Land suitability for viticulture


Material Information

Land suitability for viticulture a geographic information system-derived method for a Mediterranean type climate in California
Physical Description:
ix, 132 leaves : ill. ; 29 cm.
Watkins, Russell L
Publication Date:


Subjects / Keywords:
Viticulture -- Research -- California   ( lcsh )
Geographic information systems -- California   ( lcsh )
bibliography   ( marcgt )
theses   ( marcgt )
non-fiction   ( marcgt )


Thesis (Ph. D.)--University of Florida, 1995.
Includes bibliographical references (leaves 117-130).
Statement of Responsibility:
by Russell L. Watkins.
General Note:
General Note:

Record Information

Source Institution:
University of Florida
Rights Management:
All applicable rights reserved by the source institution and holding location.
Resource Identifier:
aleph - 002051798
notis - AKN9757
oclc - 33436629
System ID:

Full Text








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.



LIST OF TABLES ................
LIST OF FIGURES ...............
ABSTRACT .....................

1 INTRODUCTION .................
Research Problem ................
Hypothesis ......................
Setting of Study ..................

Relevance ................
Land Evaluation Literature ...
Implications of GIS and Relevant
Viticultural Literature ....... .


Literature..... ..
. . .

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 .......

.... 99
.... 107
.... 109

APPENDIX ................................ ..... 113
















SUITABILITY STUDIES ............................ 31




MAJOR FACTOR NAMES .......................... 90

SUMMARY OF STAGE ONE ........................ 93

SUMMARY OF STAGE TWO ........................ 95

SUMMARY OF STAGE THREE ...................... 97
























STUDY AREA LOCATION ..............


GENERAL GEOLOGY .................

GENERAL SOILS ....................


PROCEDURES ......................




SO IL DEPTH ........................


SOIL RUNOFF ......................

SOIL ROOTING DEPTH ................

SOIL FERTILITY .....................

SOIL MOISTURE .....................

SO IL ACIDITY ........................

. . 9

. . 13

. . 15

. . 17

.......... 40

........... 44

.......... 47

.......... 60

.. . 63

........... 64

. .. 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



Russell L. Watkins

May 1995

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

evaluation method.

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."

Research Problem

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

of formats;

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

which includes:

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,

p. 164).

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

geographic area;

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



0)' "V




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



SZinfandel vineyards

1 0 1 2 Kilometers
f -- I I I I


composes the Sierra Nevada range (McKnight, 1990; Ritter, 1978; Soil

Conservation Service, 1974). Figure 1-3 shows the general geology of the

study area.

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

General Geology
Mehrten Formation
S Valley Springs Formation
"Auriferous" Gravels
| Mariposa Formation
SLogtown Ridge Formation
F Calaveras Complex Volcanic Rocks
SMesozoic Granitics 1



0 1 2 Kilometers



General Soils
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 --



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.



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

geographic analysis.

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.

Viticultural Literature

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.

Variable Definition

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



c :



C )
0 0





0) -









o E
0 E0 (0





a ,


( -o


", >

c ?
= 0 a

r C

c c

3. 0)-

0a a

0 00

w ):s



0. >0

0 1-
0 (m

0I a

t 5 o




0 0






u 0

= c






0 0


(D 0

i -
01 CD


3 >





C >
Cj oa


- 0
a o







L i






0) _
a 2


)( D
" F

(0 t


^- 8




u m
0 <)



i ti

(D a







(1) .





'a c
Co %




o I

c t









0 0







4- -

C 0


> U)L

%- C 0
0 >






cn a~





E 1






Dutt, et al., 1981
Fisher, 1978
Galleta & Himelrick, 1990
Halliday, 1993
Onokpise & Mortenson, 1988
Rankine, et al., 1971
Rieger, 1991
Sequin, 1986
Scienza & Falcetti, 1990
Smart, et al., 1985a
Stevenson, 1986



North Florida
So. Australia
NW California
Northern Italy


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
T= Type
Te = Temperature
Tx = Texture
WH Water-holding capacity

(A) Applied = original study or experiment
(S) Summary = summary of other work or literature


Variable (by class)


slope angle (Q)
slope aspect (Q)
elevation (D)


type (D)
texture (D)
depth (Q)
permeability (Q)
available water holding capacity (Q)
drainage class (Q)
runoff class (Q)
effective root depth (Q)
fertility (Q)
soil moisture (Q)
pH (Q)
cation exchange capacity (Q)
Storie Index Rating (Q)
Storie Index Rating Grade (Q)

Geology (D)


temperature (D)
precipitation (D)


(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).

Data Sources

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.

Data Input

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
units longitude/latitude
- projected into Lambert conformal conic
- secondary projection into Universal
Transverse Mercator
- 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
supplement contours
- generalize contours
- set weed tolerance of 10 meters
- add or remove points to eliminate
flat triangles

4* --

- check plots f
- digitizer RMi
- tic overlay a

or visual alignment
3E 0.002
accuracy 3.5 meters




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

4.5 hectares).

Data Manipulation

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

name, width, type

tabular data from Soil Surveys
elevation data from USGS quads 33
geologic categories < Q
ownership of vineyards

use add from command
create errror log file

( review data for missing, altered or
mislabeled records

convert TIN model to a polygon coverage

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
statistical software


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.

Statistical Methods


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

Descriptive Statistics:
removal of incomplete records
mean, median, high, low values
scatterplots of variables
frequency distributions

^ _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
variance thresholds
n specified
Varimax orthogonal

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/)


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


Data Reduction

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

Johnston (1978).

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

Chapter 4.

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;

Cattell, 1978).

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:

E(S)=(n,(n, +n2+1))12

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.

Climate Description

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

(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

tailings, riverwash).

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


sUOTqVAZ6s qO;o




(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)

(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.

Very Low Low Moderate High Very High Variable
SCS Water-Holding Capacity Class

SVineyard (n=1,035) M Sample (n-16,747)

(Source: Soil Conservation Service, 1965; 1974)

...... ........................................

........................- -- ..... -----------..- -- -

....... ......-.. -. -... -............... ........- -

............ .. .......... ""


Slow Slow Med.

.......... ... ..................... .......

=3 Vineyard (n=1,035) M Sample (n-16,733)


(Source: Soil Conservation Service, 1965; 1974)




g 40





V. Shallow Shallow Mod. Deep Deep Very Deep Variable
SCS Root Depth Class

x3 Vineyard (n=1,035) J Sample (n-16,747)

(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

No DataVery Low Low Moderate High V. High Variable
SCS Fertility Class

|3 Vineyard (n=1,006) | Sample (n=14,385)

(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)

(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)

(Source: Soil Conservation Service, 1965; 1974)

bU ------- ---

5 0 .................
50 ........

40 -----

30 ----.---

210 -----

7 0 - - -- - - -- - - -- - - -

50 1----.............- ..................

40 1.......

30 ............

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)

(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)

(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)

(Source: Soil Conservation Service, 1965; 1974)

70 -1------------

6 0 ............................--

50 .----------

40 -

30 -4.....................

10 ..-------. -------


I I i
No Data 20.87 22.20
Percentage Sand

.. .

. .



3 Vineyard (n=1,035) M Sample (n-14,385)

(Source: Soil Conservation Service, 1965; 1974)

60 --------- ------------ -----1--- -

50 ------. --.-.. -

40 --

3 0 ....... ......................................


I 1

0o Da
No Data 11.00

13.00 33.32
Percentage Silt


i Vineyard (n=1,035) j Sample (n-14,385)

(Source: Soil Conservation Service, 1965; 1974)

70 -1 ...-........................... ..-- .--------..------.-.-...-.....-........

6 0 -.................................................. ..-... ... ..... .................................-------- .---

50 ------

2 0 ... ..... ..................

10 -1...........................---


No Data 13.73

Percentage Clay

[E= Vineyard (n=1,035) M Sample (n=14,385)

(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


Comparison Summary

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


o a



', r

*= a

(n CTNoe
o a

J== g i

tI in

aa 5

mo V

Aspect in Cardinal Direction

SVineyard (n-1,035) M Sample (n-16,915)

(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

rooting depth.

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.

Statistical Results

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

particular coefficient.

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


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


+ 7


+ + ----+ +----- +
.^- .^- .^- .^ ^


+ +


+ +
> +>
+ +




++ +





+ 2

oc m (oDl .9 0U .a) mw -




c r

0 c

I- 0

2 -
CR 0

o o
0 0

-o o

.0 CuO

o o


o C.




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


Data Set:
Factor Number



M:1 Depth
V:2 Profile Sand
S:1 Clay

M:2 Water-holding capacity
V:1 Fertility Rooting depth
S:3 *Moisture
* Not Significant Fertility
in Sample

M:3 Storie
V:3 Capability Grade
S:2 Runoff
* Significant in
Sample Only

M:4 Slope
V:4 Physiography *Aspect

* Significant in
Model Only


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%


Model Validation

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