|UFDC Home||myUFDC Home | Help|
This item has the following downloads:
1 PATTERNS AND PROCESSES OF LAND CO VER CHANGE: UNDERSTANDING TRADEOFFS AMONG ECOSYSTEM SERVICES By AMY ELIZABETH DANIELS A DISSERTATION PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLOR IDA IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY UNIVERSITY OF FLORIDA 2009
2 2009 Amy E. Daniels
3 To my family
4 ACKNOWLEDGMENTS I thank three important mentors for humanizing, and piquing my interest in, this mysterious thing called a PhD during my masters degree. Without this confluence of encouragement from Jane Southworth, Andres Guhl and Hugh Popenoe, I probably would not have considered pursuing this degree. I am deeply grateful to my advisory committee for granting me the freedom and challenge of defining my own research agenda. Th ey believed that what I wanted to do was worthwhile and this is the single-best form of support I could have asked for. I thank my advisor, Jane Southworth, for her critical res earch insights and for the illuminating and positive perspective that has guided me through my moments of fuzzies t thinking. Tom Ankersens enthusiasm and ability to think-outside-the-box have contributed immensely to this research. His contacts, experience and knowledge of Costa Rica have been invaluable. I am grateful to have been able to work under Toms gui dance in the Conservation Clinic in Costa Rica during the exploratory phase of this rese arch. Mike Binfords boundless curiosity is contagious and inspiring. I appreciate our disc ussions that taught me so much about how to approach complex, interdisciplinary research questions in a rigor ous fashion. I thank Brian Child for several immensely-helpful discussions and brainstormi ng sessions regarding th e social and economic dimensions of this research. Brians vast applied experience ha s been a refreshing and balancing influence for certain. Though not an official committee member, I also thank Graeme Cumming for his collaboration, support and sp ot-on advice about many issues. I am grateful to these collaborators: Katie Painter at UF; Kenneth Bagstad and Valerie Esposito at the University of Vermonts Gund In stitute for Ecological Economics; and Carlos Manuel Rodriguez at Conservation International. I thank Roger Medina Gonzalez and Ivan Dominguez Tec from the Autonomous University of Yucatan, for their hospitality and field
5 assistance in Mexico. I thank Juan Carlos So lano Montero for his hard work as my field assistant in Costa Rica and for his wonderful sense of humor that kept me laughing, even on those 100+ degree days en el horno Guanaco Natalia Hoyos is the god dess of organization! I greatly appreciate her remote sensing assistance. Finally, chapters one, three and four benefited from the comments of Billie Lee Turner, Thomas Rudel, and Alvaro Umana, respectively. I really appreciate the insight and feedback they each shared. Thanks go to the many individuals and ag encies in Costa Rica who supported and contributed to this effort. In particular, I would like to tha nk Orlando Matarrita, Alexander Leon and Francisco Ramirez from SINAC (MINAE); Jose Cubero, Alberto Mendez, Cristian Diaz, and Jose Angel Jimenez from FONAFIFO; Shirley Sanchez from the University of Costa Ricas Law School; from MAG; and Rafael Mendez from the Cattlemens Association. I am indebted to all those Guanacastecos who shared their time and knowledge with me during on-farm interviews. This dissertation was funded through a Univer sity of Florida Al umni Fellowship, a NASA Jenkins graduate fe llowship and several small research grants from the Ruge Foundation, Phi Kappa Phi Honor Society, and the NSF/IGER T Working Forests in the Tropics program (NSF-DGE-0221599). My heartfelt thanks go to th ese programs and, where applicable, to the taxpayers who fund them. I pledge to reciprocate the publics investment in me and to not forget the immense privilege that earn ing this degree has been. I thank the Land Use and Environmental Ch ange Institute (Christian Russell), Department of Geography and School of Natura l Resources and Environment for institutional support in the unique and challenging context of this university-wide inte rdisciplinary degree program.
6 I cannot imagine this experi ence without having my husband Paul Ghiotto to share it with. He is trul y the worlds best abogado vuelto a machitero I am grateful for all of his assistance in the field, and for his eternal optimism and fun-loving spiritthrough even the most challenging days of the Coco blues. Thanks go to my brother Blakely Da niels for a lifetime of love and support, and for shari ng his home with us during our pos t-fieldwork transition. Finally, I am forever grateful to all of my friends and family for having accepted and supported me unconditionally in reaching this goal, even when they may not have understood or shared it.
7 TABLE OF CONTENTS page ACKNOWLEDGMENTS...............................................................................................................4 LIST OF TABLES................................................................................................................. ........10 LIST OF FIGURES.......................................................................................................................11 ABSTRACT...................................................................................................................................13 CHAP TER 1 LAND USE AND THE PROVISION OF ECOSYSTEM SERVICES................................. 15 2 MILPA IMPRINT ON THE TROPICAL DRY FORES T LANDSCAPE IN YUCATAN, MEXICO: REMOTE SENSING & FIELD MEASUREMENTS OF EDGE VEGETATION........................................................................................................... 21 Summary.................................................................................................................................21 Introduction................................................................................................................... ..........22 Study Region................................................................................................................... .......26 Methods..................................................................................................................................26 Field-Based Vegetation St ructure Measurements........................................................... 26 Image Processing and Geographic Information Systmes Procedures............................. 27 Statistical Analyses.......................................................................................................... 30 Results.....................................................................................................................................31 Milpa Spatial Characteristics........................................................................................... 31 Milpa Edge Effects.......................................................................................................... 31 Discussion...............................................................................................................................33 Conclusions.............................................................................................................................37 Acknowledgements.................................................................................................................38 3 CONVERSION OR CONSERVATION? UNDERSTANDI NG WETLAND CHANGE IN NORTHWEST COSTA RICA.......................................................................................... 47 Summary.................................................................................................................................47 Introduction................................................................................................................... ..........48 Methods..................................................................................................................................51 Study Region................................................................................................................... 51 Overview of Methods and Modeling............................................................................... 52 Land Cover Data and Predictor Variables....................................................................... 53 Statistical M odel Building...............................................................................................56 Model Validation and Performance Assessment............................................................. 58 Results.....................................................................................................................................59 Model Selection...............................................................................................................59 Principal Components Analysis...................................................................................... 60
8 Spatial Predictions...........................................................................................................61 Model Performance.........................................................................................................61 Model Coefficients.......................................................................................................... 62 Discussion...............................................................................................................................63 Drivers of Wetland Change............................................................................................. 63 Assessment of Model...................................................................................................... 64 Conservation Implications............................................................................................... 66 Conclusions.............................................................................................................................69 Acknowledgements.................................................................................................................69 4 A DECADE OF PAYMENTS FOR ENVI RONMENTAL SERVICES ( PES): BUILDING ON COSTA RICAS MODEL AND APPLYING LESSONS LEARNED......81 Summary.................................................................................................................................81 Introduction................................................................................................................... ..........81 History and Trends in Co sta Ricas PES Program................................................................. 83 PES Evolution and Scheme Design................................................................................. 83 Empirical Trends for Costa Rican PES........................................................................... 86 Evaluating Costa Ricas PES Program...................................................................................89 PES Administration.........................................................................................................89 Opportunity Costs............................................................................................................93 Ecosystem Service Bundling........................................................................................... 95 Sustainable Financing...................................................................................................... 96 Equity in Funding PES....................................................................................................99 Spatial Variability and PES Targeting.......................................................................... 100 Socioeconomic Objectives............................................................................................ 104 Conclusions...........................................................................................................................106 Acknowledgements...............................................................................................................109 5 FOREST EXPANSION IN NORTHWEST COSTA RICA: CONJUNCTURE OF THE GLOB AL MARKET, LAND-USE IN TENSIFICATION AND FOREST PROTECTION..................................................................................................................... 118 Summary...............................................................................................................................118 Introduction................................................................................................................... ........118 Tempisque Basin: Geographica l and Historical Setting....................................................... 121 Forces of Landscape Change................................................................................................124 Methods................................................................................................................................127 Land-Cover Classification and Change Detection........................................................ 127 Trajectory Analysis....................................................................................................... 129 Dominant Explanat ory Trajectories..............................................................................130 Landscape Setting for Trajectories................................................................................ 131 Results...................................................................................................................................131 Land Cover Area: Net Trends....................................................................................... 131 Explanatory Trajectories...............................................................................................132 Protected and Non-Protected Landscape Comparisons................................................. 133 Landscape Setting for Dominant Trajectories............................................................... 133
9 Discussion.............................................................................................................................135 Acknowledgements...............................................................................................................143 6 CONCLUSIONS.................................................................................................................. 156 APPENDIX: MODEL BUILDING AND MODEL SELECTION STEPS FOR PREDICTI NG WETLAND CONVERSION....................................................................... 161 Conceptual Model of Wetland Change in the Tempisque Basin.......................................... 161 Flowchart of Modeling Process............................................................................................ 162 Data from Akaike Information Criterion Comp arisons for Variable Selection and Model Selection............................................................................................................................163 LIST OF REFERENCES.............................................................................................................164 BIOGRAPHICAL SKETCH.......................................................................................................183
10 LIST OF TABLES Table page 2-1 Mean milpa patch metrics acr oss three years of analysis. ................................................. 39 2-2 Mean forest structure measurements for milpa -edge and control sites.............................. 40 2-3 Mancova results for three da tes indicating the effect size ( p 2) of each covariate and the milpa indicator, along with their significance.............................................................. 41 2-4 ANCOVA/ANOVA results for the three dates in the analysis. Bonferroni-adjusted significance is indicated for the milpa indicator for each vegetation index...................... 42 3-1 Construction of wetland trajectorie s using three binary land cover maps (wetland/non-wetland).......................................................................................................71 3-2 Response variable (wetland trajectory) and proposed predictor variables and interaction terms, along with their theore tical justification, for wetland conversion model..................................................................................................................................72 3-3 Eigenvalues and percent variance acc ounted for by each of the three principal components extracted through PCA.................................................................................. 74 3-4 Rotated structure matrix for the seven original predictor variables................................... 75 3-5 Model performance statistics computed with independent observations and model coefficients estimated with (Model 1) a nd without (Model 2) wetland patch size............76 4-1 Legal status of PES modali ties over the last decade. The legal citations listed in parentheses are referenced in th e citation column on the right........................................ 110 4.2 Summary of issues and recommendati ons for Costa Ricas PES program..................... 111 5-1 Protected area in the Tempisque Basin ove r the three dates of land cover analysis........ 144 5-2 Description of land cover classes (abbreviations in parentheses).................................... 145 5-3 Of the possible 64 trajectories (left), only the dominant e xplanatory trajectories were retained for analysis of each trend of net land cover change........................................... 146 5-4 Population means ( ) of key variables representi ng physical and socioeconomic landscape setting in the Tempisque Basin, st ratified by protected and non-protected status......................................................................................................................... .......147 5-5 Population means ( ) of variables representing physical and socioeconomic landscape setting in the Tempisque Ba sin, stratified by unique land cover trajectories........................................................................................................................148
11 LIST OF FIGURES Figure page 2-1 Interactions and feedbacks between land use ( milpa agriculture) and land cover properties............................................................................................................................43 2-2 Map of Peto municipality study region in the southern, central region of Yucatan, Mexico.Peto study area is shown as a 5,4,3 (a s R,G, B) composite subset from an April 1988 Landsat 5 Image.............................................................................................. 44 2-3 Examples of intermediate remote sensi ng products used to isolate clearing edges from 2003...................................................................................................................... .....45 2-4 Flowchart illustrating the protocol for ex tracting vegetation indi ces for forest region buffering milpas (top) and background forest (bottom) for a small subset of Peto study region (RGB:4,3,2)...................................................................................................46 3-1 Map of the Temspique River Waters hed in northwestern Costa Rica............................... 77 3-2 Flowchart of image rule-based land cove r classification proce dure used to derive binary wetland and non-wetland land cover maps for 1975, 1987 and 2000 Landsat images................................................................................................................. ..78 3-3 Map of landscape features in the Temp isque Basin that influence wetland conversion..................................................................................................................... .....79 3-4 (a) Map of probability of wetland conver sion. In the non-protected watershed, probability generally increases as distance from central, lower watershed increases, suggesting a sweeping typology of conve rsion. Within PVNP, the typology is much more patchy and the probability of conversion is generally much lower, though some areas still may be at risk for conversion (b) Map of actual wetland trajectories and model prediction errors............................................................................................... 80 4-1 Timeline detailing the evolution of PES in Costa Rica................................................... 112 4-2 (a) Cumulative area in PES (thousands of ha) as a function of the cumulative budget (millions US Dollars) FONAFIFO receives to implement PES (b) Time series of cumulative PES area (thousands of ha) recr uited across all modalities (square) in contrast with net area where expired contra cts are subtracted off of the running sum (circle)..............................................................................................................................113 4-3 Time series of the recruited area per moda lity and number of trees in the agroforestry modality (SAF)................................................................................................................ 114 4-4 A conceptual multi-dimensional production possibility frontier for ecosystem services.............................................................................................................................115
12 4-5 Costa Ricas m odel for using instituti ons to bundle services linking buyers and sellers across different spatial scales................................................................................ 116 4-6 Relative sustainability/s ecurity of financing mechanisms for PES................................. 117 5-1 Map of Tempisque River Basin in northwest Costa Rica (5414 km2)............................. 149 5-2 Four class land cover maps (plus water) for 1975, 1987 and 2000................................. 150 5-3 Trends of net area land-cover change for 1975, 1987 and 2000 in the Tempisque River Watershed. The black region of each bar indicates the contribution of each land cover class within protected areas............................................................................151 5-4 Area protected through various forms of forestry incentives for the Tempisque Basin and at the national level................................................................................................... 152 5-5 Dominant land cover trajectories explaini ng the observed net area changes in land cover.......................................................................................................................... ...........1 5-6 Graph of dominant trajectories explaining the expansion of forest observed by year 2000 for protected and non-protec ted areas of the landscape.......................................... 154 5-7 Conceptual model of dominant trends accounting for forest expansion in the Tempisque Basin. The landscape is represen ted by the large recta ngle, divided into uplands (gray) and lowlands (white)................................................................................155
13 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 PATTERNS & PROCESSES OF LAND COVER CHANGE: UNDERSTANDING TRADEOFFS AMONG ECOSYSTEM SERVICES By Amy E. Daniels August 2009 Chair: Jane Southworth Major: Interdisciplinary Ecology Patterns and processes of land use and land cover change are among the most critical causes and consequences of modern-day global environmental change. Anthropogenic land use simplifies ecosystem structure and function a nd so, understanding tradeoffs in ecosystem services begins with characterizing landscape dynamics. Using case studies from Mexico and Costa Rica I identify spatial typologies of land cover change, relate these patterns to driving forces, and identify institutiona l approaches that complement ecosystem service provision. In Chapter One, I first examine the legacy of a la nd-use system indigenous to southern Mexico on the vegetation structure of tropical dry forest and then characterize changing spatial landscape patterns. Chapter Two identifies the driving forces of wetland conversion in northwest Costa Rica. Through the development of an empirical spatially-explicit convers ion model, I classify important spatial typologies of wetland loss. Recognizing the spatial patterns of conversion within the broader economic and biophysical cont ext is critical for evaluating tradeoffs and designing spatial-targeting criteria for environmental policies. In Chapter Three, I analyze PES policy and implementation in Costa Rica in or der to suggest design nuances, including the incorporation of greater spatia l nuance to enhance landscape-lev el ecosystem service provision. In Chapter Four, I synthesize landscape dynamics in a conceptual model that highlights the
14 tradeoffs in landscape-level ecosystem services and the role of conservation institutions in resulting land cover dynamics. In conclusion, la nd use patterns considered in both space and time represent an emergent expression of emphasized mix of ecosystem services, whether intended or unintended. Understanding landscape dynamics is essential in designing institutions to manage tradeoffs and encourage syne rgies necessary for societys well-being.
15 CHAPTER 1 LAND USE AND THE PROVISION OF ECOSYSTEM SERVICES Land use and land cover change are among the mo st critical causes and consequences of modern-day global environmental change (Foley et al. 2005). This is illustrated by examples across many geographic scales. Land cove r change accounts for roughly 20% of global greenhouse gas emissions (Baumert et al. 2005) and 35% of CO2 emissions, the most important greenhouse gas (Houghton & Hackler 2001). Conversi on of natural habita t to direct human land use is the greatest driver of species extirpation (Wilson 1988, Dirzo and Raven 2003). At the local scale, direct human land use is found to reduce ambient biodiversity levels below even what todays unprecedented extinction rates would predict (Polasky 2005). Interactions between these environmental-change processes may rein force or mitigate their respective negative consequences. For example, climate change resulting from rising greenhouse gas concentrations further exacerbates net losses in biodiversity. In contrast, landscapes experiencing forest transitions (i.e., significant forest regrowth) mitigate climate change through carbon sequestration associated with fore st regeneration. Secondary forest s may also better facilitate the processes that generate and maintain biodiversity, rela tive to other land uses. Land cover designations indicate generalized ecosystem propert ies and a particular range and nature of ecosystem functions. The ma ss ratio hypothesis offers a specific ecological mechanism linking generalized land cover and eco logical function. The hypothesis holds that trait values (e.g., vegetation height specific leaf area, etc.) of the dominant contributors to plant biomass in an ecosystem largely determine ecosy stem functioning (Loreau et al. 2001). Diaz et al. (2007) found that combinations of trait va lues and abiotic factors effectively explain ecosystem functions in a subalpine grassland sy stem. Ecosystem functioning, in turn, either
16 directly or indirectly characterizes ecosystem services, or the goods and servic es derived from ecosystem processes that benefit society. Land use decision-making and resulting la nd cover changes entail important tradeoffs in the provision of ecosystem services. The dominant contemporary anthropogenic land use patterns tend to simplify ecosystem structure and function, emphasizing production-oriented services like crop cultivation (F oley et al. 2005). The efficiency achieved through simplification of ecosystem functions has clear advantages. Food production at the global scale, for example, has kept pace with population growth; indee d, surplus agricultural yields constitute the foundation upon which elaborate commerce has deve loped. Nonetheless, as society proceeds domesticating the biosphere (see Kareiva et al. 2007), tradeoffs are increasingly clear. At the local scale, a vast suite of ecosystem servicesfrom infectious disease mediation to water quality regulationare foregone or drastically diminished (Rodriguez et al. 2005) through industrial, monoculture agriculture or other land uses th at appreciably simplify ecosystem processes. The role of landscape pattern Land use patterns develop in relation to phys ical gradients in the landscape. Huston (2005) holds that spatiote mporal patterns of land use are li nked to net primary productivity in different ways as economic development progre sses through agricultural, industrial and then information phases. Each phase is characterized by a unique pattern of resource dependence. Thus land use may be an emergent expression of the value that society collectively places on particular ecosystem services at a given point in time. More than representing the result of land use tradeoffs, the spatial pattern of land use and land cover also plays a critical role in the provision of ecosystem services. The arrangement of land cover patches determines the nature
17 and degree of interaction among and between biologi cal and physical elements of the landscape. For example, Ricketts et al. (2004) found that coffee plantations within roughly one kilometer of forest land cover benefited from wild pollinator species. Yields increased by 20% and the frequency of small, misshapen beans decreased by 27%. The geographic pattern of land use also confer s certain path dependencies that constrain future landscape dynamics while simultaneously ge nerating an aperture of opportunities. For example, drainage of wetlands for agriculture e liminates the ecosystem service of storm-water runoff storage but enhances the food production service. Even after agricultural abandonment, the wetland ecosystem and services it provi ded may not return depending upon the level of hydrologic and geomorphic disrupt ion involved in the drainage If the land remains dry, afforestation may ensue, illustrating how previous land use redirects future ecosystem functioning and services. The spatial pattern of habitat in a landscap e may function as biologi cal insurance in the sense that migration and dispersal across ecosy stems plays a key role in maintaining local biodiversity in heterogenous, mixed-use landscapes (Loreau et al. 2003). Thoughtful landscape planning can maximize this insurance function, al ong with other ecosystem services, for a given sum area of ecosystem or land-cover type pres ent in the landscape, even while ensuring economically-productive land uses. Polasky et al. (2005) found an L-shaped and relatively-flat efficiency frontier between aggregate economic re turns (x-axis) and defined conservation goals (y-axis) when iteratively modeling simulated land-u se patterns. These results indicate that many different landscape patterns achieve intermedia te-to-high levels of bot h objectives but that maximizing either one results in precipitous losses of the other. Eliminating externalitiesthat
18 is, charging for ecosystem service degradation and rewarding ecosystem service provision further diminishes the perceived tradeoffs between environmental and economic outcomes. The role of institutions In reality, theoretically-ideal landscape patt erns are attainable through simulation, but are unlikely to arise spontaneously or even by design in complex landscapes with myriad ownership arrangements and, often, countervailing economic incentives. Institut ions, through policies, management practices and soci al norms, may mediate land use systems, influence ecosystem dynamics and drive land cover trajectories (H arrison 1991, Chomitz and Gray 1996, Southworth and Tucker 2001, Foster et al. 2003). Diminishin g tradeoffs between the provision of ecosystem services and classic economic production functions hinges on appropriate, dynamic institutions. Protected area establishment comprises a corner stone strategy of conservation institutions and parks have been established in many landscapes Indeed, protected areas cover over 11% of the earths surface (Chepe et al. 2005). The expense, both financially and socially, of wholesale land purchases associated with tr aditional protected area establishment limits the degree to which parks may influence overall landscape composition and pattern. Park locations are often chosen by narrow criteria, like species re presentation. Alternatively, park s are established by default on low-productivity lands such as rock and ice, places that are unlikel y to be exploited even without protection (Hazen and Anthamatten 2004). The curr ent global network of protected areas may contribute to uneven protection of critical ecosystem services if reserve selection proceeds according to current strategies (Pyke 2007). Given the limited function of protected area es tablishment, along with the importance of landscape pattern in the provision of ecosystem services, institu tions that influence private property land-use choices should prove useful for shaping landscape patterns in ways that
19 enhance ecosystem service provision. One such strategy is payments for ecosystem services (PES) which attempts to eliminate externalities associated with targeted ecosystem services. PES is a nimble instrument that may be configur ed to pay a determined amount directly to land owners who provide a particular ecosystem service or suite of services for a prescribed duration. PES contracts may be geographica lly-targeted at properties impor tant for achieving particular ecosystem elements, meeting some land cover area threshold and/or which are integral components of some broad-scale beneficial land scape pattern. An import ant research theme is to determine in what contexts this shift in land tenure and economic incentive structures prescribed by PES is appropriate. Understanding tradeoffs in ecosystem servi ces begins with characterizing landscape dynamics, identifying spatial typolo gies of land cover change, rela ting these patterns to driving forces, and identifying institutional approaches that complement ecosystem service provision. I address all of these themes in the following chapte rs of this dissertation. In the first, I examine the legacy of a land-use system indigenous to Mexico on tropical dry forest structure and changing spatial landscape patter ns. To what extent does pa st land use affect ecosystem properties directly related to the provision of forest ecosystem serv ices? In the second chapter I identify the driving forces of wetland convers ion and, through a spatia lly-explicit conversion model, I classify important spa tial typologies of wetland loss. Recognizing the spatial patterns of conversion within the broader economic and biophysical context is a critical for evaluating tradeoffs and designing spatial-targ eting criteria for environmental policies. In chapter three, I analyze PES policy and implementation in Costa Rica in order to suggest design enhancements, including the incorporation of gr eater spatial nuance. In the fourth chapter, I synthesize landscape dynamics in a conceptual model that highlights the tradeoffs in landscape-level
20 ecosystem services and the role of conservation institutions in emergent land cover dynamics. Finally, in the concluding chapter I synthesize what pattern analysis, models of landscape change, a decade of empirical PES trends and an integrated understand ing of the role of conservation institutions in landscape dynamics has to say about tradeoffs in ecosystem services.
21 CHAPTER 2 MILPA IMPRINT ON THE TROPICAL DRY FOREST LANDSCAPE IN YUCATAN, MEXICO: REMOTE SENSING & FIELD ME ASUREMENTS OF EDGE VEGETATION Summary The Yucatan Peninsula hosts part of Central Am ericas largest remaining tract of tropical dry forest and has been identified as a regi on of critical landscape change. This study complements the extensive research on land cove r conversion in the region by investigating a subtle but important aspect of forest modi fication. We examine changes in the spatial characteristics of milpa cultivation plots in the swidden landscape of Peto municipality in Yucatan state from 1988 to 2003 using remote se nsing. We also test the hypothesis that milpa clearings create a discernible edge effect in terms of forest structure. Re sults indicate that spatial patterns of milpas have changed over time. The amount of milpa /forest linear interface increased over the study period. Both satellite-based ve getation indices and field-based canopy cover measurements indicated that forest buffering milpa clearings had significantly lower biomass than background forest, despite that the backgrou nd forest is itself a mosaic of successional forest stages. In contrast, there was no differen ce in stand basal area for milpa edge forest and background forest. Multivariate m odels demonstrated that the milpa edge indicator was the most important variable in explaining differences of vegetation indices for milpa edges and background forest compared with other factors that create edges in the la ndscape. Models were relatively effective in explaining mean values of vegetation indices; but th ey performed poorly in terms of explaining measures of forest vegetati on heterogeneity. Comparing model results from each date suggests that the importance of milpa edges decreases over time, possibly as a function This chapter is an authorized re-print: Daniels, Amy E., K. Painter, and J. Southworth. 2008. Milpa imprint on the tropical dry forest landscape in Yucatan, Mexico: remote sensing & field measurements of edge vegetation. Agriculture, Ecosystems and Environment 123(4): 293-304.
22 of the accumulated land use history as milpas rotate through the forest matrix. Evidence supports the notion that the eff ects of milpa land use extend be yond the clearing itself and into adjacent forest. Introduction Human land use shapes ecosystem structure and function at multiple scales of time and space (Turner et al.. 1995). Among the most signi ficant global challenges in the next century relates to management of the transformation of the earths surface occurring through changes in land use and land cover (Mustard et al. 2004). Mu ch land change science research for tropical regions focuses on deforestation using discrete land cover cla ssifications to study wholesale conversion, like deforestation (Def ries et al. 2000). Land cover modifications, such as forest degradation without wholesale cl earing, also merit attention. Such modifications may have drastic effects on ecosystem processes and services like species composition and richness (Ferguson et al. 2003), trophic pathways (Bunn et al. 1999), carbon sequestration (Lawrence and Foster 2004), and land-surface energy balance (Southworth 2004). The tropical dry forest life z one supports much of the wo rlds agriculture, having more productive soils than those of humid tropical forests (Murphy and Lugo 1986). The dry forest life zone covers 42% of all land ar ea in tropical latitudes. Desp ite this geographic predominance, fewer studies have been conducted on this forest type relative to moist and wet tropical forests (Ramankutty et al. 2006, Prez-Salicrup et al. 20 04). The Yucatan peninsula of southeastern Mexico hosts part of the largest remaining expanse of seasonally-dry tropical forest in Mesoamerica and has been identified as a hot spot for tropical deforesta tion, in part related to government-sponsored agrarian settlement programs (Chowdhury and Turner 2006). Agriculture is the major driver of land c over change in tropical regions (Lambin et al. 2001). Swidden agriculture, in pa rticular, comprises a major land use and an important resource-
23 management system in many parts of the tr opics (Coomes et al. 2000). An extensive ethnographic literature classifies and describes these cultivation systems throughout the world (Unruh 1990, Banerjee 1995, Teran and Rasmussen 1995). Milpa is the traditional form of recurrent swidden in Mesoamerica. It is base d on rotation of maize fields and fallows, during which secondary forests are established to replenish organic matter and nutrients. Milpa cultivation in the Yucatan Peninsula genera lly occurs in conjunction with communal ejido tenure. This coupled land-use and land-tenure system has sustained cu ltivation on poor soils by regulating the number and timing of milpa fields; the amount of ejidal forest land (through deliberate set-asides); and the effects of population growth on land resources through generally non-divisible inheritance rights (Plaza 2000). In 1992, however, an amendment to Article 27 of the Mexican constitution provided a multi-step process through which ejido members may elect to priv atize their long inalienable ejidal land. This amendment occurred as part of a national agenda to create an institutional framework favoring private investment, the development of a land market and productivity gains in agriculture (Johnson 2001). Other efforts aimed at modernizing Mexican agriculture include extension programs like PROCAMPO, PR ONASOL, and other economic incentives that promote sedentary, intensive cultivat ion (Chowdhury and Turner 2006). Of all forested land in Mexico, 85% occurs in ejidos, making the country unique among both developed and developing nations (White a nd Martin 2002). In Yucatan state, most ejidos are dominated by milpa land use. At broad spatial scales in the Yucatan, the precipitation gradient is the most important variable drivi ng vegetation patterns (Lawrence and Vester 2004). At local scales, like within an individual ejido however, the structure and function of Yucatecan forestsfrom litter production to biomass and soil propertiesare more strongly influenced by
24 forest age and history of cultivation than by e nvironmental gradients (Tur ner et al. 2004). The process of clearing land for swidden cultivation an d the subsequent regeneration of forest during fallow periods creates a landscape mosaic of active milpas forest patches in various stages of succession, and edge forest at the linear border between the two. Many biophysical, structural and flor istic changes occur as a result of milpa /forest adjacency and due to the use of fire in creating the milpa clearing (Eaton and Lawrence 2006). These changes may reinforce, or be reinforced by, land use feedbacks. One such feedback is that after clearing vegetation to make a field, forest at the interface imme diately experiences an increase in both photosynthetically active ra diation (PAR) and wind penetration at the ground level; altered albedo; and changes in surface ener gy and water balances. Subsequent land use in the edge forest may include collection or harves ting of fuel wood and cultivation of fruit trees (Ochoa-Gaona 2001). Both these biophysical and land-use effects reinforce longer-term changes in soil properties, floristic composition, and plant and animal dispersal (Figure 2-1). Remote sensing and geographic information science have become standard tools for addressing these complex human-environment inte ractions at the landscape-level. Remotely sensed vegetation indices have also proven useful for coarse -scale biodiversity assessment, complementing field-based surveys (Nage ndra 2001). With such great emphasis on deforestation in the land change science arena, only recently have researchers begun to address more subtle issues of re-growth/succession a nd other qualitative forest changes and feedbacks (Chowdhury et al. 2004, Moran 2004, Rudel et al.. 2005) Most such studies still rely on the use of discrete land-cover classifications; yet a more effective approach is to use the full suite of continuous data available from satellite imag ery. NDVI (normalized differenced vegetation index) has long been used as a proxy for biomas s (Jensen 1996) and to study structural forest
25 attributes like canopy architecture (Eamus 2001). Similarly, thermal band data have been useful in discriminating successional stages of forest (Southworth 2004) given that surface energy balance is linked to the character of land cover and thus past land use. With continuous data, both land-cover conversions and within-class modifications are detectable. Not only can conversion from agricultu re to successional fore st be seen after field abandonment, but also within-class changes. The latter includes processes such as changes in forest density, forest degradation or the ability to identify a gr eater number of successional stages as a forest matures (Southworth et al. 2004, Bonan et al. 2003, DeFries et al.. 2000). Illumination of within-class changes greatly enhances the ability to fore see potential wholesale conversions before they occur, a key issue for biodiversity conservation and policy monitoring. The critical link between forest cover and communal land tenure in the region underscores the importance of analyzing landscap e patterns and forest attributes since the amendment of the constitution. A spatially explic it, landscape-level analysis ma y facilitate a better understanding of the effects of broader neo-liberal policy shifts by addressing the spatial pattern of recurrent clearings, or the landscape context that influences forest/clearing adjacency within the mosaic of successional forest stages comprising swidden-dominated land use systems. In this research we examine whether the spa tial characteristics of milpa clearings have changed over time (19882003) using remote sensing. We refer to the patter ns or characteristics of forest vegetation as structure. With both image and fieldbased data we test the hypothesis that milpa clearings create detectable edge effects compared to the background forest matrix (itself a mosaic of successional stages) in terms of several measuremen ts of forest structure. With multivariate models, we also examine how this swidden-i nduced land cover juxtaposition compares with other factors that create li near edges and affect forest vegetation characteristics.
26 Study Region The study site lies in the inte rior, central portion of the Yucat an Peninsula of southeastern Mexico, within the state of Yucatan ( Figure 2-2 ). The landscape is composed of areas of mechanized, monoculture cultivation; large regions of forest within which swidden agriculture occurs (mostly in ejidos ); and built areas of roads, communities, and other infrastructure. The region has a marked seasonality, with most precipitation occurri ng between May and October. A precipitation gradient exists across the peninsula, increasing toward the south. Because of the prominent karst topography of the region, the peninsula has virtually no surface hydrography. Most of the Yucatan Peninsula fa lls within the tropical dry fore st life zone (Holdridge 1971). This includes subdeciduous tropical forest, deciduous tropical forest and scrub areas. Within this region, we selected the municipality of Peto (3296 km2), in the southern, central region of Yucatan state, as a subset for further analysis (Figure 2-2). Rural areas of the municipality are characterized by milpa agriculture within a forest matrix. Milpa farmers generally reside in comm unities and commute to their nearby milpa s daily. Methods Field-Based Vegetation Structure Measurements We conducted fieldwork in March of 2004 and 2005, late in the dry season. The pointquarter method, a variable plot size technique (Bell and Dilworth 2002), was employed at randomly-selected milpa edge and forest transects to calc ulate stand basal area and estimate canopy cover (n = 19). Each 60 m transect consiste d of three plots (n = 57), where we measured the diameter at breast height (dbh) of the nearest tree (> 2.0 cm) to the center of the plot for four quadrants, along with each trees distance from th e plot center. We also estimated canopy cover at each plot. We conducted semi-structured interviews with milpa farmers in an opportunistic
27 fashion during field work to gain a better appr eciation of the social dimensions of swidden cultivation and of land use history of key areas within the study site. Image Processing and Geographic Information Systems Procedures Three Landsat TM images covering the st udy area were obtained through the Working Forests in the Tropics (WFT) Program at the Un iversity of Florida (N SF-IGERT). All image processing was carried out us ing ERDAS Imagine 8.7. Dates selected were 1988, 1994, and 2003, with all imagery from the end of the dry season. The 1988 image represents the landscape prior to the ejido privatization amendment, whereas the 1994 and 2003 dates represent the landscape shortly after and roughly one decade after the amendment, respectively. The 2003 image was used as our base map after georefer encing it to regional t opographic maps (1:50,000), obtaining a root mean squared error (RMSE) of less than half of a pixel (< 15 m). The remaining images were then geometrically registered to the base image through image-to-image registration (RMSE < 15 m). Final positional accuracy of geoc orrected images was validated in the field with the aid of a handheld GPS ( positional error < 5.5 m). Images were calibrated, to correct for sensor gain, atmospheric distor tion, and differences due to non-anniversary image dates (Green et al. 2005). Band 6, thermal emission (low gain) was also calibrated an d converted to black body temperatures (BBTemp) in Kelvin. We subset all images to Peto municipality, ma sking out urban areas, i ndustrial agriculture, and commerciallydeveloped roadside so that only forest cover and clearings remained. Unsupervised classifications were performed for all images to identify milpa clearings, pasture clearings and forest (Figure 2-3b). Sixty-four training samples from fieldwork in March of 2004 were used for accuracy assessment (overall accuracy 88% and Kappa 0.79). Two vegetation indices were calculated for each year for th e area corresponding to the forest in the classifications. The first was the norma lized difference vegetation index (NDVI),
28 and the second was a thermal-band ratio, Milpa clearings that resulted from image classification did not yield sufficiently precise, linear edges for the purposes of th is analysis. To surmount this challenge, texture analysis was employed as follows. Principal Components Analysis (PCA) was performed for each image date on all bands minus the thermal band1. Texture analysis was performed on the first three resulting PCs for each image date using a 3 x 3 pixel window. Texture was calculated as where xij is the PC score for pixel (i,j), xc is the PC score for cente r pixel of window (kernel), and n is number of pixels in a window. The texture images clearly and precisely deline ated the edges of clearings in the forest (Figure 2-3c). Thresholding of high texture valu es, compared with the relatively low texture values of the background forest, was used to creat e an image of raster cel ls corresponding only to the edges of clearings (Figure 2-3d). These raster edges were vectorized to make polygons out of all clearings in the forest. These forest cl earing polygons for each year were overlaid on their 1 Due to their coarser spatial resolution, thermal bands were less useful in identifying milpas. For Landsat 7, resolution on the thermal band is 60 x 60 m, but on Landsat 5 it is 120 x 120 m. Band 4 (IR) Band 3 (R) Band 4 (NIR) + Band 3 (R) Band 4 (IR) Band 3 (R) Band 4 (NIR) + Band 3 (R) (2-1) Band 6 (thermal) Band 5 (MIR) + Band 4 (red) (2-2). [ (xc xi j )2 ]1/2 ( n 1) (2-3)
29 respective unsupervised classifi cations and subsets of the pol ygons were created responding only to milpa clearings (i.e., excludi ng pasture clearings). From the milpa polygon vector files for each date, we filtered patches fine r than the level of interest for this research (< 0.09 ha) and calculated milpa area, perimeter and average density for each image date. A spatially-distr ibuted random subset of 50% of the milpa polygons was selected for each year (nmilpa= 527, nmilpa= 796, and nmilpa = 819 for 1988, 1994 and 2003 respectively). This ensured sufficient sample size for multivariate analysis while reducing the problems associated with spatial autocorrelation. Th e latter inflates the ri sk of type I error in statistical analyses (Hoe ting et al.. 2006). Each milpa polygon was buffered, creating a zone corresponding to two pixels, equal to the length of the 60 m field transects. Overlapping buffer areas for closely-spaced milpa s were subtracted so that overlap was eliminated from the analysis. The mean and variance were calculated for the NDVI and thermal band ratio within the area corresponding to the milpa buffer polygons. Only forest pixels were included in the vegetation index computations so that if a given milpa was located along a road or other land cover type, only the adjacent forest pixels for that milpa were included. Further, an y effect from shadow that may affect the dependent variables was averaged out by using ratios in the vegetation indices. These vegetation index summary statistics for milpa buffers were extracted from the GIS and exported to a database (illust rated in Figure 2-4). To sample the remotely-sensed vegetation i ndices for the background forest matrix while controlling for the size and shape of the sample units (i.e., n uniquely-shaped milpa buffers), random points were generated within the backgro und forest and the coordinates of those random points were assigned to milpa buffer polygons. This randomly re-positioned the milpa buffer sample units over the background forest matrix ( illustrated in Figure 2-4). Buffer overlap was
30 again subtracted to exclude any such overlap and indices were extr acted as done for the milpa buffer regions ( nforest= 519, nforest= 776, and nforest= 796 for 1988, 1994, and 2003 respectively). Finally, distance from the sample centroid to the nearest road and the nearest forest clearing (be it milpa or pasture) was calculated for all samples. Statistical Analyses To determine whether spatial landscape characteristics of the milpa /forest mosaic have changed over time, we compared combined milpa patch metrics ( milpa area, milpa perimeter and the perimeter-to-area ratio) across three da tes using MANOVA with follow-up ANOVAs for comparisons of individual metrics across the three dates. We tested for differences in means of stand basal area and canopy cover between milpa -edge sites and forest si tes using Mann Whitney tests with Bonferoni corrections for multiple comparisons. The means and variances of both vegetation indices were compared for milpa edges and the background/control forest for ea ch image date using MANCOVAs. Milpa -edge indicator ( milpa buffer zone or forest site) was the binary, inde pendent variable of inte rest. Covariates in the model were distance to nearest road and dist ance to nearest forest clearing. The effect size for independent variables was a ssessed via partial eta-squared ( p 2) values which indicate the percent of remaining variance explained by a give n independent variable after accounting for the effects of other independent variables in the model (i.e.,, SSeffect/[SSeffect + SSerror]) (Norusis 1990). Finally, ANCOVAs were used to test for a difference of means for individual vegetation indices ( milpa -edge versus control forest sites) for each time step. ANOVAs were used when covariates were not significant for a particular index. Appropriate descriptive statistics were calc ulated for all variables prior to each analysis, including checking for correlations among dependent variables, between dependents and factor
31 levels (where applicable), and between covariates Assumptions for each statistical test or model were tested appropriately. All analyses were performed on standardized variables using SPSS 11.5. Results Milpa Spatial Characteristics MANOVA revealed that significant diffe rences of means for the combined milpa patch metrics exist across the three dates of the analysis (Wilks : 0.836; F(6, 8562) = 133.51; P=0.000). The number of milpa s increased over time from 1054 in 1988 to 1639 in 2003 (Table 2-1). Comparing individual metrics in pair-wise fash ion for the three dates showed no significant difference in mean milpa size (p > 0.45 for all cases), with the average milpa being slightly less than 1.5 ha in area. In contrast, milpa perimeter was different for each date (p < 0.00 for all cases), increasing from 5 83.95 m to 729.11 m from 1988 to 1994 but decreasing to 641.48 m by 2003. Mean perimeter to area ratio was the sa me for 1988 and 1994 at just less than 0.06 m/m2 (p = 0.633), while the ratio was 0.061 m/m2 in 2003 (p<0.000). Though st atistically significant, this difference is miniscule on the ground. Milpa Edge Effects Mann Whitney results (Table 2-2) comparing field measurements of vegetation structure for milpa -edge and control sites indicate no di fference in stand basal area for the milpa -edge (x = 31.61 m2/ha) than the control (28.3 m2/ha)(p = 0.137). Canopy cover is greater on average for the control (37.0%) compared with the milpa -edge sites (7.1%) (p < 0.00). The three separate MANCOVAs (Table 2-3) re veal significant effects of all covariates and the milpa indicator for each date (f or all independent variables, p = 0.000) and that significant differences exist among combined vegetation for milpa and control sites for each date (for all models, Wilks < 0.67; P = 0.000). The partial eta-squared ( p 2) values indicate that,
32 after accounting for the effects of the two covariates, the milpa indicator explains nearly half of the remaining variance in the co mbined vegetation indices in 1988 ( p 2 = 0.470). For 1994 and 2003, the milpa indicator respectively e xplained roughly one-third ( p 2 = 0.300) and one-fifth ( p 2 = 0.193) of the variance remaining after accoun ting for the effects of proximity to the nearest road and nearest forest clearing. Not all covariates were found to be significan t for all vegetation indices across the three years of the study. ANCOVAs (or ANOVA where a pplicable), performed separately for each time step, revealed significant effects for the milpa indicator for all four vegetation indices, for all years (p < 0.001 for all cases) (Table 2-4). For 1988, 1994 and 2003, means for both NDVI and thermal band ratio indices were greater for control sites relative to forest at milpa edges (p = 0.000). The magnitude of the difference between milpa edge and control sites decreases over time, however (Table 2-4). In terms of the indices variance for all years, milpa edge sites proved to be more heterogeneous than contro l sites (p = 0.000) with one exception. The exception is the thermal band ratio index va riance for 1994 where the control site was significantly more hete rogeneous than the milpa edge site (p = 0.000). Across all years of the study, mean values of both NDVI and thermal band ratio indices are better predicted than the vari ance of the indices as evidenced by consistently higher R2 values (Table 2-4). Except for the thermal band ratio index mean for 1994, the simple models with only three predictor variables explaine d nearly or over half of the variance in mean vegetation index values. In contrast, vegetati on heterogeneity for both indices across all years was poorly explained with the highest adjusted R2 value at 0.343 for 1988 (NDVI).
33 Discussion Swidden agriculture, or regions characterized by this land use, have primarily been examined in terms of the way that cultivation decreases the amount of forest cover (Myers 1993, Lawrence et al.. 1998, Vance and Geoghegan 2004). Of the research that has focused on the full, episodic swidden-fallow cycle (e.g., Walker 1999), none has exam ined edge effects of the agricultural clearings. Rather, focus has been on the space corresponding to th e confines of cleared patches themselves. Ecol ogical edge research has rested predominantly in the domain of conservation literature, generally examined as a result of longer-term, relatively permanent fragmentation processes (B ierregaard et al.. 1992), not as part of an episodic land use pattern (but see Ochoa-Gaona 2001). Forest age, cultivatio n history and manageme nt (land use, owner preferences) are the major driv ers of the rates and dynamics of ecosystem functioning (Eaton and Lawrence 2006). Here we have complemented the growing literat ure on land cover modification with a subtle but important aspect of swidden la nd use effects and feedbacks. Results supported the hypothesis that milpa clearings create discernible ecologica l edge effects. The milpa indicator variable is consistently th e most important vari able in explaining variance of the combined vegetation indices acro ss all dates (Table 2-3). Coupled with the finding that forest in the milpa buffers was both consistently lower in biomass and more heterogeneous (as indicated by NDVI and thermal ratio indices), this confirms that at this scale of observation, milpa land use creates a discernible edge e ffect in the forest. Forest areas buffering milpas should favor edge species, managed sp ecies and habitat generalists at the expense of mature forest species and understory plant communities. Relative to other factors that create linear land cover ju xtapositions (i.e., roads and the next-nearest clearing), local milpa edges clearly have the dominant impact on local vegetation structure, according to the relative partial eta-squared values (Table 2-3). That is, if the more dist ant edges created by the covariates
34 (other clearings in the forest or roads) affected broader spatial patterns of forest structure, the effect sizes for these variables would have been greater. This result conforms with other findings of scale-dependency in edge effects related to the ecological process in question (Baldi and Kisbenedek 1999, Huhta et al.. 1998). None theless, when these local effects of milpa edges on forest structure are multiplied across th e Yucatan landscape for thousands of milpas this local land use/cover juxtaposition is clearly important in structuring broader regional vegetation characteristics. The spat ially-radiating effects of milpa clearings should be considered in examining forest regeneration in swidden cy cles, as opposed to fo cusing only on clearings themselves. Examining the vegetation indices separately across milpa and forest sites, however, there is an appreciable discrepancy in the efficacy of the milpa indicator and covariates to predict mean vegetation heterogeneity relative to mean biomass, as indicate d by the adjusted R2 values (Table 2-4). Clearly, there are substantial drivers of spatial pa tterns of vegetation heterogeneity that were not considered he re given the relatively low R2 values (other than biophysical gradients which average out by virtue of the sampling sc heme). Since all organisms and ecosystem processes respond to environmental variability, and not to the arbitrary con cept of mean values, exploring which factors do effec tively explain vegetation heteroge neity represents an important area for future research. Lawrence and Foster (2004) f ound that after twenty-five year s of forest regeneration in this region, woody basal area had recovered to 63 % of total mature forest levels and that recovery of total live biomass took even longer. Comparing our stand basal area results with those of Read and Lawrence (2003) suggests that, on average, the successional forest matrix of ejidal land in Peto is at least ten years of age. Given that the cultivat ion history of their study
35 area is not as long as that of Peto and the fact th at Peto is drier than the southern region of the Yucatan Peninsual, the forest may be on average ev en greater than ten years in age. Since we found no significant difference between stand basal area for milpa edge forest and control forest, this suggests either that none of the ejidal ba ckground (control) forest was sufficiently mature to differ from milpa edge forest; or that in terms of this particular measure of vegetation structure, edges have no conse quence on woody biomass in a matrix of secondary, successional forest. In contrast, vegetation biomass was gr eater for background forest than for forest at milpa edges as indicated by both canopy cover and the two remotely-sensed indi ces. The latter are particularly suited to indicate th e photosynthetic capacity of ve getation. Thus, taking all three forms of forest structure measurement together results may indicate that edge effects have greater impact in terms of green, leafy biomass than woody biomass. Leaf litter contributes to the accumulation of organic matter in the soil, as well as nitrogen concentrations (Lawrence and Foster 2002). Thus, perhaps after some lag in time, even within a matrix of successional forest as in this landscape, edges may feedback to influence woody basal area through changes in nutrient cycling. The total number of milpa s has increased within Peto municipality over the three dates supporting the notion that efforts to moderni ze the swidden landscape may potentially weaken the communal land tenure institution (Klepeis and Vance 2003), which has traditionally limited the number of fields cultivated through non-divisi ble inheritance rights to ejidal land. While none of the farmers interviewed had privatized th eir land, several alluded to an existing informal land market and expressed interest in eventually pursuing the multi-step privatization process if their ejido elects; further south on the pensinsula, Abizaid and Coomes (2004) found evidence of changes in land use made in anticipation of pr ivitization. Increasing numbers of land users
36 and/or decreased fallow times appear to uni ntentionally accompany efforts to liberalize the regions agriculture (Klepeis and Vance 2003). This trend suggests a loss of agronomic services that may constrain future agricultural productivity without substantial chemical inputs, given that key aspects of forest productivity decrease significantly after two clearing cy cles (Lawrence and Foster 2004), let alone promotion of permanent cu ltivation in order to meet the government-set goal of a surplus maize (Chowdhury and Turner 2006). Compared with reported values in the literature for mean milpa area (Klepeis et al.. 2004, Vogeler 1976, Vogeler 1970), field size in Peto is smaller than average (roughly 1.5 ha in Peto compared with 4 ha). Field size in Peto is, on average, smaller than those further south on the peninsula because there, agro -forestry ejidos created in the 1960s were quite large by comparison. This smaller size and the static mean milpa area across dates for our Peto study region also may reflect both biophysi cal and social constraints to the expansion of average field size. Firstly, the cockpit karst topography, pa rticularly in the sout hwest region of the study area (Puuc Hills), affects site se lection for agriculture, as it has in this landscape for millennia (Killion et al.. 1989). Field size is necessarily limited by the area of planos or flat land suitable for cultivation. Secondly, farmers indicated that they generally work their fields alone, such that labor may constr ain the possible extent of milpas a condition confounded by outmigration of youth toward the peninsulas urban areas and coastal fringe s (Lutz et al.. 2000). Results from the analysis of milpa patch metrics suggest subtle but important changes in the spatial characteristics of the milpa /forest landscape mosaic over the course of the study. The static mean of the ratio of perimeter to area (i.e., [ milpa P/ milpa A]/n, a simple, patch-based measure of shape complexity) across the three da tes contrasts with the landscape-level trend where the average amount of edge forest created per average field size (i.e.,, [ milpa P]/n : [
37 milpa A]/n) has increased. Since mean fi eld size is constant across date s, this indicates an increased variability of milpa perimeter. This underscores the impor tance of consideri ng both patch-level and landscape-level means when analyzing spatial la nd-use patterns. Further research is needed to determine if this trend mi rrors natural variab ility driven by biophysical contours (e.g. topography) in the landscape or whether it repres ents an increase in forest-edge juxtaposition driven by policy-related land-use ch oices and land-use path dependency, or another trend such as increased milpa activity along linear featur es like roads. Lawrence and Foster (2004) found past land use to be the most important factor determining forest processes and character istics, which may provide a reason why the importance of milpa edges in explaining vegetation structure has decreased over time (Table 2-3; decrease in p 2 across dates). That is, edge effects from milpa clearings may be more significant when land use intensity is relatively low at th e landscape level when edges are created in a matrix of mature forest such that the contrast be tween edges and the forest matrix is great. Over time, as milpa s rotate through the landscape and the matrix is transformed to a mosaic of successional forest stages, ecological edges may not be as important in driving patterns of vegetation structure relative to past milpa -fallow cycles that occurred in a particular location (Eaton and Lawrence 2006). Future research which maps milpa land use cycles as they occur (in contrast to this retrospective remote-sensing approach) will help discern the dynamic relative influence of milpa edges in determining forest vegetation characteristics. Conclusions In this region, past land use is a critical determinant of th e rate and nature of forest regeneration, yet to date, most re search has examined only forest cl earings themselves. With this study, we have complemented existing research with a subtle but important aspect of milpa land
38 use effects and feedbacks by examining the impact of milpa clearings on the structure of surrounding forest. Results supported our hypothesis that th e footprint of milpa land use is broader in geographic extent than the patch of cleared land itself with associated constraints in agricultural productivity potentially transf erring to the forest region bordering each milpa site. The milpa edge indicator variable was the most impor tant factor in explai ning variance in the examined measures of forest structure, though it s influence decreased over time and appears to also be spatially scale-dependent. While mean values of various measures of forest structure were well explained by the models, factors impor tant for explaining vegetation heterogeneity were clearly omitted and likely include past la nd use. We found that the amount of edges between the forest matrix and milpas has increased over time. Thus, understanding the role of ecological edges within a swidden landscape will contribute to sensible policy formulation and management within the matrix of remaining successional forest. Acknowledgements This research was funded through the Worki ng Forest in the Tropics program at the University of Florida (National Science F oundation DGE-0221599). We would like to thank our colleagues Roger Medina-Gonzalez and Iv an Dominguez-Tec from the Autonomous University of Yucatan (UADY) for their assistance in the field. We are grateful to the residents in Peto who shared their time and knowledge with us about milpa agriculture and land use history.
39 Table 2-1. Mean milpa patch metrics across three years of analysis. Patch Metric 1988 1994 2003 Significance N 1054 1593 1639 n/a Area (ha) 1.38a 1.41 a 1.41 a All p > 0.45 (+ 1.21) (+ 1.02) (+ 1.27) Perimenter (m) 583.95 a 729.11 b 641.48 c All p < 0.00 (+ 323.20) (+ 400.17) (+ 398.77) P/A (m/m2) 0.057 a 0.058 a 0.061 b For a p=0.63 (+ 0.02) (+ 0.01) (+ 0.02) For b p<0.00
40 Table 2-2. Mean forest structure measurements for milpa -edge and control sites. Vegetation Attribute Mean Std. Dev. p Stand Basal Area Milpa edge 25.66 28.25 Control 30.67 16.09 0.137 Canopy Cover Milpa edge 7.1 4.49 Control 37.0 16.05 0.00
41 Table 2-3. Mancova results for three date s indicating the effect size ( p 2) of each covariate and the milpa indicator, along with their significance. Year Variable Wilks p 2 (df1, df2) F p 1988 d. road 0.924 0.077 (4, 1039) 21.4 0.000 d. clearing 0.916 0.084 (4, 1039) 23.7 0.000 milpa indicator 0.297 0.470 (4, 1039) 615.3 0.000 1994 d. road .949 0.069 (4, 1565) 20.90 0.000 d. clearing .964 0.077 (4, 1565) 14.55 0.000 milpa indicator .663 0.300 (4, 1565) 199.06 0.000 2003 d. road 0.927 0.042 (4, 1608) 31.87 0.000 d. clearing 0.914 0.030 (4, 1608) 37.84 0.000 milpa indicator 0.513 0.193 (4, 1608) 380.95 0.000 p 2 = SSeffect/[SSeffect + SSerror] (Norusis 1990)
42 Table 2-4. ANCOVA/ANOVA results for the three dates in the an alysis. Bonferroni-adjusted significance is indicated for the milpa indicator for each vegetation index. Year Veg. Index Milpa Edge Control df F P* Adjusted R2 1988 NDVI x 0.535 2.124 1 1569.91 0.000 0.711 (+ 0.591) (+ 0.446) BBT/R+MIRx 0.983 2.109 1 1183.32 0.000 0.697 (+ 0.329) (+ 0.485) NDVI s2 0.424 -0.512 1 337.36 0.000 0.343 (+ 0.882) (+ 0.351) BBT/R+MIR s2 1.204 0.052 1 156.10 0.000 0.167 (+ 1.386) (+ 1.211) 1994 NDVI x -0.250 0.409 1 769.28 0.000 0.482 (+ 0.394) (+ 0.338) BBT/R+MIR x -1.099 -0.920 1 111.47 0.000 0.217 (+ 0.069) (+ 0.284) NDVI s2 -0.270 -0.534 1 149.33 0.000 0.130 (+ 0.451) (+ 0.188) BBT/R+MIR s2 -0.391 0.318 1 86.22 0.000 0.139 (+ 0.849) (+ 1.130) 2003 NDVI x -0.913 0.312 1 853.05 0.000 0.530 (+ 0.751) (+ 0.480) BBT/R+MIR x 0.053 0.469 1 1127.58 0.000 0.632 (0.192) (0.180) NDVI s2 0.635 -0.408 1 207.81 0.000 0.200 (1.425) (0.563) BBT/R+MIR s2 -0.123 -0.502 1 340.27 0.000 0.289 (0.346) (0.269) Bonferroni adjustments for multiple comparisons made to significance interpretations (p must be < 0.006 for significance)
43 Figure 2-1 Interactions and feedbacks between land use ( milpa agriculture) and land cover properties. Biophysical Effects Subsequent LU Effects Firewood collection Firewood harvesting Fruit tree cultivation Increased net radiation Altered albedo surface energy balance floristic composition plant & animal dispersal soil properties Land Use Action (milpa clearing) Feedbacks Biophysical Effects Subsequent LU Effects Firewood collection Firewood harvesting Fruit tree cultivation Increased net radiation Altered albedo surface energy balance floristic composition plant & animal dispersal soil properties Land Use Action (milpa clearing) Feedbacks
44 Figure 2-2. Map of Peto munici pality study region in the southe rn, central region of Yucatan, Mexico.Peto study area is shown as a 5,4,3 (a s R,G, B) composite subset from an April 1988 Landsat 5 Image.
45 Figure 2-3. Examples of intermediate remote sensing products used to isolate clearing edges from 2003.
46 Forest, Clearings & Milpa Milpa Polygons Buffered Milpa Edges Extraction of Indices: NDVI x NDVI s2BBT/R+MIR x BBT/R+MIR s2 5 km Randomly-Located Forest Samples 60 m Buffer Milpa Forest Matrix Forest, Clearings & Milpa Milpa Polygons Buffered Milpa Edges Extraction of Indices: NDVI x NDVI s2BBT/R+MIR x BBT/R+MIR s2 5 km Randomly-Located Forest Samples 60 m Buffer Milpa Forest Matrix Figure 2-4. Flowchart illustrating the protocol for extracting vege tation indices for forest region buffering milpas (top) and background forest (bottom) for a small subset of Peto study region (RGB:4,3,2).
47 CHAPTER 3 CONVERSION OR CONSERVATION? UNDERSTANDING WETLAND CHANGE IN NORTHWEST COSTA RICA Summary Wetlands are more threatened than any othe r ecosystem type, with losses exceeding 50% of their original extent worldwide. Despite th e small portion of the Earths surface that they comprise, wetlands contribute signif icantly to global ecosystem servi ces. In this study we tested the hypothesis that the location and rate of change in wetland amount in the Tempisque Basin of northwest Costa Rica is predictable from landscap e setting. Our results de monstrate that a strong potential exists for developing predictive models of wetland conversion based on an understanding of wetland location and surrounding trends of land use. We found that topography was the single most important pr edictor of wetland conversion in this area, entraining other conversion processes, and that sp atial patterns of wetland loss c ould consistently be predicted from landscape-level variables. Areas with high est probabilities of conversion were found in the most accessible, non-protected regions of the land scape. While Palo Verde National Park made a substantial contribution to we tland conservation, our results highlight the dependence of lowerlying protected areas on upland proc esses, adding a little-addressed dimension of complexity to the dialogue about protected area management. Conservation strategi es aimed at reducing wetland loss in tropical habitats will benefit from careful analysis of the dominant land use system(s) at a relatively broad scale, and the s ubsequent development of management and policy responses that take into account dynamic opport unities and constraints in the landscape. This chapter is an authorized re-print: Daniels, Amy E., and G.C. Cumming. 2008. Conservation or conversion? Understanding wetland change in northwest Costa Rica. Ecological Applications 18(1): 49-63.
48 Introduction Wetlands are ecologically diverse habitats th at are integral to a range of ecosystem processes and provide many important ecosystem se rvices. They are also under threat in many parts of the world from convers ion to agricultural and residen tial uses. Some of the main scientific and practical challe nges for wetland conservation incl ude understanding the processes that lead to wetland conversion to other land uses; predicting wh ere and when wetlands are most likely to be lost; and developing ways of reduc ing or mitigating negative anthropogenic impacts on wetlands. In this paper we test the hypothesis that wetland c onversion is predictable from landscape location and develop a qua ntitative approach to predicting wetland losses. Our results suggest a range of management and policy actions that could reduce rate s of wetland conversion in our study site. A disproportionate percentage of threatened plants and animals are wetland obligates (Boylan and McLean 1997); although wetlands comprise less than 3% of the Earths surface, they may contribute up to 40% of the globes ec osystem services on an annual basis (Zedler and Kercher 2005). The many benefits provided by wetlands include flood protection (Hey and Philippi 1995), water quality enhancement (Jeng and Hong 2005), carbon storageif managed appropriately (Mitra et al. 2005), and nutrient cyc ling, both internally and at the upland interface (Bunn et al. 1999). Wetlands also play a critical role in the land-su rface energy balance (Meijerink et al. 2005), provide ha bitat for many species (Trebitz et al. 2005), and are important for hydrologic connectivity at the wate rshed scale and be yond (Pringle 2003). The conversion of land with re latively natural vegetation patterns to more intensive, direct human use (e.g. monocrop agriculture) is a major cause of global declines in biodiversity in recent decades (Forester and Machlis 1996, White et al. 1997, Cincotta et al. 2000). Wetlands are more threatened than any other ecosystem type (Millennium Ecosystem Assessment 2005),
49 with wetland losses exceeding 50% of the original modern-era global extent (Mitsch 2005). Both the ecological contributions of wetlands and the many threats to their persistence are best understood from a landscape perspective (Mits ch and Gosselink 2000). Wetland processes (e.g., floodwater retention and mercury methylation) and character (e.g., riparian, or coastal) vary across space and time (Haig et al 1998, Ogawa and Male 1986). Spatial and temporal context also influence anthropogenic activities in the la ndscape. Direct wetland loss often occurs in flatter areas through conversion of wetlands to ot her land uses, primarily agriculture and urban areas (Zedler and Kercher 2005). Indirect losses occur as a consequence of such things as climate change (Klein et al. 2005); water extraction (Lemly et al. 2000) and other forms of hydrologic alteration (Liu and Cameron 2001); ad jacent and upstream changes in land use (Pringle 2003); and changes in populations of wetland-depende nt wildlife (Haig et al. 1998, Semlitsch 2002). Tropical wetlands are particularly at risk from the increasing human demand for fresh water and land, dependence upon hydroelect ric energy for development, inadequate wastewater treatment, and frequent inability to craft or apply appropriate protection measures when wetland sites have been neither docum ented nor inventoried (Junk 2002). All Central American nations are signatories of the 1971 Ra msar convention, which created a framework for national and international collaboration in conserving wetlands and promoting the sensible use of wetland resources. However, anth ropogenic pressure on wetlands in the region is only expected to increase in the coming years (Ellison 2004, Junk 2002). Land cover changes in many tropical areas have been characterized and quantified extensively over the last severa l decades (Lambin and Geist 2001), and spatially-explicit models of deforestation processes have been deve loped for many regions (Irwin and Geoghegan 2001). Changes in spatial patterns of wetland mosaics resulting from wetland loss have been quantified
50 for the northeastern U.S. (Gibbs 2000). The am ount of wetland loss has been predicted for coastal Louisiana (Cowan and Turner 1988) and the U.S. Atlantic coastal plain (Koneff and Royle 2004). Yet there are few generalized wetla nd conversion prediction m odels akin to those for deforestation (but see Reyes et al. 2000).. Landscape conser vation and planning efforts are increasingly recognizing the importance of geographic settin g in defining and prioritizing wetland functions and in unders tanding how wetlands are infl uenced, both directly and indirectly, by the socio-economic drivers of landscape change (Haig et al. 1998). This is particularly true for tropical regi ons that lack long-term, spatia lly-explicit data and have been dominated by conservation efforts that fo cus predominantly on tropical forests. Our primary hypothesis in undertaking this study was that the processe s that drive wetland conversion are broad-scale and socio-economic in nature, meaning that wetland conversion should therefore be predictable from the locati on of wetlands along releva nt gradients in the broader landscape. Alternative hypotheses in clude the suggestions that wetland conversion occurs randomly; that wetland conversion is driv en primarily by local (s mall-scale) processes, and hence is not tightly aligne d with broader gradients in the landscape; and that wetland conversion is driven by such a complex, multi-laye red set of processes th at the prediction of wetland loss from a landscape settin g is essentially impossible. Our objectives in this study also included the development of a potentially gene ralizable modeling approach for predicting wetland conversion, and the formulation of ma nagement and policy recommendations for our study area, the Tempisque Basin of northwestern Costa Rica. The results suggest that although a diversity of processes contribu te to wetland conversion, the singl e most important driver of wetland conversion in our study site is topograp hy, which entrains a range of other landscapelevel processes. Our results support the hypothe sis that wetland conversion can be understood
51 through analysis of landscape context; provide a detailed example of one way in which the likelihood of wetland conversion in tropical landscapes can be predicted; and shed light on relevant management and policy concerns in our study region. Methods Study Region The Tempisque River Watershed is a 5,404 km2 basin in the province of Guanacaste, in northwestern Costa Rica (Figure 3-1). The Tempis que River originates in the northeastern area of the watershed in the Guanacaste Mountain Range which runs from nort hwest to southeast and defines the eastern border of the Tempisque basi n. The region experiences one wet and one dry season per year and has a mean temperature of 27.5oC. The mean annua l precipitation of 1817 mm falls mainly between May and November (Mateo-Varga 2001), though a break in the rainy season ( el veranillo or little summer) corresponds roughly to the month of July. Annual and inter-annual variation in rainfall affects the character and extent of active wetlands at any given time. The Tempisque Basins intricate mosaic of diverse wetlands and adjacent uplands, including wet meadows, open-water lagoons, flooded forests and riparian mangroves, provides critical habitat for resident and migratory wate r birds traveling along or over-wintering on the Central American isthmus (McCoy and Rodriguez 1994). Counts of more than 50,000 waterfowl have been made in the wetlands of Palo Verde National Park, including the endangered Jabiru stork ( Jabiru mycteria ). The economy of the Tempisque region has long been based on extensive cattle ranching, and many of the formerly forested upland areas ha ve been converted into cattle pasture (Peters 2001). Traditionally, the wetland areas of the lower basin were used for grazing during the dry season since they remained green much longer in to the dry months than upland pastures. The crash of the beef market in the 1980s, the rise of tourism, intern ational lender-forced structural
52 adjustments, and the implementation of a regi on-wide hydroelectric/ir rigation project have contributed to the transformation of land use in the Tempisque Watershed from an extensive, cattle-ranching system to a more diverse and mo re intensive agrarian system (Daniels 2004). With the implementation of irrigation (the Arenal -Tempisque Irrigation Project or PRAT as it is known by its Spanish acronym), commercial, cropbased production has become more important than cattle ranching in the lower basin. Dependi ng on the season, PRAT injects a volume of 5085 m3/s of water into the terres trial landscape year-round. Floode d rice, sugar cane, and melon cultivation are increasingly prominent features of the lower watershed and often occur at the expense of wetlands. Ensuing alterations in hydrologic connectivity, although not measured directly in this study, also c ontribute to changes in wetland lo cations and configurations. Roughly a third of the 18,760 ha Palo Verd e National Park (situated in the lower Tempisque watershed) is wetland. The park and adjacent, non-protected wetlands were declared as Ramsar sites in 1991 (Mateo-Varga 2001). We tlands and all surface waters are held in the public domain by the Costa Rican government and fall under a special category of protection. In practice, however, defining and delineating wetl ands has long challenged the Ministry of Environment since most wetland definitions ar e suitable for temperate latitudes but not necessarily applicable in tropi cal systems (Ellison 2004). For this research, protected wetlands were considered to be only those that fell within the bounds of the national park. Overview of Methods and Modeling We derived binary land cover maps from thr ee Landsat images by identifiying wetland and non-wetland classes at the pixel level. The land cover history of each pixel was traced over the three dates and compiled into a single data laye r of wetland trajectorie s. This dichotomous trajectory map represented whether each pixel ha d been conserved or converted by the final date in the image series. We then modeled this binary response variable with multiple logistic
53 regression models, using orthogona l predictors obtained via Pr inciplal Components Analysis (PCA), to explore the relevance of differe nt variables in driving landscape change. Land Cover Data and Predictor Variables We used the Ramsar definition of wetlands1 in this research, with the interpretation that flooded rice fields were not considered wetlands for the purposes of this analysis. This is also the interpretation used by the Costa Rican government. Binary land cover maps (wetland/nonwetland) were produced from a three-date, early dry-season Landsat image classification series (1975, 1987 and 2000) using a rule-based classificati on technique (path 16, row 53 for all dates; also path 17, row 53 for 1975 since watershed wa s split across multiple scenes by WRS1). The dates of the images capture b aseline land cover (1975) prior to the establishment of the National Park or the implementation of the ir rigation program, an image date after their implementation (2000), and one mid-point image (1987). Radiometric data were converted from radiance to surface reflectance (Jensen 2000) by correcting for sensor gain, atmospheri c distortion, and differences caused by nonanniversary dates. Geometric correction fo r the 2000 image was performed with sixty ground control points taken with a handheld GPS. The root mean squared error (RMSE) error achieved from a first order geometric transformation wa s 0.4880 pixels, or less than 15 m. The other images were then co-registered to the 2000, also achieving an RMSE less than 0.5. Pixel size for the 1975 MSS image mosaic was resampled to 30 x 30 m to match spatial resolution from later image dates. 1 Wetlands are areas of marsh, fen, peatland or wa ter whether natural or artificial, permanent or temporary, with water that is static or flowing, fr esh, brackish or salt, including areas of marine water the depth of which at low tide does not exceed six meters.
54 Since statistical clustering of spectral data alone has often proved ineffective for accurately classifying wetlands (Ozesmi and Bauer 2002), we employed a rule-based procedure that incorporated domain knowledge, spatial re lationships, and two distinct algorithms on spectral data (for normal and non-normal distribut ions of land cover classes) to improve the accuracy of the resulting land cove r maps (Figure 3-2). A descrip tive example of a rule used is as follows: Classify as wetland all pixels identified as wetland by the parallelepiped algorithm if they occur at < 60m elevation and do not fall within ground-mapped polygons of agricultural fields where crops are cultivated. This particular rule helped compensate for the spectral confusion of wetlands and rice fields by exploiting location information ma pped during fieldwork. Rules were applied systematically across dates so as not to bias classification or the modeling of trajectories, adjusting for differences in spatial resolution and inter-annual climate variation accordingly. Accuracy was assessed separately for each of the three dates using over 90 independent reference points for each. For 2000, reference data were co llected via field traini ng samples, while for 1987 and 1975, they were derived from higher resolution aerial photomosaics (1:35,000 and 1:20,000, respectively). While the data-collection and processing time for this classification method were costly compared with traditional remote sensing on spectral reflectance data alone, the results justify the expenditure. Accuracy of wetland classification exceeded 90% across all dates (class-wise Kappa statistics were also > 0.90). For more details on this rule-based classification, see Daniels (2006). We traced the wetland status (wetland or non-wetland) of each pixel across the three classification dates and compiled the result in to a single data layer of dichotomous wetland trajectories representing whether each pixel was conserved or converted by 2000 (Table 31). To simplify the modeling process, we considered only one-way conversions from wetland
55 land cover to other land cover types, excluding the poten tial for the reverse trend. The same simplification is typically employed for spatially -explicit deforestation modeling (Mertens and Lambin 2000). The potential for misclassificatio n of dry versus converted wetlands was reduced by a number of steps and checks, specifically (1 ) careful standardizati on and recalibration of images to compensate for differences in prio r rainfall (Danie ls 2006); (2) inclusion of a midpoint image to provide another re ference point, while increasing th e probability of unidirectional change and reducing error propaga tion across trajectories; (3) test ing for a relationship between wetland size and conversion probabil ity (this was weak); and (4) chec king for spatial pattern in pixels that were apparently converted back to wetland (this was lacking). In a basin-wide descriptive analysis of land cover change Daniels (2004) found that the major drivers of wetland conversion in the Temp isque Basin were related to agricultural intensification in the lower watershed and alterations in hydrologic connectivity and hydroperiods. To assess the rela tive importance of drivers of we tland trajectories related to these processes, a number of predictor variables and interaction terms we re considered for inclusion in the predictive model (Table 3-2). These data were compiled in a multi-layer, raster geographic information system (GIS) for the Tempisque Watershed with a spatial resolution of 30 m2. Predictors included x/y coordina tes, elevation, slope, a protect ed area indicator surface, wetland patch area, distance to nearest road, distance to the Panamerican Highway, distance to the nearest population center, and distance to the Tempisque River (features ar e mapped in Figure 3-3). The use of distance surfaces as predictors has severa l distinct advantages. First, it allows for a spatially explicit modeling fram ework and enhances our abilit y to understand the role of landscape setting in wetland convers ion. Also, this approach allows for accurate representation of important but temporally dynamic variables (e. g. distance to nearest population center remains
56 the same while the population itself may change). The rationale for testing each variable and a host of interaction terms (Table 3-2) was based on a qualitativ e understanding of changing land use systems in the Tempique Basin. Statistical Model Building For each raster cell (i.e., sample unit) in the GIS, the dependent variable (wetland trajectory) and its corresponding values for each of the ten proposed predictor variables were extracted from the GIS to a database (N = 223,689). Fourteen inte raction terms among predictors were computed and added to the suite of independent variables. We then generated random subsets for model building and independent model validation using the formula ntest/ntrain = [1 + (p-1)1/2]-1, where p, the number of predicto r variables (Schaafsma and van Vark 1979). With p equal to four for the final model, this determined the ratio of training (n = 141,813) to testing (n = 81,876) cases in our split sample approach to model validation (Ozesmi et al. 2006). A rigorous exploratory analysis was performed to better discern relationships am ong potential predictor variables and complement our ground-based und erstanding of landscape dynamics (see Daniels 2004). Potential predictors with non-significant (p>0.05) or w eak correlations (rho<0.25) were eliminated from further consideration. A series of models was developed by combining conceptual models of wetland c onversion processes, findings from the exploratory analysis, and the principle of parsimony. We compared the Ak aike Information Criter ion (AIC) across models to determine the most effective suite of predictor variables. This is computed using each models log likelihood ratio: AIC = -2(log likelihood ratio) + 2k where k represents the number of model paramete rs estimated. AIC indicates the loss of information incurred for incorrect mode l specification (Hoe ting et al. 2006)..
57 For the best models, we transformed the suite of landscape (predictor) variables along orthogonal axes via principal components analysis (PCA) to eliminate multicollinearity, which can substantially confuse interp retation of model results (Leg endre & Legendre 1998). Varimax rotation was employed to increase interpretabi lity of resulting components by enhancing the distinction between variables that significantl y load on each component and those that do not, while maintaining the cumulative variance of the struct ure matrix (McGarigal et al. 2000). PCA has been widely used with multiple regressions in spatial ecological analyses for many years. For recent examples from several different contexts see Chang et al. (2006) ; Summerville et al. (2006); Krueger et al. (2006); a nd Corbett & Anderson (2006). In ecological modeling there are well-documented tradeoffs between prediction accuracy, reliable coefficient estimati on, understanding driving mechanisms and ecological interpretation (Graham 2003). One of the central weakne sses of PCA is that components are often uninterpretable. The more variab les that are included, the less interpretable the model becomes and the more likely it is to produce statisti cally significant results that are ecologically meaningless. For this reason, we performed an initial round of variab le selection for model building as described above prior to PCA. Our goal was to not only make accurate predictions, but to understand dynamics in this wetland landscap e through interpretati on of the model. Two predictors were withheld from the PC As and added separately to the models: protected area status (falling with in PVNP or not) and wetland pa tch size. The other variables group naturally along major biophysical and soci al landscape gradie nts; and given the importance of protected areas in the conservatio n of natural habitat (Sanchez-Azofeifa et al. 2003) and the role of patch size in landscape-level ecological pr ocesses (Flather and Bevers 2002), we wanted to estimate model coefficients for these critical predictors directly. Finally,
58 variance inflation factors (VIF) indicated that multicollinearity among these two variables and the principal components used as predictors wa s not an issue (VIF < 1.2 and p < 0.001). Three prediction models with slightly different combinations of predictor variables were compared using the AIC. Jackknifing (n = 141,813 leave-on e-out iterations; see Ef ron and Tibshirani 1993 for discussion of jacknifing approach) indicated that the model was stable and all independent variables were unbiased to less than 0.001 for a ll variables. Model Validation and Performance Assessment Model validation was conducted with indepe ndent dataset (n = 81,876) that were set aside prior to building the logis tic regression model, as describe d above. The logit, predicted probability of conversion and predicted trajectory (conservation or conversion) were obtained by evaluating the final prediction equation for each independent observation. Nagelkerkes R2 values along with area under the curve (AUC) for receiver operating ch aracteristic (ROC) plots were used to assess model predicti on performance. Nagelkerkes R2 ranges from zero to one; it is considered a pseudoR2, however, in that the percent of variance explained for a dichotomous dependent variable depends on its frequency distribution (Nagelkerk e 1991). The AUC provides a threshold-independent measur e of model performance (Cumming 2000) by assessing model function independe nt of the probability cut-off used for defining the conversion trajectory. This is done by balancing the opposing goa ls of model sensitivity (the probability of predicting conversion when, in fact, conversion has occurred), with m odel specificity (the probability of predicting conservation when, in fact, conservation has occurred). AUC is calculated by a calculus-based trapezoidal rule and ranges between 0 and 1. An effective model would demonstrate a rapid rise in sensitivity (y-axi s) with little increase in the probability of producing a false positive (x-axis), having an AUC that approaches 1. By contrast, a random model would display a curve of a diagonal line be tween the origin and (1,1), with an AUC of 0.5.
59 When evaluating the model to display its prediction results in geographic space, we selected the threshold probabi lity corresponding to the highe st sensitivity and lowest complement of specificity (i.e., one minus speci ficity, the probability of falsely predicting conversion). This was 0.6, meaning that any pixel with a probability of conversion greater than 60% was mapped as a conversion. Model coefficients from the logistic regr ession equation, based on orthogonal variables, were interpreted to determine the relative im portance of independent variables and landscape gradients in predicting wetland conversion. We ev aluated the model on a pixel-by-pixel basis in order to create a spatially-explicit map of wetland conversion probability. A thorough exploration of mapped results was performed dig itally to identify salient spatial trends of wetland conversion probabilities. We also mapped re siduals and predicted trajectories from the model, along with error rates, to better discern any spatia l typologies of wetland conversion and to assess the validity of the model in geographic space (Rogerson 2001). Results Model Selection The predictor variables for th e best models included PC1, PC 2 and PC3, tenure status and, for one of the models, patch size (see Appendix A for details). AIC and R2 values for the model that included patch size were better (AIC = 119,819 and R2= 0.48) than for the model where patch size was excluded (AIC = 125,403 and R2= 0.44) (Table 3-5). Residuals became less variable and model fit improved as patch size incr eased. In contrast, howe ver, AUC was slightly higher when patch size was excluded (0.81 versus 0.79); and the model coefficient for patch size (-0.00) demonstrated that virtually no relationship existed with the dependent variable (Table 35). The observed trend for the residuals appear ed related to the fact that the number of
60 observations decreased as patch size increased (C umming 2000). After taki ng these issues into account, wetland patch size was excluded as a predictor in the final model. Principal Components Analysis The final composite predictor variables were co mprised of these variables: y coordinate, elevation, slope, distance to river, distance to nearest road, distance to population center and distance to the Panamerican Highway. PCA with varimax rotation reduced the dimensionality of these seven predictors into three orthogonal components with eigenvalues greater than one, which explained over 73% of the original varian ce (Table 3-3). Communalities were relatively high, except for distance-to-nearest-road (0.53) confirming that PCA effectively reduced dimensionality while explaining most of the varian ce in the original variables (Table 3-4). The rotated structure matrix (Table 3-4) s hows that distance to nearest road (0.70) and distance to population center ( 0.90) loaded highly onto the firs t component. This component represented the socioeconomic infrastructure gradient of the Tempisque landscape (PC1). Positive signs for both of these loadings indi cated that large distances from towns were associated with the same end of the component as large distances from roads, since population centers in the Tempisque Ba sin are connected through a tr ansportation network. The ycoordinate also loaded substa ntially, but negatively, on this component (-0.52) meaning that wetland sites become increasi ngly isolated (less socio-ec onomically connected) moving downstream (i.e., further south) in the basin. On the second com ponent (PC2), the y coordinate, distance to river, and distance to the Panameri can Highway loaded highly. The loading for y coordinate was negative (-0.78) wh ile loadings for distance to the Panamerican (0.89) and river (0.57) were positive. The distance from the high way and river increases moving south in the basin, reflecting the watersheds triangular, geomorphically-defined shape. This second component was called the north-south gradient in the landscape (PC2). Slope (0.79) and
61 elevation (0.86) loaded highly on the third component, representi ng the topographic gradient of the landscape (PC3). Spatial Predictions The probability of wetland conve rsion is mapped in Figure 3-4a. In general, the probability of conversion increased moving away from the central, lower watershed. Wetlands near the rivers discharge point within the protected area, ha d the lowest probabilities of conversion (p < 0.20) in the enti re basin. Areas along the river channel in the lower watershed also had low probabilities of wetland conversi on (around 0.25), even on the non-protected west bank of the river. At the outer reaches of the basins wetland network, among the highest and driest wetlands in the landscape, pixels had c onsistently high probabilit ies of conversion (p > 0.60). For these sites, highest conversion probabi lities tended to occur at the edges of larger wetland patches (p >0.85), with probability decreasing toward the interior of such patches (p<.60). Localized non-protected wetland patch ne tworks with high probab ilities of conversion (p>0.80) occured immediately adjacent to the nationa l park within large, corporation farms that cultivate sugar cane, cantaloupe and rice. Ar eas within the park ha d significantly lower probabilities of conversion on average, yet there we re three distinct regi ons in the park where wetlands had very high probabi lities of conversion (p>0.80). Model Performance The final model had an AUC of 0.81. This indi cates that there was an 81% chance that the conversion probability for a randomly-selecte d observation belonging to the conversion trajectory group, would be greater than that of a randomly select ed observation belonging to the conservation group (Fielding a nd Bell 1997). Figure 3-4 b illustrates wher e sites of both conversion and conservation were correctly predicted, along with errors. False negative cells (i.e., prediction of wetland conservation when conversion actually occurred) were concentrated
62 in the lower, central watershed in and around PVNP False positives (i.e., prediction of wetland conversion when the site was conserved) were clustered. Model Coefficients All predictor variab les were statistically si gnificant at p<0.001 (Table 3-5). The order of influence of landscape drivers on wetland loss dem onstrated that the topo graphic gradient (PC3) was the most important correlate of wetland c onversion, followed by socioeconomic connectivity (PC1), land tenure, and fina lly, north-south gradient (PC2). Model coefficients ( Bi) are in logodds units and so we interpreted the exponentiation of the coefficients (eB). The magnitude of the model coefficient for PC3 was 3.10 (Table 35). As PC3 scores increased, degree of landscape reliefelevation and/or slopeincreas ed. Every unit increa se in the topographic gradient raised the likelihood of wetland c onversion more than twenty-two times (eB =22.11). The coefficient of PC1 was -1.03 (Table 3-5). As PC1 scores increased, distance from built landscape features like p opulation centers and roads also in creased and y-coordinate values decreased (i.e., move southward). The negative sign on the PC1 coefficient indicated that the likelihood of conversion decreased as PC1 scores increased. For every unit increase of the socioeconomic infrastructure gradient (i.e., incr eased isolation), the likelihood of conversion was about one-third (eB = 0.36) as great. Results also demonstrated that tenure status affected the pr obability of wetland conversion ( B = -0.91). The likelihood of conve rsion decreased in the protected area; in fact conversion was less than half as likely (eB = 0.40) inside the park as in the non-protected sites, controlling for other relevant drivers of convers ion and variation in th e landscape. The fi nal predictor in the model was PC2 ( B = -0.82). As PC2 scores increased, y-co ordinate values decreased. Since the north-south gradients coeffici ent was negative, the likelihood of wetland conversion decreased moving south in the basin.
63 Discussion Drivers of Wetland Change The results provide clear support for the hypothe sis that wetland conversion is predictable from landscape setting in the Tempisque Watershe d. The importance of landscape setting in this instance provides valuable insi ghts into the process of wetla nd conversion in Costa Rica. Landscape variables proved to be strong and sign ificant correlate s of changes in wetland amount and location. The relative order of influence of the drivers of conversion from highest to lowest was topography, socioeconomic connectivity, land tenure, and north-sou th gradient respectively. This order suggests a hierarchic al structure of influences in which the physical landscape (topography) may provide the ultimate context for landscape processes that are more proximate in nature, driving conversion at finer spatial and temporal scales In other words, topography defines higher, dryer wetland areas that are more likely to be settled and converted to human land use relative to lower-lying, boggy areas. The physical landscape consequently provides a constraining influencein a hierarchical contexton socioe conomic drivers of conversion (PC1). This result makes sense both for economic efficiency and security of investment in agricultural production. It may also indicate increased likelihood of indirect conversion related to urbanization such as altered hydroperiods, changes in surface runoff patterns, and changes in water quality. As might be expected, isolated area s in the lower basin had a lower probability of conversion than sites with great er economic connectedness or proxi mity to infrastructure like towns and roads. Although tenure affected wetland conversion prob ability, given that there are no occupants or direct agricultural threats in side the park, the fact that co nversion was no less than half as likely to occur inside the park underscores the cr itical, indirect role of intraand inter-annual hydrologic alterations and cumulativ e watershed effects. While converted wetlands inside the
64 park are not precluded from being wetlands again in the future, the model illustrates sites that are presumably most sensitive to changes in hydrol ogic connectivity. The north-south gradient (PC2), may reflect the underlying influence of geomorphic configuration, thereby supporting the idea that shape and directional gradients in a watershed influence the spatial organization of natural and anthropogenic processes. These, in turn, affect location of wetlands and threats of their conversion (Huston 2005). Assessment of Model Our model of wetland conversion provides a rela tively strong fit to th e data (AUC = 0.81) by comparison to other published models. Pont ius and Schneider (2001) used two distinct methods for predicting deforestation in a Massac husetts watershed and obtained AUC values of 0.65 and 0.70, respectively. Other spatially-explic it models of land cover conversion report accuracies ranging from sixty-five to 90% (Brown et al. 2002, Pijanowski et al. 2002, Mas et al. 2004). The nature of focal land cover transitions and differences between studies in modeling techniques and performance measures make the u tility of cross-model comparisons questionable. Nonetheless, our results fall within the upper rang e of conversion predictions for upland habitats. Given that ours may be the first spatially-explic it attempt to model we tland conversion at the landscape scale in the tropics, future models should be able to improve upon this performance. The spatial pattern of pred iction errors points out the weaknesses of the modeling approach that we employed (Figure 3-4 b ). False negatives (i.e., fa ilure to predict conversion when it actually happened) were concentrated in the lower, central watershed in and around PVNP. These conversions were la rgely due to alterations in th e hydrologic regime as little mechanized agriculture or infrastructure develo pment has taken place in these locations of the basin, indicating the models weakness in pred icting indirect wetland conversion related to
65 hydrologic alterations. This shortcoming was not surprising since predictions were made using landscape variables rath er than spatially-explicit hydrologic data. The clustered spatial pattern for false pos itives (i.e., prediction of wetland conversion when the site was actually conserved) suggests th at the watershed-scale model failed to predict correctly in places where finer-scal e processes might have controll ed the pattern of conversion. For example, in the central region of the nor thern border of the pa rk, flooding has been associated with land use conversion to agricultu re and poor irrigation dr ainage. In response, officials identified a buffer zone for which speci al management provisions are hoped to prevent land use conversion and maintain the quality of wetlands inside the park. Urban (2005) points out that ecologists tend to model fine-scale pro cesses that have somewhat limited utility for studies that seek to scale their results up to entire landscapes. In cont rast, our prediction model performed well for the overall la ndscape-level process of conversion but clearly missed some of the finer scale processes occurring locally within the watershed. Perhaps this is related to a spatial filtering or averaging eff ect of using a predefined grain of analysis (i.e., the 30 m pixel of Landsat imagery). The low probability of conversion of both pr otected and non-protecte d wetlands near the basins discharge point is probably related to the fact that these areas are increasingly prone to flooding, as a consequence of upstr eam canalization and channel-wide ning activities. Much of the discharge region is largely inaccessible by land, suggesting th at the significant investment required for conversion to direct human land use has facilitated pe rsistence of even nonprotected wetlands in these isolated, seasonally-boggy and flood-prone areas. Though conversion probabilities inside th e park were low on average, conversion of upstream patches has the potential to further alter hydrologic connectivity, hyd roperiod, and water quality for
66 wetlands inside the park. Of the three distinct regions inside the park with high probabilities of conversion (>0.80), one occured adja cent to a portion of the river that was channelized in the mid-1980s. This is consistent with findings in many other watershed studies where riparian areas are degraded or convert ed by diversion and channelization (e.g., Toner and Keddy 1997, Tiner 2005). A second such region occured around an historically-maintained canal used to bring freshwater from the river into the lagoons of the park when it was a cattle ranch. In recent history, the connectivity of lagoons in PVNP and the river decreased as the channel was filled by sediment deposition (Jimenez et al. 2003). Conservation Implications The spatially-explicit map of conversion pr obabilities (Figure 3-4a ) suggests several patterns that have releva nce to conservation planning. First, remaining wetlands in proximity to agriculture and urban land uses are at extremely high risk of c onversion. Prevention of direct conversion will require active protection and manage ment. Even then, such wetlands may be at risk of degradation and/or indirect conversi on from cumulative watershed impacts (Vorosmarty and Sahagian 2000). Secondly, active maintenance of a buffer region around PVNP may serve to protect wetlands within the park from indirect conversion. As emphasized in Pringle (2001), the parks location at the bottom of the watershe d implies that successf ul conservation of its wetlands will entail an integrated, watershed-le vel approach to managing land-use, nutrient runoff and hydrologic flows. Finally, the typology of wetland conversion in the non-protected areas of the Tempisque landscape appears to be one of sweeping conversion from higher/dryer, economicallyconnected regions to more isolated, boggy sites. The park provides a defensible and successful border against direct human land use, meaning th at only indirect wetland conversion is possible within its bounds. The pattern of conversion inside PVNP is therefore more patchy and less
67 predictable in relation to lands cape-level predictor va riables than for surrounding areas. This difference in typologies is similar to results from deforestation analyses, where different spatial patterns of conversion develop depending upon the dr iving forces and their spatial organization in the landscape (Husson et al. 1995). Me rtens and Lambin (1997) found that dividing landscapes into regions with uniform deforest ation typologies enhanced model specificity. Developing a typology of wetland co nversion processes may help to better predict wetland loss in the future. The development of such a typology, however, is confounded by the inherently patchy nature of wetland occurr ence in the landscape. Whereas deforestation entails any of various patterns cutting into la rge blocks of forest, even sweeping wetland loss may appear patchy given that wetlands are dispersed and co mprise a small proportion of the surface area of any region. The spatial nature of this model aids in pointing out the need to carefully consider dominant land use systems in relation to opportun ities and constraints presented by the physical landscape, which may change over time. Extens ive cattle-ranching, the dominant economic land use in the Tempisque Basin up to the late 1970s, re sulted in deforestation of uplands, occurring at a wide range of elevations and even on sloped lands. Agricultu ral intensification, in contrast, has in part facilitated reforestation of uplands at the expense of land us e conversion in flat, lowlying areas (i.e., wetlands). Th e conservation mechanisms and polic ies that are most appropriate for the respective land use regi mes are complementary, but w ould obviously differ in their strategy and budgetary foci (e.g. up land forest reserves and ripari an buffer restoration versus wetland network preservation and monitoring of in-stream nutrien t load). Our results support Hustons (2005) proposed theory of land use change, where spatio -temporal patterns of land use
68 are linked to net primary productivity in diffe rent ways depending upon the phase of economic development and its constituent patterns of resource dependence. Though threats of direct human land use within PVNP are not an issue in this landscape, indirect conversion of wetlands (i.e., through altered hydrology) was found to represent a threat to protected wetlands; and one that was not as readily predicted by landscape variables. Our results agree with those of Cowan and Turner (1988) for Louisianas coastal region in that indirect impacts on wetlands are of ten as important as direct conve rsions. This finding adds a dimension of complexity to the dialogue about protected area management that is little addressed (but see Pringle 2001). Most discussions on the efficacy of protected areas center around the degree to which park boundaries are defensible given natural resource dependence among people living in and around parks, particularly in th e tropics (Brandon et al. 1998, Bruner et al. 2001, Terborgh 1999). The degree of land-surface co nnectivity between parks and their surrounding landscapes has been quantified and described in many case studies (Mas 2005, Ravan et al. 2005). Defries et al. (2005) analyzed 198 protected areas around the world and found a substantial increase in park isolatio n over the last twenty years. Model results confirmed that while surface connec tivity appears to be critical for landscape management and planning considerations, it is only one of many elements of ecological connectivity for which protected areas shoul d be monitored and managed. Assessing and monitoring hydrologic connectivity a nd changes therein will require an unders tanding of how all land cover and resource use at the waters hed level are connected on the surface (two dimensional), through hydrologic flows (three dime nsional), and how connectivity varies over time (four dimensional). Clearly these interli nked processes, and the landscape position of protected areas in relation to th em (Harris et al. 2005), must be considered carefully in the
69 context of integrated watershed management non-linear pattern-process relationships, and critical thresholds in drivers and responses (Bedford and Preston 1988). While this approach considers only wetland conversion (as opposed to degradation), the results may be used to focus finer analyses of wetland conditions. Future wo rk may also explore temporal aspects of wetland conversion in greater detail using a multinomial spatiotemporal model. Conclusions Model results supported our hypothesis that we tland conversion is driven by broad-scale processes related to socio-economic trends of land use. Conversion was effectively predicted from wetland location along relevant social a nd biophysical landscape gradients. This study demonstrates that a strong potential exists for developing a generalizable approach to predicting wetland conversion. By quantitatively explori ng and mapping wetland conversion probabilities, areas of the landscape in which different mechan isms of landscape change are operating can be identified and brought to the at tention of managers and policy makers. While it is encouraging that Palo Verde National Park makes a substan tial contribution to wetla nd conservation, this study also highlights the importance of lands cape context and landscape influences on hydrologic connectivity, along with the dependence of lower-lying areas on their watersheds. Ultimately, wetland conservation and the maintena nce of the many goods and services wetlands provide will have to encompass a regional perspective on ecosystem management and appropriate spatial planning within the mosaic of different land uses that influence wetland persistence. Acknowledgements This research was funded by a Fulbright Fellowship and research grants for the Tinker Foundation and University of Fl oridas Tropical Conservation and Development Program.
70 Thanks go to the Costa Rican Ministry of Environment (MINAE) and the Organization for Tropical Studies (OTS) at Palo Verde for providing some of the GIS data layers for this research.
71 Table 3-1. Construction of wetland traject ories using three binary land cover maps (wetland/non-wetland). Each raster cell in the landscape fell into one of these two trajectories of conversion or conservati on. MSS = Multispectral Sensor, TM = Thematic Mapper, and ETM = Enhanced Thematic Mapper. Date (Landsat platform) 1975 (MSS) 1987 (TM) 2000 (ETM) Trajectory Land Cover Wetland Wetland Wetland Conservation Wetland Wetland Non-wetland Conversion Wetland Non-wetland Non-wetland Conversion
72 Table 3-2. Response variable (w etland trajectory) and proposed pr edictor variables and interaction terms, along with their the oretical justification, for we tland conversion model. Raster Layer Data Source Description Justification Wetland Trajectory This analysis Cells classified as wetlands in 1975 were extracted from land cover classifications for 1975, 1987 and 2000. These cells were coded for either the conservation or conversi on trajectory depending on whether they remained wetlands or converted to another land cover/use. The dependent variable. Tenure Indicator Surface OTS1,2 A raster dummy variable coding for the park (PVNP) and the non-protected parts of the Tempisque Watershed. Protected area establishment is a key st rategy in biodiversity conservation. While many studies have assessed the effi cacy of protected areas in preventing forest conversion, little analogous work exists for wetland protection. Wetland patch size This analysis Cells from the wetland trajectory surface were clumped into patches according to an 8-neighbor rule. Area was calculated for each patch. Patch size suggests something about accessibility and degree to which other landscape variables (e.g. distance to near est road) may affect wetland habitat. Wetland size is related to cost of conversion for agricultural land uses. X Coordinate This analysis The X coordinate for the center point of each 30m2 sample cell. There are several rivers with a primary directional gradient along eastwest axis (e.g. Rio Canas and Bebed ero) whose flows, alterations, seasonal fluctuations affect wetland areas; rivers have been increasingly channelized resulting in the desiccation of many wetland areas in the floodplain Y Coordinate This analysis The Y coordinate for the center point of each 30m2 sample cell.. The Tempisque River has a primary di recti onal gradient running north-south whose flow has been significantly reduced in the last 20 years due to extraction for agricultural uses. The stream bed has also been incised and channelized, cutting off dendritic flow-like patterns in one key wetland area Digital Elevation Model (DEM) OTS1,3 15 m x,y resolution (re-sampled via nearest neighbor algorithm to 30 m resolution) and 1 m for z resolution Elevation partly controls accessibility a nd distance to river (important for irrigation). It interacts with slope. O ccurrence of wetlands in this basin is related to seasonal surface flows and intersection of land surface with seasonally saturated water table. The lower the elevation, the greater the area of up-stream land cover change and hy drological changes must be integrated and the more likely wetland conversion should be Slope Model Calculated 30 m resolution, calculated from DEM and expressed as percent Percent slope partly controls accessibility since steep areas are more expensive to construct roads on. It interacts with elevation and also controls utility of land for agricultural purposes (many crops cultivation requires 0% slope for mechanized harvest to function efficiently) Wetlands occurring at no slope should be more likely to follow the conversion trajectory.
73 Table 3-2. Continued Raster Layer Data Source Description Justification Distance to Road OTS1,4 Raster surface of 30m spatial resolution where value of each cell is the s hortest distance to a public road. Note that private roads on large, commercial farms are not included here. Roads imply changes to hydrological peri od and connectivity by altering overland water flow. Also, accessibility may play an important role in determining whether a wetland area will be drained and used for agriculture. Distance to Panamerican OTS1,4 Raster surface of 30m spatial resolution where value of each cell is the shortest distance to the Panamerican Highway This is Costa Ricas major highway controlling access to national and international markets. Areas closer to the highway would seemingly be more likely to be converted directly (for agriculture) or indirectly thr ough changes in water quality or hydroperiod. Distance to River OTS1,4 Raster surface of 30m spatial resolution where value of each cell is the shortest distance to the Tempisque and Bebedero (or confluence thereof) Rivers The likelihood of switching to 12 month, int ense cultivation system depends on the availability of water with the river bei ng one of the two major water sources. Low lying, wetland areas adjacent to the river are probably more likely to be converted to irrigated agriculture than such areas that are farther from a water source. Distance to Population Center INEC5 Raster surface of 30m spatial resolution where value of each cell is the shortest distance to one of six population centers in the watershed Proximity to towns is related to the li kelihood of lowland areas being used for agriculture or probability of incurring hydrol ogic alterations relate d to urbanization. Urban runoff contributes to eutrophication of fresh water lagoons that may result in a state change and proximity to population ce nters implies sufficient labor pool to work the land. This variable is somewhat correlated with d_rds (+0.401). Coordinate Interactions Calculated x2, y2, xy2, x2y, xy3, and x3y These terms control for possible interactions or nonlinearities) between observation coordinates. Biophysical Interactions Calculated slo.elev = slope elevation slo2.elev = slope2 elevation slo.elev2 = slope elevation2 riv.slo = distance to river slope riv.elev = distance to river elevation rv.elslo = distance to river elevation slope These terms control for possible interactions between slope, elevation and distance to river. Built Structure Interactions (Accessibility) Calculated Acs_index = distance to Panam distance to rd Acs_indx2 = distance to Panam* distance to rd asdf distance to population center These terms control for possible interactions between distance to the Panamerican Highway, distance to road, and distance to population center. 1OTS = Organization for Tropical Studies. 2Vector layer of PVNP obtained from OTS and a raster indicator su rface was created indicating cells within park and cells in rem ainder of watershed. 3Produced by the Instituto Geografico Nacional (IGN) thr ough interpolation of point data on topographic maps. 4Produced by digitizing data from topographic maps 5Population statistics were obtained from the Instituto Nacional de Estadisticas y Cens os in order to identify population center s comprising 90% of the basins population; coordinates of these six cities were taken at central square using a handheld GPS (~5m accuracy).
74 Table 3-3. Eigenvalues and percent variance accounted for by each of the three principal components extracted through PCA. Component Eigenvalues % of Variance Cumulative % 1 1.80 25.65 28.11 2 1.78 25.41 53.31 3 1.55 22.16 73.22
75 Table 3-4. Rotated structure matr ix for the seven original predictor variables. Three principal components extracted through PCA are show n with their respective loadings. Distance to population center has been abbreviated, for example, as d. pop. ctr, etc. Component PC1 PC2 PC33 Predictor socioeconomic north-south topographic Communalities y -.52 -.78 .05 .88 elevation -.19 -.11 .86 .78 slope .11 .06 .79 .64 d. river -.38 .57 .39 .63 d. roads .70 -.15 -.14 .53 d. pop. ctr. .90 .048 .09 .82 d. panamerican -.20 .89 -.11 .85
76 Table 3-5. Model performance statistics comput ed with independent observations and model coefficients estimated with (Model 1) and without (Model 2) wetland patch size. AIC = Akaike information criterion, R2 = Nagelkerkes pseudoR2 AUC = Area Under the Curve for Receiver Operation Characteristic (ROC) plots, eB = exponentiation of the coefficient. Model AIC R2 AUC Socioeconomic (eB) Northsouth (eB) TopoGraphic (eB) Tenure (eB) Patch Size (eB) Intercept (eB) 1 119,820 0.48 0.79 -.81 -.85 2.68 -1.00 -0.00 2.147 (0.45) (0.43) (14.57) (0.37) (1.00) (8.56) 2 125,404 0.44 0.81 -1.03 -.82 3.10 -.91 excluded 1.99 (0.36) (0.44) (22.11) (0.40) (7.32) *All coefficients for models 1 and 2 were significant at p<0.001 level.
77 Figure 3-1 Map of the Temspique River Waters hed in northwestern Costa Rica.
78 DomainKnowledge? Parallelpiped Classifications Maximum Likelihood Classifications Accuracy Assessment SystematicErrors? No Yes Supervised ClassificationsII. Standard ClassificationAlgorithms ImagesI. Pre-Processing Calibrations Georeferencing Accuracy Assessment RuleFormulation Adequate? Final Classifications GIS (spatialdata) Interviews& Ag. Calendars Formationof KnowledgeBase KnowledgeBased Classifications Accuracy Assessment LogicalConsistency? III. Rule-BasedClassification Figure 3-2. Flowchart of image rule-based land c over classification proce dure used to derive binary wetland and non-wetland land cover maps for 1975, 1987 and 2000 Landsat images (Reprinted with permi ssion from Daniels 2006. Incorporating domain knowledge and spatial relationships into land cover classi fications: a rulebased approach. International Journal of Remote Sensing 27(14): 2949-2975). Ag. Calendars refers to an organized matrix of data compiled from interviews with local farmers and officials at th e Ministry of Agriculture (MAG) detailing crop phenology and the timing of specific land use actions (f ield preparation; planting; pesticide, herbicide and fertilization application; harvest; post-harvest field maintenance, etc.) for all major crops.
79 Figure 3-3 Map of landscape features in the Te mpisque Basin that influence wetland conversion.
80 Figure 3-4 (a) Map of probability of wetland conver sion. In the non-protected watershed, probability generally increases as distance from central, lower watershed increases, suggesting a sweeping typology of conve rsion. Within PVNP, the typology is much more patchy and the probability of conversion is generally much lower, though some areas still may be at risk for conversion (b) Map of actual wetland trajectories and model prediction errors. Correctly predicted conserved sites (green) are located predominantly in the central, lower waters hed. Correctly predicted converted sites (red) generally decrease in frequency from the outer extremities of wetland occurrence in the watershed, relative to the di scharge point of the river, in toward the central, lower watershed. False negatives, failure to predict conversion when it actually happened (yellow), are concentrat ed in the lower, central watershed both inside and outside of PVNP. Errors of commission, false prediction of wetland conversion when the site was conserved (c yan), occur in small, cluster-like areas within the central, lower watershed, in proximity to PVNP.
81 CHAPTER 4 A DECADE OF PAYMENTS FOR ENVIRONMENTAL SERVICES (PES): BUILDING ON COSTA RICAS MODEL AND APPLYING LESSONS LEARNED Summary Costa Rica has pioneered a nation-wide payments for environmental services (PES) scheme that addresses the critical role of private property land use in the provision of ecosystem services. The scheme complements the country s lauded national park system, effectively matching it in area. We describe the origin and functioning of Costa Ricas PES. We then explore a decade of nationallevel empirical trends (19972006) which demonstrate both achievements and challenges. Costa Ricas experience highlights the real-world hurdles of PES implementation and may prove instructive to emer ging and future PES schemes. Institutionaldesign tradeoffs entail striking a balance between efficiency vers us equity in participation, production versus conservation modalities, and optimal provisioning of ecosystem services versus achievement of socioec onomic objectives. We suggest se veral design-enhancements for Costa Ricas scheme. These include decoupli ng the finance of PES monitoring from the monitoring itself; strategically targeting PES land for both ecological and social objectives; reverse auctioning PES contracts to enhance efficiency and ladde ring contracts over different time spans to enhance ecosystem service continuity The long term viability and credibility of PES as a policy tool hinges on learning from the experience of existing programs and on continual innovation. Costa Rica is well-positioned to begin pilot testing some of these nuanced PES design elements. Introduction While certainly not the only approach to cons erving and managing ecosystem services, payments for ecosystem services (PES) is the first conser vation mechanism explicitly designed to address these positive externalities. Va riants of PES have existed sin ce at least 1985 when the U.S.
82 Conservation Reserve Program began purchasin g long-term cropland retirement on U.S. farms (Szentandrasi et al., 1995). Th is voluntary program retires agricultural production in exchange for several ecosystem services including soil erosion reduction, habita t provision, and improved water quality. In the developing world, Costa Rica is not only a PES pioneer, but has successfully implemented the only nationwide program to-date. Over the last decade, PES in Costa Rica and elsewhere has evolved into a moreformalized approach to manage and sustain ecosy stem services. PES-based conservation efforts have proliferated in the deve loping world, and are being activel y promoted by international aid and conservation organizations. PES goals ma y include both ecological objectives, like biodiversity conservation (Pagio la et al., 2005a), and social be nefits like poverty alleviation (Pagiola et al., 2005b) and enhanced land tenure security (Grieg-Gran et al., 2005). As with any conservation mechanism, Costa Ricas experience illustrates that PES entails navigating a complex array of program-design tradeoffs. As PES institutions continue developing, it is important to clearly define and evaluate PES in light of specific program goals to ensure they achieve their intended objectives (Mulder and C oppolillo, 2005). Indeed, th e long-term viability and credibility of PES as a policy tool hinges on learning from Costa Ricas experience and leadership in the field. PES schemes present many complex instituti onal and political design challenges due to the broad array of issues that must be addr essed and the logistics of dealing with many stakeholders. An extensive literature exists on Costa Ricas PES (Chomitz et al., 1999; LandellMills and Porras, 2002; Rojas a nd Aylward, 2003; Zbinden and Lee, 2005; Miranda et al., 2006; Pagiola, 2006; Sierra and Russman, 2006; W under, 2005; Wunder, 2007). Our goal is to complement this body of literature by reflecting on empirical trends from 1997 to 2006. Costa
83 Ricas PES system is currently gearing up to im plement a suite of innovations and enhancements after reflecting on the first World Bank/GEF-affi liated project, Ecomarkets (World Bank 2000). This new phase represents a second round of collaboration between Costa Rica and World Bank/GEF with the goal of ma instreaming and scaling up PES through focusing on identifying and refining sustainable funding mechanisms. Our review dovetails nicely w ith this initiative. Details of Costa Ricas PES scheme have not always been consistent or well-documented in the literature, likely due to the evolving legal structure of the program along with divergence between the written laws and th eir effective regional implementation. Our objectives are to accurately describe PES design and implementation and discuss themes that are critical to the enhancement and continued evolutio n of the system. In section 2, we describe the origin and operations of Costa Ricas PES, and present national-level data to illuminate trends, achievements and tradeoffs. In section 3, we analyze several themes critical to PES systems. Costanza and Farley (this issue) further discuss the importance of these themes to successful PES programs: institutional design based on program administration and opportunity costs; ecosystem service bundling and payment levels; program fina ncing and equity; spatia l considerations for PES implementation; and finally, tradeoffs in PES systems relevant to socioeconomic objectives. Some of the challenges we identify are unique to Costa Rica; others apply to PES programs more broadly. Both theory and experiences from elsewhere offer meaningful insight for enhancing Costa Ricas PES design, while programs around the globe stand to learn much from Costa Ricas experience. History and Trends in Costa Ricas PES Program PES Evolution and Scheme Design Though currently well-known for its conservation programs, in the recent past Costa Rica had one of the highest deforestati on rates in the world; between 1986 and 1991 Costa Rica was losing
84 4.2% of remaining forest cover pe r year (Sanchez-Azofeifa et al ., 2001). To address this and other environmental issues, Cost a Rica began building a system of national parks and private reserves in the 1970s, which today encompasses ove r one quarter of the na tional territory. Yet deforestation in non-protected area s continues to occur, threateni ng to isolate protected areas as forest islands (Sanchez-Azofeif a et al., 2003). Further expans ion of non-extractive protected areas is impractical, if not in appropriate, given Costa Ricas population growth rate of 1.7% (World Bank, 2007) and lingering concerns over lack of just compensation for private property incorporated into the current park system (Steed, 2003). PES emerged in Costa Rica partly in response to the need for addressing land use choices on private property. In much of Latin America, the forestry sector has a long history of government subsidies through interest-free loans, tax exemptions, provision of seed lings, extension services and even direct payments (CIFOR, 1999). In recent deca des Costa Rica has been no exception (Brockett and Gottfried, 2002). Evolution of forestry incentiv es began in the late 1970s with tax credits aimed at offsetting the costs involved in establ ishing and managing forest plantations (Figure41). From remarkably favorable credit conditions to tradable tax vouchers, Costa Rica used subsidies to promote growth in the forestry sect or. Over time, however, international pressure mounted to eliminate such subsidies. An acute financial crisis in the early 1980s saw the country become the first in a series of Latin American nations to defau lt on internat ional loans (Lara, 1995) at a time when their per capita de bt load was among the hi ghest in the developing world (Biesanz et al., 1982). Subsidies to the fore stry sector were politic ally unsustainable since Costa Ricans failed to see much contribution fr om forestry to the local economy. The third World Bank loan negotiated during the ensuing stru ctural adjustments abolished subsidies to the forestry sector (Watson et al., 1998). Yet Costa Rica cleverly turned the subsidy concept on its
85 head by articulating the broader so cial cost of deforestation and the need to compensate private land owners for the ecosystem services their fo rest stewardship provide s. Thus, Costa Ricas archetype PES program evolved s eamlessly from the existing traj ectory of forestry incentives (Figure4-1), shifting the nomina l focus from timber to conserva tion. Capacity-building and ecological awareness played an important role in affording this policy evolution. The authorizing legislation for PES in Cost a Rica was the fourth national forestry law passed in 1996 (Ley 7575, 4-16-96, Gaceta 72, Alcance 21). Ley 7575 recognizes four environmental services provided by forest ecosystems: biodiversity, watershed function, scenic beauty, and greenhouse gas mitigation through th e storage and sequestration of atmospheric carbon. Land owners may sell their environmenta l services through one of several modalities which currently include (a) refore station through plantations, (b) pr otection of existing forest, (c) natural forest regeneration, and (d) agrofore stry systems (Gaceta 51, 3-13-07). Table 4-1 reviews the criteria and impleme ntation history for Costa Ricas PES approaches. The payment per hectare is the same for all land owners within each modality (Table 4-1). Payments occur for five years, during which the PES-related land-use restriction is supposed to be noted on the property title to ensure that th e service provision con tinues even if a property is deeded to another party. Each year a program budget and PES procedur es manual are published by the Ministry of Environment and Energy (MINAE) and the PES administrative agency, the National Forestry Financing Fund (FONAFIFO), respectively. MINAE determines the distribution of funds across modalities and also provides some direction with regard to priority zones for each method. From the publication date of these ex ecutive decrees each y ear, interested land owners meeting the requirements have fifty days to submit the n ecessary paperwork to th e appropriate regional
86 FONAFIFO offices. Generally, the program can only accommodate about a quarter of the annual applicants into the scheme. By desi gn, FONAFIFO should prioritize contracts within biodiversity conservation corrido rs identified by the GRUAS repor ts (Garca, 1996; Castillo, 2006) and through annual consultation with the national system of pr otected areas (SINAC) within MINAE (Rojas and Aylward, 2003). In practice, however, prioritization of PES contracts varies regionally. Regions th at were not targeted by the World Bank-funded Mesoamerican Biological Corridor initia tive, and/or that lack a strong civil society presence to conduct outreach, may operate on a first-come, first-serv ed model of prioritizat ion out of logistical necessity (Daniels, personal observation). Each contracted environmental service provi der must have a formal forest management plan designed by a professional forester, regente according to the specifications of the modality in which they are participating (Article 20, Ley 7575) The fixed cost of this activity is taken off of the top of the program payment and is thus proportionately higher for small holders. Other responsibilities include posting signage on the la nd declaring that it is protected from hunting, fire and logging (Article 12, Gaceta 51, 3-13-07). The same regentes that write management plans are charged with monito ring compliance with PES regulations (Article 21, Ley 7575). Regentes are required to perform a site visit every twelve months for the life of the contract (Article 10.2, Gaceta 51, 3-13-07). Empirical Trends for Costa Rican PES The mean annual PES budget over the last decade exceeds $13.3 million USD or 0.43 % of Costa Ricas 2006 national budget.1 To put this in perspective, the entire EPA budget for 2006 comprised 0.0003% of the U.S. federal budget1or three orders of magnit ude difference relative to 1 Data Sources: PES budget: FONAFIFO, Costa Rica 2006 National Budget: CIA World Fact Book, EPA 2006 Budget: EPA website, US 2006 National Budget: CIA World Fact Book
87 this single program within Costa Ricas portfolio of conservation initiatives. The extent of Costa Ricas investment in PES underscores their commitment to conserving environmental services by addressing land use management on private prop erty. It also highlights the importance of iteratively reviewing the instit utional design of PES and its im plementation in the name of enhancing efficiency and efficacy. Overall budgetary efficiency, plotted as cumulative area enrolled in PES versus cumulative PES budget, corresponds roughly to th e five year payment cycles (Figure4-2a). The slope between data points for individual years represents the gain in PES area per unit of FONAFIFOs annual budget. Recruitment of area into the PES scheme diminished per unit of the budget over the first years of the program, up through 2002. This is a function of having to spend an increasing portion of each successive annua l budget servicing contracts from past years. By 2002, the 1997 cohortthe largest in the prog rams history with 102,784 hahad finished receiving payments and program efficiency in creased markedly. Conceptualizing PES as a cumulative forest protection scheme for the provis ion of environmental services as in Figure4-2 a assumes that land owners will abide by Article 19 of Ley 7575 once the payment period has expired. That is, land owners will continue to protect and maintain forest cover as mandated under the law so that PES investments have cumu lative and lasting effects for environmental service provision. In practice, however, Article 19 is somewhat unrealistic and is weakly enforced.2 For example, within the forestry modaliti es of PES, land owners may choose not to re-plant a plantation site after timber harvest and the expiration of the PES contract. In the forest 2 Article 19 of Ley 7575 was instrumental in establishing a favorable context for PES, but is problematic in its implementation. With regard to opportunity cost, the program payment (no matter how low) technically always exceeds the land rent from the next best land use, given that it is illegal to ch ange from forest land use to another activity.
88 protection modality of PES, the common practic e of forest thinning and/or clearing of the understory (socolando ) may ensue after the payments end, making the gradual land use change difficult to detect (Danie ls, personal observations). Figure4-2 b illustrates the difference between the best and worst case PES implementation scenarios respectively. The upper curve represents conservation of all PES forest, even after paym ents endlikely an unrealistic scenario. The lower curve represents conservation of only the forest areas receiving contract payments (i.e., forest area for expired PES contracts is subtract ed off the running cumulative area). Institutional design and supporting forest conser vation policies are critical in determining where the empirical curve falls between these two extremes. Another aspect of budgetary effi ciency relates to hectares per contract for individual land owners. Figure4-3 illustrates that across some time steps (e.g., 1998 to 1999), the total PES area recruited may increase while the number of cont racts stays roughly the same. This means the area per contract is greater and the relative administ rative cost per hectare r ecruited is lower. In contrast, from 2004 to 2005, the number of contract s is constant while th e recruited area drops precipitously (i.e., area per contra ct is much smaller). This i ndicates a tradeoff between program efficiency and equitable distri bution of environmental service contracts across the range of property holdings. The overwhelming majority (89%) of recr uited PES area throughout the history of the program has been for the forest protection m odality, with only five and 6% falling in the reforestation and management modalities, respec tively. The budgetary breakdown, however, is somewhat different given that payments per hect are of the timber-related modalities is over twice the payment level for forest protection in order to cover the higher costs of planting and technical assistance (Table 4-1). A decr ease in recruited PES area from 2004 to the present reflects the
89 implementation of the agroforestry modality which is based on payments per tree rather than area of forest contracted for environmental service provision (Figure4-3). Over the last decade, Costa Ricas PES program has purchased ecosystem services from over half a million hectares of land in forest use (5,314 km2), including regenerating forest and plantations at various stages of the timber production cycle (F igure4-3). As such PES has provided a significant private-property complement to the countrys network of national parks, which comprises only slightly more area (5,415 km2). As Costa Rica begins implementing a new phase of PES (corresponding to a second Wo rld Bank/GEF-sponsored project), reflecting on institutional design at this point should enhance existing arrangem ents and facilitate innovations that further improve PES performance. Evaluating Costa Ricas PES Program PES Administration FONAFIFO, the semi-autonomous arm of MINAE that administers PES, has considerable freedom and flexibility with regard to how the program is implemented. A 1990 budgetary law (Article 32, Ley 7216, Gaceta 245, Alcance 48, 12-26-90) created the agency and charged it with financing forestry initiatives among small a nd medium-sized producers. As such, the institutional strengths of FONAFIFO arguably lie in its forestry-re lated capacities. The agency was charged with managing the PES scheme only since 1996 (Article 46, Ley 7575). FONAFIFOs Board of Directors (Article 48) is comprised of two representatives from the private forestry sector, one industrial and one small to medium-sized producer group (e.g., JUNAFORCA); one representative from the Ministry of Agri culture; one from the national banking system; and a single representative from the Ministry of Environment. The Board essentially writes the executive decrees defining explicit participation criteria, modalities and payment details in the annual PES Procedures Manual. This leadership structure and the
90 historical role of FONAFIFO prior to PES may have set forth some degree of institutional pathdependency, restricting PES design and implemen tation innovations to a degree. Political pressure from the forestry lobby has fu rther reinforced this structure. FONAFIFOs particular institutional structure has both positive and negative consequences regarding PES object ives. Benefits of the forest ry-bias to date include the development of progressive, technically-sound sma ll forestry operations that have at least nominally contributed to rural development. By facilitating the establishment of such forestry plantations, the scheme design may reduce legal and illegal logging pressure on natural forests.3 Plantations also generate carbon credits with potential for sales on the international market, thereby creating a positive feedback for PES fu nding (e.g., a current proposal for the World Banks BioCarbon Fund). The negative consequence of the institutional forestry bias from a conservation perspective is that ecosystem services provided by plantation land use are production-biased relative to those provided by natu ral forest cover. To date, the scheme has identified generalized categories of environmental services provided by land uses (i.e., modalities) already employed in pre-PES forest ry incentives (Figure4-1), as opposed to identifying ecosystem functi ons and services, and then defining with greater nuance what land cover, land use and management practices best provide these services. New modalities are currently being proposed, however, and will be regionalized acco rding to local needs (World Bank 2006). A holistic approach to forest ecosystem se rvice provision and management requires that production, consumption and conservation issues be addressed in lo ckstep to enhance net levels of service provision. The tradeoff between pr oduction and conservation modalities, however, 3 A decade ago, 50% of local timber came from natural forests compared with only 5% today (MINAE/National Forest Office, 2004).
91 has been highly politicized si nce the beginning of PES in Costa Rica. Sound planning and rational discourse sometimes get lost in the propaganda from the two artificially-distant extremes. For example, the forest manageme nt modality was eliminat ed entirely in 2002, arguably on principle alone, reflecting the delicate balancing-act that FONAFIFO and policymakers face in sustaining support for PES in Costa Rican society. Unfortunately minimal rigorous peer-reviewed research exists to obj ectively provide insights regarding the optimum distribution of PES area and funding across moda lities for a range of different economic and ecological scenarios. From 1990 through 2003, FONAFIFOs role was la rgely that of a bank. In essence, its mandate is still financial in naturecollecting, managing and di spersing funds through payments and loans (Article 46, Ley 7575). Yet, PES implementation entails a host of administrative, information-systems, and monitoring/reporting considerations which the agency accomplishes using less than 10% of its given annu al budget. In 2003, FO NAFIFO took PES field administration from SINAC through the staffing of eight regional offices ( housed within regional SINAC offices). Over time, the agency has be come savvier in managing the challenging groundbased logistics of PES implementation. Decentr alization has enhanced both efficiency and accessibility for interested landholders. The ei ght administrative zones are divided into geographic regions that do not co rrespond to natural landscap e units like watersheds, however. Monitoring for Costa Ricas PES scheme is w eak and leaves room for improvement. The duty of all field verification, mana gement plan drafting and monitoring falls, by design, to third party agronomists and foresters (regentes ) compensated by PES participants out of the program payment (Article 21 of Ley 7575). Contracted foresters may have a disincentive to report noncompliance with PES contracts since they may fail to receive compensa tion if a non-compliant
92 PES contract is disqualified. Further, regentes may lose the non-compliant contract from their portfolio of managed cont racts. Since regentes have public faith (fe publica ), there is little oversight of their work. FONAFIFOs Board of Directors has been slow to develop explicit criteria and procedures that regentes should follow during the initia l and follow-up site visits. For example, only in 2004 did the Procedures Manual specify how, where and in what units GPS points should be taken on-site by the regente to identify the property being contracted for PES (Gaceta 46, 3-5-04). Data collec ted prior to 2004 were often recorded in a variety of incompatible map datums and projections. Only in 2006 did the manual require regentes to begin mapping the actual contracted forest area within the larger landho lding. This marks a dramatic improvement as officials, research ers and conservation groups may now use remote sensing methods to complement field-based mon itoring and begin to systematically quantify the impacts of PES on forest cover. FONAFIFO has demonstrated its capacity to effectively incorporate lessons-learned by adapting its administrative desi gn. Nevertheless, the PES mon itoring mechanism still merits considerable re-thinking. First, foresters may not always be the most appropriately trained for evaluating ecosystem services or monitoring thei r provision, particularly as new modalities are added in the future. A more robust approach incorporating ecologists, h ydrologists, geographers, ecological economists and landscape planners may be beneficial. Greater monitoring oversight, including penalties for hasty technical work, is also needed. The program should move toward completely decoupling the financing of monitoring from the act of monitoring itself. For example, fees now paid directly to regentes could be deposited into a general fund for each region. Then payments could be made out of the fund to regentes randomly assigned to perform
93 follow-up visits, without regard to which regente had written the original management plan. In this way, regentes could better self-police in execu ting technical and monitoring duties. Opportunity Costs The payment amount for Costa Ricas PES program has long been a topic of debate. In theory, the payment should exceed the land rent ea rned for the next-best land use option (i.e., the opportunity cost). Payments were derived from calculating an average opportunity cost for the most immediate land use option prior to PES in itiation over a decade ago, which was assumed to be cattle ranching. Since that time, FONAFIFO has annually adjusted payments upwards, to minimally match inflation (with a marked increa se in 2005). There are several problems with this approach. Land rent for cattle ranching va ries greatly depending on location and specialty (breeding, dairy or meat). Cattle ranching was rela tively less profitable due to low beef prices at the time (Arroyo-Mora et al., 2005). And fina lly, low-intensity cattl e ranching is no longer necessarily the most immediate land use alterna tive as some regions of Costa Rica have been moving away from this extensive production mode l toward higher-intensity land uses (Daniels, in prep ). Intensive agriculture and development/urban ization are increasingl y prevalent land use options. Sites suitable for cultivating export-grade pineapple, for example, can be rented for about $390/ha per year or sold for around $5800/ha (Oviedo, 2006). Such high land rent is possible by externalizing the costs of environmenta l degradation like water pollution. As long as local to international laws and institutions fail to internalize social costs, PES may be less competitive, highlighting the importance of policy coherence in effective PES implementation (Costanza and Farley, this issu e). The PES payment of $41/ha per year for natural forest regeneration or $64/ha for forest protection is trivial for those intere sted in profits alone, if their land is suited for intensive agriculture. PES is thus generally more attractive on marginal lands
94 which may or may not provide ample levels of eco system services for a particular landscape or region (see section 3.6). Rapid development in some regions increases the need for environmental services that re duce peak stream flows and prev ent flooding (Marsik and Waylen, 2006). Nearly three million square meters of ne w construction were permitted within Costa Rica in 2004 alone (Estado de la Nacion, 2006). Yet th e very process of urbanization often precludes even the consideration of PES because of th e comparatively immense one-time profit that a landowner can earn by selling their property. Land speculation and real es tate development are particularly prevalent in coastal regions and in the urban Central Valley. Zbinden and Lee (2005) point out the need for more research on opportunity cost dynamics in Costa Rica. The long-term viability of PES depends upon addressing these difficult issues of modern-day land use competition openly without being perceive d as a threat to PE S validity and utility. Consideration of how Costa Rican land use economics have changed in recent years underscores the importance of a PES design that incorporates a feedback loop for changing economic contexts. Periodic updates regarding o pportunity costs could be used in conjunction with PES contract laddering to ensure provisi on of ecosystem services over appropriate time scales despite economic change. That is, rather than having a fixed term (currently five years) for PES contracts, laddering over different term lengths with higher payment rates for longer contracts would help ensure at least some critical level of e nvironmental service provision even when market conditions make PE S a less-attractive land use. Fu rthermore, a vast literature on adoption of conservation-friendl y management practices and la nd use decision-making suggests that the process is far more complex than acc ounting for farm profit levels alone (Godoy, 1992; Ayuk, 1997; Neupane et al., 2002; Bere ntsen et al., 2007). This sugge sts that there may be room for outreach and education to enhance considerati on of the non-monetary factors involved in the
95 decision to participate in PES (e.g., the long-held Costa Rican ideal of maintaining the small family farm appears to play an important role). Ecosystem services valuation must be tied to overall quality-of-life considerations. Improved understanding of the dynami cs between natural, social, built and human capital can help better inform appropriate land use decisions (Costanza. et al., 1997; Costanza, 2001). Ecosystem Service Bundling The natural functioning of ecosystems deliv ers inseparable bundles of ecosystem services (Brennan, 1995). Often, service delivery occurs in synergistic fashion, especially between adjacent ecosystems. Certain manage ment strategies, however, can enhance some services relative to others, or even result in their total loss. Prudent ecosystem service management requires considering complementarity (e.g., riparian forest habitat and enhanced water quality) or competition (e.g., forest ha bitat versus food production from a cleared agricultural field) among services Figure4-4 conceptually illustrates a multi-dimensional production possibility frontier for several ecosyst em services. One reasonable management objective could be to increase the volume defined by the provision le vel of interacting services. Given a target level for a focal service, anot her goal could be to achieve the corresponding maximum provision level for other bundled services as illustrated. Unfortunately, however, many ecosystem service tradeoffs are still either unknown or poorly understood (Rodriguez et al., 2006). Costa Ricas PES program bundles the sale of ecosystem services. The assumption is that the prescribed land use of a given modality w ill result in the provision of at least one or more of the four environmental services specif ied by the program. Yet the different modalities leave room for various levels of service provision, with a so mewhat nebulous link between the modality, level of service provision and flat payment rate. Diffe rentiated payments would better
96 reflect the degree of ecosystem service bundling provided by a given cont ract and have been proposed for future implementation in Costa Rica (World Bank 2006). For example, land that contains old growth forest cover would certainly store mo re carbon, while simultaneously providing greater biodiversity, than equal area of early successional forest within the same life zone. The PES program might consider paying more for the old growth forest within its forest protection modality. Differentiated payments could provide part of the missing link in the current institutional design toward maximizing the service provision volume depicted in Figure44. Furthermore, allowing graduated payments through multiple tiers of ecosystem services within a modality may increase PES retention and reenlistment. Sustainable Financing A successful PES program must have the appr opriate mechanisms and political will to capture funding from a wide range of ecosystem service beneficiaries. Costa Ricas scheme successfully exemplifies this a monopsony that captures ecosystem se rvice sales across multiple scales. The scheme indirectly connect s local, regional and international buyers of ecosystem services to individual la nd owners (Figure4-5). Locally, for example, a new decree is being phased in over the next seven years (Decreto 32868, Gaceta 21, 1-30-06) where water concessionaires pay a fixed tariff that gets inve sted in watershed protection. Internationally, Costa Rica has marketed discrete carbon storage/sequestration services such as the $2 million certified offset sold to Norway in 1997. F ONAFIFO successfully bundles ecosystem services, while simultaneously exploiting markets fo r the sale of discrete services. Sustainable financing mechanisms improve th e likelihood that resources will be available to continue funding PES programs into the indefinite future. Figure4-6 illustrates the continuum of relative financial sustainability am ong funding sources both within-country and internationally. Reliance on external loans and grants is the least secure PES financing source.
97 Costa Rica has benefited from being a pioneer in the field of PES. Yet as other nations begin or expand PES initiatives, Costa Rica may face grea ter competition for such funding, stressing the need to begin pilot-testing designs that enhanc e efficiency. Stronger international treaties on biodiversity and climate change that require payments for in ternational public goods would provide a more secure external funding source (Far ley et al., this issue). Such arrangements would reduce free riding by developed nations and contribute to PES success by enhancing the demand-side of environmental markets. Within-country, financing may entail voluntary payments, funding from the general budget, or funding from a specific activity. Vol untary purchases, like FONAFIFOs innovative Environmental Service Certificates (CSAs) sold to local utilities like Energa Global, provide funding but allow free riding by non-purchasing fi rms. CSAs are certificates of bundled ecosystem services that any entity may purchase. In practice, CSAs f unction like a donation to the PES scheme, but the concept is radically diffe rent. Great potential exists for enhancing CSA sales through an eco-frie ndly certification process for tourism-re lated businesses. Criteria for the certification might include carbon-neutrality, biodiversity conservation, and enhancing hydrologic functions (e.g., offsetting diminished aquifer recharge for each square meter of constructed surface area). Firms could provide these environmental services on their property or through purchasing CSAs. This would generate revenues for PES while internalizing some of the environmental and social costs of the larg est foreign currency-earner in the Costa Rican economy, tourism (Brockett and Gottfried, 2002). Currently, however, few tourism businesses in highly-visited areas appear to be aware of CSAs, underscoring the need for an outreach mechanism to capture such sales.
98 PES funding from a nations general treas ury risks competition from numerous other budgetary needs. Therefore taxes or fees on goods or services related to provisioning of ecosystem services are more sustainable revenue sources for PES. Costa Rica has a fuel tax, currently about 28 cents/liter of gasoline (Decree 33570, 1-8-07). By law 3.5% of the revenues should be channeled to FONAFIFO to fund PES (Article 5, Ley 8114, 7-9-01), significantly less than the originally-intended one-third of revenue s. This design is conceptually sound since it requires polluters to pay for the atmos pheric waste absorption capacity for CO2. However, revenues first pass through the Ministry of Finance where competition for other legitimate uses, e.g., the Costa Rican social security system, is understandably great. Fuel tax revenues actually dedicated to FONAFIFO do not always meet thei r intended level. Three and 0.5% of fuel tax revenues would be about $8.6 million per year (Miranda et al. 2006), but FONAFIFOs budget has been as low as $3.1 million in 2005. Government estimates of income tax evasion in Costa Rica are high (Lutz and Dal y, 1991; OGrady, 2006), though recent reforms have shown promise in turning this trend around (U maa, personal observation). The fuel tax gap may arguably be making up for much-needed revenues. The PES funding shortfall creates a gap between the supply of landowners interested in PES and the demand FONAFIFO can generate with its given budget. Both the water tariff and fuel taxes apply to goods with inelastic demand. Such taxes should be more sustainable unde r changing economic conditions th an those on goods or services with more elastic demand, like tourism. Biodiversity services have proven especially challenging for developing targeted financing instruments at the local level. Co sta Rica has devised a pa rticularly innovative and sound strategy to capitalize a trust fund (The Trust Fund fo r Sustainable Biodiversity Conservation) that will serve as the financier of last resorts. The fund will target zones of
99 globally-significant biodiversity that do not overlap with exis ting carbon and water-related PES funding mechanisms (World Bank 2006). Equity in Funding PES Sustaining funding for PES invol ves iterations of internalizin g ecosystem services at the local, regional and global levels An important component of su stainable funding mechanisms is rooted in effectively dealing with free riding the act of benefi ting from a service without paying for it (Olson, 1965). Globally, Costa Rica pr ovides biodiversity c onservation and carbon sequestration, services that yield global benefits yet most recipien ts at the global scale do not pay for these benefits (Farley et al, this issue). Both of these services are c onsidered to be non-rival and non-excludable, placing them in the public goods realm (Samuelson, 1954; Randall, 1993). Who is responsible for managing and financing th ese services? Theory advises that governments should have a significant role in managing and directing ge neral funds for non-excludable services (Randall, 1993; Daly and Farley, 2004). However, dr awing from a countrys internal general funds does not reflect the la rger benefits to global society. On a regional scale, scenic beauty is considered a non-rival but excludable and congestible service, because overuse of the land scape could diminish or potentially eradicate aesthetic qualities (Randa ll, 1993). One logical way to pres erve the quality of a congestible service is to consider it a public good, which can then be subject to user fees or other methods of management (Randall, 1993; Bengston and Youn, 2006). Obvious users include tourists; past negotiations have occurred between actors such as hotels, rafting companies and tour industries (Pagiola, 2006). Yet because ther e is such a vast range of user s in Costa Rica, identifying and maintaining a collective base from which to acqu ire funding is difficult. One way might be to implement a tourist fee, which w ould more evenly distribute the cost among the beneficiaries. Currently, visitors and citizen air travelers alike pay an exit fee of US $26; an additional fee may
100 be a simple way of using an established channe l for funding scenic beauty provision. This would widely spread the burden of payments while in creasing the funding pool. Reasonable thresholds could be established through willingness-to-pay surveys and by examining similar payment schemes. This mechanism does not burden the poor since no such exit fee exists for land-based border crossings. On a local level, the free-riding issue for water conservation stands to be controlled by the new water tariff, once implemented nationwid e. The tariff will effectively equalize costs across all concession holders. Yet, if the tariff is passed on to users, it may disproportionately burden the poor since, albeit a negligible $0.003/m3 (Article 5, Decreto 32868), it constitutes a higher percentage of their total income. The fu el tax charges citizens for carbon sequestration by forests, standardizing the funding of this service acro ss local beneficiaries; but international beneficiaries continue to free-ride (Farley et al, this issue). Free-riding also occurs for biodiversity protection, as similar such measur es do not exist to expl icitly charge local beneficiaries. Spatial Variability and PES Targeting Ecosystem processes, climate, disturbance, and characteristics of human user populations clearly vary across Costa Ricas diverse geograph y, and interact to infl uence ecosystem service provision. Yet the flat payments in Costa Rica s PES scheme to-date fail to account for this variability. Carbon storage and se questration vary greatly by forest type and successional stage (Rojas and Aylward, 2003). Landscape beauty is likely greatest in places of high visibility (e.g., along roads, mountaintops). Areas of high biodiversity value are identified through the GRUAS reports (Garca, 1996; Castillo, 2006). Hydrologic services present scientific uncertainty as well as spatial dependence on human user populations. Forest type, climate, and landscape setting are all ke y factors influencing hydrologic serv ices (Chomitz et al., 1999). De
101 Camino et al. (2002) developed a qualitative rank ing system for ecosystem service provisioning by forest type, which could provide a basis for more empirical measurem ent of service provision differences, as proposed in the next phase of Cost a Rican PES. Benefits from diverse services can be aggregated using indices (e.g., the U.S. Conservation Reserve Program environmental benefits index or Australias BushTender bi odiversity quality index (Chomitz et al., 2006). Popular wisdom suggests that forests regulat e high and low flow events, increase total water supply, and reduce erosion and sedimentat ion. Scientific evidence, though, presents a more complex and site-dependent view (Bruijnze el, 2004; Bruijnzeel, 2006; Kosoy et al., 2007). Key findings of studies relevant to Costa Rica s PES program include: 1) Runoff is less in forests, except for cloud forests. 2) Dry seas on flow and groundwater r echarge contributions from forests are site-specific, and largely de pend on local geology, tree species composition, and successional stage. 3) Peak flows are mitigated in newly regrowing forests, but full benefits are achieved once complete vegetative cover becomes esta blished. This effect is most prominent in small watersheds, and less important with increasi ng watershed area. 4) Forests encourage more rainfall only in cloud forests or over large geographic areas (e .g., the Amazon). 5) There is greater scientific consensus a bout the water quality and sediment reduction benefits provided by forests. Despite uncertainty about hydrologic serv ices, utilities in Costa Rica have renewed their CSA contracts for the purchase of environm ental services, indicating their satisfaction. On an annual basis and at the nationwide scal e, Costa Rica receives far more rainfall (170 km3/yr) than its water use (6 km3/yr, Pagiola, 2006). Despite this abundance of moisture, spatial and seasonal variability can cause serious water shortages with nationwid e consequences. Since about 80% of Costa Ricas electricity is generated in hydroelectric plan ts, the variability of rainfall relative to plant locations is a critical concern. This is partic ularly true during ENSO
102 events since water levels in the Arenal reservoi r, located in the driest Costa Rican province of Guanacaste, can be significantly diminished dur ing El Nio (Amador et al., 2000). In fact, President Arias declared a national energy crisis in March 2007 due to insufficient electric production as a function of record-low water le vels in concert with other malfunctions. Economic losses in the industrial sector alone summed to $20 million in a single week within a longer period of rolling blackouts (Avalos, 2007). This underscores the critical nature of considering spatial and temporal variability of hydrological functions. Appropriate spatial targeting of watershed services could offer great er resilience in times of climate anomalies and technical failures that aff ect national electricity supply. Just as ecosystem service provisioning varies across landscapes, opportunity costs of their protection vary as well. In a general sense, environmental markets can improve conservation efficiency over command and control regulati on by identifying specific locations or firms offering the lowest costs and greatest benefits (Tietenber g, 1989; Salzman and Ruhl, 2002; Pagiola et al., 2005a). Careful arrangement of PES payments may similarly achieve the same environmental benefits at lower costs. In a system of uniform paymen ts, however, landowners with low opportunity costs receive rent from PE S programs, reducing money available to spend elsewhere, while those with hi gher opportunity costs are unwilling to participate even if they could provide socially valuable ecosystem services. Through spatial targeting, payments can be matched to levels of service provision, eliminatin g the blunt subsidy nature of uniform payments across diverse landscapes (Salzman, 2005). Tools for spatial targeting of ecosystem se rvices have been developed and used in numerous geographic contexts a nd policy settings (Babcock et al., 1996; Babcock et al., 1997; Ando et al., 1998; Polasky et al., 2001; Stoms et al., 2004; Chomitz et al., 2006; Naidoo and
103 Ricketts, 2006; Beier and Patters on, in review). Costa Ricas PES scheme might gain from implementing a spatially-nuanced approach that employs these kinds of tools (Chomitz et al., 1999; Ferraro, 2001). Wnscher et al. (2007) highlight the efficiency gains in targeting payments to landowners based on both serv ice provision and opportunity cost. They demonstrate that conservation gains for the Nico ya Peninsula in northwe st Costa Rica would be 58-88% greater using a targeted PES system that ranked each parcels total ecosystem services score and opportunity cost of service provision. The surveys th at Wnscher et al. used to estimate opportunity costs are expensive and ti me-intensive, however, meaning they could undermine the efficiency gains of a spatial targe ting. This is particul arly true since both ecosystem functioning and opportunity costs are dynamic, requiring periodic updates that exceed the capacity of FONAFIFOs field staff. Thus while a good theoretical concept, Costa Rica needs a more straightforward method to estimate landowner costs. One solution might be reverse auctioni ng where landowners self-identify through a confidential bid. Potential serv ice providers discretely subm it to the buyerin this case FONAFIFOthe price they would accept to enro ll in the PES program (Stoneham et al., 2003; Latacz-Lohmann and Schilizzi, 2005; Salzma n, 2005; Sierra and Russman, 2006). Reverse auctioning provides several clear benefits: it can prevent collusion and bidding up of prices among landowners; it is well-suited to monopsonie s; and it can reduce, though not eliminate rent seeking (Chomitz et al., 2006) by reducing in formation asymmetry between the ecosystem service provider and buyer. Reverse auctions have been us ed in Australias BushTender program (Stoneham et al., 2003; Salzman, 2005) a nd the U.S. Conservation Reserve Program. When the coordinating agency matches areas of gr eatest benefit and lowest cost, efficiency in
104 ecosystem service provision is maximized. The buye r then accepts bids up to a budget threshold, service provision level, or cost-benefit ratio. Latacz-Lohmann and Schilizzi (2005) show evidence from experiments, models, and real-world PES data demonstrating substantial gains in total ecosystem services provision on a fixed budget by targeting services and auctioning ve rsus paying a fixed price. Efficiency gains may not be universal or measurably positive from auctioning, and can shrink over time as landowners learn how to stra tegically bid (Latacz-Lohma nn and van der Hamsvoort,1997; Latacz-Lohmann and Schilizzi, 2005). Carefu l auction design and selective information disclosure by the buyer are necessary to maintain efficiency when auctions are repeated over time. Adaptively testing spatial targeting and au ctioning methods in different parts of Costa Rica could help determine their feasibility and uti lity while advancing the state of knowledge about efficient, fair, and sustainable PES systems. Currently, the GRUAS reports and other efforts provide for basic spatial targe ting. Yet once superimposed, target areas cover 70% of the country, confounding spatial prioriti zation and meaningful clustering of PES properties (Sills et al. 2005). For many services, protecting adjacent land offers synergistic benefits. Designing proper incentives for multiple-landowner coordinati on is an important challenge for the Costa Rican PES systems (Latacz-Lohmann and Schilizzi 2005) and may include allowing groups of adjacent landowners to bid together on cons ervation contracts in an auction format. Socioeconomic Objectives The potential exists for synergy between rura l development and conservation goals, yet the relationship between PES and socioeconomic objectiv es is still largely un certain. Costa Ricas PES program aims to benefit and augment the quali ty of life for rural populations whose lands possess forest or the potential fo r forest cover through silvicultu re (Article 1 of Ley 7575). Though the law itself targets rural populations, evolution of the political discourse has re-framed
105 the issue to center on whether PES positively benefits "the rural poor." At the national level, the majority of PES participants are small and medi um landholders (Sills et al., 2005). Nonetheless, regional studies found that PES participants tend to have higher off-farm incomes, larger properties and higher levels of education than otherwise equiva lent non-participants (Zbinden and Lee, 2005). Furthermore, a recent qualitative study in the central cordill era region found that the income generated from PES is used by the majority of poor part icipants for routine household expenses, precluding its application to longer-term savings or sustained quality-of-life investments (Esposito, in prep) Thus, PES is probably contribu ting very little to enhancing the economic well-being of the poorest of Costa Ricans, since participants are on-average not among the poorest landholders; and even poor landholders are lik ely better off than the landless poor. Yet economic well-being is not equal to quality-of-life. Whether earni ng income from PES or not the poorest and richest landholders alike benefit from greater landscape levels of environmental services afforded by the program. To-date, no indicators have been designed or measured for the latter in Costa Rica with regard to enhanced provision of ecosystem services through PES. Initiatives are underway to implement a PES-impact-monitoring system to better understand the degree to which socioeconomic objectives are being met (World Bank 2006). Perhaps a germane question at this juncture is, in precisely what ways does Costa Rica hope PES benefits the rural population and what are the relative priorities (income, capacity-building or en suring a healthy, safe environment)? Several modifications over the last decade have attempted to facilitate smallholder participation. In key impoverished regions, FONAFIFO makes exceptions to the need for legal title when submitting PES applications provided th at landholders meet certain requirements (see
106 Decree 31633, La Gaceta 29, 2-11-04). Transa ction fees have been reduced, though not eliminated, by allowing smallholders to form as sociations and enter PES in bulk. FONAFIFO has streamlined their information system with other government databases to facilitate verification of requirements (Pagiola, 2006). However, the degree to which these measures have facilitated access remains un-quantified. That th ere is no shortage of willing participants demonstrates that PES is clearly attractive to a sufficient number of landholders. Yet this confounds the ability to better understand achievement of PES-related socioeconomic objectives, or even how such objectives should be defined. Despite debate in development literature about appropriate tac tics for ameliorating poverty, there is consensus that financial assist ance alone will not yield success. Rather, a combination of investments in health services, education and infrastruc ture is essential (e.g., U.N. Millennium Village projects). Costa Rica ha s long-been recognized for its extensive social services and emphasis on education. FONAFIFO is gearing up to increase collaboration with civil society to enhance outrea ch and capacity-building for marginalized groups (World Bank 2006). If PES is to better the quality-of-life for th e rural poor, perhaps an explicit, formal design linking PES participation with these broader well-be ing institutions and mechanisms is needed in the next phase of implementation. Conclusions Land use change has significant ecological impacts, and is second only to electricity/heat generation as a source of global greenhouse gases (Baumert et al., 2005). To address forestrelated land use change, Costa Rica implemented a novel, market -based conservation strategy. PEScoupled with a long-standing commitment to address deforestation and biodiversity erosionhas substantially transformed the externalized values of forests. Costa Rica designed a conceptually-sound PES finance mechanism and set an example for other countries to follow. In
107 fact, PES land now rivals the much-lauded Costa Ri can national park system in area, illustrating how significantly this strategy affect s private-property land use. We reviewed Costa Ricas experience with PE S; the lessons learned range from logistical to scientific. Even when scheme design is sound (e.g. Costa Ricas reliance on a polluter paysinspired fuel tax to fund PES), implementation can fall short of the in tended policy (e.g., cooptation for other uses). This case study illustrates that many challenges arise for PES schemes due to the complexity of working with large numbers of diverse stakeholde rs in an ever-changing economic context. We have made suggestions regarding how Costa Ricas scheme design might be enhanced (Table 4.2) as it embarks on a new phase of PES implementation through the Mainstreaming and Scaling Up PES projec t with World Bank support. These include decoupling the financing of monitoring from the monitoring itself, strate gically targeting PES land for both ecological and social objectives, an d laddering contracts over different time spans to enhance the continuity of ecosystem servi ce provision. While each of these changes offer benefits and drawbacks, their careful consid eration and use can promote future PES-based conservation in Costa Rica while providing valuable lessons for emerging programs. Rodriguez et al. (2006) stress the need to critically consider tradeoffs resulting from competing ecosystem services. Ecosystem ma nagement institutions like PES schemes are influential in tipping this balance. Our re view of Costa Ricas PES program highlights institutional design tradeoffs affecting the nature, amount and geographic arrangement of ecosystem service provision. For example, la rger PES contracts are advantageous for institutional efficiency and for meeting ecological scale-dependency in ecosystem service provision. Yet this translates in to fewer PES contracts and diminished program equity. Another example entails allocatio n of PES contracts across producti on and conservation modalities of
108 PES. Forestry initiatives may contribute to rural development and re lieve timber pressure on natural forests, while protection of natural fore st generally yields a gr eater bundle of ecosystem services. FONAFIFOs institutiona l forestry bias is arguably a ppropriate for the current mix of modalities, but may be inadequate to administer m odalities added in future iterations of the PES program. This underscores the importance of feedback mechanisms in PES design so that the institutional arrangements ma y evolve appropriately. One critical tradeoff in Costa Ricas current PES scheme design occurs between maximizing ecosystem services and achievement of socioeconomic objectives. Providing socially-optimal levels of ecosystem services an d raising the quality-of-life for the rural poor are both components of the PES program. To mana ge this tradeoff consciously, Costa Rica might explicitly define quality-of-life indicators and implement ecologically-rigorous spatial targeting criteria, as we have suggested here. We have also identi fied steps to improve program efficiencyconserving more land th rough tiered payments that sp atially target areas of high ecosystem services, combined with reverse auctioni ng to conserve land at a rate consistent with landowners opportunity cost. Reverse auctioning can reduce rents to landowners, yet for the poor, such rents could constitute valuable suppleme ntal income. Alterna tively, if benefiting the poor is the programs primary goal, PES would likely achieve lower overall levels of ecosystem service provisioning. Ecological economics seeks a sustainable economic scale, fair distribution of resources, and efficient allocation (Daly and Farley, 2004). These goals are typically ranked in that order with the understanding that mainta ining justice and efficiency is impossible in the absence of sufficient natural capital to support the human economy. While some level of winwin may be possible between the numerous trad eoffs that PES entails, a more decisive PES design is required.
109 The tradeoffs we highlight do not represent design flaws per se Rather, they are inherent elements of any PES system and serve as juncture s for critical decision-ma king on the part of the implementing agencies and supporting constituencies. We have pointed out the achievements and challenges in Costa Ricas present PES sche me, providing insight usef ul to other programs in evaluating PES design choices. As this case study illustrates, PES is a pliable conservation tool that can be molded to fit specific contexts and meet certain objectives; but tradeoffs should be anticipated and d ealt with both a priori and iteratively for the long term success of environmental service provision. Viable PES schemes hinge on innovation and Costa Rica is well-positioned to begin pilot testing some of the more-nuanced design elements we have proposed here. Acknowledgements We thank the participants from the works hop Payments for Ecosystem Services: From Local to Global held in March 2007 in Heredia, Costa Rica and hosted by the University of Vermonts Gund Institute for Ecological Economics and the Universidad Nacional de Costa Ricas International Centre of Economic Policy for Sustainable Development (CINPE), with funding from the Blue Moon Foundation We are very grateful to Dr. Alvaro Umaa for his invaluable feedback on the manuscript. Thanks go to the NASA Jenkins Fellowship program for funding fieldwork.
110 Table 4-1. Legal status of PES modalities ov er the last decade. The legal citations li sted in parentheses are referenced in t he citation column on the right. Modality Status Criteria Payment Priority Citation Forest Protection Dates from adoption of Forest Law 7575 (1996) to present (1) (2) 3.4. Between 2 and 300 hectares enrolled (2) 3.6. Maximum 600 hectares (within indigenous areas) (3) 2(a) $64 per hectare per year (provided over five year period and renewable) (2) 2.2.1. SINAC biological corridors; 2.2.2. Existing biological corridors; 2.2.3. Protection of AyA hydrologic resources; 2.2.4. Unpurchased protected areas; 2.2.5. Locations in cantons with MIDEPLAN Social Development indexes lower than 40% (1) Ley Forestal N 7575, publicado en La Gaceta 72 del 16 de Abril del 1996. (2) Reglamento N 9, Manual de Procedimientos para el pago de Servicios Ambientales, FONOFIFO, publicado en La Gaceta 51 del 13 Marzo del 2007. Reforestation Dates from adoption of Forest Law 7575 (1996) to present (1) (2) 3.1. Between 1 and 300 hectares enrolled; 3.2. Maximum 50 hectares enrolled; 3.3. Minimum 50 hectares enrolled. (3) 2(b) $816 per hectare over ten-year period (2) 2.1.1. High potential forest plantations; 2.1.2. Areas with threatened species; 2.1.3. Pastures defined as Kyoto lands; 2.1.4. Projects under natural regeneration for at least one year (3) Decreto Ejecutivo N 33226, MINAE, publicado en La Gaceta 141 del 21 de Julio del 2006. (4) Reglamento N 0, FONOFIFO, publicado en La Gaceta 1 51 del 8 Agosto del 2006. Natural Forest Regeneration Dates from first mention in 2005 to present (5,6) Minimum 2 hectares enrolled (2) 3.5. (3) 2(c) $41 per hectare per year (provided over five year period and renewable) None specified (5) FONOFIFO, http://www.fonafifo.com/text_files/ servicios_ambientales/montos.pdf, (2007) Agro-forestry Systems Dates from 2003 to present (5) (2) 3.7. Min. 350 trees, max. 3500 trees per participant; 3.8. Maximum 336,000 trees per joint project, cooperative or indigenous reserve; 3.9. Specific requirements per hectare and square km. (3) 2(d) $1.30 per tree (provided over three year period) (2) 2.3.1. Projects with organizations with FONFOFIFO agreements; 2.3.2. Land as described in ( 1)Ministerio de Agricultura y Ganadera. 1995. Metodologa para la Determinacin de la Capacidad de Uso de las Tierras de Costa Rica. San Jos, Costa Rica. 60p. 2.3.3. Areas with specific agreements with FONOFIFO (6) Reglamento N 2, FONOFIFO, publicado en La Gaceta 26 del 7 Febrero del 2005. (7) Decreto Ejecutivo N 30478 MINAE, publicado en La Gaceta 112 del 12 Junio del 2002. Forest Management Dates from adoption of Forest Law 7575 (1996) until 2002 (1, 7). Criteria determined by conservation area (8) 5.1-5.10 (8) ,540 (or about $343) per hectare (provided over five year period) (8) 5.1-5.10 Priority determined by conservation area (SINAC) (8) Decreto Ejecutivo N 30090, MINAE, publicado en La Gaceta 32 del 14 de Febrero del 2002.
111 Table 4.2. Summary of issues and reco mmendations for Costa Ricas PES program. Identified Issue Category Action Benefits Challenges Institutional forestry bias Program administration Change internal structure of FONAFIFOs board; involve other environment-related professionals in PES design & monitoring Forestry-sector bias has helped develop small plantations, reducing pressure on forests Forestry-sector bias provides equal funding for plantations, which do not provide the ecosystem services of natural forests Monitoring too closely tied to compensation Program administration Have multiple regentes monitor a contract over its lifetime (e.g. random assignment each year) More transparent reporting of success & failure Need to develop feasible way to restructure regente payments from the proposed pooled fund Amount of payment Opportunity cost Redefine payment amounts to be more sensitive to economic fluctuations; consider contract laddering Possibly greater perception of fairness to landowners, and ensure longer-term service provision Difficulty of PES competing with high-value land uses; higher payment levels could mean less total land and ecosystem services enrolled Free riding and identifying long-term funding sources Equitable financing; Sustainable financing Eliminate free riding by international & local beneficiaries; move toward more sustainable financing sources Fairer & more sustainable program financing; being implemented at national level Lack of political will to require fair international payments for global ecosystem services Level of service provision variable and loosely linked to modality Bundling, Spatial targeting Target payments to areas of high ecosystem service values, differentiate payments based on services delivered, modalities, regions of the country Could improve program efficiency and total ecosystem service provision Could reduce participation by the poor; fair measurement and payment systems must be developed Level of service provision variable and loosely linked to modality Spatial targeting Reverse auctioning to identify opportunity costs for landowners to participate Could improve program efficiency and total ES provision Could reduce participation by the poor; auctions must be carefully designed to avoid strategic bidding Poverty alleviation debatable goal of PES Socioeconomic objectives Reduce transaction costs and other barriers to entry for poor households Greater participation for poor May reduce total delivery of ecosystem service benefits Poverty alleviation debatable goal of PES Socioeconomic objectives Further research into why poorer households do not participate in PES, and if PES is appropriate for poverty alleviation Better identify potential barriers to entry for poor households Requires rigorous and political will to act on recommendations
112 197919831986198819941996I n c o m e T a x C r e d i t S o f t C r e d i t s F o r e s t P a y m e n t C e r t i f i c a t e ( C A F ) A d v a n c e d F o r e s t P a y m e n t C e r t i f i c a t e ( C A F A ) F o r e s t P a y m e n t C e r t i f i c a t e f o r M a n a g e m e n t ( C A MA ) F o r e s t C o n s e r v a t i o n C e r t i f i c a t e ( C C B ) P a y m e n t s f o r E n v i r o n m e n t a l S e r v i c e s ( P S A )123467 19935 1Decree No. 10521-AH, Sept. 1979. Income tax credit given to land owners involved in reforestation activities to offset the cos t of plantations. The concept was to promote plantations as a way of alleviating defore station pressure on natural forests. Thi s tax credit targeted large landholders since small holders generally did not pay income tax. 2COREMA-AID project. International funding helped to finan ce low-interest reforestation lo ans with long grace periods and extended repayment windows. This initiative was the first of seve ral soft credit incentives, some of which still continue in t he present (e.g. FONAFIFO-brokered loans for reforestation). 3Article 82 of the Second Forestry Law (N o. 7032, La Gaceta 13; Circulo 84 May 6, 1986) creates the Certificado de Abono Forestal (CAF). Reforestation investments in plantations ar e made up front by land owner a nd compensation is given later through a tradable tax voucher. 4Decree No. 18691-MIRENEM-H, Dec. 1988. Like CAF but compensati on is given prior to reforest ation investment so that land owners with less capital could participate. 5Decree No. 22452-MIRENEM-H, 1993 (La Gaceta 170, Alcance 6). Esta blished that scientifically -managed timber extraction from natural forests would be eligible for tax vouchers. 6Decree No. 23101-MIRENEM-H, April 1994 (La Gaceta 74). Establishe d that tax vouchers could be paid for natural forest protection (equal to the CAF vouchers paid for reforestation). 7 Fourth Forestry Law (No. 7575, Gaceta 72, Alcance 21 April 16, 1996). Article 22 affirms continuation of tax vouchers for protecting natural forest, along with othe r tax benefits. Article 24 provides that land owners voluntarily allowing forest regeneration are eligible for the same benefits. Ar ticle 29 details tax benefits for plantation owners. Figure 4-1 Timeline detailing the evol ution of PES in Costa Rica.
113 20062004 2002 2000 1998 0 100 200 300 400 500 600 8182838485868Cum. Area (thousands ha)Cumulative PES Budget (millions USD) 0 100 200 300 400 500 600 1997199819992000200120022003200420052006PES Area (thousands ha ) Cum. Total PES Area (ha) Net Total PES Area (ha) 0 100 200 300 400 500 600 1997199819992000200120022003200420052006PES Area (thousands ha ) Cum. Total PES Area (ha) Net Total PES Area (ha) Figure 4-2 (a) Cumulative area in PES (thousands of ha) as a function of the cumulative budget (millions USD) FONAFIFO receives to implement PES (b) Time series of cumulative PES area (thousands of ha) recr uited across all modalities (square) in contrast with net area where expired contra cts are subtracted off of the running sum (circle). The shaded region between the tw o curves represents the difference between the best and worst case PES implementati on scenarios respectively where 100% of PES forest is conserved even after paym ent period ends (defining upper limit of region) or where none of the contracted forest is conserved after payments end (defining the lower limit of region). PES scheme design is critical in determining where the empirical curve falls between these two extremes.
114 0 10 20 30 40 50 60 70 80 90 100 1997199819992000200120022003200420052006Area (thousands of ha ) 0 100 200 300 400 500 600Thousands of Trees in SA F Management Protection Reforestation SAF Figure 4-3. Time series of the recruited ar ea per modality and number of trees in the agroforestry modality (SAF).
115 Figure 4-4 A conceptual multi-dimensional production possibility frontier for ecosystem services. Bundling ecosystem services entails maximizing the volume defined by provision levels of in teracting services.
116 Figure 4-5 Costa Ricas model for using institut ions to bundle services linking buyers and sellers across different spatial scales. Local service Regional service Global service Local, watershed, or national-level institution (e.g.., FUNDECOR, other regional intermediaries) Landowner 1 Landowner 2 Landowner 3 Buyer 1 Buyer 2 Buyer 3 (Watershed services) (National le vel tourism) (Carbon, biodiversity)
117 Figure 4-6 Relative sustainability/security of financing mechanisms for PES Funding from Internal Internal External Internal Internal External External Revenue source Tax/fee on good or service with inelastic demand Tax/fee on good or service with elastic demand Compulsory payments under international treaty Voluntary payments Funding from general budget Grants or voluntary payments Loans Example Tax or fees on energy or water users Tax or fees on luxury goods or tourism Carbon credit or biodiversity purchases under strong international treaties Payments from firms (e.g., Cerveceria Costa Rica) General appropriation subject to renewal GEF grants; Norways preKyoto carbon credit purchase; U.S. voluntary carbon markets World Bank loans Higher Moderate Lower Degr ee of sustainability/security
118 CHAPTER 5 FOREST EXPANSION IN NORTHWEST COSTA RICA: CONJUNCTURE OF THE GLOBAL MARKET, LAND-USE INTENSI FICATION AND FOREST PROTECTION Summary Though widely documented in developed c ountries, understanding of how and whether forest transitions occur in the developing world is limited. This research entails a spatiallyexplicit, remote-sensing-based anal ysis of land cover conversion tr ends in northwest Costa Rica to explore the mechanisms driving the observed forest expansion. I assess the physical and socioeconomic landscape setting for dominant conversion patterns in light of broader conservation and development initiatives, along w ith relevant economic trends, in order to propose a conceptual model accounting for forest expansion. Results demonstrate that conversion processes are interrelate d and characterized by unique la ndscape niches. Agricultural intensification occurred in lowl and areas, facilitating reforestati on of marginal grasslands in upland areas. Protected area establishment facilita ted forest recovery. Yet forest conservation and regeneration were significant on private property as well due to the conjuncture of declining beef prices, agricultural intensification and revised forestry policies. Elements of this study provide support for aspects of the classic explanations for forest transitions while highlighting the limitations of forest transition theory to account for observed reforestation trends. Introduction Changes in forest cover have consequences for biodiversity, climate change, and water resources. For many years, the focus of land-c over change research in the tropics centered on deforestation processes (e.g., S kole et al. 1994, Lambin 1997). Approaches to studying and quantifying deforestation, however have sometimes belied the complexity of forest cover This chapter is an authorized re-print: Daniels, Amy E. (in press) Forest expansion in northwest Costa Rica: conjuncture of the global market, land-use intensification and forest protection. In Nagendra, H. and J. Southworth, editors. Reforesting Landscapes: Linking Pattern and Process. Springer, NY, NY, USA.
119 changes. Forest cover dynamics may be nuanced, multi-directional and exhibit varying degrees of reversibility. Forest degradation may precede outright deforestation (World Bank 1995) or mask long-term, qualitative forest change where ne t forest cover remains stable (de Jong et al. 2001); secondary re-growth may keep pace with concomitant deforestation (Ramankutty et al. 2007), making rates of forest-cover change mi sleading oversimplificatio ns. Soil degradation that ensues in the wake of intensive non-fore st land use may preclude near-term reforestation (Carpenter et al. 2004). Altern atively, forest transitions may o ccur in which long-term declines in forest cover halt, followed by an enduring ex pansion of forested area (Mather 1992). The drivers of these respective processes of forest change are often distin ct (Grainger 1995), but sometimes overlap or interact (Stern et al. 1992). Proximate and ultimate drivers of deforestation have been identified through historicalcomparative studies and synthetic meta-analy ses (Geist and Lambin 2002). Similarly, the concept of forest transitions has emerged to explain how and why forest expansion through secondary regrowth occurs (Mat her 1992). The concept holds th at economic development drives off-farm migration and urbani zation, contributing to spontan eous forest regeneration on agricultural lands due to land ab andonment and a lack of labor (Rudel et al. 2005; Rudel, this volume). An alternative, though not mutually excl usive, pathway to the forest transition posits that scarcity of forests and forest-derived pr oducts, along with changing social norms promoting improved land management, may motivate refo restation through tree planting (Rudel et al. 2005) and/or the establishment of forest reserves (Perz 2007). Studies from developed, industr ial countries yield evidence to support these mechanisms of forest transitions (e.g., Foster 1992, Andr e 1998, Staaland et al. 1998, Mather, Fairbairn and Needle 1999). Questions abound, however, re garding how relevant developed-world
120 experiences and observations ar e in developing countries (Ko op and Tole 1999). Perz (2007) highlights the need to incorporate other contextu al biophysical and institut ional explanations of forest expansion if the forest tr ansition concept is to prove m eaningful in the developing-world forest management and conservation dialogue. National studies with aggregated data are unlikely to address important questions about regional developing-country forest change. Subnational case studies like this are critical to understanding the relationship between development and forest cover change (Klooster 2003). In the region of Mexico and Central Ameri ca, forest cover dynamics are complex with evidence of forest recovery in some places (B ray, this volume). Costa Rica is a good case in point since it has undergone net forest expansio n in recent years. The country experienced centuries of forest loss, most pronounced from 1950 to the mid-1980s, and had one of the highest deforestation rates in the world at 3.9% per year during th is period (Leonard 1986). The inflection point in Costa Ricas fo rest-area curve occurred in th e late 1980s (Kleinn et al. 2002), representing an opportunity to study the process of reversing the countrys net forest loss to a trend of forest expansion Over the last several decades, Costa Rica built a world-renowned protected area network covering over a quarter of the national territory. The protected area network and other environmental protection polic ies undoubtedly play a vital role in abating deforestation and facilitating fore st regrowth. Yet forest expans ion does not appear to be limited to protected areas (Daniels 2004, Arroyo-Mora et al. 2005), suggesting that other important processes merit further analysis. In this study I focused on identifying dominant land cover conversion sequences that explained net forest expansion in the Tempisque Basin of northwest Costa Rica. I analyzed the physical landscape setting, along with the so cioeconomic and policy context for these
121 conversions to explore the factors that drove forest expansion in the landscape from 1975 to 2000. This time period was chosen in light of the differences in data availability for important factors related to forest cover dynamics like fore stry incentives and a mo re recent program that pays for forest ecosystem services (see Forces of Landscape Change section for details). I quantified the contributio n of forest in protected and nonprotected areas to net land-cover changes and to the landscapes dominant trajectories This allowed me to partly disentangle the effect of protected area establishment from othe r processes driving forest expansion in order to develop a conceptual model of forest recovery. By contrasting this model with explanator y mechanisms posited by the forest transition concept I examine how well the latter accounts fo r observed forest expansion in the Tempisque Basin. Many of the significant studies on forest transitions fail to explicitly consider the biophysical landscape setting and pattern of forest expansion, or how these factors interact with the land tenure and policy arena (e.g., Mather 1992, Mater 1998, Rudel et al 2005). Hence, my approach in this case study cont ributes to the development of a nua nced forest transition concept that is more robust in the developing-world context. Tempisque Basin: Geographical and Historical Setting Costa Ricas Tempisque Basin lies in th e northwestern province of Guanacaste, comprising 10% of the na tional territory (5,404 km2, Figure 5-1). Mean temperature is 27.5oC and an annual mean precipita tion of 1817 mm falls between Ma y and November (Mateo-Varga 2001). The Tempisque River runs roughly north to south through the center of the basin, increasing in volume in the southern watershed with the confluence of the Bebedero River and several other important tributaries. Thirteen Holdridge life zones (Hol dridge 1967) are found in the basin, along with myriad habita ts like tropical dry forest, mois t pre-montane forest and vast
122 seasonal wetlands. The Cordillera Guanacaste defines the eastern border of the watershed, reaching over 2000 m in elevation. With more productive soils and more readily-cl eared vegetation than the rainforest-covered regions of the country, the Tempisque Basin is the only part of lowl and Costa Rica to be continuously inhabited since the be ginning of the colonial era (Pet ers 2001). Significant forest clearing had taken place in Guan acaste by the seventeeth century (Boucher et al. 1983). The Tempisque Basin is thus hardly a contemporary agricultural frontier. Since European arrival cattle ranching was the primary cause of deforestation by generating demand for grazing land; timber was a secondary motive. Land clearing wa s reinforced by settlement policies that made partial forest clearing integral to obtaining formalized land rights. Certain physical constraints a nd socio-economic characteristics of the Tempisque Basin played an integral role in calcifying the extens ive cattle-ranching land use system that dominated the region for centuries. Because of the l ong, harsh dry season, transhumant grazing was adopted where cattle were herded upslope to cloud-bathed, evergreen pastures during the dry season. Alternatively, herds were moved down sl ope to graze in the green floodplain of the Tempisque River. This extens ive pattern of low-intensity re source management was evidenced by the basins land tenure patterns: thirteen ranc hers held 11,000 ha or more; one rancher alone controlled nearly 134,000 ha (Edelman 1985) or about a quarter of the basin. Minimal investment was made in land and thus produc tion per unit area remained low for these latifundios. Beef yields, for example, averaged about 168 kg/ha, rela tive to a possible yield of 273 kg/ha (Edelman 1985). Agricultural census data shows that ranch size and output efficiency were negatively correlated in northwest Costa Ri ca, suggesting that land concentr ation was as much speculative
123 as it was production-oriented (Taylor 1980). This begs the question of why and how latifundios, a key element of a seemingly i rrational land use system that is intimately tied to forest-cover patterns, persisted for so many centuries w ithout intensifying produc tion. Edelman (1992) details the logic of Guanacastes latifundios, highlighting the importance of institutional rent (see de Janvry 1981) including subsidized producti on and highly-favorable credit conditions. Favorable credit was facilitat ed by the powerful cattlemen lobbys connections to the nationalized banking system. W ith extensive landholdings as collateral, ranching operations were as much about concentrating credit, often invested in enterprises other than ranching, and about harvesting subsidies as it was about cattle production (Taylor 1980). Since land was passed down through families, sometimes dating as fa r back as initial land grants, the cost of land was often not a consideration in land-use decision making (Edelman 1981). The pattern of land cover and te nure resulting from the persis tence of this extensive land use system in the Tempisque Basin (up until 1970s) superficially resembled that of a hollow frontier. The hollow frontier model attempts to explain the persistence of deforested land despite population declines. It pos its that settlers clear land for agriculture, degrade soils through poor land use management, and then abandon crop cultivation plots to move further into the forest (Preston 1959, Casetti and Gauthier 1977). La nd concentration occurs in the wake of this out-migration to the frontier, or in recent years for some places, to urban centers (Browder and Godfry 1997). Despite depopulation, forests fail to regenerate because the land use adopted by the large landholders, generally cattle ranching, requires little labor to stay in production. While similar to a hollow frontier in appearance and with regard to the pattern of land tenure, a key distinction for the Tempisque Basin is that land concentration largely caused outmigration rather than resulting from it. Low-inte nsity cattle ranching in the area required about
124 six worker days per hectare per year (Taylor 1980), encouraging the hi ghest negative migration rates in the country (-22%, Universidad de Costa Rica 1976). Forces of Landscape Change Several forces aligned to set the context for landscape change in the Tempisque Basin. Completion of the Panamerican Highway in th e 1950s facilitated land-based connectivity to national markets (Peters 2001). In 1953 the cattlemens lobby effectively pushed for legislation permitting the export of live cattle (Edelman 1985) later followed by beef once the first packing house was established. Sugar cane cultivation was promoted in the Tempisque lowlands after implementation of the 1959 US-C uban trade embargo. By the late 1970s, Costa Rica was the fourth largest supplier of beef to the U.S. (Peters 2001). The average price for beef on the international market was $2.37/kg during the 1970s; but in the early 1980s the price of beef on the in ternational market fell precipitously (1985-1999 mean price was $1.36/kg, Arroyo-Mora et al. 2005). At the same time, interconnected with the global recession, Costa Rica experienced an acute economic crisis while having one of the highest per capita debt loads in the world (Harrison 1991) Interest on Costa Ricas external debt rose 15% in a short time and the nation defa ulted on international loan s (Lara 1995). In the structural adjustments that ensued at the behest of international lending institutions, Costa Rica reduced public spending, including s ubsidies for cattle production. In the mid-1970s, construction began for Cost a Ricas largest hydroelectric plant just outside the eastern border of the basin, which now generates a quarter of the nations electricity (capacity 372 MW)(DeWitt 1977). After electricity generation, water began being channeled to the lowlands of the Tempisque Basin in the late 1980s via some 234 km of canals in a project known by its Spanish acronym, PRAT ( Proyecto Riego Arenal Tempisque ). The government redistributed land outside of the eastern floodplain of the Tempisque River in 5 ha plots for
125 flooded rice cultivation using wate r from PRAT. Otherwise, irriga tion water is distributed to existing agro-industry operations, which was alr eady extracting water from the Tempisque River and the underground aquifer. In addition to providing much-needed electric ity for economic and social development, PRAT greatly affected land use potential in part s of the basin. Yea r-round cultivation became possible with steady irrigation, in creasing yields and profitabilit y. This encouraged land-use intensification at a time when cat tle ranching was hardly viable. Irrigation raised yields from 3 to 8 MT/ha for rice and from 8 to 14 MT/ha fo r sugar cane (Amador et al. 2003). Furthermore, crop diversity expanded to include cantaloupe, watermelon, sod, a nd occasionally vegetables like onion and bell pepper. By the 1970s, the establishment of protected areas also acted as a force of landscape change by protecting many remaining forested areas a nd facilitating forest regeneration in others. It was the Forestry Law of 1969 (no. 4465) that created a legal mechanism for establishing protected areas through executive decrees. Since then, the park system has grown in area and morphed through various administrative stru ctures (Boza 1993, Campbell 2002), having many implications for national development. Parks are often the centerpiece for tourism development at the local and regional levels, which has generally helped garner public support for conservation. The shift of Costa Ricas histor ically-agrarian economy toward a service economy related to tourism was highlighted in 1994 when the tourism industry surpassed all other sectors of the economy in earning foreign currency (Brockett and Gottfried 2002). In the Tempisque Basin, the first national park s were established in the early 1970s, save a part of Santa Rosa first protec ted in the mid-1960s. By 2000, four teen protected areas that at least partly intersect the basin had been estab lished across multiple categories including national
126 forests and an experimental re forestation reserve (Figure 5-1 Table 5-1). These areas comprise 47,340 ha or 9.5% of the basin. While Costa Rica s protected area system does include several inhabited indigenous reserves, none fall within the Tempisque Basin. Some area incorporated into parks in the basin, like parts of Guanacaste National Park and Palo Verde National Park, had long been used as pasture. These areas were le ft to naturally regenerate or were managed under experimental and assisted forest regeneration regimes (Janzen 2002). A final, critical influence releva nt to land use and forest cover entailed a series of revisions to the forest regulatory framework. From 1969 to 1994, an evolving series of tax-credit-based incentives was developed to encourage reforestat ion and forest protection on private property. These incentives were not universal, but ra ther implemented by landholder application upon management plan approval. Results of tax credits were not always intended, particularly in the early years, since natural forest could be cleare d and replaced by plantatio ns to receive the tax benefit. In 1996, a new forestry law (no. 7575) was passed transforming these tax credits into direct payments for compliance with one of several modalities of forest-based land use (see Daniels et al. 2009 for details). That is, landhol ders began being compensated for the sale of environmental services (i.e., payments for envi ronmental services, PES). The new forest law also disallowed changes in forest land use; exta nt forest cannot be cut down even on private property (article 19) without a permit. Implementation and monitoring of both the incen tives and prohibitions provided for in Law 7575 is ever-evolving. Measuring the impact of the incentives on forest cover, particularly prior to the PES scheme created in the new fo restry law, has been difficult. Records of properties that received tax cred its were not spatial and prog ram administration has changed multiple times through the years making it difficu lt to find reliable paper-based documentation.
127 Data for the first phase of PES (1997-2000) consist of GPS points taken within ~1 km from the actual contracted forest. Furthermore, forest contracted for PES may be divided into multiple, distinct patches. Only in 2006 did PES rules begin to require a map of the actual forest contracted for ecosystem service provision. In this analysis, non-protected landscape refe rs to all land outside of public protected areas. This designation obviously under represents forest protect ion measures by not explicitly including forests conserved due to tax incentives PES and forest protection afforded by article 19 of the 1996 forestry law. These latter forms of forest protection are dea lt with implicitly to the extent that non-spatial data would allow. Also, the appropriate selection of image dates facilitates understandin g of how the discussed factors may have impacted forest cover. Methods This analysis used the classification of La ndsat satellite imagery to quantify land cover changes that occurred from 1975 to 2000 in the Te mpisque Basin and to determine net area gains or losses of major land cover classes. Single-date land cover classifications were compiled into a trajectory layer where the valu e of each pixel indicated th e specific land cover conversion sequence that occurred in that specific locati on across the three date s of analysis. The biophysical and economic landscap e setting was calculated for dom inant trajectories to discern geographic patterns associated with particular conversion sequences These generalized patterns were used to construct a generalized conceptual model of landscape change. Land-cover Classification and Change Detection Four Landsat images were obtained to a ssess land cover dynamics over time in the Tempisque Basin. All images were acquired from the early dry season since cloud cover greatly obscures land surface features during the wet season. The first two images, 1974 (path 17, Row 53) and 1975 (Path 16, Row 53), were collected by the multispectral sensor (MSS) aboard the
128 earliest Landsat satellite. These images were stit ched together as a mosaic to create a baseline land cover representation for the study region. Si nce 85% of the resulting mosaic was comprised of the 1975 scene, this mosaic is referred to as the 1975 image. The middle time point was a 1987 image from the Thematic Mapper (TM) sens or; the final date was a 2000 image collected by the Enhanced Thematic Mapper plus (ETM+) sensor. Image dates were chosen to faci litate an understanding of the drivers of landscape change. The 1975 image represents the baseline landscape la rgely prior to protected area establishment or forestry tax credits, and when extensive ca ttle ranching was still th riving. The 1987 image coincides with the inflection of the national forest transition curve for Costa Rica, the beginning of PRAT-based irrigation, and afte r the crash of the internationa l beef market. The 2000 image serves as a follow-up date, by which time all prot ected areas in the basin had been established, but prior to effects from the sy stematic application of the new forestry law (Ley 7575). Data availability for PES after 2000 lends itself to a sp atially-explicit analysis si nce maps of properties were collected (though not the actual forest patc hes within properties). For this reason, the present study stops at year 2000. Geometric rectificati on of the 2000 image used GPS-base d ground-control points and root mean square error (RMSE) was below 0.5 or 15 m. Other dates were then co-registered to the 2000 image (RMSE < 0.5). To ensure that data varian ce across the series corresponded to changes on the ground, rather than differences at th e sensor level, all images were corrected for atmospheric haze, sensor bias and differences due to variations in solar angle across nonanniversary dates (Jensen 1996). These resulting calibrated data were unitless in the form of reflectance (albedo), the ratio of energy reflected by a surface to the energy received. To ensure uniform spatial extent across dates, pixels corresponding to clouds occurring in any image date
129 were eliminated from the other dates. The 1975 scene boundary was used in subsetting images to the watershed extent because its footprint tr uncated the northern extent of the watershed (by 110 km2). These adjustments resulted in an extent of 5,153 km2 (95.35% of actual basin). Training samples for the 2000 cl assification were collected during fieldwork in 2001 (323 reference points). Historic photo-mosaics corr esponding to the dates of the satellite imagery were used to generate training samples fo r the 1975 and 1987 image (393 and 368 reference points, respectively). Over half of the land cover training data were set aside for accuracy assessment. During fieldwork, 29 semi-structure d land use history interviews were conducted throughout the watershed to construct agricultur al calendars, verify image interpretation and better understand land-use systems and changes therein. Land cover classes employed were crop ( c ), wetland ( w ), grassland ( g) and forest ( f ) (Table 5-2). Note that there are no natural dry grasslands and grassland cover thus generally corresponds to cattle ranching land use. Land cover was independently determined for ea ch image using a rule-based classification (Daniels 2006). This approach was developed because statistical clustering of spectral data alone failed to accurately identify the land cover classes of interest. The rule-based classification technique incorporated domain knowledge, spatia l relationships and two distinct clustering algorithms. Overall classification accuracy exceeded 90% for each date and class-wise accuracy was > 85% for each date. For details see Daniel s (2006). Area for each land cover class was compared across the three-date image series to determine net-area changes across the study period. Also for each date, land cover inside and outside of protected areas was calculated for comparison. Trajectory Analysis Land-cover composition was assessed for each of the three dates to determine trends of net-area change (e.g. net increase in forest c over). These net-area trends then guided the
130 spatially-explicit analysis of la nd cover conversion sequences (i .e., what conversion processes accounted for the observed net changes in land cover). Land cover across all three date s was compiled into a single raster layer of land cover trajectories where each possible sequence of land cover conversion was given a unique value. For example, a single pixel followed over the three original land-cover maps may have transitioned from grassland (g) to forest (f) from 1975 to 1987, and then remained forest through 2000. This trajectory was given a un ique integer value and labeled g-f-f to indicate the sequence. For this analysis, a total of 43 = 64 unique trajectories was possible. The area of each trajectory was calculated for protected and nonprotected parts of th e landscape by summing over the pixels in each of the two categories. Dominant Explanatory Trajectories Because of the relatively large number of possibl e trajectories (i.e., 64 classes) in a threeimage time series, criteria were developed to determine the dominant trajectories in the landscape. If a particular land cover experien ced a net-area gain from 1975 to 2000, the goal was to understand which trajectories accounted for th e increase. For instance, for a net crop area increase over the study period, what was the predominant land cover converted to cropland that explains the net increase in crop area? More generally, for an increase in land cover x the areas of all trajectories ending in x for year 2000 (explanatory trajecto ries) were compared. Similarly, if land cover y experienced a net area decrease over the study period, all possible explanatory trajectories starting with that land cover in 1975 were compared to understand what conversion processes explain the de crease in land cover y by 2000. Dominant explan atory trajectories were identified as those accounting for the majority of net-area change for a particular land cover (Table 5-3).
131 Landscape Setting for Trajectories Five raster layers were generated to provide spatia l data for key physical and socioeconomic landscape gradients related to land-cover trajectorie s based on previous research (see Daniels and Cumming 2008). The resolution of each of these layers matched the pixel size for land-cover data (30 x 30 m). Layers incl uded a digital elevation model (DEM), slope, distance to nearest road (d.rd), distance to la rge forest (d.forest), and distance to nearest population center (d.popctr). The DEM (1 m vertical resolution) was obt ained from the Organization for Tropical Studies and slope was computed from it. Public roads were obtained from the Instituto Geografico Nacional (IGN) of Costa Rica. GPS coordina tes were recorded for the six main population centers accounting for over 95% of th e population in the watershed. Large forests were defined as those on the first image with an area > 50 ha. These patches were queried and isolated from the 1975 land cover map via an eigh t neighbor rule for patch definition. ArcView Spatial Analyst calculated distance surfaces For dominant trajectories means ( ) and standard deviations ( ) for each of the landscape setting variables were calculated. Results Land Cover Area: Net Trends Net changes in land cover areas revealed the fo llowing trends in the basin: an increase in crop area, a decrease in wetland a nd grassland, and an increase in forest (Figure 52). Crop area increased 47,829 ha from 1975 to 2000, almost ex clusively in the non-protected landscape (Figure 5-3) indicating that no threat of incursion on parks ex isted. Wetland area decreased by 7,019 ha. Nearly half of the wetlands rema ining by 2000 fell in protected areas. By 2000, grassland had decreased to n early half its 1975 extent, a lo ss of 141,102 ha. The proportion of
132 grassland falling in protected areas increased w ith each date, however, since land acquired for parks within the basin often re quired restoration from non-fore st uses like cattle ranching. For the 101,404 ha increase in forest, an increas ing proportion fell within protected areas. While only 11 ha or 0.1% (Figure 54) of modern PES was invested in reforestation in the basin, this modality was the focus of all past itera tions of forest incentives, save only Forest Conservation Tax Vouchers or CPB which was th e direct pre-cursor of PES (year 1995 only). Unfortunately, reliable data are not available for estimating how much of this net increase in forest may be attributable to early reforesta tion incentives (Figure 5-4). Of the 256,708 ha of forest by 2000, nearly 12% (26,466 ha ) fell within protected areas. Explanatory Trajectories The dominant trajectories accounting for the ne t area increase in cr opland were grassland conversions ( g-c-c with 14,290 ha and g-g-c with 16,116 ha, Figure 5-5). Only about one-fifth of total 2000 crop area had been cropland throughout ( c-c-c ), indicating a substitution of crop cultivation for former pastures with grassland co ver. Net loss of wetlands was largely accounted for by conversion to cropland, with nearly four thousand hectares converted by 1987 (w-c-c) and half again as many by 2000 (2,335 ha for w-w-c ). Wetland conversion represents a minor fraction of the total area converted to crop cove r in the Tempisque Basin over the study period. Yet the proportion of wetlands lost through agricultural convers ion (6,415 ha) is much greater, 21% of their original extent. The importance of protected area esta blishment for wetland conservation ( w-w-w ) is evidenced by the fact that over half of conser ved wetlands fell within protected areas (Figure 55) (see Daniels and Cumming 2008). Reforesta tion trajectories were dominant in explaining both net loss of grasslands and net increase in forest. A majority of grassland reforestation (77,224 ha) occurred from 1975 to 1987 (g-f-f ) and over half again as much grassland (45,981
133 ha) reforested by 2000 (Figure 5-5 ) Protected areas played a more significant role in reforestation trajectories than in explaining other trends of land-cover change. Eight percent and 7.5% of trajectory area for g-f-f and g-g-f, respectively, occurred within protected areas, sizeable considering that parks comprise le ss then 10% of the basin area Forestry incentives were likely important too, though their contribution cannot be estimated directly (Figure 5-4). Figure 5-6 illustrates that of the total 2000 forest area wi thin protected areas, 16,102 ha (53%) of it was already forest in 1975 and remained forest throughout. This f-f-f trajectory in protected areas explained more of the 2000 forest area than grassland reforestation by 1987 (g-ff ; 6,211 ha or 20%) or by 2000 ( g-g-f ; 3,451 ha or 11%) combined. The forest conservation trajectory ( f-f-f ) explained 37% (83,384 ha) of non-protec ted 2000 forest area, compared with 50% for combined g-f-f and g-g-f trajectories (31% or 71,01 3 ha and 19% or 42,531 ha, respectively). Protected and Non-Protected Landscape Comparisons Results from comparing physical and socioeconomic landscape variables for protected and non-protected areas of the Tempisque Basin revealed that areas established as parks are different from the surrounding landscape (Table 5-4). Pa rk land on average has a higher elevation (232 vs. 191 m) and on slightly steeper slopes relati ve to land outside of parks (12 vs. 10%). Protected areas also occur at greater distan ces from population centers (16.1 vs. 10.2 km) and roads (3.2 vs. 1.5 km). Protected areas pixels ar e, on average, substantially shorter distances from large forest patches (0.4 vs. 1.1 km). Landscape Setting for Dominant Trajectories The means ( ) and standard deviations ( ) of landscape variables are listed for each landscape trajectory (Table 5-5) indicating spatial patterns with regard to the physical or socioeconomic landscape setting. On averag e, trajectories accounting for wetland loss ( w-c-c
134 and w-w-c ) occurred closer to population center s and roads than conserved wetlands ( w-w-w ). Conversion to cropland also occu rred on slightly higher, more-sloped land relative to wetland conservation ( w-w-w ). The latter trajectory occu rs at the lowest elevation ( = 9 m) and slope ( = 0%), and in sites that are most-isolate d relative to infrastructure like roads ( = 2.9 km) and towns ( = 17.3 km). Explanatory trajectories for cropland gains were those of grassland conversion, where conversion to cropland in the first time step ( g-c-c) occurred, at lower elevation, gentler slopes and farther from forest and infrastructure, rela tive to conversion that oc curred in the second time step ( g-g-c ). The dominant grassland loss trajectories were g-f-f and g-g-f though clearly conversion to cropland played a significant role Reforestation in the first time step ( g-f-f ), relative to the second time, occurr ed at higher elevations (239 vs 175 m), steeper slopes (14 vs. 12%), farther from population centers (11.1 vs. 10.0 km), and effectively the same distance from roads (1.5 km vs. 1.4 km). On average the g-f-f sites occurred closer to exis ting large forest patches (1.0 km) compared with grassland reforestation in the s econd time step (1.3 km). Forest conservation ( f-ff ) occurred at highest elevations ( 281 m) and steep slopes (14%). Th is trajectory was also closer to extant forest (398 m) than any other tr ajectory, in part by de finition. And finally, f-f-f trajectories were surpassed in distance to infr astructure (d.rd 2.0 km and d.popctr 12.0 km) only by the wetland conser vation trajectory. Sites that were grassland throughout ( g-g-g ) occurred at elevations (145 m) and slopes (7%) intermediate to grasslands that reforested and grasslands converted to cropland. Similarly, g-g-g sites were intermediate, relative to conversi on to cropland and refore station trajectories, with regard to distance from large forest patches (1.4 km). On average, sites characterized by the
135 g-g-g trajectory were more distant from populatio n centers (9.8 km) and roads (1.6 km) than grasslands converted to cropland. Discussion Gains in cropland were greatest from 1987 to 2000, after the implementation of PRAT which underscores the importance of regional irri gation infrastructure in changing the land-use potential for this seasonally-arid landscape. Logically, wetland loss was also greatest in the second time step since wetlands we re predominantly converted to cr ops. Greater gains in forest area were seen in the first time step corresponding to forestry ta x credits, the establishment of several protected areas, and decrea sed profitability of cattle ranc hing resulting from a significant decline in beef prices and re duced production subsidies. Grassland loss was greatest in the first time ste p, effectively reciprocal to gains in forest area. Since this time step included the crash of beef prices, this indi cates a tradeoff between cattle ranching and forest-related land uses. The decline of grasslands in the second time step was also substantial, however, owing to continued reforestation and intensification facilitated by PRAT. Grasslands in the lowland areas of the watershed were converted to crop cultivation. Also in the second time step, incremental additi ons of forest on private land was facilitated through forestry incentives and PES. For example, about 204 ha of reforestation were afforded during this period through the reforestation modality of PES (Figure 5-4). Results suggest that conversion processe s had unique domains within the broader landscape. Wetland conservation ( w-w-w ) occurred on the lowest-l ying, flattest and mostisolated regions of the basin, whereas t hose wetlands first converted to cropland (w-c-c ) were on higher, faintly-sloping land. These patterns reflect the importance of both drainage and accessibility for crop cultivation. In especially boggy sites, crop yields may be diminished by
136 seasonal flooding. Or the required investment for wetland drainage may deter conversion in such locations. Grassland areas undergoing intens ification through c onversion to cropland were located on lower and flatter land. In contrast, grassl ands on higher, sloped landsmarginal areas for intensive cultivation or pastur e usegenerally became reforeste d. These findings concur with results from an analysis of forest cover cha nge for a broader region of northwest Costa Rica (Arroyo-Mora et al. 2005), also described in Kozak (this volume) for the Polish Carpathians. Further, in agreement with patterns observed in developed-world fo rest transitions, both agriculturally-marginal and isolated lands prefer entially experienced reforestation (Mather and Needle 1998). Grassland areas where forest re generated were farther from population centers and roads, but closer to large patches of extant forest, than was grassland that converted to cropland. A substantial portion of the net increase in forest area observed by year 2000 occurred due to reforestation and forest conservation on pr ivately-held land. This provides a quasi-control for the influence of protected areas on forest expansion in the landscape, lending support to evolving forestry incentives and th e decreased profitability of cattle-ranching as drivers of forest recovery. The latter set a favorable context for the transfor mation of the dominant land use system, reinforced by PRATs implementation. Low land rent from cattle ranching also contributed to spontaneous forest regrowth by facilitating the retire ment of marginal areas of the landscape. Coupled with the forestry-incentive tax cred its and later PES, re latively unprofitable cattle ranching enhanced the economic attractive ness of forest-related land use. Since no systematic spatial record s are available for landholders that part icipated in the forestry-incentives program, their historic contributi on to forest recovery was not explicitly quantified (but see
137 Sanchez-Azofeifa et al. 2007 which used GPS poi nts taken in the general vicinity of PES projects from 1997-2000, as opposed to polygons needed for this spatially-explicit analysis). Protected areas were established preferen tially in locations where forest had been conserved; on average, parks had higher elevatio ns, steeper slopes and were more isolated. This is not unique among analyses of protected area site selection patterns (e.g., Rouget et al. 2003, Southworth et al. 2004). Precise ly these characteristics contribut e to the persistence of forest cover by making such areas less competitive fo r economically-driven and deforesting land uses (noted in several case studies in this volume). These results, in a ddition to the forest persistence trajectory ( f-f-f) on marginal lands in the non-protected la ndscape, suggest that forest in some now-protected locations would have been conserved during the st udy period due to the landscapes physical constraints al one. In terms of forest e xpansion within protected areas, reforestation of grasslands nearly equaled extant forest that became park land. Trends in the non-protected landscape suggest that some of these grasslands in lowland areas may have been converted to cropland while uplands may have re forested even without protected area status (exemplifying Graingers point that under certain conditions, The history of this landscape underscores that forest cover expands and shrinks, in part, with economic trends that influence the dominan t land use system(s). Establishing protected areas in perpetuity ensures that at least some minimum forest remains regardless of the economic context, at least in Costa Ri can, where protected area boundaries are generally respected. While there is forest cover in the non-protected landsca pe that appears relativ ely unthreatened either due to geographic setting or short-term (5 yea r) protection through PES, deforesting or forest degrading land uses for these such areas is conceiva ble in the future. Further, most of the basins non-protected forest is in varyi ng successional stages. Protected area establishment may not have
138 been determinant of forest expans ion in the basin, but is likely a critical element for the longterm recovery of old-growth tropical dry forest one of the most endangered ecosystems in the world. The unique landscape niches associated with particular trajectories, along with the broader drivers of landscape change, can be synthesized in a conceptual model explaining observed forest expansion in the Tempisque Basin (Figure 5-7). Grasslands in sufficiently flat, accessible areas intensified from pasture use to crop cultivation; on more marginal, isolated grasslands, or those falling with in protected areas, reforestation occurred. Forces behind forest recovery in the basin are multiple and interrelated as with other case studies in this volume (e.g., the combination of economic and political driver s of Vietnams transition, see Meyfroidt and Lambin). The causal structure that had long ensured the maintenance of cleare d land got rearranged (see Perz and Almeyda, this volume) in the Tempis que Basin. The conjuncture of protected area establishment, declining beef prices, diminish ed subsidies for beef production, implementation of regional irrigation and the advent of forest incentives mani fested in significant forest recovery. The decreased profitability of cat tle ranching appears to have set a favorable context, whereby its synergy with land-use intensification and forest protection, acted at a critical moment in the history of the landscape to favor forest e xpansion. Had nothing changed regarding the economics of cattle ranching, the imp act of forest incentives and ir rigation on forest cover in the Tempisque Basin would have probably been diminished. Models attempting to explain forest-cover dynamics may fail to address mechanisms of non-forest land-cover change concomitant with changes in forest area. Understanding the relationship between different c onversion processes in landscapes where forest expansion has
139 been observed will contribute to a more nuanced unde rstanding of forest recovery. In the case of the Tempisque Basin, considering the landscape as a forest/non-forest dichotomy would have failed to reveal the clear trend of land-use intensif ication, a critical component of landscape change. Intensification, while beneficial for fo rest recovery, has been detrimental to wetlands given that roughly two-thirds of non-protected we tlands were converted to cropland. The ways in which landcover patterns are affected by sh ifts in the dominant land use system (e.g. from extensive cattle ranching to intensive use) must be anticipated to ensure policy coherence. For instance, a system of wetland conservation or restoration on private landanalogous to tax incentives or, more recently, the PES scheme for forestsmay mitigate negative tradeoffs from the indirect competition between reforestation a nd wetland conservation, processes that interact despite occurring in distinct niches of the landscape. The proposed model of forest expansion (F igure 5-7) begs several important questions about the longevity of the obs erved forest recovery in th e Tempisque Basin. The unique landscape niches of different traj ectories highlight the need for spatially-conscious management decisions regarding how shifts in land use ar e facilitated or constrained by the physical landscape. Significant portions of the non-protected landscape, occurring on sites marginal for cultivation or pasture use, appear to compri se conditional forests, land that is forested depending upon the economic and policy context. If forest protection (article 19 of Ley 7575) is not fully implemented, if PES initiatives are el iminated, and/or if extensive (as opposed to intensive) cattle ranching is profitable again to the extent that it had been, will forests on marginal areas persist as implied in the enduring forest recovery concept of a forest transition? Forest protection mechanisms and incentives li ke PES are clearly critical to facilitating secondary forest regeneration and c onservation. In Costa Rica, these factors were locked in at an
140 optimal time when land-use economics were being reorganized. Recent trends (i.e., post-2000) certainly support this notion in that there are now more head of cattle in Guanacaste than in the best years of extensive ranching model of the pastyet on far less pasture. The Ministry of Agriculture (MAG) still subsidizes cattle produc tion, but of the intensive sort. Synthesizing production patterns and forest recovery patterns suggests that Co sta Rica has crafted relatively effective institutions for addressing historic defo restation threats. But are these institutions equipped to conserve the observed forest recovery for the long-term (see Daniels et al. 2009)? Demographic factors have often figured prom inently in explaining forest cover trends (Meyer and Turner 1992). Indeed, about half of the variance in the extent of deforestation can be explained by population in the long run, though the relationship is far from simple or static (Mather et al. 1998). Similarly, one of the major forest transition pathways is thought to be the demographic trend of out-migration in rural lands capes where forest regeneration occurs in the wake of a diminished labor pool and associated land abandonment (Rudel et al. 2005, Taff et al. this volume). In northwest Costa Rica, however, Harrison (1991) found that forest cover was not correlated with population trends from 1950 to 1984, the most intense period of deforestation (Harrison 1991). Given the labor-saving, extens ive land tenure and land-use system in place during the centuries of deforestati on, the lack of correlation is intu itive and noted to characterize much of Latin America (Sloan 2007). The relationship between forest cover and population in recent years for the Tempisque Basin has not been examined. Yet from 1984 to 2000, corresponding roughl y to the second time step in this analysis (1987 to 2000), the populat ion growth rate for Guanacaste was nearly 2% (compared with less than 1% from 1973 to 1984, r oughly the first time step in this study). Trends of net emigration diminished four-fold in the period from 1995 to 2000 relative to trends
141 in the 1970s. Net negative migration even reversed in some cantons to net positive immigration.6 With half again as many people in 2000 as at the beginning of the study (represented by 1973 census data) a nd a declining or reversing tr end for out-migration in the province, forest recovery does not appear to be driven by depopulation. This suggests that population patterns were independent of deforestation or reforest ation of the Temspique Basin. A carefully-designed analysis of demographic tr ends with disaggregated data and corresponding land cover maps is needed to test this. Other broad classes of explanations for fore st transitions discussed in the literature find greater support in the Tempisque Basin. Graing er (1995) describes several mechanisms that resulted in forest transitions over the cour se of economic development. Agricultural intensification leads to abandonment of marginal lands which, in tur n, reforest. He attributes this process to industrialization/ur banization and decreased competitiveness of small-scale agriculture. While intens ification was certainly an important component of forest expansion in the Tempisque Basin, its drivers were distinct fr om Graingers generalized scenario. This study illustrates the importance of context-specific factor s like PRAT, in explaining forest transitions. Grainger also describes a mechanism whereby improved land management and a shift in attitudes about forest that occurs through development. This ce rtainly characteri zes the case of the Tempisque Basin, and Costa Rica more gene rally, over the last several decades (also see Nagendra, this volume). The shift in forest valuation was not driven only by nature-based tourism, but was facilitated by exogenous forces like the global recession, crash of the beef market and reduced public spending. The latter circumstances brought about a new economic context and a reorganization of national development goals. Hen ce, according to the experience
142 of the Tempisque Basin, a robust forest trans ition theory cannot neglect to incorporate both context-specificity and non-local influe nces on land use and land cover. Agricultureextensive or intensiveis hardly the major threat to forests anymore in the Tempisque Basin. Results illustrate that low-in tensity land uses account for an ever-diminishing area in the Tempisque Basin. Perhaps the greatest uncertainty regarding the future of forest cover and the longevity of th e observed forest expansion is landscape gentrification and associated real-estate development. The forest ry laws ban on forest land-use change proves costly to enforce and arguably al together unrealistic. Pr operty values in the basin have increased steeply from real and speculativ e land investment but PES does not exceed the opportunity cost of land development which externalizes ecosystem degradation. About 24,853 ha (Figure 5-4) of forest were protected through tax vouchers (17,000 ha) and direct payments (7,853 ha), representing 10% of the 2000 fo rest area in the basin. Yet once these ecosystem-service contracts expire, the fate of these forested lands is uncertain. A critical but largely absent dimension in fore st transition theory is the role of regional and international timber trade in facilitating forest recovery at local and regional scales. At the national level, Costa Rican timber imports have increased dramatically in recent years (ONF 2006), underscoring that forest recovery does not imply a net decrease in timber demand. Analogous to the forest transition path of ma ny developed nations, fo rest conservation and regeneration in Costa Rica could be simply affo rded by displaced deforestation and/or expansion and intensification of timber plantations elsewhere. Ultimately, context-specific case studies must be linked to broader global forest cover trends and the supply/demand of fo rest goods in order to appropria tely contextualize regional or national-level forest recovery, it s driving forces, and its degree of permanence. These are all
143 critical points for post-Kyoto REDD initiatives (see also Grainger, th is volume). As of now, the issues of land development pressure, the timb er trade and deforestation displacement are not explicitly addressed in the forest transition literature for developing countries. The phenomenon of forest recovery in the Temp isque Basin is instruct ive. It casts local forest expansion, not as an inevitable by-pr oduct of economic development as suggested by forest transition theory, but as a process driven by the conjuncture of ke y global, national and local factors affecting land use patterns like th e international beef market; national forest protection policies and irrigation development; and local land use intensification respectively. Evidence from this research suggests that patterns of forest cover like those characterized by the hollow frontier and forest transition concep ts are not necessarily competing models in explaining forest-cover trends after the passing of an agricultural frontier. Rather, these models may describe distinct phases of a longer-term pro cess of forest change in different economic and development contexts. The challenge, insofar as the provision of forest goods and services is concerned, lies in understanding th e levers that maintain a favorable context for the long-term conservation of this forest recove ry in a dynamic, developing economy. Acknowledgements This research was funded through a Fulb right Fellowship and a NASA pre-doctoral fellowship. Many thanks are owed to the Orga nization for Tropical Studies, Jose Cubero of FONAFIFO, and Francisco Ramiriz and Orla ndo Matarrita of SINAC/MINAE for their assistance and support. Helpful comments a nd suggestions by reviewers are especially appreciated.
144 Table 5-1. Protected area in th e Tempisque Basin over the three dates of land cover analysis. Year Protected Area (ha) intersecting watershed Protected Area (ha) contained in watershed Percent of Watershed* 1975 55,114 6,208 1.2 1987 76,472 28,176 5.6 2000 127,561 47,340 9.5 after eliminating pixels that were not represented across all three image dates either due to cloud cover in one or more dates or WRS1 scene truncation. The latter eliminated most of Santa Rosa and Guanacaste National Parks.
145 Table 5-2. Description of land cover classes (abbrevia tions in parentheses). Class Description Crop ( c ) Dominant crops include rice, suga r cane and melon; includes all phenological stages of each crop (includi ng bare soil associated with preand post-harvest crop cover on agricultural plots Wetlands ( w ) RAMSAR definition of wetland was employed1 with the exception that flooded rice fields were consider ed crops; includes open water, mangroves, flooded forests, fresh ma rshes with emergent and floating vegetation, fresh meadows of grasses and sedges; includes all phenological stages of each sub-classes Grasslands ( g) Includes strict pasture of grasses on ly, pasture with occasional trees, recently fallowed pasture scrub (<3 y ears) still dominated by grasses, grassy areas along roads, and finall y, regions frequently burned that cannot support appreciable woody growth Forest ( f ) Limestone forests, deciduous lowlan d forest, evergreen lowland forest, and pre-montane moist forest; rainy season canopy closure > 35% 1 Areas of marsh, fen, peatland or water, whether natural or artificial, permane nt or temporary, with water that is static or flowing, fresh, brackish or sa lt, including areas of marine water the depth of which at low tide does not exceed six meters.
146 Table 5-3. Of the possible 64 traj ectories (left), only the dominant explanatory trajectories were retained for analysis of each trend of net land cover change. Explanatory trajectories account for the majority of net area change (gain or loss). Explanatory Trajectories1 All possible trajectories Forest Gain Cropland Gain Wetland Loss 1. g-g-g g-f-f g-c-c w-c-c 2. g-g-f g-g-f g-g-c w-w-c 3. g-g-c f-f-f c-c-c w-w-w n. etc. 64. w-g-c
147 Table 5-4 Population means ( ) of key variables representi ng physical and socioeconomic landscape setting in the Tempisque Basin, st ratified by protected and non-protected status. D.popctr, D.rds and D.forest signi fy distance to nearest population center, nearest road and large forest patc h (> 50 ha in 1975), respectively. N Elevation (m) Slope (%) D.popctr (m) D.rd (m) D.forest (m) NonProtected 3,813,062 191 (205) 10 (16) 10,197 (6,121) 1,527 (1,379) 1,108 (1,409) Protected 394,436 232 (291) 12 (18) 16,129 (5,988) 3,198 (2,248) 415 (790) Standard deviations ( ) in parentheses.
148 Table 5-5 Population means ( ) of variables representing physical and socioeconomic landscape setting in the Tempisque Basin, stra tified by unique land cover trajectories. Tempisque Basin for. D.popctr, D.rds and D.forest signify distance to nearest population center, nearest road and large forest patch (> 50 ha in 1975), respectively. The letters in each trajectory name represen t land cover for 1975-1987-2000, respectively, with the following abbreviations: c = crop, g = grassland, f = forest, and w = wetland. N Elevation (m) Slope (%) D.popctr (m) D.rd (m) D.forest (m) g-f-f 858,042 239 (202) 14 (18) 11,119 (6661) 1,527 (1451) 1,010 (1297) g-g-f 510,910 175 (182) 12 (17) 9,961 (6315) 1,416 (1475) 1,261 (1419) w-c-c 43,977 13 (12) 1 (2) 11,450 (4350) 2,555 (1547) 1,430 (1236) w-w-c 27,295 13 (9) 0 (2) 9,434 (5133) 1,810 (1480) 1,537 (1320) g-c-c 158,955 20 (16) 1 (2) 7,891 (3700) 1,397 (1033) 2,050 (1669) g-g-c 180,829 46 (73) 2 (5) 8,420 (4838) 1,153 (1027) 1,871 (1567) f-f-f 1,105,410 281 (252) 14 (18) 12,035 (6356) 2,000 (1609) 398 (994) g-g-g 934,159 145 (154) 7 (13) 9,755 (6096) 1,600 (1689) 1,357 (1403) c-c-c 65,815 18 (15) 1 (2) 7,698 (3435) 1,326 (1063) 2,250 (1573) w-w-w 84,159 9 (6) 0 (2) 17,310 (6897) 2.946 (1596) 1,375 (1066) Standard deviations ( ) in parentheses.
149 Figure 5-1 Map of Tempisque River Basin in northwest Costa Rica (5414 km2). Shaded polygons represent protected areas within, or that partly in tersect with, the basin. For this analysis, portions of protected areas falling outs ide the basin were excluded. Protected areas are labeled as follows: ( 1) Santa Rosa National Park (38,656 ha, 1966+), ( 2) Guanacaste National Park (34,651 ha, 1991), ( 3) Horizontes Experimental Forestry Station (7330 ha), ( 4) Rincon de la Vieja Na tional Park (14,161 ha, 1974), ( 5) La Virgen State Farm (1,923 ha), ( 6) Las Delicias State Farm (1,378 ha), ( 7) Lomas de Barbudal Biological Reserve (2,645 ha, 1986), ( 8) Palo Verde National Park (18,410 ha, 1977+), ( 9) Mata Redonda Lagoon (372 ha, 1994) ( 10) Corral de Piedra Wetland (2,484 ha, 1994), ( 11 ) Madrigal Lagoon (12 ha, 1994) (12 ) Taboga Forest Reserve (303 ha, 1978) ( 13) Diria Nacional Forest (13,402 ha combined, 1991), ( 14) Barra Honda Nacional Park (2,297 ha, 1974).
150 0 15 60 30 1975 1987 2000 Crop Wetland Grassland Forest Water Figure 5-2 Four class land cover maps (plus water) for 1975, 1987 and 2000.
151 Figure 5-3 Trends of net area land-cover ch ange for 1975, 1987 and 2000 in the Tempisque River Watershed. The black region of each bar indicates the contribution of each land cover class within protected areas. 0 50 100 150 200 250 300Area (thousands ha) Protected Non-protectedCropsWetland GrasslandForest'75 '75 '75 '75 '87 '87 '87 '87 '00 '00 '00 '00
152 Incentive1 Active Period Basin (ha) National (ha) IDR 1969 1988 unknown 35,597 CAF 1986 1995 unknown 38,086 CAFA 1988 1995 unknown 33,818 CAFMA 1993 1995 unknown 22,120 CPB 1995 17,0002 22,199 PES 1997-2000 7,8533 256,520 Tempisque Basin PES 0.1% 2.6% 97.2% Management Protection Reforestation National PES85.7% 8.6% 5.6% 1See Daniels et al. 2009 for details on the evolution of incentives. Abbreviations are acronyms for the following titles that have been translated to English: Income Ta x Deduction, Certified Forest ry Tax Voucher, Advanced Certified Forestry Tax Voucher, Advanced Certified Forest Management Tax Voucher, Forest Conservation Tax Voucher, and Payments for Environmental Services. 2Estimated by Ing. Francisco Ramirez, Area de Conservacion de Guanacaste (SINAC). No other data were available. 3These data are from 1999 and 2000 either because no projects were located in the basin from 1997-1998 or because of poor data management when PES field administration changed from SINAC to FONAFIFO. Figure 5-4 Area protected through various forms of forestry incentives for the Tempisque Basin and at the national le vel. The pie charts illustra te the break-down of 1997-2000 PES area across the three forestry modalities.
153 Crop Expansion0 3 5 8 10 13 15g-c-c g-g-c c-c-cThousands ha Protected Non-protected Wetland Loss0 3 5 8 10 13 15w-c-c w-w-c w-w-wThousands ha Forest Expansion0 20 40 60 80 100g-f-f g-g-f f-f-fThousands ha Figure 5-5 Dominant land cover trajectories explai ning the observed net area changes in land cover. The black region of each bar indicates the contribution of respective trajectories within protected areas. The three letters in each trajectory name represent land cover for 1975-1987-2000, respec tively, using these abbreviations: c = crop, g = grassland, f = forest, and w = wetland.
154 0 10 20 30 40 50 60f-f-f g-f-f g-g-fPercent of Forest in 2000 Non-protected Protected Figure 5-6 Graph of dominant trajectories explaini ng the expansion of forest observed by year 2000 for protected and non-protected areas of the landscape. The bars indicate the percent of non-protected (gray) or protected (black) forest in year 2000 explained by each trajectory.
155 Distant from Infrastructure Forest Protection, PA Establishment & Incentives/PES Lowland Upland Irrigation Proximity to Infrastructure Grassland Crops Forest Wetland Drainage Intensification Reforestation Decreased CattleRanching Profitability Forest Reforestation Distant from Infrastructure Forest Protection, PA Establishment & Incentives/PES Lowland Upland Irrigation Proximity to Infrastructure Grassland Crops Forest Wetland Drainage Intensification Reforestation Decreased CattleRanching Profitability Forest Reforestation Figure 5-7. Conceptual model of dominant tr ends accounting for forest expansion in the Tempisque Basin. The landscape is represen ted by the large recta ngle, divided into uplands (gray) and lowlands (white). The four land cove r classes analyzed in this study are represented by small rectangles with arrows depicting the dominant trajectories (e.g. grassland conversion to cr ops). Ovals around th e landscape depict the forces driving observed processes of landscape change. The dominant reforestation trend occurred in upland areas (black). Lowland reforestation was a minor factor in accounting for forest expansion (gray).
156 CHAPTER 6 CONCLUSIONS At the time scale of an average human lifes pan, humans may be the most dominant force shaping ecosystem function and structure (Meyer and Turner 1994). Society has domesticated landscapes and simplified ecosystem functioning to enhance the ra te and amount of particular ecosystem goods produced or capable of being extracted at any singl e location (e.g. food and fiber production through crop cul tivation, timber production through managed forestry systems, or energy extraction through foss il fuel mining). Ecosystem specialization, by necessity, expands the human footprint or dimensions of th e resource catchment from which the needs of a typical person are met, particularly in the deve loped world. Ecosystem specialization is thus both a cause and a consequence of globalization as disparate and remote locations become increasingly interdependent through trade. For ces in distant locations drive local land cover change. These changes alter the global mix of ecosystem services, but may have disproportionate effects on local ecosystem service provision like water quality control or the maintenance of soil productivity. Processes of environmental change and ecosystem service tradeoffs are complex, and multidirectional system. Conservation dialogues and financial investments focusing narrowly on traditional protect ed areas are therefore ever more irrelevant (Kareiva et al. 2007). Prot ected areas undoubtedly serve an important function within conservation agendas; but traditional parks cannot cover sufficient area to preclude difficult present and future decisions about resour ce and land use on the margin. Minimizing tradeoffs between conservation and economic production functions depends in part on the spatial pattern of societys resource and land use (Polasky et al. 2005). As revealed in chapter two, spatial juxtapositions, like milpa /forest adjacency in the tropical dry forest of the Yucatan, affect ecosystem structure. Increasing ly, scientists highlight the importance of past
157 land use on contemporary ecosystem properties (Fos ter et al. 2003). Rarely, however, are spatial patterns of past land uses considered as emphasi zed in chapter two. These results underscored the importance of characterizing spatial patterns over time when attempting to understand the legacy of past land use systems and also the ne ed to measure changes to ecosystem structure within a given land cover class (forest in this case). Results showed that the trait values or forest structure for vegetation occurring at milpa edges were different fr om background forest. If basal area roughly indicates carbon storage, just one particular ecosystem service, then minimizing the length of edge per area of milpa within ejidos would enhance this partic ular service. The wetland conversion model in chapter three provided a detailed case study that crystallized an important ecosystem tradeoff o ccurring around the globe: the loss of ecosystem services provided by wetlands in exchange for cr op cultivation. Whether this transformation of the lower Tempisque basin is reversible remains to be seen and hinges on the degree to which previous landscape hydrologic c onnectivity has been severed through geomorphic alterations. For the foreseeable future, these low-lying, flat areascultivable now that they are drained will likely remain in crop land use. Costa Rica has a biofuel mandate that will likely keep the demand for sugar cane high in the co ming years. This spatially-ex plicit model not only estimates the probability of conversion based on past convers ion patterns, but it can be used as a spatialtargeting policy tool. For example, if a wetland -analogue to the forest-based PES system were implemented, this model would iden tify the sites most likely to be converted to other land uses and aid in the differentiation of payments according to th e opportunity cost. Chapter five emphasized the critical links be tween institutions and landscape patterns. For much of the Tempisque Basin s history, the prevailing extensive latifundio assured that extensive cattle ranching was the predominant land use system, at the expense of forest cover.
158 Once the global market acted to remove the pi llars propping up this antiquated institution, the landscape was subject to reorganization. Both agricultural intensif ication and forest conservation policies transformed land-cover patterns in r ecent decades. Protected area establishment certainly facilitated conservation of existing forest and encouraged some degree of reforestation, though much of the land incorporated into park s was disproportionately forested already. Further, much of the gain in forest cover wa s achieved through reforestation on private land, particularly upland areas. New institutional driving forces in the landscape facilitated forest regrowth and thus forest-related ecosystem services like carbon sequestration. Though land -use intensification released marginal lands from extensive cattle ranc hing for reforestation, inte nsification played an important role in the loss of wetlands in the Temp isque Basin. Wetlands and forests were not in direct competition since these ecosystems o ccur in distinct domain s of the landscape. Institutions did, however, act to create an indirect tradeoff between these two ecosystem types and their respective ecosystem services. Forest protection through park establishment and PES alone may not have been sufficient to transform land cover patterns and the mix of ecosystem services in the Tempisque Basin. But these f actors acting in just th e right economic context locked the landscape on a new traject ory. The longevity of this fo rest transition depends largely on the degree to which Law 7575 is enforced and the extent to which opportunity costs associated with PES continue to increase in th e basin (relative to one-time land sales for realestate development or land rent from intensive agriculture). The analysis of Costa Ricas PES scheme in chapter five demonstrates both the progress and room for improvement within this innovative inst itutional scheme. Development agencies seek to replicate this model in other parts of the developing world. Yet the context elsewhere
159 may not be as amenable due to confounding factor s like abject poverty, a lack of administrative experience, the absence of land titles, and/or a hi story of political violence, for example. Costa Ricas experience may minimally serve as a base line from which othe r programs may build, tailoring implementation according to the unique challenges presented. Several suggestions for PES include dec oupling the monitoring of PES from the mechanism financing PES monitoring; strategica lly targeting contract s in order to seek landscape-pattern-based synergies; and reverse auctioning contract s in order to optimize service provision for each dollar spent. Many of the challenges for Costa Ricas PES lay in the actual ground-based implementation, not in technical or policy realms. Even when the design is sound, implementation can be difficult because the hum an-resources and capacity-building aspects of PES are largely unappreciated. This is a particularly important and timely conclusion from this research. The Norwegian government recently announced that it would contribute $500 million per annum to forest conservation activities that redu ce emissions from deforestation and degradation (United Nations 2008). This immense investment will double the tota l amount of foreign aid dedicated to forests and forestry programs worldwide (Zarin 2008). The dearth of communication between field officers and high-ranking officials in the PES administration is, perhaps, in large part by design. PES administrators and financiers at the World Bank anxiously, and righ tfully, await returns on their investment of public funds in the PES sc heme. Yet, rather than openly assessing, conceding and re-thinking program weaknesses, the channels for communicating constructive criticism or suggesting program design enhancements are not well-established. Perhaps, in large part, the administration legitimately fears that admitting shortcomings in the PES scheme may deflect potential carbon offset investments and th at small Costa Rica would be by-passed for
160 investments in tropical defo restation hotspots like Brazil a nd Indonesaia. These complex institutional dynamics underscore the critical importance and challenge of defining PES baselines and measuring additionality in ways th at do not penalize Costa Rica for its decades of continually-improved forest stewards hip and environmental management. Hence, the real challenge in making use of spatial analyses like those performed in chapters two, three and five, lie s in developing the institutions to evaluate and manage the inherent tradeoffs in ecosystem services that aris e. Ultimately these decisions about tradeoffs are often value-based. If the aggregat e global land use and land cover pa ttern is explicitly framed as representing the collective ecosystem services most valued at a particular moment in time, this may better facilitate dialogue and deliberate choi ces regarding their tradeoffs, many of which are irreversible and redirect ecosyst em functioning in uncertain ways (Rodriguez et al. 2005). Land change science offers a particularly powerful appr oach for understanding the patterns and drivers of land conversion and, thus, broad-scale tradeoffs in ecosystem services.
161 APPENDIX MODEL BUILDING AND MODEL SELECTION STEPS FOR PREDICTING WETLAND CONVERSION Conceptual Model of Wetland Change in the Tempisque Basin
162 Flowchart of Modeling Process
163 Data from Akaike Informatio n Criterion Comparisons for Variable Selection and Model Selection 1. Variables were eliminated if their removal resu lted in a lower AIC relative to the saturated model. 2. The "asc.index" variable refers to an accessibility index (distance to road distance to Panamerican). 3. The saturated model included all of the variables listed in the "variable removed" column plus patch size and the tenure indicator. As explained in the te xt, we wished to quantify the contribution of these variables for specific reasons, so their rem oval was not considered at this stage. 4. Model A included these variables in the PCA: y -coordinate, elevation, slope, distance to river, distance to population center and acs.index. 5. Models B and C included these variables in the PCA: y -coordinate, elevation, slope, distance to river, distance to population center, distance to road and distance to Panamerican in the PCA. 6. Model C was the final, selected model despite a slightly higher AIC value. This is because for models B and C we also evaluated R2 and AUC values using inde pendent data. See text for details.
164 LIST OF REFERENCES Amador JA, R.E. Chacon, and S. Laporte. 2003. Climate and climate vari ability in the Arenal River Basin of Costa Rica. In H. F. Diaz and B. J. Morehouse, editors. Climate and water: transboundary challeng es in the Americas. Kluwer Academic Publishers, Boston, USA. Amador, J.A., S. Laporte, and R.E. Chacon. 2000. Cuenca del Rio Arenal: analisis del los eventos Nino de los anos 1992-93, 1994-95 y 1997-98. Topicos Meteorologicos y Oceanograficos 7(1): 1-20. Ando, A., J. Camm, S. Polasky, and A. Solow. 1998. Species distributions, land values, and efficient conservation. Science: 279: 2126-2128. Arroyo-Mora, J. P., G. A. Sanchez-Azofeifa, B. Rivard, J. C. Calvo, and D. H.Janzen. 2005. Dynamics in landscape struct ure and composition for the chorotega region, costa rica from 1960 to 2000. Agriculture Ecosystems & Environment 106: 27-39. Avalos, A.R. 2007. Ice suspende coretes de elec tricidad en todo el pais La Nacion. Sabado 5 de mayo. Available at http://www.nacion.com/ln_ee/2007/mayo/05/pais1085589.html Accessed on May 5, 2007. Ayuk, E. T. 1997. Adoption of agroforestry tech nology: The case of live hedges in the central plateau of Burkina Faso. Ag ricultural System s 54: 189-206. Babcock, B.A., P.G. Lakshminarayan, J. Wu, and D. Zilberman. 1996. The economics of a public fund for environmental amenities: A stu dy of CRP contracts. American Journal of Agricultural Economics 78: 961-971. Babcock, P.G., P.G. Lakshminarayan, J. Wu, and D. Zilberman. 1997. Targeting tools for the purchase of environmental amenitie s. Land Economics 73(3): 325-339. Baldi, A., and T. Kisbenedek. 1999. Species-specific distribution of reed -nesting passerine birds across reed-bed edges: effects of spatia l scale and edge type. Acta Zoologica Academiae Scientiarum Hungaricae 45: 97-114. Banerjee, A. K. 1995. Rehabilitation of de graded forests in Asia. World Bank. Bierregaard, R.O., Lovejoy, T.E., Kapos, V., Augusto dos Santos, A., Hutchings, R.W., 1992 The Biological Dynamics of Tropical Rainfo rest Fragments. BioScience 42(11): 859866. Baumert, K., T. Herzog, and J. Pershing. 2005. Navigating the numbers: greenhouse gas data and international climate policy. World Re sources Institute, Wa shington D.C., USA. Bedford, B.L., and E.M. Preston. 1988. Developi ng the scientific basis for assessing cumulative effects of wetland loss and degradation on la ndscape functions: Status, perspectives, and prospects. Environmental Management 12(5): 751-771.
165 Beier, C., and T. Patterson. In review. Targeted social-ecological vulnernability at the landscape scale: ecosystem services, social importance and disturbance. Submitted to Ecosystems. Bengston, D.N., and Y.C. Youn. 2006. Urban C ontainment Policies and the Protection of Natural Areas: The Case of Seouls Greenbelt. Ecology and Society 11: 3. Berentsen, B.M., A. Hendriksen, W. Heijman, and H. van Vlokhoven. 2007. Costs and benefits of on-farm nature conservation. Ecological Economics 62: 571-579. Biesanz, R., K.Z. Biesanz, and M.H. Biesan z. 1982. The Costa Ricans. Waveland Press, Prospect Heights, IL, USA. Bonan, G.B., S. Levis, S. Sitch, M. Vertenstein, and K.W. Oleson. 2003. A dynamic global vegetation model for use with climate models : concepts and description of simulated vegetation dynamics. Global Change Biology 9: 1543-1566. Bunn, S. E., P. M. Davies, and T. D. Mosisch. 1999. Ecosystem measures of river health and their response to riparian and catchment degradation. Freshwater Biology 41: 333-345. Boylan, K.D., and D.R. Mclean. 1997. Linking species loss with wetlands loss. National Wetlands Newsletter 19(6): 13-17. Boucher D.H., M. Hansen, S. Risch et al. 1983. Agriculture. In D.H. Janzen, editor. Costa Rican Natural History. University of Chicago Press, Chicago, IL, USA. Boza M.A. 1993. Conservation in action past, pres ent, and future of the national park system of Costa-Rica. Conservation Biology 7: 239-247. Brandon, K., K.H. Redford, and S.E. Sanderson, edito rs. 1998. Parks in pe ril: people, politics, and protected areas. Island Pr ess, Washington, D.C., USA. Brennan, A., 1995. Ethics, ecology and economics. Biodiversity and Conservation, 4 (8): 798811. Brockett C., and R. Gottfried. 2002. State policies and the preservati on of forest cover: lessons from contrasting public-policy regimes in Costa Rica. Latin American Research Review 37(1): 7-40. Browder J.O., and B.J. Godfrey. 1997. Rainforest cities: urbanization, development, and globalization of the Brazilian Amazon. Columb ia University Press, NY, NY, USA. Bruijnzeel, L.A. 2004. Hydrological functions of tropical forests: Not seeing the soil for the trees? Agriculture, Ecosystems, and Environment 104: 185-228.
166 Bruijnzeel, L.A. 2006. Cloud forests, hydrology and payment for Ecosystem Services in Costa Rica. http://www.iucn.org/themes/cem/news/newsletter/2006/cem_newsletter03_2006.htm Bruner, A.G., R.E. Gullison, R.E. Rice and G. Fonseca. 2001 Effectiveness of parks in protecting tropical biodivers ity. Science 291: 125-128. Brown, D.G., P. Goovaerts, A. Burnicki, and M. Y. Li. 2002. Stochastic simulation of landcover change using geostatistics and genera lized additive models. Photogrammetric Engineering & Remote Sensing 68(10): 1051. Bunn, S.E., P.M. Daview and T.D. Mosisch. 1999. Ec osystem measures of river health and their response to riparian and catchment de gradation. Freshwater Biology 41: 333-345. Campbell, L.M. 2002. Conservation narratives in Costa Rica: Conflict and co-existence. Development and Change 33: 29-56. Carpenter F.L., J.D. Nichols, and E. Sandi. 2004. Early growth of native and exotic trees planted on degraded tropical pasture. Forest Ecol Manag 196 (2-3): 367-378. Andre MF (1998) Depopulation, land-use change and transformation in the French Massif Central. Ambio 27: 351-353. Casetti E, and H.L. Gauthier. 1977. Formali zation and Test of Hollow Frontier Hypothesis. Economic Geography 53: 70-78. Castillo, E.A. 2006. Proyecto Ecomercdos: Cons ultoria Para Actualizacion de la Propuesta Technica de Ordenamiento Territorial Con Fines de Conservacion de Biodiversidad en Costa Rica (GRUAS II). Chang, C.R., P.F. Lee, M.L. Bai, and T.T. Li n. 2006. Identifying the scale thresholds for fielddata extrapolation via spatia l analysis of landscape gradients. Ecosystems 9: 200-214. Chomitz, K.M., E. Brenes, and L.Constantino. 1999. Financing environmental services: The costa rican experience and its implications. Science of the Total Environment 240: 157169. Chomitz, K.M., G.A.B. Fonseca, K. Alger, D.M. Stoms, M. Honzk, E.C. Landau, T.S. Thomas, W.W. Thomas, and F. Davis. 2006. Viable reserve networks arise from individual landholder responses to conservation initiatives. Ecology and Society 11: 40. Chowdhury, R.R., L.C. Schneider, Y. Ogneva-Himme lberger, P.M. Mendoza, S.C. Villar, and A. Barker-Plotkin. 2004. Land Cover a nd Land Use: Classification and Change Analysis. Pages 105-141 in : B. L. Turner II, J. Geoghegan, D. Foster, editors. Integrated Land-Change Science and Tropica l Deforestation in the Southern Yucatn: Final Frontiers. Clarendon Press, Oxford, UK.
167 CIA, 2007. The World Fact Book Costa Rica. Available at https://www.cia.gov/cia/publications/factbook/geos/cs.htm l, accessed on May 2, 2007. CIFOR,1999. Your subsidies are my incentives. POLEX. Available at http://www.cifor.cgiar.org/Publica tions/Polex/polexdetail.htm ?pid=19 accessed on December 18, 2006. Cincotta, R.P., J. Wisnewski, and R. Enge lman. 2000. Human population density in the biodiversity hotspots. Nature 404: 990-992. Coomes, O. T., F. Grimard, G. J. Burt. 2000. Tropical forests and shifting cultivation: secondary forest fallow dynamics among trad itional farmers of the Peruvian Amazon. Ecological Economics 32: 109-124. Corbett, E., and R. C. Anderson. 2006. Landsca pe analysis of Illinois and Wisconsin remnant prairies. Journal of the Torre y Botanical Society 133: 267-179. Costanza, R., d'Arge, R., de Groot, R., Farber S., Grasso, M., Hannon, B., Limburg, K., Naeem, S., O'Neill, R.V., Paruelo, J., Raskin, R.G., Sutton, P., van den Belt, M., 1997. The value of the world' s ecosystem services and natural capital Nature 387: 253-260. Costanza, R. 2001. Visions, Values, Valuation, and the Need for an Ecological Economics. BioScience 51(6): 459-468. Costanza, R. and J. Farley. 2009. Introduction: Theory and Practice of Payments for Ecosystem Services: The Heredia Declaration. Ecological Economics. In press. Cowan Jr., J.H., and R.E. Turner. 1988. Mode ling wetland loss in coasta l Louisiana: geology, geography and human modifi cations. Environmental Management 12: 827-838. Cumming, G.S. 2000. Using betwee n-model comparisons to fine-tune linear models of species ranges. Journal of Biogeography 27: 441-455. Daly, H.E. and J. Farley. 2004. Ecological Econ omics: Principles and Applications. Island Press, Washington, DC, USA. Daniels, A.E. 2004. Protected area management in a watershed context: a case study of Palo Verde National Park, Costa Rica. Thesis. University of Florida, Gainesville. Daniels, A.E. 2006. Incorporating domain knowle dge and spatial relationships in land cover classifications. International Journa l of Remote Sensing 14(20): 2949-2975. Daniels A.E., G.S. Cumming. 2008. Conversio n or conservation? Understanding wetland change in northwest Costa Rica. Ecological Applications 18: 49-63.
168 Daniels A.E., V. Esposito, K. Bagstad et al. 200 9. A decade of PES: building on Costa Ricas model and applying lessons learned. Ec ological Economics, special issue. De Camino, R., O. Segura, L.G. Arias, and I. Prez. 2002. Costa Rica: Forest Strategy and the Evolution of Land Use. World Bank Evaluation Country Case Study. Series 148. The World Bank, Washington, DC. Defries, R., A. Hansen, A.C. Newton, and M.C. Hansen. 2005. Increasing isolation of protected areas in tropical forests over the past twen ty years. Ecological Applications 15(1): 1926. DeFries, R.S., M.C. Hansen, and J.R.G. Townsend. 2000. Global continuous fields of vegetation characteristics: a linear mixt ure model applied to multi-year 8 km AVHRR data. International J ournal of Remote Sensing 21(6-7): 1389-1414. A. De Janvry. 1981. The agrarian question and re formism in Latin America. The Johns Hopkins University Press, Baltimore. de Jong W, L. Freitas, J. Baluarte et al. 2001. Secondary forest dynamics in the Amazon floodplain in Peru. Forest Ecology and Management 150:135-146. DeWitt R.P. 1977. The Inter-American Development Bank and political influence, with special reference to Costa Rica. Praeger, NY, NY, USA. Eamus, D. 2001. How does water balance influence net primar y productivity? A discussion. NEE Proceedings Workshop, pp. 62-70. Eaton, J. and D. Lawrence. 2006. Woody debris stocks and fluxes during succession in a dry tropical forest. Forest Ecology and Management 232: 46-55. Edelman M. 1981. Apuntes sobre la consolidacion de las haciendas en Guanacaste. Avance de Investigacin No. 44. Instituto de Inves tigaciones Sociales, Facultad de Ciencias Sociales. Universidad de Cost a Rica. San Jose, Costa Rica. Edelman M. 1985. Extensive Land-use and the Logic of the Latifundio a Case-Study in Guanacaste Province, CostaRica. Human Ecology 13:153-185. Edelman M. 1992. The logic of the latifundio : the large estates of northwestern Costa Rica since the late nineteenth century. Stanford University Press, Stanford, CA, USA. Efron, B., and R.J. Tibshirani. 1993. An introdu ction to the bootstrap. Chapman and Hall. NY, NY, USA. Estado de la Nacion. 2006. Duodecimo inform e estado de la nacion en desarrollo humano sostenible. Programa Estado de la N acion, San Jose, Costa Rica. (ISBN: 9968-806-374).
169 Ellison, A.M. 2004. Wetlands of Central America. Wetlands Ecology and Management 12: 355. Farley, J., A. Aquino, A. Daniels, D. Lee, and A. Krause. In press. Global mechanisms for sustaining and enhancing PES syst ems. Ecological Economics. Ferguson, B. G., J. Vandermeer, H. Morales, and D.M. Griffith. 2003. Post-agricultural succession in El Peten, Guatemala. Conservation Biology 17: 818-828. Ferraro, P.J. 2001. Global habitat protection: Li mitations of development interventions and a role for conservation performance paymen ts. Conservation Biology 15(4): 990-1000. Fielding, A.H., and J.F. Bell. 1997. A review of methods for the assessmen t of prediction errors in conservation presence/absence models. Environmental Conserva tion 24(1): 38-49. Flather, C.H., and M. Bevers. 2002. Patchy reac tion diffusion and populat ion abundance: the relative importance of habitat amount and arrangement. The American Naturalist 159: 40-56. Forester, D.J., and G.E. Machlis. 1996. M odeling human factors that affect the loss of biodiversity. Conservation Biology 10(4): 1253-1263. Gibbs, J.P. 2000. Wetland loss and biodiversit y conservation. Conservation Biology 14(1): 314-317. Foster D.R. 1992. Land-use history (1730-1990) and vegetation dynamics in central NewEngland, USA. Journal of Ecology 80:753-772. Garca, R. 1996. Propuesta tecnica de ordenami ento territorial con fines de conservacion de biodiversidad: protyecto GRUA S. Ministerio del Ambien te y Energa (MINAE) and sistema Nacional de Areas de Conservacion (SINAC), Costa Rica. Godoy, R. A. 1992. Determinants of sma llholder commercial tree cultivation. World Development 20: 713-725. Geist H.J., E.F. Lambin. 2002. Proximate causes and underlying driving forces of tropical deforestation. Bioscience 52:143-150. Grainger A. 1995. The forest transition: An alternative approach. Area 27:242-251. Green, G.M., C.M. Schweik, and J.C. Ra ndolph. 2005. Retrieving Land-Cover Change Information from Landsat Satellite Im ages by Minimizing Other Sources of Reflectance Variability. Pages 131 in E.F. Moran, E. Ostrom, editors. Seeing the Forest and the Trees: Human-Environment In teractions in Forest Ecosystems. MIT Press, MA, USA.
170 Grieg-Gran, M., I. Porras, and S. Wunder. 2005. How can market mechanisms for forest environmental services help the poor? Prel iminary lessons from Latin America. World Development 33: 1511-1527. Haig, S.M., D.W. Mehlman, and L.W. Oring. 1998. Avian movement and wetland connectivity in landscape conservation. Cons ervation Biology 12(4): 749-758. Harris, M.B., W. Tomas, G. Mour ao, C.J. Da Silva, E. Guimaraes, F. Sonoda, and E. Fachim. 2005. Safeguarding the Pantanal wetlands: threats and conserva tion initiatives. Conservation Biology 19(3): 714-720. Hey, D.L., and N.S. Philippi. 1995. Flood reduc tion through wetland restoration: the Upper Mississippi River Basin as a case history. Restoration Ecology 3: 4-17. Hoeting, J.A., R.A. Davis, A.A. Merton, and S.E. Thompson. 2006. Model selection for geostatistical models. Ecologica l Applications 16(1): 87-98. Holdridge, L.R., 1971. Forest environments in tropi cal life zones; a pilot study. Pergamon Press, NY, NY, USA. Huhta, E., J. Jokimaki, and P. Helle. 1998. Predation on artificial nests in a forest dominated landscapethe effects of nest type, patch size and edge structure. Ecography 21(5): 464-471. Hunter, J.R. 1994. Is Costa Rica truly conser vation-minded. Conservation Biology 8: 592-595. Husson. A., J. Font, H. Jeanjean, C. Miquel, H. Puig, and C. Solier. 1995. Study of Forest/Nonforest Typology of Fragmentation. TREES Series B, Research Report No. 2. European Commission, EUR 16291 EN. Huston, M.A. 2005. The three phases of land-us e change: implications for biodiversity. Ecological Applicat ions 15(6): 1864-1878. Irwin, E.G., and J. Geoghegan. 2001. Theory, da ta, methods: developing spatially explicit economic models of land use change. Agri culture, Ecosystems and Environment 85: 723. James P.E. 1959. Latin America. Odyssey Press, NY, NY, USA. Janzen D.H. 2002. Tropical dry forest: area de conservacion de Guanacaste, northwestern Costa Rica. In M.R. Perrow, and A. J. Davy, editors. Handbook of Ecological Restoration, Volume 2, Restor ation in Practice. Cambridge University Press, Cambridge, UK.
171 Jeng, H. and Y. Hong. 2005. Assessment of a natural wetland for use in wastewater remediation. Environmental Mon itoring and Assessment 111(1-3): 113-131. Jensen, J.R. 2000. Remote Sensing of the E nvironment: An Herat Resource Perspective. Prentice Hall. Upper Saddle River, NJ, USA. Jensen, J. R., 1996. Introductory Digital Image Processing: A Remote Sensing Perspective. Prentice Hall, NJ, USA. Jimenez, J.A., E. Gonzalez, and J. Cal vo. 2003. Recomendaciones tecanicas para la restauracion hidrologica del Parque Nacional Palo Verde: casos humedal Palo Verde y La Bocana. http://www.ots.ac.cr/en/paloverde/docs/pv_bocana.pdf Johnson, N.L. 2001. Tierra y libertad: will tenu re reform improve productivity in Mexicos ejido agriculture? Economic Developm ent and Cultural Change 49(2): 291-309. Junk, W. 2002. Long-term environmental tren ds and the future of tropical wetlands. Environmental Conservation 29(4): 414-435. Killion, T. W., J. A. Sabloff, G. Tourtellot, N. P. Dunning. 1989. Intensive Surface Collection of Residential Clusters at Terminal Classi c Sayil, Yucatan, Mexi co. Journal of Field Archaeology 16: 273-294. Kleinn C., L. Corrales, D. Morales. 2002. Forest area in Costa Rica: A comparative study of tropical forest cover estimates over time. Environmental monitoring and assessment 73: 17-40. Klein, E., E.E. Berg, and R. Dial. 2005. Wetland drying and succession across the Kenai Peninsula Lowlands, south-centr al Alaska. Canadian Journa l of Forest Resources 35: 1931-1941. Klepeis, P., C. Vance. 2003. Neoliberal policy and deforestation in so utheastern Mexico: an assessment of the PROCAMPO program. Economic Geography 79: 221. Klepeis, P. 2003. Development policies and trop ical deforestation in the southern Yucatan peninsula: centralized and decentralized approaches. Land Degredation and Development 14: 541-561. Klepeis, P. 2004. Forest extraction to theme pa rks: the modern history of land change. Pages 39-62 in B. L. Turner II, J. Geoghegan, D. Foster, editors. Integrated Land-Change Science and Tropical Deforestation in the S outhern Yucatn: Final Frontiers. Oxford Geographical and Environmental Stud ies. Clarendon Press, Oxford, UK. Klooster, D. 2003. Forest transitions in Mexi co: Institutions and fo rests in a globalized countryside. Professional Geographer 55: 227-237.
172 Koneff, M.D., and J.A. Royle. 2004. Modeling wetland change along the Un ited States Atlantic Coast. Ecological Modelling 177: 41-59. Koop G, and L. Tole. 1999. Is there an environm ental Kuznets curve for deforestation? Journal of Development Economics 58: 231-244. Kosoy, N., M. Martinez-Tuna, R. Muradian, an d J. Martinez-Alier. 2007. Payments for environmental services in watersheds: Insi ghts from a comparative study of three cases in Central America. Ecologi cal Economics 61(2-3): 446-455. Krueger, R.C., B. L. Kerans, E. R. Vincent, and C. Rasmussen. 2006. Risk of Myxobolus cerebralis infection to rainbow trout in the Madison River, Montana, USA. Ecological Applications 16: 770-783. Lambin E.F. 1997. Modelling and monitoring la nd-cover change processes in tropical regions. Progress in Physical Geography 21: 375-393. Lambin, E.F., and H.J. Geist. 2001. Global land -use and land-cover change: what have we learned so far? Global Change Newsletter. http://www.geo.ucl.ac.be/LUCC/pdf/Pages%20from%20NL%2046.pdf Lam bin E.F., B.L. Turner, and H.J. Geist. 2001. The causes of land-us e and land-cover change: moving beyond the myths. Global Environmental Change 11: 261-269. Lambin, E.F., B.L. Turner II, H. Geist, S. Agbola, A. Angelsen, J.W. Bruce, O. Coomes, R. Dirzo, G. Fischer, C. Folke, P.S. George K. Homewood, J. Imbernon, R. Leemans, X. Li,, E.F. Moran, M. Mortimore, P.S. Ramakrishnan, J.F. Richards, H. Sknes, W. Steffen, G.D. Stone, U. Svedin, T. Veldkamp, C. Vogel, and J. Xu. 2001. Our emerging understanding of the causes of land-use and -cover change. Global Environmental Change 11: 261. Landell-Mills, N., and I.T. Porras. 2002. Silver bullet or fools gold? A global review of markets for forest environmental services and their impact on th e poor. International Institute for Environment and Development, London. Lara, S., T. Barry, and P. Simonson. 1995. Inci de Costa RicaThe e ssential guide to its politics, economy, society and environment. Resource Center Press. Albuquerque, New Mexico, USA. Latacz-Lohmann, U., C. van der Hamsvoort. 1997. Auctioning conser vation contracts: A theoretical analysis and an application. American Journal of Agricultural Economics, 79(2): 407-418. Latacz-Lohmann, U., S. Schilizzi. 2005. Auctions for conservati on contracts: A review of the theoretical and empiri cal literature. Report to the Sc ottish Executive Environment and Rural Affairs Department. Project No. UKL/001/05.
173 Lawrence, D., H. Vester, D. Prez-Salicrup, R. Eastman, B. L. Turner II, and J. Geoghegan. 2004. Integrated Analysis of Ecosystem Interactions with Land-Use Change: The Southern Yucatn Peninsular Region. Pages 277-292 in R. DeFries, G. Asner, and R. Houghton, editors. Ecosystem Interactions with Land Use Change. American Geophysical Union, Washington, D.C., USA. Lawrence, D.C., R. Peart, and M. Leighton. 1998. The impact of sh ifting cultivation on a rainforest landscape in West Kalimantan: spatial and temporal dynamics. Landscape Ecology 13: 135-148. Lawrence, D., and D. Foster. 2002. Changes in forest biomass, litter dynamics and soils following shifting cultivation in southern Mexico: An overv iew. Interciencia 27: 400408. Lawrence, D., and D. Foster. 2004. Recovery of Nutrient Cycling and Ecosystem Properties following Swidden Cultivation: Regional and Stand-Level Constraints. Pages 81-104 in B. L. Turner II, J. Geoghegan, D. Foster, editors. Integrated Land-Change Science and Tropical Deforestation in the Southern Yu catn: Final Frontiers, Clarendon Press, Oxford, UK. Legendre, P., and L. Legendre. 1998. Numerical Ecology (2nd Ed.). Elsevier, NY, NY, USA. Lemly, A.D., R.T. Kingsford, and J.R. Thompso n. 2000. Irrigated agriculture and wildlife conservation: conflict on a global scale. Environmental Management 25(5): 485-512. Leonard H.J. 1986. Recursos naturals y desa rrollo economico en America Central: un perfil ambiental regional. Seria tecni ca, informe tecnico/CATIE, No. 127. Liu, A.J., and G.N. Cameron. 2001. Analysis of landscape patterns in coastal wetlands of Galveston Bay, Texas, USA. Landscape Ecology 16(7): 581-595. Lutz, E., and H. Daly. 1991. In centives, regulations and sustaina ble land use in Costa Rica. Environmental and Resource Economics 1: 179-194. Lutz, W., L. Prieto, and W. Sanderson. 2000. Population, Development and Environment on the Yucatan Peninsula. International Institute for Applied Systems Analysis. Research Report RR-00-14. Marsik, M., and P. Waylen. 2006. An applicati on of the distributed hyd rologic model CASC2D to a tropical montane watershed. Journal of Hydrology 330: 481-495. Mas, J.F., H. Puig, J.L. Palacio, and A. Sosa -Lopez. 2004. Modelling deforestation using GIS and artificial neural networks. Environm ental Modeling and Soft ware 19(5): 461-471.
174 Mas, J.F. 2005. Assessing protected area effectiveness using surrounding (buffer) areas environmentally similar to the target area. Environmental Monitoring and Assessment 105(1-3): 69-80. Mateo-Varga, J. 2001. Caracteristicas generale s de la cuenca del Rio Tempisque. Pages 32-72 in J.A. Jimnez and E. Gonzalez, editors. La Cuenca del Rio Tempisque, Perspectivas para un Manejo Integrado. Organizacin para Estudios Tropicales, San Jose, Costa Rica. Mather A.S. 1992. The Forest Transition. Area 24: 367-379. Mather A.S., and C.L. Needle. 1998. The fore st transition: a theoretical basis. Area 30: 117124. Mather A.S., J. Fairbairn, and C.L. Needle. 1999. The course and drivers of the forest transition: the case of France. J ournal of Rural Studies 15: 65-90. McCoy, M.B., and J.M. Rodriguez. 1994. Catta il (Typha dominguensis) eradication methods in the restoration of a tropical, seasonal, freshwat er marsh. In W.J. Mitsch, editor. Global Wetlands: Old World and New. Elsevier, NY, NY, USA. McGarigal, K., S. Cushman, and S. Stafford. 2000. Multivariate Statistics for Wildlife and Ecology Research. Springer-Verlag, York, PA, USA. Meijerink, A., A. Gieske, and Z. Vekerdy. 2005. Surface energy balance using satellite data for the water balance of a traditional irrigatio n-wetland system in SW Iran. Irrigation and Drainage Systems 19(1): 89-105. Mertens, B. and E.F. Lambin. 1997. Spatial mo deling of deforestation in southern Cameroon. Applied Geography 17(2): 143-162. Mertens, B. and E.F. Lambin. 2000. Land cover change trajectories in southern Cameroon. Annals of the Association of Am erican Geographers 90(3): 467-494. Meyer W.B., and B.L. Turner. 1992. Human-Population Growth and Global Land-use Cover Change. Annual Review of Ecology and Systematics 23: 39-61. Millennium Ecosystem Assessment. 2005. Ecosystems and Human Well-being: Wetlands and Water Synthesis. World Resources Institute. Washington, D.C., USA. http://www.millenniumassessment.org//proxy/Document. Miller, P. M. 1999. Effects of deforestation on seed banks in a tropi cal deciduous forest of western Mexico. Journal of Tropical Ecology 15: 179-188.
175 Miranda, M., C. Dieperink, P. Glasbergen. 2006. Costa Rican environmental service payments: the use of a financial instrument in partic ipatory forest management. Environmental Management 38: 562-571. Mitsch, W.J. 2005. Applying science to conserva tion and restoration of the worlds wetlands. Water Science & Technology 51(8): 13-26. Mitsch, W.J., and J.G. Gosselink. 2000. The va lue of wetlands: importance of scale and landscape setting. Ecological Economics 35: 25-33. Mitra, S., R. Wassman, and P.L. Vlek. 2005. An appraisal of globa l wetland area and its organic carbon stock. Curre nt Science 88(1): 25-35. Moran, E.F. 2004. Inferring the Behavior of H ouseholds from Remotely Sensed Changes in Land Cover: Current Methods and Future Di rections. Pages 23-47 in M.F. Goodchild editor. Spatially Integrated Social Scien ce. Oxford University Press, Oxford, UK. Mulder, M., and P. Coppolillo. 2005. Conserva tion: Linking Ecology, Economics and Culture. Princeton University Press, Princeton, NJ, USA. Murphy, P. G., and A. E. Lugo. 1986. Ecology of Tropical Dry Forest Annual Review of Ecology and Systematics 17: 67-88. Mustard J.F., R.S. Defries, T. Fisher, and E. F. Moran. 2004. Land-use and land-cover change pathways and impacts. Pages 411-429 in G. Gu tma, A.C. Janetos, C.O. Justice, E.F. Moran, J.F. Mustard, R.R. Rindfuss, D. Skol e, B.L. Turner II, and M.A. Cochrane, editors. Land change science: observing, m onitoring and understanding trajectories of change on the earths surface. Kluwer, Dordrecht, The Netherlands. Myers, N. 1993. Tropical Forests the Main Deforestation Fronts. Environmental Conservation 20: 9-16. Nagelkerke, N.J. 1991. A note on a general defi nition of the coefficient of determination. Biometrika 78(3): 691-692. Nagendra, H. 2001. Using remote sensing to assess biodiversity. Inte rnational Journal of Remote Sensing 22: 2377-2400. Naidoo, R., and T.H. Ricketts. 2006. Mapping the economic costs and benefits of conservation. PLOS Biology 4(11): 2153-2164. Neupane, R. P., K. R. Sharma, and G. B. Thapa. 2002. Adoption of agrofore stry in the hills of Nepal: a logistic regression analysis. Agricultural Systems 72: 177-196. Norusis, M. 1990. SPSS Advanced Statisti cs Users Guide, SPSS Inc., Chicago, IL.
176 Notman, E., Gorchov, D.L., 2001. Variation in post-dis persal seed predation in mature Peruvian lowland tropical forest and fallow agricultural sites. Biotropica 33, 621-636. Ochoa-Gaona, S., 2001. Traditional land-use systems and patterns of forest fragmentation in the highlands of Chiapas, Mexico. Environmental Management 27: 571-586. Ogawa, H., and J.W. Male. 1986. Simulating the flood mitigation role of wetlands. ASCE Journal of Water Resources Pl anning and Management 112: 114-127. OGrady, M.A. 2006. Costa Rican poverty fighte r. The Wall Street Journal. May 5, 2006. Olson, M. 1965. The Logic of Collective Action. Harvard University Press, Cambridge, MA. Oviedo, E. 2006. Estado paga poco por conservar bosques. La Nacion, 20 de noviembre. Ozesmi, S.L., C.O. Tan, and U. Ozesmi. 2006. Methodological issues in building, training, and testing artificial neural networks in ecologi cal applications. Ecol ogical Modelling 195: 83-93. Ozesmi, S.L., and M.E. Bauer. 2002. Satellit e remote sensing of wetlands. Wetlands Ecology and Management 10(5): 381-402. Pagiola, S., A. Arcenas, and G. Platais. 2005a. Can payments for enviro nmental services help reduce poverty? An exploration of the issues and the evidence to date from Latin America. World Development 33: 237-253. Pagiola, S., P. Agostini, J. Gobbi, C. de Haa n, M. Ibrahim, E. Murgueitio, E. Ramirez, M. Rosales, and J. P. Ruiz. 2005b. Paying for biodiversity conservation services Experience in Colombia, Costa Rica, and Nicaragua. Mountain Research and Development 25: 206-211. Pagiola, S. 2006. Payments for environmental services in Costa Rica. MPRA Paper No. 2010. Revised version of a paper presented at the ZEF-CIFOR workshop on Payments for environmental services: Methods and desi gn in developing and developed countries, Titisee, Germany, July 15-18 2005. Perez-Salicrup, D. R., S. Schnitz er, and F. E. Putz. 2004. Co mmunity ecology and management of lianas. Forest Ecology and Management 190: 1-2. Perz, S.G. 2007. Grand theory and context-spec ificity in the study of forest dynamics: Forest transition theory and other directions Professional Geographer 59: 105-114. Peters, G. 2001. La cuenca del Tempisque: una perspective historica. Pages 1-21 in J. Jimenez and E. Gonzalez, editors. La Cuenca del Rio Tempisque: Perspectivas para un Manejo Integrado. Organizacion para Estudios Tropicales, San Jose, Costa Rica.
177 Pijanowski, B.C., D.G. Brown, B. A. Shellito, and G.A. Manik. 2002. Using neural networks and GIS to forecast land use changes: a land transformation model. Computers, Environment and Urban Systems 26: 553-575. Plaza, C.R. 2000. Gender roles, inheritance patt erns, and female access to land in an ejidal community in Veracruz, Mexico. Pages 161-173 in A. Zoomers, G. van der Haar, editors. Current land policy in Latin Am erica: regulating land tenure under neoliberalism. Royal Tropical Institut e, Amsterdam, The Netherlands. Polasky, S., J.D. Camm, and B. Garber-Yonts. 2001. Selecting biol ogical reserves costeffectively: An application to terrestrial vertebrate conservation in Oregon. Land Economics 77(1): 68-78. Pontius, R.G., and L.C. Schneider. 2001. La nd cover change model validation by an ROC method for the Ipswich watershed, Massachus etts, USA. Agriculture, Ecosystem and Environment 85: 239-248. Pringle, C.M. 2001. Hydrologic connectivity a nd the management of biological reserves: a global perspective. Ecological Applications 11(4): 981-998. Pringle, C. 2003. What is hydrologic conneci tivty and why is it ecologically important? Hydrological Processes 17(13): 2685-2689. Ramankutty N, H.K. Gibbs, and F. Achard. 2007. Challenges to estimating carbon emissions from tropical deforestation. Gl obal Change Biology 13: 51-66. Ramankutty, N., L. Graumlich, F. Achard, D. Alves, A. Chhabra, R. DeFries, J. Foley, H. Geist, R. Houghton, K. Goldewijk, E.F. Lambin, A. Millington, K. Rasmussen, R. Reid, and B.L. Turner II. 2006. Global Land Cover Change: Recent Progress, Remaining Challenges. Pages 9-40 in E. Lambin, and H. Geist, editors. Land Use and Land Cover Change: Local Processes, Global Impact s. Springer Verlag, NY, NY, USA. Randall, A. 1993. The Problem of Market Failure In: Economics of the Environment. R. Dorfman and N. Dorfman, eds. Norton, New York. Rodriguez, J. P., Beard, T. D., Bennett, E. M., G. S. Cumming, S. J. Cork, J. Agard, A. P. Dobson, and G. D. Peterson. 2006. Trade-offs across space, time, and ecosyst em services. Ecology and Society 11: 28. Ravan, S., A.M. Dixit, and V.B. Mathur. 2005. Spatial analysis for iden tification and evaluation of forested corridors between two protected areas in Central India. Current Science 88(9): 1441-1448. Read, L., and D. Lawrence. 2003. Recovery of biomass following shifting cultivation in dry tropical forests of the Yucatan. Ec ological Applications 13: 85-97.
178 Reyes, E., M.L. White, J.F. Martin, G.P. Kemp, J.W. Day, and V. Aravamuthan. 2000. Landscape modeling of coastal habitat change in the Mississippi delta. Ecology 81: 2331-2349. Rogerson, P.A. 2001. Statistical Methods for Ge ography. Sage Publications. Thousand Oaks, CA, USA. Rojas, M., and B. Aylward. 2003. What are we learning from experiences with markets for environmental services in Costa Rica? A review and critique of the literature. International Institute for Environment and Development, London, UK. Rouget M, D.M. Richardson, and R.M. Cowling. 2003. The current configuration of protected areas in the Cape Floristic Region, South Africa reservatio n bias and representation of biodiversity patterns and processes. Biological Conservation 112: 129-145. Rudel T.K., O.T. Coomes, E. Moran. 2005. Fo rest transitions: towa rds a global understanding of land use change. Global Environmental Change 15:23-31. Rudel, T.K., O. Coomes, E.F. Moran, A. Angelsen, F. Achard, E. Lambin, and J. Xu. 2005. Forest transitions: towards an understa nding of global land use change. Global Environmental Change 15(1): 23-31. Salzman, J., and J.B. Ruhl. 2002. Paying to prot ect watershed services: Wetland banking in the United States. In S. Pagiola, and J. Bishop, editors. Selling Forest Ecosystem Services: Market-based mechanisms for c onservation and development. Earthscan, London, UK. Salzman, J. 2005. Creating markets for ecosystem services: Notes from the field. New York University Law Review 80: 870. Samuelson, P.A. 1954. The pure theory of public expenditure. The review of Economics and Statistics 36 (4): 387-389. Sanchez-Azofeifa, G.A., G.C. Daily, A. Pfaff, and C. Busch. 2003. Integrity and isolation of Costa Ricas nacional parks and biological re serves: examining the dynamics of landcover change. Biological Conservation 109: 123-135. Sanchez-Azofeifa G.A., A. Pfaff, J.A. Roba lino, et al. 2007. Costa Ricas payments for environmental services program: intention, implementation and impact. Conservation Biology 21: 1165-1173. Sanchez-Azofeifa, G., R. C. Harriss, and D. L. Skole. 2001. Deforestation in Costa Rica: A quantitative analysis using remote se nsing imagery. Biotropica 33: 378-384. Schaafsma, W. and G.N. van Vark. 1979. Cl assification and discrimination problems with applications. Part IIa. Statistica Neerlandica 33: 91-126.
179 Semlitsch, R.D. 2002. Critical elements for bi ologically based recovery plans of aquaticbreeding amphibians. Conservation Biology 16(3): 619-629. Sierra, R., E. Russman. 2006. On the efficiency of environmental service payments: A forest conservation assessment in the Osa Peninsula, Costa Rica. Ecological Economics 59(1): 131-141. Sills, E., G. Hartshorn, P. Ferraro, and B. Sp ergel. 2005. Evaluation of the World Bank-GEF ecomarkets project in Costa Rica. Blue Ribbon Panel Report. North Carolina State University, Durham, NC, USA. Skole D.L., W.H. Chomentowski, W.A. Salas et al. 1994. Physical and Human Dimensions of Deforestation in Amazonia. Bioscience 44: 314-322. Sloan, S. 2007. Fewer people may not mean more forest for Latin American forest frontiers. Biotropica 39(4): 443-446. Southworth, J. 2004. An assessment of Landsat TM band 6 thermal data for analysing land cover in tropical dry forest regions. International Journal of Remote Sensing 25: 689706. Southworth J., D.K. Munroe, and H. Nagendr a. 2004. Land cover change and landscape fragmentationcomparing the utili ty of continuous and discre te analyses for a western Honduras region. Agriculture, Ecosystems and Environment, 101(2-3): 185-205. Southworth J, H. Nagendra, L.A. Carlson et al 2004. Assessing the impact of Celaque National Park on forest fragmentation in west ern Honduras. Applied Geography 24: 303-322. Staaland H., O. Holand, C. Nellemann, et al. 1998. Time scale for forest regrowth: Abandoned grazing and agricultural areas in southern Norway. Ambio 27: 456-460. Steed, B. 2003. Completing the mosaic: the cons ervation of private lands in Costa Rica. Journal of Land, Resources, a nd Environmental Law 173: 173-180. Stern P.C., O.R. Young, and D. Druckman. 1992. Global environmental ch ange: understanding the human dimensions. National Acad emy Press, Washington, D.C., USA. Stoms, D.M., K.M. Chomitz, and F.W. Davis. 2004. TAMARIN: a landscape framework for evaluating economic incentives for rainfo rest restoration. Landscape and Urban Planning 68: 95-108. Stoneham, G., V. Chaudri, A. Ha, and L. Strappazon. 2003. Auctions for conservation contracts: An empirical examination of Vict orias BushTender trial. Australian Journal of Agricultural and Reso urce Economics 47(4): 477-500.
180 Szentandrasi, S., S. Polasky, R. Berrens, J. Leonard. 1995. Conserving biological diversity and the conservation reserve program. Growth and Change 26: 383-404. Summerville, K. S., C. J. Conoan, and R. M. St eichen. 2006. Species traits as predictors of lepidopteran composition in restored and remnant tallgrass prairies. Ecological Applications 16: 891-900. Taylor J.E. 1980. Peripheral capitalism and rural-urban migration: a study of population movements in Costa Rica. Latin Am erican Perspectives 7(2-3): 75-90. Terborgh, J. 1999. Requiem for Nature. Island Press. Washington, D.C., USA. Teran, S., and C. H. Rasmussen. 1995. Genetic Diversity and Agricult ural Strategy in 16thCentury and Present-Day Yucatecan Milpa Agriculture. Biodiversity and Conservation 4: 363-381. Tietenberg, T.H. 1989. Economic instruments for environmental regulation. Oxford Review of Economic Policy 6(1): 17-33. Tiner, R.W. 2005. Assessing cumulatie loss of wetland functions in the Anticoke River Watershed using enhanced national wetlands inventory data. Wetla nds 25(2): 405-419. Toner, M., and P. Keddy. 1997. River hydrology a nd riparian wetlands: a predictive model for ecological assembly. Ecologica l Applications 7(1): 236-246. Trebitz, A.S., J.A. Morrice, D.L. Taylor, R.L. Anderson, C.W. West and J.R. Kelly. 2005. Hydromorphic determinants of aquatic habita t variability in Lake Superior coastal wetlands. Wetlands 25(3): 505-519. Turner, M.G., R.H. Gardner, R.V. O'Neill. 1995. Ecological Dynamics at Broad Scales. BioScience 45: S29-S35. Turner II, B.L., J. Geoghegan, D. Foster. 2004. Integrated Land-change Science and Tropical Deforestation in the Southern Yucatn: Fi nal Frontiers. Clarendon Press, Oxford, UK. Universidad de Costa Rica. 1976. La poblacion de Costa Rica. Instituto de Investigaciones Sociales. Editorial Universidad de Costa Rica: 9091. San Jose, Costa Rica. Unruh, J. D. 1988. Ecological Aspects of Site Recovery under Swidden-Fallow Management in the Peruvian Amazon. Agroforestry Systems 7: 161-184. Unruh, J.D. 1990. Iterative increase of economi c tree species in manage d swidden-fallows of the Amazon. Agroforestry Systems 11(2): 175-197. Urban, D.L. 2005. Modelling ecological proces ses across scales. Ecology 86(8): 1996-2006.
181 Vance, C. and J. Geoghegan. 2004. Modeling the determinants of semi-subsistance and commercial land uses in an agricultural fr ontier of southern Mexico: a switching regression approach. International Re gional Science Revi ew 27(3): 326-347. Vogeler, I.K. 1970. Frontier settlements in south-eastern Campeche: A report for the 1970 National Geographic Society-Tulane Univer sity Archaelogical Project at Becan, Campeche, Mexico. Vogeler I.K. 1976. The Dependency Model A pplied to a Mexican tropical frontier region. Journal of Tropical Geography 43: 63-68. Vorosmarty, C.J., and D. Sahagian. 2000. Anth ropogenic disturbance of the terrestrial water cycle. BioScience 50(9): 753-765. Walker, R.T. 1999. The Structure of Uncultiv ated Wilderness: Land Use beyond the Extensive Margin. Journal of Regional Science 39(2): 387-410. White, A. and A. Martin. 2002. Who owns the worl ds forests? Forest tenure and public forests in transition. Forest Tre nds. Washington, D.C., USA. Watson, V., S. Cervantes, C. Castro, L. Mora, M. Solis, I.T. Porras, and B. Cornejo. 1998. Making Space for Better Forestry: Costa Ri ca Country Study. Policy that Works for Forests and People Series No. 6. International Institute for Environment and Development, London, UK. White, D., P.G. Minotti, M.J. Barczak, J.C. Si fneos, K.E. Freemark, M.V. Santelmann, C.F. Steinitz, A. Ross Kiester, and E.M. Preston. 1997. Assessing risk to biodiversity from future landscape change. C onservation Biology 11: 349-360. World Bank. 1995. Mexico resource conservation and forest sector review. Natural Resoruces & Rural Poverty Operations Division C ountry Department II Latin America & the Caribbean Regional Office, Washington, D.C., USA. World Bank. 2000. Costa Rica Ecomarkets Project. Project Appraisal. Environmentally and Socially Sustainable Development, Latin America and the Caribbean Regional Office, Washginton, D.C., USA. World Bank. 2006. Scaling up and mainstreaming payment for environmental services in Costa Rica. Project Appraisal. Environmentall y and Socially Sustainable Development, Latin America and the Caribbean Regional Office, Washginton, D.C., USA. World Bank. 2007. Key Development Da ta & Statistics. Available at http://web.worldbank.org/WBSI TE/EXTERNAL/DATASTATISTICS/ access ed May 2, 2007.
182 Wunder, S. 2005. Payments for environmen tal services: Some nuts and bolts. CIFOR Occasional Paper No. 42. Center for Inte rnational Forestry Research, Jakarta, Indonesia. Wunder, S. 2007. The Efficiency of Payments for Environmental Se rvices in Tropical Conservation. Conserva tion Biology, 21(1): 48-58. Wnscher, T., S. Engel, and S. Wunder. 2007. Spatial targeting of paym ents for environmental services: A tool for boosting conservation benefits. Presented at the Workshop on Carbon Sequestration in Agriculture and Fo restry conference, Thesaloniki, Greece, June 27, 2007. Zbinden, S., and D. R. Lee. 2005. Paying fo r environmental servic es: An analysis of participation in Costa Rica's PSA pr ogram. World Development 33: 255-272. Zedler, J.B., and S. Kercher. 2005. Wetland res ources: status, trends, ecosystem services and restorability. Annual Review of Environmental Resources 30: 39-74.
183 BIOGRAPHICAL SKETCH Amy Daniels grew up in Apalachicola, FL, graduating from Apalachicola High School in 1995. She went on to Wesleyan College to earn her B.A. in biology in 1999. In the years between college and graduate school, Amy worked as a resource management biologist for the Apalachicola National Estuarine Research Reserve (ANERR) in Florida and as a counselor for a non-profit AIDS-care foundation in New York. She also travel ed independently throughout Central America, learning Spanish and volunteering for a grassroots conser vation organization in Guatemala. During her masters degree Amy received a Fulbright Fellowship to investigate landscape transformation in northwest Costa Rica and a Coca-cola World Citizenship Fellowship to work as an agricultural development consultant for a United States Agency for International Development-funded project in El Salvador. Amy graduated w ith an M.S. in Interdisciplinary Ecology in May 2004. Deciding that the graduate-s chool lifestyle was too great to give up, Amy began her PhD the following year. Amy had the pr ivilege of independently teaching Physical Geography two semesters during her doctoral studie s, a rewarding experience that helped her discover a great love for teaching. After gra duating Amy received a Presidential Management Fellowship to work in the Global Change Progr am of the United States Forest Services Research and Development Office. Amy is married to a wonderful attorney-turne d-diplomat, Paul Ghiotto. Together they enjoy traveling, canoeing, spending time with family and friends, and tryi ng just about anything new.