LAND SUITABILITY FOR VITICULTURE:
A GEOGRAPHIC INFORMATION SYSTEM-DERIVED METHOD FOR
A MEDITERRANEAN TYPE CLIMATE IN CALIFORNIA
RUSSELL L. WATKINS
A DISSERTATION PRESENTED TO THE GRADUATE SCHOOL
OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT
OF THE REQUIREMENTS FOR THE DEGREE OF
DOCTOR OF PHILOSOPHY
UNIVERSITY OF FLORIDA
Many thanks to the numerous people who have contributed to this study
in one form or another. Special thanks to my Chair, Dr. Cesar Caviedes, for
teaching me the value of a holistic, experiential approach to science, and to my
Co-chair, Dr. Joann Mossa, for her unflagging support and emphasis on the
details necessary for a successful research project. To Dr. Timothy Fik of
course goes my gratitude for his patience with and insight into my statistical
education, and most importantly, his friendship. Thanks to Dr. Scot Smith for
his timely advice and continual support of my doctoral efforts, and to Dr. Harry
Paul for his patience with my applied, non-philosophical approach to a topic we
are both intensely interested in.
Mention must be made of a few other special individuals. Dr. Peter
Waylen, who threw me a lifeline when I was awash in a sea of data, and to my
"External Research Committee", Thomas Olmsted, Richard Browning, and John
Newton, without whose continual encouragement and research support, both
liquid and financial, this study would not have come about. Finally, a wish of
gratitude to parents, for whom no additional words are needed.
TABLE OF CONTENTS
LIST OF TABLES ................
LIST OF FIGURES ...............
1 INTRODUCTION .................
Research Problem ................
Setting of Study ..................
2 RELEVANCE OF RESEARCH /
Land Evaluation Literature ...
Implications of GIS and Relevant
Viticultural Literature ....... .
. . .
. . .
3 METHODOLOGY ...................
Description of Technique and Process ...
Selection of Approach ...............
Data Collection and Description ........
Data Input ........................
Data Manipulation ..................
Statistical Methods ..................
4 RESULTS ........................
Climate Description .................
Vineyard and Sample Area Comparison ...
Statistical Results ...................
Interpretation of Results ..............
5 CONCLUSIONS ....................................
Spatial Accuracy and Resolution Limitations
Strengths of the Method ..............
Opportunities for Future Research .......
APPENDIX ................................ ..... 113
LIST OF TABLES
ENVIRONMENTAL FACTORS AND KEY RESULTS
OF PREVIOUS VITICULTURAL LAND
SUITABILITY STUDIES ............................ 31
SUMMARY OF SOIL VARIABLES COLLECTED
FROM THE VITICULTURAL LITERATURE .............. 34
VARIABLES USED IN STUDY BY CLASS .............. 35
SUMMARY OF CORRELATION MATRIX RESULTS ....... 87
MAJOR FACTOR NAMES .......................... 90
SUMMARY OF STAGE ONE ........................ 93
SUMMARY OF STAGE TWO ........................ 95
SUMMARY OF STAGE THREE ...................... 97
LIST OF FIGURES
STUDY AREA LOCATION ..............
ZINFANDEL VINEYARD LOCATION ........
GENERAL GEOLOGY .................
GENERAL SOILS ....................
DATABASE CONSTRUCTION PROCEDURES
INFORMATION MODEL CONSTRUCTION
SUMMARY OF STATISTICAL PROCEDURES
MEAN MONTHLY PRECIPITATION ........
STORE LAND CAPABILITY ............
SO IL DEPTH ........................
SOIL WATER-HOLDING CAPACITY ........
SOIL RUNOFF ......................
SOIL ROOTING DEPTH ................
SOIL FERTILITY .....................
SOIL MOISTURE .....................
SO IL ACIDITY ........................
. . 9
. . 13
. . 15
. . 17
.. . 63
. .. 66
LIST OF FIGURES (continued)
FIGURE 4-10: CATION EXCHANGE CAPACITY ................... 73
FIGURE 4-11: SOIL PERMEABILITY .......................... 74
FIGURE 4-12: NATURAL SOIL DRAINAGE ...................... 76
FIGURE 4-13: SOIL SAND CONTENT .......................... 77
FIGURE 4-14: SOIL SILT CONTENT ........................... 78
FIGURE 4-15: SOIL CLAY CONTENT .......................... 79
FIGURE 4-16: SLOPE ANGLE ................................ 81
FIGURE 4-17: SLOPE ASPECT ............................... 82
Abstract of Dissertation Presented to the Graduate School
of the University of Florida in Partial Fulfillment of the
Requirements for the Degree of Doctor of Philosophy
LAND SUITABILITY FOR VITICULTURE:
A GEOGRAPHIC INFORMATION SYSTEM-DERIVED METHOD FOR
A MEDITERRANEAN TYPE CLIMATE IN CALIFORNIA
Russell L. Watkins
Chairman: Dr. Cesar Caviedes
Major Department: Geography Department
This study has applied geographic information system (GIS) techniques
as the basis for an understanding of the relationship of environmental variables
and viticultural land suitability. Spatial analysis of a viticultural region in a
Mediterranean-type climate in northeastern California, incorporating the
Shenandoah Valley and Fiddletown Approved Viticultural Areas, was
undertaken using the Arc/Info GIS software.
The research problem focuses on the development of measurable
environmental criteria based on the characteristics of existing Zinfandel vineyard
and non-vineyard areas, to evaluate new land areas for viticultural suitability.
Twelve soil and two topographic variables were compared, including Storie
Capability Index, depth, percent content of sand, silt and clay, water-holding
capacity, natural runoff, effective rooting depth, fertility, moisture, acidity, cation
exchange capacity, slope angle and slope aspect.
Factor analysis was initially used to reduce the number of variables
describing the data sets. Similarity analysis using the Mann-Whitney U test
was applied to discriminate the factor descriptions of the physical characteristics
of Zinfandel vineyards from non-vineyard areas. Upon measuring the similarity
of the data sets containing factor coefficients, no statistically significant
difference was found between the medians of the environmental variables.
Similarity analysis of the observations on the complete variable data set
identified differences at the 95% confidence level between Zinfandel vineyard
and non-vineyard areas in the medians of the slope angle, slope aspect, Storie
Index, soil depth, water-holding capacity and cation exchange capacity
variables. Effective rooting depth, natural runoff, and clay content were
significantly different at the 85% confidence level. This level represented a
clear break in the structure of the data sets. Overall, 53% of the variables can
be identified as having characteristics that are significantly dissimilar between
the Zinfandel vineyard and non-vineyard areas.
A reliance on secondary data and the inherent power and flexibility of
geographic information system techniques are the main ingredients for a
location and crop independent suitability reconnaissance. The method
developed here provides the foundation for the successful construction of a
predictive model of viticultural land suitability.
Wine is a wonderful example of a multifaceted geographic phenomenon.
The questions of origin and means of diffusion loom large in any historical
study. In addition, the regionalization of wine production areas encompasses a
spatial dimension that has both physical and cultural components.
Although the origin of grape (Vitis vinifera) cultivation, or viticulture, is
unclear, most researchers generally agree that the cultivation of vines for wine-
making originated some time before 4000 B.C. and possibly as early as
6000 B.C. The location was in the mountainous region between the Black Sea
and the Caspian Sea, on the borders of the modern states of Turkey, Syria,
Iraq, Iran and the Soviet Union (Unwin, 1991; Nuiez and Walker, 1989; de Blij,
1983). The first instances of wine production, or vinification, are subject to
conjecture only, and are thought to have been the result of natural fermentation
resulting from the interaction of wild yeasts with grapes in storage (Unwin, 1991;
de Blij, 1983; Dickenson and Salt, 1982; Stanislawski, 1975).
Numerous factors have historically influenced the geographical pattern of
viticulture and viticultural decision-making, including physical environment, local
cultural, political and economic influences. For example, the early location of
vineyards near coastlines, along rivers, and within coastal valleys indicates the
influence of physical environment through geographical accessibility and
transportation (Peacock and Williams, 1986; Panella, 1981; Drinkwater, 1983;
Weiter-Matysiak, 1985). Initially, valleys and rivers represented areas of
settlement penetration and important means of transportation (Smith, 1967).
Increased accessibility through improvements in transportation means and
routes, and changes in market patterns from export to local consumption,
contributed to the movement of viticultural activities into interior areas and
expansion around settlements, within regions environmentally suited for grape
growing (Unwin, 1991; Dickenson and Salt, 1982).
A strong identification of wine with specific regional and local
environments, and an emphasis on the study of environmental factors in the
scientific viticulture literature, has given rise to what has been referred to as
classic wine production areas such as Bordeaux, Rhine, Chianti and Napa.
This regional identity has been reinforced by the designation of grape
production areas known as DOC's (denominazione di origine controllata) in
Italy, Appellation d'Origine Control6e (AOC's) in France, and Approved
Viticultural Areas (AVA's) in the United States. Requirements such as grape
production limits and minimum percentages of specific grapes in wines have
institutionalized these identities (Stevenson, 1986; de Blij, 1985; Wasserman and
Wasserman, 1985; Bureau of Alcohol, Tobacco and Firearms, 1986).
This research explores the regional aspects of wine grape production, or
viticulture, purely from a physical geographic perspective. The analysis of the
environmental characteristics of an established wine region is undertaken in an
attempt to discern if a unique combination of physical parameters in vineyards
can be identified. Development of a geographic information system (GIS)
method to assess land suitability for current and potential viticultural regions in
Mediterranean-type climatic regions is the primary focus of the proposed
What is implied by the term suitability? After reviewing the land evaluation
literature (notable summaries include Beek, 1978; Stewart, 1968; Davidson,
1986), McRae and Burnham (1981) provide the most widely accepted definition,
by way of contrasting land capability and land suitability. Capability refers to the
potential of land areas for general uses, such as agriculture, forestry, or
rangeland. Suitability refers to the potential for use in production of a specific
crop or plant species. The premise then, is that an evaluation of the resource
base is a necessary first step in the determination of land use potential. The
natural progression of such an analysis could incorporate cultural and
economic considerations such as land tenure patterns and value, transportation
facilities, and market conditions, among others.
The approach selected in this study allows for a spatial and quantitative
analysis of environmental factors. As the following text will indicate, this
approach is in sharp contrast, with a few exceptions, to the bulk of previous
viticultural research. It integrates and builds on theories and techniques of
agricultural and regional geography, land evaluation, GIS spatial analysis and
modeling, and viticulture (Ilbery, 1985a, 1985b; Stewart, 1968; Davidson, 1986;
Berry, 1968; Bowler, 1990; Frank, et al., 1991; Lowry, 1968; Gadille, 1967;
Dickenson and Salt, 1982; de Blij, 1983; Newman, 1986; Unwin, 1991). This
integration is achieved by means of systematic geography, GIS and statistical
techniques which relate environmental factors within a viticultural land suitability
Review of the existing literature, and personal communications with
researchers in the fields of viticulture, geography, and geographic information
systems, has shown a paucity of applications of GIS in this area (de Blij, 1991;
Elliot-Fiske, 1991; Rodolfi, 1991; Unwin, 1991; Dickenson, 1991; Scienza, 1991;
Baxevanis, 1992). However, GIS techniques are beginning to be applied to
determine land suitability for selecting optimal sites for vineyards in northern
Italy (Scienza, 1992; 1990), and in the Champagne (Laville et al., 1992) and
Loire Valley (Morlat and Asselin, 1992) regions of France. These studies focus
on matching cultivars to environment, the "genotype x environment" approach
(Onokipse and Mortenson, 1988). Environmental variables, as well as chemical
components of grapes, and taste determinations are being evaluated to place
the grape within what is described as the "optimal quality environment."
Having set the stage for this research within the larger context of
viticulture and regional geography, the remainder of this chapter will address
the hypothetical aspects and methodological constraints of the chosen
approach. In addition, a description of the study area will be presented.
The research problem addressed in this study is the development of
measurable environmental criteria based on the characteristics of existing
vineyard areas to evaluate new land areas for viticultural suitability. In seeking
to fill a perceived void in the viticultural geographic literature, this analysis
highlights soil and topographic variables. As discussed in the Literature Review
(Chapter 2), much of the viticultural land evaluation research has focused
primarily on the influence of climatic criteria. In addition, the emphasis here is
placed on measurable, or quantitative criteria, as opposed to that based on
opinion, observation or folklore. A series of excellent examples of the latter are
provided by Dickenson (1989), describing generalizations in studies on
viticultural geography. In discussing the lack of scientific literature in English,
Dickenson identifies "nine axioms" generalized from empirical analyses of
European practices (Dickenson, 1989, p. 3). Included amongst these are the
role of slope, soils, climate, physiology, vineyard management, and the artwork
of the individual.
Specific objectives of this research are to
a) identify applicable environmental variables through review of previous
studies, expert interview, direct observation, and statistical correlation analysis;
b) construct a GIS data model integrating environmental data in a variety
c) conduct a statistical analysis of vineyard location, soil and
topographic variables to describe existing spatial patterns and land suitability;
d) validate the suitability of the method through its application to
existing and potential viticultural sites in the study area.
The principal hypothesis of this research is that a statistically significant
difference could exist between soil and topographic characteristics of vineyard
and non-vineyard sites within the study area. In this study, the term vineyard
refers to those sites growing predominantly Zinfandel grape types. What is
most important in this study is the development of a methodological procedure
a GIS framework, allowing for the integration of a variety of data
types, scales and formats;
a general applicability to multiple climate types, grape types, and
other crop types; and
-the use of spatial occurrence to develop quantitative measures of
similarity (correlation coefficients and factor loadings).
The last part requires some consideration. Rather than use results from
field trials or lab experiments to establish evaluation criteria, variables have been
drawn from the literature and grower opinions. These variables are then used
to describe vineyard and non-vineyard areas. No attempt was made to
subjectively weight the influence of variables on vineyard location according to
productivity figures or expert opinion. The foundation of the method rests on
spatial analysis, defined by Goodchild (1991, p. 42) as "a set of techniques
whose results are dependent on the locations of the objects or events being
analyzed, requiring access to both the locations and the attributes of the
objects." The method was constructed and validated based solely on the
spatial location and association of variables as delimited by vineyard and non-
vineyard areas. Spatial location and association are included in a group of
spatial analysis variables termed "geographic primitives" (Meentemeyer, 1989,
Setting of Study
The actual study area boundaries are derived from the Shenandoah
Valley and Fiddletown Approved Viticultural Areas (AVA), established in
January 1983, and includes portions of Amador and El Dorado counties (see
Figure 1-1). An AVA can be defined as "a delimited grape growing region with
geographic features which set it apart from surrounding areas" (Bureau of
Alcohol, Tobacco and Firearms, 1986). These areas are established by the
Bureau of Alcohol, Tobacco and Firearms (BATF) upon petition from local
growers and wineries. The name can then be used on wine labels as an
appellation, or indication of the origin of the grapes used to produce at least
85% of that wine.
The study area was chosen for the following reasons:
it is a well-defined (both physiographically and politically), small
it falls within a Mediterranean-type climatic region;
the Zinfandel cultivar is extensively cultivated; and
grapevine management practices are largely uniform.
The discrete physiographic and political definition of the area is reflected
in its designation as an AVA and in the established history of viticulture there.
The size of the area is significant in that the necessary digital database did not
exist prior to this study, and had to be constructed. A much larger area would
have been unfeasible if a database at the same level of detail had to be
generated. The Mediterranean-type climate is representative of the most
important viticultural climatic zone worldwide, and provides the opportunity for
comparison and application of this approach in other areas. The uniformity of
EL DORADO COUNTY
FIGURE 1-1: STUDY AREA LOCATION
grape cultivar and grapevine management practices is significant in terms of
minimizing variation in location and productivity of vineyards. Common
practices include head pruning, irrigation to supplement Spring precipitation,
fertilization to counteract nitrogen and boron deficiencies, and sulfur dusting to
prevent powdery mildew. The reader is referred to Plaister (1980) and Winkler
et al. (1974) for comprehensive discussions of these management practices.
Historical Description of Study Area
The development of viticulture in California began with the arrival of the
Spanish in the mid-sixteenth century, and was sustained by Catholic priests to
satisfy the need for sacramental wine. Initial attempts in California to use the
native grapes (Vitis californica and V. girdiana) are generally acknowledged as
being a failure (Teiser and Harroun, 1983). A species of Vitis vinifera, the
"Mission" grape was widely planted thereafter. Early accounts from travelers
indicate that the physical environment was very favorable for viticulture (Teiser
and Harroun, 1983). The history of the development of the wine industry in
California is long and varied, affected by patterns of settlement and expansion,
national policy and natural hazards such as frost and pests. A comprehensive
description of this history into the early 1980s can be found in Teiser and
Harroun (1983), and de Blij (1983).
The Sierra Foothills Region is famous historically as a destination in the
Gold Rush days of California. It is also this Gold Rush that brought viticulture to
the region in the mid-nineteenth century. One of the earliest recorded
statements on viticulture mentioned the Davis Ranch in Shenandoah Valley,
which was settled in 1859 (Plaister, 1980). In 1860 it was reported that there
were nearly 500 acres of vines in Shenandoah Valley (Plaister, 1980). In the
early to mid-1860s, the wine business declined due to a lack of quality and
transportation to markets in the eastern United States (Barbour, 1984). Much
of the land in Shenandoah Valley was used to produce grain to feed livestock,
including wheat, barley and oats. Some of the area was in orchards, producing
peaches, plums, nectarines and apricots for sale to local towns (Farnham,
1984). The 1870s saw a viticultural revival due to opening of quartz mines in
the area. The major market for grapes was home winemakers in the local area
and in nearby Sacramento (Barbour, 1984). The Sierra Foothills region boasted
more wineries than Napa and Sonoma counties in 1870 (Baxevanis, 1990).
Little is known of this region in the period from 1880 to 1960 (Farnham,
1984). The area was "rediscovered" in the early 1960s by wine makers and
merchants from the San Francisco area (Barbour, 1984). Beginning in the
1970s, viticulture expanded rapidly in this region, with the establishment of
23 new vineyards in the decade of the 1980s, as estimated by Baxevanis
(1990). Within this region, Amador and El Dorado counties (see Figure 1)
provide an example of rapid and concentrated viticultural development.
In 1992 these counties accounted for 2,281 acres of wine grapes, or
approximately 0.7% of the statewide total acreage (California Agricultural
Statistics Service, 1993). The most common grapes are Zinfandel, Merlot, and
Chardonnay. Recent efforts have resulted in an increase of the acreage of
some of the "classic" Italian grapes, such as Sangiovese and Nebbiolo.
General Physical Characteristics
The study area consists of approximately 8,296 hectares (20,500 acres).
Ninety-three Zinfandel vineyards have been identified, comprising 499 hectares
(1,200 acres), 72% (67) of which are at or below 5 hectares in size. A majority
of the vineyards are found in the 400 to 500 meter (approximately 1,200 to
1,500 foot) elevation range. Figure 1-2 shows the location of the vineyards.
In a transition zone between Mediterranean-type and Mountain climatic
regions, the temperatures in degrees Fahrenheit generally range from the low
20s to the mid-90s. Annual precipitation occurs primarily in the winter months,
ranging from 30 to 40 inches (McKnight, 1990; Baxevanis, 1990). A more
detailed description of the climate can be found in Chapter 4.
Geologically, the Sierra Nevada region was subject to intense folding and
metamorphism, changing marine sediments into slates, shales and other
metamorphics. In the Late Eocene volcanic activity occurred, producing lava
flows and ash. A major uplift along faults on the eastern face occurred in the
early Pleistocene, exposing portions of the huge granitic batholith which
1 0 1 2 Kilometers
f -- I I I I
FIGURE 1-2: ZINFANDEL VINEYARD LOCATION
composes the Sierra Nevada range (McKnight, 1990; Ritter, 1978; Soil
Conservation Service, 1974). Figure 1-3 shows the general geology of the
The topography consists of a series of long, gently sloping ridges with
deep river canyons dividing them (Baxevanis, 1990). Shenandoah Valley
exhibits a clear decrease of elevation from northeast to southwest. This gradual
slope is characteristic of the Sierra Nevada mountain range, a fault-block feature
formed by the uplift and folding of a huge granitic batholith (McKnight, 1990;
Ritter, 1978). A series of long, linear slopes occur in the southeastern and
western portions of the study area. A deeply incised river channel defines the
northern boundary of Shenandoah Valley.
Surface soil profiles exhibit the results of colluvial and alluvial depositional
processes. Colluvial soils are generally found on the lower portion of slopes,
while alluvial soils are typically found at the "toe" or bottom of the slope and in
flatter areas (Figure 1-4). The surface and soil profile are dramatically altered in
some areas due to the effects of placer mining. Placer mining is a hydraulic
technique involving the use of large hoses to wash away the soil cover in
mining areas (Ritter, 1978). Some 15% of the study area (approximately 1,200
hectares) are classified in the placer tailings/riverwash category. Predominant
soil types by area are classified as Sierra coarse sandy loam, Ahwanee loam,
Auburn silt loam, and Sierra sandy clay loam, respectively. The Sierra soil
S Valley Springs Formation
| Mariposa Formation
SLogtown Ridge Formation
F Calaveras Complex Volcanic Rocks
SMesozoic Granitics 1
0 1 2 Kilometers
FIGURE 1-3: GENERAL GEOLOGY
Riverwash; Placer Diggings N
I Sierra Coarse Sandy Loam
K Ahwanee Loam A
% Auburn Silt Loam
Sierra Sandy Clay Loam
1 0 1 2 Kilometers
E7 I --
FIGURE 1-4: GENERAL SOILS
series is described as "well-drained, deep and moderately deep soils formed in
material from granitic rock" (Soil Conservation Service, 1965, p. 117). The
Ahwanee soil series consists of "well-drained to somewhat excessively drained,
mostly moderately deep soils formed from weathered granitic rock" (Soil
Conservation Service, 1965, p. 84). Auburn soil series are well drained and
shallow to moderately deep, commonly formed in metabasic igneous rock and
metasedimentary rock (Soil Conservation Service, 1965, p. 88). It is interesting
to note here that none of this soil series is found in the vineyard areas.
RELEVANCE OF RESEARCH AND PREVIOUS LITERATURE
Presented in this chapter is a discussion of the relevance of this research
in the greater context of the regional geography, land evaluation, and viticultural
literature. Considered as a multidisciplinary approach, aspects of each of these
subfields define and contribute to the method developed. To place this
contribution in a plausible context, a small sampling of relevant studies are
identified and reviewed.
There is a raging debate in the wine literature--both scientific and
popular--that has been going on for decades, certainly since the 1940s. This
debate centers on which components of the physical environment have the
greatest impact on the productivity and quality of grapes, and subsequently on
the flavor and quality of wine. There are two decidedly entrenched camps:
terroir and climate.
Adhering most rabidly to the terroir view are French researchers, with
supporters in other areas of Europe (Enjalbert, 1983; Seguin, 1986, Wallace,
1972). The definition of the term terroir has evolved over the years from
describing the ambient environment--climate and land--to describing soils,
oftentimes specifically very local soils. That is, the properties and micro-
properties of the local soils affect grapevine productivity and convey a unique
set of components to the grape, and hence to the flavor and quality of the wine.
Agreement on the definition of terroir remains elusive. As described by Seguin
(1986), it refers to the soil and topographic environment, while Berry (1990)
includes aspect and climate as part of the grape growing territory.
The climate camp originated in the United States, commonly attributed to
work by Amerine and Winkler at the University of California Davis in the late
1930s and early 1940s (Amerine and Winkler, 1944). They determined that
because grapes grow on a variety of soil types, but are limited both in terms of
geographic distribution and quality, that climate, primarily temperature, has the
greatest impact on productivity, character and quality (Amerine and Winker,
1944; Winkler et al., 1974). In fact, Amerine and Roessler, in their book Wines.
Their Sensory Evaluation, (1983) cite climate in the index 27 times, but have no
listing at all for soils.
The suitability technique promoted by Winkler is regionalization and
grape cultivar-matching based on heat summation. Heat summation is
described in detail under the Climate Description section of Chapter 4. An
examination of the viticultural literature illustrates a strong emphasis on climatic
studies, including Becker (1984), Bell (1980), Coombe (1987), Fisher (1978),
Kliewer and Smart (1989), Koblet (1984), Smart (1987), and Winkler (1949).
Accepting the influence of climate in determining viticultural suitability on
a macro scale, this research adopts themes from the terroir approach. This is
accomplished through identification of distinctions in pedologic and topographic
influences within a regional climate type accepted as suitable. Integrated within
the framework of land evaluation techniques are aspects of GIS and regional
Land Evaluation Literature
A comprehensive chronological review of the land evaluation and
suitability ranking literature can be found in Ryder (1989). Early development
and widespread application of land evaluation techniques can be found in
anthologies of the American and Australian literature (Davidson, 1986; Stewart,
1968; Nortcliff, 1987).
Land evaluation techniques and applications in western Europe are
reviewed in proceedings of a seminar edited by Haans, Steur and Heide (1984).
Emphasizing a quantitative approach, Beek et al. (1986) discuss the evolution
and future direction of land evaluation.
Italian researchers have expanded this tradition, especially in the areas of
viticulture and agricultural land use planning on the regional and local scale
(Costantini, 1987; Scienza and Falcetti, 1991; Scienza, 1992, 1990; Falcetti,
1992; Costantini and Pinzauti, 1992). One very typical example is provided by
Rodolfi (1988) in his examination of land capability, topography and soil erosion
in the vineyard regions of Tuscany, central Italy.
There are numerous examples of similar approaches to agricultural land
evaluation in the literature. Many use existing, commonly available data for
suitability determination (e.g. Soil Conservation Service soil surveys, county
parcel maps, U.S. Geological Survey topographic data) and some computer
techniques, primarily for data manipulation.
Approaches to land suitability determination and ranking can be divided
into two classes: identification of homogeneous land units or utilization types;
and classification of soil characteristics. The first category follows a
methodology established by the Food and Agricultural Organization (FAO) of
the United Nations (FAO, 1976). This methodology evaluates a range of
physical characteristics including, but not limited to, water supply, soil type,
physiography, and transportation facilities to determine the suitability of a region
for a particular use (FAO, 1976; Diamond, 1984; Andriesse and Scholten, 1983).
Typical of this approach is work by Dent (1978) which incorporates cultural
management practices and soil classification into a multiple regression-based
predictive model for wetland rice yield. Summaries of applications in Latin
America based on the concept of Land Utilization Types (LUT) were compiled
by Beek (1978).
The second approach emphasizes the role of soil properties and
characteristics, such as texture, structure, organic content, and moisture, in
land evaluation. A ranking scheme is developed based on the suitability or
limitations of these factors for agricultural or urban development. This approach
is employed by the Soil Conservation Service (SCS) of the U.S. Department of
Agriculture (USDA) (Klingebiel and Montgomery, 1961; Singer, 1978). Another
example is the use of cluster analysis techniques by Rogoff et al. (1980) to
group soils with similar qualities for land development and evaluate the relative
costs of treatments to overcome development limitations
Of special significance to this research are the studies by Schreier and
Zulkilfi (1983) and Hiley et al. (1992). Working in British Columbia, Schreier and
Zulkilfi (1983) used interviews of growers, and factor analysis to select relevant
soil and site parameters. These parameters were grouped based on similarity
using cluster analysis techniques, and the resulting soil management groups
were related to crop productivity. The authors concluded that although a
significant quantitative relationship between soil groups and yield could not be
established due to differential management practices, within the management
practice groups an indication of economic potential can be determined
(Schreier and Zulkilfi, 1983). In a similar approach, Hiley et al. (1992), working
in Alberta grouped soil types by management class and related these classes
based on a ratio of their respective yields. Similarity of the ratios was tested
using the Mann-Whitney U test. The findings indicate that this technique
provides an acceptable relative measure of productivity, but that "more precise
data on weather patterns and soil classes would provide greater accuracy"
(Hiley et al., 1992, p. 17).
Implications of GIS and Relevant Literature
The geographic information system (GIS) is used to record and
manipulate geographic information through a variety of operations: input and
display of geographic data, query of the geographical database for specific
facts, comparative analysis of different data types for one area in reference to
a specific problem, and complex spatial modeling used to test alternative plans
(Dickinson, 1989). The development of GIS capabilities and their application
across a number of disciplines have been rapid in recent years (Chrisman
et al., 1989; Tomlinson, 1987; Rhind, 1987).
Of significance to this research project is the application of GIS
techniques to viticultural land suitability and spatial analysis. GIS techniques are
beginning to be applied to determine land suitability for selecting optimal sites
for vineyards in northern Italy (Scienza, 1992), and in the Champagne (Laville,
Moncomble and Panigai, 1992) and Loire Valley (Morlat and Asselin, 1992)
regions of France. Input variables to the Scienza (1992) GIS model were
lithology, soil type, elevation, slope, precipitation, temperature, and Winkler and
Huglin indexes (i.e. indices of number of days per year with suitable
temperature ranges). The results of this application are currently being
analyzed. The French study in Champagne used GIS to analyze cartographic
techniques, including map scale and grid cell sampling, for characterizing the
environment (terroir). In the Loire Valley, GIS techniques were used to consider
geology, soil and climate factors in determining rootstock suitability and wine
quality. In northern Chile, Romero (1990) is using GIS to analyze physical
characteristics, transportation facilities and market location with respect to land
suitability for table grape production.
Many studies of the spatial pattern of viticulture and vinification have
addressed the influence of environmental factors. Although the bulk of articles
indicate dominant factors from a holistic perspective, no consistent relationship
between environmental factors and viticultural productivity emerges.
Relatively recent approaches to determination of environmental influences
and suitability have taken a physiological and/or quantitative tack. Using a
technique that is widely cited and gaining in popularity, known as the "genotype
x environment" approach, Onokpise and Mortensen (1988) evaluated
productivity of grape cultivars in northern Florida. This technique essentially
stratifies local climates or environments based on the productivity of specific
grape cultivars. Along a similar vein, but using a different technique, Dutt et al.
(1981) suggest matching cultivars to location based on mean annual soil
temperature at a depth of 50 cm. This technique was applied in the southwest
United States, and provides a more general regional delineation. In attempting
to determine areas of origin and the influence of environmental variables several
European researchers have analyzed various chemical constituents of the
grape, with limited success. Notable among these attempts are Etievant et al.
(1988) and Forina et al. (1986). Although a correlation between chemical
constituents of color and some phenols was found with soil type, the results
were not statistically significant.
Historically, a more descriptive regional approach to land suitability was
taken. The work of Olmsted (1956) stands out as a very typical work, in his
regionalization and description of vineyard and orchard areas of the United
States. One notable Australian example is the assessment of the prospects and
costs of viticulture in the Barossa region, provided by Smith (1970), where the
environmental and economic advantages and constraints in this region are
examined. Assessing the historical patterns of settlement, transportation and
access to markets, as well as the influence of physical factors, Weigand (1954)
provides a regional description of viticulture in southwest France. More
recently, the classic geographic regional analysis can be seen in work by de Blij
(1983), in which physical, cultural, economic and political aspects of viticulture
and wine are examined.
In an examination of the Finger Lakes region in New York state, Newman
(1986) describes an evolving regional identity based on changes in the location
of vineyards, grapes and management practices. Following on the idea of
evolving regions, and applying techniques from economic geography, Peters
(1987) uses location quotients to identify shifts in regions of cultivar
specialization in California. In a departure from much of the literature on the
geography of viticulture, Moran (1993) offers an alternative explanation of the
role of the region. While acknowledging the importance of the physical
environment, his argument is that politics and economics have a greater role in
the development of regional identity and characteristics. This argument is
supported by comparing and contrasting wine production regions in France and
Description of Technique and Process
In discussing continuous themes in geographic research, King (1969)
speaks of analysis of degree and direction of correspondence or variance of
spatial patterns. "In some cases this analysis is driven by hypotheses about
functional relationships or causal mechanisms between the phenomena in
question; in other situations, the analysis is exploratory and seeks to derive
inductive generalizations concerning the covariation of spatially distributed
phenomena" (King, 1969, p. 117). Following this exploratory theme, the
approach as described below was chosen.
A comprehensive description of the data collection and analysis
techniques is presented in this chapter. It opens with a discussion on the
selection and inherent caveats of the approach, followed by identification of the
variables analyzed. A detailed description of the variables is then presented.
Data sources and techniques for manipulation and analysis of the data are
discussed. The final section addresses statistical techniques, assumptions and
relevant issues employed in the model, sample and validation data set analyses.
Selection of Approach
Selection of this approach was driven by the desire to develop an
innovative and efficient method for land suitability evaluation. Geographic
information system (GIS) software is an evolving tool, comprised of a set of
methods for data manipulation and spatial analysis. Capabilities of GIS relevant
to this research are the ability to integrate a diverse array of data types and to
compare and associate data features spatially. By taking advantage of these
capabilities and techniques, the intent of this study is to develop a widely
applicable computerized method using existing technology and sources of data.
The greatest cost in most resource oriented studies lies in data collection
and conversion into a digital format (Montgomery and Schuch, 1993). A
reliance on existing secondary and tertiary data minimizes these costs, and
utilizes data efficiently in a familiar and commonly available form. While
minimizing expense, the use of secondary data can also impose limitations on
the scope and techniques employed in an analysis. Considerations include the
accuracy, format and timeliness of the available data. Scale, resolution, data
collection and manipulation error, and intended use of original data are all
factors that can affect accuracy and format. Timeliness includes date of
collection, time and interval of record. Scale, resolution and sources of error
will be discussed in detail in Chapter 5, in the context of their influence on the
results of this study.
The statistical analysis component of this study follows closely the
methods for agricultural suitability assessment employed by Schreier and Zulkilfi
(1983), and to a lesser degree, Hiley et al. (1992). Reviews of these and related
works are presented in Chapter 2 under the Land Evaluation subsection.
Data Collection and Description
Data collection was subdivided into four main stages. Stage one
consisted of a literature review of land evaluation, and viticultural topics,
including an extensive viticultural geography literature. The results of this stage
formed the bulk of Chapter 2 and provided the initial list of environmental
variables for consideration.
The second stage involved mail and phone contact including:
letters sent to established researchers in this topical area to
determine if there were any existing works on viticultural land
suitability using GIS techniques;
-the location of sources of information, addresses and contacts; and
-establishment of initial contact and arrangement of interviews with
growers, county extension agents, the Property Appraiser's office,
and other points of contact.
The third stage consisted of a site visit to conduct interviews with
vineyard owners/managers and agency personnel, inspect the vineyards and
gather data from state and county offices. Revisions of the data comprised the
fourth stage, in which inconsistent information or mistakes were rectified, and
additional data were gathered by mail and phone survey.
The number and type of variables selected for this research evolved over
time as a result of data availability and type. Variables are classified as
descriptive and quantitative. Descriptive variables included the climate data,
due to scale limitations (addressed in Chapter 5), and the nominal soil and
geology properties, due to their limited value in the statistical analysis.
Quantitative variables were the ordinal, interval and ratio level soil properties,
and slope aspect and angle.
An initial listing of variables analyzed in previous studies is compiled in
Table 3-1. Revisions were made subsequent to visiting the study area,
interviewing vineyard owners/manager and agency personnel, and evaluating
available data. Considering the scale limitations of the climate data and the
difficulties encountered in obtaining complete production data, the focus was
placed on soil, geologic and physiographic information. A detailed review of the
literature revealed the soil properties most commonly investigated. Table 3-2
presents these properties organized by study and location.
Presented in Table 3-3 are the variables which formed the basis of this
analysis. The classification of the variables by data type in the context of this
study is also presented in this table. Soil variables are organized based on Soil
(D '- (
0 E0 (0
= 0 a
t 5 o
%- C 0
TABLE 3-2: SUMMARY OF SOIL VARIABLES SELECT ED FROM THE
Dutt, et al., 1981
Galleta & Himelrick, 1990
Onokpise & Mortenson, 1988
Rankine, et al., 1971
Scienza & Falcetti, 1990
Smart, et al., 1985a
A = Acidity (pH)
C = Composition
CEC = Cation Exchange Capacity
Col = Color
D = Depth
Dr = Drainage
F = Fertility
M = Moisture
N = Nitrogen
OM = Organic matter
P = Phosphorous
Pe = Permeability
Pot = Potassium
S = Structure
Te = Temperature
Tx = Texture
WH Water-holding capacity
(A) Applied = original study or experiment
(S) Summary = summary of other work or literature
TABLE 3-3: VARIABLES USED IN STUDY BY CLASS
Variable (by class)
slope angle (Q)
slope aspect (Q)
available water holding capacity (Q)
drainage class (Q)
runoff class (Q)
effective root depth (Q)
soil moisture (Q)
cation exchange capacity (Q)
Storie Index Rating (Q)
Storie Index Rating Grade (Q)
(Q) = Quantitative
(D) = Descriptive
Conservation Service categories. Definitions of these categories are presented
in the Appendix. Slope angle, or steepness, is calculated in percentages.
Aspect, or cardinal direction of slope exposure, is initially measured in degrees
clockwise from North (0 ). In this form it is a circular variable, with no
dimension of magnitude or relative scale (Kimmerling and Moellering, 1990).
Consequently, in preparation for the statistical analysis, a sine transformation
was applied to the aspect data. A commonly used technique in forestry site
index suitability studies (Beers et al., 1966; Trimble and Weitzman, 1956), it
takes the form:
Slope aspect in degrees is represented by (x), while 90 represents a neutral
weighting coefficient. That is, no particular aspect is emphasized, rather all
directions are treated equally. One is added to the sine function to shift the
result to a range between 1 and 2.
The Zinfandel grape type was used as the basis for vineyard selection.
This cultivar was chosen due to its widespread occurrence in the study area
and associated consistency in vineyard and grapevine management practices.
Some background information on the grape is appropriate here. The origin and
routes of dispersion of this grape remain a mystery. One of the earliest
citations of a cultivar thought to be Zinfandel comes from an 1830 seed
catalogue of a New York nursery (Muscatine et al., 1984). The country of origin
was cited as Hungary. In the early 1850s, several different efforts were
undertaken to produce wine from various cuttings brought from the East Coast,
and known as Zinfandel. One of the concentrated areas of cultivation was
around Sacramento in the late 1850s. In 1967, a plant pathologist with the U.S.
Department of Agriculture studied a cultivar in the Puglia region of southeastern
Italy, locally known as Primitivo di Gioia. This grape was found to be identical
to the Zinfandel grape, but there is no historical information of any connection
or diffusion of the Primitivo (Muscatine et al., 1984; Ramey, 1977). Whatever its
origin, Zinfandel is distinctly identified with California, and is second only to
cabernet Sauvignon in acreage (California Agricultural Statistics Service, 1993).
Locational data refers primarily to the GIS model base map, vineyard,
and cultivar locations. The base map source was obtained in digital form. This
format is known as Digital Line Graph (DLG), which contains cultural features
and surface water bodies (U.S. Geological Survey, 1984).
Vineyard and cultivar locations are derived from the literature, Amador
County Property Appraiser maps, U.S. Geological Survey (USGS) 7.5 minute
orthophoto quadrangles, aerial photographs obtained from the Amador County
Surveyor's Office and the National Aeronautics and Space Administration
(NASA), the California Agricultural Extension Service (CAES), and vineyard
owners/managers. Vineyard management practice data sources included the
literature, interviews with vineyard owners/managers, CAES and the California
Agricultural Statistics Service (CASS).
Environmental variables collected for analysis can be grouped into three
major categories: soils, geology, and topography. Soils data were obtained
from the Soil Conservation Service (SCS), including the Amador County office
and the published Soil Survey (Soil Conservation Service, 1974; 1965).
Additional information was obtained from the AES and vineyard owners/
managers. Geologic data was obtained from various USGS sources and the
California Division of Mines and Geology (1987). The primary topographic data
source is the USGS 7.5 minute topographic map series. This information was
supplemented by limited field measurements of vineyard site elevation, slope
angle and aspect.
The heterogeneity of the information collected provides ample justification
for the use of GIS techniques from a purely data input, formatting and
manipulation perspective. The raw database included the following:
maps of different projection, scale, date and media
aerial photographs of different scale and date
tabular data e.g. ownership, soil properties
-interview and anecdotal information
-miscellaneous e.g. site photos and measurements
The multi-scale, multi-format, multi-type capabilities of GIS are well
documented in numerous articles and books (Dickinson, 1989; Burrough,
1986), and were essential for this land suitability evaluation research.
Figure 3-1 illustrates the major steps in construction of the digital
database using the Arc/Info GIS software. A 1:100,000 scale USGS digital line
graph generalized to the 1:24,000 scale, including roads and hydrography,
formed the base map. The Fiddletown and Amador City quadrangles were
converted from DLG to Arc/Info format and joined, taking care to edgematch
the boundary, road and hydrography features. The soils layer was created
using a zoom transferscope to "rubbersheet" the 1:20,000 aerial photos with soil
phase delineations to the 1:24,000 scale USGS orthophotoquads. A widely
used technique (Whiting, 1993, personal communication), this involved
registration and adjustment in scale and location of soil polygons using photo-
identifiable registration points from both sources. Orthophotos are rectified
aerial photographs which are distortion corrected to represent the true location
of ground features. Sources of error typically corrected in orthophotos include
distortions due to relief, atmospheric attenuation, camera tilt and lens effects.
The resultant photograph is standardized in terms of scale, coordinate system,
origin and distortion. Vineyard locations were also transferred from aerial
- soils data transferred to orthophoto quad
- vineyards transferred from parcel maps
and aerial photographs
- sample points generated from basic program
- tics derived from corners of quad sheets
- projected into Lambert conformal conic
- secondary projection into Universal
- 2 quads in DLG format joined and edge matched
- vineyards, soils and geology polygons digitized
- point coverage generated from sample points
- select high information points to
- generalize contours
- set weed tolerance of 10 meters
- add or remove points to eliminate
- check plots f
- digitizer RMi
- tic overlay a
or visual alignment
accuracy 3.5 meters
FIGURE 3-1: DATABASE CONSTRUCTION PROCEDURES
photos and maps to the orthophotos. The orthophoto delineations were
subsequently digitized, as were geology data from the California Division of
Mines and Geology. Resulting data layers, or coverages, consisted of
collections of soil, vineyard and geologic polygons. A sample point coverage
was generated following a technique outlined in the sampling section of this
chapter. All map layers are standardized in terms of reference geographic
coordinates, projection and scale. The geographic units used are meters, the
projection is Universal Transverse Mercator (zone 10) and the scale is 1:24,000.
The Arc/Info Triangulated Irregular Network (TIN) module was used to
model valley geomorphology, including elevation, slope angle and aspect. The
concept of the TIN model is based on "a set of irregularly-spaced points
connected into a network of edges that form space-filling, non-overlapping
triangles" (Kumler, 1992, p. 2). The algorithm follows a two step process.
Thiessen polygons are created around points to establish proximity or
adjacency of points. The Delaunay triangulation procedure is then used to
establish lines of interpolation between the centers of adjacent Thiessen
polygons (Kumler, 1992, Lee, 1991). The TIN topology is composed of a series
of points with x, y locational coordinates and z elevation values connected by
planar triangles (ESRI, 1992). A triangular pattern removes any ambiguity in
interpolation of elevation, slope angle and slope aspect. It includes information
on boundary nodes, arcs, and triangle areas, allowing for the calculation and
graphic representation of a number of surface characteristics. Most significant
to this research are slope aspect and angle, while other characteristics include
slope length and surface area. Important advantages of this technique over the
more traditional grid method are lower memory storage requirements and
higherachievable resolution. Fewer points are required to represent the surface,
and these points can be concentrated in high information areas (e.g. peaks,
ridges, valleys, stream channels, and areas of rapid variation in relief). Elevation
data necessary for the construction of the TIN were derived from the USGS
7.5 minute topographic quadrangles (e.g. Amador City and Fiddletown). The
spot elevation points were combined with scanned contour lines from both
topographic quadrangles to produce a TIN model. Model resolution was driven
by the heterogeneity of slope aspects, and to a lesser degree, slope angles
present in many vineyards. Minimum planar triangle area was approximately
43.3 meters as determined by "weeding out", or deleting elevation points within
a distance of ten meters of each other. This yields approximately 104 TIN
triangles, or polygons within the smallest mapped vineyard (approximately
A necessary component for any type of spatial analysis is the association
of spatial location with the aspatial attribute data. One of the strengths of GIS
technology and a critical aspect of this study is this capability. Combining the
locational map base of polygons and points with attribute data defines the
characteristics of the variables at those locations. This allows for the
comparison of locations by their characteristics. In this case the comparisons
are between vineyards, and between vineyards and non-vineyard areas. Due to
the magnitude of the data set, this critical stage of the analysis is only made
possible using GIS technology.
Characteristics of environmental factors and related data comprise the
relational database (the Info portion of Arc/Info) in both polygon and point
formats (e.g. vineyard versus sample data sets). Figure 3-2 provides an outline
of the basic data manipulation process. The raw data layers were processed to
establish topology, or more simply, the spatial relationships between point, line
(arc) and polygon features. The reader is referred to Environmental Systems
Research Institute (1992) for a description of the Arc/Info data model. Tabular,
or attribute data, were added to the coverages (processed map layers) by
creation of related files in the Info relational database. This was achieved by
defining the necessary items to be contained in the Info file and adding the
relevant data from the original source. An example would be selected soil
properties associated with soil phases identified in the study area. Slope angle
and aspect information was generated from the TIN model and a slope polygon
layer was created. All attribute data were attached to the relevant polygon
using unique identification numbers.
As discussed previously, all data layers were collections of delineated
polygons or points, with relevant attribute data attached. Thus generating
DEFINE INFO ITEMS
name, width, type
CREATE ASCII FILES OF ATTRIBUTES
tabular data from Soil Surveys
elevation data from USGS quads 33
geologic categories < Q
ownership of vineyards
ADD ASCII FILES TO INFO PAT FILES en
use add from command
create errror log file
QUALITY ASSURE DATA
( review data for missing, altered or
GENERATE SLOPE ANGLE AND ASPECT DATA
convert TIN model to a polygon coverage
GENERATE DATA SETS
create vineyard and non-vineyard data
sets of observations on environmental
intersect TIN polygons with vineyard,
soil and geology coverages
stratify sample point coverage for
vineyard and non-vineyard areas
intersect sample point coverage with
TIN, vineyard, soil and geology covers
export INFO files with attribute data,
creating ASCII files for import to SAS
FIGURE 3-2: INFORMATION MODEL CONSTRUCTION PROCEDURES
profiles of vineyard characteristics becomes a matter of intersecting vineyard
polygons with soil, geologic and slope polygons. This process can be
visualized simply as overlaying all polygons and inserting a pin through the
composite at desired locations. Intersection of the data layers occurred in two
passes, due to the difference in minimum polygon size between the vineyard
and slope layers. That is, due to the resolution of the TIN model, its minimum
polygon size was smaller than that of the vineyard layer (i.e. 43.3 m2 versus
4.5 hectares). Pass one generated 539 slope, 105 geology, and 307 soil
polygons on intersection with 93 vineyard polygons. Averaged over the number
of vineyard polygons, this yields 5.68 slope, 3.27 soil, and 1.12 geology
polygons per vineyard. Bearing in mind that the smallest vineyard contains
approximately 104 TIN polygons (triangles), it can be seen from these figures
that a substantial amount of slope data would be lost as it would have to be
averaged over each vineyard. In order to utilize the slope data, these polygons
became the minimum resolution of the intersection procedure. Pass two, an
intersection of all data layers with the slope layer, generated a total of
1,035 vineyard, soil, and geology polygons. Procedures for generation of the
sample point data are described below.
This section describes the statistical methodology utilized in the research.
Figure 3-3 is a flow diagram of this methodology. The basis for this approach is
presented, as well as a description of each statistical technique. The strengths
and weaknesses of these statistics are also discussed.
Selection of a statistical methodology is driven by the nature of the
research question and the type of data required to address this question. A
comprehensive discussion of the general considerations inherent in this process
is not appropriate here. Specific techniques employed within the general
approach are determined by the type of data used. The reader is referred to
any number of basic statistical texts which will provide a relevant explanation
(Clark and Hosking, 1986; Barber, 1988; King, 1969).
The question to be answered in this research was stated previously in
Chapter 1 under the Research Problem subsection. The approach chosen to
address this question involved sampling and correlation analysis to construct a
statistical data model. Validation of the model incorporated the same
techniques, and was accomplished using comparative statistics. Variables
chosen for construction of the physical model and their data type are listed in
Table 3-3. It can be seen that data types include nominal, ordinal, interval and
ratio. Analysis options are constrained by the level of measurement of each
/ Systematic random sample of vineyard
Spolygon identification numbers for creation )
of the Validation and Model data sets. V
SStratified systematic unaligned random
Ssample for point intersection data to create 03
Sample data set in non-vineyard area. m
removal of incomplete records
mean, median, high, low values
scatterplots of variables
^ _i -
/-Generation of correlation matrices:
Spearman's r between variables
for all data sets x
Factor Analysis: 0
creation of initial factor pattern -
selection of factors (n) 0
scree plot Z
Similarity Analysis of Model, Validation and Sample data sets:
Mann-Whitney U test comparing:
rotated factor coefficients <
named major factor coefficients C)
frequency distribution of observations C/)
FIGURE 3-3: SUMMARY OF STATISTICAL PROCEDURES
data type, as represented by magnitude and relative scale. Magnitude is
measured in terms of level above an origin or absolute zero. Scale refers to the
distance between individual observations on common variables, e.g. 2 is twice
as large as 1. Data types containing magnitude and scale are typically interval
and ratio, and allow for the use of more powerful analytic and parametric
techniques. Accepting the analytic limits of nominal and ordinal data,
nonparametric techniques were employed in order to utilize the relevant data
set as fully as possible. Nonparametric or "distribution-free" techniques are so
named as they have no requirements or assumptions regarding the shape of
the sample probability distribution (Siegel, 1956, p. 3). Parametric statistics
generally require an assumption that the sampled variables be normally
distributed, as shown in a basic bell-shaped curve. That is, the variables follow
a "specific mathematical expression which defines the characteristic bell shape"
(Barber, 1988, p. 177), with a mean of 0 and a standard deviation of 1.
Using the discussion above as a foundation, a description of the
techniques applied in this study is presented. The methodology can logically be
divided into three segments: 1) sampling and data collection; 2) data reduction;
and 3) model validation and analysis.
Three basic data sets were derived from two sources. The vineyard data
set was taken from the intersection of vineyard and environmental variable
polygons as described under the GIS methodology. Sampling techniques were
used for two purposes: development of a vineyard data set for model
validation, and for generation of a data set for the non-vineyard remainder of
the study area. A systematic random sample of the vineyard data set was
generated using a random number table. This sample formed the model
validation data set. A sample of 35 was selected from a possible total of
93 vineyards. This number was chosen in order to leave an adequate number
of vineyards (58) for model construction. In addition, it exceeds the optimum
minimal number of sample observations (31) often cited in the statistical
literature (Barber, 1988, p. 249).
Generation of the sample data set was accomplished using the
systematic unaligned sampling technique, following the design suggestions of
Berry and Baker (1968). This design is advantageous for analysis involving
land cover data and includes: avoidance of the periodicities inherent in the
spatial systematic sample; good coverage over an area; efficiency; and
applicability to most population distributions (Clark and Hosking, 1986; Webster,
1977; Berry and Baker, 1968). The point selection procedure involved
overlaying a grid on the study area and selecting a random x,y coordinate in
the uppermost left cell. Subsequent sample points utilize the original x or y
coordinate, in terms of row or column, and new random values for the x or y
coordinate respectively (Clark and Hosking, 1986; Berry and Baker, 1968).
Square grid cells 67.26 meters per side were used. This interval was chosen as
it represented the minimum vineyard size in the study area (4.52 hectares). A
basic program generated the sample point Universal Transverse Mercator
coordinates in a format compatible for importation to Arc/Info. A point
coverage was created from the sample point ASCII file. Intersecting the point
coverage with the previous intersection coverage (e.g. soils, geology, slope)
produced the initial sample data set. Existing vineyard areas were stratified out
of this sample by using the vineyard coverage as an erase coverage to delete
all points falling within vineyard polygons. The final sample point intersection
coverage consisted of 17,017 points containing location and environmental
variable attributes. Sample points with incomplete observation records on
variables were discarded. A total of 11,326 points remained for the statistical
As a consequence of the large number of variables initially selected for
analysis in this study (15), some form of data reduction was required. Factor
analysis was applied to these variables in order to reduce the number of
variables based on the interrelationships identified in a correlation matrix. By
assessing the communality, or covariance of the variables, a realistic
approximation of the physical/environmental system can be constructed. That
is, it provides a versatile set of procedures for discerning the simultaneous
variation in many variables (Davies, 1984; Cattell, 1965). Covariance can be
thought of as a measure of association between variables (Kim and Mueller,
1978b). It is based on the deviations of variable observations from the mean
value of those observations. Notable comprehensive discussions on the
derivation and use of this technique, as well as reviews of applications can be
found in Davies, (1984), Cattell, (1978), Kim and Mueller, (1978a; 1978b) and
A brief description and definitions of several key components of the
technique are useful here. A description, or re-description of the relationships
between variables in a data set according to their strength of association, or
correlation is the basis of the technique. Two primary approaches in factor
analysis are commonly referred to as R-mode and Q-mode. R-mode identifies
the patterns or structure inherent in the data set as evidenced by the similarity
between variables. In the context of this study, it was used to describe the
patterns apparent in the spatial location of the environmental variables listed
previously in Table 3-3. The observations on these variables were their
occurrence within vineyard or non-vineyard areas. Q-mode analysis is
commonly used to describe the similarity structure of the observations on the
variables. That is, to describe the vineyard in terms of the patterns of variables
that occur within them. It was not possible to undertake a Q-mode analysis,
due to the size of the data set, and the limitations of the statistical software
(SAS-PC). In addition, with only 15 observations (the variables from R-mode)
on 1,035 variables (the observations from R-mode; i.e. the number of vineyard
subpolygons after intersection with the slope layer), it would be difficult to
achieve a statistically significant result. Consequently, an analysis of frequency
of variable occurrence by vineyard and non-vineyard area is presented in
Results of the R-mode analysis are presented in Chapter 4. Primary
objectives of this analysis were to: identify an underlying pattern or structure in
the data set; rewrite or transform the data set into a more parsimonious form;
and use the statistics generated in both a descriptive context, and as input to
the Mann-Whitney U test procedure for additional analysis (Davies, 1984;
Discerning patterns in data sets is predicated upon an examination of the
factors selected to describe the data set. Factors can be thought of as vectors,
or new variables that geometrically summarize the variability of each study
variable. As stated clearly by Davies (1984, p. 36) "These factors, axes or
vectors--the terms are used interchangeably--can also be considered as
surrogate variables, in the sense that the variability of each variable is rewritten
as a linear function of these new reference vectors." A table of factor loadings
is produced when this procedure is run using the SAS statistical software.
Factor loadings represent the weighted value of variability explanation achieved
by each factor, analogous to the correlation between the variable and the
factor. Remember that the factors represent columns, and the variables rows,
in a matrix. Summing the squares of the factor loadings across the rows
provides a measure of the amount of variance in a variable explained by a
particular factor solution. Summing the squares of the loadings down the
columns provides a measure of the variance in the data set explained by each
factor. The square root of the column value is known as the eigenvalue
(Davies, 1984; Kim and Mueller, 1978a).
Parsimony is a widely cited term as an objective of factor analysis
(Davies, 1984; Kim and Mueller, 1978a; Henshall and King, 1966). In the literal
sense, it refers to frugality or economy in approach. Applied to factor analysis,
it refers to a redescription, or synthesis of a data set into a simpler, more
discrete structure. This is accomplished by reducing the number of data
variables through compression into a lesser number of uncorrelated factors.
The last objective consists of two components, descriptive and analytic.
Descriptively, the results of the factor analysis are used to identify and discuss
those variables which are shown to be associated one with the other. These
variables can be represented by naming the major factors which describe them.
This serves to both confirm intuitive perceptions of, and explain physical
relationships between variables. Analytically, the results are then used in a
subsequent nonparametric statistical procedure, the Mann-Whitney U test,
described under the Model Validation subsection below.
Typically, the first step in factor analysis is the construction of a
correlation matrix of variables in the study data set. A correlation matrix of the
observations on the variables listed in Table 3-3 was constructed using
Spearman's r., a rank correlation coefficient. This is an ordinal measure of
association between variables, based on rank. The concepts of variable
magnitudes, or distance between variables are not applicable to ordinal data.
The statistic is calculated by ranking observations on variables, taking the
differences between ranks and squaring them. Ties in rankings are resolved by
averaging the rank over the number of affected observations. Squares are
summed over all the variables, and standardized so the results are between
-1 and +1 (Barber, 1988; Siegel, 1956). The significance of the statistic can be
calculated by: z = r,/n-1 (Barber, 1988, p. 380).
Spearman's r,, statistic provides the advantages of few data structural
requirements, making no assumptions about the linearity of the relationship
between variables, or the shape of the sample distribution. It provides a
measure of the strength of association between variables, having been shown
to be 91% as efficient as the Pearson r parametric correlation coefficient. That
is, "...if a correlation between X and Y exists in that [a] population, with
100 cases Spearman will reveal that correlation at the same level of significance
which Pearson attains with 91 cases" (Siegel, 1956, p. 213). One obvious
disadvantage is the loss of data associated with the lack of detail in the
magnitude of and relationships between observations and variables, which is
inherent in interval and ratio data. This information can be useful in describing
and explaining patterns of variation in the data set. It has been noted that
Spearman's r., can also "give undue weight to the middle range of a
distribution, since the difference between 50 and 51 is treated the same as that
between 1 and 2 or 90 and 91" (Davies, 1984, p. 120). In the context of factor
analysis however, several researchers have cited the similarity in factor loadings
derived from Spearman and Pearson coefficients of a data set (Davies, 1984;
Moser and Scott, 1962).
An initial, or pattern factor procedure is performed next to discern the
basic structure of the data. At this stage the number of factors used to
describe the data is left unspecified. Using the SAS software, the initial number
of factors defaults to the number of variables in the data set (SAS Institute,
1982). Determination of the correct number of factors to extract is the subject
of an ongoing debate in the literature (Davies, 1984; Cattell, 1978; Kim and
Mueller, 1978b). As suggested by a number of researchers, and discussed in
Chapter 4, several "rules of thumb" were used in this study to select the
appropriate number of factors. The intuitive concept is selection of the fewest
number of factors which explain the greatest amount of variance in the data set.
The next step is the choice of rotation strategy to apply. An explanation of the
derivation of rotation strategies is beyond the scope of this discussion. Once
again, the reader is referred to the texts mentioned above for a complete
explanation. Suffice to say that a rotation technique is applied to maximize the
factor loadings of a variable on a particular vector or factor, and minimize those
loadings on the remaining factors. This leads back to the objective of
parsimony in describing the data set. That is, based on the strength of the
loadings, major and minor factors can be identified and associated with the
variables. Finally, in preparation for additional statistical tests, the rotated factor
loadings can be standardized by dividing each loading in a column by the sum
of that column to generate a factor coefficient.
Model Validation and Analysis
All three data sets were used in the model validation and analysis phase.
Model validation and analysis were accomplished by applying the same
technique to the validation and sample data sets. Continuing the nonparametric
approach, the Mann-Whitney U test was chosen for model validation and
analysis. Consistent with the use of ordinal and higher level data in this study,
Mann-Whitney provides a method for testing "whether two independent groups
have been drawn from the same population" (Siegel, 1956, p. 116). It has
been cited by Siegel (1956, p. 116) as "one of the most powerful of the
nonparametric tests", and a "useful alternative to the parametric t-test when the
researcher wishes to avoid the t-test's assumptions, or when the measurement
in the research is weaker than interval scaling."
The basis of this test is that if two independent samples are drawn from
the same population, then the mean ranks of the two samples should be
equivalent. Two independent samples x and y of size n, and n, are combined
and the observations ranked in ascending order. Observations with the same
value (ties) are assigned the average of their rank. The test statistic (S) is the
sum of the ranks from the first sample, x. For sample sizes (n) larger than 10,
the sampling distribution of S is approximately normal with a mean and
standard deviation as follows:
o2=(n n2(nl +n2+1))/12
The hypothesis that the difference between the sum of ranks of samples x and
y equals zero can be tested in the conventional manner for a one- or two-tailed
test (Barber, 1988; Siegel, 1956). Addressing the efficiency of the Mann-
Whitney U test, Mood (1954) cites its rating as approaching 95.5% of a t-test of
suitable data as the sample size (n) increases. The results of this procedure
are presented and discussed in the following chapter.
This chapter presents the results from both a visual and statistical
analysis of the data sets. A simple analysis and description of the regional
climate is presented first. This includes an annual summary of temperature
and precipitation, as well as a regional identification based on heat summation.
A comparison of vineyard polygon and sample point data histograms is then
presented. Following is a discussion of the results from the correlation and
factor analysis. Validation of the method discusses the results of the Mann-
Whitney U test of the data set distributions. Finally, the results are interpreted in
terms of variable and vineyard relationships, and the strengths and weaknesses
of the method are identified.
As presented previously in the Introduction, the general climate
according to the K6ppen classification system is described as Mediterranean,
transitional to Mountain. Within this category, the average temperature of the
coldest month is between 32 Fahrenheit (0 C) and 64 F. (18 C) while the
average temperature of the warmest month is above 50 F. (10" C). In terms
of precipitation, the driest summer triad (i.e. June, July, August) has less than
one-third the average precipitation of the wettest winter triad (i.e. December,
January, February) (McKnight, 1990; Trewartha, 1957). Figure 4-1 illustrates
the rainfall pattern, which once again closely follows Koppen's description. This
graph is based on up to twenty years of record, from the Plymouth and Dexter
Ranch-Fiddletown stations. These stations are located in the lower southwest
corner and just outside the upper northeast boundary of the study area,
From a descriptive perspective, the heat summation technique, first
espoused by Amerine and Winkler (1944) is most relevant. Heat summation is
typically described as the sum of the mean monthly temperature greater than
50" Fahrenheit. (10 Centigrade). Typical period of summation is April
through October (i.e. "the growing season"). This threshold was chosen
because it is the temperature at which the vine emerges from dormancy and
shoot growth begins. Summation is expressed in terms of degree days, or the
summation of the number of degrees above the threshold times the number of
days in the month (Winkler et al., 1974). Summation classes of degree days
formed the basis for the delineation of five climatic regions in California by
Winkler et al. (1974). Using records of unspecified periods from the Placerville
and Camino stations in El Dorado County, Winkler classified the area as being
in Regions II and III respectively. The mean degree days calculated from
I I i I I .I T I
Jan. Feb. Mar. Apr. May Jun. Jul. Aug. Sep. Oct. Nov. Dec.
Summary of Monthly Mean (1948-1968)
-- E Fiddletown --+-- Plymouth
FIGURE 4-1: MEAN MONTHLY PRECIPITATION
(Source: National Oceanic and Atmospheric Administration)
records of the Shenandoah Valley station places the study area within Region II
based on mean degree days for the period of record.
Vineyard and Sample Area Comparison
As used in this study, vineyard refers to data from Zinfandel grape type
vineyards only. A visual comparison of raw data histograms of variables for
both vineyard polygons and sample points yielded some interesting descriptive
trends. These histograms were constructed using the raw data obtained from
intersecting the appropriate coverages as discussed previously. Intervals for
the variables Storie Index, permeability, water-holding capacity, effective root
depth, drainage, runoff and fertility were based on classes established by the
Soil Conservation Service (SCS). Intervals for the variables acidity (pH), depth,
moisture, slope angle and slope aspect were established based on the data,
while ensuring that a minimum of 11 class intervals were maintained. Eleven
was chosen as the threshold by applying the widely used k suggested interval
formula to the vineyard data set (Griffith and Amrhein, 1991; King, 1969). This
data set was the smaller of the two (n= 1,035). The natural data thresholds or
breaks exceeded k in every instance. A brief description of these trends is
presented, followed by the graphs. Definitions of the variables derived from the
Soil Surveys can be found in the Appendix (Soil Conservation Society, 1965;
1974). It should be noted that the class labeled as Variable shown on several
graphs, refers to areas of soil disturbance and non-natural mixing (e.g. mine
Figure 4-2 shows the relationship of the Storie Land Capability Index
between the sample and vineyard data sets, respectively. Very briefly, the
Storie Index considers four general factors for the evaluation of land capability
for crop production. Those factors are: "a) characteristics of the soil profile
and soil depth; b) the texture of the surface soil, c) slope, and x) other factors,
such as drainage, alkali and erosion" (Soil Conservation Service, 1965, p. 28).
This is a multiplicative parametric approach to land evaluation. That is, the
product of the factor scores equals the Index. Grades of soil capability are
developed based on classes of the Index ratings. The reader is referred to
Storie (1933) for a complete description of the technique. Approximately 28% of
the sample points lie at index level 3, which falls in the category of "lands that
are not suited to agriculture" (Soil Conservation Service, 1965, p. 28). The
remaining sample points are largely evenly distributed, below 5% of the total, up
to a maximum index of 68, denoted as "good soils, well-suited for agriculture"
(Soil Conservation Service, 1965, p. 28). Well over half of the vineyard
polygons lie at an index of 42 or higher, up to 68. This places them in the
"fairly well-suited for agriculture" class (Soil Conservation Service, 1965, p. 28).
Variations in soil depth can be seen in Figure 4-3. Approximately one-third of
the sample points suggest a moderate soil depth, at 30 to 35 inches, and nearly
FIGURE 4-2: STORE LAND CAPABILITY
(Source: Soil Conservation Service, 1965; 1974)
0 5 10 15 20 25 30 35 40 45 50 55 60 65 70 75+
Depth in Inches
73 Vineyard (n=1,035) Sample (n=13,087)
FIGURE 4-3: SOIL DEPTH
(Source: Soil Conservation Service, 1965; 1974)
44% with very deep soils of 75 or more inches. The vineyards exhibit a distinct
predominance of deep soils (75+ inches), with approximately 72% of the
polygons in this category.
The sample point and vineyard observations on soil water holding
capacity, are shown in Figure 4-4. The greatest number of sample points
(approximately 33%) are in the moderate class, followed by 27% in the high
class. The vineyard data differs in that the largest percentage falls in the high
class (59%), 22% in the moderate class, and each of the remaining classes
having less than 10%.
Both the sample point and vineyard data sets are centered on the
medium runoff class (Figure 4-5), with the sample observations at 32% and the
vineyard observations at 48%. It should be noted that the sample points are
more evenly distributed, with a substantially higher percentage of observations
in the very rapid class. The vineyard observations exhibit a noticeably higher
percentage (27%) in the slow class, against 13% in the sample observations.
In terms of effective rooting depth, again the sample points show a more
even distribution than the vineyard points (Figure 4-6). Vineyards have a
distinct predominance of observations in the deep class (65%), while the
moderately deep (36%) and deep (29%) classes combine for approximately
65% of the total sample observations.
I I I I I
Very Low Low Moderate High Very High Variable
SCS Water-Holding Capacity Class
SVineyard (n=1,035) M Sample (n-16,747)
FIGURE 4-4: SOIL WATER-HOLDING CAPACITY
(Source: Soil Conservation Service, 1965; 1974)
........................- -- ..... -----------..- -- -
....... ......-.. -. -... -............... ........- -
............ .. .......... ""
Slow Slow Med.
.......... ... ..................... .......
=3 Vineyard (n=1,035) M Sample (n-16,733)
FIGURE 4-5: SOIL RUNOFF
(Source: Soil Conservation Service, 1965; 1974)
V. Shallow Shallow Mod. Deep Deep Very Deep Variable
SCS Root Depth Class
x3 Vineyard (n=1,035) J Sample (n-16,747)
FIGURE 4-6: SOIL ROOTING DEPTH
(Source: Soil Conservation Service, 1965; 1974)
Soil fertility (Figure 4-7) is dominated by the moderate class in both
sample and vineyard data sets. This class comprises approximately 65% of
the sample observations, and 88% of the vineyard observations.
Soil moisture again shows a great deal of correspondence between the
sample points and the vineyards. Both data sets are dominated by the 9% soil
moisture level (Figure 4-8) with the sample having 56% and the vineyards
59% of their total observations here. In addition, the 6% and 8% moisture levels
show similar patterns, although of different magnitudes, between the data sets.
Distributions of acidity of the soil (pH) are shown in Figure 4-9. The
sample data are characterized by a pH level of 6, with over 80% of the points in
this category. The vineyards have a different pattern, with less than 25% of the
polygons having a pH of 6, and 60% having a pH of 7.
Although there appears to be a similar distribution pattern between the
sample and vineyard data sets for cation exchange capacity (CEC), the
magnitudes of concentration are different (Figure 4-10). Approximately 65% of
the sample points fall between 10 to 12 mEq/100g, while approximately 70% of
the vineyards are in the same range. The highest concentration of sample
points is at 11, while for vineyards it is 12 mEq/100g. The sample data set also
exhibits higher percentages of points at 16 and 21mEq/100g.
Soil permeability (Figure 4-11) is virtually the same for both data sets.
Eighty-one percent of the sample points, and 90% of the vineyard points fall
I I I I I I
No DataVery Low Low Moderate High V. High Variable
SCS Fertility Class
|3 Vineyard (n=1,006) | Sample (n=14,385)
FIGURE 4-7: SOIL FERTILITY
(Source: Soil Conservation Service, 1965; 1974)
-9 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17
Moisture Percentage (-9=No data)
S3 Vineyard (n=1,006) m Sample (n-13,643)
FIGURE 4-8: SOIL MOISTURE
(Source: Soil Conservation Service, 1965; 1974)
-9 0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5 5.5 6 6.5 7
Soil pH (-9 = No data)
f 3 Vineyard (n=869) = Sample (n-14,111)
FIGURE 4-9: SOIL ACIDITY
(Source: Soil Conservation Service, 1965; 1974)
bU ------- ---
5 0 .................
7 0 - - -- - - -- - - -- - - -
50 1----.............- ..................
2 0 1 ....................
I II I I i I j i I l i I ] I I I I
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31
2 4 6 8 10 12 14 16 18 20 22 24 26 28 30 32
Sodium (Na) in mEq/100 grams of Soil
I7M Vineyard (n-869)
M Sample (n-13,643)
FIGURE 4-10: CATION EXCHANGE CAPACITY
(Source: Soil Conservation Service, 1965; 1974)
Very Slow Mod. Slow Mod. Rapid Very Rapid
Slow Moderate Rapid Variable
SCS Permeability Class
1(3 Vineyard (n-1,035) M Sample (n-17,017)
FIGURE 4-11: SOIL PERMEABILITY
(Source: Soil Conservation Service, 1965; 1974)
within the moderate permeability class. A small percentage (less than 5 each)
falls within the moderately rapid and rapid classes for the sample points.
With some minor variations, both data sets are dominated by the well
drained class of natural soil drainage. This class comprises approximately
83% of observations in both data sets (Figure 4-12). The sample data set
shows some excessively and moderately drained observations, while the
vineyards show less than 10% of the polygons are moderately drained.
Figures 4-13 through 4-15 are histograms of soil sand, silt, and clay
content respectively, for the vineyard and non-vineyard data sets. The
percentage content classes were derived from the textural classes established
by the Soil Conservation Service (1975). These classes establish the ranges of
percentage sand, silt, and clay content that define a particular soil series. Two
methods were used to established the values used in this study. Laboratory
analysis data measuring actual content of representative profiles were available
for the Sierra coarse sandy loam and Auburn silt loam soil series. The mean
value for these analyses, averaged over the effective root depth, were used for
the values shown in the figures. Median class values derived from the Soil
Conservation Service (1975) categories were used for the Ahwanee loam, and
Sierra sandy clay loam, due to the absence of laboratory analysis data. The
median values were also averaged over the effective root depth of the soil
profile. A small percentage (approximately 2%) of the vineyard data set had
No Data Poor Moderate S/W Excess. Variable
Very Poor S/W Poor Well Excess.
SCS Drainage Class
Vineyard (n=1,035) M Sample (n-16,745)
FIGURE 4-12: NATURAL SOIL DRAINAGE
(Source: Soil Conservation Service, 1965; 1974)
6 0 ............................--
10 ..-------. -------
I I i
No Data 20.87 22.20
3 Vineyard (n=1,035) M Sample (n-14,385)
FIGURE 4-13: SOIL SAND CONTENT
(Source: Soil Conservation Service, 1965; 1974)
60 --------- ------------ -----1--- -
50 ------. --.-.. -
3 0 ....... ......................................
No Data 11.00
i Vineyard (n=1,035) j Sample (n-14,385)
FIGURE 4-14: SOIL SILT CONTENT
(Source: Soil Conservation Service, 1965; 1974)
70 -1 ...-........................... ..-- .--------..------.-.-...-.....-........
6 0 -.................................................. ..-... ... ..... .................................-------- .---
2 0 ... ..... ..................
No Data 13.73
[E= Vineyard (n=1,035) M Sample (n=14,385)
FIGURE 4-15: SOIL CLAY CONTENT
(Source: Soil Conservation Service, 1965; 1974)
incomplete observations on one or more texture classes, and were
subsequently grouped in the "No Data" category.
As seen in Figure 4-13, both the vineyard and non-vineyard (sample)
data sets fall in primarily the 22.20% and 25.00% sand classes, which describe
the Sierra coarse sandy loam and Ahwanee loam, respectively. The results for
the silt and clay (Figures 4-14 and 4-15) analyses also fall within the percentage
classes which define the Sierra and Ahwanee series.
Slope angle, compared in Figure 4-16, show the bulk of the vineyard
observations (approximately 70%) between 2% and 10% slopes, with a distinct
skewed distribution towards the steeper angles. The sample distribution has
greater than 50% of its observations at a slope angle of 12% or steeper, up to a
maximum of 92% slope.
Slope aspect, shown in Figure 4-17 exhibits a distinct predominance of
westerly slopes in the vineyard data set. More than 68% of the observations fall
within the western aspect categories, with west and west southwest containing
the highest percentages. In the sample data set western and northerly aspects
The preceding visual comparison can be summarized into several
apparent differences and similarities between the vineyard and non-vineyard
(sample) areas. Vineyards can be characterized as occupying lands more
J== g i
NNE NE ENE E ESE SE SSE S SSW SW WSW W WNW NW NNW N
Aspect in Cardinal Direction
SVineyard (n-1,035) M Sample (n-16,915)
FIGURE 4-17: SLOPE ASPECT
(Source: TIN Model)
capable for agriculture than non-vineyard areas based on the Storie Index.
Soils are deeper and have a higher available water-holding capacity (i.e. there
is more water available within the rooting depth to the grapevine roots). This is
also apparent in the higher percentage of vineyards with a deep effective
The surficial runoff rate is one indicator of the difference in surface
morphology between vineyard and non-vineyard areas. The predominance of
the medium class indicates more gradual slopes in vineyard areas,
concentrated in the 2% to 10% slope angle range. This is reaffirmed by the
frequency of lower slope angle values occurring in the vineyards. In addition,
the presence of rapid and very rapid class observations in the non-vineyard
areas reflect its greater variability in surface morphology, and generally steeper
slopes. This is consistent with the requirement for vineyards that are
"navigable" by farm equipment. Soil fertility and moisture levels are very similar
for both areas, based on the Soil Conservation Service classes. Soils are more
acid in the non-vineyard areas than in the vineyards. Although the apparent
difference is one unit (6 versus 7), the pH scale is logarithmic, with a change in
1 unit equivalent to a multiplier of 10. Thus the non-vineyard areas are 10 times
more acidic (in terms of free hydrogen ions) than the vineyards (Galletta and
Himelrick, 1990). Soil acidity affects availability of nutrients and biological
activity in the soil--the higher the pH, the lower the availability of organic such
as nitrogen, potassium and phosphates, and also microorganism activity. In
addition, acidic soil facilitates the availability of certain minerals which can be
harmful to plants in high concentrations, including zinc, manganese and copper.
One estimate of the optimum soil pH, 6.5, within the desirable range of 5.5 to
7.0, is provided by Galletta and Himelrick (1990). This range encompasses
most of the variation found in the study area.
Related to acidity in terms of soil nutrients is cation exchange capacity
(CEC). This is the amount of the negative charge of the soil, or the number of
exchangeable cations a soil can hold (Galletta and Himelrick, 1990; Thorne and
Peterson, 1954). It affects plant nutrient availability, and is a function of soil
texture, acidity, moisture levels, and clay and organic matter content. Typically
the CEC level decreases with increasing acidity, and can give some inference
as to the presence of clay and organic matter in the soil (Buol, 1989). Both
data sets show a predominance of a moderate CEC, although the non-
vineyards show a greater variability in CEC levels. CEC levels can be modified
by the application of lime, fertilizers, and mineral supplements. The variability in
the non-vineyard areas may reflect the occurrence of other agricultural activities
such as row crops and orchards in the study area.
Permeability and natural drainage are virtually the same for both areas,
although once again there is greater variability in the non-vineyard areas. This
relationship reflects the soil texture, with the higher levels of sand and lower
levels of silt and clay content predominant in the non-vineyard areas.
Patterns of slope angle show a greater variability, and substantially
steeper slopes in the non-vineyard areas. This may reflect a cultural vineyard
management effect, once again due to the need for farm equipment navigability
in the vineyards. Patterns of slope aspect show a strong southern and western
trend for the vineyards, which maximizes the daily solar radiation received.
Following from the methodology detailed in Chapter 3, the results will be
presented in three sections: correlation, factor analysis, and application of the
Mann-Whitney U test.
The results of the Spearman's r. correlation procedure provided marginal
insight into relationships between the variables, and more importantly, formed
the basis for the factor analysis.
Cross-correlation between the variables yielded some very general
trends, especially between the model and validation data sets. Once the
original data for all three sets were cleaned of observations with incomplete
records (i.e. those not containing values for each variable), the resulting sample
sizes were n =537 for the model, n= 273 for the validation, and n= 11,326 for the
sample data sets. The correlation matrices were examined using an arbitrary
Spearman correlation coefficient threshold of 0.50. All of the coefficients above
this level were significant at the 99% confidence level. Substantial differences
were identified as those being 10 or more "correlation units" above or below any
The variables runoff, fertility and CEC all showed higher correlations with
the other variables in the model and validation data sets. Soil moisture showed
lower correlations in these same sets. The variables slope angle and aspect
(after sine transformation) exhibited a random pattern of correlation between all
three data sets. Unfortunately, as mentioned above, this technique provides
little useful insight in the variance and relationships of the observations. Only
general patterns of interaction amongst variables can be discerned. The results
indicated that certain of the variables behave consistently within the vineyard
data sets, and that there appears to be a random distribution of slope angles
and aspect. Table 4-1 summarizes these results. Once again, the real value of
this matrix lies in its role as the interim product for input to the factor analysis
Following on the discussion of the factor analysis procedure presented in
Chapter 3, this section presents the analysis of results for the chosen approach.
An initial pattern factor analysis (FA) was undertaken using the squared
correlation coefficients from the Spearman's r, matrix. Coefficients are squared
to remove the negative sign. The initial FA yields the same number of factors
+ + ----+ +----- +
.^- .^- .^- .^ ^
oc m (oDl .9 0U .a) mw -
as variables (15 in this study) and the contribution of each factor in explaining
the variance of the observations on the variables. An eigenvalue represents
that portion of the variance of the ranked observations on each variable
explained by each factor.
Since the primary intent here was to reduce the number of variables for
subsequent analysis, a scheme had to be developed to reduce the number of
factors. This problem, the choice of the appropriate number of factors, was
alluded to previously as the source of much debate. It was decided that two
so-called "rules of thumb" would be used to assist in this problem. The first is
known as the Scree Test, as named by Cattell (1965), and refers to the position
of each factor plotted against its eigenvalue. The point at which the slope of
the curve begins to level off in a consistent horizontal direction indicates where
the factors are measuring random errors (Davies, 1984; Kim and Mueller, 1978).
The second rule of thumb is advanced by Davies (1984) among others, and is
known as the "five percent rule." This rule informs that a threshold value of 5%
of the total variance of the data set be established, and that all factors
explaining less than this threshold be discarded. Applying both of these rules
to the initial FA of the three data sets resulted in generally good agreement.
The Scree test showed a clear break at 5 factors for the validation and
sample data sets, and an acceptable difference for the model data set. The
5% threshold provided very good agreement with the sample data set and an
acceptable level for the model and validation data sets. The factors were then
rotated using the Varimax rotation option in the SAS (1988) software, to
maximize the variable loadings, or correlations on the 5 chosen factors.
A brief description of the structure of the data sets, based on factor
loadings, is appropriate. It is important to keep in mind that the relationship
between factors and variables, and amongst variables reflects association, or
correlation, but not cause and effect. Structure in the data set refers to the
existence of patterns of association among variables, as identified by the
factors. An analysis of the loadings of the variables on each factor was
conducted by highlighting all factor loadings above 0.317. This threshold
identifies those factors which account for 10% or more of the variance of each
variable. This threshold has been accepted as a reasonable starting point for
data set interpretation (Davies, 1984; Cattell, 1978). In practice, the lowest
loading used in this analysis was 0.46 (approximately 21% of the variance
accounted for). Based upon the level and arrangement of loadings, four
consistent factors were discernible from the model and validation data sets, and
three from the sample data set. These factors were named: Profile, Fertility,
Capability, and Physiography. It should be noted that the Physiography factor
in the validation data set (and all other validation factors) lacked a significant
loading on aspect. The factor numbers and the variables comprising each are
presented in Table 4-2. The absence of a physiography factor in the sample
data set reflects the lack of a significant factor loading for the aspect variable,
and the inclusion of the slope loading (in this data set only) with the Capability
TABLE 4-2: MAJOR FACTOR NAMES
V:2 Profile Sand
M:2 Water-holding capacity
V:1 Fertility Rooting depth
* Not Significant Fertility
V:3 Capability Grade
* Significant in
V:4 Physiography *Aspect
* Significant in
M# = Model Factor
S# = Sample Factor
V# = Validation Factor
factor. Incidentally, this is consistent with the determination of the Storie Land
Capability Index rating, as discussed previously.
Factor 5 in each data set can be considered a minor factor, due to its
general lack of, and variability, in significant loadings. In the model data set
there were no significant loadings, while the validation set had a significant
loading on soil fertility. The sample data set had a significant loading on soil
moisture. Although the results for this factor were marginal, it is generally
considered better to extract a greater number of factors rather than a lesser
number, to avoid excluding major factors (Davies, 1984). In addition, the four
major factors identified not only provide a parsimonious description of a
complex data set, but they explain a substantial amount of the variance in that
data set. The common variance explained in the model data set exceeds 94%,
while the validation and sample factors explain approximately 96% and 90%
The final phase of the statistical analysis involved comparison of the
sampling distributions of the data sets. As discussed in Chapter 3, the Mann-
Whitney U test is an appropriate method to compare the mean ranks of each
distribution. The null hypothesis was that the data sets were all drawn from the
same population. That is, that there is no statistically significant difference
between the values of the mean ranks of the data sets as represented by the