Integrated Approach to Predictive Modeling: A Case Study from the Upper Xingu (Matto Grosso, Brazil)

Material Information

Integrated Approach to Predictive Modeling: A Case Study from the Upper Xingu (Matto Grosso, Brazil)
Copyright Date:


Subjects / Keywords:
Archaeology ( jstor )
Forests ( jstor )
Global positioning systems ( jstor )
Landscapes ( jstor )
Modeling ( jstor )
Pixels ( jstor )
Predictive modeling ( jstor )
Remote sensing ( jstor )
Soils ( jstor )
Vegetation ( jstor )

Record Information

Source Institution:
University of Florida
Holding Location:
University of Florida
Rights Management:
Copyright Joseph Christian Russell. Permission granted to the University of Florida to digitize, archive and distribute this item for non-profit research and educational purposes. Any reuse of this item in excess of fair use or other copyright exemptions requires permission of the copyright holder.
Embargo Date:
Resource Identifier:
670352233 ( OCLC )


This item is only available as the following downloads:

Full Text




Copyright 2005 by Joseph Christian Russell


iii ACKNOWLEDGMENTS I would like to acknowledge a ll those who played an inte gral part in my study’s development. My thanks go to my comm ittee members: Dr. Ken Sassaman, Dr. Michael Binford, Dr. Richard Stepp, and most especially to Dr. Michael Heckenberger. The data presented in this dissertation came as a direct result of funding secured by Dr. Heckenberger, and it is thanks to his dedica tion to his students, and to his spirit of academic collaboration that this study was ma de possible. I would like to thank my fiance, Michelle, without whose support none of this would have been possible. Additionally, I would lik e to thank Cooper, for remindi ng me that everything can be made better through the judicious use of sleep. Finally, I exte nd heart-felt thanks to my dog, Jasper, for reminding me daily that there is certainly more to life than writing, and that running around outdoor s and catching sticks has its own rewards.


iv TABLE OF CONTENTS page ACKNOWLEDGMENTS.................................................................................................iii LIST OF TABLES............................................................................................................vii LIST OF FIGURES.........................................................................................................viii ABSTRACT....................................................................................................................... ..x CHAPTER 1 INTRODUCTION........................................................................................................1 Selection of a Region....................................................................................................1 Synthetic Approaches...................................................................................................6 Organization.................................................................................................................9 2 AMAZONIAN ECOLOGY........................................................................................12 Xingu Indigenous Park...............................................................................................14 Description of the Region...........................................................................................18 Early Attempts to Characterize the Region................................................................20 Varzea Terra Firme Dichotomy.................................................................................24 Importance of the Upper Xingu..................................................................................40 3 PREDICTIVE MODELING AND GIS......................................................................42 Defining the Model.....................................................................................................42 Predictive Modeling and Archaeology................................................................42 Differing Approaches to the Modeling Problem.................................................49 Model Testing......................................................................................................58 Refining the Model.....................................................................................................60 Criticisms of GIS and Predictive Models............................................................60 What Drives the Model?......................................................................................65 Meshing of Archaeological Theory with Predictive Models..............................69 Integration of Landscape.....................................................................................73 Refined Approach.......................................................................................................74


v 4 GLOBAL POSITIONING SYSTEM (GPS)..............................................................78 Using GPS in Archaeological Survey.........................................................................78 GPS Use in the Upper Xingu......................................................................................88 5 REMOTE SENSING..................................................................................................99 Brief Overview...........................................................................................................99 Processing Imagery for Analysis..............................................................................103 6 METHODS...............................................................................................................121 Traditional Analysis Techniques..............................................................................122 Realization of an Integrated Approach.....................................................................124 Pre-classification Techniques...................................................................................129 Classification............................................................................................................131 Results.......................................................................................................................133 Data Transformations...............................................................................................134 Normalized Differential Vegetation Index (NDVI)..........................................136 Transformed Normalized Differe ntial Vegetation Index (TNDVI)..................141 Transform Vegetation Index (TVI)...................................................................142 Simple subtraction Ve getation Index (SVI)......................................................142 Devised Band Ratios.........................................................................................142 Soil Adjusted Vegetation Index (SAVI)............................................................142 Tassel Cap Transformation................................................................................143 Decorrelation.....................................................................................................145 Principal Components Analysis........................................................................146 Revised Methodology...............................................................................................149 7 RESULTS.................................................................................................................153 Post-Classification Procedur es and GIS Manipulation.............................................153 Discussion of the Classification Process..................................................................164 8 DISCUSSION...........................................................................................................174 Summary and Conclusions of Methods....................................................................175 Land Use/Land Change in Amazonia.......................................................................188 Perspectives.......................................................................................................188 Future Directions...............................................................................................191 APPENDIX A CONFUSION MATRICES, CLASS SPECTRAL PROFILES, AND SEPARABILITY VALUES.....................................................................................194


vi LIST OF REFERENCES.................................................................................................217 BIOGRAPHICAL SKETCH...........................................................................................268


vii LIST OF TABLES Table page 6-1. The PCA statistics for 2002 (August) transformation.............................................147 6-2. The PCA statistics for 2003 (May) transformation.................................................148 7-3. The 2002 (August) accuracy results for combined classes......................................158 7-4. The 2003 (May) accuracy results for combined classes..........................................158 A-1. The 2002 (August) combined classes confusion matrix.........................................194 A-2. The 2003 (May) combined classes confusion matrix.............................................197 A-3. The spectral profiles for the 20 02 (August) combined classification.....................200 A-4. The spectral profiles for the 20 03 (May) combined classification.........................205 A-5. The separability values for the 2002 (August) classes...........................................210 A-6. The separability values for the 2003 (May) classes................................................213


viii LIST OF FIGURES Figure page 1-1. Overview of Brazil and loca tion of the Upper Xingu study area................................3 1-2. Map of the Upper Xingu project study area showing known ar chaeological site locations (adapted from unpublished project imagery)..............................................4 2-1. Overview of study area and Xingu Pa rk southern boundaries in the Upper Xingu basin, Matto Grosso state, Brazil (5, 4, 3 color composite, May 2003)...................16 4-1. GPS survey map of MTFX-6.....................................................................................90 4-2. GPS survey map of MTFX-13...................................................................................90 4-3. 5, 4, 3 composite Landsat 7 image (A ugust 2002) with generalized legend for a rough interpretation of land-cover classes...............................................................91 4-4. MTFX-6 and MTFX-13 survey data in tegrated into a GIS with a 5, 4, 3 color composite from La ndsat 7 imagery..........................................................................92 4-5. GPS survey map of MTFX-18 integr ated into a GIS with a 5, 4, 3 color composite from La ndsat 7 imagery..........................................................................93 4-6. GPS survey maps of MTFX-19 and MTFX 20 integrated into a GIS with a 5, 4, 3 color composite from Landsat 7 imagery.................................................................94 4-7. Composite GPS survey map of the Nokugu cluster of sites integrated into a GIS with a 5,4,3 color composite from Landsat 7 imagery.............................................95 4-8. Composite GPS survey map of Kuikugu cluster of sites integrated into a GIS with a 5, 4, 3 color composite from Landsat 7 imagery...........................................96 6-1. Methodological flowchart........................................................................................128 6-2. Detail of the Upper Xingu st udy region (2002 NDVI transform)...........................139 6-3. Detail of the Upper Xingu st udy region (2003 NDVI transform)...........................140 7-1. Detail of the 2002 (August) supervised classification.............................................156


ix 7-2. Detail of the 2003 (May) supervised classification.................................................157 7-3. Detail of predicted site locations (P C1 of vegetation indices composite, PC1 of soil indices composite, PC2 of Landsat 7, August 2002).......................................161


x Abstract of Dissertation Pres ented to the Graduate School of the University of Florida in Partial Fulfillment of the Requirements for the Degree of Doctor of Philosophy INTEGRATED APPROACH TO PREDICTIVE MODELING: A CASE STUDY FROM THE UPPER XINGU (MATTO GROSSO, BRAZIL) By Joseph Christian Russell May 2005 Chair: Michael Heckenberger Major Department: Anthropology The objective of this research was to a ssess the degree to which an implementation of GPS-based survey techniques, and the extr action of information through a combination of GIS and remotely sensed imagery, in conjunction with more formal archaeological techniques and ethnoarchaeol ogical research methods, can add to the formation of a predictive modeling effort aimed at developi ng a heuristic model cap able of identifying additional significant site locations base d on the vegetative si gnature unique to anthropogenic soils in the area. The paucity of data for the Upper Xingu region of Brazil (and Amazonia in general) makes it an excellent location for the use of new technologically-informed approaches to st udying tropical land use, and providing an initial step toward an empirical understandi ng of the anthropogenic aspects of landscape formation. Important questions facing Amazonian an thropology today ar e the nature of prehistoric and present human-environmen t interaction, and th e impacts of human


xi settlement in non-western tropical settings. This dissertation a ddresses some of the problems of quantifying and qualifying the na ture of the landscape transformation resulting from long-term human occupation a nd settlement of the Upper Xingu region of southern Amazonia. I have examined the nature of the Xinguano impact on their local environment, how they transformed their im mediate landscape, and how the results of this work undermine current theories of a fundamentally non-dynamic environment coupled with models of low-density population. There is a growing acknowledgement by arch aeologists of the f undamental utility of GIS, GPS survey, and the inte gration of remotely sensed im agery into current studies. This begs more in-depth investigations in to the extent to which each technology may be utilized. The application of the predictive model of ar chaeological site locations described in this study allowed rapid identifica tion of areas with a high probability of past human occupation. This has the potential to direct more efficient archaeological exploration of the Indigenous Park of the Xingu. In all, some 1800 km2 of predicted site locations were identif ied within the region of study. During the course of the Upper Xingu Project, headed by Michael Heckenbe rger, a small sample of this region was discovered to have undergone massive altera tion by human agents (Heckenberger et al. 2003). The pattern of anthropogenic vegeta tion uncovered through this investigation adds to a mounting body of evidence that w ould seem to reaffirm this hypothesis.


1 CHAPTER 1 INTRODUCTION Selection of a Region In anthropology, geography, geology, ecol ogy, and numerous other disciplines, a consensus of opinion has been reached regarding the dramatic impact human society can have on the world around it. Often times, th ese arguments are couched in terms of the adverse nature of human impact on the e nvironment and the best ways to minimize potentially harmful consequences of human ac tivities. In much the same vein, few would argue that modern ecosystems are static co mmunities. Instead, they are viewed as manifestations of continuous and complex interactions among organisms and between organisms and the abiotic elements of their environment. Such interactions are not simply modern phenomena. This complex inte rplay of actors and elements has always existed, and the state of modern ecosystems is very much a product of their histories. The sum of the analyses in this disserta tion represent an effo rt to understand how humans and the environment within which they live act upon one another over time, manifested as transformations writ large on the landscape. My study sought to add to a body of evidence informing on larger questions cr itical to the debate about the role of humans in the formation of landscapes, a nd the impacts of long-term human occupation on the environment in non-western tropical se ttings. Specifically, th is research posits how one might be able to quantify what thos e impacts might look like in the present day by deriving a specific signature for vegetation growing in past settlement areas, and then asking how one might then identify such areas when a predictive model based on that


2 signature is extrapolated out to the larger region. The met hodological approach used in this dissertation is an initial step to an empirical understanding of the human-environment dynamic at play in the Upper Xingu. The rationale behind my study was to understand the Upper Xingu region, not just as an object of study, but also as the home for thousands of people who are coping with a heterogene ous and diverse environment on a daily basis in the present, mimicking the lifeways of an cestral populations who did much the same in the past (Heckenberger 1996, 2005), and provi ding a further point of discussion on questions of what exactly is the nature of the human impact on the environment in this region, and how do these impacts challenge theories of a fundamentally non-dynamic environment? Simply stating that human beings have a dramatic impact on their surrounding environments would seem, on the surface, a common-sensical premise. However, in the minds of some, this has yet to be proven w ithin Amazonia. Certainly little archaeology has been conducted in Amazonia (and especi ally in the Upper Xingu), and yet some researchers would have others believe that the human populations of this region were somehow less capable of transforming thei r surroundings to meet their needs. Additionally, the Upper Xingu, while researched, remains largely untreated in the anthropological literatu re. Using data gathered first-hand, and informed by the efforts of other researchers currently wo rking (or who have worked) in this and similar regions, I tried to understand, and to qualify and qua ntify, the effects of long-term human occupation and transformations of the e nvironment that accompany such activities through an integrated approach of utilizing GPS survey, geographic information systems,


3 and remotely sensed imagery. My study focused on the area occupied by the Indigenous Park of the Xingu, as well as surro unding territory (Figure 1-1). Figure 1-1. Overview of Brazil and location of the Upper Xingu study area The bulk of this dissertation is predicated on the results of more than a decade of anthropological research conduc ted by Dr. Michael Heckenberger (University of Florida, Department of Anthropology). The ethnographic materials used in this dissertation are a result of Dr. Heckenberger’s ongoing partne rship with the region’ s Kuikuro. The bulk of the GPS survey was conducted by myself and other project members in cooperation with Kuikuro informants as part of the larger National Science Foundation funded project “Late Prehistoric Social Complexity in Southern Amazonia (Upper Xingu, Brazil).” The primary purpose of this research was to document regional settlement patterns (size, placement, and form of occupation sites) and determine if there was a rank-order settlement hierarchy. Research to date is summarized in The Ecology of Power: Archaeology and


4 Memory in the Southern Amazon (Heckenberger 2005) and some recent findings have been reported in Science (Heckenberger et al. 2003). Heckenberger’s efforts in the Kuikur o study area have identified numerous prehistoric settlements (Figure 1-2), prim arily through major earthworks. Detailed mapping was conducted over most major earthw orks at six prehistoric archaeological sites, demonstrating a remarkably complex a nd integrated regional plan. The earthworks include: 1) excavated ditches in and around an cient settlements (up to 2.5 km long and 5 meters deep); 2) linear mounds bordering majo r roads and circular plazas (averaging about 0.5 to 1.0 meter high); and 3) a variety of wetland features (bridges, weirs, artificial Figure 1-2. Map of the Upper Xingu projec t study area showing known archaeological site locations (adapted fro m unpublished project imagery)


5 ponds, raised causeways, canals, and other st ructures) that are often less obvious but clearly integrated into the re gional plan (Heckenberger et al. 2003; Heckenberger 2005). The research conducted thus far in as part of Upper Xingu study has revealed a rank-order hierarchy, including a number of la rger first-order occ upation sites, with multiple plazas, and earthen feat ures that seem to indicate the remains of large population centers (Heckenberger et al. 2003; Heckenberg er 2005). There also appear to be some second-order settlements, with single or double plazas, comparatively fewer earthworks, with evidence of ties to larger centers, as we ll as some third-order plaza villages that have one plaza, are much smaller, and have earth works only in the immediate area of the plaza (Heckenberger et al. 2003; Heckenberger 2005). Thus, the Upper Xingu provides a wide arra y of data sources upon which to draw, including preserved landscapes , indigenous populations, and archaeological sites. What is lacking is any meaningful attempt to produce information using methods that are technologically based. Thus, this region provi des an opportunity to form some guidelines for integrating new technologies with traditiona l archaeology to refute older theories and perhaps posit new questions. The model of low density prehistoric populations in the Amazon is outdated, and generally agreed on to no longer accurately refl ect past settlement patterns. In light of recent modern research into the nature of human-environment interactions, we have begun to understand the Amazon, and the people who have occupied that region, in a new light. At issue is the vast territory to be covered, and a means by which we can quantify and qualify precisely what these interactio ns are, and how humans transform their environment over time. The amount of land area to be studied seems, on the surface,


6 prohibitively expansive for traditional ar chaeological investigative techniques. Thankfully, new technologies vastly increase our ability to operate ove r larger scales and with greater efficiency and rapi dity than previously possible. Synthetic Approaches Use of newer technologies that can be in tegrated into most methodologies to alter how we archive, manage, analyze, and disp lay data is becoming more widespread in archaeological research. The potential of thes e tools can be realized only if their varied functions are correctly underst ood, and if they can effectively interact with one another. If recent literature is any i ndication, the archaeological wo rld has become increasingly curious and interested in th e application of Geographic Information Systems (GIS), Global Positioning Systems (GPS), and Remo te Sensing (RS) to anthropological problems. The main impediments to full adop tion of these new tools (and a move away from entrenched methodologica l approaches) hinges on the di fficulty in using software packages, which are sometimes complex to im plement and operate; and the difficulty in integrating varied datasets, in different form ats, coming from different sources, and based on fundamentally different models of analysis. The answer to these problems will be the availability of interopera ble geoprocessing and imageprocessing technologies. By meshing traditional archaeology with an integrated approach to these new technologies, this dissertation attempte d to understand the long-term aspects of human/environment interaction in the Uppe r Xingu region by interpreting archaeological evidence and remotely sensed imagery w ithin the context of a new methodological approach to data collection, processing, and anal ysis. This integrativ e approach sought to extract the strengths from each system to contribute to a synthesized methodological approach for predicting site locations with in the Upper Xingu of Brazil using spectral


7 signatures of specific training sites as the ba sis of defining location parameters. The goal of this research is not to replace grou nd-truthed or ground-informed archaeological investigations; merely to develop a method to increase the efficiency with which specific regions of interest might be targeted. These findings will have implications, not just for archaeologists, but also, to other disciplines , as well as advancing our general knowledge of the region. This study resulted in a realized model, but perhaps its va lue is greater in demonstrating how various analysis and data collection platforms may be integrated. One of the greatest obstacles to successful synthesis of tec hnology into the design of an archeological model is that the archaeologica l data are often widely varied in content (anthropic or natural facts, s ites, features, historical texts) , thus the information is not homogeneous (aerial/digital/near infra-red photographs, remote sensed images, drawings, plans, contour surveys, paper/digital maps, CAD layers, GIS databases, field notebooks, and so on). Such is the power of a GIS that va ried data sources can be treated as different layers of information. Thus, a predictive model derived from the informed use of multiple data sources can be realized. An elementary assumption in any predictive model of prehistoric activity loci is that some locations are more suitable for conduc ting specific activities than others, or that some locations were of greater strategic, religious, or other cu ltural importance. It is also assumed that any attributes contributing to th is differential suitability can be quantified and represented using modern data sources, or can be represented by proxy data sources. To date, the majority of predictive models formulated by archaeologists in the United States demonstrated that the variables c hosen by archaeologists to correlate with


8 archaeological sites often repr esent noncultural aspects of the landscape. Though these variables form the backbone of the “predi ction,” even the archaeologists themselves oftentimes explicitly acknowledged that the variables might have no causal relationship with the placement of sites. Such variab les are considered indicators. While these variables often have no causal relationship w ith the placement of sites, when taken in combination, they invariably re sult in the discovery of site s. Contemporary efforts at archaeological prediction explicitly or imp licitly assume that environmental factors are the primary (even exclusive) determinants of most human behavior. The causal link between site locations and natural, independent variab les is usually considered multivariate, meaning that sites were positioned to take advantage of an optimal combination of all the resources critical to the cultural group under investigation (Ebert and Kohler 1988:107). The fundamental flaw in models of this sort (setting aside the environmentally deterministic overtones) is the subjectivity involved in assessing the potential of prehistoric archaeological site existence wh ile relying primarily on the expertise and experience of archaeologists. While subj ective knowledge should never be discounted (and, in fact, may be one of the most valuable resources available to the modeler), such efforts are flawed from the sta ndpoint of replicablity outside of very specific regions. My model differs from many of the existing m odels in its attempt to utilize indigenous knowledge and contemporary cultural patterns , and incorporate that understanding into the modeling process. Any effort to predict where people chos e to settle in the past requires some understanding of the people and their activities. Initially, it might seem burdensome and


9 problematic to introduce anthropological expl anation and culturally informed variables into a discussion of practical ar chaeological prediction. However, it is only in the context of explanation and explanatory modeling utiliz ing data from numerous sources to derive variables outside of simplistic geomorphol ogical feature proximics so common in modeling to date that archaeologists can su ccessfully predict the locations and other characteristics of the materials that make up the archaeological record. Variables that determine the locations of occupational sites are not static properties of the environment that can be measured easily from topographic or environmental maps (Ebert and Kohler 1988:158-159). No one would argue that geographical or environmental variables (often characterized by landscape featur es such as river flats or te rraces, joining or narrowing bodies of water, soil drainage among others) are effective for finding sites, but it is questionable whether they are finding the tota l range of site types available or only the sites they expect to find. Models based on culturally uniform vari ables are valid to a certain extent, and often times will find more sites than would be found by pure chance. However, adding other types of data, and intr oducing various platform s of data collection broadens the scope of modeling efforts, expand ing the range of identified sites, ultimately proving more useful for managing archaeologi cal resources in areas where insufficient archaeological work has been conducted. Organization The organization of this dissertation reflect s the multi-disciplinary character of the research. Each chapter is devoted to a topic supporting th e development of a methodological approach to the integrative use of technology in the Upper Xingu region to expand our understanding of human-environm ent interaction in that region. Chapter 2 describes the theoretical, environmental, and cultural setting for the study. The Upper


10 Xingu is placed squarely into the context of the Amazon region and the ongoing debate within anthropological circle s governing the capacity of the environment to support large-scale settlement of human populations. The next three chapters provide an introduc tion to each of the systems that will be used is provided, including a short synopsis of how each software/data collection platform operates. Chapter 3 addresses the theoretical development of predictive modeling approaching in gene ral, and in anthropology in particular. The provided overview of predictive models includes comm entary on their inception, development, and the fine-tuning of their components. In a ddition, some of the inherent weaknesses of current modeling trends are demonstrated, a nd a list of critiques of predictive models outlined. A short section is also provided regarding the development of predictive models pertinent to these investigations. Th is chapter also presents the first of three technologies used in this study, geographic info rmation systems (or GIS), and explains its utility, especially in light of modeling efforts Chapter 4 contains an expl anation of the second utili zed technology, GPS. The overall system of GPS is explained in deta il. Additionally, arguments are made for the greater inclusion of GPS in archaeological su rvey, including an explanation of the many benefits to using GPS over traditional surv ey methodologies. Finally, I present the GPS survey data collected in the Upper Xingu, demonstrating the util ity of GPS survey methods as a critical aspect of integrative approach methodol ogies. Data collected in the field supported the creation of a database in cluding size of archaeological sites, and was essential to the overall mapping of the project area. Moreover, spectral data referring to


11 the specific regions of interest were tied directly to georefer enced points and were integrated into the database. Chapter 5 expounds on the utility of remote sensed imagery, and specifically Landsat imagery, to provide a backdrop for the formulation of an overall approach methodology. In Chapter 6, the integrative approach is finally realized. As this study involves the use of Landsat imagery, in combination with a variety of image processing techniques, GPS surveyed archaeological data sets, and the integration of thes e two types of data within a GIS to form a cohesive whole, this chapter is devoted to the method of classification of the overall image, and the resu lts of those classifications, as well as the output of the final analysis product. Chapter 7 presents the analysis of the clas sified imagery, as well as a reporting of the findings and their implications to the la rger arguments regardi ng the degree to which populations with the Xingu altered their surrounds. In the final chapter, concluding remarks are addressed. Questions and hypotheses are discussed based on the main findings of the dissertation. Further studies are also suggested in search of a be tter understanding of the relationship of humans and their environment in the Amazon.


12 CHAPTER 2 AMAZONIAN ECOLOGY The Upper Xingu region of Brazil provide s an excellent backdrop for addressing questions relating to the nature of the la ndscape transformation th at has occurred as a result of long-term prehistoric human occupa tion, specifically in re gards to discussing what the dynamic interaction of the Xingua no and local environments, and how they transformed their landscape. The paucity of data for this region makes it an excellent candidate for the use of new approaches to re searching land use in tropical settings. This dissertation sought to apply t echnologically-informed research techniques to provide an initial step towards an em pirical understanding of landscape formation through the formation of heuristic model designed to pred ict archaeological site locations through the spectral signature of anthropogenic vegetation. The Amazon retains a prominent position in both Western popular and scientific imagination as a region relatively untouc hed by human hands. This concept of a “pristine” environment is be lied by the archaeological evid ence suggesting the presence of large, densely settled, and integrated regional populat ions (Heckenberger 1996; Heckenberger et al. 1999; H eckenberger et al. 2003; Heck enberger 2005; Petersen and Heckenberger 2001; Porro 1996; Roosevelt 1980, 1991; Whitehead 1998), a concept that has long been suggested by several researcher s in the area (e.g., Ca rneiro 1970; Denevan 1976; Lathrap 1968, 1970). Demonstrable evidence for large, complex social formations in the Upper Xingu, and the extensive influe nce these populations may have had on the natural environment has important implicati ons for how anthropologi sts, ecologists, as


13 well as researchers in many other discipline s, view the non-Western peoples indigenous to the region, and the nature of the landscape transforma tion these peoples may have performed. The attention to anthropogenic forests, beginning two decades ago, was a paradigm shift in Amazonian ecology and ethnology, and has motivated research on humanenvironmental interactions in the region (Balee 1987, 1989, 1993; Balee and Campbell 1990; Brondizio 1996; Brondizio et al. 1994, 1996, 2002; Carniero 1983, 1985, 1987, 1995; Heckenberger 1996, 1998, 1999, 2005; H eckenberger et al. 1999, 2001, 2003). However, perceptions of the transformativ e nature of human occupation on surrounding environments are certainly not a new concep t. As early as the mid 1800s, von Humbolt pondered “plant geography,” a collective study of vegetation, and ar gued, “the vegetation of a region was an expression of the physical environment, and also a direct influence on Mankind, both materially a nd spiritually” (Nicolson 1987: 177). George Marsh’s Man and Nature or Physical Geography as Modified by Human Action (1864) also grew out of this growing acknowledgement of the fundame ntal link between human populations and the environments within which they lived. This concept of the dynamic nature of human/environmental interaction remained an undercurrent through much of the 20th century in anthropology (Boas 1887), ecol ogy (Clements 1936; Sayce 1938), and in general writings and thought (Sauer 1956; Bates 1956; Mumford 1956). This trajectory of literature regarding the anthropogenic natu re of environments, and the transformative processes human actors perform in the forma tion of landscapes formed the foundation for much of modern thought on the subject.


14 In light of the pedigree and abundant nature of literature regarding these issues, it does seem surprising that theories of a f undamentally non-dynamic environment would have persisted for so long, or that some re searchers would have ot hers believe that the human populations of this region were so mehow less capable of transforming their surroundings to meet their needs. Modern considerations of the Amazon demonstrate a near-universal agreement of the widespread anthropogenic transformations of vegetation. Indeed, Heckenberger et al. ( 2003) effectively correlated spatia l and structural patterns of vegetation and archaeological re mains with settlement distribution, adding to the body of evidence of these large-scal e transformations. Through th e use of an integrated technological approach, I sought to add to this body of evidence, specifically in describing how researchers might assess the ex tent of landscape tr ansformation through a heuristic modeling effort. Xingu Indigenous Park As stated in the introductory chapter, this dissertation centered on the Xingu Indigenous Park (or PIX after the Brazilian designation of Parque Indigena do Xingu), specifically encompassing most of the Uppe r Xingu cultural area, as well as areas immediately to the north of the Xingu Ri ver proper (Figure 2-1). The Upper Xingu contains a number of divers e ethnic groups including the Aueti, Kalapalo, Kamayur, Kuikuro, Matipu, Mehinaku, Nahuku, Trumai, Wauj a, and Yawalapiti. While retaining distinct cultural differences , these groups maintain a high ly interconnected network of specialized trade, marriages, and inter-village rituals. The PIX is located in the northeastern part of th e State of Mato Grosso, in the southern part of the Brazilian Amazon. Comprising some 26420.03 km2, the park encompasses enormous biodiversity and a number of distinct ecological communitie s, ranging from savannas and dryer, semi


15 deciduous forests to the south, to the ot her end of the scale with Amazonian ombrophyllous forest to the north (includi ng extremely dense primary and secondary forest cover, fields, flood land forests, terra firme forests, and forests on terra preta ). Climatically, the region alternates between a wet season (November to April) and a dry season during the rest of the year. The Park was formed July 31, 1961, with adjustments made in 1968, and 1971, with the final perimeter demarcation complete d in 1978. Deemed a “National Park,” this region was intended to serve both as a protected ecosyste m and as a haven for the indigenous populations that guid ed its creation. With the formation of Funai in 1967 (replacing the SPI, or Indian Protection Service), the “National Park” designation was dropped in favor of the “Indigenous Park” moniker.


16 Figure 2-1. Overview of st udy area and Xingu Park southe rn boundaries in the Upper Xingu basin, Matto Grosso state, Braz il (5, 4, 3 color composite, May 2003) The Indigenous Park of the Xingu consists of three generalized regions: one to the north (known as the Lower Xingu), one in th e central region (the Middle Xingu) and one to the south (the Upper Xingu). The southern regions harbor the primary feeder rivers of the Xingu river system and contain a closely kni t federation of distin ct cultural groups. With final demarcation of the boundaries of the park in 1978, only a few years elapsed before the first incidents i nvolving encroachment into pa rk boundaries. At first, trespassing was in the form of hunters and fisher men at an individual, low-impact level.


17 However, by the end of the Twentieth Century, numerous forest fires on cattle ranches to the northeast, as well as the encroachment of lumber companies to the west began to compromise the park borders. Additionall y, occupation and agricu ltural practices near the headwaters of the park river system cau sed increasing pollution of the internal water supply. This pollution, in addition to othe r forms of encroachment on park boundaries, are among the most pressing issues to the popula tions resident in the region. Indigenous groups in the park, however, have proactiv ly in defended the s overeignty of park boundaries though litigation. Two te rritorial lawsuits, in partic ular, resulted in the Wawi and Batovi Indigenous Lands, of the Suy a nd the Wauja respectivel y. Ratified in 1998, these expanded areas brought the total land area of the Park to approximately 2,797,491 hectares. The question of monitoring th e territorial boundaries is pa ramount in the minds of the indigenous population, often discussed in meetings of leaders, assemblies of the Xingu Indigenous Land Associati on (ATIX), and in meetings with the Funai and federal and state environmental agencies (IBAMA and the State Environmental Foundation, or FEMA). Eleven vigilance posts have been es tablished thus far to protect and buffer areas that allow direct access to the Park, as in the case where highwa y BR-080 borders park boundaries or where major wate rways intersect the park. While vital to protecting the interests of the indigenous populations, this series of posts alone is insufficient deterrent to boundary violations. Other systems have been put into place to assist the park inhabitants in ma intaining the integrity of park borders. For example, a coordinated partnership between AT IX and ISA led to an ongoing project (the Borders Project), that systematically maps deforestation for trend analysis and use in


18 litigation, as well as id entifies new vectors of settlement and occupation of areas in close proximity to park borders. This joint partnership has ta ken on the task of construc ting training programs for the vigilance post inhabitants, restoring a nd maintaining boundary markers delimiting the physical limits of the park, and maintaining a ge o-referenced database of all ranch owners bordering the Park. The Borders Project a llows park inhabitants to closely follow breaches of park boundaries, mobilizes reside nt communities, and provides a medium for dialog through inter-village discussions and with the publ ic agencies responsible (FUNAI, IBAMA, and the state government). Description of the Region The most striking feature of the Amazon Ba sin are its diversity of ecosystems and the conglomeration of many distinct and diffe rent cultures (contrary to traditional portrayals of limited ecological and cultural di versity). Though research efforts into every aspect of the Amazon (f rom anthropological studies to botanical categorizations), have made numerous leaps from the relative paucity of information catalogued by the mid-20th century, we still lack a robust understand ing of the complexity of the region. Efforts to construct prehistory and protohi storic periods of tr opical lowland South America are severely hindered by the fragme ntary and often unreliable nature of the cultural data in some instances, and by the sh eer lack of any real information at all in others. Despite the efforts of numerous resear chers to begin to understand the vast areas of the Amazon Basin, a great deal of the ar ea is still unknown archeo logically. Thus, the Amazon has often been portrayed as a pristine habitat, with indigenous peoples viewed as little more than Stone Age peoples living in harmony with the ancien t, unchanged forest. Such models have been summarily disman tled by evidence of significant changes in


19 environment and human adaptation before th e European conquest of the Americas. Perhaps the most important pieces of eviden ce to date suggest th at ancient Amazonians made considerable impacts on the habitat in ar eas adjacent to their settlements, and that some distinctive forest patterns once thought purely natural now may be seen as having been influenced by past human activities. Anthropologists have made numerous atte mpts to categorize (and organize) the Amazon into a number of distinct cultura l areas, and much of the current thought governing potential social complexity in th e Amazon can be attri buted to an evolving understanding of the cultural history of indigenous populations. Early studies on the region tried to explain the appa rent lack of social complexity by the existence of “limiting factors,” thought responsibl e for relatively low population densities (Heckenberger 1996, 1998, Heckenberger et al. 1999, 2003). Meggers ( 1954) suggested that levels of social complexity are limited by the agricultural poten tial of soils in the Amazon. Others have argued that protein sources, not necessarily the production of highly caloric crops, have limited population densities throughout the region (Gross 1975; Ross 1978). Recent archeological evidence from the re gion shatters many preconceived notions of the nature of complexity in pre-Colu mbian Amazonia, proving the existence of complex prehistoric cultures in the Amaz on (Lathrap 1970; Myer s 1973; Roosevelt 1987, 1989a, 1989b; Heckenberger 1999, Heckenberg er et al 2003). Agronomic and biophysical surveys demonstrate a diversity of soils and land resources, including areas suitable for large-scale agriculture (F alesi 1974; Cochrane and Snchez 1982; Nicholaides et al. 1984). Th e very process of coloni zation, evidenced today in ethnographic studies of population migra tions, and shown in the past through


20 archaeological investigations, suggests that previous views of the Amazon were overly simplistic, and that much more complex sy stems are in play, requiring a fundamental shift in the way researchers c onstruct their questions and seek answers if we hope to ever understand the potential of the Amazon for deve lopment in the past and in the present (Moran 1981, 1984; Schmink and Wood 1984; Bunker 1985; Hecht and Cockburn 1990; Stewart 1994). Early Attempts to Characterize the Region Steward provided the first substantia l body of information governing cultural development in South America with the publ ication of the Handbook of South American Indians (HSAI). In his gene ral theory, Steward promoted the viewpoint of cultural ecology, which portrayed cultures as func tionally integrated wholes, possessing technology designed to maximize ecological expl oitation. In Amazonia, the environment was viewed as the primary limiting factor in cultural development (Roosevelt 1980: 3). He classified South American cultures accord ing to a list of characteristics focused on exploitation of local environments. The fo cus on environment and material culture, especially those items utilized in expl oiting resources, over the symbolic-ritual complexes enabled Steward to create ci rcumscribed geographic boundaries wherein shared cultural traits delin eated “culture areas.” Steward’s Handbook laid the groundwork for an environmental determinist, or ecofunctionalist (Heckenberger 2005), perspectiv e in Amazonian scholarship where direct linkages were suggested between population de nsity and natural resources. Steward was the first to maintain that variance in populat ions was due to ecologi cal constraints, and for a number of decades this line of r easoning persisted. Typical exchanges between Amazonian scholars consisted of debate over the role of the envi ronment in conditioning


21 human adaptive strategies and guiding the course of cultural developm ent in the region. The environment was viewed in terms of abso lute constraints it posed to human societies and cultural evolution, as typifi ed in Betty Meggers’ (1954, 1996) version of Steward’s environmental determinist viewpoint, which she believed would bett er explain cultural development in general and tropical lowland culture in particular. Meggers seized on Steward’s “Tropical Fore st Tribes” with its core of unifying traits. Using Steward’s definitions, the Tr opical Forest culture embodies a number of signature traits, notably a re liance on hunting and/or fishi ng, gathering, and rudimentary slash-and-burn cultivation fo r subsistence. She strengt hened Steward's argument by collating evidence on the poverty and fragility of tropical soils. Meggers believed that both patterns of settlement and formation of a synthesized culture were a direct result of the amount and quality of cultivable land av ailable to a given population. Meggers in effect argued that the highly leached soil substrate and rapid decomposition of organic matter in humid tropical climates imposed ab solute limitations on the possibilities of intensive cultivation and soil im provement (which in turn were determinants of the slashand-burn cultivation system prevalent among upland horticulturalists by ethnographic accounts), resulting in local population pre ssure which could be felt even under low density unless people lived in small, nomadic communities. In “Environmental Limitation on the Development of Culture” (1954), Meggers asserted that the primary locus of interac tion between a culture and its environment can be found in subsistence activitie s, and that the most vital as pect of the environment, from the point of view of the participants, is its suitability for food production. She outlined the importance of agriculture in cultural de velopment and identified the effects of soil


22 fertility, climate and rainfall as critical to the productivity of agriculture. In turn, productivity regulates populati on size and other factors relate d to the survival of the group (1954). Her working hypothesis was that there must exist a casual relationship between the type of environment and the maximum cultural development that a given environment can support. Meggers’s hypothesis was the central them e in discussions concerning Amazonian cultural development for decades, centeri ng on whether or not various subsistence economies could provide the basis for sedent ism and population growth. In general, Meggers’s assumed that the limiting factors on population size and density were tied to the ability of human cultures to adapt to heterogeneous environm ental conditions. She dismissed the possibility of the presence of large-scale sedentary populations in Amazonia, due to environmentally limiti ng factors inconducive, in her mind, to supporting large populations. In essence, Megge rs’ dismissed any ability of prehistoric populations to manage environmental conditi ons to achieve levels of subsistence necessary for population growth and mana gement (Gross et al. 1979; Moran 1982). Utilizing the trope of the “Tropical Fore st Tribe” (denoting a limited population and essentially placing a cap on the deve lopmental sequence of the peoples under investigation), Meggers has tena ciously held onto a belief th at the small populations that characterize the impermanent ethnographic comm unities of modern times can sufficiently characterize the full extent of variability of cultural groups of the past (Roosevelt and Meggers 1996, 2001; Heckenberger et al. 1999, 2001). Deemed the “Standard Model” (Viveiros de Castro 1996), th is outdated point of view was subsumed by revisionist thinking expanding on Julian Steward’s initia l observations (1946) that perhaps the


23 different environments represented by floodpl ain and upland areas might hold the key to some of the cultural variation he observed. In gross terms, regional specialists typically delineate two ecological zones within Amazonian: terra firme was defined as land not subject to annual inundation, with elevations varying from immediately above flood level to several hundred meters, and varzea as the flood plain of “white water” rivers, laden with fertile silt. Some authors question, howev er, whether their distinction is particularly relevant to cultural diversity, at least when framed in such simplistic, dichotomous terms (Whitehead 1996; Heckenberger et al. 1999, Heckenberger 2005). Certainly there is widespread acknowledgeme nt on the part of most researchers in the Amazon that, at least within the varzea regions, large, sedent ary societies could and did exist. Archaeological evidence points to ceramic-making Archaic Sedentary Shellfishers (evidenced by fluvial and maritime shell middens) that formed settled villages as early as 7500 BP. By 6000 to 4000 BP, there is evidence of initial riverine agriculture based on diversified root crops, fru it trees, and numerous other tropical forest resources in parts of Amazonia (Heckenbe rger et al. 1999; Lathrap 1970, 1977; Neves 2001; Neves et al. 2003; Oliver 2001; Petersen et al. 2001; Roosevelt et al. 1991). Later (2000-3000 BP) sites on the Orinoco River de monstrate the widespread adoption of ceramic griddles (postulated to have been us ed in manioc preparat ion) and pottery forms generally associated with the Amazonian Barrancoid, or Inci sed-Modeled ceramic tradition, providing eviden ce of early complex societies with intra-regional interaction. By around 3000 BP, sites in the Central Amazon and the middle Orinoco exhibit evidence of institutionalized social hierarchy, with a grow th of regional chiefdom-like


24 organization by about 2000 BP (Carnier o 1995; Heckenberger 2001; Lathrap 1977; Rossevelt et al. 1991). The Marajoara culture at the mouth of the Amazon (developing in situ by ca. AD 400) has been revealed as a complex chie fdom boasting numerous villages or towns containing numerous domestic structures , cemeteries, and ceremonial complexes (Roosevelt 1991, 1994, 1999; Scheen 1997, 2001, 2004)). The complexity of the mound construction suggests long-term habitation and ce remonial use of the sites, sometimes in groups of up to 40 related mounds; these are in terpreted as small regional polities Scheer 2005). Cultural distinct polychrome po ttery began to be produced by AD 400-600, reaching a florescence by 1000 AD. This mound-building culture used ceramic urns for burial, both in cemeteries and accompanyi ng single/few individual burials. Within the middle Amazon, Amazonian Barr ancoid persisted until about 900 AD at sites like Acutuba, Manacupuru, and Paredao (Heckeberger et al. 1998). Incised and punctuated styles became dominant along the Lower Amazon by 1000 AD. The best examples of this ceramic style are Tapajonica from Santarem, at the mouth of the Tapajos River, and the related Konduri style upstr eam around the Trombetas River. When Europeans encountered the Tapajos in the 1500s, they observed a continuous strip of settled villages along the rive rbanks, organized as chiefdoms, with village chiefs subordinate to a paramount chief (Medin a 1998, Porro 1996). Agricultural production likely centered on the cultivation of both manioc and maize (Roosevelt 1991). Varzea Terra Firme Dichotomy To explain archaeological evidence of co mplex social organization in numerous floodplain areas and Contact-period accounts of large Amerindian groups with complex social structure, many researchers (e .g. Carneiro 1970, 1986, 1995; Deneven 1984, 1996;


25 Lathrap 1960, 1970; Meggers 1996; Roosevelt 1980, 1989, 1991, 1994) have hypothesized a varzea model emphasizing that the rich floodplain soils could, in fact, support large, fully sedentary villages (s ee Heckenberger et al. 1999). Termed the “ varzea terra firme dichotomy,” this hypothesis was designed to account for the distribution of fine ceramics, large populations, and monument al architecture within the floodplain regions of the Amazon and its Andean derived tributaries. These elements were believed to be representative of comp lex social formation suggesting the possibility of early chiefdom-level organization of cultu res within this region. Proponents of this model suggest that the higher produ ctivity of the soils located in varzea regions provided the basis for sedentary populations to settle in to these riverine areas, and, conversely, that the reduced fertility of soils in the terra firme limit growth (thus, inherent in this theoretical model is a bias ag ainst the formation of comple x social organization in the terra firme regions). This revisionist view accounted for the increasing archaeologi cal evidence of the formation of complex social organization in the varzea , as well as the seeming lack of complexity in the upland areas. In a bizarre leap of logic, many were led many to believe that the ethnographic present of displaced, disenfranchised groups could sufficiently explain the nature of terra firme social organization in th e past, but, in an amazing display of duplicitous logic, they found that the mode l was insufficient to explain the level of social organization in floodplain ar eas clearly documented in Contact periods accounts. Continued excavation of sites within the floodplains, and the lack of any real archaeological investigations into the devel opment of more complex social organization in the upland areas, seemed to bolster the varzea hypothesis.


26 This varzea terra firme dichotomy became the launching point of a number of discourses attempting to unravel the apparent lack of developm ent in interfluvial regions. Lathrap (1968) maintained an adaptivis t view, expanding on the division between terra firme -interfluvial groups and varzea populations engaged in mo re sedentary root-crop farming combined with fishing and the hunti ng of aquatic mammals and reptiles. Lathrap placed primacy on the contrast between the tw o ecological zones in terms of differential productivity, both in agriculture a nd animal bio-mass. Lathrap’s thesis is that most of the groups inhabiting the tropical inland are the remains of evolved agricultural societies forced into an environment unsuitable to the basic economic pattern (1968). They had to rely on the hunting of forest game to provide the protein and fat essent ial to the diet. As game was often scarce, they were forced into nomadism, a decline in agricultural productivity, and a still greater dependence on wild food (Lathrap 1968). This led to Lathrap’s formation of the “Tropical Forest Culture” (1970), significantly different from early categorizations of upland settlement gr oups in that it addressed shared economic and cultural patterns independe nt of Morgan-esque evolut ionary stages of cultural development. Carneiro also felt that earlier explanati ons were too simplistic and static. Late nineteenth century travelers' opinions that the soils of the region were rich and productive were quickly followed by suggestions that soil s were so poor that they could not support complex cultures or intensive cultivation. Wh ile it is true that tropical rain forests typically have the highest biomass and species diversity of any terrestrial ecosystem, the selfsame climatic conditions that create an id eal environment for a wide variety of flora is also ideal for chemical weat hering, erosion, and the destruct ion of humus. Soils in many


27 parts of the Amazon Basin are generally cons idered poor, leached, acidic, and capable of producing for only a couple of years before being abandoned due to fertility decline (McNeil 1964). The soils of the upland tropical forests ar e dominated by highly acidic and nutrient-poor zones. The bi ological diversity of the upla nd forest regions is directly related to nutrient recycling and a number of complex, extremely localized adaptations of plants regional conditions, as well as land management approaches practiced by local populations in the past and pres ent rather than any inherent qualities of the soils of the regions (Posey and Balee 1989). Furthermore, indigenous technologies and a wide range of domesticated and semi-domesticated plan ts provide diverse avenues for increased productivity. Within the some floodplain regions, soils are rejuvenated each year. As a matter of course, it is more useful to say the humid tr opics include some of the poorest and some of the richest soils in the world (Sanchez 1976, 1977). In 1954, Meggers set out to demonstrate that the poor and acid soils of the humid tropics could only support smallscale societies living by swidden cultivation beca use soil fertility could only be sustained from the nutrients released by the slash-a nd-burn method of cultivation. Denevan (1976: 225) called attention to the ex tremely low estimates of population density made by Julian Steward (1946), who suggested densities of barely 0.2 persons per square kilometer for lowland South America. Of course overall population density would have been highly variable, due to any number of localized conditions like produc tivity of soils, availability of edible flora, and concentrations of suitable fauna. In general, it has been suggested (and may well be true) that the upland interfluvial forests had lower population densities than savannas and floodplains in lowland South


28 America. Nonetheless, it remains to be wide ly demonstrated that this is uniformly or even typically the case. Preh istoric populations ov er the centuries seem to have brought about significant environmental modification th at has enhanced the net returns to human populations and ensured that familiar environm ents are present from one generation to the next. For a long time afte r the initial arriva l of the Europeans, the upland forests offered refuge to native populat ions fleeing from the incursio ns of colonial and national societies. Carneiro challenged several of Meggers' notions about the limits imposed on Amazonian culture by a rainforest envi ronment (Carneiro 1970, 1983, 1985, 1995). In a study among the Kuikuru of the upper Xingu River, he estimated the carrying capacity of a slash-and-burn system, and his calculations suggested a much denser population limit and a more elaborate cultural development than Meggers' es timations (Carneiro 1995). With ethnographic data from the Kuikuru, Carn eiro states that "with slash-and-burn as the only limiting factor, a village of some 2,000 persons could live on a permanent basis where the Kuikuru do now” (Carneiro 1961: 232 cf. Carneiro 1995: 48). It turned out that any soil limitations could be corrected through the addition of organic matter, mulches, irrigation, and overall proper soil ma nagement. Carneiro's studies flew in the face of Meggers’ ecological limitation of so il fertility, and his studies have mostly focused on the role of other ecological factor s related to food crops in the evolution of Amazonian chiefdoms. Carneiro developed a hypothesis to explain the rise of complex social organization in the varzea that actually began to trend away fr om environmental determinism, seeking explanation in “such things as the availability of varzea , the huge quantities of aquatic food resources found in the waters of the Amazo n, and other ecologica l factors, including


29 population pressure and warfare” (1995: 52) . Roosevelt repudiated Carneiro’s environmental circumscription theory, o ffering her maize hypothesis (1980) as an alternative means of overco ming the believed environmental limitations of the Amazon and providing the impetus for the formation of chiefdom-level social organization. Roosevelt’s maize hypothesis has itself been brought under severe scrutiny (Carneiro 1995) and all but dismissed by Roosevelt herself. To date, the general disagreement between all parties involved ove r the nature of environmenta l limitations has been heated, and there has been an unfortunate tend ency to focus on single limiting factors (Beckerman 1979 and Gross 1975 on protein availability; Meggers 1954 vs. Carneiro 1957 on soil fertility). In spite of an ever-increasing body of knowledge about numerous aspects of Amazonia demonstrating irrefutably just how va riable tropical habitats can be, there still seems to be some impetus to hold onto vestigial linkage with existing hypotheses based on earlier viewpoints. For some inexplicab le reason, many researchers in the Amazon continue to suggest that the fi ndings from one site are genera lizable to the entire region. Most anthropologists wholehearted accept the varzeaterra firme dichotomy, categorizing as terra firme adaptation a slew of findings from ecologically disparate regions as the Xingu Basin, the Rio Negro Basin, and the cent ral Brazilian savannas. At its heart, however, the dichotomy is re plete with false assumptions and broad generalizations that render the entire basis of the theorem untenable. Scientific studies across numerous disciples have be gun to recognize the complexity of the of the region, and the fall acy in derivation to so simple a picture (Moran 1993, 1995). On the one hand, some basic findings were generalized for the


30 entire region as 'unifying principles': a gene ral scarcity of nutrients in the soil (Sombroek 1984); a tightly closed, continuous recycling of nutrients within the biomass of the forest (Jordan 1989); extreme diversity of the biota (Prance and Lovejoy 1985); and the regional recycling of a large part of the ra inwater, crucial for the maintenance of a climate affected by pluvial processes (Schuba rt and Salati 1980; Dickinson 1987). On the other hand, the heterogeneity of such a broad territory was recognized in several fields. Among other regional elements, Amaz onian river types were differentiated into whitewater (e.g., Solimes-Amazonas, Madeira) , clearwater (e.g., Ta pajs, Xingu), and blackwater (e.g., Negro) (Sioli 1984). Sombro ek (1984) divided soils into three major categories: well-drained soils of the uplands, imperfectly drai ned soils in the sedimentary parts of the region, and p oorly drained soils in varzeas and igapos. Pires (1984) and Pires and Prance (1985) have describe d several vegetation types. A myriad of papers have addressed the human dimension behind the environmental heterogeneity (Balee 1989; Deneven 1984; Hames and Vickers 1983; Meggers 1974; Moran 1974, 1995; Moran and Herrera 1984; Posey and Balee 1989; Sponsel 1992). Recent scholarship has begun to refute previous positions, providing evidentiary information regarding the numerous techniqu es used by indigenous groups in social (regional trade, kinship ties ), technological (food storage, resource procurement) and ecological (settlement in proximity to multiple ecological zones) realms to buffer from potential environmental constraints (Denevan 1996; Posey 1983). In addition, substantial evidence is available for the modification of the landscape by na tive populations, leading to the creation of novel resources and demons trating a greater degree of human agency with the environment (Balee 1989; Clay 1988; Denevan 1992). Perhaps the most


31 powerful evidence comes from Heckenberger’s 1999 publication that demonstrated the existence of large, fully sedentary populations in a broad range of ecological settings in pre-contact Amazonia (Heckenberger et al. 1999), further discrediting environmental determinants of cultural evolution. The tropical rain forest has long had a reputation for being pristine. Strong evidence suggests, however, that much of th e vegetative cover in the Amazon Basin is actually the result of careful cultivation, rooted in an thropogenic processes. The high biodiversity of the Amazonian Basin places it unquestioningly at the top of a list of the most diverse ecosystems on the planet. However, the scale of “ecosystem” puts researchers at a disadvantage to talk about the reasons why we see the variability of flora and fauna, where the origins of such biodiversit y herald. In keeping with principles of historical ecological theory, the resultant image that we have of the ove rall distribution of so many different types of organisms is actual ly the result of natural disturbances in environmental systems (meaning naturally occurring changes to the environment) being met with human response, and vise versa. Brown and Lugo (1990: 4) estimate that today about forty percent of the trop ical forest in Latin America is secondary as a result of human clearing and that most of the remainder has had some modification despite current low population densities. Many soils of th e lowland humid tropics have long been considered too infertile to support sustainable agriculture. The primary thrust of this argument relies on the assumption that due to the rapid decomposition of organic matter as a result of high temperatures, large amounts of precipitation, and the lack of stabilizing minerals in the humid tropics, sustainable agriculture simply w ould not be possible. Increasingly strong ev idence refutes such re search, instead implying that permanent or


32 semi-permanent agriculture can itself create su stainable fertile soils in a cyclical selfsustaining and a self-renewing. Modifications of human origin have dr amatic impacts on regrowth opportunities for various flora (and, by extension, fauna). For example, species composition in fallow periods of slash-and-burn horticultural te chniques (adopted by many ingenious groups) differ significantly from natural gaps in canopy, altering “species composition of the mature forest on a long-term scale" (Wal schburger and Von Hildebrand 1991: 262). This is not to suggest that human m odification of tropical forests, be it past or present, is limited to a simple “slash and burn” model. Rather, the Amazonian Basin is composed of a sequence of crafted landscapes, operating as a functional system at local levels, and mirrored in its functionality at more regional levels. Large expanses of what are viewed as “natural” forested areas were and are cr eated landscapes conformi ng to a larger design in which the kinds, numbers, and distributions of useful species are managed by human populations. As previously indicated, a v ital mechanism in forest management is manipulation of slash-and-burn fa llow periods in an effort to replenish or attract specific types of vegetation or faunal sp ecies of interest to local popu lations. The effort expended in selection, transplant ation, and protection of specific forms of “wild” vegetation is a premeditated labor that belies the simplicity with which some regard slash-and-burn cultivation, suggesting instead a preconceive d notion of well-manicured environments (Denevan and Padoch 1987). Balee (1987, 1989) has expounded on the “anthropogenic" forests in Amazonia in which numerous species have been carefully manipulated to increase the availability of specific species without adversely affec ting local biodiversity. These include specialized forests (baba ssu, Brazil nuts, lianas, palms, bamboo),


33 composing approximately 12% of the total upl and forest in the Brazilian Amazon (Balee 1989: 14). Balee (1989:14) conc ludes "large portions of Amazonian forests appear to exhibit the continuing effects of past human interference." What has become painfully obvious is that older models attempting to characterize the variation in social organi zation and complexity in the occupation of upland and riverine ecological zones based upon the differentia l perceived productivity of these regions have vastly underestimated th e ability of indigenous populations to fully exploit a given environment. While the agricu ltural potential of soils in the interfluvial regions of the Amazon may not have the same nutrient level, they are, nonetheless, more than sufficient to allow intensive producti on of manioc. This coupled with the high availability of protein in the form of aquatic resources coul d provide more than adequate calories to sustain large populat ions, allowing sedentary grou ps to grow and thrive. Manioc cultivation, even in the poorest soils, has been hypothesized to be perfectly adequate for providing a stable economic foundation for large, settled populations (Carneiro 1987, 1995; Heckenberger 1998; Heck enberger et al. 1999; Petersen et al. 1999.). Anthropogenic soils, known as terra preta do ndio, anthrosols, and terra preta arqueolgica , or Amazonian Dark Earths (ADE)(Smith 1980; Kern and Kampf 1989; Kern et al. 2003; Woods et al. 2003), are highly coveted areas for cultivation (Smith 1980). The presence of terra preta along the edges of the larg e floodplains, as well as in the uplands, has been hypothesized to demarcat e areas of intense cu ltivation (Balee 1989: 10-12; Smith 1980). Terra preta anthrosols are closely asso ciated with flat-tops of escarpments of the well-dr ained Tertiary Plateau ( Terra firme ), where they form patches resulting from ancient middens (waste depos its). These soils contain charcoal and


34 cultural waste from prehistoric burning and settlement with high carbon, nitrogen, calcium, and phosphorus content leading to a specific vegetative signature (Kern and Kampf 1989; Zech et al. 1990; Ke rn and Costa 1997; and Glaser et al. 2000; Woods et al. 2003) and are often associated w ith types of forests that are believed to be the product of long-term management by native Amazonian s (Balee 1989). When several species dominate in terra firme forests, human activity is us ually implicated (Anderson 1983; Balee 1989; Balee and Campbell 1990). In num erous published inventories of ‘‘virgin’’ tropical rain forest in the Lower Amaz on, certain species are more common than expected (Campbell et al. 1986) and inferring ei ther direct human intervention, or, at the very least, significant human-induced impacts on local ecosystems. Most importantly, anthropogenic upland forests that grow on past occupation sites are commonly extensive enough to be detected in remote sensing data due to the distinctive texture and reflectivity of their vegetation. Debate may have once centered on theories of natural origins for terra preta (Smith 1980), it is now indisputable that such deposits are, in fact, human artifacts (Heckenberger et al. 1999; Woods 1995; Woods and McCann 1999). It is highly improbable upland soils well outside the nor mal floodplain environments could ever attain such high concentrations of P, Ca, K, and C without some sort of anthropogenic influence, evidenced by the chemical co mposition (Eidt 1977), spatial distribution (typically on high landforms where depositi onal processes could onl y be cultural), and the presence of artifacts throughout the modified horizon. Sites of terra preta usually occur in areas averaging 20ha (Smith 1980; Zech et al. 1990; McCann et al. 2001), but very large terra preta sites up to 350ha have also b een reported (Smith 1999). Smith


35 (1999) has shown evidence that terra preta is not restricted to varzea , but also occurs on terra fimre regions as well. The similarity of th e texture and mineralogy with that of surrounding soils (Zech et al. 1990), in c onjunction with the occurrence of preColumbian ceramics and other arti facts in the u pper horizons of terra preta soils (Sombroek 1966; Smith 1980) seems to suggest th at the occurrence of this “black earth” is actually an indicator of human activity, and methodical enhancing of poorer soils with fertilizing materials through th e introduction of soil organi c matter (SOM) and nutrients (Sombroek 1966; Smith 1980; Zech et al. 1990). Research published by Zech et al. (1990) and supported by later scholars (M cCann et al. 2001; Woods and McCann 1999) suggests that chemical and bi ological processes work in c onjunction to form deposits of terra preta once a threshold of nutri ent retention capacity an d biological activity is reached as a result of cultural activity (burning, deposition, etc.). The absence of significant ADE deposits in many contemporary Amerindian and mestizo settlements (Heckenberger et al. 1999) w ould suggest that itinerancy, low-density settlement, and traditional slash-and-burn agricultural practices are not the patterns of settlement and land-use that led terra preta production. The ability to enrich localized environments to suit changing cultural demands would indicate that human agency is an incredibly powerful force (not theory, but and actualizing mechanism. The possibility that these anthrosols were intentionally modified to expand livelihood opportunities in previously impoverished landscapes, as suggested by Herrera et al. (1992) and Mora et al. (1991) is an incredibly powerful statement of the impact that humans, historically or prehistorically, had in constr ucting landscapes. An emerging view stresses the ecological praxis of native Amazonians (intensive ag ricultural practices and intentional soil


36 modification), or stated more simply, ‘‘soi ls were a constraint, but people overcame them’’ (Mann 2000: 788). However, we mu st consider the possibility that terra preta may well be a simple byproduct of changes in settlement and live lihood. Selection of settlement locals are certainly intentional, as are the choices of agricultural practices undertaken in a particular environment, and waste treatment and ot her activitie s carried out within settlements. But does the inten tionality of acts leading to the production of enriched soils necessarily mean that th e end result is also intentional? Terra preta has thus generated a great deal of interest am ong the research community in Amazonia owing to a broad recognition of the import of discussi ons of “black earth” formation and its ties to Amazonian cultural history (Whitehead 1998). Research on terra preta has suggested that, even in areas with low agricultural potential, soils could be modified in such a way as to dramatically increase productive potential. To assess intentionality and the degree of intentionality in terra preta formation, it is important first to identify be nefits derived from these soils once formed. Despite widespread acceptance of anthropogenic origins and soil enrichment (Kern et al. 1989; Woods 1995; Woods and McCann 1999), evidence for the importance of these soils in traditional agriculture has been mostly anecdotal. Furthermore, historical factors that played a role in defini ng the cultural importance of terra preta through time, leading to its treatment as an important resource, ha ve yet to be addressed. McCann et al. (2001) discuss indigenous practices of soil inoc ulation with ash, organic material, and microorganisms, and the implications of th ese practices for tropical soil management. Yet linkages between intentional agricultural intensification and terra preta formation, and between terra preta and an increase in human carrying capacity, have been


37 insufficiently discussed. Furthe rmore, while the presence of terra preta continues to provide evidence of relative ly permanent settlement in Amazonia prior to European arrival (Denevan 1996; Smith 1980), the suggesti on that these richer pockets of soil made possible this transition to a more sedentar y lifestyle is as ye t unproven, but a highly seductive hypothesis. If Pe terson et al. (1999) are correct in their assumption that terra preta was the result of human occupation rather than the genesis of it, then certainly it would have served as an important incentiv e for continuous occupa tion. Additionally, if, as has been suggested, (Peterson et al. 1999) terra preta was, in fact, th e direct result of intensive occupation, then th is would have created something analogous to a living organism (self-perpetuating and self-sustaini ng as accumulations of human waste acted as a vector of renewal). Contributions of recent historical ecologi cal literature indica te indigenous groups of the Amazon Basin may have actively modifi ed the environments in which they live (Balee 1989, 1998; Posey 1985, 1998). Mo st literature in this ve in refutes the concept of human adaptation to environment, but the va rious researchers place different levels of emphasis on the environment in this dialecti c. Whitehead (1998) takes perhaps the strongest stance against environmental dete rminism, proposing a fully ‘‘historical’’ ecology, wherein human agency provides an independent variable in the equation of environmental dynamics maximizing the role of human decision-making (“ecological praxis”) in constructing the la ndscape. This is not meant to remove outside pressures, however. Localized cultural behavior practic es are greatly influenced by cultural and environmental contexts, and broader cultu ral processes (regional interactions), environmental impacts of past cultural activity, and unstructu red or unmodified


38 environments themselves act to affect change and effect decision-ma king at the local and individual level. Politicaleconomic influences can have a massive impact in subsuming human agency in localized environments. A dditionally, past actions of local populations (landscape degradation, soil enrichment , and other anthropogenic impacts on the environment) affect the resources available to future occupants. Within a historical context, the very nature of a changing envi ronment and the varied properties of any given setting will undeniably interact with cultural processes to in fluence the range outcomes. Outdated models emphasizing uniform infertility of soils across the broad landscape of Amazon are patently false. The view that terra firme adaptations were ecologically restrictive, dicta ting the absence (or at least th e near absence) of sedentism simply reifies the “Tropical Forest Tribe” m odel (now some 60 years old). The Amazon, however, has been revealed as an extrem ely diverse environmental complex with a multiplicity of ecosystems. The original m odel, with its archaic belief in homogenous ecozones, has been proven insufficient to describe the diversity revealed through ethnographic study (Morgan Schmidt, a PhD ca ndidate, Department of Geography, UF, is currently conducting pedoarchaeological studies related to the Upper Xingu project and a detailed discussion of ADE in th e Upper Xingu is not given here). Those that seek to make a distinct ion between the fertile soils of the varzea and the comparative infertility of soils in upland regions have also fallen by the wayside, suffering from generalizing principles when ev idence suggests that simple divisions into varzea and terra firme cannot encapsulate the diversity of the Amazonian ecosystem. Numerous floodplains soils along “black wa ter” or “clear water” rivers have demonstrated soil characteristics even poor er than many of the “impoverished” upland


39 soils (Peterson et al. 1999: 11) . It is important to note th at a number of this model’s proponents have acknowledged the likelihood of both widespread sedentary populations and social complexity in both varzea settings and in bluff regions along the margins of upland and riverine areas (Carneiro 1986, 1995; Lathrap 1970; Meggers 1996; Roosevelt 1980, 1994), but the idea that cultures exhibiti ng these traits were rest ricted to these areas is simply untenable. The over-emphasis of the varzea as the sole area of agri cultural potential has been endemic in much of the though governing the po ssibility of resource potential si gnificant enough to provide a basis for sedentary populatio ns and social complexity. In regions like the lower Negro and Upper Xingu, sedent ary villages were supported by intensive terra firme agriculture (Denevan 1992; Heckenberg er 1998; Heckenberg er et al. 1999; Peterson et al.1999). Additionally, the vast potential of aquatic resources, both in floodplain and upland areas would suggest that people would have been “pulled” to all riverine settings, not just those located in varzea floodplains (Carneiro 1995; Heckenberger 1998; Peterson et al. 1999). Even allowing for strict adherence to ecologically determinist principles g overning the “carrying capacity” of the terra firme , certainly the combination of manioc agri culture and aquatic re sources could have provided a substantial subsiste nce base allowing for much la rger prehistoric populations that what we have observed in the ethnogra phic present. If we couple that with the evidence that past populations were actively transforming their landscape into a hugely productive agricultural zone through intensive occupation (producing the terra preta soils we find today), then we have a powerful model describing how Amerindian populations were not only selec tive in their use of the la ndscape (Heckenberger 1999),


40 but rather than degrade ar eas through intensive occupa tion, soils could be improved through both intentional and uni ntentional modifications. An inescapable conclusion is that we can no longer view prehis toric settlement of Amerindian populations as small, impermanen t, or autonomous groups barely scraping by an existence within a hom ogenous landscape. Nor can we allow the projection of the ethnographic present (replete with the dise nfranchised, displaced, decimated, vastly impacted results of European contact and expa nsion) onto the past. We must also avoid the trap of strict adherence to old theories of environmental determinism. Certainly local ecology differentially constrains cultural development, but the relationship between culture and environment is a dynamic one. Cultu re can be molded to fit an environmental setting, and environment can be modified to better suit a particular cultural need. The central questions we must ask ourselves in adopting these deterministic perspectives are can we ever hope to adequately understand th e full scope of variabili ty of the Amazonian ecosystem as a whole, or be able to characte rize the range of cultural adjustments to it? Can we ever hope to separate constraints im posed by the environment and contrast them to those imposed internally? The view promot ed here demands that any statements of the casual nature of environmental factors on cultural change that do not address the reciprocal nature of the relationship should be greeted with skepticism. Importance of the Upper Xingu The Upper Xingu provides a unique example of an Amazonian cultural lifeway that supported large, densely settled, and integrat ed regional populations over the past 1000 years (Heckenberger et al. 2003). Archaeol ogical evidence has show n (Heckenberger et al. 2003) evidence of large, well-engineered publ ic works (such as plazas, roads, moats, and bridges) in and between pr e-Columbian settlements.


41 This would seem to suggest a highly modi fied environment, having much more in common with other regional-scale complex pr ehistoric societies found elsewhere in the Americas than with the commonly held misbe lieve of small-scale, localized populations constrained by their environment, and rele gated to the lower echelons of cultural complexity. But can the archaeological eviden ce alone speak to the questions of to what extent the environments were modified, and if there was a pattern to the modification? In short, yes, but only with great difficulty. The nature of the region is such that archaeological research is time consuming. Site density, even within the relatively small geographic confines of the Upper Xingu archeological work conducted by Mike Heckenberger since 1996, seems substantial enough to place th e feasibility of garnering either enough time or money to fully survey the entirety of th at area, let alone the extents of the PIX, utilizing traditional survey techniques far beyond our reach. And yet the import of discovering that largescale regional complex societies did, in fact, exist in this area, and the possibility of dr amatically reversing years of researchers relegating tropical forest regions, such as the Upper Xingu, to so mething akin to a “cult ural backwater” begs further investigation. The questi on is, how can we approach studi es of this area in such a way to mitigate both the time and expense of traditional survey? Through the integration of new technologies, we can provide a mean s to augment current research, including archaeological and ethnographi c fieldwork, with remote-sensed data analysis, GPS surveying, unified in a geographic information systems database, in an attempt to further illuminate the interplay of human populati ons and their environment in non-western tropical forest settings.


42 CHAPTER 3 PREDICTIVE MODELING AND GIS Any sort of in-depth study seeking to address research question in Amazonia utilizing tradition field techniques will, at some point, become hampered by issues of accessibility and other logistical problems. The Upper Xingu is no exception, as the area is indeed difficult to access physically and acad emically, with little data available for the region. This makes the use of new technologi es, including GPS, GIS, and remote-sensed data (especially satellite imagery) even more critical to a successful exploration of some of the issues mentioned in the previous chapter. Each of these tools can augment traditional archaeological exploration, w ith the additive effect producing a robust database of information, leading to more eff ective models capable of moving us closer to answering questions of long-term human-envi ronment interaction, landscape change, and cultural development in the region. This di ssertation sought to fo rmulate an inductive predictive model (not as an explanatory devi ce, but rather as a he uristic model) using vegetative signatures of known ar chaeological sites to extrapol ate out to a wider scale of possible site locations within a specific region of the Xingu. Defining the Model Predictive Modeling and Archaeology An archaeological predictive model is a tool that indicates the relative probability of encountering an archaeological site. Pa rker (1985) sees predictive modeling as a natural outgrowth of the theories and methodol ogies of spatial archaeology and predictive modeling has become the focus of a number of archaeological studies (e.g., Allen et. al.


43 1990; Judge and Sebastien 1988; Kvamme 1992) . Of particular relevance here, predictive modeling is an avenue of res earch within archaeology that has gained prominence over the past two decades, specifically with the development of new technologies of remote-sensing, mapping, and GI S. Predictive modeling for archaeology is defined as a “... simplified set of test able hypotheses, based either on behavioral assumptions or on empirical correlations, which at a minimum attempts to predict the loci of past human activities resu lting in the deposition of ar tifacts or alteration of the landscape” (Kohler 1988:33). To implement a predictive model, a set of georeferenced parameters (factors) that are related to site occurrence is collected, stored, and manipulate d. Thus, the underlying theoretical basis for modeling site locati ons is the hypothesis that environmental attributes can be correlated with archaeol ogical site locations. Known archaeological evidence has to be acquired and classified into chronologically and typologically homogeneous groups. By confr onting landscape and archaeologi cal data sets, a heuristic or statistical model is then built up which links the spatial variability of such parameters with the occurrence of the sites (Jacoli a nd Carrara 1996). The integration of GIS, remote-sensing techniques, and GPS surv ey may provide new opportunities for identifying, analyzing, and in terpreting archaeol ogical sites makes it possible to both define the relations of known sites with the en vironmental context, and identify new sites. The theoretical and applied aspects of conducting predictive modeling in conjunction with archaeological studies are re latively new. They have th eir basis in studies conducted during the 1950s and 1960s, but remained th e bailiwick of relatively few until they gained prominence during the late 1970s, 1980s , and early 1990s (coinciding with a surge


44 in cultural resource management in the Un ited States). The literature concerning predictive modeling increased e xponentially during that period of time (thus, much of my discussion is framed by the character of the lit erature of that period). And while, in the past, many archaeologists steadfastly refuse to accept the possible value of predictive models in determining possible site loca tions (Kohler and Parker 1986:396), with predictive models were viewed by many as an expensive exercise to discover the obvious, and regarded as suspect or unreliab le or being limited in value (Kohler and Parker 1986:398), it is importa nt to remember that archaeo logists, on the whole, have moved beyond these myopic viewpoints. Most researchers would now acknowledge the utility of predictive models as an impor tant element of contemporary archaeological investigations, with even great er potential in the future, part icularly in little know and/or remote regions. In point of fact, most cu rrent modeling efforts have begun to develop entirely new methodological approaches in to the modeling process, recognizing the importance of the theoretical contributions of landscape archaeol ogy, historical ecology, and advances in thought rega rding settlement patterning (C rumley and Marquardt 1990; Crumley 1994; Lookabill 1998; Witcher 1999; Chuchill 2000; Perkins 2000; Wise 2000; Wheatly and Gillings 2002) The strength of predictive modeling is in providing a unified framework that includes testing and self-corre ction components, and contains an element that goes far beyond mere number crunching to arrive at an accurate depiction of the past. The mathematical “purity” of predictive modeling effo rts (a result of at least two generations of archaeologists’ processual, positivist approaches) is one of the greatest strengths of current modeling programs, allowing for a replicability of the model, and allowing for


45 “tweaking” of a model by alte ring simple (or sometimes complex) variables. However, strict adherence to only those variables that can be enumerated has led to an exclusion of those elements of culture that make it so uniqu e. The archaeological past is composed of tens of thousands or more unique cultural gr oups: different times, different peoples, and different places. Each culture has its ow n unique history, its ow n chronology of events, and it is this uniqueness that gets lost in the midst of current predictive modeling efforts. Incorporating culture history into modeling efforts can breathe “life” into the model, adding new dimensions and increasing the ability of the model to accurately predict, or at least approximate, cer tain behaviors. Models are partial representations of a th eory, formulated in a manner that enables the archaeologists to test the theory by means of empirical data. When new patterns are found, hypotheses are formed to explain them, and models are built to test the hypotheses (Warren 1990). While there are many approaches to predictive modeling, all must choose between various kinds of units of anal ysis, dependent and independent variables, types of models and decision rules, and mode ling testing procedures. The effective and efficient application of the predictive archaeo logical models of the past was severely hampered by the labor required to manually measure map variables in large-scale projects. In fact, past app lications of high-resolution models were virtually impossible without restricting sample size and the range of variables investigat ed. Nearly all of these limitations have been successfully ove rcome in recent years, however, through the application of GIS. There are two traditional ways to go about developing models of where prehistoric peoples located themselves. One approach lies in a close examination of the


46 anthropological and historical l iterature of a region in an effort to deduce the kinds of locations that past peoples may have select ed to place their camps and settlements. Researchers must determine relevant variables an d then arrive at specific values for each of those variables (a difficult task even in th e present, infinitely more so when dealing with materials from the more distant past). For example, we can assume that nearly all peoples would build their residences on le vel ground, but exactly how level does the ground have to be? A permanent settlement w ould almost certainly need to be located near a secure water source, but how near, how secure, and of what type? Furthermore, other variables may come into play, such as defense, that make a narrow focus on single or even few variables unrealistic. Like wi se, there are always exceptions to the rule (some people do live on slopes, over water, or position their settlements between, not near, water sources). The degree of freedom in selection of these variables, and the error tolerances inherent in the establishment of any fixed value for any one variable, could result in outcomes with widely varying results. In the face of such issues, the task of accu rately describing the past seems daunting. However, utilizing advances in both theory and methodology, it may be possible to gain insight into those environmental and cultura l variables deemed of critical importance by the actual groups under study, thus revealing a more robust and more useful form of predictive modeling. First, any modeling e ffort should incorporate some element of archaeological field survey. By measuring and analyzing environmental variables at known archaeological sites, it is conceivable to ascertain which variables may have been critical in deciding on past si te location with some degree of success. By coupling this process with extensive use of available ethnog raphic materials and analogy, particularly


47 within specific historic trajectories, one can more accurately determine which features may have been the most important, eliminating those that have little or no bearing for the culture concerned, simplifying th e variables, and creating a more elegant model. The reason such studies have not historically been popular is that the amount of labor involved often makes them impractical. Ho wever, the advent of GIS has permitted virtually any sort of map variable to be computer-encoded, and combined with other variables to yield complex modeling outcomes over larger areas with more precision, better error correction capabilities, less labo r, and the ability to easily model several different iterations of element values. Data updates and corrections, once slow and costly, can be conducted quickly and effici ently. Using this technology, measurements can be completed in a matter of seconds. The only limits to constructing complex simulations and models of be havior are limited only by imagination, as the time required for processing data sets has been relega ted to insignificance (Kvamme 1989). Many archaeologists of the period recogni zed the value of GIS-based analyses, especially in the realm of archaeological pred ictive modeling. For the most part, archaeologists would agree that the spatial di stribution of sites is largely dependent on a wide spectrum of features (suc h as landform, soil type, wate r proximity, vegetation cover, climatic conditions, etc.) that characterize the environmental context where sites are located. Throughout the later 1980s and 1990s, predictive models were increasingly applied in investigations attempting to bot h explain the spatial distribution of sites already known, and predict where new site s are most likely to occur (Kvamme 1989; Warren 1990). Still, it is cr itical to recognize the limitations of a GIS, and of predictive modeling within a GIS environment. “Predic tive models are probabi lity statements; they


48 are not “facts,” and cannot subst itute for facts in any applicat ion requiring the use of hard data about specific individua ls as decision making criter ia” (Wildesen 1974 in Kohler 1988). Judicious use of GIS can give practical inte gration of spatial st ructures (habitations, soils, river drainage), but to gain a real understanding of pa st and current relationships among environmental and human systems require s that archaeologists research culturally specific temporal and spatia l perspectives. Much curre nt research in GIS and archaeology revolves around ways to incorpor ate landscape perspectives as well as spatial archaeology. The development of more methodological approaches to the integration of GIS requires a fundamental understanding of how a GIS enables the creation, organization, and management of mu ltidimensional archaeological data sets. Vector GIS is essentially a simple relati onal database wherein records describe the attributes of a real world geographic entity that is th en linked to a geographically referenced digital representation of vect or information (point , line, or polygon) (Maschner 1996). A GIS commonly incorporat es several different types of geographic entities. Similar mapped geogr aphic entities are generally or ganized and stored in their own georeferenced layers. When georeferenced layers are placed in the same projected coordinate system, the layers can be overlai n, allowing the researcher to investigate the relationships and patterns between phenomena by examini ng and analyzing alternate combinations of layers, the response to a query, or the results of an analysis. All archaeological data has a spatial asp ect; individual artifact s have provenience, features are located within a site, sites are part of a mosaic of ha bitation and activity areas, travel routes and borders, and a ll are located in, and affected by, diverse


49 environments. How one manages these da ta can have a dramatic impact upon the efficiency and practicality of attaining resear ch goals. Two benefits of using GIS for archaeological data management are the ra pid creation of maps to visualize the archaeological record and the ability to query the database using spatial parameters. The scale of the GIS can range from individua l excavation units to regional or national registries. More than simply a two-dimensional ma p design program, the real power of GIS lies in its modeling abilities. The gene ral mapping capabilities of GIS are a basic function, and although it is generally more e fficient than manual or computer assisted means, it only begins to address the potential of GIS for archaeological data management. These visual representations of archaeological data in a georeferenced map can be used in to produce archaeological "sensi tivity" maps that indicate which geographical locations are more sensitive than others for cultural resources. The dependability of such predictability models is a func tion of their performance. This can be examined and tested by comparing the results of a predictive model to archaeological field survey results. By comparing model predictions against known arch aeological site locations, it is possible to determine, with specifiable confidence, how ac curately a model performs. In fact, this very approach gives us confidence in a model a nd allows us to use it as a predictive tool. Differing Approaches to the Modeling Problem Prior research and settlement theory de monstrated that open-air site placements were most often a function of a matrix of environmental fact ors that have been found to be quite consistent from study to stud y (e.g., Jochim 1976; Roper 1979; Shermer and Tiffany 1985). As a rule, the vari ables chosen are restricted to those that reflect relatively stable landform characteristics through time, such as elevation, slope, and aspect, to


50 insure that there is some correspondence be tween modern map-measured data and the prehistoric-early historic envi ronment. Some potentially important variables, such as plant community composition and water table elevation, are notoriously sensitive to climatic changes and, as a result, are difficu lt to use without recourse to proxy measures (Kohler and Parker 1986: 415). Since the particular environmental variab les most suited to a particular model depend in part on the physical nature of the region under investigation and cannot be determined completely a priori (that is, without analysis ), most modeling projects initially measure a relatively large number of landform, hydrological, soil, and geologic characteristics, including slope, aspect, elev ation, local relief, la ndform type, horizontal distance to the nearest permanent water and st ream confluence, and distance to streams. Characteristics with low predictive po wer are filtered out throughout the model development process. In larg e areas, some characteristics may prove to be important in only some sub-areas, necessitating the construc tion of multiple models. Justification for the adoption of specific characteristics a nd procedures for operationalizing their measurement can be found in Hasenstab ( 1990), Kvamme (1986), Kvamme and Kohler (1988), Parker (1985, 1986), and Roper (1979), am ong other authors. General reviews of the issues involved are provided by Judge a nd Sebastian (1988), Kohler and Parker (1986), and Kvamme (1990). Kvamme likens inductive pred ictive models to the supe rvised classification of remote sensed imagery: (1) training sa mples representing the locations of known archaeological classes of interest (e.g. classe s representing different types of prehistoric activity, time periods, cultural types, or amount s of activity) are established through on-


51 the-ground field survey; (2) a statistical or othe r classifier is devel oped for these classes, including an archaeologically “empty” class, based on patterns exhibited by the classes on the GIS encoded environmental data; (3) th e classification rules are applied on a cellby-cell bases by the GIS to classify the entire region of study, where each cell is assigned to the archaeological class to which its envir onmental characteristics are most similar, or the empty class; and (4) the accuracy of the cl assification is assessed through test samples of known class membership. The classification of the region resulting from the approach constitutes a location model of the region’s archaeological phenomena as defined by the sample data (Kvamme 1989). Although this "empirical correlation" pro cedure was, by necessity, used in the formation of earlier predictive locational mode ls, the importance of social and political factors in the spatial location of settlements have been recognized and incorporated into more recent efforts (Crumley and Marquardt 1990, Madry and Crumpley 1990, Madry 1996), and, arguably, have been in place sin ce the advent of central-place theory (Christaller 1966) and earlier gravity models . As a result, cons ideration of land-use choice derived from “habitual behavior” de rived from cultural norms, traditions and spiritual proscriptions became the norm in de veloping models, rather than an overriding consideration of the economic attractiveness of a specific lo cality (Kohler and Parker 1986:435 citing Wright and Dirks 1983). Fact ors related to actions having little archaeological visibility, such as spiritual influences, may have resulted in activities being located in less '”typical” locations. C hoice of activity location may also be the result of historical events th at override environmental consid erations. This is where the practice of forming predictive models beco mes less science and more of an art.


52 Researching historical events within a cu lture, becoming intimate with the ethnographic material, and incorporating it into the vari able-forming process is perhaps the most difficult part of model building. This require s extensive data, and even that may not be enough to fully encapsulate even a small range of behaviors. Despite the conundrum of establishing ties to the cultural history of a site, some archaeologists have made valiant attempts to bring the ethnographic past into their modeling efforts. Flannery (1976) and Re ynolds (1976) discuss social factors that condition site placement. Jochim (1976:12) de tails criteria of economic relevance and assumes that "the determination of resource use tends to precede and condition the site placements and demographic arrangements of a hunter-gatherer group.” There are as many procedur es developed for quantit atively determining the correlation between environment and site pres ence as there are researchers working on such problems (see Warren 1990, Kohler and Parker 1986, Carr 1985 for some of the more highly utilized methodologies) Of par ticular interest are logistic regression (numerical approach), weighted value appr oaches, and Dempster-Shafer theory, also known as weight-of-evidence modeling. Logis tic regression’s established use, broad literature, and numerous case makes it partic ularly well suited for predictive modeling endeavors. Predictive models utilizing a numerical approach employ multivariate statistics as a means of identifying associa tions among variables. Because statistical methodology discovers meaningful associa tions among variables from known site information, it is important that the known site information be representative of the site population. The most intricate and complex proba bilistic designs are of little use if the population sample is not the same as the ta rget population. It is then beholden upon


53 archaeologists employing the numerical approach to ensure the validity of their assertions by carefully evaluating the nature of the existing database. The data must be representative, and, perhaps more importantl y, the researcher shoul d determine whether known site locations reflect the actual distri bution of archaeologica l sites, or simply reflect where archaeologists ha ve conducted their surveys. It is also critical for the archaeologist to recognize that the physical and cultural environment has changed over time (a careful co nsideration of a diachronic perspective). Too often, we are lulled into a synchronic view of the environment. To the contrary, the environment is dynamic, and it is only with an eye towards a diachronic study that we can even begin to comprehend the past (changes in environment may well have affected the choice of activity location through tim e). Kohler and Parker state that: “... Despite numerous studies in diverse areas indicating change in site location through time in response to changes in adaptation type, and despite evidence that within any adaptation type, functional subsets of sites may have differing environmental determinants, most empirical correlative models aggregate sites of all types and ages together for prediction" (1986:408). This remains a chief problem with numerically derived models. Often times these models show a lackadaisical treatment of, or completely fail to address temporal considerations. There are some researchers th at have chosen to avoid the issue of time altogether, and develop generalized models that, by their very nature, are extremely limited in their applicability (and perhaps not of any use at all). Other researchers have sidestepped temporal considerations by sugge sting that discernibl e patterns of human behavior crosscut considerati ons of time (Kvamme 1992:23). The numerical approach certainly retains value as an approach to predictive modeling, allowing the researcher to uncover associations between site locations and variables. The drawback to this approach, however, is that it requires a high degree of


54 statistical training and competence in order to develop the model, interpret the results, and replicate the outcome. Invariably, a great deal of interpretation is required to relate the results to real-wor ld geography. More often than not , the generalizing nature of the models make them unsuitable for addressing i ssue of localized cultural practice (thus of little use to researchers interested in speci fic peoples at specific times). While the numerical approach can and is still used to generate valid predictive models, many archaeologists have turned to different approaches to he lp eliminate some of these complications. The weighted value approach, or “graphica l approach,” details the development of predictive models utilizing environmental vari ables, but distinguishes itself by using a graphical methodology derived from map overlay techniques. Advances in GIS applications have allowed researchers gather data from different va riables represented on separate computer map layers . These map layers can be combined in ways that can readily identify areas spatially associated w ith landscape characteristics determined to be pertinent to the questions being asked by the re searcher. Utilizing different combinations of variables, the researcher is able to address various st ages in the predictive modeling process, fine-tuning the outcome of the model. Database queries of overlain data sets allow researchers to extract those geographica l regions that contain desired associations between variables and sites, which can then be evaluated using stat istical techniques. Statistical techniques are no l onger used as the primary mean s of discovering sites, but rather as a method of evaluating the strength of the association between variables and sites after the model is applied.


55 Within these parameters, models either are physically generated by an intersection or weighted value method. The intersecti on method begins with the basic assumption that all variables used in th e generation of a predictive model contribute equally to the determination of site location potential. Ca lculating high, medium, low potential areas is simply a process of determining where the gr eatest number of variable s that converge in a given location. The weighted value method be gins with the basic assumption that each variable contributes differently to the final determination of site location potential. This is accomplished by developing and applying a weighting scale that effectively ranks variables numerically. Site potential is determined by th e arithmetic addition of all variables. Areas of high potential will have the largest numeric values and areas of low potential will have smallest numeric values. During the development of numerical predictive models, a number of issues mu st be considered. These include the “representativeness” of the variables to the behaviors being m odeled, the quality of databases consulted, the scale at which m odeled should take place, and the manner in which potential is presented. Weight-of-evidence modeling, perhaps, holds great potential for further investigation, as it is perhap s better suited to the uncertainties associated with archaeological data and environments that ca nnot be directly observed. Dempster-Shafer theory is a “soft” decision support tool that works extremely well w ith Fuzzy Set theory or Bayesian statistics. It allows for and deals with incomplete datasets, expert knowledge, anecdotal evidence, experience, and ignorance. Each line of evidence used in the model (like the proximity to fresh wate r resources) supports one of three hypotheses: (1) that sites are located near water (site), (2) that sites are not located near water


56 (nonsite), or (3) that we are uncertain about which hypothesis the evidence supports (site/nonsite). Each line of evidence is weighted and a cumulative probability map generated. Conversely, logistic regression analyses can be considered a “hard” decision tool because it does not account for fuzzy datase ts or uncertainty; it is assumed datasets are ‘perfect.’ Although this ca n be seen as a detriment, the statistical robustness of logistic regression makes it a more clean-cut predictive tool. L ogistic regression has advantages over other statistical methods such as discriminate function analysis; (1) it has less restrictive assumptions; (2) tends to be a more powerful and consistent; (3) accepts a mix of scales of data (nominal, ordina l, interval and ratio) (Warren 1990). Both univariate and multivariate statistical models are commonly used to identify variables on which the distri butional differences of depe ndent variables (resources present/absent) are most pronounced. A variet y of statistical tests are performed on the univariate descriptive statistics to elicit trends , and a logistic regres sion technique is most often used to explore multivariate differences. Many researchers have adopted multiple logistic regression models for analysis. Th e rationales are varied, but primarily center on the fundamental nature of such models in that they make no assumptions about the distribution of the data (representing a nonpara metric technique), are robust classifiers regardless of distributional form (an essent ial attribute in environmentally diverse regions), and can handle nominal, ordinal, a nd interval level independent variables. Although multiple discriminant analysis, ma ximum distance classifiers, quadratic classification procedures, and maximum likelih ood distance classificat ion techniques all have their adherents (e.g., Bradley et al . 1986; Custer et al. 1986; Kvamme 1983, 1985,


57 1986; Parker 1985; 1986; Warren 1990; see Kohl er and Parker 1986, and Kvamme 1988; and 1990, for general reviews). The findings of these two different modeling st rategies are used, at least initially, to formulate decision rules (in es sence, describing weights of environmental variables that indicate whether archaeological resources are li kely present or not). These decision rules can then be applied to any unsurveyed area to determine whether the model specifies that a site will be present. Since a fundamental issue in locational modeling is in determining what weight or rank should be applied to each of the independent variables investigated, a decision point will be select ed by calibrating the decision rule at first to sample data (Kvamme 1988). Presumably, the application of decision rules captures a pattern left behind in the archaeological record. In archaeology, a loca tional model is essentially a decision rule that assigns specific areas to ce rtain classes based on environmental or other non-archaeological char acteristics of the area under que stion. The model predicts archaeological sites when it assigns value to an area of unknown archaeological class membership representing either a presence or absence or archaeo logical components. This pattern recognition a nd classification methodology f acilitates exploration by abstracting the environmental patterns exhi bited by surveyed regions that contain archaeological resources and mapping them across unsurveyed regions through GIS. GIS is notorious for the amount of initial processing that is required unless the researcher is fortunate enough to acquire existing digital data (although it may be a wolf in sheep’s clothing with uncertain data qual ity, undesired projections and compatibility issues make using pre-existing digital data troublesome sometimes). At issue is the fundamental truism that raw ar chaeological data is can be ex tremely hard to come by in


58 many areas, much less data sets in a digital fo rmat. Analog maps have to be scanned or digitized, contour lines need to be generate d to interpolate a digital elevation model (DEM) (Kvamme 1990), and all layers need to be georeferenced so they share the same projection and datum. Even data in digital form almost alwa ys needs to be processed in some way, either to make it compatible with a specific GIS format, or to classify or otherwise restructure the data itself. Once all the data is in the GIS, formatted, georeferenced, and processed the act ual analysis can be conducted. Model Testing Simply stated, an archaeological predictive model is simply a set of decision rules that assign areas in a study to one of a num ber of mutually exclus ive categories based on environmental or other non-archaeological ch aracteristics of the locations. Assessment of model performance and accura cy are obviously necessary, for, at the very least, a predictive model must be able to perfor m better than random chance alone. Model testing involves the determination of the a priori or chance probability of the occurrence of certain archaeological even ts and an independent test of the model’s effectiveness against this probability. Presumably, the identification of the key non-archaeological characteristics of particular lo cations that are associated with the presence or absence of an archaeological resource is a guarantee that the predictive model will be more effective than the random model, but it must be demons trated that this is so. In addition, a good test will specify the degree of effectivene ss of the predictive ar chaeological model over the random-chance model. An a priori probability is the probability that a random geographical location does or does not contai n archaeological resources. As a randomchance locational model, it provid es a baseline that helps defi ne what other models must accomplish. In regional studies, random-chan ce models can be calculated by determining


59 the relative frequency of the presence or absence of resources in a random sample of surveyed areas (e.g., Kvamme 1983, 1988; Pa rker 1985:187). For instance, if 100 surveyed locations contain resources and 900 do not in a surveyed sample of 1000, then the probability that a given location contains resources by chance is 0.1 or 0.9 that it does not. Since the probability of correctly attri buting a location positive for resources is no better than chance, these probabilities can be considered random-chance. The predictive power of a model is determ ined by calculating its percent of correct predictions in the test sample and comparing this percent wi th the likelihood of a correct prediction by chance alone. These calculatio ns determine the model's specific percent of predictive accuracy over chance alone. The main method of assessing model performance in archaeology is usually some fo rm of cross-tabulation that compares the actual and model assigned presence or absenc e of resources. One of a number of statistical tests can then be used to determ ine the significance of these frequencies (e.g., Congalton et al. 1983; Kvamme 1988, 1990). Si nce the performance or accuracy of the model is evaluated statisticall y, field data must be collected within a sampling framework that utilizes the basic princi ples of elementary probability theory. This foundation also gives a model the ability to assign a proba bility to the occurrence of archaeological resources in a given area. The effective and efficient application of early predictive archaeological models was crippled by the sheer volume of labor re quired to manually measure map variables (Judge and Sebastian 1988). For all practical purposes, the applica tion of high-resolution models was impossible without severely rest ricting sample size and/or the range of variables investigated. The advent of GIS, however, has allowed modelers to overcome


60 these constraints by automating much of the entire process (e.g., Kvamme 1986, 1989, 1990). Unlike traditional database manageme nt systems, GIS has a spatial or mappable component that allows the capture, effici ent manipulation, analysis, and storage of geographical information. In addition, GIS is easily capable of producing maps of this information in various formats on a video monitor or on paper, and, because the information is coded electronically, it is very easy to update and improve models in relatively short timeframes. Refining the Model Criticisms of GIS and Predictive Models There have been a number of criticisms of archaeo logically-based predictive modeling efforts utilizing GIS as the primary analysis tool. As mentioned before, the power of GIS lies in the visual display of vol umes of information, and the ability to make such databases infinitely more accessible. However, the graphical power of GIS can create its own compounded conundrum of issues . Models generated in GIS environments can easily become just another “pretty pict ure,” completely bereft of theory or explanatory power. The main mechanism fo r many inductive locati onal models is the statistical analysis that determines which environmental variable(s) are indeed influencing site selection, or at the least, correlate with site presence. Thus, many predictive models generated using GIS reek of an environmental deterministic approach, and, even worse, seem static and do not ade quately reflect the dynamic adaptability of humans, or the dynamic nature of the natural environment. This sp eaks to the accuracy of such derived modeling approaches. Dete rmining the accuracy of a model involves comparing virtual indicators to actual circ umstances. Ideally, development of a model should be from a random archaeologi cal site sample so that i nherent biases are removed.


61 Sample-based modeling approaches in ar chaeology face ubiquitous problems that limit the predictive accuracy of models and should be considered when using modeling results. A key problem is the accessibility of archaeological resour ces. In any region, many archaeological sites will have been dest royed by erosion or human activities. Other sites will be deeply buried, well hidden in sealed rock shel ters or under dense vegetation, covered by towns or lakes, on property to which access is denied or so small as to easily fall between transects in a fiel d survey. Predictive models, because they are necessarily based on sample survey data, are only sensi tive to the types of archaeological resources included in the initial samples. This means that they are usually only sensitive to certain types of surface distributions, for the distribut ion of geologically buried sites is rarely explored in a systematic manner. A second problem is the difficulty in archaeology of satisfying statistical assumptions, such as the requirement of multivariate normality or homogeneity of variance. For this reason, m odelers usually remain somewhat skeptical of statistical indicators of the importance of independent variables in the developmental phase of model building and employ robust mathematical pr ocedures to identify decision rules. It is in the better-controlled testing phase where the requirements of statistical assumptions can be more fully met that statistical infere nce and probability theory play their primary role. A third problem is the presen ce of patterned variation (i.e ., spatial autocorrelation) in the distribution of archaeological phenomena. Its existence violates the assumption of independent observations and generally result s in overestimates of the significance of


62 independent variables. This problem can be partially controlled by adopting a sampling procedure that widely separa tes surveyed parcels of land. Many other kinds of problems exist. The im portance in site location of social and political factors ("sociocultura l noise"). The difficulty of co nsidering such factors is one reason why nearly all predictive archaeologi cal models have accura cy rates less than 85 percent. It is also the main reason why field survey must remain an integral component of cultural resource management. The most common critique of archaeological predictive modeling is that it is often not grounded in anthropological or archaeologica l theory. Predictive models have been primarily descriptive, for the most part c onfirming what we already know. Trends in modeling efforts have led to simple aggrega tions of sites and the flattening of the temporal dimension resulting in models that adequately predict the average known site, but fail to address the maximums of variability. In a standard distribution of sites, a large proportion of sites will fall within a middle-ra nge, which we can address as “type-sites,” or, more to the point, “common.” However, it is the uncommon, or non-type sites from which archaeologists may be able to glean th e most information. Because a large number of current correlational models are based on observed archaeological sites, they will inevitably be biased towards predictions of sites about which information already exists rather than for “outliers,” which may have a greater potential for new information. By associating sites representing many different functional, chro nological, and cultural types into a single open-air class, a gr eat deal of locationa l, or cultural, vari ability is introduced to the modeling problem. Nevertheless, in past literature, some researchers asserted that


63 there were common locational tendencies that ma y crosscut functional categories, such as preferences for level ground or pr oximity to water (Kvamme 1985). Predictive modeling has become critical as a means of identifying landscape variables that are consistently correlated with known site distributions. By identifying these correlates, researchers are better able to identifying uninvestigated localities that have a high probability of containing site s based upon their geographic similarity to known sites. There is a dange r in this, however. Simply identifying new site locations based upon the attributes of known site locatio ns is not really making progress in the investigation of the unknown. Instead, we are simply investigating more of the same sorts of sites, with the added bonus of identif ying areas that have not yet been surveyed (we are merely modeling existing assumptions and expectations). This is not to say that such undertakings are any less important than other pursuits, but I submit that we, as archaeologists, can take it one step farther. For us to make predictions of the unknown, we must step outside what is “expected” and employ a modeling rationale that does not build exclusivity into its results. Thus, the underlying flaw in correlation models is exposed. Such models are exceedingly good at illustration the probable location of any number of like sites based upon an approximate “type,” but without a more substantial theoretical foundation, they cannot be expect ed to produce information governing “why” or “how” such sites came to be. The importance of infusing some measur e of anthropological or archaeological theory into modeling efforts remains a critical goal. Butzer states, “When the intellectual framework is too narrow, the results, no matter how elaborat ely programmed, cannot hope to allow high level interpretations” (B utzer 1978). And Ebert states “predictive


64 modeling will be transformed into a wo rthwhile adjunct to archaeology and archaeological thinking only by the formula tion of a body of explanatory propositions linking contemporary correlations with the pa st. In other words, it is productive, explanatory thought, and not com puters, that can potentially raise predictive modeling above an anecdotal level (Ebert 2000). Our ability to generate predictive models rests on two fundamental assumptions: First, that prehistoric people made settlement choices based on particular characteristics in the natural environment; second, that thos e chosen environment factors can be mapped within the modern envi ronment in a given area of interest . With this in mind, and with a sufficient sample, it should be possibl e to distinguish between places where archaeological sites should or should not appear. Thus, in orde r to implement a predictive model, first a set of georef erenced parameters related to site occurrence must be collected, stored, and manipulat ed. Known archaeological ev idence has to be acquired and classified into chronologically a nd typologically homogeneous groups. By combining both landscape and arch aeological data sets, a heuris tic or statistical model is then constructed, allowing the archaeologist to link the spatial variability of such parameters with the occurrenc e of the sites (Kvamme 1989). Our ability to grasp and accurately measure the variability between elements of site and non-site areas is critical for accurately modeling settlement patterns across the landscape. These predictive models help not only to location possible sites, but they make it possible for land mangers to get a sense of the expected distribution of the resources under their care. The goal, then, of predictive modeling is to establish a correlation between certain environmental pa rameters and known archaeological site


65 locations, build a statistical model based on that relationship, and apply the model to unsurveyed land. GIS remains the most capable tool for performing these tasks. GIS has spawned a revolution in spatial thinking, maki ng a fundamental change in the way human spatial behavior is studied. The physical space occupied by a population serves as the primary frame of reference by which they locate themselves and their varied activities relative to all other occurrences. There are many cultural in fluences upon space and place: communication routes, personal taste, landscap e history, proximity to other sites, aesthetics, and ritual, to name but a few. They have been omitted, for the most part, because they are unknown in sufficient detail to allow for their oper ationalization in a GIS framework. If the archaeological record was that complete, ther e would be no need for predictive models in the first place. However, this is not to say that there is no possibility for either a standalone cognitive model or a coupled cognitiveenvironmental model. “If people’s actions are systematically patterned by their beliefs, the patterning (if not th eir beliefs, as such) can be embodied in the arch aeological record” (Renfrew et al. 1982:11). Research (like viewshed and cost surface analyses) continue s to delve into how human behavior could be incorporated in a GIS framework. What Drives the Model? In essence, the necessary information needed to generate accurate modeling projects depends on the purpose of the project. In most academic projects, the goal is to model the locational behavior of different f unctional, chronological, and cultural types of occupations (components). By contrast, the goal of most cu ltural resource management projects is to conserve res ources and limit cost by identi fying areas with and without resources regardless of the nature of the re sources themselves. Given this goal, and the


66 difficulty involved in clearly identifying m eaningful functional a nd cultural types of occupations that are securely anchored in time in most archaeological sites, it is not surprising that, in early modeling attemp ts (in the 1970s, 1980s, and 1990s) the most frequently used dependent variable in these contexts were relatively simplistic, binary cases of the presence or absence of arch aeological materials (e.g., Bradley et al. 1986; Kvamme 1984, 1986, 1990; Parker 1985; Wa rren 1990). Because these approachs lumped occupations of various kinds together , it incorporateed a grea t deal of locational variability that reduces the potential pred ictive power of those models (e.g., Judge 1973; Roper 1979). However, such approaches had the advantage of minimizing complexity by focusing on defined events that form a mutually exclusive, exhaustive, and nonambiguous partitioning of the region being i nvestigated and of producing large sample sizes, because of the use of the singl e “resource present/absent” class. Such approaches also depended on co mmon locational tendencies that were perceived to crosscut functional and cultural categories, such as proximity to water and preference for level ground, and that many loca tions in a region were unsuitable for most kinds of activities for similar environmental re asons, such as the pres ence of swamps or very steep slopes (e.g., Kvamme 1985; Kv amme and Jochim 1989). Many powerful predictive models have been built using this simple solution to defining all the possible events that can occur in a land par cel (e.g., Kvamme 1989, 1990; Parker 1985). Other choices of dependent variables in early locational modeling efforts included multiple site types (e.g., Kvamme 1988; Parker 1986), counts of artifact density (Green 1973; Nance et al. 1983; Zubrow and Harba ugh 1978), and various measures of site significance (e.g., James and Knudson 1983). The advantages and disadvantages of these


67 choices are reviewed in Kohler and Park er (1986), Judge and Sebastian (1988), and Kvamme (1990). A variety of independent va riables have been used in archaeological models of locational behavior, including sociocultural and radiometric (e.g., Custer et al. 1986) characteristics and positional parameters (Parker 1985). Historically, many modeling projects have focused on the economic com ponent of site location, as environmental factors were generally consid ered intimately related to locational decisions by groups without advanced transportation. It is the archaeological signatures of these populations that were the main focus of many cultural re source management surveys during the past few decades in most parts of North Ameri ca (e.g., Jochim 1976; Wood 1978). The crux of the argument historically centered on the belief that th ese types of societies placed emphasis on economic transactions within a regional environment and that these populations tended to minimize the time and effo rt they expended in these transactions (Kvamme 1990: 271). The effect was to encourage location cl ose to important environmental resources. The focus on envi ronmental or biophysical characteristics of geographical locations (such as slope, soil type, elevation, plant community type, and distance to water) is also a pr actical one as these variables are relatively easy to identify today through measurements or observations made on maps, aerial photographs, remotely sensed data sets, and even computer-generated spatial information sources, such as GIS. Environmentally based predictive locational mo dels work by correlati ng the location of a sample of sites with the environmental characteristics of the land parcels they are in and predicting that other, unknown sites will be present in parcels with similar sets of


68 characteristics. The goal is to define those characteristics of physical locations that have some bearing on the distribution of ar chaeological resources in a study area. In some ways, these approaches to modeling mirrored the processual/postprocessual debate that still rages within archaeological circles. While this study does tend to emphasize a recording and analysis of pattern in a processual, positivistic way, it is by no means limited to the theoretical bent of the Processual School proponents. In fact, much of this dissertation is grounded in a historicist, post-processual approach. The middle ground, however, seems to be the be st place to situat e this research. Within archaeological regional or landscape st udies, current or in the past, with or without the use of GIS, the general approach has been c oncerned with discerning and interpreting patterns of archaeological land use and settlement. The archaeological record certainly retains pattern, but patte rn can also be found in numerous factors affecting our interpretation of the record (namely erosion and deposition, land use, and research bias). This raises the specter of how can we interpret th e archaeological record if we cannot separate the pa tterns we wish to illumina te from the “background noise” wrought by other processes? If human patter ning is only one of a number of factors determining the patterns we find in the arch aeological record, what hope do we have of modeling these processes, and wh at, in the final analysis, does this say about the models we develop? An examination of the literature reveals some of the most basic environmental variables used in predictive models: elev ation, slope, aspect, and distance to water (Kvamme 1985; Parker 1985; Altschul 1990; Carmichael 1990; Warren 1990). Culturally relevant environmental variables can actually be derived from any number of


69 sources (ethnographic analogy, Jochim 1976; simulation models, Gunn 1979; impressionistic models, Gardner 1978; correl ation studies, Bettinger 1977; or linear programming approaches, Keene 1979). Dean ( 1983:11) has raised an issue that, rather than attempting to find areas that match a number of criteria, people may actually look for only a select few critical variables in their surroundings when identifying and selecting activity locations. Many would d ecry such an approach as environmentally determist, especially in light of pendulum-like swings of archaeological theory from the scientific, rule-findi ng approach of the 1960s and 1970s to the humanistic, historicist approach of postmodernism in 1980s a nd 1990s where many believed that such approaches were a thing to distance oneself from for fear of rebuke. This highlights concerns about the analysts' ability to choose “meaningful” environmental variables for inclusion in the modeling process. Meshing of Archaeological Theo ry with Predictive Models Anthropologists and archaeol ogists are acutely aware of the spatial as pect of the material culture they aggressively utilize in cultural reconstruction efforts. However, the methodological approaches to concepts of space and place have been inconsistent from one theoretical paradigm to the next. 19th cen tury Diffusionist thinki ng resulted in the formation of culture area types, such as the re gional divisions laid out by Alfred Kroeber. A fundamental shift in theore ctical approach resulted in many archaeologists adopting a methodology of charting similar artifact traits to generate spatia lly and temporally defined archaeological traditi ons and phases. In Europe, the Austro-German school of anthrogeography (1880-1900) introduced the noti on of the “Kulturkriese”; a formalized methodology of mapping cultural behaviors and material cultural initially over large regions, and, with later refinement, smaller spatial scales. Spatially considerations


70 remained prevalent in European archaeology. “Comparative analysis of archaeological distribution maps had become a standard, if in tuitive procedure in European archaeology” (Clarke 1977: 2). Archaeologists formally tr ained or teaching in geography, such as O. G. S. Crawford and H. J. Fleure, publis hed archaeological findings in geography publications including the Royal Geographical Society’ s Geographical Journal and National Geographic. In the 1930’s, C. F ox’s “Archaeology of the Cambridge Region” combined archaeological and environmental distribution maps over time similar to modern GIS-assisted research. By th e 1930’s and 1940’s, American anthropology had enthusiastically begun to reject diffusionist theories in favor of re-emerging evolutionary explanations. “Anthropology turned inward and sought to demonstrate the roles of history, place, and locality as the primar y means by which an unde rstanding of human cultural diversity could be appreciated. “Sp ace, thus, became passive and sterile as an analytical concept” (Aldende rfer and Maschner 1996:6). Ecological anthropological theory and the adoption of the ecosystem concept (Julian Steward) was key to the reintegration of spatial thinking in anthropology, at least at smaller scales (Aldenderfer and Maschner 1996:7). Again, paradigmatic shifts in theoretical approaches drove archaeologists to place increasing emphasis on spatial analysis as central aspect of their rese arch, and a fundamental means of answering archaeological questions. Willey, in his cla ssic Prehistoric Settlement Patterns in the Viru Valley (1953), defined settlement patter ns as “the way in which man disposed himself over the landscape on which he lived. It refers to dwellings, to their arrangement, and to the nature and disposition of other build ings pertaining to community life” (Willey


71 1953). Ecological spatial vari ability was soon incorporated into more generalized settlement pattern research as its popularity grew. The emergence of the New Archaeology (and with it a drive for quantitative, scientific means of explanation) drove neoscenitific archaeologists to seek answers outside their discipline. Th ey integrated concepts such as Von Thunen’s economic distance, Christaller’s centra l place, and Chisholm’s catchment area in an effort to standardize the practice of spatial analysis , to produce a more formulaic means of approaching such analysis (Clarke 1977) . Clarke’s 1977 seminal treatise Spatial Archaeology (as well as Models in Archaeo logy (1972) and Analytical Archaeology (1968)), placed spatial studies at the forefront in methodologi cal approaches to answering archaeological questions. Clarke de fined “spatial archaeology” as: “the retrieval of information from ar chaeological spatial re lationships and the study of the spatial consequences of form er hominid activity patterns within and between features and structures and their ar ticulation within sites, site systems, and their environments: the study of the flow and integrati on of activities within and between structures, sites and resource spaces from the micro to the semi-micro and macro scales of aggregation” (Clarke 1977:9) The growth of landscape archaeology from its nacient form (spatial archaeology, which focused on the spatial analysis, as well as examining demographics, as well as social and economic interaction) drew arch aeologists to examine issues revolving around social aspects of the land, at how people pe rceived the landscape. “Landscape is the spatial manifestation of the relations between humans and their environment” (Crumley and Marquardt 1990). Sites were not viewed as independent units, but ra ther, as a part of a network of habitation areas, roads, ritu al spaces, rivers, landforms, and resource extraction sites “that societies use and imbue with meaning” (Crumley and Marquardt 1990). Landscapes were relative, particular to the individual or group of individuals


72 experiencing it. Landscape archaeologists we re concerned with the social creation of place, not necessarily the analysis of space (Knapp and Ashmore 1999). New approaches are allowing researchers to examine the dynamic interchange between humans and their environment. Often, anthropologists have assumed that "culture has triumphed over na ture.” This conclusion is simply “wrong-headed” and overly simplified, and harkens back to an outdated nature-culture dualism. As anthropologists, we are cons tantly bombarded by introductor y anthropological texts that tell the story of human evolu tion in environmental terms, a nd further confiscating readers by separating our “evolution” from our “h istory” and denying the environment a meaningful role in the development of cu lture. Instead, values , beliefs and issues, history, and culture constitute the key elements of the explanatory framework. Too often, we find a schism within anthropological circ les: in one camp, cultural materialists and other environmental determinists, while diametrically opposed are groups of anthropologists who claim the transcendental nature of culture to rise above the environment within which it is placed. The first is an exercise in humility, denying the interaction of man with his e nvironment and the capacity of human cultures to alter their landscapes in dramatic ways. The second is the ultimate form of hubris, disavowing the massive controls and constraints nature can put upon culture. While few are actually at either end of this theoretical spectrum, many anthropologists find themselves sliding towards on end or the other. Historical eco logical approaches have actually taken steps to put us in the center of this continuum. Historical ecology is merely a framework of perspective, allowing researchers to inve stigate the recursive nature of humanenvironment interaction (Crumley 1994).


73 Integration of Landscape Historical ecology gives us a powerful t ool with which we can modify static models and make them dynamic. Inco rporating a study of dialectical humanenvironmental relationships within a give n study area requires in teractive long-term sequences and a study of changing landscapes. Th is integration of lands cape is critical in our development of a working model. Th e landscape is a human construction of the environment, imbued with cultural and social significance and givi ng an area a sense of “place” (Crumley 1994; Tilley 1994; Balee 1998; Knapp and Ashmore 1999). In today’s terminology, the concept of landscape ha s begun to incorporate socio-symbolic dimensions, allow researchers to envision the la ndscape as “an entity that exists by virtue of its being perceived, experienced, and c ontextualized by people” (Knapp and Ashmore 1999:1). This broader definition of landscape has al lowed anthropologists in particular to stress the relationship of people to their environment both horizontally through space and “vertically” through time. Thus, evidence for the historical interrelatedness of humans and environments may be read in the lands cape. By identifying the mark that human cultural practice has made upon its environment, changing human attitudes may also be identified and their effects studied. By u tilizing historical ecological approaches, we surpass the limitations of simple landscape eco logy (the study of st ructure, function, and change of a heterogeneous land area composed of interacting ecosystems) and move into a framework within which we can conduct a study of past ecosystems by charting the change in landscapes over time, and utilizing data recovered for these studies to add an element of history to our selected predictive modeling variables. The symbolic structures that form an environment do not operate in dependently of the pe ople who conceptualize


74 them. Attempts to resolve a symbolic or ideological landscape w ithout a discussion of the “practice” of the indivi duals inhabiting that landscap e quickly become impossibly convoluted; ultimately remaining unresolved (s imply a “you can’t get there from here” problem). What implications does this have for the future of predictive models? We need to decide the scale at which we feel most comfortable applying our model. Regional models may posses generalizing principles th at allow us to understand human interaction at a large scale, but the details of that inte raction become fuzzy due to the resolution of the model. More accurate studies can only be performed utilizing specific perspectives. This dramatically increases th e accuracy of the model, but vastly reduces its generalizing potential. It becomes an ex ercise in selecting the lesse r of two evils based upon the questions being asked by the researcher. Our job may not be a futile as it sounds . Granted the above choice sounds like a difficult one to make. In either instance, however, the introduction of historically informed environmental analyses into such studies offers an important opportunity for anthropologists, and archaeol ogists to make strides toward s improving the way in which they model human-environmental interaction. Th e true test will be to see if we can get beyond simple numerical representations of cultural phenomena and start understanding the way in which cultural la ndscapes are produced. Refined Approach Kohler and Parker (1986:433) have stated "Perhaps in building predictive models we are too ready to make the assumption that only a complex multivariate model can adequately account for human locational behavior , when in fact, a few (proxy?) variables, observed in the highly correlated data base th at is our environment, may be sufficient for


75 forming locational decisions". I take issue with this, for as I see it, human behavior simply cannot be modeled upon a few variables, and correlational data my not be as solid as it appears on the surface. Certainly, we can find correlation in slope, distance to water, and other environmental variables to the occurrence of certain site groups, but such correlations may well be the result of erosion, secondary deposition, or differential visibility of archaeological materials on the surface, rather than directly related to human/environmental interaction. If we rec ognize anything at all, it should be that the patterns of human behavior we are attempti ng to uncover can and do occur at different spatial and temporal scales. This dissertati on is about the detecti on and description of such patterns, and the development of a heur istic device as a starting point for discussion about how Xinguanos may have modified their environment, and how we can assess the extent of those modifications through the develo pment of a predictive model. Since it is absolutely critical that we first understand how the ways in which we study the past affect our understanding of it, much of this research is given over to a me thodological approach, including how we develop new ways of coll ecting and recording data, how we analyze the data and determine some structure or pattern, and how we attach meaning to the results of our analyses. The modeling approach described in this dissertation is based on the assumption that human behavior is driven by a number of cultural and environmental factors. The physical locations of past occupational areas are assumed to have been the result of informed choice on the part of the indigenous groups being c onsidered in this research (i.e., the choice of where to settle was arri ved at based upon some inter-group interaction,


76 as well as an interaction of the population with their environment). This means that significant regional patterning should exist in the distribution of archaeological resources, an implication of the assumption supported by numerous studies of settlement data (e.g., Judge 1973; Kvamme 1985; Roper 1979). It is outside the purview of this research to derive what environmental factors may have been at play, nor can we easily derive what cultural factors may have influenced settlement choices (although enthographic accoun ts certainly give us insight into what kinds of locations were more suitable than othe rs). In fact, we would be quite remise if we based our model on simple correlations of prox imity to water, or soil drainage values. Archaeologists often take a sta tic, classificatory approach to the environment, even when the human variables happened to be considered part of a dynamic system. Church et al. decry the use of modern environmental da ta in APMs as well: “all the predictive models to date have relied on variable e xpressed in the contemporary environment. To expect a model based on present-day condi tions to be of use to modeling the site locations of say, Paleoindian sites, is a tenuous assumption at best” (Church et al. 2000). The pattern must be found outside of these types of relationships. While geographic information systems can give practical integra tion of spatial structures (habitations, soils, river drainage), practical unde rstanding of past and curre nt relationships among these environmental and human systems require a culturally specific temporal and spatial perspective applied at a regional scale. Two broad assumptions must be made before a predictive model may be employed. First, archaeological resource locations must be assumed to be nonrandomly distributed with respect to identifiable environmental variables; and second, site samples can be


77 obtained that are sufficiently representative of resource locations of the region under study. To satisfy this second assumption, this study utilized targeted sampling, based upon informed knowledge derived from ethnographi c investigations about what sorts of resources may have been of particular interest to the progenitors of the archaeological site locations in the Upper Xingu. The process of targeted sampling will be described in more detail in the remote sensing portion of this dissertation. The goal of this research is to find new ways of studying past human/environmental interaction utilizing technologi cal advances which can aid in the discerning of pattern over large spatial scales. By utilizing an integrated approach to these technologies, it may be possible to formulate a model of what sorts of areas might be likely to contain traces of past human occupation, and thus c ontribute to a larger understanding of the interplay of humans and the landscape within which they existed in the Upper Xingu region.


78 CHAPTER 4 GLOBAL POSITIONING SYSTEM (GPS) The first steps in providing information fo r the development of any modeling effort are the collection of spatial a nd attribute data. For the purpo ses of this study, a synthetic approach was utilized, incor porating GPS spatial data colle cted over the course of two field seasons, which was then augmented by ethnoarchaeological da ta, and unified in a GIS database to present a unified platform fo r analysis of remotely sensed imagery. The GPS was used as both survey tool and as a general storage device for georeferenced attribute data, thus serving two critical functions duri ng the course of this project. In the following text, a brief description of the ove rall system behind GPS is provided, as well as an in-depth treatment of how such a system may be more effectively incorporated into archaeological investigations. Using GPS in Archaeological Survey Historically, archaeologists have been highly skeptical of the degree to which measurements obtained via GPS signals can be relied upon and utilized in mapping and surveying. To be sure, there are a number of sources of possible e rror and biases that enter into the process of obt aining precise positioning with a Global Positioning System. The combined magnitude of these error factor s influences the accuracy of the positioning results. Biases may be defined as being t hose systematic errors that cause the true measurements to be different from observed measurements by a constant, predictable, or systematic amount. Biases may have physical bases, but may also enter at the data processing stage through imperfect knowledge of constants, for example any "fixed"


79 parameters such as the satellite orbit, station coordinates, velo city of light, etc. Both the positions of each satellite in the constellat ion and their on-board atomic clocks are constantly monitored, each of the satellite ve hicles drift slightly from their predicted orbits over time, and their on-board atomic clocks are can never remain completely synchronized. Additionally, satellite transmi ssion can often be disrupted as it travels through both the troposphere and ionosphere. Signals reaching the ground antenna are subject to "multipathing.” Random measurem ent errors can dramatically affect the accuracy of precise position observations. Random errors are simply unpredictable events (in magnitude and sign) and are due to any one on many factors. The chief sources of random error are the "resolution" of the measurement scale, random internal instrumental effects, and, very rarely but qui te possible, some exte rnal, highly localized condition such as micro-meteorological events , local signal interfer ence, and the like. Systematic errors, on the other had, occur according to some pa ttern. For example, errors of this type may be induced by the in strument, the observer, the physical or environmental conditions, but at a constant ma gnitude, or such errors might be the result of incorrect application of ca libration data. Finally, gross e rrors are the result of overt blunders. By their very nature, these types of errors are often so conspicuous that they can be easily identified and corrected im mediately. If, however, these errors go unnoticed, they can have detrimental impacts on the final product. GPS relative positioning, also called di fferential positioning, employs two GPS receivers simultaneously tracking the same satellites to determine their relative coordinates. Of the two receive rs, one is selected as a reference, or base, which remains stationary at a site with precisely known c oordinates. The other receiver, known as the


80 rover or remote receiver, ha s its coordinates unknown. Th e rover receiver may or may not be stationary, depending on the type of the GPS operation. Thus, higher accuracies are generally possible if the relative pos ition of two GPS receivers, simultaneously tracking the same satellites, can be derived. The principle behind this observation is qui te simple: because a wide variety of sources of error can ultimately affect the abso lute position of two or more GPS users to almost the same extent, these errors largely cancel each other out when differential or relative positioning is executed. There are a wide variety of differential positioning procedures, but each method shares a common thread in that the position of the GPS receiver of interest is derived relative to another fixed, or reference, receiver whose absolute coordinates in the satellite datum ar e assumed to be known. Thus, the ability to derive precise positions via a GPS essentially requires measurement of the baseline components between simultane ously observing receivers. Ultimately, accuracy is dependent on a number of variables. Precision in position measurement clearly depend on whether the user is moving or stati onary as stationary observations permit an improvement in precis ion due to the effect of averaging of positions over time. Accuracy is also depende nt on whether or not the data can must be processed in the field, or if it can be subjected to post-pro cessing at a later date before implementation. Real-time positioning requires a "robust" but less precise technique to be used. The luxury of post-processing the da ta permits more sophisticated modeling and processing of GPS data to minimize the magnit ude of residual biases and errors. Again, the level of measurement noise has a cons iderable influence on position precision, with carrier phase measurements permitti ng a higher accuracy than pseudo-range


81 measurements. Accuracy is influenced by th e degree of redundancy in the measurements (for example: the total numb er of SVs within view of the receiver (dependent upon the elevation cutoff angle, the number of receiver tracking channels, the ability of the system to track other GPS constellations like GLOS NASS, and the number of observations. GPS relative positioning has the advantage of consistently providing a higher degree of accuracy than autonomous positioni ng observation. Even higher levels of accuracy can be obtained by utilizing carrier -phase measurements as opposed to pseudo range measurements, primarily due to the principle that measurements of two (or more) receivers simultaneously tracking any given SV will elicit similar errors and biases (Langley 1993). The similarity of error measurement is a function of the distance between the two receivers (the closer they are in proximity to one another, the more similar the levels of error will be). If th e difference between the two measurements is calculated, then it should follow that any sim ilarity in error should cancel out and be removed form the measurement equation. Sta tic GPS surveying is a relative positioning technique that depends on th e carrier-phase measurements (Hoffmann-Wellenhof et al. 1994). A static methodology is designed to utili ze two or more stationary receivers to simultaneously track the same SVs. A desi gnated base receiver is set up over a point with precisely known coordinates, such as a geodetic benchmark. The other receiver is set up over an unknown point. By collecting simu ltaneous measurements at both the base and remote receivers over a specified peri od of time, a large amount of averaged positions can then be post-processed to de termine the precise location of the unknown point in relation to the coor dinates of the geodetic benchm ark. More applicable to archaeological survey, however, is the fast, or rapid, static survey ing technique. Rapid


82 static survey also uses two or more receive rs simultaneously tracking the same satellites and using carrier-phase measurements. The difference from the aforementioned static technique is that rapid static survey requires that only the ba se station remain stationary over a known point. The rover need remain stationary over unknown points for only a short period of time, and then can be moved to another point (Hoffmann-Wellenhof et al. 1994). Post-processing of data co llected in either of these st atic modes may elicit either a fixed solution or a float solution (dep endent on whether enough common data was collected from the receivers for the software to fix ambiguity parameters at integer). Fast-track, or “stop-and-go,” provides for th e most rapid form of static survey. This method also employs two or more GPS receivers simultaneously tracking the same satellites with a stationary ba se receiver and any number of rovers that travel between unknown points, taking very brief (oneto tw o-second recording rate for a period of about 30 seconds per each stop) measurements. As long as the initial integer ambiguity can be determined (through initialization), one can hope for approximately centimeterlevel positioning accuracy as long as ther e is a minimum of four common satellites simultaneously tracked by both the base and the rover receivers at all times. In order to perform properly, the rover and base station must always ma intain at least four common satellites, even when the rover is being m oved, otherwise the initial ization process must be repeated again. The last carrier-phase relative positioning technique that will be discussed is called RTK surveying. Again, it employs two or more receivers, with one operating as a base station, and one or more rovers. The method is most useful when: 1) the survey involves a large number of unknown points located in the vicinity th e base station; 2) the


83 coordinates of the unknown points are required in real time; a nd 3) the line of sight, the propagation path, is relative ly unobstructed (Langley 1998). This is actually the preferred method by many users because of the relative ease of use and the ability to determine position in real-time as opposed to the other methods that require some degree of post-processing. The base receiver measurements and co ordinates are transmitted to the rover receiver a radio beac on. The built-in software in a rover receiver combines and processes the GPS measurements collected at both the base and th e rover receivers to obtain the rover coordinates. More commonly used methods include real -time differential G PS (DGPS), a codebased relative positioning t echnique that employs two or more receivers simultaneously tracking the same satellites and based on the f act that the GPS errors in the measured pseudo ranges are essentially the same at bot h the base and the rover, as long as the baseline length is within a few hundred k ilometers. Again, base receiver remains stationary over the known point. The built-i n software in the base receiver uses the precisely known base coordinates as well as the satellite coordinates, derived from the navigation message, to compute the ranges to each satellite in view. The software measures the difference between the comput ed ranges and the measured code pseudo ranges to obtain the pseudo range errors (o r DGPS corrections). These corrections are transmitted in a standard format called Radio Technical Commission for Maritime Service (RTCM) to the rover. The rover then applies the DG PS corrections to correct the measured pseudo ranges at the rover. Fina lly, the corrected pse udo ranges are used to compute the rover coordinates.


84 The accuracy obtained with this method va ries between a sub meter and about 5m, depending on the base-rover distance, th e transmission rate of the RTCM DGPS corrections, and the performance of the C/ A-code receivers (Langley 1998). Higher accuracy is obtained with short base-rover se paration, high transmission rate, and carriersmoothed C/A-code ranges. Real-time meas urements are often preferable over postprocessed observations because the positioning da ta as well as the accuracy measures can be obtained while remaining in the field (obviously this is a critical issue for archaeologists as it allows for continuous data collection without the need to return to a home office or otherwise connect to some data-transfer syst em to access post-processing data files). The ability to receive corrections in the field inevitably leads to higher productivity compared with post processing. Post processing, how ever, will generally lead to more accurate results, primarily becau se of the inherent abilities of the post processing software to do editing and cleaning of the collected GPS data. Real-time DGPS operations require a comm unication, or radio, link to transmit the information from the base receiver to the r over receiver. DGPS corre ctions are typically transmitted at 200 Kbps utilizing very high and ultrahigh frequency (VHF/UHF) bands (OMNISTAR, ; RACAL LandStar, Dedicated radio link equipm ent is available to consumers to transmit base station information using the VHF/UHF band. While expensive, these types of radio link systems are able to provide exceptional line-of-si ght coverage and have the ability to penetrate into build ings and other obstructions. A number of various GPS correction services are readily availabl e at various levels of accuracy and cost. Very high levels of accu racy can be achieved utilizing any of the


85 highly precise permanent GPS reference st ation networks established by several organizations around the world (IGS and the Continuously Operating Reference Station (CORS). These services are available free of charge. The Canadian Active Control System (CACS) is another regional GPS servic e, which is available to users for a small charge. Reference stations within these systems receive signals on a continuous basis, and thus provide some of the most accurate corrections possible. A number of other countries have established their own internal networks of reference stations along coastal areas (primarily designed to enhance th e safety of marine navigation), which continuously broadcast real-t ime DGPS corrections in RTCM format. GPS receivers capable of accepting RTCM corrections can reac h levels of accuracy in the range of sub meter to a few meters. At the commercial level, two real-time DGPS correction services are widely used. One broadcasts the DGPS corrections through FM broadcast stations, while the other transmits through communicatio n satellites. These systems are called wide-area differential GPS (WADGPS). Both systems require a special receiver to decode the DGPS correction information, whic h would be interfaced to the GPS rover receiver to output positional information at the meter-level accuracy. WADGPS systems have several advantages over conventiona l single-station DGPS systems, including coverage of large, inaccessibl e regions using fewer reference stations. The most useful WADGPS correction systems for remote acces sibility are provided by OMNISTAR and RACAL LandStar. These systems use satell ite data link, with OMNISTAR operating in the C-band of the frequency spectrum, while th e LandStar service operates in the L-band. To access either service, a subscriber need s the system data receiver to receive and decode the DGPS corrections. The data recei ver must be interfaced to a differential-


86 ready GPS receiver to obtain the corrected posit ion. Accuracy of the order of a sub meter to a few meters can be obtained, depending mainly on the GPS receiver type. Advances in commercially available GPS equipment are also improving the use of Global Positioning for precise positioning appl ications. Many olde r receivers relied exclusively on the use of th e coarse/acquisition (C/A) pseudo-random code signals transmitted by the SVs. Many of the newer models, however, have integrated carrierphase receivers. In older models, the receiv er was designed to monitor a smaller number of channels, sequencing through every visible sa tellite to obtain positioning information. Multichannel GPS units can track several sate llites simultaneously, allowing the receiver to monitor carrier phase signals and calculate accurate positions at a much faster rate. In simple navigation applications, input from a single receiver is adequate. However, the more precise-positioning demands of arch aeological mapping and surveying require a much greater degree of accur acy and thus must depend on carrier-phase observations taken at by at least two receivers at regular intervals. The measurement accuracy of the carrier phase is about 1/100 of a cycle, which amounts to 2 mm distance for the 19 cm Ll carrier. Measurements of phase on the Lband carrier signals thus have millimeter random error, while pseudo-range measurements made with the aid of the time signals modulated on the carrier waves are betw een 100 and 1000 times noisier. It is the resolution of this carrier wave cycle th at permits high degrees of accuracy in archaeological applications of global positioning for the purposes of recording positional information and for extensive survey use. There are several advantages that arch aeologists may find wh en utilizing GPS satellite surveying techniques for archaeol ogical site mapping and general survey.


87 Perhaps most importantly, intervisibility be tween stations is not necessary. This advantage cannot be overstated enough, especi ally in cases such as the Upper Xingu project, involving vast areas of unsurve yed area and extremely thick undergrowth. Secondly, because GPS uses radio frequencie s to transmit the signals, the system is independent of weather conditions. Additi onally, much of the hardware is weather resistant, so the process of data collection can continue under a variety of climatic conditions. Because of the generally ho mogeneous accuracy of GPS surveying, the traditional task of planning a network of intervis ible transit stations is no longer relevant. Points can be taken where they are required a nd need not be located at evenly distributed sites. Because intervisibility of stations is not requisite, and conventional network design strategies can be cast aside, GPS surveying is a more efficient, more flexible, and less time consuming method of mapping. In addi tion, because GPS is in operation 24-hours a day, 7 days a week, data collection can be done at any time. Finally, high accuracy can be achieved with relatively little effort, un like conventional survey techniques. However, a few disadvantages need to be taken into consideration. Because station intervisibility is not nece ssary, GPS survey methods are especially attractive to researchers working in rugge d terrain, or covering exceptionally large regions. The relative ease of use, however, is often offset by the logistical problems of transporting and supporting both the technological side, the technical support to run the system. If additional units are put into opera tion, costs and logistical issues could rise even dramatically. Perhaps most important of all, GPS requires that there be no obstruction to the antenna of th e receiver. Thus, a clear view of the overhead sky must be maintained, meaning that overhanging branches or structures have to be removed (though


88 the antenna can be raised above the obstr uction). The benefit gained through the georeferenced positioning of GPS (GPS coordi nates are provided in the earth-centered, earth-fixed coordinate system defined by the GPS satellite ephemeredes) means that any positions collected need to be transformed into a local geodetic system before they can be integrated with results from conventional surveys. However, GPS technologies are becoming more and more accurate all the time, vertical measurements retain significant error, thus limiting the ability of GPS to be used in acquiring accurate three-dimensional maps (GPS vertical positions also must be reduced to the employed geoid). Finally, GPS requires extensive training to be an effec tive replacement. It involves a significant investment of both time and financial resour ces, and requires the development of new procedures and strategies for planning, field operation, and da ta analysis. The inherent accuracy of GPS may be the la rgest barrier to its widespr ead acceptance, since; under proper survey conditions, the measurement s obtained via GPS are often more accurate than surrounding control marks established by traditional survey methods, and integration of the two methods requires manipulation of one coordinate system on another in most cases. GPS Use in the Upper Xingu The Upper Xingu Project initially employe d a Trimble Pathfinder GPS 12-channel Pro XRS receiver attached to a TSC1 data logg er. This unit is capable of receiving C/A code with carrier-phase filt ering and has instantaneous fu ll wavelength carrier-phase measurement. The data logger was connected to compact dome antenna and was held in the operator’s hand. The data logger was attached to the GPS receiver by a short cable, and information was displayed on a small LCD screen. The interface of the data logger consisted of an easy-to-use menu-driven pr ogram. Users input data through a keyboard,

PAGE 100

89 numeric pad, directional arrows, enter keys, and a range of function keys. Customized database "libraries" can be loaded into the da ta logger to make data acquisition easier, but we found the "generic library" to be so simp le, flexible, and easy to use that it was unnecessary to modify it. The GPS unit is easily operated by one person in the field, although two people were often used to ensure sa fety of the operator in remote locations. Pathfinder Office differential correction soft ware was loaded onto a laptop computer resident at the base camp. The Pro XRS mode l allows differential correction to be done in real time using either a control beacon or via a satellite signa l provided by one of several satellite companies. The Upper Xingu Project leader chose a s ubscription to the Omnistar satellite differential correction service due to the remote location of the field sites. The GPS rover data were downloaded from the data logge r daily. Since data were collected using DGPS, no post-processing was needed to make use of the data instantly in the field (though the data was subjected to post-processing once downloaded onto systems back in the United States. Maps were generated each night on the laptop, both to identify errors in the data collection and to gui de further research efforts. This collection of equipment was used to rapidly complete the survey a nd mapping of surface archaeological features in the defined research area. Figures 3-1 and 3-2 illustrate the effectiveness of GPS survey for rapidly detailing overall site ex tent and providing good coverage of the overall feature space, including a delineation of the presence of circular ce ntral plazas, radiating causeways, and defensive berms, often noted as principle landscape features of prehistoric and historic Xinguano settlement patterns (Heckenberger et al. 2003; Heckenberger 2005).

PAGE 101

90 Figure 4-1. GPS survey map of MTFX-6 Figure 4-2. GPS survey map of MTFX-13 Figures 4-4 through 4-8 detail different areas of the Upper Xingu research project. These figures are for illustra tive purposes, meant to showcas e the results of ongoing work in fully surveying multiple sites throughout the area. A 5, 4, 3 color composite of a Landsat 7 image (August 2002) is used as a backdrop for the GPS layer overlay. Figure

PAGE 102

91 4-3 is an expanded version of this composite backdrop, co mplete with a generalized legend explaining how to interpre t different land-cover classes. Figure 4-3. 5, 4, 3 composite Landsat 7 imag e (August 2002) with generalized legend for a rough interpretation of land-cover classes

PAGE 103

92 Figure 4-4. MTFX-6 and MTFX-13 survey data integrated into a GIS with a 5, 4, 3 color composite from Landsat 7 imagery

PAGE 104

93 Figure 4-5. GPS survey map of MTFX-18 in tegrated into a GIS with a 5, 4, 3 color composite from Landsat 7 imagery.

PAGE 105

94 Figure 4-6. GPS survey maps of MTFX-19 a nd MTFX 20 integrated into a GIS with a 5, 4, 3 color composite from Landsat 7 imagery

PAGE 106

95 Figure 4-7. Composite GPS survey map of th e Nokugu cluster of sites integrated into a GIS with a 5,4,3 color composite from Landsat 7 imagery

PAGE 107

96 Figure 4-8. Composite GPS survey map of Kuik ugu cluster of sites integrated into a GIS with a 5, 4, 3 color composite from Landsat 7 imagery The most significant source of operato r error occurred wh en the number of satellites visible to the local base station dropped to a level where it was difficult to establish three-dimensional positions. Differential knowle dge of the operation of the

PAGE 108

97 GPS unit meant that some of the less-capable operators believed such issues could be overcome by selecting the two-dimensional op tion on the local base station and rover receivers, requiring one less sa tellite than the 3-D option. Unfortunately, this option requires an accurate elevation to be determined and entered into the local base station receiver. The error was apparent once the data was differentially corrected. It was immediately apparent that the points collected in this manner were significantly offset in comparison to the data collected using full 3-D mode. These and other instances of operator error were resolved by forcing less-kno wledgeable operators to re-read portions of the manuals that came with the equipment, and retracing those features collected using this form of mapping. Once our initial prob lems were resolved, the equipment proved easy to use. A major advantage of GPS survey is having the ability to make plan maps of archaeological sites and features with the in strument. This capability is not generally appreciated, but advancements in GPS technol ogy now make it possibl e to create feature maps with comparable accuracy to detailed tape and compass maps. In most cases the level of accuracy provided by the GPS receiver (20 to 70 cm) is the same or greater than the discrepancy introduced by the value judgment made by the archaeologist as to where to place the G PS antenna to record a point. Increased acceptance of GPS use in archaeological mappi ng and survey seems assured, particularly as the cost of GPS systems drops and new higher productivity techniques are developed. However, for all its technical advantages , there remain a number of substantial differences between GPS mappi ng techniques and the traditi onal transit survey still utilized by most of the archaeological world. To actually make GPS survey and mapping a viable endeavor to most researchers, techniques will have to be developed to reduce the

PAGE 109

98 significant post-processing time still required to permit GPS positional data to complement more traditional survey methodol ogies (and hence maintain compatibility with the geodetic framework al ready established in most regi ons under investigation). To their credit, the primary manu facturers of GPS receiver equipment continue to improve upon the ergonomics of their models, striving to make equipment that is ever more "userfriendly." Truly, this is where the battle will be won, when even the most technologically inept can effectively put GPS equipment into the field and expect to see a significant return on their investment.

PAGE 110

99 CHAPTER 5 REMOTE SENSING Brief Overview As mentioned earlier, there is nothing ne w about the process of utilizing remotely sensed information to aid in archaeological investigations. There have been numerous efforts among researchers to bring the inhe rent value of remotely sensed data (and specifically the widely-available Landsat imagery) to bear on archaeological investigations. The vast majo rity of cases have seen the implementation of Landsat data sets to delineate environmental “zones” that can be correlated in a general way with archaeological site distributi ons (Schalk and Lyons 1976). Th e principle applications of Landsat imagery by archaeologists who work w ith remote sensing techniques typically revolve around attempts to identify very spec ific landforms or anth ropogenic signatures (crop marks, shell scatters, or structural rema ins) to locate archaeological sites (Ebert and Lyons 1976). However, the 30-meter spatial resolution of even the more current Landsat imagery is still far to coarse to identify any but the most gargantuan of individual features. The solution to this problem is not to bemoan the resolution of the Landsat imagery, but rather identify the proper spatial scale at which to conduct research. Instead of trying to identify archaeological features (many of which are surely smaller than 30 meters on a side) on the ground by analyzing th e satellite imagery, perhaps researchers would be better served by attempting to id entify larger signature s of human activity. One approach to this problem is to identif y those environmental features that would be represented at the scale of Landsat imag ery that might have some bearing on site

PAGE 111

100 placement. The issue of how to link envi ronmental variables (determined by prior predictive models to be critical determinan ts of site locations) with specific on-theground locations is often poorly addressed. Mo st summaries of predictive modeling in archaeology do not consider this rather cr itical problem (e.g., Bettinger 1977; Jochim 1976; Johnson 1977) and predictiv e studies (e.g., Gunn 1979; Jochim 1976; Wood 1978) usually deal with this problem in an impressionistic manner. “Remote sensing” is simply a catchall te rm encompassing a number of different techniques of acquiring and processing informa tion gathered by distant sensor platforms. For the purposes of this study, remotely sensed data will specifically refer to information gathered from satellite sensor platforms. In the past few decades, there has been a growing interest in the applic ation of such data sets to answering fundamental questions in social science research. The growing appreciation of scie ntists for the usefulness of these types of data and the ever -increasing availability of su ch data are equally culpable for the expanded use of remotely sensed info rmation. The real key to the usefulness of these data sets lies in the ability of RS information to provide a means of measuring numerous dependent variables that may have an impact on human settlement and distribution patterns. The La ndsat series of sensors has provided some of the most widely distributed RS data to date, encomp assing a series of satellites specifically engineered to provide high-resolution imager y of Earth’s land surfaces. Originally designed to provide unclassified data for use in land and wa ter resource assessment, the uses of data from Landsat have been expa nded to address everything from visualizing population movement and migration, to monitori ng global deforestati on and fire damage, to estimating soil moisture and land-cove r classifications. NASA recognized the

PAGE 112

101 potential of using space tec hnology to study the Earth's environment as early as the 1960’s. A concerted effort was made to a dvance a program for developing a remote sensing platform designed to analyze both la nd and water based features from space. Earth Resources Technology Satellite (ERTS) , later re-named Landsat 1, was launched by NASA in July, 1972. Landsat 1 carried both a television camera a sensor package called the MultiSpectral Scanner (MSS, which provided swat h-like scans of the Earth’s surface in different portions of the electromagnetic sp ectrum. The MSS proved a valuable sensor package, and was included on the next four La ndsat satellites. By the launch of Landsat 4 in July, 1982, a newly improved multispectra l sensor called the Thematic Mapper was incorporated into the sensor platform. Th e Thematic Mapper adde d three new spectral bands and provided improved spatial resolution of 30 meters (compared to the MSS resolution of 80 meters). Landsat 7 served as the basis for an improved sensor array called the Enhanced Thematic Mapper Plus (ETM+). The ETM+ provides seven channels in the visible, near, mid, and ther mal infrared channels, along with a 15-meter spatial resolution panchromatic sensor. Additionally, Landsat 7 was designed with increased storage and data transmission capabili ties, dramatically increasing the ability of the platform to acquire and transmit images within a given timeframe. The data for a single scene taken by the Landsat 7 ETM+ includes image data for each of the bands and for the panchromatic sensor and is stored as image data. The images are usually corrected for radiometric and geometric distortions and are then made available as a complete set of raw data.

PAGE 113

102 The spectral response of Band 1 is in the visible portion of the electromagnetic spectrum that corresponds with blue-green light. Energy at this portion of the electromagnetic spectrum is easily scattered by particles in the atmosphere, often giving images in this band a hazy appearance. This band is capable of being transmitted through water and is especially sensitive to particle s suspended in water (such as sediments and algae). Data from this band can be used with bands 2 and 3 to create "true" color composite images, which most closely approx imate how the scene would appear to the human eye. The spectral response of Ba nd 2 is in the visible portion of the electromagnetic spectrum that corresponds with green light. It can be used with bands 1 and 3 to create "true" color composite images . The spectral response of Band 3 is in the visible portion of the electroma gnetic spectrum that corresponds with red light. It is also one of the three component bands used to create "true" color composite images. The spectral response of Band 4 is in th e Near Infrared (N IR) portion of the electromagnetic spectrum. This form of infrared sits just outside the visible red light portion of the electromagnetic sp ectrum. This form of radiation is reflected to a high degree off leafy vegetation since chlorophyll (the green pigment in green vegetation) reflects much of the NIR that reaches it (it ha s a high albedo in this band). The spectral response of Band 5 is in the Middle Infrar ed (Mid-IR) portion of the electromagnetic spectrum. This portion of the spectrum is sens itive to variations in water content in both leafy vegetation and so il moisture. The spectral respons e of Band 6 is in the Thermal Infrared portion of the electromagnetic spectrum . Thermal infrared is radiation that is detected as heat energy; therefore, the thermal IR band effectively measures the temperature of the surfaces it scans. Band 6 on the ETM+ sensor can distinguish

PAGE 114

103 temperature difference of about 0.6 Celsius, which allows it to detect relatively small differences in land and water surface temp eratures. The cooling effect of water evaporating from vegetation can be detect ed, assisting in efforts to map land use characteristics of a region. Spectral response of Band 7 is in the Middle Infrared (MidIR) portion of the electroma gnetic spectrum. This portion of the electromagnetic spectrum is sensitive to moisture and thus resp onds to the moisture contents in soils and vegetation. This band is useful in detecting moisture levels in leafy vegetation and thus provides a means to monitor productivity a nd identify areas under cultivation. The panchromatic band is composed of a black and white sensor with a 15 m spatial resolution. The higher resolution of this da ta assists land-use researchers by making identification of sm aller objects easier. Processing Imagery for Analysis Image processing is the first step in assemb ling remotely sensed data into a usable product. The raw data is take in its un-p rocessed form, manipulated using a combination of raw computer hardware power and a variet y of software modules, and rendered into a final form. There have been many texts wr itten about this topic, including those by Campbell (1996), Lillesand and Kiefer (1994) , Jensen (1996), and Cracknell and Hayes (1991). The authors may differ on the fine r points of how one goes about processing images, however, there remains a consistent th read or basic agreement on the major steps involved. There are at least four primary stages of image processing that must be undertaken at the onset of a ny analysis of remotely sensed imagery: preprocessing, classifying, post processing, and an assessm ent of accuracy for the final product. Preprocessing is preparing raw digital data for the main analysis, usually classification. According to Campbell (1996), preprocessing can be separated into three

PAGE 115

104 functions: feature extraction, ra diometric corrections, and geometric corrections. Feature extraction is a process of determining which ar eas of the raw data imagery to utilize. Multiple bands of information are contained in Landsat TM and other reflectance-based Earth observation data. It is a precarious balancing act, deciding which bands may be kept and which may be discarded, but necessary as each information band requires storage space and exponentially increases processing time. The combinations of bands chosen for processing are dictated by the type s of feature extraction required. It is this process of feature extraction, determined by the types of cl assification being undertaken, that effectively determines which bands are critical to the analysis. In some cases, research goals may dictat e that a greater amount of information (read: more bands) is of more import than th e efficiency of the processing. A common compromise between information detail and effici ency of the analysis of the data is to utilize principal component analysis (PCA). PCA creates a few new assimilated bands with most of the variation of the original data, but w ithout the redundancy and noise. PCA achieves thus data reduction by looking for correlations between bands and uses these correlations to redu ce any redundancy. Davis (1986) and Gould (1967) are excellent resources for more detailed disc ussions of principle component analysis. Another strategy in feature extraction is arith metic operations. These operations can take the form of simple band ratios, vegetation i ndices, the tasseled cap transformation (Crist 1983), and even multitemporal combinations. After a determination of which imagery ba nds are of the greatest value to the analysis, the bands themselves must be subjec ted to a process of ra diometric correction. Radiometric correction is an algorithmic proce ss of altering the brightness values of the

PAGE 116

105 original data to resolve sens or specific-noise or to comp ensate for various atmospheric effects such as haze or surface shadows ( like those found on slopes). Sensor-specific noise can show up as striping, which can be attr ibuted to errors in data transmission or collection, or to actual sens or malfunction (Lillesand and Kiefer 1994). Whereas sensor noise can be eliminated mathematically with relative ease, atmospheric interference, due primarily to the augmentation of measured ra diation as it passes from the sun through the atmosphere to the Earth’s surface and then b ack through the atmosphere to be recorded on the Landsat instrument. The most common culprits of atmospheric interference are innumerable particulates in the atmosphere which can cause problems ranging from total obscuring of ground features (e.g., clouds or sm oke), or by altering, to varying degrees, the spectral values (Lavreau 1991). Many re searchers have sought methods of correcting or minimizing these effects (Chavez 1996; Hill and Sturm 1991; Lavreau 1991). Additional factors that can influence record ed brightness values include reflectance of the target, angle of the sensor, solar eleva tion angle, and slope and aspect of the target in relation to the solar direc tion. The reflectance of the targ et is dependent on localized land cover, and thus is typically the variab le actually sought by the analysis. Sensor angle and solar elevation angle essentially remain constants and are not factors in interpreting images due to th e sun-synchronous orbits of Earth observation satellites (Campbell 1996). Topographic effects, or the influence of the slope and aspect of localized elevation values on reflectance va lues, while an important factor in some studies, especially in mountainous environmen ts, has no bearing on these discussions due to the lack of elevation with in the study area under question.

PAGE 117

106 Following radiometric correction, the next st ep in the process of data preparation before a classification can be performed requires that the satellite data be registered with a consistent georeferencing system. Such referenci ng allows all geospatial data to be used readily in a GIS. First, a standard co rrection is made to adju st the effects of the Earth’s rotation on the satellite image, and th en the image is georeferenced to a users defined map projection and to a coordinate sy stem. While it is technically true that geometric corrections can actually be perfor med at any stage in image processing, it is imperative the data geometrically corrected befo re ancillary data are brought in to aid in radiometric correction, classification, or accur acy assessment. Geometric corrections as well as the other steps in pr eprocessing must be done with caution. Most of these processes alter the pixel reflectance values in some way. Alteration of these values may affect classification accuracy, thus, it is usuall y best to alter these values as little as possible (Campbell 1996). Raw data streams from the Landsat platform are of little use initially. After being transmitted to a ground-based station, the data streams must be processed and converted into a usable format, often times into a form of imagery that provides a visualization of the data collected by the sensor. Satellite im age data is sent from the satellite to the ground station in a raw digital fo rmat (essentially a stream of numerical data). Each byte in the data stream corresponds to a single pi xel element. The numerical value of the pixel, known as its Digital Number (DN), is tr anslated into a gray-shade. These pixels, when aggregated and then arranged in prope r sequence form an image wherein varying shades of gray represent the di screte energy levels detected on within a measured area. Since a satellite image is act ually a collection of numeric data, the underlying dataset can

PAGE 118

107 be manipulated using algorithms (mathematical equations) that correct for errors (like atmospheric interference), georeference or regi ster the data set to a specific geographical reference point, or extract information that may not be readily apparent to create a limitless number of derived products generated simply by performing calculations on the raw numerical data. In terms of optical perception, the human eye can only perceive, on average, about 16 shades of gray. The inability of the hum an eye to differentiate gray-shaded images necessitates the conversion of gray-scale satellite images into color derivatives by assigning a specific digital number (DN) valu e (or ranges of DN values) to specific colors, thereby increasing the contrast of particular DN values with the surrounding pixels in an image. The real advantage of utilizi ng digital imagery is that it allows the researcher to exercise some control over the image info rmation through manipulation of the digital pixel values in an image. Many of the imag es distributed to end users have often times already been subjected to a number of differe nt forms of image enhancement (radiometric corrections for illumination, atmospheric infl uences, and sensor characteristics may be done prior to distribution of data to the user , for example). However, the image may still not be optimized for visual interpretation. Satellite-based sensor platforms must be designed to cope with levels of target/background energy ty pically encountered during routine image acquisition. Since the spectra l variation across bands may vary widely depending on the type of land cover under st udy, no generic radiometric correction could optimally account for and display the optim um brightness range and contrast for all targets. Thus, a much more targeted enhan cement of specific images often is necessary.

PAGE 119

108 Pixel values within each band for Landsat imagery can rang in value (DN) from 0 to 255. The most basic form of image enha ncement , termed contrast enhancement, involves changing the original values so that more of the available range is used, thereby increasing the contrast between targets a nd their backgrounds. By manipulating the range of digital values in an image, graphica lly represented by its hi stogram, it is possible to increase the fidelity of specific features in relation to background information within an image. The simplest type of enhancemen t is a linear contrast stretch. This involves identifying lower and upper bounds from the histogram (usually the minimum and maximum brightness values in the image) a nd applying a transformation to stretch this target range to fill the full range of histogram values. This type of contrast enhancement reassigns pixel contrast so th at light toned areas appear li ghter and dark areas appear darker, with the net effect of making visual interpretati on a much easier enterprise. A linear stretch creates a uniform distribu tion of the input range of values across the full range, but this may not always be th e most appropriate way to enhance an image (this is especially true if th e original input range is itself not uniformly distributed). To create a contrast enhancement of a less even ly distributed input range, a more targeted histogram stretch can be utili zed to assign more display valu es (range) to the frequently occurring portions of the histogram. In this way, the detail in these areas retains greater fidelity relative to those areas of the or iginal histogram where values occur less frequently. By expanding vari ous portions of th e original target spectral histogram utilizing a number of differ ent stretching techniques, re searchers can achieve better contrast of desired target areas.

PAGE 120

109 One important advantage of Landsat Thema tic Mapper (TM) is its ‘synoptic view’, which makes it possible to study large portions of the Earth’s surface at a relatively low cost. Sensor systems of higher spatial and spectral resolution do not yet provide this large area view at a low cost. For this reason, improvement of classification methods of Landsat sensor data should focus on problems faced by large-area projects. Instead, this type of research has often been limited to small image subsets, sometimes comparable in size to that of only a few aerial photographs . Consequently, in novative classification methods that produce more accurate results, but require more software, hardware or ground data, have rarely been applied to large areas. The most critical (and the most error-prone) st ep in the process of image analysis is the actual classification of i ndividual pixel elements. The intent of the classification process is to assign each pixe l in a digital image to one of several land cover classes. This categorized data may then be used to produce thematic maps of the land cover present in an image. Simple visual classi fication relies on the analyst's ability to use visual elements (tone, contrast , shape, etc) to classify an image. This method may seem the easiest to employ, but, in fact, such methods tend to have too great an error margin built into the analysis to be useful for anythi ng more than simple “first-run” classification or verification of image anal ysis via other means. The two primary methods of image classificat ion involve either sp ectral pattern or spatial pattern recognition (Lillesand and Ki efer 1994). Spatial patte rn classification is produced through informed relationships betw een neighboring pixels based on statistical analysis of certain elements of each cell. Spec tral pattern classificati on relies on statistics of the brightness values for each pixel of the image in each band. Spectral pattern

PAGE 121

110 classification is the most traditional approach, and the approach that will be sued in this project. Rather than relying on the ability of the human eye to accurately perceive differences, these software based solutions in stead use the actual DN of each pixel to more accurately assign image elements to specific categories. The result of a classification is that all pixels in an image are assigned to particular classes or themes resulting in a classified image that is essent ially a thematic map of the original image. Various themes, then, actually represent cate gorized pixels sorted into a number of spectral classes. Spectral cla sses are groups of pixels that have nearly uniform spectral characteristics. It is the task of the analys t to observe and understa nd the various spectral classes and convert them into information classes, the various themes or groups the analyst is attempting to identify in an imag e. Information classes may include such classes as varied types of forest, various ag ricultural crop types, in land bodies of water, or urbanized areas. Thus, the real task of the human element of image classification is matching the spectral classes in the data to the information classes of interest. While image analysis can be performed on a single spectral band, it is simple not the most effective means of deriving accu rate classification thresholds. One band classification is usually very di fficult to classify since it is entirely within the realm of likelihood that a number of different surface types may exhibit the same spectral value within any given band. Thus, any spectral cl asses in a single band classification will likely contain several information classes, and distinguishing betw een them would be difficult. Multispectral classification, utilizi ng two or more bands of information, allows for the combination of digital numbers to identify in a much more specific way the spectral signatures of the spectral classes pr esent in the image. A greater number of

PAGE 122

111 bands nearly ensure a more accurate differentiation of different cover classes. Normally, multispectral data are used to perform the clas sification, with the spectral pattern present within the data for each pixel is used as th e numerical basis for categorization (Lillesand and Kiefer 1994). Common classification proce dures can be broken down into two broad subdivisions based on the method used: supervised classi fication and unsupervised classification. With supervised classification, the researcher locates areas on th e unmodified image for which he knows the type of land cover, defines a polygon around the known area, and assigns that land cover class to the pixe ls within the polygon. Thus, a supervised classification requires the input of known ground locations with known cover types. This process is continued until a st atistically significant number of pixels exist for each class in the classification scheme. Then, multispectral data from the pixels in the sample polygons are used to train a classification algorithm. Once trained, the algorithm can then be applied to the entire image and a final classified image is obtained. The numerical information in all sp ectral bands for the pixels co mprising these areas is used to "train" the computer to recognize spectrall y similar areas for each class. Supervised classification thus utilizes acquired (ground-truthed) knowledg e of different classes in a scene to aid in identifying re presentative samples of different surface cover types termed training sites (areas that are designed to iden tify the spectral characte ristics of each class of interest). These are referred to as “tra ining areas.” Once training sites have been established, the numerical information in all of the image's spectral bands is used to define the spectral "signature" of each cla ss. Image analysis software is coded to recognize these signatures and attr ibute then to specific classes, which the software then

PAGE 123

112 compares to every other pixel in the image. On-screen digitizing (or other methods) can then be used to designate these areas as trai ning sites. The computer then analyzes these sites and computes statistics about the reflec tance values of each category, according to a specific algorithm. Next all of the pixels are examined and, based on their brightness values, placed in the category with the clos est (stated sta tistical bounds) trai ning values. This method requires a large amount of ancilla ry data (ground data) for training sites, prior to classification. Thus, in a supervis ed classification, the analyst starts with information classes and uses these to define sp ectral classes. Each pixel in the image is then assigned to the class that it most closely resembles. Problems that could reduce thematic accuracy in supervised classification include: lack of training data for every land cover t ype, poor quality of ground data, low quantity of training sites, or a poor dist ribution of training sites. A paucity of adequate training data could cause large areas not to be classified or to be labeled as an incorrect class, depending on the classification method utilize d. Inaccurate or imprecise training data will inevitably have a dramatic affect the tr aining statistics, consequently reducing the accuracy of the classification procedure. If training sites are dist ributed poorly (for example, training sites for each class are cluste red rather than being evenly spread) then spatial autocorrelation problems may occur (C ampbell 1981). If the number of training areas for each category is low, classification statistics may be skewed, depending on the classifier. Training sites are used to form different classes, and the software calculates the mean spectral signature for each class ba sed on the Digital Numbers corresponding to each pixel element in each band involved in the classification procedure. When the

PAGE 124

113 classifier is trained and applie d to the image, the Euclidean distance is calculated between the spectral signature of each pixel and the mean spectral si gnature of each class, and pixels are assigned to the cla ss to which the distance is the lowest. Minimum distance classifies image data on a database file using a set of 256 possi ble class signature segments as specified by the signature parame ter. Only the mean vector in each class signature segment is used. The chief drawback to this method is that every pixel within the image is relegated to a class. While this may be the intent of th e classification, it does not necessarily result in the most accurate ou tput. Under this method of classification, it is entirely possible to have outlying pixels (elements that in no way resemble any of the designated classes) be lumped into a class. The Parallelpiped classification scheme improves on the Minimum Distance to Means algorithm by calculating the mean spect ral signature for each class, and then generating an additional parallelpiped image consisting of as many dimensions as there are bands used in the classification (in the case of a multiple band classification utilizing 3 bands, the resulting parallelpiped image would be a 3-dimensional cube). The boundaries of the generated image are one st andard deviation away from the mean. When the classification scheme is applied to the image, those pixels falling within a particular class's parallelpipe d are assigned to that class, with outliers assigned to an unclassified category. Obviously, this method is more accurate than the first since it restricts the assignment of classes to only t hose pixels that fall within one standard deviation of the class mean; however, the fi nal result produces a num ber of unclassified pixels.

PAGE 125

114 Maximum Likelihood operates under the assumption that values for each pixel element within a training site are distribu ted normally. This procedure calculates the mean spectral signature and a covariance matrix for all spectral bands within a specific class’s training data. When applied to the image, the Maximum Likelihood algorithm uses each pixel's spectral signature and class st atistics to calculate th e probability of each class belonging to that pixel. Whichever cl ass has the highest probability of affinity for that particular pixel is then assigned. Maximum Like lihood is one of the most commonly used supervised classifiers and generally outputs class maps with high classification accuracy, and, given the speed of modern co mputers, is relatively efficient in its computational demands. In essence, the process of unsupervised cl assification works exac tly opposite that of supervised classification. Unsupervised classification, sometimes referred to as clustering, does not require large amounts of initial input (Jensen 1996). The basic premise is that values within a given c over type should be close together in the measurement space (i.e. have similar gray le vels), whereas data in different classes should be comparatively well separated (L illesand and Kiefer 1994). Although these clusters are not always equivale nt to actual classes of land c over, this method can be used without having prior knowledge of the gr ound cover in the study site. Thus, the computer looks for logical clustering of refl ectance values. The analyst may enter initial input like the number of desired thematic or information classes, threshold standard deviations for separating clusters, or the number of passes the algorithm makes across the image The pixels in an image are examine d, without user input, by the image analysis software, and classified into spectral classes. Spectral classes ar e grouped first, based

PAGE 126

115 solely on the numerical information in the da ta, and are then matched by the analyst to information classes (if possible). Programs, called clustering algorithms, are used to determine the natural (statistical) groupings or structures in the data. Usually, the analyst specifies how many groups or clusters are to be looked for in the data. In addition to specifying the desired number of classes, the analyst may also specify parameters related to the separation distance among the clusters a nd the variation within each cluster. The culmination of this iterative clustering process may result in some clusters that the analyst will want to subsequently combine, or cluste rs that should be broken down further each of these requiring a further appl ication of the clustering algor ithm. It is then up to the analyst to translate these clusters into in formational classes. According to Campbell (1996), there are several advantag es and disadvantages to this type of classification. The advantages are: low amount of a priori knowledge of the area is required, potential for human error is reduced, additi onal relevant classes may be identified which may have otherwise been overlooked, and the output classes are spectrally consistent. Disadvantages include lack of analyst cont rol and the possibility of the spectrally clustered output being inconsistent with th e desired informational classes. There are other disadvantages. Spectral relationships change over seasons and years. These changes limit applications to a single time period. There is also a need for detailed ground data when translating th e clustered output from spect ral classes to information classes. The identity of the spectral class is not be initially known, and must be compared to classified data or some other form of reference data to ground truth the spectral classes. The timing of required ground da ta is just the opposite of supervised classification (Lillesand and Ki efer 1994). Thus, in a s upervised classification, the

PAGE 127

116 observer first defines classes based on inform ed knowledge, and then attempts to quantify their spectral separability. In an unsuperv ised approach, the software determines spectrally separable class, but does so wit hout any basis for the formation of specific categories. Post-processing of classified imagery invol ves the preparation of the thematic layer for accuracy assessment, map output, and furthe r analysis. A common procedure in postprocessing is smoothing of the classified im age, which removes the “salt and pepper” appearance on the map caused by spectral vari ability in pixel-based classification algorithms (Lillesand and Kiefer 1994). Smoothing can be accomplished using various filters. Other types of post-processing may involve conversion of the image data from raster to vector, alteration of pixel size, cr eating a new color palette, or forming mosaics of separate images to cover an entire study area. The final procedure in image analysis should always be an assessment of the accuracy of the analysis. Accuracy at this ph ase refers to how close the classified pixels are to the actual land cells. Due to the arithmetic nature of GIS overlays, errors are propagated through each step of a geographic anal ysis, and so it is essential to keep track of error in individual map la yers (Janssen and van der Wel 1994). There has been much attention given to the accuracy assessment of land cover maps in the last decade (Verbyla 1995). Typical accuracy reports include overa ll accuracy, user’s and producer’s accuracy for each category, an error matrix, and the Kappa statistic. Overall accuracy is a simple measure derived by taking the number of tota l reference cells and dividing it by the number of correctly classified cells. The user’s accuracy refers to the number of reference points within a class that were actu ally classified as that class. Producer’s

PAGE 128

117 accuracy refers to the number of pixels that we re labeled a certain cla ss that were verified by the reference data. Both of these statisti cs can be calculated using the error matrix. The error matrix is a table with the reference data on the horizontal ax is and the classified data on vertical axis. The Ka ppa statistic is a measure of how much better the results of classification are versus random pixel a ssignment (Congalton and Mead 1983). The Kappa unit expresses the percent improvement of the classification results over random classification. This statistic can also be calculated from the error matrix. Congalton (1991) and Congalton (1996) provided thorough reviews of this accuracy assessment research and of the current standards. Vegetation is the most dynamic element of the landscape from a remote-sensing perspective. Diachronic anal ysis of vegetative cover must take into account seasonal changes and their effects on the overall spectr ographic signature of any given region. To remove this sort of bias form the analysis, it is often preferable to use images that are recorded at times when the development of ve getation is at an identical or very similar stage. By breaking the vegetation down in to component species, mapping of biogenic emissions can be greatly improved. This can be performed using supervised classifications and will allow vegetation inventories to be produced quickly and efficiently. Vegetation mapping derived from the use of remotely sensed imagery is primarily classified though the cross correlati on of various bands a nd the application of band ratios. Numerous ratios have been te sted; however, Tucker (1979) using Landsat Multi Spectral Scanner (MSS) data, found the re d and near-infra-red ratio was seven to 14 percent better at detection of vegetation than the green/red band combination methods previously used for vegetation detection. The main vegetation component that the band

PAGE 129

118 combinations detect is the green leaf biom ass or green leaf area. These ratios are primarily used for vegetation density meas urement. There are a number of other properties of vegetation evidenced in spectra visible to satellite imaging equipment that can be used to classify flora into various cat egories. Several studies have measured the total biomass of vegetation and attempted to measure and predict evaporation of water to investigate atmospheric interactions (Gholz et al . 1997; Montieth 1976; Gholz 1982) and correlating leaf biochemistry to the spectral properties of vegetation for the purposes of remote sensing. Internal leaf pigments a nd cell structure, as well as reflection or scattering of light, have been addressed (B uschmann and Nagel 1991). Ultraviolet light effects on vegetation are also important fo r remote sensing of vegetation due to absorption and reflection characteristics (M azzinghi et al. 1994). One form of a vegetation index is a band ratio of the re d band with the near infrared band. An abundance of indices are available for dete ction of vegetation from remote sensing (Elvidge and Chen 1995; Jackson 1983). In its simplest form, the division of the near infrared band by the red band correlates to vegetation density and health (greenness). Other more advanced forms of indices include corrections for the influence of soil and atmosphere. Generally, most vegetation i ndices are ratios that eliminate shadowing effects through highlighting the difference in reflectance between two image bands. Removal of shadow and albedo effects from vegetation indices can offer improvements in classification (Qi et al . 1995). Vegetation cover maps are composed of polygons with a content, structure, and composition matching a type description. In a Landsat TM image with a nominal pixel size of 30m, a given vegetation type may o ccupy anywhere from a few, to several

PAGE 130

119 hundred, or even thousands of pixels. Spect ral reflectance values of pixels in close spatial proximity within a given stand tend to be similar. Conversely, the prevalence of spatially contiguous vegetation improves th e likelihood that adjoining image pixels belong to the same cover type cl ass. Contextual classificatio n exploits these relationships among neighboring pixels as opposed to a perpixel classification that derives a cover type from the information linked to a single pixel. Stable, consistent, and predictable relationships among neighboring pixels can be quantified and used to improve the classification accuracy. Further refinement of methodology has led to a substantial number of contextual methods derived from Markov random fields, spatial st atistics, Bayesian methods, fuzzy logic, segmen tation, texture, or neural nets. In the context of this research, remote sensing analysis relates specifically to a human dimension of landscape formation in that it captures the outcomes of human actions writ large in the vegetation itself. Such analysis provides both spatial and temporal information for land use and land c over analysis. For the purposes of this research, pixels components of the satellite images are assigned to land cover classes and classification maps are created. This provi des a means of quickl y assessing areas of similarity and dissimilarity to our classifica tion schema, and allows for rapid aggregation of spectral values within a specific threshold of similarity. An integrated approach to remote sensi ng is not a new concept. Many researchers in the Amazon have used a combination of re motely sensed data in combination with other technologies. Utilizi ng maps based on a visual in terpretation of Landsat TM imagery, Alves (1999) was able to analyze geographical patterns of deforestation for states, municipalities, and road buffers. Skole and Tucker (1993) using Landsat TM

PAGE 131

120 images and GIS integration, mapped land use/land change for the entire Brazilian Amazon. Deforestation, fragmented fore st, defined as areas smaller than 100 km2 surrounded by deforestation and edge effects with in 1 km into forest from adjacent areas of deforestation were measured for 1978 and 1988. The findings supported analyses on the effects of human colonization within the region, specifically tracing rates of deforestation (Skole et al. 1994). Eva a nd Lambin (1998) outline a number of more recent initiatives utilizing an integrative appr oach of combining multiple sensor platforms to aid in estimating biomass burning at a re gional scale. Wood and Skole (1998) have linked satellite, census, and field observation da ta to analyze trends in deforestation. Several other research initiatives using re mote sensing and GIS techniques have taken place at distinct sites and more detailed scal es in the Amazon. A number of studies relating to physical, biologi cal, and social processes ha ve aided researchers in understanding how human decisions affect lo cal and regional land use (Mausel et al. 1993, Moran et al. 1994, Skole et al. 1994, Brondi zio et al. 1996). A number of other recent publications adequately discuss the progression of sensor technologies and techniques to monitor land use and land cha nge, specifically within the Amazon (Adams et al. 1995; Foody et al. 1996; Steininger 1996; Saatchi et al . 1997, Yanasse et al. 1997).

PAGE 132

121 CHAPTER 6 METHODS Increasing complexity in instrumentati on and enhanced capabilities in of data production and methods of analysis in curre nt analysis programs have led to new approaches and to a more integrative visi on about Land Use/Land Change (LULC) within and across research sites (Burrough and Frank 1995). The use of integrated approaches to available technology has allowed the burgeon ing of a new era in Amazonian research. The evolution of these tools has caused a fundamental shif t in the way we approach investigative strategy and implementation. The possibility of testing spatial models using georeferenced databases and algorithms to m easure spatial heterogeneity has opened new pathways to research issues of archaeologi cal investigations. A new era of ecosystems spatial dynamics studies necessitate new qua ntitative methods capable of analyzing patterns, determining the importance of spa tial processes, and developing models about landscapes (Gardner and Turner 1991; Fortin 1999). Therefore, many ecological studies have described features in the landscape by number, divers ity, distribution, complexity, and dispersion of spatial components (Jurdant 1977; Do mon et al. 1989). Advanced airborne and satellite technol ogies, image processing and analysis, and the ever-increasing capabilitie s of advances in technology to provide more efficient means of analyzing spatial data through GI S and associated software have rapidly increased the development and testing of new quantitative methods of spatial assessment (Goodchild et al. 1992, 1993; Sample 1994; Bu rrough and McDonell 1998). The variety of aerial and orbital data in distinct spatia l, temporal, and spectral resolutions have

PAGE 133

122 required the generation of digital image processi ng techniques in appl ications related to the characterization and management of natu ral resources (Johanns en and Sanders 1982; Szekielda 1986, 1988; Richards 1993; Jens en 2000; Lillesand and Kiefer 2000). The exclusive use of Landsat TM imagery in this particular study is purposeful. There are a number of exceptionally good reason s to use Landsat imagery for studies of the Brazilian Amazon, and especially concerning the human dimensions of environmental change. Both types of Lands at imagery (earlier MSS and later TM data) cover a broad spatial extent. Each individual scene acquired by either platform covers is approximately 185 kilometers wide, based on Land sat’s large instantaneous field of view. Perhaps more importantly, however, the Lands at platform acquires scenes of all the Earth’s terrestrial surface betw een 81 N and 81 S latitude at regular intervals (Campbell 1996: 162), with the archives dating back to 1972. While a mere 30 or so odd years is decidedly small in comparison to the history of human occupation of the landscape, these regular observations provide a substantial tem poral range to capture many of the humaninduced changes that have occurred in Amaz onian landscapes. Perhaps most importantly for the purposes of these investigations, Landsat TM instrumentation provides a relatively high degree of spectral resolution in comparison to competing platforms. Traditional Analysis Techniques Techniques of image analysis have alr eady been thoroughly de fined as varied methods for displaying and inte rpreting band-to-band variations of multispectral satellite images. The most common approaches to im agery analysis comprise numerous variants of single-band analyses, color composite ge neration, band ratioing, vegetation indices, principal component analysis (PCA), and classi fication. Several of th ese techniques were used in the course of this study, and most will be thoroughly explained in coming text, so

PAGE 134

123 at this point simple descripti ons seem to be the most expedi ent way of dealing with this subject. A single-band analysis typically involves a simple display of individual bands. Color compositing techniques superimpose three bands together, displaying each band of information using three primary light colors : blue, green, and red. The limitations of simple color composite analysis is that thr ee, and only three, bands may be used at any one time. Band ratioing is the use of a variety of band math ematic functions to compare various bands comprising an image. Vegetation indices is a catch-all term referring to a set of some of the more popular band-rati oing techniques specifica lly geared towards vegetation separability and clas sification (and thus of particul ar interest to this study). One of the limitations of ratios is that th ey generally do not take advantage of all the available information contained in multisp ectral images, usually using only two bands (Adams et al. 1995). Additionally, band rati o techniques specifically geared towards vegetation analysis can be in fluenced by many factors not associated with vegetation itself (e.g., soil background and sensor differences) (Campbell 1996). Principle Component analysis will be explained in more detail later in this chapter. PCA is often used to remove interband correl ations that typically exist within multispectral image data. PCA identifies linear combinat ions of the original band data of an image to produce component images representing th e axes of maximum variation (Campbell, 1996). An assessment of variance among multispectral image components allows the researcher to “compress” the data by utilizing PC bands in the place of the original band

PAGE 135

124 data (Lillesand and Kiefer 2000). The use of PCA distinguishes itself form pervious methods by utilizing all the information contained in a multispectral image. At its core, a classification analysis utilizes a classification algorithm (chosen from among a sizable number of different options) wh ich then assigns individual pixels of a multispectral image to discrete categories. The goal of this type of analysis is to greatly simplify continuous image data (7 bands of data for TM) using quantitative techniques for identification of spectrally similar landcover classes within the image (Lillesand and Kiefer 2000). As previously mentioned, classi fication of satellite im agery falls into two basic camps, arguably with a third “hybridi zed” variant. Unsupervised classification procedures group or cluster the multispectral va lues of the image into distinct classes (e.g., water, soil, and vegetation) based sole ly on the image statistics and produce a new raster displaying the class de signations within the image. Supervised classification involves the use of ground-truthed data sets to “train” the clas sifier algorithm to assign appropriate clusters of data to certain land-cover classes. Many Amazonian researchers, in particular, are using imag e classification techniques to aid in the understanding of the human dimensions of environmental change (Brondizio et al. 1994; Lee and Marsh 1995; Moran et al. 1994). Again, one of the strengths of these methods is that all available band information is included in the analysis. As to the primary weaknesses, each pixel may belong to only one land-cover class and most classifications are extremely sensitive to errors induced by surface reflectant anoma lies, as well as non-surface sources of variability (sensor calibration, illumina tion, and atmospheric differences). Realization of an Integrated Approach The term integrated GIS (or IGIS) has b een used extensively within the GIS and remote sensing research community to mo re accurately describe the methodological

PAGE 136

125 inclusion of both RS and GIS information, and the integration between these two analysis platforms into a singular analytical technique (Faust et al. 1991, Laue r et al. 1991, Star et al. 1991). Many researchers have already made use of a number of different image analysis systems and GIS platforms within individua l projects in order to reap the maximum advantage of available functionality. The r ecent trend has been towards multi-system use of multi-format spatial databases. In practice, the actual transfer of spatial data between the two platforms is complicated, primarily ow ing to the unique way each system handles storage and processing, resulting in a number of cross-platform issues that need to be addressed. Firstly, there can be subtle differences in spatial models even within the same generic format (geo-relation feature versus obj ect-oriented vector models) which can be hamper information exchange at the attribute level. Additionally, the conversion of data between formats can lead to some generaliza tion and loss of accurac y. One is dependant on the export utility of each platform in its tran slation of native data to other formats. In order to minimize any loss of quality the data should be left in its native format, but proprietary formatting is the life-blood of many of these systems (and so the motivation to truly integrate is not present). Finally, at the most funda mental level (i.e. actual machine code), these systems are still either vector-based or raster-based, placing inherent limits on the functionali ty of applications to data outside their primary domain. Even in the case of ENVI, which does have built-in functionality to access and display vector data from a GIS system, the system still has limited vector GIS functionality for spatial decision analysis. The reverse is just as true, with current GIS software containing

PAGE 137

126 a fundamental weakness as components of inte grated systems, requiring conversion into a native format to achieve interoperability. There are many case studies involving subs tantial manual and digital cartographic interaction of both types of platforms, highlig hting the issues with interoperability, as well as studies wherein satellite images were classified and then vectorised in order to port the feature classes into a GIS module. This is a te sted methodology that has been successfully implemented many times over, but the processing steps must be well defined in advance. Kontoes et al. (1993) used GI S derived data to post-process a classified image utilizing data from both a raster based image processing system and a vector GIS. By using data resulting from digitized maps, the researchers were able to use coregistered raster forms that could be porte d into the image processing software and be integrated into the classification methodology. Utilization of data interchange format polygons in combination with the layered attr ibute data stored in a GIS, allows the polygons, as well as the attributes attached to those polygons, to be transferred into an image analysis system. There, such inform ation can be used in, for example, image segmentation, to aid selection of training sites in supervised classifi cation, or in image enhancement tasks. The issue, however, is th at once transferred out of their native GIS, many GIS functions can no l onger be applied to the pol ygons (e.g., selection/editing, attribute query, topological queri es). Johansen et al. (1994) have presented GIS as a tool for the integrated analysis and interpreta tion of remote sensing based maps with georeferenced in situ or model environmental informati on. In their analysis, involving both unsupervised and supervised classifications, the authors used a vector-based GIS, a raster-based image processing system, a nd a data visualization program. The

PAGE 138

127 “integrated” nature of such a methodology is in the passing of data between the various platforms. This approach to integration certainly is wrought with problems, but is necessary until fully integrated systems with capabilities for both raster and vector analyses become more prevalent. The value of GIS and associated software is in their ability to provide a data structure to efficiently stor e and manage ecosystems data for large areas, as well as enable both the aggregation and disaggregati on of data between multiple scales, support spatial statistical analysis, improve informa tion-extraction for remotely sensed imagery, and provide input data and parameters for various forms of mode ling (Haines-Young et al. 1996). This has led to an increasing number of applications for synthesized technological approaches. I have established the re lationship between these vari ous technological realms. The issues I shall attempt to resolve is the following: 1. Are there specific reflective signature s useful in delineating vegetative types within the study area? 2. Is there a significant difference in re flective values between vegetation within the localized boundaries of know n historical occupation site s derived from GPS survey and vegetation surrounding those sites? 3. Can we use the selected signatures of vegetation present within anthrosols to extrapolate out to regional-scale mode ls of unknown occupati onal site locations? 4. Can such a process be applied to a pr edictive model at an even broader scale, using analytical principles under discussion?

PAGE 139

128 Figure 6-1. Methodol ogical flowchart Addressing the issue Determine the questions to be asked Provide units of analysis Ethnographic Fieldwork GPS Survey Image Processing Data Correction Integration of survey data sets Establish ground control points Image Classification Classification of imagery Sorting of class features Summation of error GIS Component Data input Data Manipulation Analysis Predictive modeling Analysis and Synthesis Components

PAGE 140

129 Remote sensing, GIS, GPS survey techniques, and spatial analysis must play a central role in providing elements for this di scussion. We must first define out units of analysis. Units of analysis for the region in question must be centered on biophysical characteristics, socioeconomic context, a nd the spatial-temporal arrangements of occupation. Pre-classification Techniques Several pre-processing techniques were carr ied out prior to classification. The first step was to correct geometric distortions present in the raw Landsat TM images. Geometric rectification is the process of image adjustment to a pre-established coordinate system (Lillesand and Kiefer 2000). Before remo tely sensed data can be effectively used, the imagery must be geometrically corrected. It is often the case that imagery that has already been geometrically corrected is ava ilable to the user. However, researchers working in remote areas will likely find that geometrically corrected data are not available for the study area. Fortunately, many software packages provide geometric correction modules. The process of geometric correction (co mmonly referred to as georeferencing) involves match points in the imagery with know n coordinates collected in the field with GPS or recorded from map data known as ground control points (GCPs) By using GCPs in combination with identified pixels in the map imagery, one can apply a coordinate system to the data according to a transfor mation. More dispersed and more numerous GCP's will improve the accuracy of georeferencing (Jensen 1996). The accuracy of image georeferencing determines the spatial integrity of information derived from image processing performed on that scene. If th e data are not properl y georeferenced then

PAGE 141

130 identification of features w hose locations were not know a priori remains in significant doubt. Imagery used in this analysis was regi stered based on identifi able control points. Radiometric correction of satellite imag ery was utilized to account for path radiance (the random entry of energy into a detector's field of view at a given pixel through the process of scattering, causes additive offsets). It is assumed that atmospheric factors creating additive offsets are fairly homogenous and increase pixel values evenly throughout an entire scene. To correct for a dditive path radiance e ffects, pixel digital number minimums are subtracted from each ba nd of the data so that the DN minimum for all bands is equal to zero. The RS imagery package from RSI called ENVI has a built-in function for radiometric correction of Landsat ETM+ images. Species classification and vegetation sepa rability of tropical forest satellite coverages is usually dependant on superv ised and/or unsupervised classification techniques. The spectral complexity of tropica l forest classes has further led to numerous suggestions for procedures and techniques to improve classifications including for example stratification by eco logical zone (Thenkabail 1999; Helmer et al. 2000), topographic normalization (Colby and Keating 1998), spatial filtering (Hill and Foody 1994), image segmentation (Hill 1999), object-or iented classifications (Foody et al. 1996), vegetation indices (Boyd et al. 1996; He lmer et al. 2000) and multi-temporal image data (Lucas et al. 1993). However, no standardized classifi cation approach has been developed for tropical forest mapping as the approaches vary according to objectives and scale of study, e nvironmental settings and so ftware abilities (Thenkabail 1999).

PAGE 142

131 The prospect of separability of tropical forest types using Landsat Imagery is quite poor in tropical environments owing primarily to the rapid regrowth of vegetation, as well as the consistently hi gh level of greenness and dens ity of the vegetation canopy (Salas and Brunner 1998). Add to this equati on the complexity of the reflectance patterns due to variegation within canopy types (Hill 1999) and the 30m2 grain of Landsat imagery, and the result is a challenging task to define and classify vegetation types. The textural complexity becomes especially evid ent in higher resoluti on imagery of tropical environments as the in-class spectral differen ce is significant relative to the between-class spectral variation (Thenkabail 1999; Hill 1999) . Two near cloud-free Landsat 7 ETM+ images, from 19 May 2003 and 04 August 2002, we re acquired and co -registered using GPS surveyed ground control poin ts to rectify the August im age to a 2003 image using a nearest neighbor resampling rou tine that maintained the orig inal 30m spatial resolution. All visible and infrared bands were available for the analyses. However, the thermal infrared bands were excluded due to their lower spa tial resolution. Classification The explicit reason for an adoption of an integrative approach for the purposes of this research is combining vector information with image classificati on in the selection of training areas. Critical to th e realization of an integrat ed solution for classification purposes would be maintaining the flow of information between the two platforms, and maintaining links between the raster image a nd the vector dataset. Both unsupervised and supervised classification al gorithms were utilized. Thir ty information classes form the basis of the supervised image classificat ions used in this research. Training and reference data were extract ed based on GPS survey point s and polygons (ground truthed information), as well as known areas of discre te land-cover types that would have been

PAGE 143

132 difficult to gain access into for the purposes of collecting GPS referenced data (river/lake regions, recently burned ranching areas, and the like). Training data displayed an intentionally high degree of spect ral separability overall base d on statistics extracted for the radiometrically calibrated imagery. Ho wever, the various vegetation classes had a great deal of spectral overlap, which was cer tainly to be expected. Generally, it is preferable to select unimodal training data. Th is way an analyst is sure to select training data from a single spectral class. The hazard of utilizing multi-modal training data is the ever-present threat of a conflation of multiple spectral groups into a single training class. However, as this project aims to find br oad classes relevant to Xinguano settlement patterns, some multi-modal spectral groups we re necessary. Four classes selected from image data contained multi modal training data: cultural clearings, manioc fields, bare soils, and water bodies. Cultural clearings are heterogeneous spaces, and these training sites were as restricted as possible to preven t contamination of the sp ectral signature from surrounding vegetation. Rather th an take broad swaths of c overage from singular sites, small areas were selected from a number of sites to insure that, although the individual training site samples in each known archaeol ogical site were small, overall, a large sample of the specific spectra in question was represented for classification purposes. Gardens and fields were much more problema tic to identify tight signatures for as they are in various stages of re-growth and canopy development at different stages of cultivation. Garden sites compounded the i ssues of identifying spectra unique to regrowth in anthropogenic soils, as Xi nguanos place a premium on the richness of terra preta soils, and thus tend to select such sites for cultivation. This brought up an

PAGE 144

133 interesting dilemma, since training areas desi gnated as known site locations were often also representative of known cultivated locations. Results The results of these investigations were highly mixed. The calibrated imagery offered too little separation of vegetative classe s to be of any use at all. The analysis portion of this research hinged on the ability of the classification algor ithms to be able to distinguish between different vegetated cla sses within an expans e of highly vegetated land-cover. One of the primary issues to c ontend with when attempting to classify the complex vegetative features found throughout th e study area is relate d, in part, to the spatial configuration of agricu ltural fields, the re-use of ar chaeological sites as garden plots, and different stages of secondary su ccession within both ag riculturally active and inactive regions. The relatively small size of each of these types of land cover, and the mixed spectral responses of pixels representing their classes, are responsible for difficulty in proper classification, and misclassifications in both supervised and unsupervised approaches. Several studies have shown that these problems can be overcome utilizing data with higher spatial reso lution, the integratio n of detailed field data to support the classification process (Mausel et al. 1993, Li et al. 1994, Brondizio et al. 1996), the use of spectral mixture analysis (Adams et al. 1995), object-based classi fiers (Foody et al. 1996), indices (Steininger 1996), and hybrid techniques. The difficulty I initially faced can certainly be attributed to a lack of distinction between the classes under consid eration resulting in a miscla ssification of the images. Simple supervised classification of the imagery was completely unsuccessful due to the inability of the imagery to encompas s enough variability among the vegetative components of the landscape to be able to accurately distinguish between vegetation

PAGE 145

134 growing within known site locations and ve getation growing outside of archeological sites. A maximum likelihood algorithm was c hosen, but the error in the classification was simply at an unacceptable level. Rela tively higher accuracy values were found for the 2002 classifications versus the 2003 classifications. Th e higher accuracy in 2002 is certainly due to the timing of the imagery during the middle of the dry season for the region. Unsupervised classifications yielded poorer results yet, even when flexibility was given to the number of classe s that might be formed. Overall, I was wholly unimpressed with the ability of any of ENVI’s built-in supervised or unsupervised classification schemes to delineate different types of ve getation, let alone discriminate vegetative materials reclaiming areas laden with anthr opogenic soils. If the approach adopted was unable to distinguish between known sites and “natural” vegetative growth, it was readily apparent that the primary goal of the rese arch, to provide a successful methodological approach to integrating GIS, GPS, and Remote Sensed imagery into a meaningful predictive modeling tool for past occ upation sites, was simply untenable. A new approach was needed, one that sp ecifically addressed and emphasized the variability within the land-cover classes sel ected. Thus, I incorporated a number of different indices and transforma tions in an attempt to provid e some measure of contrast between the vegetative components of the imagery. Data Transformations Utilizing the IR bands in combination with visible wavelength bands to study vegetation is broadly termed as the Vegetati on Indices (VI) approach to analysis. The derivation of vegetation indice s is loosely defined as "m athematical transformations designed to assess the spectral contribution of vegetation to multispectral observation"

PAGE 146

135 (Elvidge and Chen 1995). The basic premise behind the use of vegetation indices rests on an assumption that selected algebraic comb inations of remotely sensed spectral bands could correlate to the presence of green vege tation in the pixels scene. The essential characteristics of most vege tation spectra provide for chlo rophyll pigment absorption in the red (R) visible bands contra sted against the high reflectivit y of plant materials in other spectrum. Jordan (1969) is credited with first co mbining near infrared and red spectral responses into a ratio that was then shown to correlate highly with leaf-area index. In the interim, a vast number of spectral band combinations have been studied as measures of vegetation, resulting in a vast number of publications that discuss R and NIR use of the different indices to estimate ve getation variables such as percent green cover, leaf area index (LAI), absorbed photos ynthetically active radiation an d others either for general vegetation studies or related to forestry (Fisher 1994; Huete et al. 1994; Myneni and Williams 1994; Spanner et al. 1994) . Various vegetation indi ces have been proposed, modified, analyzed theoretically, compare d, summarized, categorized, and criticized. R and NIR combinations are typically expressed as a ratio, a slope, or other formulation that can generally be separated into three categories: intrinsic indices, soil-line related indices, and atmospheric-corrected indices (Rondeaux et al. 1996). The first types, ratio-based indices, generally address the characterist ic chlorophyll absorpti on by vegetation in the red portion of the spectrum and high reflectan ce by vegetation in th e near-infrared portion (Tucker 1979). Ratio-based indices include the simple ratio or simple vegetation index (variously SR or SVI in the common literature) developed by Jordan (1969), the normalized difference vegetation index (NDV I) developed by Rouse et al. (1973), and various modified versions of NDVI designed to address its sensitivity to factors such as

PAGE 147

136 soil variability and atmospheric conditions. A second type of index is termed soil-line based or an orthogonal index, centered on a line in spectral space (assuming two dimensions, a plane in three dimensions, or a hyperplane in higher dimensions) along which bare soils of differing brightness will lie . Vegetation increases perpendicularly to the soil line. Kauth and Thomas (1976) de veloped their “Tasseled Cap” transformation for Landsat MSS data, the second compone nt of which has become known as the greenness index, which is sometimes called the green vegetation index (GVI). Crist and Cicone (1984) have extended the analysis to six bands of Landsat Thematic Mapper (TM) data (excluding the thermal infrared band ), and it is this vari ation that has been used during the course of this analysis. The following section describes the extent to which various transformations were applied to the data. Transfor mations do not require training and test data or even the c onstruction of information classes. Transformations are operations that place the data onto more readily interpretable axes. Supervised classification training and test data were us ed in this study to extract summary statistics and multivariate descriptors of analytically important land cover classes in transformed data. Transformations can be powerful sources of information for image interpretation particularly for historic imagery where reconstr uction of training data may not be possible. However, a ssumptions supporting transformation operations must be carefully considered. Normalized Differential Vegetation Index (NDVI) Born out of the need to produce accura te information on ve getation at a global scale, remote sensing scientists have devel oped a number of differe nt transformations of multispectral data that are broadly described as vegetation indices. These transformations of multispectral remotely sensed data are designed specifically to derive information on

PAGE 148

137 canopy characteristics relating to total biomass, overall pr oductivity, leaf area index (LAI), amount of photosyntheti cally active vegetation (PAR), and percent of vegetated ground cover (Jensen 1996). First, it should be understood that a great many of the hundreds of variations of vegetation indi ces available often contain redundant information, stemming from the fact that disc rimination of vegetative attributes often takes place within the same narrow parts of the electromagnetic spectrum. Rather than compare the results of different vegetation indice s, this research identifies one of the most commonly employed vegetation indices and applie s it to the problem of locating historic and prehistoric settlement area s and forest modifications. The normalized differential vegetation i ndex (NDVI) is defined by the equation: NDVI = (TM 4 TM 3) / (TM 4 + TM 3). Resu lts of laboratory and fi eld studies indicate that NDVI is strongly correlated with frac tions of active photoabs orbant vegetation and leaf area index (LAI) (Teillet et al 1997; Walter-Shea 1997). As such, NDVI has become a valuable analysis tool for research into everything from global carbon cycling models to commercial crop studies. A high NDVI valu e is an indicator of high fractions of photoabsorbant vegetation, as well as a high LAI within observed pixel elements. Low NDVI values are usually indicative of a rela tive paucity of phot osynthetically active vegetation, as well as a low leaf area inde x. If NDVI values ar e positively correlated with the amount of photoabsorbant vegetation or LAI contained in a pixel, then NDVI should show differences between disturbed canop ies and climax forest cover. One would expect recent clearings to ha ve very low NDVI values due to the high composition of bare soils and dead desiccating vegetation e xposed after such activit ies have taken place relative to surrounding healthy climax canopy. By that reasoning, any culturally active

PAGE 149

138 areas, including circular-plaza villages or recently cleared rocas, should produce low NDVI values due to the existence of, in the case of villages, large cen tral plazas of bare soil surrounded by thatched-roof housing, or bare soils in the case of recent clearing for planting or other activities. Only after habitation locations have been abandoned and vegetation allowed to recolonize should fo rmer residential occ upation sites or other culturally active areas begin to re turn relatively high NDIV values. One would expect bodies of water to have very low NDVI values since IR is absorbed by water and these land cover types contain almost no photoabsorbant matter. Cultural clearings and bare soils (including recently burned ranching/agricultural fields immediately surrounding the park region) exhibit mean NDVI values that are one standard deviation below any of the vegetated land cover classes due to sparse vegetation and the quantity of exposed soil. It is inte resting to note that the category defined as generalized soil actually has a hi gher NDVI value than the cultural clearing class. This is likely due to grass cover th at quickly covers any undi sturbed ground. Both soil and cultural clearings have large standard deviations. The cu ltural clearing class has large spread because the entities comprising the group are hete rogeneous. Soil on the other hand should be more homogenous. Different kinds of soils re flect and absorb energy in unique ways, but the variability observed in th ese data is probably be tter explained by the presence of vegetation in pixels otherwise c onsistent with the mean values for a general soils classification. Net primary productivity of Neotropical fo rests tends to decrea se with the age of the stand. Thus, younger regenerating patche s of forest should have higher NDVI. Complex limb architecture of climax forests should produce the most self-shading of any

PAGE 150

139 vegetated pixel, while older gardens and actively cultivated gardens have less complex limb architecture thereby creati ng a less obstructed path for radiant energy from the sun to travel between photoabsorba nt leaf surface and the spac e borne detector. Visual inspection of NDVI transformed data demonstr ated adequate separability of habitation settlements (Figures 6-2 and 6-3). Figure 6-2. Detail of the Upper Xi ngu study region (2002 NDVI transform) Upon initial inspection, it was possibl e to discriminate and identify known settlements as well as places known to cont ain areas under active cu ltivation. Piecewise linear contrast enhancement pr ovided a good way to isolate various land cover types.

PAGE 151

140 Adjustments made using linear contrast enha ncement drew out ripa rian corridors and recent meanders of major river systems. This gradient of productivity becomes apparent in NDVI transformed data, with these co rridors appearing mo re productive because nutrients are deposited in thes e areas during flooding events. Figure 6-3. Detail of the Upper Xi ngu study region (2003 NDVI transform) NDVI may be helpful in defining edges of some features once they have been identified, but as a settlement discovery tec hnique, it is woefully inadequate given the circumstance of this particular region. In this particular case, NDVI values for both riparian corridors and regrow th material in known archaeolo gical sites were close enough

PAGE 152

141 that it was difficult to separate these two classe s within a single dimension of data. Shape and context play signif icant roles in finding past settleme nt sites and discriminating them from riparian corridors. Typically, Xinguano se ttlements are located near rivers, as are riparian corridors. However, the circular plaza construction of settlement areas contrasts to the riparian corridors’ far more linear shape. The NDVI image is a single layer of 8 bit data rather than seven layers of raw data or any of the other mu lti-layered image transformations. Since NDVI creates single band 8 bit grayscale data, it has limited value for automated class extraction on its own. Suspecting the data reduction that take s place during the calculation of NDVI was responsible for its poor discriminating abiliti es, the methodological pr ocedure turned to producing a number of different vegetative indices (VI) for comparison to see which had the most separability of vegetative classes. Transformed Normalized Different ial Vegetation Index (TNDVI) Transformed Normalized Difference Vegeta tion index (TNDVI) is the square root of the NDVI. It has higher coefficient of dete rmination for the same variable and this is the difference between TNDVI and NDVI. The formula of TNDVI is designed to produce positive values and the variances of th e ratio are proportional to mean values. TNDVI indicates a relation between the am ount of green biomass found in a pixel (Senseman et al. 1996). The values from the Transform normalized vegetation index range from 0 to 1.0 and can be used to find th e leaf area index. Values near one indicate plant vigor and low vegetation cover near zero. The equation for the TNDVI is sqrt(((TM4TM3) / (TM4 +TM3 )) + 0.5). The 0.5 is added to the division to keep the value positive.

PAGE 153

142 Transform Vegetation Index (TVI) The transform vegetation index is nothing more than square root of NDVI. At times this particular index is shown with the same formula used in TNDVI transformations. It is important to note, howev er, that in this study, the two differ, if only slightly. Simple subtraction Vegetation Index (SVI) This is really just a simple band math expression (SVI = TM4-TM3) Devised Band Ratios In an effort to attempt to discriminate vegetative elements as much as possible, I introduced two additional devised ratios. Ratio 1 was dropped from subsequent analysis, however, due to its poor discrimination abilities. Ratio 1: TM2/TM4*TM3 Ratio 2: TM4*TM3/TM2 (an expansion on simple green ratio) Soil Adjusted Vegetation Index (SAVI) SAVI is the Soil Adjusted Vegetation Index that was introduced by Huete (1988). This index attempts to bridge ratio-bas ed indices and perpendicular indices by acknowledging that the isovegetation lines are not parallel, and that they do not all converge at a single point. The initial c onstruction of this index was based on measurements of cotton and range grass ca nopies with dark and light soil backgrounds, and the adjustment factor L was found by trial and error until a factor that gave equal vegetation index results for the dark and lig ht soils was found. The result is a ratiobased index where the point of convergence is not the origin. The convergence point ends up being in the quadrant of negativ e NIR and red values, which causes the

PAGE 154

143 isovegetation lines to be more parallel in th e region of positive NIR and red values than is the case for RVI, NDVI, and IPVI. Huete (1988) does present a theoretical basis for this index based on simple radiative transfer, so SAVI probably has one of the better theoretical backgrounds of the vegetation indices. However, the theoretical development gi ves a significantly different correction factor for a leaf area index of 1 (0.5) than resulted from the empirical development for the same leaf area index ( 0.75). The correction fact or was found to vary between zero for very high densities to one fo r very low densities. The standard value typically used in most applica tions is 0.5 that is for intermed iate vegetation densities. Modified Soil Adjusted Vegetation Index MSAVI2 is the second Modified Soil Adjusted Vegetation Index that was developed by Qi et al. (1994) as a recursion of MSAVI. Basi cally, they use an iterative process and substitute 1-MSAVI (n-1) as th e L factor in MSAVI (n). They then inductively solve the iteration where MSAVI (n) =MSAVI (n-1). In the process, the need to precalculate WDVI, NDVI, and the need to find the soil line are eliminated, hence the formula is a much easier index to both code and to implement. MSAVI2 = (1/2)*(2(NIR+1)-sqrt ((2*NIR+1)^2-8(NIR-red))) Tassel Cap Transformation The Tassel-Cap transformation was prim arily developed for and tested in agricultural applications of remote sensing data. Kauth and Thomas (1976) developed the tassel cap transformation originally for application to Landsat Multispectral Scanner (MSS) data. However, Crist and Cicone ( 1984) developed a similar transformation to TM data occupying the same spectral regions that Kauth and Thomas (1976) examined (Jensen 1996: 183). Given its utility in othe r agricultural settings, Tassel-Cap should be

PAGE 155

144 an effective means of transforming multi-spectral TM data into information more readily applicable to developing settlement modeling data. The tassel cap transformation can be described as a vegetation index, but mathematically it is a factor analysis. TM data are highly correlated permitting band ratio transformations like NDVI, but because of this, the effective dimensionality of TM data may be less than the total number of ba nds recorded (Crist and Cicone 1984: 334). Knowing that high correlations ex ist within the data indicates that factor transformations may be particularly effective at reducing di mensionality while maintaining variability. Tassel-Cap is derived from a rotation of principle components. However, the axes are rotated according to a set of coefficients. St andard uncalibrated coefficients are applied in this use of tassel cap (Crist and Cicone 1984, Jensen 1996: 185). If data are distributed into two perpendicular planes, th en PCA may not be effective at defining the actual planes of variation along which data resides (Crist and Cicone 1984: 345). If this is the case and factors are rotated, they ma y intersect more meaningful axes of variation describing the data. Tassel cap factors may not necessarily be perfectly orthogonal (Crist and Cicone 1984) . Laboratory and fi eld studies of agricultural canopies indicate that TM data is distributed along two roughly pe rpendicular planes with a transitional zone spanning between these two (Crist and Cicone 1984: 46). Fully vegetated test plots define a plane of vegeta tion while bare test pl ots define a plane of soil. The third transition pl ane, roughly forming a right triangle between vegetation and soil in feature space, was defined by data fr om partially vegetated plots containing both vegetation and soil.

PAGE 156

145 The plane of vegetation can be defined along two rotated axes of variation: greenness and brightness. The Greenness axis accounts for contrast that exists between near infrared and visible bands, while Brightness is a partial sum of data in all bands. Crist and Cicone's (1984: 347) experimental data indicate that over ti me re-vegetation of plots from "maximum vegetative developmen t (high Greenness) to maturity" produces movement primarily along the plane of vege tation rather than the transition zone. Given previous agricultural applications of the Tasse l-Cap transformation, one can make several predictions regarding Xinguano settlements. One would expect recently cleared plots of land to exhibit high values in the plane of soil. However, the relative brightness in this plane may be influenced by soil moisture. Recently planted gardens should fall along the transition plane, until th e garden canopy begins to close. Once garden canopy has closed blocking the soil su bstraight, pixels c overing these kinds of spaces should exhibit very high greenness and brightness values. On the other hand, climax forest should be characterized by pixels with low values in the plane of soil with high greenness value but the self-shading of c limax forest should produce relatively low brightness values compared to recent cleari ngs. This should permit a quick estimation about land usage and relative age of vegeta tion within the study ar ea, in addition to providing a means of discriminating diffe rent vegetation types across the region. Decorrelation Decorrelation stretching enhances the color separation of an image with significant band-band correlation. The exaggerated colors improve visual inte rpretation and make feature discrimination easier. Decorrelation stretch is ba sed on a principal component transformation of correlated multispectral image data. In general, highly correlated image channels, such as the Red, Green and Blue (RGB) channels in Thematic Mapper

PAGE 157

146 images, show subtle differences well, but th ese colors are not clearly related to the different surface types. One can utilize simple contrast exaggeration to expand the range of intensities of highly corre lated images, but contrast ex aggeration does little to expand the range of colors. To enha nce the color in highly correlate d images requires a selective exaggeration the least correlated portion of the spectral data (that is, one must decrease the correlation). Decreasing the correlation of spectral da ta corresponds to exaggerating the color saturation without changing th e distribution of hues (or relative color composition). The decorrelation stretch process involv es three fundamental steps: First, a principal-component transformation is appl ied with the rows and columns of the eigenvector matrix transposed. Second, contra st equalization is applied by a Gaussian stretch, so that histograms of all principal components approximate a Gaussian distribution of a specified vari ance. Third, a coordinate transformation that is the inverse of the principal component rotation is applie d so that the data are projected in their original spectral channels, using eigenv ectors as weightings for each principal component. This inverse operation maximizes the spectral separability of different surface types in the restored sp ectral channels. The decorrel ation stretched images that are created by this process can also be us ed as components for making color composites. Principal Components Analysis The fact that data are strongly correlat ed in more than one band (Sabins 1987: 261) allows for application of ratio transf ormations like the NDVI. However, these strong correlations, either positive or negativ e, create redundancy in multivariate data. Principal components transformations are a common means of reducing dimensionality in multivariate data. PCA has been widely appl ied and well discussed in the analysis of

PAGE 158

147 multispectral satellite data as it permits the capture of variability in multivariate measurements while at the same time reducing dimensionality. Defining new composite axes, PCA provide s a basis for investigating the primary sources of input variation in multivariate or multispectral data (Baxter 1994: 48; Jensen 1996: 172). In this application of PCA eige nvalues and eigenvectors for the components Table 6-1. The PCA statistics for 2002 (August) transformation Band Min Max Mean SD 1 -0.014 0.439 0.059 0.050 2 -0.016 0.489 0.046 0.044 3 -0.015 0.443 0.036 0.043 4 -0.022 1.039 0.159 0.127 5 -0.020 0.617 0.101 0.099 6 -0.019 0.605 0.043 0.061 Eigen Values (% variance) 1 0.030 2 0.006 3 0.000 4 0.000 5 0.000 6 0.000 Eigenvectors Eigenvec Band 1 Band 2 Band 3 Band 4 Band 5 Band 6 1 0.278 0.246 0.223 0.676 0.522 0.289 2 0.041 -0.022 -0.210 0.674 -0.550 -0.500 3 0.597 0.484 0.405 -0.229 -0.439 0.032 4 0.449 0.104 -0.341 -0.189 0.462 -0.650 5 0.438 -0.142 -0.690 -0.012 -0.153 0.537 6 0.415 -0.821 0.387 0.027 -0.003 -0.058 are calculated from a correlation matrix ra ther than a covariance matrix. Using a correlation matrix produces standardized PC transformations that are more readily applied to discrimination and change detec tion problems (Jensen 1997: 179; Siljestroem Ribed and Lopez 1995). By creating new comp osite axes (PC), transformation of raw multispectral data may produce results that are interpreted more easily than raw imagery

PAGE 159

148 (Jensen 1996: 172.) For TM data, PC1 generally describes the vast majority of variability in the measurements, with each band successively describing less variability between input bands. Examination of eigen values and eigen vectors demonstrates that component one explains an enormous amount of the variation in the 2002 image (72.5%) with component 2 adding an additional 13.5% and PC3 a mere 1%. In PC1, eigen vectors for all the bands of the image data are positively correlated. PC2 shows negative correlation in the eigen vectors for bands 2, 3, 5, and 6. Table 6-2. The PCA statistics for 2003 (May) transformation. Band Min Max Mean SD 1 -0.014 0.436 0.055 0.046 2 -0.016 0.479 0.042 0.040 3 -0.015 0.441 0.030 0.034 4 -0.022 1.033 0.156 0.121 5 -0.020 0.613 0.093 0.086 6 -0.020 0.601 0.037 0.049 Eigen Values (% variance) 1 0.027 2 0.002 3 0.000 4 0.000 5 0.000 6 0.000 Eigenvectors Eigenvec Band 1 Band 2 Band 3 Band 4 Band 5 Band 6 1 0.269 0.237 0.194 0.722 0.494 0.259 2 0.083 -0.014 -0.197 0.606 -0.593 -0.485 3 0.639 0.486 0.387 -0.269 -0.362 0.044 4 0.451 0.049 -0.420 -0.192 0.486 -0.587 5 0.427 -0.262 -0.612 0.014 -0.192 0.581 6 0.355 -0.798 0.472 0.023 0.029 -0.115 This correlation means that when measur ed reflectance is high in, for example, band 3, it is very likely to be low in ba nd 4. Likewise, when band 4 registers a high digital number band 3 is likely to be low. If band 3 records reflectance wavelengths corresponding to red light (known to be an important absorption band for green

PAGE 160

149 vegetation), and band 4 is attributed to a portion of the near-infrared spectrum (also highly reflective in vegetation), then PC2 s hould be showing primarily differences in vegetation coverage. PC3 eigen vectors show that TM bands 4 and 5 are negatively correlated to all other bands. Bands 5 (eigen vector = 0.439) and 1 (eigen vector = 0.597) have the strongest negative correlations. Band 1 r ecords reflectance in a section of blue wavelengths of light corresponding to peak transmittance frequencies for water. Therefore, rivers and atmospheric moistu re should be highly re flective in band 1. Conversely, band 5 is sensitive to the amount of moisture in green plants. These relationships between bands 1 and 5 indicate that PC3 is primarily relating information of differences of moisture content. Principle component analysis of the 2003 image lik ewise revealed th at the majority of the variance between the various bands is explained primarily by the first three PC bands (90%, 7.7%, and 1.3% respectively). Again, PC1 possessed positive correlation across all bands. PC2 once more demonstrated negative correlation in bands 2, 3, 5, and 6. Finally, PC3 once again showed negative co rrelation in bands 4 and 5, with bands 1 and 5 representing the great est negative correlation. Revised Methodology After combing the available literature for a possible so lution, I decided that some adjustments were necessary to achiev e a higher accuracy for the final LULC classifications. First, it was obvious that the spectral sim ilarity of the classes I had chosen would not permit an accurate classifi cation of the original calibrated imagery. Second, the emphasis on vegetative separabili ty provided by the NDVI and Tassled Cap transforms came at the cost of data reduction, and thus had a direct impact on the ability

PAGE 161

150 of the classifier to make fine distinctions between different classes. What was needed was an expansion of the number of data se ts that provided information about the minute differences between the va rious types of vegetation. First, I made a distinction between th e five vegetation indices without a soil reflectance control (NDVI, Tassled Cap Green ess, TNDVI, SVI, and TVI) and the two indices with a control for soil (SAVI and MSAV I). The first five were stacked together to form VI composite images for 2002 and for 2003. I then produced a PCA transform of these 5 indices to reduce the amount of da ta. SAVI and MSAVI were layered and transformed by PCA into two bands as wells. As previously mentioned, PCA is often used to “compress” data sets and reduce the amount of data that must be considered. I chose PC1 and PC2 from the vegetation indices composite, as well as PC1 of the soil -adjusted indices (for this study the first PC acquired from the 2 soil indices contains spectral information adequate for the classification to normalize the effects that emer ge due to the different soil types of the areas with low canopy of vegetation). I laye red these bands with PC2 and PC3 from the original calibrated images. As previous ly explained, an examination of principal components eigenvector loadings was undert aken to determine which PC possessed information that could be related directly to the spectral signatures of vegetation. Eigenvector loadings for PC2 indicated that PC2 described the difference between the visible channels TM2 and TM3, and the infrared (IR) channels TM5 and TM6. Eigenvector loadings for PC3 indicated that TM3 and TM4 were negatively correlated owing to absorption of chlorophyll in TM3 and high reflectance in TM4. Operating under an informed assumption that PC2 and PC3 were, in fact, describing differences in

PAGE 162

151 vegetative cover, PC2 and PC3 were selected as for inclusion in the final analysis composite. Decorrelation stretch results for bands 3 and 4 from the original calibrated images rounded out the final two bands of info rmation used in the re vised classification. Even though these bands still show the pr operties of the origin al bands, the color separation of these bands are enhanced w ith significant band to band correlation. A decrease in correlation of sp ectral data will result in an exaggeration of the color saturation without changing the distributi on of hues (or relative color composition) (Gillespie et al. 1987). In the final analysis of the data, an assumption was made that by selecting PC1 and PC2 of vegetation indices, PC1 of soil indices, PC2 and PC3 of raw bands and DC3 and DC4 as the primary bands for analysis and classification, one could effectively remove much of the redunda nt data among the multivariate datasets, maximizing the potential of the classification algorithm to delineat e between vegetative cover classes with more accuracy. Landscape attributes observed during da ta collection provided much of the information used in refining the classificat ion, producing “theme” la yers re-incorporated into the GIS, which was then used to produ ce a landscape attribute informed descriptive model of the sample area. From this de scriptive template, areas matching specific spectral profiles are extracted, allow us to model where uninvestigated or unknown habitation sites might be located. These areas are treated as potentia lly containing highly visible archaeological remains. The importan ce of such efforts should be apparent when compared to traditional sampling strategies, especially in reference to the relative efficiency of traditional methods in data co llection over such large spatial extents. Developed to predict archaeological site yi eld potential on non-su rveyed areas, this

PAGE 163

152 approach first derives descriptive spectr al attributes of know n “sites” and their encapsulating context, then extrapolates and the template to surrounding areas sharing the same attributes.

PAGE 164

153 CHAPTER 7 RESULTS Post-Classification Procedures and GIS Manipulation Supervised image classification is essentially a three-stage process. First, a number of training pixels, representati ve of their respective classes, are located in the image under consideration. These training pixels are used to calculate descriptive statistics for each class, thus defining a generalized sp ectral signature corresponding to each selected category. Second, based on the derived class desc riptions, each pixel is then allocated to the class with which it has the greatest similar ity, as assessed relative to the classifier’s decision rules. For supervised classifi cations using a standard maximum likelihood classification algorithm, this process labels each pixel as belonging to the class with which it has the highest poster ior probability of membership (Lillesand and Kiefer 2000, Campbell 1996). Finally, the accuracy of the classified image is assessed with respect to set of pixels for which reference or ground da ta on class membership is available. Testing of the classification is usually ba sed on an error matrix, demonstrating the correspondence between the pred icted and the actual classes of membership for an independent testing set, and fr om which it is possible to de rive a range of quantitative measures of classification accuracy. The e nd result of the classification process is effectively a thematic map depicting the spatial distribution of the selected classes accompanied by an accuracy statement. The acc uracy of the classification is dependant on a number of variables, related to both th e methodological approach selected and to the nature of the classes and remotely sensed data itself (Campbell 1996).

PAGE 165

154 The supervised classification used in this study remains conventional in all aspects, only the methodological approach differs from tradition. Thus, the output for each pixel subjected to classification comprises only th e code of the class with which it has the highest strength of membership. Often refe rred to as a “hard” or “crisp” form of classification (based on conventional cris p set theory), each pixel has a single membership in a mutually exclusive classifi cation scheme (full membership to the named class, and zero membership to other classe s). The training stage of the classification should, ideally, also be repres entative of conventional cris p set theory, with training pixels selected such that they belong (or ar e assumed to belong) to their named classes with full membership and have zero membership to other classes. However, in the case of this study, the nature of the classes under investigati on (primarily those that are cultural in nature) have no firm, defined boundari es, and thus are not easily incorporated into such a theoretical mode. There are other methodologies for dealing with pixel membership (including fuzzy-set theory and lo gistic regression), but this dissertation did not address the application of these methods to this specific problem (leaving these other treatments as objectives in future work). Instead, GPS survey coupled with ethnographic information were used to select training sites, at times unavoi dably containing mixed spectra. Conventional classification me thodologies of remotely sensed imagery assume that the area under investigation is composed of a number of unique, internally homogeneous classes that remain mutually exclusive, thus any classification based on remotely sensed data and ancillary data can be used to iden tify these classes with the aid of ground data (Townshend 1981, Lillesand and Kiefer 2000). It is certainly th e case that such

PAGE 166

155 assumptions are often invalid in areas where the classes exist as continua rather than as a mosaic of discrete classes (for example, vege tation coverages, which are rarely internally homogeneous and mutually exclusive). As a result, the classes overlap and are not separated by sharp boundaries (Wood and Foody 1993, Kent et al. 1997). Allowances must also be made for the complex relations hip between spectral re sponses recorded by a remote sensor and the corresponding ground si tuations. Similar entities at different locations may possibly exhibit varied spect ral responses, and, conversely, completely dissimilar entities may exhibit very sim ilar spectral responses (Forster 1983). To perform a conventional classification, then, requires an acknowledgement of a number of fundament sources of error. As addr essed later in this chap ter, it is a balancing act between the accuracy of the classification, and the quality of the classification, necessitating that allowances be made in so me cases for the sake of cost, expediency, ease of interpretation or repli cability, and always mindful of what goals the end-user of the final classification product has in mind. In that spirit, no sub-pi xel classification was undertaken, though it is definitely a technique that would be quite valuable to future offshoots of this research. For the scope of this dissertation, however, pixels were assumed to have belonged exclusively to one class. It should be re-iterated that membership to a specific classificatory cat egory was further simplified by the combining of classes into more useful entities for aidi ng in the discussion of probable site locations, and, more broadly, the distribution of an thropogenic soils resulting from long-term human occupation and interaction within the region of interest. Each of the classes was then converted into a vector format by the ENVI software package. Each class represented a single la yer of data, and each layer was ported into a

PAGE 167

156 GIS. The layers were then stacked on t op of one another to d uplicate the original classified image, but with each class broken into its own dataset. Figure 7-1. Detail of the 2002 (Augus t) supervised classification

PAGE 168

157 Figure 7-2. Detail of the 2003 (M ay) supervised classification

PAGE 169

158 Table 7-3. The 2002 (August) accuracy results for combined classes Class Prod. Acc.(%) User Acc.(%) Prod. Acc.(pixels) User Acc.(pixels) Water Body 99.64 99.94 10577/10615 10577/10583 Forested 70.24 62.86 347/494 347/552 Savannah 91.44 85.85 267/292 267/311 Inlet Regions 22.30 35.98 68/305 68/189 Transitional Vegetation 41.74 46.60 48/115 48/103 Bare Soils 96.15 43.86 25/26 25/57 Aldea 96.36 92.98 53/55 53/57 Pasture Outside Park 99.18 100.00 2528/2549 2528/2528 Culturally Active/ Rosas 86.07 97.19 346/402 346/356 Anthropogenic vegetation 78.45 40.63 91/116 91/224 Actively cultivated sites 40.85 61.70 58/142 58/94 Table 7-4. The 2003 (May) accuracy results for combined classes Class Prod. Acc.(%) User Acc.(%) Prod. Acc.(pixels) User Acc.(pixels) Water Body 99.16 99.73 10526/10615 10526/10555 Forested 67.81 56.49 335/494 335/593 Savannah 77.05 87.89 225/292 225/256 Inlet Regions 38.36 68.02 117/305 117/172 Transitional Vegetation 41.74 30.77 48/115 48/156 Bare Soils 100.00 42.62 26/26 26/61 Aldea 96.77 21.13 30/31 30/142 Pasture Outside Park 78.82 99.74 774/982 774/776 Culturally Active/ Rosas 81.59 98.20 328/402 328/334 Anthropogenic vegetation 75.86 30.14 88/116 88/292 Actively cultivated sites 6.34 75.00 9/142 9/12 Utilizing the maximum likelihood classifier once again in conjunction with the ground-truthed data set, I was able to pr oduce classified images with much higher

PAGE 170

159 accuracy overall than in any previous iteration. For 2002, the confusion matrix reported an overall accuracy of 86 .1326% with a Kappa Coefficient of 0.7949. For 2003, the confusion matrix reported an overall accura cy of 82.7279% and a Kappa Coefficient of 0.7457. The original 30 classes were then combined into more manageable categories. The final combined classification of the 2002 a nd 2003 images consisted of 12 categories. The confusion matrix for the combined categ ories of the 2002 classification reported an overall accuracy of 95.3478% with a Kappa Coefficient of 0.9025. For the 2003 combined classification categories, the confus ion matrix reported an overall accuracy of 92.500% with a Kappa Coefficient of 0.8024. Perhaps more interesting than the strengt h of the model based on measurements of overall accuracy are the achievements of the mo del within these specific classes denoting probable site locations. This di ssertation is specifically interested in vegetative signatures of anthropogenic soils in the region of study. To that end, the classes of most interest are “Anthropogenic vegetation” and “Actively cul tivated sites.” The combination of new, naturally occurring succ essional species, in addition to the Xinguano cultural practice of reusing past occupational sites as garden (due primarily to the richness of the anthrosols contained in such areas), the recent burning than accompanies new plantings, and the presence of older, possibly anthropogenic, vegetative classes w ould create an overall spectral target that would be di fficult, at best, to classify in to a discrete category, thus it was also deemed appropriate to maintain separate classificatio ns of probable site locations based upon the level of contemporar y modification of those regions. The active cultivation of former occupational site s is bound to produce highly mixed spectral

PAGE 171

160 responses, and thus represents the underl ying cause of the rela tively low accuracy exemplified by the “Actively cultivated sites” category. The results for the category of “Anthropogenic vegetation” are much more promising, demonstrating a better than 75% accuracy on the producer side of the classifi cation. Based upon these results, I feel confident in stating that the methodology used in this study has a grea t deal of potential for extracting probable historic and prehis toric Xinguano occupati onal sites from the surrounding vegetation, so long as they are not under active cultivation or otherwise in the process of cultural modifi cation (burning, clearing, etc.). The rise in accuracy utilizing composite im ages derived from the application of the methodology described above suggests that th is process, undertaken to maximize the information contained in the imagery while minimizing redundant data, allowed for the expression of data substantive enough that th e classifier could succe ssfully discriminate between vegetation classes that initially po ssessed very similar spectral reflectance values. The results of the classifica tions suggest that Principle Component transformations of both raw imagery bands and a sequence of vegetation indices can extract valuable vegetation cove rage information, and distill it into a concentrated form by creating a new variable set, eliminating much of the interband correlation, and greatly reducing the dimensionality of the data. The two classes of particular interest to this study (Actively cultivated sites, Anthropogenic vegetation), representing vegeta tion located in the an thropogenic soils of former occupation sites, were unified into si ngle “predicted sites” layers by year, and exported as shapefiles. The shapefiles were then overlain on the original images, and

PAGE 172

161 layered with the GPS surveyed inventory of known site locations and their extents for comparison. Figure 7-3. Detail of predicte d site locations (PC1 of ve getation indices composite, PC1 of soil indices composite, PC2 of Landsat 7, August 2002)

PAGE 173

162 This “tethering” of GPS-survey data to geo-re ctified satellite images allows a researcher to extract the exact location of classified features, as well as providing the ability at some future point to expand on this study and perhap s do further analysis of the relationships of features to other environmental and cultural variables. Due to the relatively low predictive value of the “Actively cultivated sites” layer, it was discarded from a final assessment of pr edicted site locations. Instead, this study relied upon the “Anthropogenic vegetation” la yers for 2002 and 2003. These layers were overlain on top of one another. An intersec tion of the two layers was performed (Figure 7-3), insuring that only those regions th at had been categorized as “Anthropogenic vegetation” in both the 2002 and 2003 scenes were utilized in the last stage of analysis. Analysis of this final layer suggest ed that there are approximately 1800km2 of vegetation within this region that exhibit spectral values similar to that of the vegetation within known occupation sites that are not currently under cultivation. By utilizing this final layer, one can readily assess areas of part icular interest for future archaeological exploration based upon the size, density, and patt ern of vegetation within these regions of interest, and, given the fact that these regi ons are geo-rectified, one can readily navigate to them using GPS. This final unification in GIS enabled a visualization of probable site locations in relation to survey ed feature components, and th us, a ready iden tification of those areas that hold the most potential of containing residua l occupational or cultural modified soil components. The predictive value of this methodol ogy relies on the differential organic components of anthropogenic soils, and the ma nifestation of those differences in the

PAGE 174

163 types of vegetation that grow in those soils . These culturally modified soils promote distinctive species of vegetation to thrive in successional stages , distinguishing these areas from surrounding vegetative communitie s. This, coupled with the Xinguano practice of transplanting or cultivating cultu rally significant species in and around occupational sites (a practice that continues today as it did in the past), allows researchers to characterize these areas as vastly differe nt from surrounding environments. While this is not an absolute model in any sense of the word, the course of these investigations has laid open a number of intere sting possibilities. A unified approach, utilizing image processing, GIS, and GPS survey is a viab le one. Each of these components has complemented the capabilities of the others in measurable ways. These investigations have also shown that it is possible, with some degree of accuracy, to separate out distinctive vegetative signatures over extrem ely large spatial extents. While both the methodological approach and the results of these analyses are hardly groundbreaking within geographical sciences or within the remote sensing community, the implications of this study for an integrative approach avai lable to archaeologist s are of substantial import. This study has demonstrated that, no t only is GPS capable of accurate ground survey, it may actually be a preferred method of collecting survey data if those data are then to be processed within a GIS or image processing environment. The use of numerous vegetative indices in combination with decorrelation and principle component analysis is an approach that allows us to sp ecifically target varia tions within vegetation, and provide a sound platform upon which to pe rform informed classification based upon data collected in the field. This has fa r-reaching implications within the archaeological community, especially in expanding the methodological approaches to designing

PAGE 175

164 predictive models, which seem to remain tied to geomorphological characteristics (proximity to water, elevation, and soil type s) that themselves ar e highly subject to temporal transformation and are dependent, of ten times, on the accuracy of the original classifier, as well as the digitizer of these in formation layers. The procedures laid out in this study show that, rather th an remain dependant on scarce or expensive data sets, there are viable alternatives to ga ining an understanding of the human dimension as it plays out across a landscape in a timely, cost-effective way. Discussion of the Classification Process Map products derived from remote sensing are usually critical components of a GIS. Remote sensing is an important t echnique to study both spatial and temporal phenomena (monitoring). Through the analysis of remotely sensed data, one can derive different types of information that can be comb ined with other spatial data within a GIS. The integration of the two technologies crea tes a synergy in which the GIS improves the ability to extract information from remotely se nsed data, and remote sensing in turn keeps the GIS up-to-date with actual environmental information. As a result, large amounts of spatial data can now be integrated and analy zed. This allows for better understanding of ecological processes and better insight in the effect of human activities. GIS and remote sensing can thus help people to take informed decisions about their environment. Like in all models, however, both maps and thematic da ta are abstractions or simplifications of the real world. Therefore, GIS and remote sensing can complement but never completely replace field observations. GPS survey data with was exported to a GIS for thematic development. Without direct observations in the field, the classification categories extracted from the imagery using regions of interest derived from thematic classes designated in the GIS environment would not have been possible. The lack of overall

PAGE 176

165 separation of vegetative classes in the study area necessitated extens ive field survey and categorization by both visual in spection and with the help of indigenous informants. Again, this requires redressing the issue of classificatory ac curacy. In this analysis, (as is the case in most current research), a confusion matrix was employed to estimate the efficacy of the classifier algorithm in appr oximating real-world distinctions between various coverage classes. Using a confus ion matrix to estimate accuracy is a sound approach. However, obtaining a reliable conf usion matrix continues to prove the most critical part of accuracy assessment. Pote ntial problems are the subjectivity inevitably induced by the choice of the classification sc heme (labels), the tr aining samples (in the case of supervised classification), and the reference data sampli ng size and strategy. When the classes are well separated in a feature space and there is no overlap between the distributions of th e categories, most classifiers should return the same result, which will hardly depend on the choice of training sample. When the choice of classes produces overlapping spectra or spectra that are multimodal, the classifiers may disagree, depending on the a priori information that is incorporated into the classification models. Moreover, even a given classifier can produce different results when trained with a different data set. In its purest form, a classifier is simply an automated method of pattern recognition. Pattern recognition is defined as a process wherein a form of decision rule is applied to produce a categorical identification sequence that assigns each object to a discrete class (I use pattern id entification here inte rchangeably with pattern classification, or, more simply classification). In image pr ocessing software platforms, such as the one

PAGE 177

166 used in this analysis, the process of classi fication utilizes the co sen algorithm to label each pixel composing the image as representing particular ground cover types, or classes. There are inherent issues when dealing w ith classifications of remotely sensed images. The fact remains that the real-wor ld components of the sa tellite image does not, generally consist of neatly spaced, homogene ous parcels of land. On must also accept that no natural or man-made environment is c onstructed to fit the raster model of data collection used by many sensor platforms. Se ldom can we expect true ground conditions to be accurately modeled by remote sensing platforms. The implication of these fundamental truths of image processing nece ssitates an acknowledgement on the part of the researcher that, to some degree, there mu st be a corruption of the class signatures. Rigorous methodologies have been applied to th ese issues, resulting in claims that these effects can be accounted for, modeled, or reduced using various procedures (linear mixture models, fuzzy membership functions). As observed earlier, however, some level of generalization is necessary, and for the deve lopment of this analysis, it is assumed that each pixel belongs to one class only, accepting a certain degree of error, and realizing that this may have implications for the appl icability of coarse da ta (e.g., Landsat data). As mentioned previously, selected features should be chosen by optimizing a criterion, estimated from the ground-truthed trai ning data. For each category used in the classification, it is critical to question how we ll the data represent the class overall. In essence, do the selected regions adequately sa mple the feature space for each class? The Issues of mislabeling of pixel elements are coupled to the “representativeness” of the training data and are a direct result of issues related to mixed pixels (class overlap), transition zones, dynamic zones, within-cla ss variability (covariance), limited training

PAGE 178

167 data, resolution issues (as well as any number of uncounted variables) since classification is directly impacted by affected by the s cale as well as the spatial and spectral characteristics of the image data . In terms of spa tial resolution, if the size of a specific type of object imaged is smaller than the instantaneous field of view (IFOV) of the sensor, the resulting pixel will contain other types of objects as well, thus giving rise to the problem of mixed pixels. Th e separability issues I encounte red are, in part, a result of this mixed pixel effect, as well as the other issues mentioned above. In terms of spectral characteristics, energy transfer from ne ighboring pixels results in mixed spectral signatures within a single pixel. Land us e/cover datasets are, therefore, spectrally ambiguous within the measur able spectral resolution. Remote sensing research has focused c onsiderable time and energy towards finding empirical relationships between vegetation indices measured in situ and local spectral reflectance in the hopes of defining the rela tionship between observable remote sensed data and actual biophysical properties of ve getative matter. By realizing an empirical relationship between vegetation indices and spec tral reflectance, scie ntists have provided a means to extrapolate these relationships to whole landscapes. For example, NDVI has been correlated to vegetation health, biom ass, LAI, productivity, and fractional ground cover among many physical properties. Unfortuna tely, all of these de rived relationships are site specific. Often times, other aspect s of the observed landscap e (soil reflectance, reflectance of a thick vegetation canopy, and the attenuation coefficient for radiation in the canopy that affect a given pr edicted reflectance), directly impact measured reflectance values, making these empirical relationships difficult to extrapolate across regions. Perhaps this, then, is the greatest difficulty faced in this analysis. Fieldwork was

PAGE 179

168 restricted to a narrowly defined geographi c portion of the region. The ground-truthed data collected within this relatively small portion of the overall scenes’ extents means that narrowly defined spectra signature in formation was extrapolated across a much larger spatial extent. While certainly not the optimal method of ensuring a highly accurate land cover classification, this approa ch was necessitated by the larger goals of this project. The most difficult part of dealing with thes e very issues is that they are so difficult to quantify. We can attempt to reduce classi fication error by utiliz ing carefully defined classes and increasing the number of classes. It is well established that there is a strong relationship between the number of classes, the optimum number of features and the size of the training set. When the ratio of th e number of training samples to the number of feature measurements is small, the estimates of the discriminant functions are inevitably reduced, which may, in turn, influence the qua lity of the result. We often times bandy about the term “accuracy,” as if one could obj ectively determine the relation between the spectral signature of a single element in a re mote-sensed image and a specific category. This is valid only in select cases in which the physical para meters used to describe the training data directly corres pond to the physical parameters sensed by the remote-sensing instrument. In most cases this relation is, to say the least, tenuous. It would be more profitable to speak in terms of map signature consistency rather than speak of “accuracy” would be more appropriate to talk about the relative accu racy of the classification. The use of ground-truthed data for both the tr aining and validation of the classification remains one of the most critical aspects of the methodology presented in this research,

PAGE 180

169 and should be seen as an integral part of the accuracy assessment of an image classification method. A common method for classification accuracy assessment is the error matrix. The error matrix compares the relationships be tween ground-truth data (reference data) and classified results category-by-category. From the error matrix, some important measures can be derived, such as overall accuracy, pr oducer's accuracy, and user's accuracy. Many works have provided the meanings and calculation methods for these measures (Congalton 1991; Richards 1993; Janssen and Wel 1994; Campbell 1996; Jensen 1996). Another method to interpret the classification ac curacy is to calculate Kappa coefficients (Ma and Redmond 1995; Jensen 1996; Kalkahan et al. 1997). It measures the difference between the agreement between reference data and classification results and the chance of agreement between the reference data and a random classifier. The Kappa coefficient is computed as: r i i i r i r i i iix x N x x x N1 2 1) ( ) ( where r is the number of rows in the error ma trix, is the number of observations in row i and column i in the error matrix (i.e., th e corrected classified number), and are the marginal total in row i and column i respectively, and N is the total number of observations included in the error matrix. I have provided the confusion matrices, in their entirety, in th e Appendix of this study. The confusion matrix is a square array of numbers, laid out in rows and columns,

PAGE 181

170 which express the number of sample units assign ed to a particular category relative to the actual category as verified by ground-truthed information or reference data set. The importance of the confusion matrix is that it permits both descri ptive and analytical statistics to be calculated. Descriptive techniques are relatively simple and include computation of the overall accuracy (division of the total correct by the total number of units) and the individual class accuracy. The latter measure can be expressed in two ways: by calculating the producer’s accuracy, wh ich is based on the reference data (error of omission), or the user’s accuracy, based on th e total number of pixels classified within a specific category (commission e rror). Analytical statistical calculations are useful for comparing different classifi cation methods. Previously mentioned, the most common type of discrete multi-variate techniques for statistical tests on the classification accuracy is termed “Kappa analysis.” The estimate of kappa (also called the KHAT statistic) gives a measure that indicates if the confusion ma trix is significantly di fferent from a random result. The kappa analysis also can be used to compare different matrices from different classifiers and to determine if one result is significantl y better than the other. As has been previously stated and totally contrary to logic, the term “accuracy” is not only hard to quantify, but also hard to define, and the same holds for quality. To some, the capability of the methodology presente d here to accurately identify areas where Xinguano occupation sites lie might seem lacklu ster. The relatively high accuracy of the classifier for finding pixels belonging to the “Anthropogenic vegetation” may indeed be offset by the extremely low predictive valu e of the process for finding pixels in the “Actively cultivated sites” category. However, the quality of the classification must be defined in a completely different light. The difficulty of the task at hand, considering the

PAGE 182

171 relatively small region from which samples were taken, and the spatial extent to which those training data were then extrapolated certainly had dramatic effects on the efficacy of this analysis. However, the met hodology by which those results were obtained highlights several key features of interest to archaeologists in particular. First, the computational complexity of the methodologi cal process, while high, is by no means extreme. The indices utilized in this st udy, while simple, remain tried and test as effective means of discriminating among the t ypes of coverage clas ses under scrutiny. Second, the descriptional comp lexity of the generated results has been kept to a minimum. Thus, I tether the term accuracy to quality, well aware that I cannot do this except by creating a link between the accuracy, and the objectives and requirement of the end users of such a methodology. Make caref ul note that the term “users” does not necessarily need relate to those who perform image processing and manipulation, but rather, the end user of the fi nal product, be they field arch aeologists, cultural resource managers, ethnologists, historians , or representative of any num ber of other disciplines. Thus, the quality of the classi fication is rather subjective, tied to a specific user’s objective and their requirements. Remote-sensing image interpretation has a wide range of uses across numerous disciplines, and in any discussion of the resu lts of this process and these analyses, it would be more appropriate to indicate how the results come to bear on a specific problem. This methodology was designed with a Neotropical archaeologist in mind, one who has a limited budget and must quickly target areas of interest for the sake of cost and expediency. With these goals in mind, I have developed a system that 1) has shown an overall accuracy over the course of two years of better than 90% for the combined land

PAGE 183

172 cover classes chosen for the study; 2) there is a relatively high probability that areas identified through this classification as bei ng past occupational sites of Xinguano peoples are, in fact, actual site locations; 3) that the classification has been effectively applied over a large spatial extent, leaving open th e question of how far the results of this methodology might be extrapolated or if it might have import in other geographic regions; 4) that the process re plicated by other researches and produce the same results; 5) that the classes chosen for this study pe rmit vegetation and other land cover classes to be used as surrogates for other types of activity (deforestation studies, comparison of Xinguano cultivation and its impacts on the rege nerative capacity of cleared field areas versus the impacts of ranchers on the periphery of the park); 6) that the classification system can be used during both wet and dry seasons (with drier periods producing slightly better overall results); 7) that the categories may be aggregated into a more generalized classification scheme or separated into more dis tinct categories; and 8) that multiple types of land uses can be recognized. Predictive modeling has become critical as a means of identifying landscape variables that are consistently correlated with known site distributi ons. By identifying these correlates, researchers are better able to identifying uninvestigated localities that have a high probability of containing site s based upon their geographic similarity to known sites. There is a dange r in this, however. Simply identifying new site locations based upon the attributes of known site locatio ns is not really making progress in the investigation of the unknown. Instead, we are simply investigating more of the same sorts of sites, with the added bonus of identif ying areas that have not yet been surveyed (we are merely modeling existing assumptions and expectations). This is not to say that

PAGE 184

173 such undertakings are any less important than other pursuits, but I submit that we, as archaeologists, can take it one step farther. For us to make predictions of the unknown, we must step outside what is “expected” and employ a modeling rationale that does not build exclusivity into its results. Thus, the underlying flaw in correlation models is exposed. Such models are exceedingly good at illustrating the probable location of any number of like sites based upon an approximate “type,” but without a more substantial theoretical foundation, they cannot be expect ed to produce information governing “why” or “how” such sites came to be. Here is wher e the integration of r eal, ground-truthed data and true ethno archaeological re search really comes into pla y. In the case of the Upper Xingu project, the Kuikuru themselves have be en instrumental in helping us to identify the locations of past habitation areas. Ra ther than simply seek ing correlations of brightness values of remote sensed images to specific environmental features, this work aims to establish a deeper unde rstanding of the sorts of vari ables critical in settlement location planning of current resident populations to make informed inferences about the activities of related groups in the past.

PAGE 185

174 CHAPTER 8 DISCUSSION This study has focused on a few key themes within the Upper Xingu region, beginning with a concentration on the structur e of the region landscape, and some insight into how those structures ma y have changed under the tenure of Xinguano settlement. It attempts to specifically develop a means that can be utilized to pred ict site locations, and, thus, primary areas of anthr opological alteration. The ecol ogical concept of landscape as a product of human/environment interaction ha s already been discussed at some length. This research has focused on the Xinguano la ndscape as a complicated mosaic of patches or ecosystems relevant to the phenomenon of distribution of prehistoric and historic Xinguano settlements. Building on findings desc ribed in the last chapter, land cover classification is used as a proxy for lands cape transformation to understand how this region has evolved from a complex mosaic of human-induced transformations of vegetative content in AD 1500 to a less disturbed environment covered by closed forests. Landscape structure is defined by the spat ial relationships among ecosystems, and landscape function is related to the interactions among the spa tial elements (i.e., flows of energy, materials, and species). Landscape cha nge is the alteration in the structure and function of the ecological mosaic over time (Forman and Godron 1986). This research has not focused on the function of the Xi nguano landscape or most of the vegetative classes defined for the study area, although its fi ndings can be used for this purpose in the future.

PAGE 186

175 Summary and Conclusions of Methods Scale is the temporal or spatial dimens ion of an object or process, characterized by both grain and extent. Resolution is the precision of measurem ent (grain size, if spatial). Grain is the finest level of spatial resolution possible with a given data set (pixel size for raster data). Extent is the size of the study area or the duration of time under consideration. Each of these parameters wa s kept under control to allow for an accurate analysis of landscape structure. In genera l, spatial aggregation tends to reduce the variation in spatial mosaics (Burt and Barb er 1996). The qualitative and quantitative changes in measurements across spatial scal es differ depending on how scale is defined. Therefore, measurements carried out at diffe rent scales may not be comparable. In addition, the exact relationship between cl asses varies across landscapes, creating difficulties in extrapolating from one regi on to another (Meentemeyer and Box 1987). Diversity, for example, will decrease with increasing grain size, with the more rare classes becoming lost as grain becomes coarse r, and dispersed classes dropping off at a must faster rate than more aggregated ones. Each coverage was classified through the lens of Landsat TM images, using geometri cally and radiometrical ly corrected images from the same geographic region, separated te mporally by a mere nine months. Thus, grain size is equivalent. The extent of each landscape was defined by the extent of the scene, as also mentioned before. This analysis hinges on determining a quant itative methodology that can be utilized to analyze and describe the st ructure of landscapes. Since much of predictive modeling is still dominated by empirical approaches and cas e studies, there exists a pressing need to create some sort of standardized approac h. A multi-temporal approach was used to characterize landscape in the study area, owing to the two years over which survey and

PAGE 187

176 other fieldwork took place. The recoded classes were based on the main processes occurring in the study area and affecting landscape transformation, primarily vegetation recovery through succession, and land occupation through agriculture conversion. The delimitation of landscape boundaries for calculati on of metrics is centr al, however, as has been discussed, a major issue involved with the calculation of any metric associated with theses classifications would need to address some fundamental sources of error, the most important of which the subjective nature of deciding where settlement boundaries can be delineated. Survey would seem to suggest th at no hard and fast boundaries apply in the Xinguano cases. Transitions of vegetative clas ses are much more gradual, clines rather that abrupt shifts in composition. The aim of this research was to develop a methodology useful for rapidly assessing a given area for Xinguano settlements using mu ltispectral remotely sensed data. While this model could certainly be applied to locating current settlements, archaeologists will likely find more utility in the predictive asp ects of the model and it ability to accurately delineate prehistoric or histor ic settlements hat have been abandoned. Cultural features on the landscape are surrounded by various types of climax fore st and are often located in close proximity to water bodies. Of particular interest to this resear ch is a separation of forest that has been modified recently or in the past from areas that would not seem to hold as high a probability of containing an archaeological site. Stated another way, feature discrimination and edge detection require the separation of cultural entities from a surrounding unmodified forest baseline. However, species composition creates conditions that produce a variable and he terogeneous baseline. Additionally, as culturally modified clearings age they abso rb and reflect light in different ways.

PAGE 188

177 Literature on anthropological applications of space borne multispectral data like Landsat emphasizes classical set theory appr oaches to analysis. Discussion tends to focus on using the signal of known locations to produce classifications of image data. To date, this approach has been extremely successful, and thus, co mmonly employed as a means of extracting information, most especi ally in the Neotropics. In most cases, however, only portions of scenes are utilized. The dispersed cluste ring settlement pattern noted among the Xinguanos may encompass even greater range than can be captured in a single Landsat scene, but the complications en countered during the course of this study would certainly be exponentially increased in one were to utilize multiple scenes in the classification process. The larger the scale of the analysis, the mo re difficulty signature extension would pose. The use of transformations in concert with image interpretation, as well as manual feature extraction through the use of ground-truthed data in combination with GPS for accurate position measurements is an under exploited methodology in anthropological and archaeological applications of remo te sensing. The importance of image transformation cannot be stress ed enough. While it is true that, in and of themselves, image transformations do not automate featur e definition, they do have the ability to dramatically increase the level of information available for interpretation across space and through time. This is especially true of research focused on vegetation, as the baseline may be quite heterogeneous. Activ ely cultivated land, wh ether on the site of previous occupational debris or in newly cleared areas, as we ll as other cultural features, may indeed be distinct from the surrounding vegetation. However, at issue is not appearance, but if, in fact, these cultural featur es are statistically dis tinct from other kinds

PAGE 189

178 of vegetation coverage within the limits of the data collection abilities of the remotely sensed imagery platform, or even within all contexts. Recalling that, in large part, multispectral remote sensing of vegetation relies on differences in the pattern of canopy shading, it stands to reason that while multispectral data are quite useful for discri minating material differences in land cover, particularly in the context of geology, medium resolution de tectors like Landsat may not discriminate differences in vegetated land cover through ma terial specific absorption bands. Landsat has sufficient spectral resolution to discri minate between vegetation and soil based on variation in energy absorpti on, but the detector may not have sufficient spectral resolution to detect differences in sub-classe s of vegetation without the addition of other analysis tools to either increase the sens itivity of the data to variations between reflectance measurements, or by utilizing methodologies that s eek to push analysis to the sub-pixel level (e.g. Spectral Mixture Analys is). The Landsat platform can, however, distinguish structural differenc es in vegetated canopies that are consistently related to land cover classes, and it is th ese structural differences that separate cultural features from climax forest. Thus, context is absolute ly critical to feature identification. Simply examining a 5-4-3 color composite with a piecewise linear contrast enhancement or histogram equalization applied to unprocessed data can generate an enormous amount of information through visual inspection. Patte rn recognition capabilities make any human observer likely to be able to distinguish archaeo logical sites from surrounding vegetation with little instruction. The regular patterning of Xinguano ar chaeological sites stands, at times, in sharp contrast to the seemingly irre gular patterns of many ve getative features. It is important to note that expl oration of raw geometrically and radiometrically corrected

PAGE 190

179 data does have significant merit in attempts to develop settlement information, and predict the location of archaeolo gical sites. However, 8-b it data displayed on a cathode ray tube can only represent 256 values in three bands: red, gr een, and blue. Multispectral TM data is collected in seven bands each with 8-bit quantization. Visual exploration of raw imagery excludes four dimensions of da ta variability. The strength of digital multispectral data is the ability to examine multivariate relationships, and the ability to quantify what patterns actually make a featur e cultural rather than natural, as well as allow for replicability. The point of this exercise is to demonstrate that there does exist a methodology that provides for objective assessmen t of the potential of a specific region to contain archaeological sites derived from an inductive model formed by an amalgamation of different anthropological data sources, some traditional and some more technologically informed.. Classifications and transformations build relationships between variables producing an image that contains information rather than only brightness valu es in a given portion of the electromagnetic spectrum. Image pro cessing creates a scene composed of values that can be related more directly to real world entities in the st udy area. By knowing how radiant energy interacts with matter, by in telligently sampling multiple absorption bands of energy, and by examining the relationshi ps between variable outcomes in many wavelengths, one can see that remote sensing is a powerful tool for analysis of spatial phenomena. Land-use/land-cover classification in Amaz onia is, at best, an extremely complex, arduous task, and its degree of difficulty increa ses with the number of classes one wants to distinguish. The spectral signatures detected by Landsat TM sensors produce images

PAGE 191

180 that, more often than not, contain mixed responses for the heterogeneous tropical environment. What would appear to be a straightforward case of assessing and defining the separability of vegetative cover classes b ecomes complicated by the fact that distinct vegetation types simply do not exist. Cover classes, instead, are forced to encompass scenarios within the complex dynamics of vegetation clearing or recovering, and assigning these cover classes is a tricky en terprise, attempting to tease out minute variations in the spectral signatures of vegetation at va rious stages of successional growth, within a variety of soil environments. The area under question is remote and quite large, with a lack of r eady access. The paucity of data with direct bearing on this research, such as soil maps and extensive ar chaeological investigations, also complicates the process of classification. A number of issues creep into the eq uation. Some problems faced are simply operational. For example, the choice of an August image from 2002 and a May image from 2003 might, at a glance, seem problematic . August comes at the end of a very dry period in the Xingu. Vegetation seems, on visu al inspection, more clearly delineated in May. On the other had, comes at the tail end of the wet season, thus vegetation is suffused with moisture and at perhaps it greenest. Optimally, both images under analysis would have been from the August time period, the loss of Landsat 7’s scan-line corrector (SLC) in late May of 2003 mean t that any imagery collected for August would have been SLC-off data, vastly comp licating the analysis. Other questions are more methodological, such as the reasonable implementation of a classification system designed to describe the variance of vegetation class dynamics at a scale and complexity that would permit the separation of vegetation located in

PAGE 192

181 anthropogenic soils from surrounding cover. It would seem obvious that the best way to overcome the difficulty of discriminating ve getation types would be an increase in fieldwork efforts to gather more ground-truth data. In the case of the Upper Xingu, the spatial extent is extensive. As the efforts of this study were to provide for predictive modeling of site locations outside the park boundaries, the entire scene was utilized, providing enough area to include multiple scenar ios within the distinct landscapes and classificatory outcomes. During fieldwork, da ta was collected as GPS survey was being conducted. Indigenous informants often acco mpanied the survey team, providing an incredible resource for identification of different vegetation types, as well as assisting in the location of archaeological sites. Accuracy assessment of the classification process brought attention to some critical issues. The original classifi cation schema seemed to have a great deal of difficulty distinguishing between vegeta tion within archaeological si tes and vegetation comprising the surrounding environment. Perhaps the di stinction between thes e two areas has been exaggerated more than is warra nted. It is entirely con ceivable that the anthropogenic nature of the soil content within archaeo logical sites has less bearing on the flora composition of those areas than does the leg acy of a managed landscape. Perhaps the similarity between the vegetati on internal to any given site and external to it cannot be completely delineated. One way of reducing the risk of misclassification would be to group classes such as archaeological site, and the areas surrounding the sites into a single class of “culturally modified.” While this would likely have made quite a bit more sense from some perspectives (not the least of which would be ease and accuracy of the

PAGE 193

182 classification itself), it would come at the cost of generalizing covera ge classes to prevent the predictive model from localizing areas with a high probability of being “sites.” Additionally, there was no real distinction at all between culturally cleared areas within the park boundaries (resulting from i ndigenous land-use techniques) and burned or cleared pasture areas outside the park boundaries, save for the massive size of the external clearings, as well as their more regu lar shape. This point is important enough to be driven home again. The overall land management practiced by the indigenous populations within the park showed small, disp ersed, cleared areas, with plenty of climax forest between these regions. Pasturela nd outside the park boundaries indicated a wholesale clear-cutting of what were obviously recently forested areas. Perhaps even more distressing was the proximity of these cleared pastures to the park boundaries. Nearly the entire southern bounda ry of the park had cleared pasture all the way up to the boundary itself. In some cases, there appeared to even be some encroachment into the park itself, a possibility that, while outside the scope of these inve stigations, certainly warrants further investigation at a later time. Based purel y on reflectance, however, the similarity of values for indigenous clearings and external pasturel ands, explained by the greater contribution of soil spectral response to the signature of spar sely covered grassy vegetation. The rapid and aggressive regrowth of s econdary vegetation in tropical areas has long been a topic of discussi on. Different regeneration pa tterns occur depending on land management practices following deforesta tion. Perhaps one important question not covered by this research is the role of speci es composition within the different stages of regrowth, and within different regions cultivated by differe nt cultural groups. Studies

PAGE 194

183 have shown that disturbance from slash-andburn agriculture affects species composition much more than stand structure and bioma ss (Uhl 1987). Although such an issue is of central relevance to the maintenance of local and regional biodiversit y, the utilization of remotely sensed data to classify distinct vegetative communities on the basis of species composition remains little more than a distant possibility. At best, current applications using imagery at present resolutions limits the scale at which researchers may conduct investigation. Presently, one can only hope to recognize different st ructural patterns and processes. Even so, the enhanced capabilities provi ded by an ever increasing repertoire of applications and sensor technol ogies are changing the face of analysis for Earth surface feature information, spawning new approaches and moving researchers towards a more integrative vision about LULC change within and across research sites (Burrough and Frank 1995). The expansion, refinement, a nd integration of remote sensing and geoprocessing techniques provide for the manipul ation of spatial data at several scales, further highlighting the mutual benefits of closer links between Geographic Information Systems (GIS) and methods of spatial data an alysis. There are quite a number of major hurdles to integrating cross-pl atform spatial datasets, and t hough these issues have been well documented in general GIS literature (s ee Ehlers 1992; Goodchild et al. 1992), most have received little attention overall. The primary impediments to a fully integrated system of image processing and spatial anal ysis stem from a) The different spatial resolutions of the various data sets; b) the prevalence of e rrors in individual data sets compounded by integration; c) physical constraint s in integrating rasterand vector-based data sets effec tively; and d) matching data sets captured over appropriate time intervals.

PAGE 195

184 To be truly successful, integra tion of satellite imagery, GPS data sets, and GIS analysis, the researcher must consider possible cons traints caused by the differences in their various spatial resolutions. I refer not only to the value of the actual datasets themselves, but also to the degree of compatibility between each of the data sets. Specifically, one should question whether the varied resolutions of the source data actually compliment each other. Can survey data collected form differential GPS instrumentation, providing accuracy down to a sub-meter level, be fully integrated into the 2030 meter scale of midresolution satellite imagery, combined with the scale of DEM maps or LULC classification digitized into GI S data layers? They certainly can compliment one another, so long as the scalar issues are addressed (in the case of this st udy by limiting the scope of the analysis to the sensor platfo rm with the lowest resolution). It is a simple fact that some form of error will likely be enco untered in any given spatial data set. These errors may result during data co llection, manipulation, interpretation, or presenta tion (e.g. image capture for remote sensing data and the digitizing of data for input into GIS). Thes e types of error can, for the most part, be mitigated through judicious use of data qua lity assessment, carefully noting throughout the process potential error sources and incorpor ating error estimates. The implication of the raster/vector data representation of spatial data is an on-going issue of relevance to many GIS based investigations. Ehlers (1992) has suggested that a fully integrated single system is the only effective way of resolvi ng the different representations of space as given currently by rasterand vector-based data sets. One of the most critical aspects of integra ting data is to insure that data capture occurs either at precisely the same time or within a reasonable span (such as within the

PAGE 196

185 same season). If one lacks a good temporal ma tch, it can frequently result in a degraded data synthesis, introducing yet another source of error into the equation. There seems to be a real lack of recognition or failure to document how data integration problems have been overcome in the current literature, making the comparison of results between archaeological survey studies, LULC classi fication and inventory studies, and overall the approaches taken to the spatial analysis of processed imagery problematic and. in some cases, impossible. In the same way that arch aeologists have establis hed tried-and-tested methods for the analysis and comparison of tr aditional field mapping techniques, there is a need for the GIS/RS community (specifica lly of archaeologists who wish to avail themselves of these technologies) to develop a similar approach towards an established methodology or, at the very least, a strategy fo r the integrated use of multi-source digital data sets for featur e detection, classificat ion, and modeling. I propose a multisource strategy that relies first on the use of GPS in fieldwork survey situations. Not only has it been show n that GPS is an extremely effective way to collect survey data (especially in inho spitable, or, in some cases, impassible environments), but GPS can allow for target ing of specific ground-truthed control point for georeticifcation of imagery data, as well as precise collection of regions of interest (ROIs) for later classificati on. Researchers can avail themselves of common remote sensing algorithms, such as filtering and textural enhancement procedures enabling feature detection, as well as classification (especially in conjunction with ground-truthed data sets), resulting in raster data that can provide detailed information over large spatial extents, at regular intervals, and for relatively little cost. It is though a GIS that these various data sets (vector and raster) can be fused together into unified spatial data.

PAGE 197

186 The potential for using digital remotely sens ed data as a primary source of data for GIS has received considerable attention in re cent years, particular ly in applications related to natural resources, as well as rese arch conducted at large spatial scales. Data remains the most expensive component of a GI S, and the need for inexpensive, accurate, and current GIS data has created a huge demand for remotely sensed im agery of all types. The fundamental issue at hand, how ever, is the integration of a vector data structure into a raster model, and thus GIS-based data sets remain separated from remotely sensed data at a fundamental level. The v ector processing capabilities of GIS can easily be used to delineate areas in an image for processing. The real power of image processing platforms derives from their ability to utilize pattern recognition, e dge extraction, and segmentation algorithms, processes not yet av ailable at equivalent functio nality levels in commercial GIS applications. In a perfectly integrated system, one can see th e utility of vector polygons extracted from images using edge det ection methods that can then be modified according to rule sets based on the value of the image pixels. The argument has been made that most image processing platforms have adequate vector conversion utilities such that the burden of integrated desi gn should fall on the shoulders of the GIS community. However, close inspection of rast er-derived lines and boundaries is that they contain a stepped appearance be traying their raster origin. On the other hand, one would expect GIS functionality to incorporate so me form of pixel interpolation to produce smooth boundaries, especially cons idering that both vector and raster data are often used within the confines of this one platform. Perhaps the termi nology, then, is at fault. The integrated approaches discussed here (and the approach taken in this study) should more likely be deemed an “interfacing” of the tec hnologies rather than a true “integration.”

PAGE 198

187 Consider, though, that the most important concern is to have GIS users using remotely sensed data, and remote sensing scientists using GIS in combination to maximize data extraction, and not to quibble over the semantics of the situati on. It could be well argued that the best way to achieve this goal is to provide a software environment wherein a researcher could use the best format for their specific data and have the software able to provide a level of functi onality facilitating cross platform integration. Remote sensing studies over the past few years have steadily moved from empirically based image classifications, mappi ng, and land-use/land-c over inventories to more deterministic modeling of scene char acteristics including more knowledge-based image interpretation. GIS re searchers have expanded from simple map overlay and relational models to spatiall y distributed simulation modelin g, veiwshed analyses, and informed three-dimensional analytical mode ls. To progress, researchers in remote sensing and GIS will need to possess hardware and software that can aid in the integration of these platforms, including im proved interfaces between image processing, GIS, database management, and statistical soft ware packages. It will be difficult going, and the reluctance of commercial software deve lopers to take the lead in these efforts suggests that such a product will have to ri se from open-source efforts. The potential, though, is far to tantalizing to not work towa rds this common goal. The possibility of testing spatial models through the use of ge oreferenced databases and algorithms to measure spatial heterogeneity, to fully inte grate remotely sensed imagery with GIS applications, to effectively turn the establ ished methodological approaches to endeavors like survey, mapping, classification, and multisp ectral, mutli-temporal analyses on their

PAGE 199

188 head lays bear a fundamentally new way in which researchers can attempt to understand land-use, as well as human impacts, in the Amazon. Land Use/Land Change in Amazonia Perspectives The search for some means of quantitativ ely analyzing and desc ribing the structure of landscapes has become a high priority in both social and environm ental sciences. In the midst of contemporary archaeological methodologies still dominated by empirical approaches and case studies, perhaps it is time to explore different means of information gathering and modeling. This dissertation represents an effort to develop a more integrative scientific vision addressing the complex interactions between people and the environment, carried out through a mergi ng of the old and the new, traditional approaches to data collection and utilizati on of new technological advances to augment and refine a new model of huma n-driven landscape formation. Approaches may differ slightly in th e examination of questions of humanenvironmental interaction at multiple temporal and spatial scales. Human ecological studies form the foundation of new methodolog ies, like historical ecology and landscape ecology, investigating the scien ce of human alteration of land scapes. Historical ecology in particular has provided new schema for researching human ecosystems. However, these new ways of looking at anthropological problems have not been fully integrated into anthropological theory. In additi on, there remains a re lative paucity of methodological procedure and stan dardization in research. This makes the study of the human dimensions of landscape change a real challenge. In the current scientific environment, prio rities for research have seen a decided shift towards an inclusion of the study of processes across temporal, spatial, and socio-

PAGE 200

189 cultural axis (Arrow et al. 1995; Stern et al. 1992; Turner et al. 1990). Issues of land use/land change are central to these new lines of inquiry, du e to their intricate dynamics and its consequences in landscape structur e and function (Lambin 1997; Turner et al. 1995; Turner 1997). This is particularly re levant in the Upper Xingu because of the landscape transformation evidenced in arch aeological investigations, and, more specifically, its implications for shifting th e view of prehistoric human-environmental interaction from a stagnant, limiting, and reac tionary subsistence to a dynamic interplay of long-term wide-scale landscape alteration. In this sense, this i nvestigative research into alternate methodological approaches to detecting th ese changes can offer an opportunity to broaden the disc ussion of the human dimensions of landscape pattern and process. This dissertation has sought to create a hybrid approach of traditional techniques and an integrative use of widely availabl e technologies to cr eate a new means of answering these questions of landscape tran sformation in a more efficient, expedient way, and over a larger spatial scale than woul d be feasible using full-scale excavation. The landscape characterizations presented in this study permit us to attain a broader perspective regarding the larg er debate about the role of humans in the formation of landscapes, and the impacts of long-term hu man occupations on the environment in nonwestern tropical settings. As knowledge about these processes increase s through studies at several scales and approaches, there is growing agreement on the need to investigate culture and context, before testing general hypot heses about human-induced ecological outcomes (Lambim 1997). Thus, this study provides a methodologi cal approach as an exploratory step towards a broader understanding of these issu es. Through this base-line extraction of

PAGE 201

190 features from image data, we can set the ground work for future research into issues of settlement pattern analysis and a deeper understanding of landscape transformation through time. It should be remembered howev er, that while the techniques described in this dissertation may be of use elsewhere, when dealing with the complex relationships between people and the environment, differe nt environmental outcomes may result from similar socioeconomic dynamics, and distinct socioeconomic dynamics may produce similar environmental outcomes. Simply put, while this dissertation may provide a useful alternative approach to res earch in the Upper Xingu, its usefulness may be limited. Only further testing in other regions will bear out the utility of this specific methodology beyond the cultural and geographic confin es of this particular study area. Beyond the elaboration of a methodologica l strategy based on multi-disciplinary integration for the study of human-envi ronmental interaction and landscape transformation in the Upper Xingu, the results of this dissertation can contribute to the body of work regarding the use of new sens or platforms and processing techniques to improve the classification of different t ypes of forest, specifically relating to anthropogenic alterations of vegetation struct ure. Achieving more accurate information on these types of classificator y categories represents a cr ucial step to inform ongoing research into settlement pattern analysis a nd the size and nature of Pre-Columbian human occupation of the Upper Xingu, as well as the human dimensions of environmental change. Although the majority of the Upper Xingu region is composed of closed tropical forest, the landscape retains a degree of distinctive ecological zones. Ac ting as clines, these zones blend into one another (again, the pr obably source of much of the error in the

PAGE 202

191 classification schema presented in this diss ertation) making distin ct boundaries difficult to define. During the course of the Upper Xingu Project, headed by Michael Heckenberger, a small sample of this regi on was discovered to have undergone massive alteration by human agents (Heckenberger et al. 2003). Ba sed on his findings, Heckenberger (2005) hypothesized a significant portion, perhaps as much as fifty percent or more, of the upland (non-inundated or terra firme ) forests and adjacent wetlands, are the result of human transformati on of the landscape. Balee (1 989) has also suggested that one of the most striking feat ures of the region is the degr ee to which it has been altered by the Xinguanos, creating vast anthropogenic landscapes. Heckenberger (2005: 32) has suggested that “the conclusion that much of the landscape was not only anthropogenic in origin but intentionally cons tructed and managed is inescap able the more the scale of ancient settlements and their ‘monuments’ (e.g., plaza and causeway peripheral mounds and massive ditches) are investigated.” The pattern of anthropogenic vegetation uncovered through this investigation adds to a mounting body of evidence that would seem to reaffirm this hypothesis. Today, I would not accept any part of the fo rest to be “pristine” without a detailed examination on the ground, at least in the te rritory of the Kuikuro (Xinguano Carib) community where the present study largely t ook place. In place of small paths in the forest and minor openings related to plaza villages and gardens, I now envision tree lined causeways, well maintained and broad roads, large, patchy tracts of agricultural fields, leading out from the towns and v illages that make up the skeleton of Xinguano history, and an equa lly constructed we tland environment, including major transportation canals, managed ponds, reservoirs to improve fishing, drinking and bathing reservoirs, rais ed causeways, wells, raised fields, road systems, among other features (Heckenberger 2005: 33). Future Directions Within anthropological study, we often find dur ing the course of our research that as we attempt to answer a handful of ques tions, we often succeed in raising dozens of

PAGE 203

192 new questions. The findings presented in Chapte rs 6 and 7 shed a great deal of light on the issues posited in Chapter 1 with re gard to the nature of the anthropogenic transformation of much of the Upper Xingu region. This research has made important steps to refuting the hypothesis of static, small-scale communities, instead building on the hypothesis of large-scale, interconnected communities dramatically altering their environment to suit their needs. At the same time, however, in keeping with the tone of this new methodological approach to answering such questions, it is readily apparent that further studies will be necessary to uncover other socioecological processes that affected the trajectories of landscape transformation. The findings of this research can be built on to delve further into the Xingu region a ddressing still more issues including: How can other remote-sensing platforms (i.e. soil-resistivity, ground penetrating radar) improve our understanding of the formation of terra preta ? How do the patterns uncovered by this cla ssification fit add to our knowledge of Xinguano settlement patterning? What are the implications do the presence of Pre-Columbian large-scale sedentary populations in the Upper Xingu have for fina lly putting to rest the Tropical Forest Tribe concept? Does the dramatic alteration of landscape on such a large scale and the apparent interconnectedness of communities add to the growing body of research into possibility of chiefdom formation in Amazonia? These questions should be addressed in a multi-disciplinary fashion, with social sciences and environmental sciences helping to identify driving forces affecting LULC dynamics and landscape change. If nothing else, the research presented here should highlight the value of integrating new form s of data collection into the methodological approaches to such questions. In this se nse, various initiatives are appropriate to maximize our data collection efforts including:

PAGE 204

193 Increasing temporal resolution (i.e., other satellite image dates) to have a better control over LULC dynamics Testing new sensors to scale the analys is down and up (e.g., IKONOS and MODIS) Testing new geoprocessing techniques to improve the accuracy of LULC classifications (e.g., sub-pixel classifica tion, linear mixture models, multivariate analysis, spatial-spectral classifiers) Increasing the use of ancillary data for the study of landscape structure and change (specifically soil studies) Increasing fieldwork efforts to gather data on recent trends regarding actors' landuse decisions We should prioritize further research in to ADE, more extensive archaeological investigations of the probable site locatio ns suggested by the classifications presented here (in essence, further gr ound-truthing of findings), and th e additional integration of ethnographic information into the archaeology of the region, combined with the use of GPS survey, GIS, and additional LULC assessments to aid in our achieving a finer understanding of Xinguano's decisions regardin g land use, both in the past and in the present.

PAGE 205

194 APPENDIX A CONFUSION MATRICES, CLASS SPECT RAL PROFILES, AND SEPARABILITY VALUES Table A-1. The 2002 (August) combin ed classes confusion matrix Overall Accuracy = (14408/15111) 95.3478% Kappa Coefficient = 0.9025 Ground Truth (Pixels) Class Water bodyForested SavannahInlet areas Culturally active Unclassified 38000 2 Water body 10577006 0 Forested 034711135 0 Savannah 0626714 15 Inlet areas 097068 0 Culturally active 0001 346 Transitional 061413 5 Bare soil 0000 31 Anthropogenic vegetation 036046 0 Pasture outside park 0000 0 Aldea 0000 0 Actively cultivated site 02022 3 Total 10615494292305 402 Ground Truth (Pixels) Class Transitional Bare soil Anthropogenic vegetation Pasture outside park Aldea Unclassified 00017 0 Water body 0000 0 Forested 39080 1 Savannah 4000 0 Inlet areas 20170 0 Culturally active 2100 1 Transitional 48000 0 Bare soil 02500 0 Anthropogenic vegetation 110910 0 Pasture outside park 0002528 0 Aldea 0004 53 Actively cultivated site 9000 0 Total 115261162549 55

PAGE 206

195 Ground Truth (Pixels) Actively cultivated siteTotal Unclassified 057 Water body 010583 Forested 11552 Savannah 5311 Inlet areas 5189 Culturally active 5356 Transitional 17103 Bare soil 157 Anthropogenic vegetation 40224 Pasture outside park 02528 Aldea 057 Actively cultivated site 5894 Total 14215111 Ground Truth (Percent) Class Water bodyForestedSavannahInlet areas Culturally active Unclassified 0.360.000.000.00 0.50 Water body 99.640.000.001.97 0.00 Forested 0.0070.243.7744.26 0.00 Savannah 0.001.2191.444.59 3.73 Inlet areas 0.0019.640.0022.30 0.00 Culturally active 86.07 Transitional 1.24 Bare soil 7.71 Anthropogenic vegetation 0.00 Pasture outside park 0.00 Aldea 0.00 Actively cultivated site 0.000.400.007.21 0.75 Total 100.00100.00100.00100.00 100.00 Ground Truth (Percent) Class Transitional Bare soil Anthropogenic vegetation Pasture outside park Aldea Unclassified 0.00 Water body 0.00 Forested type 33.910.006.900.00 1.82 Savannah 3.480.000.000.00 0.00 Inlet areas 1.740.0014.660.00 0.00 Culturally active 1.743.850.000.00 1.82 Transitional 41.740.000.000.00 0.00 Bare soil 0.0096.150.000.00 0.00

PAGE 207

196 Anthropogenic vegetation 9.570.0078.450.00 0.00 Pasture outside park 0.00 Aldea 96.36 Actively cultivated site 7.830.000.000.00 0.00 Total 100.00100.00100.00100.00 100.00 Ground Truth (Percent) Class Actively cultivated siteTotal Unclassified 0.000.38 Water body 0.0070.04 Forested type 7.753.65 Savannah 3.522.06 Inlet areas 3.521.25 Culturally active 3.522.36 Transitional 11.970.68 Bare soil 0.700.38 Anthropogenic vegetation 28.171.48 Pasture outside park 0.0016.73 Aldea 0.000.38 Actively cultivated site 40.850.62 Total 100.00100.00 Class Commission (Percent) Omission (Percent) Commission (Pixels) Omission (Pixels) Water body 0.060.366/1058338/10615 Forested type 37.1429.76205/552147/494 Savannah 14.158.5644/31125/292 Inlet areas 64.0277.70121/189237/305 Culturally active 2.8113.9310/35656/402 Transitional 53.4058.2655/10367/115 Bare soil 56.143.8532/571/26 Anthropogenic vegetation 59.3821.55133/22425/116 Pasture outside park 0.000.820/252821/2549 Aldea 7.023.644/572/55 Actively cultivated site 38.3059.1536/9484/142 Class Producer Acc. (Percent) User Acc. (Percent) Producer Acc. (Pixels) User Acc.(Pixels) Water body 99.6499.9410577/1061510577/10583 Forested type 70.2462.86347/494347/552 Savannah 91.4485.85267/292267/311 Inlet areas 22.3035.9868/30568/189 Culturally active 86.0797.19346/402346/356

PAGE 208

197 Transitional 41.7446.6048/11548/103 Bare soil 96.1543.8625/2625/57 Anthropogenic vegetation 78.4540.6391/11691/224 Pasture outside park 99.18100.002528/25492528/2528 Aldea 96.3692.9853/5553/57 Actively cultivated site 40.8561.7058/14258/94 Table A-2. The 2003 (May) combin ed classes confusion matrix Overall Accuracy = (12506/13520) 92.5000% Kappa Coefficient = 0.8024 Ground Truth (Pixels) Class Water bodyForested SavannahInlet areas Culturally active Unclassified 89010 0 Water body 1052610015 0 Forested 03350106 12 Savannah 002259 13 Inlet areas 0100117 0 Culturally active 0042 328 Transitional 036212 14 Bare soil 0000 35 Anthropogenic vegetation 0136044 0 Pasture outside park 0000 0 Aldea 0000 0 Actively cultivated site 0000 0 Total 10615494292305 402 Ground Truth (Pixels) Class Transitional Bare soil Anthropogenic vegetation Pasture outside park Aldea Unclassified 00080 0 Water body 0040 0 Forested 4202213 0 Savannah 6000 0 Inlet areas 8020 0 Culturally active 0000 0 Transitional 48003 0 Bare soil 02600 0 Anthropogenic vegetation 70880 0 Pasture outside park 100774 1 Aldea 000112 30 Actively cultivated site 3000 0

PAGE 209

198 Total 11526116982 31 Ground Truth (Pixels) Actively cultivated siteTotal Unclassified 1171 Water body 010555 Forested 63593 Savannah 3256 Inlet areas 35172 Culturally active 0334 Transitional 14156 Bare soil 061 Anthropogenic vegetation 17292 Pasture outside park 0776 Aldea 0142 Actively cultivated site 912 Total 14213520 Ground Truth (Percent) Class Water bodyForestedSavannahInlet areas Culturally active Unclassified 0.840.000.340.00 0.00 Water body 0.00 Forested 0.0067.810.0034.75 2.99 Savannah 0.000.0077.052.95 3.23 Inlet areas 0.00 Culturally active 81.59 Transitional 0.000.6121.233.93 3.48 Bare soil 8.71 Anthropogenic vegetation 0.0027.530.0014.43 0.00 Pasture outside park 0.00 Aldea 0.00 Actively cultivated site 0.00 Total 100.00100.00100.00100.00 100.00 Ground Truth (Percent) Class TransitionalBare soil Anthropogenic vegetation Pasture outside park Aldea Unclassified 0.00 Water body 0.00 Forested type 36.520.0018.971.32 0.00 Savannah 0.00 Inlet areas 6.960.001.720.00 0.00 Culturally active 0.00

PAGE 210

199 Transitional 41.740.000.000.31 0.00 Bare soil 0.00100.000.000.00 0.00 Anthropogenic vegetation 6.090.0075.860.00 0.00 Pasture outside park 0.870.000.0078.82 3.23 Aldea 96.77 Actively cultivated site 2.610.000.000.00 0.00 Total 100.00100.00100.00100.00 100.00 Ground Truth (Percent) Class Actively cultivated siteTotal Unclassified 0.701.26 Water body 0.0078.07 Forested type 44.374.39 Savannah 2.111.89 Inlet areas 24.651.27 Culturally active 0.002.47 Transitional 9.861.15 Bare soil 0.000.45 Anthropogenic vegetation 11.972.16 Pasture outside park 0.005.74 Aldea 0.001.05 Actively cultivated site 6.340.09 Total 100.00100.00 Class Commission (Percent) Omission (Percent) Commission (Pixels) Omission (Pixels) Water body 0.270.8429/1055589/10615 Forested type 43.5132.19258/593159/494 Savannah 12.1122.9531/25667/292 Inlet areas 31.9861.6455/172188/305 Culturally active 1.8018.416/33474/402 Transitional 69.2358.26108/15667/115 Bare soil 57.380.0035/610/26 Anthropogenic vegetation 69.8624.14204/29228/116 Pasture outside park 0.2621.182/776208/982 Aldea 78.873.23112/1421/31 Actively cultivated site 25.0093.663/12133/142 Class Producer Acc. (Percent) User Acc. (Percent) Producer Acc. (Pixels) User Acc.(Pixels) Water body 99.1699.7310526/1061510526/10555 Forested type 67.8156.49335/494335/593

PAGE 211

200 Savannah 77.0587.89225/292225/256 Inlet areas 38.3668.02117/305117/172 Culturally active 81.5998.20328/402328/334 Transitional 41.7430.7748/11548/156 Bare soil 100.0042.6226/2626/61 Anthropogenic vegetation 75.8630.1488/11688/292 Pasture outside park 78.8299.74774/982774/776 Aldea 96.7721.1330/3130/142 Actively cultivated site 6.3775.009/1429/12 Table A-3. The spectral profiles for th e 2002 (August) combined classification Aldea Basic Stats Min Max Mean Stdev Num Eigenvalue Band1 -0.233 0.000 -0.177 0.050 1.000 0.880 Band2 -0.589 -0.126 -0.478 0.103 2.000 0.023 Band3 -0.240 -0.098 -0.180 0.036 3.000 0.001 Band4 -0.520 0.268 -0.318 0.158 4.000 0.000 Band5 1.126 5.003 3.367 0.929 5.000 0.000 Band6 -0.255 -0.074 -0.201 0.044 6.000 0.000 Band7 0.002 0.047 0.018 0.008 7.000 0.000 Eigenvector Band1 Band2 Band3 Band4 Band5 Band6 Band7 Band1 -0.045 -0.094 -0.032 -0.084 0.990 -0.031 0.003 Band2 -0.177 -0.350 -0.002 -0.896 -0.122 -0.167 -0.019 Band3 -0.154 -0.097 0.633 -0.073 0.020 0.711 0.236 Band4 0.143 0.821 0.387 -0.315 0.064 -0.226 -0.030 Band5 0.054 -0.364 0.604 0.243 -0.012 -0.625 0.225 Band6 0.132 -0.160 0.288 0.043 0.009 0.063 -0.932 Band7 -0.950 0.167 0.032 0.155 -0.017 -0.143 -0.156 Bare Soils Basic Stats Min Max Mean Stdev Num Eigenvalue Band1 -0.064 0.024 -0.032 0.023 1.000 0.082 Band2 -0.210 -0.014 -0.122 0.048 2.000 0.008 Band3 -0.107 -0.059 -0.077 0.012 3.000 0.000 Band4 -0.160 0.152 -0.057 0.082 4.000 0.000 Band5 0.030 1.189 0.454 0.283 5.000 0.000 Band6 -0.100 -0.060 -0.085 0.010 6.000 0.000 Band7 -0.007 0.010 0.001 0.003 7.000 0.000

PAGE 212

201 Eigenvector Band1 Band2 Band3 Band4 Band5 Band6 Band7 Band1 -0.053 -0.129 -0.035 -0.071 0.987 -0.010 0.010 Band2 -0.197 -0.342 0.022 -0.905 -0.121 -0.102 -0.002 Band3 -0.052 0.328 0.835 -0.142 0.061 0.363 0.191 Band4 -0.101 0.634 -0.481 -0.290 0.046 0.489 -0.173 Band5 0.020 -0.017 -0.258 -0.016 -0.020 0.104 0.960 Band6 0.019 -0.597 -0.050 0.140 -0.060 0.779 -0.108 Band7 0.972 0.019 0.003 -0.228 0.039 0.031 -0.025 Culturally active Basic Stats Min Max Mean Stdev Num Eigenvalue Band1 -0.083 0.280 0.037 0.078 1.000 0.250 Band2 -0.249 0.334 -0.021 0.132 2.000 0.048 Band3 -0.198 -0.048 -0.092 0.015 3.000 0.000 Band4 -0.172 1.314 0.293 0.307 4.000 0.000 Band5 -0.627 1.859 0.642 0.424 5.000 0.000 Band6 -0.111 0.061 -0.051 0.036 6.000 0.000 Band7 -0.025 0.030 0.003 0.007 7.000 0.000 Eigenvector Band1 Band2 Band3 Band4 Band5 Band6 Band7 Band1 -0.142 -0.245 -0.010 -0.514 0.808 -0.055 0.011 Band2 -0.142 -0.217 0.018 -0.763 -0.583 -0.099 -0.011 Band3 -0.018 0.684 0.570 -0.269 0.061 0.329 0.151 Band4 -0.015 0.611 -0.742 -0.205 0.039 -0.090 -0.157 Band5 0.065 0.216 0.259 0.027 0.033 -0.931 0.119 Band6 0.220 0.020 0.230 -0.045 0.029 -0.037 -0.946 Band7 -0.952 0.066 0.071 0.190 -0.027 -0.054 -0.211 Non-anthropogenic forest Basic Stats Min Max Mean Stdev Num Eigenvalue Band1 0.109 0.271 0.192 0.024 1.000 0.043 Band2 0.215 0.439 0.327 0.030 2.000 0.009 Band3 -0.074 0.018 -0.024 0.013 3.000 0.000 Band4 0.529 1.143 0.827 0.096 4.000 0.000 Band5 -0.881 0.652 -0.090 0.205 5.000 0.000 Band6 0.018 0.062 0.041 0.008 6.000 0.000 Band7 -0.008 0.017 0.004 0.004 7.000 0.000 Eigenvector Band1 Band2 Band3 Band4 Band5 Band6 Band7 Band1 -0.068 -0.113 -0.026 -0.115 0.984 -0.001 0.014 Band2 0.197 0.151 -0.079 0.952 0.140 0.072 -0.016

PAGE 213

202 Band3 0.047 0.783 0.560 -0.114 0.093 0.190 0.120 Band4 -0.005 0.535 -0.772 -0.166 0.025 0.132 -0.269 Band5 0.024 0.240 -0.018 0.025 0.030 -0.968 0.044 Band6 0.002 0.045 -0.288 -0.018 -0.018 0.059 0.954 Band7 0.977 -0.080 -0.016 -0.196 0.035 Inlet Areas Basic Stats Min Max Mean Stdev Num Eigenvalue Band1 0.057 0.314 0.199 0.056 1.000 0.105 Band2 0.072 0.444 0.311 0.075 2.000 0.031 Band3 -0.086 0.003 -0.035 0.015 3.000 0.001 Band4 0.285 1.383 0.907 0.228 4.000 0.000 Band5 -0.641 0.907 0.099 0.275 5.000 0.000 Band6 -0.031 0.087 0.046 0.022 6.000 0.000 Band7 -0.012 0.023 0.006 0.007 7.000 0.000 Eigenvector Band1 Band2 Band3 Band4 Band5 Band6 Band7 Band1 -0.157 -0.200 0.010 -0.574 0.777 -0.045 0.020 Band2 -0.138 -0.166 0.028 -0.744 -0.625 -0.087 -0.004 Band3 0.037 0.816 0.477 -0.234 0.046 0.197 0.098 Band4 0.000 0.455 -0.837 -0.158 0.022 0.074 -0.248 Band5 0.013 0.226 -0.044 0.032 0.025 -0.935 0.267 Band6 0.056 0.021 0.261 0.008 0.027 -0.269 -0.925 Band7 -0.976 0.091 0.027 0.189 -0.033 -0.001 -0.049 Water Bodies Basic Stats Min Max Mean Stdev Num Eigenvalue Band1 -0.479 -0.089 -0.420 0.019 1.000 0.291 Band2 -1.581 -0.196 -1.336 0.072 2.000 0.005 Band3 -0.030 0.083 0.038 0.005 3.000 0.000 Band4 -0.919 0.015 -0.849 0.026 4.000 0.000 Band5 2.238 6.126 4.144 0.537 5.000 0.000 Band6 -0.027 -0.022 -0.024 0.001 6.000 0.000 Band7 0.057 0.161 0.138 0.012 7.000 0.000 Eigenvector Band1 Band2 Band3 Band4 Band5 Band6 Band7 Band1 -0.031 -0.067 0.003 0.006 0.997 0.000 0.019 Band2 0.118 0.917 -0.069 0.367 0.064 0.009 -0.024 Band3 0.202 -0.379 0.043 0.898 -0.027 0.029 0.079 Band4 0.039 -0.072 -0.242 0.058 0.015 -0.024 -0.965 Band5 -0.058 -0.062 -0.964 0.010 -0.008 0.069 0.243 Band6 -0.266 0.015 0.072 0.036 -0.007 0.959 -0.052

PAGE 214

203 Band7 0.932 -0.033 -0.031 -0.233 0.028 0.271 0.027 Pasture outside park Basic Stats Min Max Mean Stdev Num Eigenvalue Band1 -0.188 -0.056 -0.122 0.016 1.000 0.194 Band2 -0.511 -0.255 -0.376 0.030 2.000 0.008 Band3 -0.274 -0.118 -0.169 0.023 3.000 0.000 Band4 -0.589 0.092 -0.229 0.101 4.000 0.000 Band5 0.690 4.610 2.172 0.436 5.000 0.000 Band6 -0.275 -0.108 -0.157 0.020 6.000 0.000 Band7 -0.027 0.026 0.003 0.006 7.000 0.000 Eigenvector Band1 Band2 Band3 Band4 Band5 Band6 Band7 Band1 -0.001 -0.019 -0.038 0.141 0.989 0.003 0.007 Band2 -0.183 -0.323 0.028 -0.909 0.125 -0.142 -0.002 Band3 -0.116 0.118 0.672 -0.100 0.039 0.684 0.203 Band4 -0.094 0.036 -0.660 -0.122 -0.007 0.686 -0.262 Band5 -0.008 -0.933 0.022 0.301 -0.061 0.158 0.094 Band6 0.060 -0.081 0.333 0.034 0.013 -0.020 -0.937 Band7 0.970 -0.046 0.002 -0.194 0.027 0.124 0.057 Savannah Basic Stats Min Max Mean Stdev Num Eigenvalue Band1 -0.288 -0.119 -0.211 0.028 1.000 0.106 Band2 -0.665 -0.164 -0.424 0.080 2.000 0.011 Band3 -0.022 0.044 0.012 0.013 3.000 0.000 Band4 -0.801 -0.300 -0.575 0.085 4.000 0.000 Band5 0.767 2.576 1.514 0.321 5.000 0.000 Band6 -0.113 -0.047 -0.082 0.013 6.000 0.000 Band7 0.035 0.072 0.054 0.007 7.000 0.000 Eigenvector Band1 Band2 Band3 Band4 Band5 Band6 Band7 Band1 -0.051 -0.167 -0.014 -0.037 0.984 0.006 0.017 Band2 0.213 0.551 -0.039 0.793 0.134 0.047 -0.015 Band3 -0.244 0.338 0.633 -0.175 0.040 0.582 0.232 Band4 0.033 0.718 -0.078 -0.504 0.107 -0.459 -0.044 Band5 0.037 -0.172 0.530 0.184 -0.018 -0.614 0.527 Band6 -0.113 -0.062 0.514 0.095 0.010 -0.217 -0.814 Band7 0.937 -0.072 0.217 -0.206 0.031 0.156 -0.053

PAGE 215

204 Actively cultivated site Basic Stats Min Max Mean Stdev Num Eigenvalue Band1 0.074 0.239 0.159 0.034 1.000 0.049 Band2 0.154 0.364 0.257 0.046 2.000 0.020 Band3 -0.077 -0.006 -0.035 0.013 3.000 0.000 Band4 0.366 1.051 0.716 0.141 4.000 0.000 Band5 -0.498 0.559 0.046 0.215 5.000 0.000 Band6 -0.003 0.057 0.024 0.014 6.000 0.000 Band7 -0.011 0.017 0.006 0.005 7.000 0.000 Eigenvector Band1 Band2 Band3 Band4 Band5 Band6 Band7 Band1 -0.096 -0.147 -0.011 -0.254 0.951 -0.014 0.018 Band2 0.187 0.207 -0.047 0.909 0.294 0.089 -0.007 Band3 0.022 0.749 0.582 -0.188 0.075 0.210 0.121 Band4 -0.004 0.539 -0.752 -0.188 0.032 0.150 -0.293 Band5 -0.057 -0.260 -0.023 -0.011 -0.037 0.960 0.076 Band6 -0.024 0.098 -0.305 -0.016 -0.014 -0.058 0.945 Band7 -0.975 0.081 0.017 0.197 -0.033 -0.032 -0.027 Anthropogenic vegetation Basic Stats Min Max Mean Stdev Num Eigenvalue Band1 0.120 0.227 0.168 0.022 1.000 0.038 Band2 0.197 0.363 0.279 0.033 2.000 0.008 Band3 -0.057 -0.005 -0.028 0.011 3.000 0.000 Band4 0.594 0.979 0.759 0.085 4.000 0.000 Band5 -0.342 0.556 0.082 0.193 5.000 0.000 Band6 0.014 0.052 0.033 0.008 6.000 0.000 Band7 -0.001 0.016 0.007 0.003 7.000 0.000 Eigenvector Band1 Band2 Band3 Band4 Band5 Band6 Band7 Band1 -0.064 -0.129 -0.038 -0.091 0.985 -0.004 0.011 Band2 0.200 0.218 -0.046 0.941 0.127 0.086 -0.010 Band3 0.025 0.798 0.514 -0.195 0.107 0.182 0.122 Band4 -0.003 0.477 -0.792 -0.169 0.020 0.156 -0.303 Band5 0.030 0.254 -0.118 0.014 0.025 -0.933 0.224 Band6 0.031 0.005 0.301 0.017 0.025 -0.255 -0.918 Band7 0.976 -0.080 -0.015 -0.196 0.034 0.014 0.021 Transitional Basic Stats Min Max Mean Stdev Num Eigenvalue Band1 0.083 0.293 0.209 0.045 1.000 0.062

PAGE 216

205 Band2 0.088 0.431 0.310 0.060 2.000 0.036 Band3 -0.088 -0.011 -0.047 0.015 3.000 0.001 Band4 0.404 1.320 0.909 0.192 4.000 0.000 Band5 -0.793 0.393 -0.166 0.235 5.000 0.000 Band6 -0.033 0.077 0.037 0.020 6.000 0.000 Band7 -0.014 0.011 -0.001 0.005 7.000 0.000 Eigenvector Band1 Band2 Band3 Band4 Band5 Band6 Band7 Band1 -0.131 -0.186 -0.006 -0.436 0.870 -0.037 0.017 Band2 0.160 0.175 -0.046 0.836 0.484 0.090 -0.001 Band3 0.037 0.808 0.483 -0.219 0.080 0.218 0.106 Band4 -0.033 0.435 -0.833 -0.159 0.015 0.186 -0.235 Band5 0.062 0.291 -0.079 0.006 0.032 -0.949 0.061 Band6 0.089 -0.009 0.248 -0.009 0.025 -0.079 -0.961 Band7 0.971 -0.088 -0.058 -0.193 0.031 0.046 0.075 Table A-4. The spectral profiles for the 2003 (May) combined classification Aldea Basic Stats Min Max Mean Stdev Num Eigenvalue Band1 -0.145 -0.036 -0.082 0.023 1.000 0.165 Band2 -0.215 -0.020 -0.124 0.041 2.000 0.020 Band3 -0.002 0.160 0.084 0.041 3.000 0.001 Band4 -0.503 0.144 -0.125 0.142 4.000 0.000 Band5 1.298 3.029 2.209 0.402 5.000 0.000 Band6 -0.077 0.017 -0.019 0.021 6.000 0.000 Band7 0.020 0.061 0.047 0.009 7.000 0.000 Eigenvector Band1 Band2 Band3 Band4 Band5 Band6 Band7 Band1 -0.034 -0.057 0.035 0.128 0.989 0.023 0.015 Band2 -0.072 -0.236 0.232 -0.928 0.099 -0.114 -0.026 Band3 -0.537 -0.230 -0.706 -0.123 -0.001 0.338 0.175 Band4 0.122 0.857 -0.299 -0.304 0.103 0.143 -0.193 Band5 0.528 -0.138 -0.542 -0.075 0.048 -0.564 0.288 Band6 0.340 0.063 0.175 -0.085 -0.004 0.529 0.750 Band7 0.544 -0.360 -0.180 -0.042 0.006 0.504 -0.534 Bare soils Basic Stats Min Max Mean Stdev Num Eigenvalue Band1 -0.123 0.015 -0.019 0.029 1.000 0.043 Band2 -0.203 0.008 -0.035 0.044 2.000 0.010 Band3 0.028 0.118 0.050 0.019 3.000 0.000

PAGE 217

206 Band4 -0.586 0.039 -0.133 0.133 4.000 0.000 Band5 0.841 1.668 1.106 0.178 5.000 0.000 Band6 -0.065 -0.030 -0.038 0.007 6.000 0.000 Band7 0.016 0.037 0.024 0.005 7.000 0.000 Eigenvector Band1 Band2 Band3 Band4 Band5 Band6 Band7 Band1 -0.125 -0.202 0.085 -0.519 0.816 -0.028 0.018 Band2 0.131 0.150 0.019 0.800 0.565 0.035 -0.015 Band3 -0.254 0.040 -0.918 0.003 0.070 0.187 0.226 Band4 0.022 0.767 0.024 -0.225 0.071 0.442 -0.400 Band5 0.447 0.371 0.083 -0.153 0.035 -0.037 0.793 Band6 0.365 0.227 -0.325 -0.109 0.058 -0.762 -0.337 Band7 -0.754 0.398 0.191 0.074 -0.009 -0.430 0.214 Culturally active Basic Stats Min Max Mean Stdev Num Eigenvalue Band1 -0.113 0.150 0.025 0.066 1.000 0.217 Band2 -0.186 0.162 0.012 0.082 2.000 0.055 Band3 -0.048 0.141 0.047 0.031 3.000 0.000 Band4 -0.596 0.807 0.141 0.304 4.000 0.000 Band5 0.014 2.740 1.177 0.409 5.000 0.000 Band6 -0.068 0.007 -0.024 0.015 6.000 0.000 Band7 0.003 0.056 0.022 0.010 7.000 0.000 Eigenvector Band1 Band2 Band3 Band4 Band5 Band6 Band7 Band1 -0.128 -0.164 0.056 -0.507 0.835 -0.016 0.019 Band2 0.116 0.129 0.001 0.821 0.542 0.047 -0.004 Band3 -0.381 0.034 -0.856 0.020 0.020 0.303 0.172 Band4 0.280 0.905 -0.141 -0.236 0.088 -0.071 -0.120 Band5 0.543 -0.251 -0.413 -0.020 0.030 -0.621 0.289 Band6 -0.365 -0.033 -0.174 0.081 0.005 -0.536 -0.736 Band7 -0.564 0.269 0.209 0.081 -0.021 -0.477 0.575 Non-anthropogenic forest Basic Stats Min Max Mean Stdev Num Eigenvalue Band1 0.140 0.270 0.190 0.020 1.000 0.043 Band2 0.203 0.354 0.271 0.021 2.000 0.010 Band3 -0.168 -0.063 -0.109 0.018 3.000 0.000 Band4 0.640 1.264 0.912 0.102 4.000 0.000 Band5 -0.843 0.299 -0.241 0.203 5.000 0.000 Band6 0.006 0.044 0.029 0.006 6.000 0.000 Band7 -0.015 0.028 -0.001 0.004 7.000 0.000

PAGE 218

207 Eigenvector Band1 Band2 Band3 Band4 Band5 Band6 Band7 Band1 -0.066 -0.081 0.066 -0.146 0.981 0.000 0.012 Band2 0.145 0.114 0.017 0.968 0.162 0.050 -0.011 Band3 -0.162 0.244 -0.925 -0.006 0.069 0.170 0.157 Band4 -0.155 0.820 0.261 -0.104 0.028 0.362 -0.306 Band5 0.032 -0.109 0.206 -0.018 -0.032 0.688 0.685 Band6 0.138 -0.435 -0.138 -0.004 -0.010 0.603 -0.640 Band7 0.951 0.219 -0.100 -0.175 0.062 -0.030 0.050 Inlet areas Basic Stats Min Max Mean Stdev Num Eigenvalue Band1 0.147 0.374 0.219 0.033 1.000 0.071 Band2 0.219 0.414 0.298 0.030 2.000 0.022 Band3 -0.171 -0.049 -0.113 0.020 3.000 0.000 Band4 0.674 1.901 1.095 0.176 4.000 0.000 Band5 -1.082 0.429 -0.287 0.245 5.000 0.000 Band6 0.015 0.075 0.039 0.009 6.000 0.000 Band7 -0.022 0.005 -0.004 0.005 7.000 0.000 Eigenvector Band1 Band2 Band3 Band4 Band5 Band6 Band7 Band1 -0.107 -0.102 0.047 -0.441 0.884 -0.015 0.016 Band2 0.112 0.082 0.033 0.875 0.459 0.050 -0.006 Band3 -0.138 0.267 -0.925 -0.012 0.058 0.174 0.145 Band4 -0.089 0.855 0.274 -0.112 0.027 0.298 -0.289 Band5 0.052 0.329 -0.101 0.000 0.036 -0.932 -0.094 Band6 -0.006 -0.257 -0.215 0.020 0.008 0.027 -0.942 Band7 0.973 0.076 -0.101 -0.161 0.053 0.094 -0.004 Water Bodies Basic Stats Min Max Mean Stdev Num Eigenvalue Band1 -0.436 0.305 -0.277 0.182 1.000 1.775 Band2 -1.406 0.380 -0.761 0.501 2.000 0.114 Band3 -0.174 0.696 0.376 0.251 3.000 0.033 Band4 -1.306 1.594 -0.695 0.586 4.000 0.001 Band5 -1.365 8.820 2.453 1.110 5.000 0.000 Band6 -0.199 0.075 -0.013 0.025 6.000 0.000 Band7 -0.021 0.129 0.084 0.037 7.000 0.000 Eigenvector Band1 Band2 Band3 Band4 Band5 Band6 Band7 Band1 -0.132 -0.357 0.176 -0.385 0.822 -0.003 0.027

PAGE 219

208 Band2 -0.124 -0.229 0.100 -0.806 -0.519 -0.052 -0.002 Band3 -0.008 -0.758 0.442 0.416 -0.230 0.056 0.013 Band4 0.694 0.183 0.500 -0.095 0.043 -0.443 -0.168 Band5 0.555 -0.449 -0.692 -0.050 0.018 -0.089 0.010 Band6 -0.093 -0.050 -0.073 0.005 0.014 0.126 -0.984 Band7 -0.411 -0.087 -0.170 0.130 -0.008 -0.880 -0.057 Pasture outside park Basic Stats Min Max Mean Stdev Num Eigenvalue Band1 -0.169 0.293 0.082 0.066 1.000 0.540 Band2 -0.388 0.365 0.026 0.097 2.000 0.093 Band3 -0.131 0.415 0.144 0.065 3.000 0.000 Band4 -1.428 1.541 0.206 0.422 4.000 0.000 Band5 -0.759 5.552 1.213 0.660 5.000 0.000 Band6 -0.297 0.060 -0.069 0.039 6.000 0.000 Band7 -0.020 0.045 0.001 0.007 7.000 0.000 Eigenvector Band1 Band2 Band3 Band4 Band5 Band6 Band7 Band1 -0.080 -0.122 0.084 -0.449 0.877 -0.042 0.009 Band2 -0.096 -0.123 0.028 -0.863 -0.474 -0.070 -0.002 Band3 -0.348 -0.012 -0.744 -0.050 0.038 0.558 0.100 Band4 -0.365 -0.300 0.616 0.086 -0.060 0.615 -0.094 Band5 -0.064 -0.875 -0.151 0.173 -0.041 -0.296 0.297 Band6 0.474 -0.335 -0.179 -0.031 0.014 0.184 -0.772 Band7 -0.708 0.046 -0.071 0.111 -0.009 -0.427 -0.545 Savannah Basic Stats Min Max Mean Stdev Num Eigenvalue Band1 -0.086 0.102 0.004 0.044 1.000 0.156 Band2 -0.146 0.116 -0.015 0.059 2.000 0.052 Band3 -0.009 0.168 0.057 0.030 3.000 0.000 Band4 -0.510 0.546 0.014 0.224 4.000 0.000 Band5 0.460 3.224 1.239 0.388 5.000 0.000 Band6 -0.065 0.011 -0.031 0.012 6.000 0.000 Band7 0.008 0.048 0.023 0.007 7.000 0.000 Eigenvector Band1 Band2 Band3 Band4 Band5 Band6 Band7 Band1 -0.063 -0.102 0.067 -0.152 0.979 -0.004 0.013 Band2 0.158 0.190 -0.025 0.950 0.180 0.048 -0.012 Band3 -0.384 -0.109 -0.842 0.047 0.027 0.295 0.205 Band4 0.178 0.906 -0.251 -0.235 0.088 -0.020 -0.144 Band5 0.527 -0.190 -0.384 -0.028 0.030 -0.675 0.285

PAGE 220

209 Band6 -0.326 -0.103 -0.187 0.083 0.002 -0.458 -0.794 Band7 -0.638 0.272 0.202 0.092 -0.021 -0.494 0.474 Actively cultivated site Basic Stats Min Max Mean Stdev Num Eigenvalue Band1 -0.263 0.038 -0.101 0.084 1.000 0.506 Band2 -0.452 0.050 -0.158 0.128 2.000 0.048 Band3 -0.002 0.198 0.087 0.041 3.000 0.001 Band4 -0.677 0.222 -0.324 0.234 4.000 0.000 Band5 0.749 3.608 1.727 0.689 5.000 0.000 Band6 -0.068 -0.004 -0.032 0.015 6.000 0.000 Band7 0.015 0.079 0.042 0.019 7.000 0.000 Eigenvector Band1 Band2 Band3 Band4 Band5 Band6 Band7 Band1 -0.101 -0.160 0.053 -0.169 0.965 0.014 0.025 Band2 0.185 0.259 -0.050 0.919 0.226 0.033 -0.016 Band3 -0.496 -0.717 -0.127 0.315 -0.117 0.300 0.142 Band4 0.280 -0.382 0.843 0.114 -0.054 -0.163 -0.151 Band5 0.437 -0.395 -0.373 0.030 0.005 -0.597 0.396 Band6 0.602 -0.102 -0.036 -0.117 0.010 0.723 0.300 Band7 0.279 -0.284 -0.358 -0.017 0.020 0.056 -0.842 Anthropogenic vegetation Basic Stats Min Max Mean Stdev Num Eigenvalue Band1 -0.366 0.223 0.038 0.206 1.000 1.676 Band2 -0.995 0.301 0.017 0.389 2.000 0.038 Band3 -0.152 0.496 -0.008 0.174 3.000 0.007 Band4 -0.971 1.111 0.384 0.707 4.000 0.000 Band5 -0.547 2.650 0.589 0.998 5.000 0.000 Band6 -0.014 0.081 0.017 0.016 6.000 0.000 Band7 -0.007 0.084 0.024 0.032 7.000 0.000 Eigenvector Band1 Band2 Band3 Band4 Band5 Band6 Band7 Band1 -0.158 -0.291 0.126 -0.537 0.765 -0.011 0.024 Band2 0.117 0.409 -0.209 0.605 0.639 0.030 -0.003 Band3 -0.045 0.652 -0.520 -0.546 -0.061 -0.024 0.018 Band4 0.506 0.402 0.642 -0.170 0.034 -0.311 -0.208 Band5 0.783 -0.381 -0.458 -0.079 0.033 -0.148 0.047 Band6 0.275 0.064 0.120 -0.107 0.004 0.934 -0.150 Band7 0.119 0.111 0.185 -0.024 -0.009 0.086 0.965 Transitional Basic Stats Max Stdev Num Eigenvalue

PAGE 221

210 Min Mean Band1 -0.081 0.135 0.025 0.063 1.000 0.259 Band2 -0.151 0.143 0.018 0.078 2.000 0.043 Band3 -0.022 0.141 0.042 0.035 3.000 0.000 Band4 -0.412 0.763 0.185 0.267 4.000 0.000 Band5 0.426 2.661 1.252 0.469 5.000 0.000 Band6 -0.047 0.007 -0.018 0.011 6.000 0.000 Band7 0.008 0.048 0.023 0.010 7.000 0.000 Eigenvector Band1 Band2 Band3 Band4 Band5 Band6 Band7 Band1 -0.107 -0.141 0.059 -0.382 0.905 -0.011 0.018 Band2 0.148 0.142 0.026 0.886 0.413 0.033 -0.012 Band3 -0.399 0.136 -0.848 0.040 0.046 0.278 0.149 Band4 0.337 0.890 -0.078 -0.235 0.086 -0.103 -0.126 Band5 0.556 -0.294 -0.459 -0.021 0.029 -0.584 0.230 Band6 -0.528 0.248 0.195 0.081 -0.019 -0.576 0.531 Band7 -0.326 -0.040 -0.148 0.071 0.004 -0.488 -0.792 Table A-5. The separability va lues for the 2002 (August) classes (Jeffries-Matusita, Transformed Divergence) Water Bodies [Yellow] 10615 points: Non-anthropogenic forest [Maroon] 494 points: (2.00000000 2.00000000) Savannah [Purple] 292 points: (2.00000000 2.00000000) Inlet areas [Blue1] 305 points: (2.00000000 2.00000000) Culturally active [Blu e2] 402 points: (2.00000000 2.00000000) Transitional [Blue3] 115 points: (2.00000000 2.00000000) Bare soil [Yellow3 ] 26 points: (2.00000000 2.00000000) Anthropogenic vegetation [Green] 116 points: (2.00000000 2.00000000) Pasture outside park boundary [Cyan] 2549 points: (2.00000000 2.00000000) Aldea [Green1] 55 points: (2.00000000 2.00000000) Actively cultivated site [Red] 142 points: (2.00000000 2.00000000) Non-anthropogenic forest [Maroon] 494 points: Water Bodies [Yello w] 10615 points: (2.00000000 2.00000000) Savannah [Purple] 292 points: (2.00000000 2.00000000) Inlet areas [Blue1] 305 points: (1.45663532 1.87789118) Culturally active [Blu e2] 402 points: (1.99995519 2.00000000) Transitional [Blue3] 115 points: (1.56012337 1.94761477) Bare soil [Yellow3 ] 26 points: (2.00000000 2.00000000) Anthropogenic vegetation [Green] 116 points: (1.26343197 1.37078161) Pasture outside park boundary [Cyan] 2549 points: (2.00000000 2.00000000)

PAGE 222

211 Aldea [Green1] 55 points: (2.00000000 2.00000000) Actively cultivated site [Red] 142 points: (1.70913325 1.99990950) Savannah [Purple] 292 points: Water Bodies [Yello w] 10615 points: (2.00000000 2.00000000) Non-anthropogenic forest [Maroon] 494 points: (2.00000000 2.00000000) Inlet areas [Blue1] 305 points: (2.00000000 2.00000000) Culturally active [Blu e2] 402 points: (1.99999997 2.00000000) Transitional [Blue3] 115 points: (2.00000000 2.00000000) Bare soil [Yellow3 ] 26 points: (2.00000000 2.00000000) Anthropogenic vegetation [Green] 116 points: (2.00000000 2.00000000) Pasture outside park boundary [Cyan] 2549 points: (2.00000000 2.00000000) Aldea [Green1] 55 points: (1.99999960 2.00000000) Actively cultivated site [Red] 142 points: (1.99999988 2.00000000) Inlet areas [Blue1] 305 points: Water Bodies [Yello w] 10615 points: (2.00000000 2.00000000) Non-anthropogenic forest [Maroon] 494 points: (1.45663532 1.87789118) Savannah [Purple] 292 points: (2.00000000 2.00000000) Culturally active [Blu e2] 402 points: (1.98013285 2.00000000) Transitional [Blue3] 115 points: (0.92378415 0.99379191) Bare soil [Yellow3 ] 26 points: (1.99999959 2.00000000) Anthropogenic vegetation [Green] 116 points: (1.35877650 1.92027862) Pasture outside park boundary [Cyan] 2549 points: (2.00000000 2.00000000) Aldea [Green1] 55 points: (2.00000000 2.00000000) Actively cultivated site [Red] 142 points: (1.14340612 1.43507853) Culturally active [Blue2] 402 points: Water Bodies [Yello w] 10615 points: (2.00000000 2.00000000) Non-anthropogenic forest [Maroon] 494 points: (1.99995519 2.00000000) Savannah [Purple] 292 points: (1.99999997 2.00000000) Inlet areas [Blue1] 305 points: (1.98013285 2.00000000) Transitional [Blue3] 115 points: (1.95837189 1.99999983) Bare soil [Yellow3 ] 26 points: (1.90475146 2.00000000) Anthropogenic vegetation [Green] 116 points: (1.99976972 2.00000000) Pasture outside park boundary [Cyan] 2549 points: (1.99998766 2.00000000) Aldea [Green1] 55 points: (1.99952856 2.00000000) Actively cultivated site [Red] 142 points: (1.88989977 1.99238816) Transitional [Blue3] 115 points: Water Bodies [Yello w] 10615 points: (2.00000000 2.00000000) Non-anthropogenic forest [Maroon] 494 points: (1.56012337 1.94761477) Savannah [Purple] 292 points: (2.00000000 2.00000000) Inlet areas [Blue1] 305 points: (0.92378415 0.99379191) Culturally active [Blu e2] 402 points: (1.95837189 1.99999983) Bare soil [Yellow3 ] 26 points: (1.99999778 2.00000000)

PAGE 223

212 Anthropogenic vegetation [Green] 116 points: (1.69947247 1.99275216) Pasture outside park boundary [Cyan] 2549 points: (2.00000000 2.00000000) Aldea [Green1] 55 points: (2.00000000 2.00000000) Actively cultivated site [Red] 142 points: (1.20400230 1.52254514) Bare soil [Yellow3] 26 points: Water Bodies [Yello w] 10615 points: (2.00000000 2.00000000) Non-anthropogenic forest [Maroon] 494 points: (2.00000000 2.00000000) Savannah [Purple] 292 points: (2.00000000 2.00000000) Inlet areas [Blue1] 305 points: (1.99999959 2.00000000) Culturally active [Blu e2] 402 points: (1.90475146 2.00000000) Transitional [Blue3] 115 points: (1.99999778 2.00000000) Anthropogenic vegetation [Green] 116 points: (2.00000000 2.00000000) Pasture outside park boundary [Cyan] 2549 points: (2.00000000 2.00000000) Aldea [Green1] 55 points: (1.99999964 2.00000000) Actively cultivated site [Red] 142 points: (1.99412867 2.00000000) Anthropogenic vegetation [Green] 116 points: Water Bodies [Yello w] 10615 points: (2.00000000 2.00000000) Non-anthropogenic forest [Maroon] 494 points: (1.26343197 1.37078161) Savannah [Purple] 292 points: (2.00000000 2.00000000) Inlet areas [Blue1] 305 points: (1.35877650 1.92027862) Culturally active [Blu e2] 402 points: (1.99976972 2.00000000) Transitional [Blue3] 115 points: (1.69947247 1.99275216) Bare soil [Yellow3 ] 26 points: (2.00000000 2.00000000) Pasture outside park boundary [Cyan] 2549 points: (2.00000000 2.00000000) Aldea [Green1] 55 points: (2.00000000 2.00000000) Actively cultivated site [Red] 142 points: (1.57018403 1.99927390) Pasture outside park boundary [Cyan] 2549 points: Water Bodies [Yello w] 10615 points: (2.00000000 2.00000000) Non-anthropogenic forest [Maroon] 494 points: (2.00000000 2.00000000) Savannah [Purple] 292 points: (2.00000000 2.00000000) Inlet areas [Blue1] 305 points: (2.00000000 2.00000000) Culturally active [Blu e2] 402 points: (1.99998766 2.00000000) Transitional [Blue3] 115 points: (2.00000000 2.00000000) Bare soil [Yellow3 ] 26 points: (2.00000000 2.00000000) Anthropogenic vegetation [Green] 116 points: (2.00000000 2.00000000) Aldea [Green1] 55 points: (1.99986922 1.99999504) Actively cultivated site [Red] 142 points: (2.00000000 2.00000000) Aldea [Green1] 55 points: Water Bodies [Yello w] 10615 points: (2.00000000 2.00000000) Non-anthropogenic forest [Maroon] 494 points: (2.00000000 2.00000000) Savannah [Purple] 292 points: (1.99999960 2.00000000) Inlet areas [Blue1] 305 points: (2.00000000 2.00000000)

PAGE 224

213 Culturally active [Blu e2] 402 points: (1.99952856 2.00000000) Transitional [Blue3] 115 points: (2.00000000 2.00000000) Bare soil [Yellow3 ] 26 points: (1.99999964 2.00000000) Anthropogenic vegetation [Green] 116 points: (2.00000000 2.00000000) Pasture outside park boundary [Cyan] 2549 points: (1.99986922 1.99999504) Actively cultivated site [Red] 142 points: (1.99999927 2.00000000) Actively cultivated site [Red] 142 points: Water Bodies [Yello w] 10615 points: (2.00000000 2.00000000) Non-anthropogenic forest [Maroon] 494 points: (1.70913325 1.99990950) Savannah [Purple] 292 points: (1.99999988 2.00000000) Inlet areas [Blue1] 305 points: (1.14340612 1.43507853) Culturally active [Blu e2] 402 points: (1.88989977 1.99238816) Transitional [Blue3] 115 points: (1.20400230 1.52254514) Bare soil [Yellow3 ] 26 points: (1.99412867 2.00000000) Anthropogenic vegetation [Green] 116 points: (1.57018403 1.99927390) Pasture outside park boundary [Cyan] 2549 points: (2.00000000 2.00000000) Aldea [Green1] 55 points: (1.99999927 2.00000000) Table A-6. The separability va lues for the 2003 (May) classes (Jeffries-Matusita, Transformed Divergence) Water Bodies [Yellow] 10615 points: Non-anthropogenic forest [Maroon] 494 points: (1.99815024 2.00000000) Savannah [Purple] 292 points: (1.99179107 2.00000000) Inlet areas [Blue1] 305 points: (1.99762303 2.00000000) Culturally active [Blu e2] 402 points: (1.98447784 2.00000000) Transitional [Blue3] 115 points: (1.98329705 2.00000000) Bare soil [Yellow3 ] 26 points: (1.99736425 2.00000000) Anthropogenic vegetation [Green] 116 points: (1.79684895 1.99998233) Pasture outside park boundary [Cyan] 2549 points: (1.99989850 2.00000000) Aldea [Green1] 55 points: (1.99052248 2.00000000) Actively cultivated site [Red] 142 points: (1.94201223 2.00000000) Non-anthropogenic forest [Maroon] 494 points: Water Bodies [Yello w] 10615 points: (1.99815024 2.00000000) Savannah [Purple] 292 points: (1.99999346 2.00000000) Inlet areas [Blue1] 305 points: (1.07920289 1.76202451) Culturally active [Blu e2] 402 points: (1.99996033 2.00000000) Transitional [Blue3] 115 points: (1.99953199 2.00000000) Bare soil [Yellow3 ] 26 points: (2.00000000 2.00000000) Anthropogenic vegetation [Green] 116 points: (1.94755274 2.00000000)

PAGE 225

214 Pasture outside park boundary [Cyan] 2549 points: (1.99999960 2.00000000) Aldea [Green1] 55 points: (2.00000000 2.00000000) Actively cultivated site [Red] 142 points: (1.99999997 2.00000000) Savannah [Purple] 292 points: Water Bodies [Yello w] 10615 points: (1.99179107 2.00000000) Non-anthropogenic forest [Maroon] 494 points: (1.99999346 2.00000000) Inlet areas [Blue1] 305 points: (1.99999539 2.00000000) Culturally active [Blu e2] 402 points: (0.59405983 0.70226112) Transitional [Blue3] 115 points: (0.93992958 1.14855702) Bare soil [Yellow3 ] 26 points: (1.55669068 1.96453161) Anthropogenic vegetation [Green] 116 points: (1.99993939 2.00000000) Pasture outside park boundary [Cyan] 2549 points: (1.99973826 2.00000000) Aldea [Green1] 55 points: (1.86154996 1.99999970) Actively cultivated site [Red] 142 points: (1.61879949 1.99999989) Inlet areas [Blue1] 305 points: Water Bodies [Yello w] 10615 points: (1.99762303 2.00000000) Non-anthropogenic forest [Maroon] 494 points: (1.07920289 1.76202451) Savannah [Purple] 292 points: (1.99999539 2.00000000) Culturally active [Blu e2] 402 points: (1.99997316 2.00000000) Transitional [Blue3] 115 points: (1.99969323 2.00000000) Bare soil [Yellow3 ] 26 points: (2.00000000 2.00000000) Anthropogenic vegetation [Green] 116 points: (1.93948949 2.00000000) Pasture outside park boundary [Cyan] 2549 points: (1.99997083 2.00000000) Aldea [Green1] 55 points: (2.00000000 2.00000000) Actively cultivated site [Red] 142 points: (1.99999987 2.00000000) Culturally active [Blue2] 402 points: Water Bodies [Yello w] 10615 points: (1.98447784 2.00000000) Non-anthropogenic forest [Maroon] 494 points: (1.99996033 2.00000000) Savannah [Purple] 292 points: (0.59405983 0.70226112) Inlet areas [Blue1] 305 points: (1.99997316 2.00000000) Transitional [Blue3] 115 points: (0.44881983 0.50181532) Bare soil [Yellow3 ] 26 points: (1.69078470 1.99996587) Anthropogenic vegetation [Green] 116 points: (1.99959838 2.00000000) Pasture outside park boundary [Cyan] 2549 points: (1.99488084 1.99999996) Aldea [Green1] 55 points: (1.83189435 2.00000000) Actively cultivated site [Red] 142 points: (1.60733774 1.99976504) Transitional [Blue3] 115 points: Water Bodies [Yello w] 10615 points: (1.98329705 2.00000000) Non-anthropogenic forest [Maroon] 494 points: (1.99953199 2.00000000) Savannah [Purple] 292 points: (0.93992958 1.14855702) Inlet areas [Blue1] 305 points: (1.99969323 2.00000000) Culturally active [Blu e2] 402 points: (0.44881983 0.50181532)

PAGE 226

215 Bare soil [Yellow3 ] 26 points: (1.86435250 1.99999531) Anthropogenic vegetation [Green] 116 points: (1.99737630 2.00000000) Pasture outside park boundary [Cyan] 2549 points: (1.99554694 2.00000000) Aldea [Green1] 55 points: (1.86699267 2.00000000) Actively cultivated site [Red] 142 points: (1.72551744 1.99999777) Bare soil [Yellow3] 26 points: Water Bodies [Yello w] 10615 points: (1.99736425 2.00000000) Non-anthropogenic forest [Maroon] 494 points: (2.00000000 2.00000000) Savannah [Purple] 292 points: (1.55669068 1.96453161) Inlet areas [Blue1] 305 points: (2.00000000 2.00000000) Culturally active [Blu e2] 402 points: (1.69078470 1.99996587) Transitional [Blue3] 115 points: (1.86435250 1.99999531) Anthropogenic vegetation [Green] 116 points: (1.99999994 2.00000000) Pasture outside park boundary [Cyan] 2549 points: (2.00000000 2.00000000) Aldea [Green1] 55 points: (1.97474179 2.00000000) Actively cultivated site [Red] 142 points: (1.80953818 2.00000000) Anthropogenic vegetation [Green] 116 points: Water Bodies [Yello w] 10615 points: (1.79684895 1.99998233) Non-anthropogenic forest [Maroon] 494 points: (1.94755274 2.00000000) Savannah [Purple] 292 points: (1.99993939 2.00000000) Inlet areas [Blue1] 305 points: (1.93948949 2.00000000) Culturally active [Blu e2] 402 points: (1.99959838 2.00000000) Transitional [Blue3] 115 points: (1.99737630 2.00000000) Bare soil [Yellow3 ] 26 points: (1.99999994 2.00000000) Pasture outside park boundary [Cyan] 2549 points: (1.99989408 2.00000000) Aldea [Green1] 55 points: (1.99947366 2.00000000) Actively cultivated site [Red] 142 points: (1.99580144 1.99999993) Pasture outside park boundary [Cyan] 2549 points: Water Bodies [Yello w] 10615 points: (1.99989850 2.00000000) Non-anthropogenic forest [Maroon] 494 points: (1.99999960 2.00000000) Savannah [Purple] 292 points: (1.99973826 2.00000000) Inlet areas [Blue1] 305 points: (1.99997083 2.00000000) Culturally active [Blu e2] 402 points: (1.99488084 1.99999996) Transitional [Blue3] 115 points: (1.99554694 2.00000000) Bare soil [Yellow3 ] 26 points: (2.00000000 2.00000000) Anthropogenic vegetation [Green] 116 points: (1.99989408 2.00000000) Aldea [Green1] 55 points: (1.99990138 2.00000000) Actively cultivated site [Red] 142 points: (1.99992405 2.00000000) Aldea [Green1] 55 points: Water Bodies [Yello w] 10615 points: (1.99052248 2.00000000) Non-anthropogenic forest [Maroon] 494 points: (2.00000000 2.00000000) Savannah [Purple] 292 points: (1.86154996 1.99999970)

PAGE 227

216 Inlet areas [Blue1] 305 points: (2.00000000 2.00000000) Culturally active [Blu e2] 402 points: (1.83189435 2.00000000) Transitional [Blue3] 115 points: (1.86699267 2.00000000) Bare soil [Yellow3 ] 26 points: (1.97474179 2.00000000) Anthropogenic vegetation [Green] 116 points: (1.99947366 2.00000000) Pasture outside park boundary [Cyan] 2549 points: (1.99990138 2.00000000) Actively cultivated site [Red] 142 points: (1.74285492 1.99994819) Actively cultivated site [Red] 142 points: Water Bodies [Yello w] 10615 points: (1.94201223 2.00000000) Non-anthropogenic forest [Maroon] 494 points: (1.99999997 2.00000000) Savannah [Purple] 292 points: (1.61879949 1.99999989) Inlet areas [Blue1] 305 points: (1.99999987 2.00000000) Culturally active [Blu e2] 402 points: (1.60733774 1.99976504) Transitional [Blue3] 115 points: (1.72551744 1.99999777) Bare soil [Yellow3 ] 26 points: (1.80953818 2.00000000) Anthropogenic vegetation [Green] 116 points: (1.99580144 1.99999993) Pasture outside park boundary [Cyan] 2549 points: (1.99992405 2.00000000) Aldea [Green1] 55 points: (1.74285492 1.99994819)

PAGE 228

217 LIST OF REFERENCES Achard, F., H. Eva and P. Mayaux 2001 Tropical forest mapping from coarse spatial resolution satellite data: production and accuracy assessment issues. International Journal of Remote Sensing 22(14):2741-2762. Adams, J., D. Sabol, V. Kapos, R. Almeida, D. Roberts, M. Smith and A. Gillespie 1995 Classification of Multispectral Images Based on Fractions of Endmembers: Application to Land-C over Change in the Brazilian Amazon. Remote Sensing of Environment 52(2):137-154. Agrawal, A. and C. Gibson 1999 Enchantment and Disenchantment: The Role of Community in natural Resource Conservation. World Development 27(4):629-649. Aldenderfer, M. S. and H. D. G. Maschner 1996 Anthropology, Space and Geographic Information Systems . Spatial Information Series. Oxford University Press, New York. Allen, K. M. S., S. W. Green and E. B. W. Zubrow 1990 Interpreting Space: GIS and Archaeology . Taylor & Francis, London; New York. Allen, K. M. S., S. W. Green and E. B. Zubrow 1996 Iroquian Landscapes: People, Environments and the GIS Context . In New Methods, Old Problems: GIS in Modern Archaeological Research , edited by H. D. G. Maschner, pp. 198-222. Southern Illinois University. Allen, T. F. H. and T. B. Starr 1982 Hierarchy: Perspectives for Ecological Complexity . University of Chicago Press, Chicago. Altschul, J. H. 1989 Man, Models and Management: An Overview of the Archaeology of the Arizona Strip and the Management of its Cultural Resources, edited by US Bureau of Land Management, USDA Forest Service, Washington, D.C. Alves, D. S. 1999 An Analysis of the Geographical Patte rns of Deforestation in the Brazilian Amazon in the 1991-1996 Period. Proceedings of the 48th Annual Conference of the Center for Latin American Studies .

PAGE 229

218 Alves, D. S., J. Pereira, C. de Sousa, J. Soares and F. Yamaguchi 1999 Characterizing Landscape Changes in Central Rondonia Using Landsat TM Imagery. International Journal of Remote Sensing 20(14):2877-2882. Anderson, A. B. 1983 The Biology of Orbignya martiana (PALMAE), a Tropical Dry Forest Dominant in Brazil . PhD Dissertation, University of Florida, Gainesville. 1990 Alternatives to Deforestation: St eps Toward Sustainable Use of the Amazon Rain Forest . Columbia University Press, New York. 1992 Land-use Strategies for Successful Extractive Economies in Amazonia. Advances in Economic Botany 9:67-77. Anderson, J. R. 1976 A Land Use and Land Cover Classification System for Use With Remote Sensor Data . U.S. Govt. Printing Office, Washington, D.C. Anselin, L. 1989 What Is Special About Spatial Data ? Alternative Perspectives on Spatial Data Analysis . National Center for Geographic Information and Analysis, Santa Barbara, California. Arrow, K., B. Bolin, R. Costanza, P. Dasgupta and C. Folke 1995 Economic Growth, Carrying Capacity and the Environment. Science 268(5210):520. Ashmore, W. and A. B. Knapp 1999 Archaeologies of Landscape: Contemporary Perspectives . Blackwell, Malden, Mass. Asrar, G., R. B. Myneni and B. J. Choudhury 1992 Spatial Heterogeneity in Vegeta tion Canopies and Remote-Sensing of Absorbed Photosynthetically Ac tive Radiation—A Modeling Study. Remote Sensing of Environment 41(2-3):85-103. Baerwald, T. J. 1991 Social Sciences and Natural Resources. Renewable Resources Journal Autumn: 7-10. Baker, A. R. H. and G. Biger 1992 Ideology and Landscape in Histor ical Perspective; Essays on the Meanings of Some Places in the Past . Cambridge University Press, Cambridge; New York, NY.

PAGE 230

219 Baker, W. L. 1995 Longterm Response of Disturbanc e Landscapes to Human Intervention and Global Change. Landscape Ecology 10(3):18. Balee, W. 1987 Cultural Forests Of The Amazon. Garden 11:12-14. 1989 The Culture of Amazonian Forests. Advances in Economic Botany 7:1-21. 1993 Indigenous Transformation of Amaz onian Forests—An Example from Maranhao, Brazil. Homme 33(2-4):231-254. Balee, W. and D. Campbell 1990 Evidence For The Successional Status Of Liana Forests (Xing River Basin, Amazonian Brazil). Biotropica 22:36-47. Balee, W. L. 1998 Advances in Historical Ecology . Columbia University Press, New York. Balick, M. 1984 Ethnobotany of Palms in the Neotropics . In Enthobotany in the Neotropics , edited by G. Prance and J. Kallunki, pp. 9-23. New York Ethnobotanical Gardens, New York. Basso, K. H. 1996 Wisdom Sits in Places: Landscape and Language among the Western Apache . 1st ed. University of Ne w Mexico Press, Albuquerque. Bates, M. 1956 Summary Remarks: Process. In Man’s Role in Changing the Face of the Earth , edited by W. Thomas, pp. 1136-1140. University of Chicago Press, Chicago. Baxter, J. M. 1994 Exploratory Multivariate Analysis in Archaeology . Edinburgh University Press, London. Becker, B. K. 1982 Geopoltica da Amaznia: A Nova Fronteira de Recursos . A Terra e o Homem. Zahar Editores, Rio de Janeiro. Beckerman, S. 1979 The Abundance of Protein in Amazonia: A Reply to Gross. American Anthropologist 81:533-560. 1983 Does Swidden Ape the Jungle? Human Ecology 11(1):1-12.

PAGE 231

220 Behrens, C. A. 1992 A Formal Justification for the Application of GIS to the Culture Ecological Analysis of Land-use Intensif ication and Deforest ation in the Amazon . In The Anthropology of Human Behavi or through Geographic Information Analysis , edited by M. S. Aldenderfer and H. Maschner, pp. 16. University of California, Santa Barbara. 1994 Untitled—Introduction. Human Ecology 22(3):243-247. Behrens, C. A., M. G. Baksh and M. Mothes 1994 A Regional-Analysis of Bari LandUse Intensification and Its Impact on Landscape Heterogeneity. Human Ecology 22(3):279-316. Bellehumeur, C. and P. Legendre 1998 Multiscale Sources of Variation in Ecological Variables: Modeling Spatial Dispersion, Elaborating Sampling Designs. Landscape Ecology 13(1):11. Bender, B. 1993 Landscape: Politics and Perspectives . Explorations in Anthropology. Berg, Providence. Berkes, F., C. Folke and J. Colding 1998 Linking Social and Ecological System s: Management Practices and Social Mechanisms for Building Resilience . Cambridge University Press, Cambridge, New York, NY. Berry, B. 1964 Approaches to Regional Analysis. Annals of the Association of American Geographers 54:2-11. Berry, J. K. 1995 Spatial Reasoning for Effective GIS . GIS World Books, Fort Collins, Colorado. Bettinger, R. 1977 Predicting the Archaeological Po tential of the In yo-Mono Region of Eastern California . In Conservation Archaeology , edited by M. B. Schiffer, pp. 217-225. Academic Press, New York. Bian, L. and R. Butler 1999 Comparing Effects of Aggregati on Methods on Statis tical and Spatial Properties of Simulated Spatial Data. Photogrammetric Engineering and Remote Sensing 65(1):12.

PAGE 232

221 Bilsborrow, R. and D. Hogan (editor) 1996 Population and Deforestati on in the Humid Tropics . Oxford University Press, Oxford. Binswanger, H. P. 1991 Brazilian Policies that Encour age Deforestation in the Amazon. World Development 19(7):821-829. Boas, F. 1887 The Study of Geography. Science 9: 137-141. Boaz, J. and E. Uleberg 1995 The Potential of GIS-based Studies of Iron Age Cultural Landscapes in Eastern Norway . In Archaeology and Geographical Information Systems: A European Pperspective , edited by G. R. Lock and Z. Stancic, pp. 249-260. Taylor & Francis, London. Boserup, E. 1965 The Conditions of Agricultural Growth; The Economics of Agrarian Change Under Population Pressure . G. Allen & Unwin, London. Boyd, D. S., G. M. Foody, P. J. Curran, R. M. Lucas and M. Honzak 1996 An Assessment of Radiance in Landsat Tm Middle and Thermal Infrared Wavebands for the Detection Of Tropical Forest Regeneration. International Journal of Remote Sensing 17(2):249-261. Bradley, J. E., W. R. Killam, G. R. Burns and M. A. Martorano 1986 An Archaeological Survey and Pred ictive Model of Selected Areas of Utah's Cisco Desert. Cultural Resources Series 18 . Brandt, R., B. Groenewoudt and K. Kvamme 1992 An Experiment in Archaeologica l Site Location: Modeling in the Netherlands Using GIS Techniques. World Archaeology 24(2):268-282. Brondzio, E. S. 1996 Forest Farmers: Human and Landscape Ecology of Caboclo Populations in the Amazon Estuary . PhD Dissertation, University of Indiana, Bloomington. Brondizio, E. S., E. F. Moran, P. Mausel and Y. Wu 1994 Land-Use Change in the Amazon Estuary—Patterns of Caboclo Settlement and Landscape Management. Human Ecology 22(3):249-278. Brondizio, E., E. Moran, P. Mausel and Y. Wu 1996 Land Cover in the Amazon Estuary: Linking of the Thematic Mapper with Botanical and Historical Data. Photogrammetric Engineering and Remote Sensing 62(8):10.

PAGE 233

222 Brondizio, E., S. McCracken, E. Moran, A. Siqueira, D. Nelson and C. RodriguesPedraza 2002 The Colonist Footprint: Toward a Conceptual Framework of Land Use and Deforestation Trajectories Among Sm all Farmers in the Amazonian Frontier . In Land Use and Deforestation in the Amazon , edited by C. Wood and R. Porro. University of Florida Press, Gainesville. Browder, J. O. 1986 Logging the Rainforest: A Political Economy of Timber Extraction and Unequal Exchange in the Brazilian Amazon . PhD Dissertation, University of Pennsylvania. Browder, J. O. and B. J. Godfrey 1997 Rainforest Cities: Urbanization, D evelopment and Globalization of the Brazilian Amazon . Columbia University Press, New York. Brown, M. K. 1981 Predictive Models in Illinois Archaeology: Report Summaries . Illinois Dept. of Conservati on, Division of Histor ic Sites, Chicago. Brown, S. and A. Lugo 1990 Tropical Secondary Forests. Journal of Tropical Ecology 6:1-32. Bunker, S. G. 1985 Underdeveloping the Amazon: Extraction, Unequal Exchange and the Failure of the Modern State . University of Illinois Press, Urbana. Burkey, T. V. 1989 Extinction in Nature Reserves: Th e Effect of Fragmentation and the Importance of Migration Between Reserve Fragments. Oikos 55:75-81. Burrough, P. A. 1981 Fractal Dimensions of Landscap es and Other Environmental Data. Nature 294:240-242. 1990 Methods of Spatial Analysis in GIS. International Journal of Geographical Information Systems 4(3):221-223. Burrough, P. A. and A. U. Frank 1995 Concepts and Paradigms in Spatial Information: Are Current Geographical Information Systems Truly Generic? International Journal of Geographical Information Systems 9(2):101. Burrough, P. A., R. McDonnell and P. A. Burrough 1998 Principles of Geographical Information Systems . Spatial Information Systems. Oxford University Press, Oxford; New York.

PAGE 234

223 Burt, J. E., G. M. Barber and G. M. Barber 1996 Elementary Statistics for Geographers . 2nd ed. Guilford Press, New York. Buschmann, C. and E. Nagel 1991 Reflection Spectra of Terrestrial Vegetation as Influenced by PigmentProtein Complexes and the Intern al Optics of the Leaf Tissue . Paper presented at the International Geoscience and Remo te Sensing Symposium, Espoo, Finland. Bush, M., D. Piperno and P. Colinvaux 1989 A 6000 Year History of Am azonian Maize Cultivation. Nature 340:303305. Butzer, K. W. 1978 Towards and Integrated, Contex tual Approach in Archaeology. Journal of Archaeological Science 5:191-193. 1980 Context in Archaeology. Journal of Field Archaeology 7:417-422. 1982 Archaeology as Human Geology: Me thod and Theory for a Contextual Approach . Cambridge University Press, Cambridge. Campbell, D., G. Prance, U. Maciel 1986 Quantitative Ecological Inventory of Terra Firme and Vrzea Tropical Forest on the Rio Xingu, Brazilian Amazon. Brittonia 38(4):369-393. Campbell, J. B. 1981 Spatial Correlation Effects upon Accur acy of Supervised Classification of Land Cover. Photogrammetric Engineering and Remote Sensing 57:355-363. 1996 Introduction to Remote Sensing . 2nd ed. Guilford Press, New York. Campbell, J. B. and J. O. Browder 1995 Field Data-Collection for Remo te-Sensing Analysis Spot Data, Rondonia, Brazil. International Journal of Remote Sensing 16(2):333-350. Cao, C. and N. Lam 1997 Understanding the Scale and Resoultion Effects of Remote Sensing and GIS . In Scale in Remote Sensing and GIS , edited by D. Quattrochi and M. Goodchild, pp. 57-72. Lewis Publishers, Boca Raton. Carmichael, D. L. 1990 GIS Predictive Modeling of Prehis toric Site Distributions in Central Montana . In Interpreting Space: GIS and Archaeology , edited by K. M. S. Allen, S.W. Green and E. Zubrow, pp. 216-225. Taylor & Francis, London.

PAGE 235

224 1994 Sacred Sites, Sacred Places . One World Archaeology 23. Routledge, London; New York. Carneiro, R. 1960 Slash and Burn Agriculture: A Closer Look at its Implication for Settlement Patterns . In Men and Cultures , edited by A. F. C. Wallace, pp. 229234, Philidelphia. 1970 A Theory on the Origin of the State. Science 169:733-738. 1983 The Cultivation of Manioc Among th e Kuikuru Indians of the Upper Xingu . In Adaptive Responses of Native Americans , edited by R. Hames and W. Vikers, pp. 65-111. Academic Press, New York. 1985 Slash-and-Burn Cultivation among the Kuikuru and Its Implications for Cultural Development in the Amazon Basin . In Native South Americans: Ethnology of the Least Known Continent , edited by P. Lyon, pp. 73-91. Waveland Press, Illinois. 1987 Indians of the Amazonian Forest . In People of the Tropical Rain Forest , edited by J. S. Denslow and C. Padoch, pp. 25-36. University of California Press, Berkley. 1995 History of Ecological Interpretations of Amazonia: Does Roosevelt Have It Right? In Indigeous Peoples and the Future of Amazonia: An Ecological Anthropology of an Endangered World , edited by L. Sponsel, pp. 45-65. University of Arizona Press, Tuscon. Carr, C. and University of Arkansas Faye tteville (Institute for Quantitative Archaeology) 1985 For Concordance in Archaeological A nalysis: Bridging Data Structure, Quantitative Technique and Theory . Westport Publishers; Institute for Quantitative Archaeology University of Arkansas, Fayetteville. Carr, T. and M. Turner 1996 Investigating Regional Lithic Proc urment Using Multi-spectral Imagery and Geophysical Exploration. Archaeological Prospection 3:109-127. Castro, Eduardo V. d. 1984 Proposta pap um II Encontro Tupi. Revista de Antropologia 27-28:403407. 1992 From the Enemy's Point of View: Hu manity and Divinity in an Amazonian Society . University of Chicago Press, Chicago. 1996 Images of Nature and So ciety in Amazonian Ethnology. Annual Review of Anthropology 25:179-200.

PAGE 236

225 Castro, Eduardo V. d., M. C. d. Cunha and S. Dreyfus 1993 Amaznia: Etnologia e Histria Indgena . Ncleo de Histria Indgena e do Indigenismo: Fundao de Amparo Pesquisa do Estado de So Paulo, So Paulo, Brazil. Cecchi, G., P. Mazzinghi, L. Pantani, R. Valentini, D. Tirelli and P. Deangelis 1994 Remote-Sensing of Chlorophyll-a Fluorescence of Vege tation Canopies:1. Near and Far-Field Measurement Techniques. Remote Sensing of Environment 47(1):18-28. Chavez, P. S. 1996 Image-based Atmospheric Correction—Revisited and Improved. Photogrammetric Engineering and Remote Sensing 62(9):1025-1036. Chen, R. 1998 Appendix A . In People and Pixels: Linking Remote Sensing and Social Science , edited by R. Rindfuss, P. Stern and D. Liverman. National Academy Press, Washington, D.C. Chernela, J. M. 1993 The Wanano Indians of the Brazil ian Amazon: A Sense of Space . 1st ed. University of Texas Press, Austin. Cherry, J. F., J. L. Davis and E. Mantzourani 1991 Landscape Archaeology as Long-Term History: Northern Keos in the Cycladic Islands from Earliest Settlement until Modern Times . UCLA Institute of Archaeology, Los Angeles. Chomentowski, W., B. Salas and D. Skole 1994 Landsat Pathfined Project A dvances Deforestation Mapping. GIS World 7:34-38. Choudhury, B. J. 1987 Relationships between Vegetation In dexes, Radiation Absorption and Net Photosynthesis Evaluated by a Sensitivity Analysis. Remote Sensing of Environment 22(2):209-233. Christaller, W. and C. W. Baskin 1966 Central places in Southern Germany . Prentice-Hall, Englewood Cliffs, N.J. Church, T., R. Brandon and G. Burgett 2000 GIS Application in Archaeol ogy: Method in Search of Theory . In Practical Applications of GIS for Archae ologists: A Predicti ve Modeling Toolkit , edited by K.L. Westcott and R.J. Brandon, pp. 135-151. Taylor & Francis, London.

PAGE 237

226 Clarke, D. L. 1972 Models in archaeology . Methuen, London. 1977 Spatial Archaeology . Academic Press, London; New York. Clarke, D. L. and B. Chapman 1978 Analytical Archaeology . 2d ed. Columbia University Press, New York. Clastres, P. 1987 Society Against the State: E ssays in Political Anthropology . Zone Books; Distributed by The MIT Pr ess, Cambridge, Mass. Clay, J. W. 1988 Indigenous Peoples and Tropical Fo rests: Models of Land Use and Management from Latin America . Cultural Survival Report 27. Cultural Survival, Cambridge, Mass. Clements, F. E. 1916 Plant Succession: An Analysis of the Development Of Vegetation . Carnegie Institution of Washington, Washington. 1928 Plant Succession and Indicators; A Definitive Edition of Plant Succession and Plant Indicators . The H.W. Wilson Comp any, New York City. 1936 Nature and Structure of the Climax. The Journal of Ecology 24: 253-284. Cleveland, W. and R. McGill 1985 Graphical Perception and Graphica l Methods for Analyzing Scientific Data. Science 229:828-833. Cochrane, T. and P. Sanchez 1982 Land Resources, Soils and Their Management in the Amazon Region: A State of Knowledge Report . In Amazonia: Agriculture and Land-use Research , edited by S. Hecht, pp. 137-209. Proceedings of the Internation Conference on Amazonian Agriculture and Land-use Resear ch. Centro de Investigacion Agricola Tropical, California. Colby, J. D. and P. L. Keating 1998 Land Cover Classification Using La ndsat TM Imagery in the Tropical Highlands: The Influence of Anisotropic Reflectance. International Journal of Remote Sensing 19(8):1479-1500. Conant, F. P. 1994 Human Ecology and Space Age Technology: Some Predictions. Human Ecology 22(3):405.

PAGE 238

227 Conese, C. and F. Maselli 1991 Use of Multitemporal Information to Improve Classification Performance of TM Scenes in Complex Terrain. International Journal of Photogrammetry and Remote Sensing 46(4):187-197. Congalton, R. 1996 Accuracy Assessment: A Critical Component of Land Cover Mapping . In Gap Analysis: A Landscape Approach to Biodiversity Planning , edited by J. M. Scott, T.H. Tear and F.W. Davis, pp. 119-131. American Society for Photogrammetry and Remote Sensing, Betheda. Congalton, R. and K. Green 1998 Assessing the Accuracy of Remotely Sensed Data: Principles and Practices . Lewis Publishers, Washington, D.C. Congalton, R. G. 1991 A Review of Assessing the Accuracy of Classifications of Remotely Sensed Data. Remote Sensing of Environment 37(1):35-46. Congalton, R. G. and R. A. Mead 1983 A Quantitative Method to Test for Consistency and Correctness in Photointerpretation. Photogrammetric Engineering and Remote Sensing 49(1):6974. Congalton, R. G., R. G. Odeerwald and R. A. Mead 1983 Assessing Landsat Classification A ccuracy Using Discrete Multivariate Analysis Statistical Techniques. Photogrammetric Engineering and Remote Sensing 49:1671-1678. Cooke, R. and D. Harris 1970 Remote Sensing of the Terrestrial Environment: Principles and Progress. Transactions of the Institute of British Geographers (50):1-23. Cooney, G. 1999 Social Landscapes in Irish Prehistory . In The Archaeology and Anthropology of Landscape , edited by P. J. Ucko and R. Layton, pp. 46-64. Routledge, London. Cordell, L. S. 1997 Archaeology of the Southwest . 2nd edition. ed. Academic Press, San Diego, Calif. Corlett, R. T. 1995 Tropical Secondary Forests. Progress in Physical Geography 19(2):159.

PAGE 239

228 Costanza, R., L. Wainger, C. Folke and K.-G. Maler 1993 Modeling Complex Ecological Economic Systems. BioScience 43(8):545. Coulson, R., C. Lovelady, R. Flamm, S. Spradling and M. Saunders 1991 Intelligent Geographic Information Systems for Natural Resource Management . In Quantitative Methods in Landsc ape Ecology: The Analysis and Interpretation of Landscape Heterogeneity , edited by M. Turner and R. Gardner, pp. 153-172. Springer-Verlag, New York. Cowgill, G. 1990 Toward Refining Concepts for Full-coverage Survey . In The Archaeology of Regions: A Case for Full-Coverage Survey , edited by S. K. a. S. K. Fish, pp. 249-260. Smithsonian Institution Press, Washington, D.C. Cox, C. 1992 Satellite Imagery, Aerial Photogr aphy and Wetland Archaeology: An Iterim Report on an Application of Remote Sensing to Wetland Archaeology. World Archaeology 24(2):249-267. Cracknell, A. P. and L. Hayes 1991 Introduction to Remote Sensing . Taylor & Francis, London; New York. Crist, E. P. and R. C. Cicone 1984 Application of the Tasseled Cap Concept to Simulated Thematic Mapper Data. Photogrammetric Engineering and Remote Sensing 50(3):343-352. Crist, E. P. and R. J. Kauth 1986 The Tasseled Cap De-Mystified. Photogrammetric Engineering and Remote Sensing 52(1):81-86. Crumley, C. 1994 Historical ecology: cultural knowledge and changing landscapes . 1st ed. School of American Research advanced seminar series. School of American Research Press, Santa Fe. Crumley, C. and W. Marquardt 1990 Landscape: A Unifying C oncept in Regional Analysis . In Interpreting Space: GIS and Archaeology , edited by K. M. S. Allen, S.W. Green and E. Zubrow, pp. 73-79. Taylor & Francis, London. Custer, J., T. Eveleigh, V. Klemas, I. Wells 1986 Application of LANDSAT Data and Synoptic Remote Sensing to Predictive Models for Prehistoric Archaeological Sites. American Antiquity 51(3):573-588.

PAGE 240

229 Dale, V. and S. Pearson 1997 Quantifying Habitat Fragmentation Due to Land-use Change in Amazonia . In Tropical Forest Remnants: Ecolog y, Management and Conservation of Fragmented Communitites , edited by W. Laurance and R. Bierregaard, pp. 400409. Univeristy of Chicago Press, Chicago. Dale, V. H., R. V. O'Neill, M. Pedlowski and F. Southworth 1993 Causes and Effects of Land-Use Change in Central Rondonia, Brazil. Photogrammetric Engineering and Remote Sensing 59(6):997. Dale, V. H., R. V. O'Neill, F. Southworth and M. Pedlowski 1994 Modeling Effects of Land Manage ment in the Brazilian Amazonian Settlement of Rondonia. Conservation Biology 8(1):196-206. Dale, V. H., S. M. Pearson, H. L. Offerman and R. V. O'Neill 1994 Relating Patterns of Land-Use Ch ange to Faunal Biodiversity in the Central Amazon. Conservation Biology 8(4):1027-1036. Dalla, B. L. 1996 Modeling Prehistoric Land Use in Northern Ontario . In New Methods, Old Problems: Geographic Information System s in Modern Archaeological Research , edited by H. D. G. Maschner. Southe rn Illinois University, Springfield. 2000 Protecting Cultural Resources Throu gh Forest Management Planning in Ontario Using Archaeological predictive Modeling . In Practical Applications of GIS for Archaeologists: A Predictive Modeling Kit , edited by K. L. Westcott and R.J. Brandon, pp. 77-99. Taylor and Francis, London. Dangermond, J. 1992 What is Geographic Information Systems (GIS)? In Geogrpahic Information Systems (GIS) and M apping: Practicies and Standards , edited by A. Johnson, C. Petterson and J. Fulton, pp. 1117. American Society for Testing and Materials, Philadelphia. Davis, F. W., D. M. Stoms, J. E. Estes, J. Scepan and J. M. Scott 1990 An Information Systems Approach to the Preservation of Biological Diversity. International Journal of Ge ographical Information Systems 4(1):55. Davis, J. C. 1986 Statistics and Data Analysis in Geology . 2nd ed. Wiley, New York. de Castro, F., M. C. Silva-Forsberg, W. Wilson, E. Brondizio and E. Moran 2002 The Use of Remotely Sensed Data in Rapid Rural Assessment. Field Methods 14(3):243-269.

PAGE 241

230 Dean, J. S. 1983 Environmental Aspects of Modeling . In Theory and Modeling: Refining Survey Strategies for Locating Prehistoric Heritage Resources , edited by L. S. a. J. S. G. Cordell, pp. 11-27. Cultural Resources Document 3. Southwestern Region: Forest Service, USDI. Decker, D. 2001 GIS Data Sources . J. Wiley, New York. DeMers, M. N. 2000 Fundamentals of Geographic Information Systems . 2nd ed. J. Wiley, New York. Denevan, W. 1976 The Aboriginal Population of Amazonia . In The Native Population of the Americas in 1492 , edited by W. Denevan, pp. 205-234. University of Wisconsin Press, Madison. 1984 Ecological Heterogeneity and Horizont al Zonation of Agriculture in the Amazon Floodplain . In Floodplain Expansion in Amazonia , edited by M. Schmink and C. Wood. University of Florida Press, Gainesville. 1992 Stone Versus Metal Axes. Journal of the Steward Anthropological Society 20(1&2):153-165. 1996 A Bluff Model of Riverine Sett lement in Prehistoric Amazonia. Annals of the Association of American Geographers 86(4):28. Denevan, W. M. and C. Padoch 1987 Swidden-fallow Agroforest ry in the Peruvian Amazon . Advances in Economic Botany, v. 5. New York Botanical Garden, Bronx, N.Y. Dickinson, R. E. and United Nations University. 1987 The Geophysiology of Amazonia: Ve getation and Climate Interactions . Wiley Series in Climate and the Biosphere. Wiley for the United Nations University, New York. Domon, G., M. Garipy and A. Bouchard 1989 Ecological Cartography and Land-Use Planning: Trends and Perspectives. Geoforum 20(1):69-82. Douglas, I. 1999 Hydrological Investigations of Forest Disturbance and Land Cover Impacts in South-East Asia: A Review. Philosophical Transactions of the Royal Society of London Series B-Biological Sciences 354(1391):1725-1738.

PAGE 242

231 Dunnell, R. and W. Dancey 1983 The Siteless Survey: A Regional Scale Data Collection Strategy . In Advances in Archaeological Method and Theory , edited by M. B. Schiffer, pp. 267-287. vol. 6. Taylor & Francis, London. Ebert, J. I. 1992 Distributional Archaeology . 1st ed. University of New Mexico Press, Albuquerque. 2000 The State of the Art in "Inductive" Predictive Modeling . In Practical Applications of GIS for Archaeol ogists: A Predictive Modeling Kit , edited by K. L. Westcott and R.J. Brandon, pp. 129-134. Taylor and Francis, London. Ebert, J. I. and T. Kohler 1988 The Theorectical Basis of Arch aeological Predictive Modeling and a Consideration of Appropri ate Data Collection Methods . In Quantifying the Present and Predicting the Past: Th eory Method and Application of Archaeological Predictive Modeling , edited by J. Judge and L. Sebastian, pp. 97172. Bureau of Land Management, Denver. Ebert, J. I. and T. Lyons 1976 The Role of Remote Sensing in Regional Archaeological Research Design . In Remote Sensing Experiments in Cultural Resource Studies , edited by T. Lyons, pp. 5-10. University of New Mexico Press, Albuquerque. Ebert, J. I., E.L. Camilli and M.J. Berman 1996 GIS in the Analysis of Di stributional Archaeological Data . In New Methods, Old Problems: GIS in Modern Archaeological Research , edited by H. D. G. Maschner, pp. 25-37. Southern Illinois Universi ty, Springfield. Effland, R. 1979 Statistical Distribution Ca rtography and Computer Graphics . In Computer Graphics in Archaeology , edited by S. Upham, pp. 17-29. Anthropological Research Papers No. 15. University of Arizona Press, Tempe. Ehlers, M. 1992 Remote Sensing and Geographic Info rmation Systems: Image-Integrated Geographic Information Systems. Geographic Information Systems (GIS) and Mapping-Practices and Standards ASTM STP 1126:53-67. Eidt, R. 1977 Detection and Examination of Anthrosols by Phosphate Analysis. Science 197:1327-1333.

PAGE 243

232 Elvidge, C. D. and Chen, Z. 1995 Comparison of Broad-Band and Na rrow-Band Red and Near-Infrared Vegetation Indices. Remote Sensing of Environment 54:38-48. Erickson, C. 1995 Archaeological Methods for the St udy of Ancient Landscapes of the Llanos de Mohos in the Bolivian Amazon . In Archaeology in the Lowland American Tropics: Current Analyt ical Methods and Applications , edited by P. W. Stahl, pp. 66-95. Cambridge University Press, New York. Eva, H. and E. F. Lambin 1998 Remote Sensing of Biomass Bu rning in Tropical Regions: Sampling Issues and Multisensor Approach. Remote Sensing of Environment 64:292-315. Falesi, I. C. 1974 Soils of the Brazilian Amazon . In Man in the Amazon , edited by C. Wagley, pp. 201. University of Florida Press, Gainesville. Faust, N. L., W. Anderson and J. Star 1991 Geographic Information Systems an d Remote-Sensing Future Computing Environment. Photogrammetry Engineering and Re mote Sensing Experiments in Cultural Resource Studies 57:655-668. Fedje, D. and T. Christensen 1999 Modeling Paleoshorelines and Locati ng Middle Holocene Coastal Sites in Haida Gwaii. American Antiquity (64):635-652. Fish, S. K. and S. A. Kowalewski 1990 The Archaeology of Regions: A Case for Full-Coverage Survey . Smithsonian Series in Archaeological Inquiry. Smithsonian Institution Press, Washington; London. Fisher, A. 1994 A Model for the Seasonal Variati on of Vegetation Indices in Coarse Resolution Data and its Inversi on to Extract Crop Parameters. Remote Sensing of Environment 48:220-230. Flannery, K. V. 1976 The Early Mesoamerican Village . Academic Press, New York. Flowers, N. 1994 Subsistence Strategy, Social Organiza tion and Warfare in Central Brazil in the Context of European Penetration . In Amazonian Indians from Prehistory to the Present: Anthropological Perspectives , edited by A. C. Roosevelt, pp. 249269. University of Arizona Press, Tuscon.

PAGE 244

233 Foody, G. M. and P. J. Curran 1994 Estimation of Tropical Forest Ex tent and Regenerative Stage Using Remotely-Sensed Data. Journal of Biogeography 21(3):223-244. Foody, G. M., G. Palubinskas, R. M. Lucas, P. J. Curran and M. Honzak 1996 Identifying Terrestrial Carbon Sinks: Classification of Successional Stages In Regenerating Tropical Fore st from Landsat TM Data. Remote Sensing of Environment 55(3):205-216. Foresman, T. W. 1998 The History of Geographic Informati on Systems: Perspectives from the Pioneers . Prentice Hall Series in Geographic Information Science. Prentice Hall PTR, Upper Saddle River, NJ. Forman, R. T. T. and M. Godron 1986 Landscape Ecology . Wiley, New York. Forster, B. 1983 Some Urban Measurements from Landsat Data. Photogrammetric Engineering and Remote Sensing 49:1693-1707. Fortin, M. J. 1999 Spatial Statistics in Landscape Ecology . In Landscape Ecological Analysis: Issues and Applications , edited by J. Klopatek and R. H. Gardner, pp. 253. Springer-Verlag, New York. Franchetto, B. and M. Heckenberger 2000 Os Povos do Alto Xingu: Histria e Cultura . Editora UFRJ, Rio de Janeiro. Freeman, A. 2000 Application of High-re solution Alluvial Stratigraphy in Assessing the Hunter-Gatherer/Agricultural Transiti on in the Santa Cruz River Valley. Geoarchaeology 15(6):559-589. Frohn, R. C. 1998 Remote Sensing for Landscape Ecol ogy: New Metric Indicators for Monitoring, Modeling and A ssessment of Ecosystems . Lewis Publishers, Boca Raton. Fullagar, R. and L. Head 1999 Exploring the Prehistory of H unter-gatherer Attachments to Place . In The Archaeology and Anthropology of Landscape , edited by P. J. Ucko and R. Layton, pp. 322-335. Routledge, London.

PAGE 245

234 Fuller, R., J. Sheail and C. Barr 1994 The Land of Britian, 1930-1990: A Comparative Study of Field Mapping and Remote Sensing Techniques. Geographical Journal 160(2):173-184. Gaffney, V. and M. van Leusen 1995 Postcript . In Archaeology and Geographical Information Systems: a European Perspective , edited by G. R. Lock and Z. Stancic, pp. 367-382. Taylor & Francis, London. Gaffney, V., Z. Stancic and H. Watson 1995 The Impact of GIS on Arch aeology: A Personal Perspective . In Archaeology and GIS: A European Perspective , edited by G. R. Lock and Z. Stancic, pp. 211-229. Taylor & Francis, London. 1996 Moving from Catchments to Cognition: Tenative Steps Toward a Larger Archaeological Context for GIS . In Anthropology, Space and Geographic Information Systems , edited by M. S. Aldenderfer and H. Maschner, pp. 132-154. Oxford University Press, Oxford. Gardner, R. H. and M. G. Turner 1991 Future Directions in Quantitative Landscape Ecology . In Quantitative Methods in Landscape Ecology: The Anal ysis and Interpretation of Landscape Heterogeneity , edited by M. G. Turner and R. H. Gardner, pp. 519-526. SpringerVerlag, New York. Gardner, W. M. 1978 Comparison of Ridge and Valley, Blue Ridge, Piedmont and Coastal Plain Archaic Period Site Distribution: An Id ealized Transect (P reliminary Model) . In Middle Atlantic Archaeological Conference , Rehoboth Beach, Delaware. Geoghegan, J. 1998 Socializing the Pixe l and Pixeling the Social . In People and Pixels: Linking Remote Sensing and Social Science , edited by D. M. Liverman. National Academy Press, Washington, D.C. Gholz, H. L. 1982 Environmental Limits on Above Ground Net Primary Production, Leaf Area and Biomass in Vegetation Zones of the Pacific Northwest. Ecology 63(2):469-481. Gholz, H. L., Nakane, K. and Shimoda, H. 1997 The Use of Remote Sensing in th e Modeling of Forest Productivity . Forestry Sciences Vol. 50. Kluw er Academic Publishers, Boston.

PAGE 246

235 Gillespie, A. R., Kahle, A.B. and Walker, R.E. 1987 Color Enhancement of Highly Corr elated Images: Decorrelation And Hsi Contrast Stretches. Remote Sensing of Environment 20:209-235. Gillings, M. 1995 Flood Dynamics and Settlement in the Tisza Valley Northeast of Hungary: GIS and the Upper Tisza Project . In Archaeology and Geographical Information Systems: A European Perspective , edited by G. R. Lock and Z. Stancic, pp. 67-84. Taylor & Francis, London. Glaser B, B. E., Haumaier L, Guggenberger G, Zech W 2000 Black Carbon in Density Fractions of Anthropogenic Soils of the Brazilian Amazon Region. Organic Geochemisty 31:669-678. Gonzalo, A. 1999 The Perception of Landscape Among the Q'eqchi' . In The Archaeology and Anthropology of Landscape , edited by P. J. Ucko and R. Layton, pp. 254-263. Routledge, London. Goodchild, M., R. Haining and S. Wise and 12 others 1992 Integrating GIS and Spatial Data An alysis: Problems and Possibilities. International Journal of Ge ographical Information Systems 6(5):407-424. Goodchild, M. F. 1992 Geographical Information Science. International Journal of Geographical Information Systems 6(1):31-46. 1996 GIS and Spatial Analysis in the Social Sciences . In Anthropology, Space and Geographic Information Systems , edited by M. S. Aldenderfer and H. Maschner. Oxford University Press, New York. Goodchild, M. F. and National Center for Geographic Information & Analysis (U.S.) 1992 Spatial Analysis Using GIS: Seminar Workbook . 2nd ed. National Center for Geographic Information & Analysis, Santa Barbara, CA. Goodchild, M. F., B. O. Parks and L. T. Steyaert 1993 Environmental Modeling with GIS . Oxford University Press, New York. Goodland, R. J. A. and H. S. Irwin 1975 Amazon Jungle: Green Hell to Red De sert? An Ecological Discussion of the Environmental Impact of the High way Construction Program in the Amazon Basin . Developments in Landscape Manageme nt and Urban Pla nning; 1. Elsevier Scientific, Amsterdam; New York.

PAGE 247

236 Gould, P. 1967 On the Geographical Interpretation of Eigenvalues. Transactions, Institute of British Geographers 42:53-86. Gow, P. 1995 Land, People and Paper in Western Amazonia . In The Anthropology of Landscape: Perspectives on Place and Space , edited by E. Hirsch and M. O'Hanlan, pp. 43-62. Clarendon Press, Oxford. Grant, L. 1987 Diffuse and Specular Characte ristics of Leaf Reflectance. Remote Sensing of Environment 22(2):309-322. Green, E. L. 1973 Location Analysis of Prehistoric Ma ya Sites in Northern British Honduras. American Antiquity 38:279-293. Green, S. W. 1990 Sorting out Settlement in South eastern Ireland: Landscape Archaeology and GIS . In Interpreting Space: GIS and Archaeology , edited by K. M. S. Allen, S.W. Green and E. Zubrow, pp. 356-363. Taylor & Francis, London. Gregor, T. 1977 Mehinaku: the Drama of Daily Life in a Brazilian Indian Village . University of Chicago Press, Chicago. Gross, D. 1975 Protein Capture and Cultural Development in the Amazon Basin. American Anthropologist 77(3):526. Gross, D. R., G. Eiten, N. M. Flowers, F. M. Leoi, M. L. Ritter and D. W. Werner 1979 Ecology and Acculturation among Na tive Peoples of Central Brazil. Science 206(30):1043-1050. Gumerman, G.J. and T. Lyons 1971 Archaeological Methodology and Remote Sensing. Science 172(3979):126-132. Gunn, J. 1979 Occupation Frequency Simulation on a Broad Ecotone . In Transformations: Mathematical Approaches to Culture Change , edited by C. Renfrew and Kenneth L. Cooke. Academic Press, New York. Guyot, G. and X. F. Gu 1994 Effect of Radiometric Corrections on NDVI-Determined from Spot-HRV and Landsat TM Data. Remote Sensing of Environment 49(3):169-180.

PAGE 248

237 Haffer, J. 1991 Mosaic Distribution Patterns of Ne otropical Forest Birds and Underlying Cyclic Disturbance Processes . In The Mosaic-Cycle C oncept of Ecosystems , edited by H. Remmert, pp. 83-105. Ecologica l Studies. Springer-Verlag, Berlin. Haines-Young, R., D. R. Green and S. H. Cousins (editor) 1996 Landscape Ecology and Geographic Information Systems . Taylor & Francis, London. Hames, R. 1983 The Settlement Pattern of a Yanomamo Population Bloc . In Adaptive Responses of Native Americans , edited by R. Hames and W. Vikers, pp. 192-229. Academic Press, New York. Hames, R. and W. Vikers (editor) 1983 Adaptative Responses of Native Amazonians . Academic Press, New York. Harris, T. and G. Lock 1995 Towards and Evaluation of GIS in European Archaeology . In Archaeology and Geographical Information Systems: A European Perspective , edited by G. R. Lock and Z. Stancic, pp. 349356. Taylor & Francis, London. Hasenstab, R. and B. Resnick 1990 GIS in Historical Predictive Modeling . In Interpreting Space: GIS and Archaeology , edited by K. M. S. Allen, S.W. Green and E. Zubrow, pp. 284-306. Taylor & Francis, London. Hecht, S. B. and A. Cockburn 1990 The Fate of the Forest: Develope rs, Destroyers and Defenders of the Amazon . 1st U.S. ed. Harper Perennial, New York, NY. Heckenberger, M. J. 1996 War and Peace in the Shadow Of Empi re: Sociopolitical Change in the Upper Xingu of Southeastern Amazonia, A.D. 1400-2000 . PhD Dissertation, University of Pittsburg. 1998 Manioc Agriculture and Seden tism in Amazonia: The Upper Xingu Example. Antiquity 72:633-648. 1999 Enigma das Grandes Cidades: Co rpo Privado e Estado em Amazonia . In A Outra Margem do Occidente , edited by A. Novaes, pp. 125-152. Companhia das Letras, Sao Paulo. 2005 The Ecology of Power: Culture, Place and Personhood in the Southern Amazon, A.D. 1000-2000 . Routledge, New York; London.

PAGE 249

238 Heckenberger, M., J. Petersen and E. G. Neves 2001 Of Lost Civilizations and Primitiv e Tribes, Amazonia: Reply to Meggers. Latin American Antiquity 12(3):328-333. Heckenberger, M., J. Petersen, E. Neves 1999 Village Permanence in Amazonia: Two Archaeological Examples from Brazil. Latin American Antiquity 10(4):353-376. Heckenberger, M., A. Kuikuro, U. Kuikuro, J. C. Russell, M. Schmidt, C. Fausto and B. Franchetto 2003 Amazonia 1492: Pristine Forest or Cultural Parkland? Science 301(5640):4. Helmer, E. H., S. Brown and W. B. Cohen 2000 Mapping Montane Tropical Forest Successional Stage and Land Use With Multi-Date Landsat Imagery. International Journal of Remote Sensing 21(11):2163-2183. Herrera, L. F., L. Cavelier, C. Rodriguesz and S. Mora 1992 The Technical Transformation of an Agricultrual System in the Colombian Amazon. World Archaeology 24:98-113. Hill, J. D. 1988 Rethinking History and Myth: Indige nous South American Perspectives on the Past . University of Illinois Press, Urbana. Hill, J. D. and F. Santos-Granero 2002 Comparative Arawakan Histories: Rethinking Language Family and Culture Area in Amazonia . University of Illinois Press, Urbana. Hill, J. D. and B. Sturm 1991 Radiometric Correction of Multitem poral Thematic Mapper Data for Use in Aricultural Land-Cover Classi fication and Vegetation Monitoring. International Journal of Remote Sensing 12(7):1471-1491. Hill, R. A. 1999 Image Segmentation for Humid Tropical Forest Classification in Landsat TM Data. International Journal of Remote Sensing 20(5):1039-1044. Hill, R. A. and G. M. Foody 1994 Separability of Tropical Rain-For est Types in the Tambopata-Candamo Reserved Zone, Peru. International Journal of Remote Sensing 15(13):2687-2693. Hofmann-Wellenhof, B., H. Lichtenegger and J. Collins 1994 Global Positioning System: Theory and Practice . 3rd ed. Springer-Verlag, Wien; New York.

PAGE 250

239 Hord, R. M. and W. Brooner 1976 Land-Use Map Accuracy Criteria. Photogrammetric Engineering and Remote Sensing 42:671-677. Huete, A., C. Justice, and H. Liu 1994 Development of Vegetation a nd Soil Indices for MODIS EOS. Remote Sensing of Environment 49:224-234. Huete, A. 1988 A Soil-Adjusted Vegetation Index (SAVI). Remote Sensing of Environment 25:295-309. Hunt, E. D. 1992 Upgrading Site-catchment Analyses w ith the Use of GIS: Investigating the Settlement Patterns of Horiticulturalists. World Archaeology 24(2):283-309. Jackson, R. D. 1983 Spectral Indices in n-Space. Remote Sensing of Environment 13:409-421. Jacoli, M. and A. Carrara 1996 GIS-based Multivariate Models fo r Identifying Archaeological Sites, Calabria, Southern Italy. Proceedings of the International Congress on "Science and Technology for the Safeguard of Cultu ral Heritage in The Mediterranean Basin" . Nov. 27-Dec. 2, 1995, Catania. James, S. E. and R. Knudson 1983 Predicting Site Significance: Management Applications of HighResolution Modeling . Paper presented at the 48th Annual Meeting of the Society for American Archaeology, Pittsburgh. Janssen, L. and F. van der Wel 1994 Accuracy Assessment of Satellit e Derived Land Cover Data: A Review. Photogrammetric Engineering and Remote Sensing 60:419-426. Jensen, J. R. 1986 Introductory Digital Image Processi ng: A Remote Sensing Perspective . Prentice-Hall, Upper Saddle River, N.J. 2000 Remote Sensing of the Environment: An Earth Resource Perspective . Prentice Hall, Upper Saddle River, N.J. Jensen, R. R., G. Yu, P. W. Mausel , V. Lulla, E. Brondizio and E. Moran 2004 An Integrated Approach to Amazon Research: The Amazon Information System. Geocarto International 19(3):5.

PAGE 251

240 Jochim, M. A. 1976 Hunter-Gatherer Subsistence and Se ttlement: A Predictive Model . Academic Press, New York. Johannsen, C. J., J. L. Sanders, and Soil Conservation Society of America. 1982 Remote Sensing for Resource Management . Soil Conservation Society of America, Ankeny, Iowa. Johansen, M. E., Tommervik, H., Gune riussen, T. and Pedersen, J. P. 1994 Using a GIS (ArcInfo) as a Tool for Integration of Remote Sensed and InSitu Data in an Analysis of the Air Po llution Effects on Terrestrial Ecosystems in Varanger (Norway) and Nikel-Pechenga (Russia). Proceedings of International Geoscience and Remote Sensing Symposium 2:1213-1216. Johnson, A. W. and T. K. Earle 1987 The Evolution of Human Societies : From Foraging Group to Agrarian State . Stanford University Press, Stanford, Calif. Johnson, G. 1977 Aspects of Regional Analysis in Archaeology. Annual Review of Anthropology 6:479-508. 1981 Monitoring Complex Systems Integr ation and Boundary Phenomena with Settlement Size Data . In Archaeological Approaches to the Study of Complexity , edited by S. E. van der Leeuw, pp. 143-188. Giffen Instituut woor Prae-en Protohistorie, Amsterdam. Jordan, C. F. 1969 Derivation of Leaf-Area Index from Quality Combinations for Monitoring Vegetation. Ecology 50:663-666. 1989 An Amazonian Rain Forest: The Stru cture and Function of a Nutrient Stressed Ecosystem and the Impac t of Slash-and-Burn Agriculture . Man and the Biosphere Series; v. 2. UNESCO; Pa rthenon Pub. Group, Paris, France. Joseph, J. W. 2000 Archaeology and the African-Ameri can Experience in the Urban South . In Archaeology of Southern Urban Landscapes , edited by A. L. Young, pp. 109-126. University of Alabama Press, Tuscaloosa. Judge, J. and L. Sebastian (editor) 1988 Quantifying the Present and Predic ting the Past: Theory, Method and Application of Archaeol ogical Predictive Modeling . Bureau of Land Management, Denver.

PAGE 252

241 Judge, W. J. 1973 Paleoindian Occupation of the Cent ral Rio Grande Valley in New Mexico . 1st ed. University of New Mexico Press, Albuquerque. Jurdant, M. 1977 L'inventaire du Capital-Nature: M thode de Classification et de Cartographie cologique du Territo ire: 3me Approximation . Srie de la Classification cologique du Territoire; no 2. Service des tudes cologiques Rgionales Direction Rgiona le des Terres Pches et Environnement Canada, Qubec. Kalkahan, M. A., R. M. Reich and R. L. Czaplewski 1997 Variance Estimates and Confidence Intervals for the Kappa Measure of Classification Accuracy. Canadian Journal of Remote Sensing 20(3):210-216. Kaplan, E. D. 1996 Understanding GPS: Principles and Applications . Artech House, Boston. Kauth, R. J. and G. Thomas 1976 The Tassled Cap: A graphic Desc ription of the Spectral-Temporal Development of Agricultural Crops Seen by Landsat. In Proceedings of the Symposium on Machine Processing of Remotely Sensed Data , West Lafayette, Indiana. Keene, A. S. 1979 Economic Optimization Models a nd the Study of Hunter-Gatherer Subsistence Settlement Systems . In Transformations: Mathematical Approaches to Culture Change , edited by C. Renfrew and K. Cooke, pp. 369404. Academic Press, New York. Kent, M., W. J. Gill, R. E. Weaver and R. P. Armitage 1997 Landscape and Plant Commun ity Boundaries in Biogeography. Progress in Physical Geography 21(3):315. Kern, D. C. and M. Costa 1997 Os Solos Antropicos . In Caxiuana , edited by P. Lisboa, pp. 105. MCT/CNPq, Belem. Kern, D. C. and N. Kmpf 1989 O Efeito de Antigos Assentamentos Indgenas na Formao de Solos com Terra Preta Arqueolgica na Regio de Oriximin-Pa. Revista Braileira de Cincias do Solo 13:219-225.

PAGE 253

242 Kern, D., G. D’Aquino, T. Rodrigues, F. Fr anzo, W. Sombroek, T. Myers and E. Neves 2003 Distribution of Amazonian Dark Earths in the Brazilian Amazon . In Amazonian Dark Earths: Origin, Properties, Management , edited by J. Lehmann, D. Kern, B. Glaser, & W. Woods, pp. 51-75. Kluwer Academic Publishers, Netherlands. King, T. F. 1984 An Overview of Archaeological Law at the Federal Level. American Archaeology 4:115-118. Kintigh, K. W. 1990 Comments on the Case for Full-coverage Survey . In The Archaeology of regions: A Case for Full-Coverage Survey , edited by S. K. Fish and S. Kowalewski, pp. 237-242. Smithsonian In stitution Press, Washington, D.C. Knapp, A. B. and Wendy Ashmore 1999 Archaeological Landscapes: Constructed, Conceptualized, Ideational . In Archaeologies of Landscape: Contemporary Perspectives , edited by W. Ashmore and A. Knapp, pp. 1-30. Blackwell, Malden MA. Kohler, T. A. 1988 Predictive Location Modeling: History and Current Practice . In Quantifying the Present and Predicting th e Past: Theory Method and Application of Archaeological Predictive Modeling , edited by J. Judge and L. Sebastian, pp. 19-59. Bureau of Land Management, Colorado. Kohler, T. A. and S.C. Parker 1986 Predictive Models for Archaeological Resource Location . In Advances in Archaeological Method and Theory , pp. 397-452. vol. 9. Academic Press, New York. Komjathy, A. and University of New Bruns wick. Dept. of Geodesy and Geomatics Engineering. 1997 Global Ionospheric Total Electron Content Mapping Using the Global Positioning System , Dept. of Geodesy and Geomatic s Engineering University of New Brunswick. Koninck, R. and International Develo pment Research Centre (Canada). 1999 Deforestation in Viet Nam . International Development Research Centre, Ottawa, ON. Kontoes, C., G. G. Wilkinson, A. Burril, S. Goffredo and J. Megier 1993 An Experimental System for the In tegration of GIS Data in Knowledgebased Image Analysis for Remote Sensing of Agriculture. International Journal of Geographical Information Systems 7(3):247-262.

PAGE 254

243 Koussoulakou, A. and E. Stylianidis 1999 The Use of GIS for the Visual E xploration of Arch aeological SpatioTemporal Data. Cartography and Geographic Information Science 26(2):152160. Kowalewski, S. A. 1990 Merits of Full-Coverage Survey . In The Archaeology of Regions: A Case for Full-Coverage Survey , edited by S. K. Fish and S. Kowalewski, pp. 33-85. Smithsonian Institution Press, Washington, D.C. 1995 Large Scale Ecology in Aboriginal North America . In Native American Interactions: Multiscalar Analyses and Interpretations in the Eastern Woodlands , edited by M. Nassaney and K. Sassama n, pp. 147-174. University of Tennessee Press, Knoxville. Kowalewski, S. A. and S. Fish 1990 Conclusions . In The Archaeology of Regions: A Case For Full-Coverage Survey , edited by S. K. Fish and S. Kowalewski, pp. 261-277. Smithsonian Institution Press, Washington, D.C. Kruckman, L. 1987 The Role of Remote Sensi ng in Ethnohistorical Research. Journal of Field Archaeology 14(3):343-351. Kvamme, K. L. 1983 Computer Processing Techni ques for Regional Modeling of Archaeological Site Locations. Advances in Computer Archaeology 1:26-52. 1984 Models of Prehistoric Site Location Near Pion Canyon, Colorado . In Papers of the Philmont Conference on th e Archaeology of Northeastern New Mexico , edited by C. Condie, pp. 347-370. Pr oceedings of the New Mexico Archaeological Council. vol. 6, Albuquerque. 1989 Geographic Information Systems in Regional Archaeological Reserch and Data Management . In Archaeological Method and Theory , edited by M. B. Schiffer, pp. 139-203. vol. 1. University of Arizona Press, Tuscon. 1990 One-Sample Tests in Regional Archaeo logical Analysis: New Possibilites Through Computer Technology. American Antiquity 55(2):367-381. 1992 A Predictive Site Location Model on the High Plains: An Example with an Independent Test. Plains Anthropologist 37(138):19-40. 1995 A View From Across the Water: The North American Experience in Archaeological GIS . In Archaeology and GIS: A European Perspective , edited by G. R. Lock and Z. Stancic, pp. 1-14. Taylor and Francis, London.

PAGE 255

244 1996 Investigating Chipping Debris Scatte rs: GIS as an Analytical Engine . In New Methods, Old Problems: Geographi c Information Systems in Modern Archaeological Research , edited by H. D. G. Maschner, pp. 38-71. Southern Illinois University. 1999 Recent Directions and Developments in Geographic Information Systems. Journal of Archaeological Research 7(2):153-201. Labovitz, M. L. and J. W. Marvin 1986 Precision in Geodetic Correction of Tm Data as a Function of the Number, Spatial-Distribution and Success in Matc hing of Control Points—A Simulation. Remote Sensing of Environment 20(3):237-252. Ladefoged, T., S. McLachlan, S. Ross, P. Sheppard and D. Sutton 1995 GIS-based Image Enhancemen t of Conductivity and Magnetic Susceptibility Data from Uretur ituri Pa and Fort Resolution. American Antiquity 60(3):471-481. Lambin, E. F. 1997 Modeling and Monitoring Land C over Change Processes in Tropical Regions. Progress in Physical Geography 21(3):375-393. Langley, R. B. 1983 The GPS Observables. GPS World 4(4):52-59. 1998 RTK GPS. GPS World 9(9):70-76. Lathrap, D. W. 1968 The "Hunting" Economies of the Tropi cal Forest Zone of South America: An Attempt at Historical Perspective . In Man the Hunter , edited by R. Lee and I. Devore, pp. 23-29. Univeristy of Chicago Press, Aldine. 1970 The Upper Amazon . Praeger Publishers, New York. 1973 Alternative Models of Population M ovements in the Tropical Lowlands of South America. Acatas y Memorias del 39 Congresso Internacional de Americanistas 4:13-23. 1977 Our Father the Cayman, Our Mother the Gourd: Spinden Revisited, or a Unitary Model for the Emergence of Agriculture in the New World . In Origins of Agriculture , edited by C. Reed, pp. 713-751. 1985 Jaws: The Control of Power in the Early Nuclear American Ceremonial Center . In Early Andean Ceremonial Centers , edited by C. Donnan, pp. 241-267. Dumbarton Oaks, Washington.

PAGE 256

245 Lauer, D., Estes, J., Jensen, J., Greenlee. D. 1991 Institutional Issues Affecting the In tegration and Use of Remotely Sensed Data and Geographic Information Systems. Photogrammetric Engineering and Remote Sensing 57(6):647-655. Lavreau, J. 1991 De-hazing Landsat Thematic Mapper Images. Photogrammetric Engineering and Remote Sensing 57(10):1297-1302. Law, B. E. and R. H. Waring 1994 Remote-Sensing of Leaf-Area I ndex and Radiation Intercepted by Understory Vegetation. Ecological Applications 4(2):272-279. Layton, R. and P. Ucko 1999 Introduction . In The Archaeology and Anthropology of Landscape , edited by P. J. Ucko and R. Layton, pp. 1-20. Routledge, London. Lee, C. T. and S. E. Marsh 1995 The Use of Archival Landsat MSS and Anolliary Data in a GIS Environment to Map Historical Change in an Urban Riparian Habitat. Photogrammetric Engineering and Remote Sensing 61(8):10. Lillesand, T. M. and R. W. Kiefer 1994 Remote Sensing and Image Interpretation . 3rd ed. Wiley & Sons, New York. 2000 Remote Sensing and Image Interpretation . 4th ed. Wiley & Sons, New York. Limp, W. F. 1991 Continuous Cost Movement Models . In Applications of Space Technology in Anthropology , edited by C. Behrens and T. Sever, pp. 237-250. NASA. Lock, G. R. and Z. Stanicic 1995 Archaeology and Geographical Information Systems: A European Perspective . Taylor & Francis, London. Lookabill, Anna B. 1998 Predictive model for locating vaccinium -huckleberry processing sites in the northern Cascades of Washington. Northwest Anthropological Research Notes 32(2): 173-181.

PAGE 257

246 Lu, D., P. Mausel, E. Brondizio and E. Moran 2002 Assessment of Atmospheric Correc tion Methods for Landsat TM Data Applicable to Amazon Basin LBA Research. International Journal of Remote Sensing 23(13):2651-2671. 2003 Classification of Successional Fore st Stages in the Brazilian Amazon Basin. Forest Ecology and Management 181(3):12. 2004a Change Detection Techniques. International Journal of Remote Sensing 25(12):37. 2004b Relationships between Forest Stand Parameters and Landsat TM Spectral Responses in the Brazilian Amazon Basin. Forest Ecology and Management 198(1):20. Lucas, R. M., M. Honzak, G. M. Foody, P. J. Curran and C. Corves 1993 Characterizing Tropical Secondary Forests Using Multitemporal Landsat Sensor Imagery. International Journal of Remote Sensing 14(16):3061-3067. Ma, Z. and R. L. Redmond 1995 Tau Coefficients for Accuracy Asse ssments of Classification of Remote Sensing Data. Photogrammetric Engineering and Remote Sensing 61(4):5. Madry, S. and C. Crumley 1990 An Application of Remote Sensi ng and GIS in a Regional Archaeological Settlement Pattern Analysis: The A rrouz River Valley, Burgundy, France . In Interpreting Space: GIS and Archaeology , edited by K. M. S. Allen, S.W. Green and E. Zubrow, pp. 364-380. Taylor & Francis, London. Madry, S. and L. Rakos 1996 Line-of-sight and Cost-surface Tech niques for Regional Research in the Arroux River Valley . In New Methods, Old Problems: GIS in Modern Archaeological Research , edited by H. D. G. Maschner, pp. 1-23. Southern Illinois Univeristy, Springfield. Malczewski, J. 1999 GIS and Multicriteria Decision Analysis . J. Wiley & Sons, New York. Mann, C. C. 2000 Earthmovers of the Amazon. Science 287:787-789. Maracci, G., Schmuck, G., Hosgood, B. and Andreoli, G. 1991 Interpretation of Reflectance Spect ra by Plant Physiological Parameters . Paper presented at the International Ge oscience and Remote Sensing Symposium, Espoo, Finland.

PAGE 258

247 Marble, D. F. 1990 The Potential Methodological Impact of Geographic Information Systems on the Social Sciences . In Interpreting Space: GIS and Archaeology , edited by K. M. S. Allen, S.W. Green and E. Z ubrow, pp. 9-21. Taylor & Francis, London. Marble, D. F., H. Calkins, and D. Peuguet 1984 Basic Readings in Geographic Information Systems . SPAD Systems Ltd., New York. Mark, D. M. and A. Frank 1996 Experimental and Formal Models of Geographic Space. Environment and Planning 23:3-24. Marquardt, W. and C. Crumley 1987 Theoretical Issues in the Analysis of Spatial Patterning . In Regional Dynamics: Burgundian Landscapes in Historical Perspectives , edited by C. Crumley and W. Marquardt, pp. 1-18. Academic Press, New York. Marsh, G. P. 1864 Projected or Possible Geogr aphical Changes by Man. In Man and Nature or Physical Geography as Modified by Human Action , pp. 437-465. Charles Scribner, New York. Martin, P. 1963 The Last 10,000 years . University of Arizona Press, Tuscon. Martin, W. A. 1991 Assessing Feature Function and Sp atial Patterning of Artifacts with Geophysical Remote-sensing Data. American Antiquity 56(4):701-720. Maschner, H. D. G. 1996 New Methods, Old Problems: Geographi c Information Systems in Modern Archaeological Research . Center for Archaeological Investigations Southern Illinois University at Ca rbondale, Carbondale, Ill. Mausel, P., Y. Wu, Y. Li, E. F. Moran and E. S. Brondizio 1993 Spectral Identification of Succession al Stages Following Deforestation in the Amazon. Geocarto International 8(4):61-81. Maybury-Lewis, D. and J. Bamberger 1979 Dialectical Societies: The G and Bororo of Central Brazil . Harvard University Press, Cambridge.

PAGE 259

248 McCann J.M. and D.W. Meyer 2001 Organic Matter and Anthrosols in Amazonia: Interpreting the Amerindian Legacy . In Sustainable Management of Soil Organic Matter , edited by B. B. Rees, CD Campbell and CA Watson, pp. 180. CAB International, Wallingford. McNeil, M. 1964 Lateritic Soils. Scientific American 211(5):96-102. Medina, J. T., G. Carvajal and H. C. Heaton 1988 The Discovery of the Amazon . Dover Publications, New York. Meentemeyer, V. and E. O. Box 1987 Scale Effects in Landscape Studies . In Landscape Heterogeneity and Disturbance , edited by M. G. Turner, pp. 1534. Springer-Verlag, New York. Meggers, B. J. 1954 Environmental Limitations on the Development of Culture. American Anthropologist 56:801-824. 1971 Amazonia: Man and Culture in a Counterfeit Paradise . Aldine Atherton, Chicago. 1974 Environment and Culture in Amazonia . In Man in the Amazon , edited by C. Wagley, pp. 91-110. University of Florida Press, Gainesville. 1987 The Early History of Man in Amazonia . In Biogeography and Quaternary History in Tropical America , edited by T. C. Whitmore and G. Prance, pp. 151174. Clarendon Press, Oxford. 2001 The Continuing Quest for El Dorado: Round Two. Latin American Antiquity 12:304-325. Meggers, B. J. and C. Evans 1961 An Experimental Formulation of Ho rizon Styles in the Tropical Forest Area of South America . In Essays in Precolumbian Art and Archaeology , edited by S. Lathrup, pp. 372-388. Harvard University Press, Cambridge. Meggers, B. J. and E. Brondizio 2003 Revisiting Amazonia Circa 1492. Science 302(5653):2067. Mertes, L., D. L. Daniel, J. M. Melack, B. Nelson, L. A. Martinelli and B. R. Forsberg 1995 Spatial Patterns of Hydrol ogy, Geomorphology and Vegetation on the Floodplain of the Amazon River in Brazil from a Remote-Sensing Perspective. Geomorphology 13(1-4):215-232.

PAGE 260

249 Montieth, J. L. 1976 Vegetation and the Atmosphere . Vol. 2: Case Studies. Academic Press, New York. Mora, S., Herrera, L.F., Cave lier, I., & Rodrguez, C. 1991 Cultivars, Anthropic Soils and Stability. A Preliminary Report of Archaeological Research in Araracuara, Colombian Amazonia. Latin American Archaeology Reports No 2 . Morain, S. A. 1998 Brief History of Remote Sensing . In People and Pixels: Linking Remote Sensing and Social Science , edited by D. M. Liverman, pp. 28-50. Smithsonian Institution Press, Washington, D.C. Moran, E. 1981 Developing the Amazon . Indiana University Press, Bloomington. 1982 Human Adaptability: An Introduc tion to Ecological Anthropology . Westview Press, Boulder, Colorado. 1983 The Dilemma of Amazonian development . Westview Press, Boulder, Colo. 1984 Amazon Basin Colonization. Interciencia 9(6):377. 1993 Through Amazonian Eyes: The Hu man Ecology of Amazonian Populations . University of Iowa Press, Iowa City. 1995 Disaggregating Amazonia: A Strate gy for Understanding Biological and Cultural Diversity . In Indigenous Peoples and th e Future of Amaznia , edited by L. Sponsel, pp. 71-95. University of Arizona Press, Tuscon. Moran, E., E. Brondizio, P. Mausel and Y. Wu 1994 Integrating Amazonian Vegeta tion, Land-use and Satellite Data. BioScience 44(5):329. Moran, E., E. Brondizio, J. Tucker, M. da S ilva-Forsberg, S. McCracken and I. Falesi 2001 Effects of Soil Fertility and LandUse on Forest Succession in Amazonia. Forest Ecology and Management 139(1):16. Moran, E. and R. Herrera 1984 Human Ecology in the Amazon. Interciencia 9(6):342. Moran, E., A. Packer, E. Brondizio and J. Tucker 1996 Restoration of vegetation cover in the eastern Amazon. Ecological Economics: The Journal of the Interna tional Society for Ec ological Economics 18(1):14.

PAGE 261

250 Mosteller, F. and J. W. Tukey 1977 Data Analysis and Regression: A Second Course in Statistics . AddisonWesley, Reading, Mass. Mudar, K. 1999 How Many Dvaravati Kingdoms? Locational Analysis of the First Millenium A.D. Moated Settlements in Central Thailand. Journal of Anthropological Archaeology 18:1-28. Muller, J.-P. 1988 Key Issues in Image Understanding in Remote Sensing. Philosophical Transactions of the Royal Society of L ondon Series A-Mathematical and Physical Sciences 324(1579):381-395. Mumford, L. 1956 Summary Remarks: Prospect. In Man’s Role in Changi ng the Face of the Earth , edited by W. Thomas, pp. 1141-1152. University of Chicago Press, Chicago. Myers, T. P. 1973 Toward the Reconstruction of Preh istoric Community Patterns in the Amazon Basin . In Variation in Anthropology , edited by D. W. Lathrap and J. Douglas, pp. 233. llinois Archaeological Survey, Urbana, Il. Myneni, R., S. Maggion, J. Iaquinto, J. Priv ette, N. Gobron, B. Pinty, D. Kimes, M. Verstraete and D. Williams 1995 Optical Remote-Sensing of Vegetation—Modeling, Caveats and Algorithms. Remote Sensing of Environment 51(1):169-188. Myneni, R. and D. Williams 1994 On the Relationship between FAPAR and NDVI. Remote Sensing of Environment 49:200-211. Nance, C. R., H. Holstein and D. C. Hurst 1983 Evaluation of Multiple Regression M odels Predicting Archaeological Site Distributions at Fort McClellan, Alabama . In Society for American Archaeology , Pittsburgh. Neves, E. 2001 Indigenous Historical Trajectori es in the Upper Rio Negro Basin . In Unknown Amazon , edited by C. McEwan, C. Barreto and E. Neves. British Museum, London.

PAGE 262

251 Neves, E. G., J.B. Peterson., R.N. Bartone and C.A. da Silva 2003 Historical and Soci o-cultural Origins of Amazonian Dark Earths . In Amazonian Dark Earths: Origin, Properties, Management , edited by J. Lehmann, D.C. Kern, B. Glaser, & W.I. Woods, pp. 29-49. Kluwer Academic Publishers, Netherlands. Nghiem, S. V., T. Letoan, J. A. Kong, H. C. Han and M. Borgeaud 1993 Layer Model with Random Spheroid al Scatterers for Remote-Sensing of Vegetation Canopy. Journal of Electromagnetic Waves and Applications 7(1):4975. Nicholaides, J., D. Bandy, P. Sanchez, J. Villachica, A. Coutu and C. Valverde 1984 Continuous Cropping Potential in the Upper Amazon Basin . In Frontier Expansion in Amazonia , edited by M. Schmink and C. Wood, pp. 337-365. University of Florida Press, Gainesville. Nicolson, M. 1987 Alexander von Humboldt, Humboldtia n Science and the Origins of the Study of Vegetation. History of Science 25: 167-194. Nunez, M., A. Vikkula and T. Kirkinen 1995 Perceiving Time and Space in an Isostatically Rising Region . In Archaeology and geographical informati on systems: a European perspective , edited by G. R. Lock and Z. Stancic, pp. 141-151. Taylor & Francis, London. Nyerges, A. and G. Green 2000 The Ethnography of Landscape: GIS a nd Remote Sensing in the Study of Forest Change in West African Guinea Savanna. American Anthropologist 102:271-289. Oliver, J. 2001 The Archaeology of Forest Foragi ng and Agricultural Production in Amazonia . In Unknown Amazon , edited by C. McEwan, C. Barreto and E. Neves, pp. 50-85. British Museum, London. Oliveria, A. E. 1994 The Evidence for the Nature of the Process of Indigenous Deculturation and Destabilization in the Brazilian Am azon in the Last Three-hundred Years . In Amazonian Indians from Prehistory to th e Present: Anthropological Perspectives , edited by A. C. Roosevelt, pp. 95-119. University of Arizona Press, Tuscon. Olsson, G. 1965 Distance and Human Interacti on; A Review and Bibliography . Regional Science Research Institute, Philadelphia.

PAGE 263

252 Orton, C. 2000 Sampling in Archaeology . Cambridge University Press, Cambridge, U.K.; New York. Parker, S. 1985 Predictive Modeling of Site Set tlement Systems using Multivariate Logistics . In For Concordance in Archaeologi cal Analysis: Bridging Data Structure, Quantitative Technique and Theory , edited by C. Carr, pp. 173-207. Westport Publishers, St. Loius. 1986 The Role of Geographic Information Systems in Cultural Resource Management . In Geographic Information Systems in Government , edited by B. Opitz, pp. 133-140. A. Deepak Publishing, Virginia. Parrington, M. 1983 Remote Sensing. Annual Review of Anthropology 12:105-124. Parsons 1972 Archaeological Settlement Patterns. Annual Review of Anthropology 1:127-150. Pashley, E. F. 1966 Structure and Stratigraphy of the Ce ntral, Northern And Eastern Parts of the Tucson Basin, Arizona. Paul, C. and A. Mascarenhas 1981 Remote Sensing in Development. Science 214(4517):139-145. Paynter, R. 1983 Expanding the Scope of Settlement Analysis . In Archaeological Hammers and Theories , edited by J. Moore and A. Keene, pp. 233-275. Academic Press, New York. Perkins, Philip 2000 A GIS investigation of site locati on and landscape relationships in the Albegna Valley, Tuscany. In Computer Application and Quantitative Methods in Archaeology , edited by K. Lockyear, T. Sly, a nd V. Mihailescu-Birliba, pp. 133140. BAR International Series 845, Archaeopress, Oxford, England. Petersen, J. and M. Heckenberger 2001 Gifts from the Past: Terra Preta a nd Prehistoric Amerindian Occupations in Amazonia . In The Unknown Amazon , edited by C. McEwan, C. Barreto and E. Neves, pp. 86-105. British Museum, London.

PAGE 264

253 Peuquet, D. J. 1984 Conceptual Framework and Comp arison of Spatial Data Models. Cartographica 21(4):66-113. Peuquet, D. J. and D. F. Marble 1990 Introductory Readings in Geographic Information Systems . Taylor & Francis, London. Philipson, W. R., D. K. Gordon, W. D. Philpot and M. J. Duggin 1989 Field Reflectance Calibration with Grey Standard Reflectors. International Journal of Remote Sensing 10(6):1035-1039. Pickles, J. 1997 Tool or Science? GIS, Technoscience and the Theoretical Turn. Annals of the Association of American Geographers 87(2):363-372. Pinty, B. and M. M. Verstraete 1998 Modeling the Scattering of Light by Homogeneous Vegetation in Optical Remote Sensing. Journal of the Atmospheric Sciences 55(2):137-150. Pires, J. M. 1984 The Amazonian Forest . In The Amazon: Limnology and Landscape Ecology of a Mighty Tropical River and Its Basin , edited by H. Sioli, pp. 581-601. Dr. W. Junk Publishers, Netherlands. Pires, J. M. and G. T. Prance 1985 The Vegetation Types of the Brazilian Amazon . In Amazonia: Key Environments , edited by G. T. Prance and T. E. Lovejoy, pp. 109-165. Pergamon Press, Oxford. Plog, F. 1990 Some Thoughts on Full-coverage Survey . In The Archaeology of Regions: A Case for Full-Coverage Survey , edited by S. K. Fish and S. Kowalewski, pp. 243-248. Smithsonian Institution Press, Washington, D.C. Pope, K. and B. Dahlin 1989 Ancient Maya Wetland Culture: New Insights from Ecological and Remote Sensing Research. Journal of Field Archaeology 16(1):87-106. Porro, A. 1996 O Povo das guas: Ensaios de Etno-histria Amaznica . Petrpolis, Sao Paulo.

PAGE 265

254 Posey, D. A. 1983 Indigenous Ecological Knowledge and Development of the Amazon . In The Dilemma of Amazonian Development , edited by E. Moran. Westview Press, Boulder. 1985 Indigenous Management of Tropical Fo rest Ecosystems: The Case of the Kayap Indians of the Brazilian Amazon. Agroforestry Systems 3:139-158. 1994 Environmental and Social Implications of Preand Postcontact Situations on Brazilian Indians: The Kaya po and a New Amazonian Synthesis . In Amazonian Indians from Prehistory to th e Present: Anthropological Perspectives , edited by A. C. Roosevelt, pp. 271-286. University of Arizona Press, Tuscon. 1998 Diachronic Ecotones and Anthropoge nic Landscapes: Contesting the Consciousness of Conservation . In Advances in Historical Ecology , edited by W. Balee, pp. 104-118. Columbia Un iversity Press, New York. Posey, D. A. and W. L. Bale 1989 Resource Management in Amazonia: Indigenous and Folk Strategies . Advances in Economic Botany, v. 7. New York Botanical Garden, Bronx, N.Y. Posey, D. A., J. Frechione, J. Eddins, L. F. Da silva, D. Myers, D. Case and P. Macbeath 1984 Ethnoecology as Applied Anthropology in Amazonian Development. Human Organization 43(2):95-107. Prance, G. T. and T. E. Lovejoy 1985 Amazonia . 1st ed. Key environments. Pergamon Press, Oxford Oxfordshire; New York. Price, J. C. 1987 Calibration of Satellite Radiomet ers and the Comparison of Vegetation Indexes. Remote Sensing of Environment 21(1):15-27. Price, J. C. and W. C. Bausch 1995 Leaf-Area Index Estimation from Visible and near-Infrared Reflectance Data. Remote Sensing of Environment 52(1):55-65. Qi, J., Chehbouni, A., Huete, A.R., Kerr, Y.H. and Sorooshian, S. 1994 A Modified Soil Adjusted Vegetation Index. Remote Sensing of Environment 48:119-126. Qi, J., Moran, M.S., Cabot, F. and Dedieu, G. 1995 Normalization of Sun/View Angl e Effects Using Spectral Albedo-Based Vegetation Indices. Remote Sensing of Environment 52:207-217.

PAGE 266

255 Ramenofsky, A. F. and A. Steffen 1998 Unit Issues in Archaeology: Measuring Time, Space and Material . University of Utah Press, Salt Lake City. Rapp, G. R. 1975 The Archaeological Field Staff. Journal of Field Archaeology 2:229-237. Rapp, G. R. and C. L. Hill 1998 Geoarchaeology: the Earth-Scien ce Approach to Archaeological Interpretation . Yale University Press, New Haven. Renfrew, C., M. J. Rowlands and B. A. Segraves, 1982 Theory and Explanation in Archaeology: The Southampton Conference . Academic Press, New York. Reynolds, R. 1976 Linear Settlement Systems on the Uppe r Grijalva River: the Application of a Markovian Model . In The Early Mesoamerican Village , edited by K. V. Flannery, pp. 180-194. Academic Press, New York. Ribed, P. S. and A. M. Lopez 1995 Monitoring Burnt Areas by Prin cipal Components-Analysis of Multitemporal Tm Data. International Journal of Remote Sensing 16(9):15771587. Richards, J. 1998 Recent Trends in Computer Applications in Archaeology. Journal of Archaeological Research 6(4):331-382. Richards, J. A. 1993 Remote Sensing Digital Image Analysis: An Introduction . 2nd ed. Springer-Verlag, Berlin; New York. Richardson, A. J. and C. L. Wiegand 1977 Distinguishing Vegetation from Soil Background Information. Photogrammetric Engineering and Remote Sensing 43(12):1541-1552. Rondeaux, G., Steven, M. and Baret, F. 1996 Optimization of Soil-Adju sted Vegetation Indices. Remote Sensing of Environment 55(95-107). Roosevelt, A. C. 1980 Parmana: Prehistoric Maize and Ma nioc Subsistence along the Amazon and Orinoco . Academic Press, New York.

PAGE 267

256 1987 Chiefdoms in the Amazon and Orinoco . In Chiefdoms in the Americas , edited by R. Drennan and C. Uribe, pp. 153-184. University Press of America, Lanham. 1989a Lost Civilizations of the Lower Amazon. Natural History 2:76-83. 1989b Resource Management in Amazonia Before the Conquest: Beyond Ethnographic Projection . In Resource Management in Amazonia: Indigenious and Folk Strategies , edited by D. A. Posey and W. Balee, pp. 30-62. Advances in Economic Botany Vol. 7, New York. 1991 Moundbuilders of the Amazon: Geophy sical Archaeology on Marajo Island, Brazil . Academic Press, San Diego. 1994 Amazonian Anthropology: Strategy for a New Synthesis . In Amazonian Indians from Prehistory to the Present: Anthropological Perspectives , edited by A. C. Roosevelt, pp. 1-29. University of Arizona Press, Tuscon. 1999 The Development of Prehistoric Complex Societies: Amazonia, A Tropical Forest . In Complex Societies in the Ancient Tropical World , edited by E. Bacus, pp. 13-34. Archaeological Papers of the American Anthropological Association 9. Roosevelt, A. and B. J. Meggers 1996 Amazonian Indians from Prehistory to the Present. The Journal of the Royal Anthropological Institute 2(1):2. Roper, D. C. 1979 Archaeological Survey and Settlement Pattern Models in Central Illinois . MCJA special paper; no. 2. Kent St ate University Press, Kent, Ohio. Rose, M. R. and J. H. Altschul 1988 An Overview of Statistical Me thod and Theory for Quantitative Model Building . In Quantifying the Present and Predic ting the Past: Theory Method and Application of Archaeol ogical Predictive Modeling , edited by J. Judge and L. Sebastian, pp.173-256. Washington, D.C. Ross, E. B. 1978 Food Taboos, Diet, and Hunting Stra tegy: The Adaptation to Animals in Amazonian Cultural Ecology. Current Anthropology 19:1-36. Rouse Jr., J. W., Haas, R. H., Sc hell, J. A., & Deering, D. W. 1973 Monitoring Vegetation Systems in the Great Plains with ERTS . In Third Earth ResourcesTechnology Satellite-1 Sy mposium. Technical presentations, section A , edited by S. C. Freden, E. P. Me rcanti, & M. Becker. vol. 1. National Aeronautics and Space Administratio n (NASA SP-351, Washington, D.C.

PAGE 268

257 Rowe, J. 1953 Technical Aids in Anthropology: A Historical Survey . In Anthropology Today , edited by A. Kroeber, pp. 895-940. Univ ersity of Chicago Press, Chicago. Ruggles, C. and D. Medyckyj-Scott 1996 Site Location, Lanscape Visa bility and Symbolic Astronomy . In New Methods, Old Problems: Geographic In formation Systems in Modern Archaeological Research , edited by H. D. G. Maschner, pp. 127-146. Southern Illinois University, Springfield. Ruggles, C., D. Medyckyj-Scott and A. Gruffyd 1993 Multiple Veiwshed Analysis Using GIS and Its Archaeological Application . In Computing The Past: Computer Applications and Quantitative Methods in Archaeology , edited by J. Andresen, T. Mads en and I. Scollar. Aarhus University Press, Denmark. Saatchi, S. S., J. V. Soares and D. S. Alves 1997 Mapping Deforestation and Land Use in Amazon Rainforest by Using SIR-C Imagery. Remote Sensing of Environment 59:191-202. Sabins, F. 1987 Remote Sensing Principles and Interpretation . W.H. Freeman and Co., New York. Sader, S. A., T. Sever, J. C. Smoot and M. Richards 1994 Forest Change Estimates for the Northern Peten Region of Guatemala 1986-1990. Human Ecology 22(3):317-332. Saffirio, J. and R. Hames 1983 The Forest and the Highway . In The Impact of Contact . 6 ed. Working Papers on South American Indi ans. Bennington College, Bennington. Salas, B. and J. Brunner 1998 Techincal Guide For Change Detection with Multitemporal Landsat Data . University of New Hampshire, Co mplex Systems Research Center. Salthe, S. 1988 Notes Towards a Formal History of the Levels Concept . In Evolution of Social Behavior and Integrative Levels , edited by G. Greenberg and E. Tobach, pp. 251-254. Lawrence Erlbaum Associates, Hillsdale. Sample, V. A. 1994 Remote Sensing and GIS in Ecosystem Management . Island Press, Washington, D.C.

PAGE 269

258 Snchez, P. A. 1976 Properties and Management of Soils in the Tropics . Wiley, New York. Sanchez, P. A. and North Carolina Agricultural Experiment Station. 1977 A Review of Soils Research in Tropical Latin America . North Carolina Agricultural Experiment Station in cooperation with the U.S. Agency for International Development, Raleigh. Sauer, C. 1956 Summary Remarks: Retrospect. In Man’s Role in Changing the Face of the Earth , edited by W. Thomas, pp. 1131-1135. University of Chicago Press, Chicago. Saunders, N. 1994 At the Mouth of the Obsidian Cave: Diety and Place in Aztec Religion . In Sacred Sites, Sacred Places , edited by D. L. Carmichael, pp. 172-183. Routledge, London. Savage, S. 1990 GIS in Archaeological Research . In Interpreting Space: GIS and Archaeology , edited by K. M. S. Allen, S.W. Green and E. Zubrow, pp. 22-32. Taylor & Francis, London. Sayce, R. 1938 The Ecological Study of Culture. Scientia 63: 279-285. Schaan, D. 1997 Evidncia Arqueolgica e Organiz ao Social na Fase Marajoara. Estudos Ibero-Americanos 23(1):97-114. 2001 Into the Labyrinths of Marajoara Po ttery: Status and Cultural Identity in an Amazonian Complex Society . In Unknown Amazon , edited by C. McEwan, C. Barreto and E. Neves, pp. 108-133. British Museum, London. 2004 The Camutins Chiefdom: Rise and Development of Social Complexity on Marajo Island, Brazilian Amazon. PhD Dissertation, University of Pittsburgh. Schalf, R. and T. Lyons 1976 Ecological Application of Landsat Imagery in Archaeology: An Example from the San Juan Basin, New Mexico . In Remote Sensing Experiments in Cultural Resource Studies , edited by T. Lyons, pp. 173-186. University of New Mexico Press, Alburquerque. Schmink, M. and C. H. Wood 1984 Frontier Expansion in Amazonia , University of Florida Press, Gainesville.

PAGE 270

259 1992 Contested frontiers in Amazonia . Columbia University Press, New York. Schubart, H. and E. Salati 1980 Natural Resources For Land Use In The Amazon Region: The Natural Systems . In Amazonia: Agriculture and Land Use Research ., edited by S. Hecht, pp. 211-239. Centro Internacional de Agri cultura Tropical, Cali, Colombia. Sebastian, L. and W.J. Judge 1988 Predicting the Past: Correla tion, Explanation and the Use of Archaeological Models . In Quantifying the Present and Predicting the Past: Theory Method and Application of Ar chaeological Predictive Modeling , edited by J. Judge and L. Sebastian, pp. 1-18. Bureau of Land Management, Denver. Senseman, G. M., C.F. Bagley, S.A. Tweddale 1996 Correlation of Rangeland Cover Meas ures to Satellite-Imagery-Derived Vegetation Indices. Geocarto International 11(3):29-38. Sever, T. 1998 Validating Prehistoric and Current Social Phenomena Upon the Landscape of the Peten, Guatemala . In People and Pixels: Linki ng Remote Sensing and Social Science , edited by D. M. Liverman, pp.145-163. Smithsonian Institution Press, Washington, D.C. Shermer, S. J. and J. A. Tiffany 1985 Environmental Variables as Factors in Site Location: An Example From the Upper Midwest. Midcontinental Journal of Archaeology 10:215-240. Siljestroem Ribed, P. and A. Moreno Lopez 1995 Monitoring Burnt Areas by Princi pal Components Analysis of MultiTemporal TM Data. International Journal of Remote Sensing 16(9):1577. Singh, A. 1987 Spectral Separability of Tr opical Forest Cover Classes. International Journal of Remote Sensing 8(7):971-979. Sioli, H. 1984 The Amazon: Limnology and Landscape Ecology of a Mighty Tropical River and Its Basin . Monographiae Biologi cae; v. 56. W. Junk; Distributors for the U.S. and Canada Kluwer Academic Publishers, Dordrecht Netherlands. Skole, D. and C. J. Tucker 1993 Tropical Deforestation and Habitat Fr agmentation in the Amazon: Satellite Data from 1978 to 1988. Science 260:1905-1910. Skole, D. L., W. H. Chomentowski , W. A. Salas and A. D. Nobre 1994 Physical and Human Dimensions of Deforestation in Amazonia. BioScience 44(5):314-322.

PAGE 271

260 Smith, C. A. 1976a Analyzing Regional Social Systems . In Regional Analysis. Volume II: Social Systems , edited by C. A. Smith, pp. 3-20. Academic Press, New York. 1976b Regional Economic Systems: Li nking Geographical Models and Socioecomic Problems . In Regional Analysis. Volume I: Economic Systems , edited by C. A. Smith, pp. 3-63. Academic Press, New York. Smith, N. J. H. 1980 Anthrosols and Human Carrying Capacity in Amazonia. Annals of the Association of American Geographers 70:553-566. 1999 The Amazon River Forest: A Natura l History of Plants, Animals and People . Oxford University Press, New York. Sombroek, W. G. 1966 Amazon Soils: A Reconnaissance of the Soils of the Brazilian Amazon Region . Centre for Agricultural Publicatio ns and Documentation, Wageninger. 1984 Soils of the Amazon Region . In The Amazon: Limnology and Landscape Ecology of a Mighty Tropical River and Its Basin , edited by H. Sioli, pp. 521– 535. Dr. W. Junk Publishers, Netherlands. Spanner, M., L. Johnson, J. Miller, R. McCr eight, J. Freemantle, J. Runyon and P. Gong 1994 Remote Sensing of Seasonal Leaf Area Index across the Oregon Transect. Ecological Applications 4(2):258-271. Sponsel, L. E. 1992 The Environmental History of Amazonia: Natural and Human Disturbances and the Ecological Transition . In Changing Tropical Forests: Historical Perspectives on Today’s C hallenges in Central and South America , edited by H. K. Steen and R. P. Tucker . Forest Historical Society, Duram. Stahl, P. W. 1995 Archaeology in the Lowland American Tropics: Current Analytical Methods and Applications . Cambridge University Press, Cambridge England; New York. Star, J., Estes, J. and Davis, F. 1991 Improved Integration of Remote Sensing and Geographic Information Systems: A Background to NCGIA Initiative. Photogrammetric Engineering and Remote Sensing 57(6):643-645. Steiniger, M. 1996 Tropical Secondary Forest Regrowth in the Amazon. International Journal of Remote Sensing 17(1):9-27.

PAGE 272

261 Stern, P. C., O. R. Young and D. Druckman 1992 Global Environmental Change: Understanding the Human Dimensions . National Academy Press, Washington, DC. Steward, J. 1938 Basin-plateau aborigin al sociopolitical groups. Bureau of American Ethnology Bulletin 150 . 1946 Handbook of South American Indians. U.S. Government Printing Office, Washington, D.C. Steward, J. H. and L. C. Faron 1959 Native peoples of South America . McGraw-Hill, New York. Stewart, D. I. 1994 After the Trees: Living on the Transamazon Highway . 1st ed. University of Texas Press, Austin. Stine, R. and D. Lanter 1990 Considerations for Archaeology Database Design . In Interpreting Space: GIS and Archaeology , edited by K. M. S. Allen, S.W. Green and E. Zubrow, pp. 80-89. Taylor & Francis, London. Teillet, P. M., K. Staenz and D. J. Williams 1997 Effects of Spectral, Spatial and Ra diometric Characteristics on Remote Sensing Vegetation Indices of Forested Regions. Remote Sensing of Environment 61:139-149. Thenkabail, P. S. 1999 Characterization of the Alternat ive to Slash-and-Burn Benchmark Research Area Representing the Congolese Rainforests of Africa Using NearReal-Time SPOT HRV Data. International Journal of Remote Sensing 20(5):839877. Theodoraus, D. and F. LaPena 1994 Wintu Sacred Geography of Northern California . In Sacred Sites, Sacred Places , edited by D. L. Carmichael , pp. 20-31. Routledge, London. Thomas, D. H. 1975 Non-site Sampling: Up the Creek Without a Site? In Sampling in Archaeology , edited by J. Mueller, pp. 61-81. University of Arizona Press, Tuscon.

PAGE 273

262 Tilley, C. Y. 1994 A Phenomenology of Landscape: Places, Paths and Monuments . Berg, Oxford, UK; Providence, R.I. Tipps, B. and A. Schroedl 1984 Site Density Estimation . In Investigations of the Southwestern Anthropological Research Group , edited by R. Euler and G. Gumerman, pp. 168175. Museum of Northern Arizona, Flagstaff. Tobler, W. 1979 Cellular Geography . In Philosophy in Geography , edited by S. Gale, pp. 379-386. Reidel Publishing Co., Holland. Tomlinson, R. 1998 The Canada Geographic Information System . In The History of Geographic Information Systems: Perspectives from the Pioneers , edited by T. Foresman, pp. 21-32. Prentice Hall PTR, New Jersy. Tosta, N. 1991 More Than Maps: Asking the Right Questions. Geographic Information Systems 2:46-48. Tottrup, C. 2002 Deforestation in the Uppers Ca River Basin in North-central Vietnam: A Remote Sensing and GIS Perspective . Geographica Hafniensia C12. University of Copenhagen. Townshend, J. 1981 Terrain Analysis and Remote Sensing . Allen & Unwin, London; Boston. Trimble, N. L. 1997 Characterizing Accuracy of the Trimble Pathfinder Mapping Receivers . /dsweb/Get/Document-9718/acc_sum.pdf . Tucker, C. J. 1979 Red and Photographic Infrared Li near Combinations for Monitoring Vegetation. Remote Sensing of Environment 8:127-150. Tucker, J. M., E. S. Brondizio and E. F. Moran 1998 Rates of Forest Regrowth in Eastern Amazonia: A Comparison of Altamira and Bragantina Regions, Pa ra State, Brazil (in Spanish). Interciencia 23(2):10. Turkey, J. 1977 Exploratory Data Analysis . Addison-Wesley, Massachusetts.

PAGE 274

263 Turner, B. L. 1997 The Sustainability Principle in Global Agendas: Implications for Understanding Land-Use/Cover Change. The Geographical Journal 163(2):8. Turner, B., D. Skole, S. Sanderson, G. Fische r, L. Fresco and R. Leemans (editor) 1995 Land-Use and Land-Cover Change Science/Research Plan . Stockholm, Sweden: International Geosphere-Biosphere Programme. Turner, B., R. Kasperson, W. Meyer, K. Dow, D. Goulding, J. Kasperson, R. Mitchell, and S. Ratick 1990 Two Types of Global Environmental Change: Definitional and SpatialScale Issues in Their Human Dimensions. Global Environmental Change 1(1):14-22. Turner, M. 1989 Landscape Ecology: The Eff ect of Pattern on Process. Annual Review of Ecology and Systematics 20:171-197. Uhl, C. 1987 Factors Controlling Succession Follo wing Slash-and-Burn Agriculture in Amazonia. Journal of Ecology 75:377-407. van Leusen, M. 1993 Cartographic Modeling in Cell-based GIS . In Computing The Past: Computer Applications and Quan titative Methods in Archaeology , edited by J. Andresen, T. Madsen and I. Scollar, pp. 105-123. Aarhus Univeristy Press, Denmark. Verbyla, D. L. 1995 Satellite Remote Sensing of Natural Resources . Lewis Publishers, Boca Raton. Vinta-Finzi, C. and E. Higgs 1970 Prehistoric Economy in the Mount Carmel Area of Palastine: Site Cachment Analysis. Proceedings of the Prehistoric Society 36:1-37. Wallace, A. F. C. 1960 Men and Cultures; Selected Papers . University of Pennsylvania Press, Philadelphia. Walschburger, T. and von Hildebrand, P. 1991 The First 26 Years of Forest Rege neration in Natural and Man-Made Gaps in the Colombian Amazon . In Rain Forest Regeneration and Management , edited by A. Gomez-Pompa, T. C. Whitmore and M. Hadley, pp. 257-263. UNESCO, Paris.

PAGE 275

264 Walter-Shea, E., J. Privette, D. Cornel, M. Mesarch, and C. Hays 1997 Relations Between Directional Spec tral Vegetation Indices and Leaf Area and Absorbed Radition in Alfalfa. Remote Sensing of Environment 61:162-177. Wandsnider, L. 1998 Regional Scale Processes a nd Archaeological Landscape Units . In Unit Issues in Archaeology , edited by A. F. Ramenofs ky and A. Steffen, pp. 87-102. University of Utah Press, Salt Lake City. Warren, P., D. Gordon, W. Philpot and M. Duggin 1989 Field Reflectance Calibration with Grey Standard Reflectors. International Journal of Remote Sensing 10(6):1035-1039. Warren, R. 1990 Predictive Modeling of Archaeological Site Location . In Interpreting Space: GIS and Archaeology , edited by K. M. S. Allen, S.W. Green and E. Zubrow, pp. 201-215. Taylor & Francis, London. Waters, M. 1986 The Geoarchaeology of Whitewater Draw . Anthrpological Papers No. 45. University of Arizona Press, Tuscon. 1988 The Impact of the Fluvial Pr ocess and Landscape Evolution on Archaeological Sites and Settlement Pattern s Along the San Xavier Reach of the Santa Cruz River, Arizona. Geoarchaeology 3:205-219. Waters, M. R. and D. D. Kuehn 1996 The geoarchaeology of place: The e ffect of geological processes on the preservation and interpretation of the archaeological record. American Antiquity 61(3):483-497. Weill, L. R. 1997 Conquering Multipath: The Gps Accuracy Battle. GPS World 8(4):59-66. Wells, D. and N. Beck 1986 Guide to GPS Positioning . Canadian GPS Associat es, Fredericton, N.B., Canada. Whalen, M. 1990 Sampling Versus Full-coverage Survey . In The Archaeology of Regions: A Case for Full-Coverage Survey , edited by S. K. Fish and S. Kowalewski, pp. 219236. Smithsonian Institution Press, Washington, D.C.

PAGE 276

265 Wheatley, D. 1995 Cumulative Viewshed Analysis: A GIS-based Method for Investigating Intervisability and its Archaeological Application . In Archaeology and Geographical Information Systems: A European Perspective , edited by G. R. Lock and Z. Stancic, pp. 171185. Taylor & Francis, London. 1996 The Use of GIS to Understand Region al Variation in Earlier Neolithic Wessex . In New Methods, Old Problems: Ge ographic Information Systems in Modern Archaeological Research , edited by H. D. G. Maschner, pp. 75-103. Southern Illinois Univeristy, Springfield. Wheatley, D and M. Gillings 2002 Spatial Technology and Archaeology . Taylor & Francis, NY. Whitehead, N. 1998 Ecological History and Historical Ecology: Diachronic Modeling Versus Historical Explanation . In Advances in Historical Ecology , edited by W. Balee, pp. 30-41. Columbia University Press, New York. Whitmore, T. C. 1990 An Introduction to Tropical Rain Forests . Clarendon Press; Oxford University Press, Oxford. Wildesen, L. E. 1974 Archaeologists and Planners: The Use and Misuse of Predictive Models . In Annual Meeting of the Society for California Archaeology , Riverside. 1984 Locating Significant Archaeological Sites by Landform Analysis in Central Oregon. Bureau of Land Management, Princeville District. Willey, G. R. 1953 Prehistoric Settlement Patterns in the Viru Valley, Peru . Bureau of American Ethnology Bulletin 155. Bureau of American Ethnology, Washington, D.C. Winterhalder, B. 1994 Concepts in Historical Ecology . In Historical Ecology: Cultural Knowledge and Changing Landscapes , edited by C. Crumley. School of American Research Press, Santa Fe. Wise, Alicia L. 2000 Building theory into GIS-based landscape analysis. In Computer Application and Quantitative Methods in Archaeology , edited by K. Lockyear, T. Sly, and V. Mihailescu-Birliba, pp. 141-147. BAR Inte rnational Series 845, Archaeopress, Oxford, England.

PAGE 277

266 Witcher, R. E. 1999 GIS and landscapes of perception. In Geographical information systems and landscape archaeology . Edited by M. Gillings, D. Mattingly and J. v. Dalen. Oxford, Oxbow Books pp. 13-22. Wobst, H. M. 1983 We Can't See the Forest for th e Trees: Sampling and the Shapes of Archaeological Distributions . In Archaeological Hammers and Theories , edited by J. Moore and A. Keene, pp. 37-85. Academic Press, New York. Wood, C. H. and D. Skole 1998 Linking Satellite, Census and Survey Data to Study Deforestation in the Brazilian Amazon . In People and Pixels: Linking Remote Sensing and Social Science , edited by D. Liverman, E. Moran, R. Rindfuss and P. Stern, pp. 70-93. National Academy Press, Washington, D.C. Wood, J. J. 1978 Optimal Location in Settlement Sp ace: A Model for Describing Locational Strategies. American Antiquity 43:258-270. Wood, T. F. and G. Foody 1993 Using Cover-type Likelihoods and Typicalities in a Geographic Information System Data Structure to Map Gradually Changing Environments . In Landscape Ecology and GIS , edited by R. Haines-Young, D. R. Green and S. H. Cousins, pp. 141-146. Taylor & Francis, London. Woods, W. I. 1995 Comments on the Black Earths of Amazonia . In Papers and Proceedings of Applied Geography Conferences 18, edited by F. Schoolmaster, pp. 159-165. Applied Geography Conferences, Denton, Texas. 2003 Development of Anthrosol Research . In Amazonian Dark Earths: Origin, Properties, Management , edited by J. Lehmann, D.C. Kern, B. Glaser, & W.I. Woods, pp. 3-14. British Museum, London. Woods, W. I., and J. McCann 1999 The Anthropogenic Origin and Persistence of Amazonian Dark Earths. Yearbook Conference of Latin Americanist Geographers 25:7-14. Wright, D., M. Goodchild, and J. Proctor 1997 GIS: Tool or Science? Demystifyi ng the Persistent Ambiguity of GIS as a "Tool" Versus "Science". Annals of the Association of American Geographers 87(2):346-362. Wrigley, N. 1977 Probability Surface Mapping: An Intr oduction with Examples and Fortran Programs . Geo Abstracts, University of East Anglia, Norwich, U.K.

PAGE 278

267 Wust, I. 1994 The Eastern Bororo from and Archaeological Perspective . In Amazonian Indians from Prehistory to the Present: Anthropological Perspectives , edited by A. C. Roosevelt, pp. 315-342. University of Arizona Press, Tuscon. Yanasse, C., S. Sant’Anna, A. Frery, C. Renn, J. Soares and A. Luckman 1997 Exploratory Study of the Rela tionship between Tropical Forest Regeneration Stages an d SIR-C L and C Data. Remote Sensing of Environment 59:180-190. Yentsch, A. E. 1996 Introduction: Close Attention to Place-Landscape Studi es by Historical Archaeologists . In Landscape Archaeology: Reading and Interpreting he American Historical Landscape , edited by R. Yamin and K. Metheny, pp.1-16. University of Tennessee Press, Knoxville. Zech, W., L. Haumaier and R. Hempfling 1990 Ecological aspects of soil organic matter in tropical land use . In Humic Substances in Soil and Crop Sciences: Selected Readings , edited by C. McCarthy, R. Malcolm and P. Bloom, pp. 187. Amer ican Society of Agronomy and Soil Science Society of America, Madison. Zeidler, J. 1995 Archaeological Survey and Site Di scovery in the Forested Neotropics . In Archaeology in the Lowland American Tr opics: Current Analytical Methods and Applications , edited by P. W. Stahl, pp. 7-40. Cambridge University Press, New York. Zubrow, E. B. W. and J. W. Harbaugh 1978 Archaeological Prospecting: Kriging and Simulation. In Simulation Studies in Archaeology , edited by I. Hodder, pp. 109-122. Cambridge University Press, Cambridge.

PAGE 279

268 BIOGRAPHICAL SKETCH Joseph C. Russell was born in Gainesville, Florida. After completing his master’s in anthropology at Florida State University, he returned to his hometown to complete his doctoral program at the University of Florida. Not specifically tied to the state of Florida, he is currently enjoying both the weather a nd the academic climate of the University, with no immediate plans to abandon either.