Adina Merenlender Final Report

A Spatially Explicit Vineyard Expansion Model: Addressing Crop Production, Public Policy, and Environmental Concerns

Final Report - September 1999

Principal Investigator:
Adina Merenlender
University of California, Berkeley
Hopland Research and Extension Center
4070 University Rd.
Hopland, CA 95449
(707) 744-1270, adina@nature.berkeley.edu

Research Team:
Colin Brooks, Emily Heaton, David Newburn

Cooperators:
Sonoma County Grape Growers Association
Circuit Rider Production Inc.

Location of project: Sonoma County

Commodities: winegrapes

Funding:
Funded FY 1998-99: $20,000
Funded FY 1997-98: $20,000
Funded FY 1996-97: $22,000

Summary

Vineyards are expanding rapidly in California's north coast due to a booming wine market. In Sonoma County, much of this expansion is occurring on hillsides that harbor California's remaining oak woodlands. These natural areas support a majority of the region's biodiversity, provide ecosystem goods and services and scenery valued by residents. This project was designed to provide an assessment of where vineyards have been planted, where they are likely to be developed, and address the natural resource and environmental policies that are associated with agricultural expansion in Sonoma County. The specific objectives of our research were to 1) integrate vineyard maps with physical and habitat data for Sonoma County in a GIS to assess patterns of vineyard development; 2) make a model of areas suitable for future vineyard development; 3) use the GIS to evaluate proposed and adopted regulations; and 4) evaluate the risk of habitat loss and fragmentation. This research project provided analysis that is designed to promote sustainable agriculture at the landscape scale. Also, this information is incorporated into a planning tool that is being used by Sonoma County for planning purposes thereby meeting the University of California Sustainable Agriculture Research and Education Program's priority to integrate crop production issues with public policy issues and community development.

Background

California wines have become extremely popular nationally and internationally, leading to increased demand for wine grapes. In 1996 California wine grapes were worth over $1 billion, as compared to $647 million in 1988. This increase is due both to increased price per ton and continued establishment of approximately 40,000 hectares of new vineyards throughout California in the past decade, a 30 percent increase in vineyard land. The California wine boom has been an economic blessing for many rural landowners who can now reap significant value from developing their land into vineyard. In the current market, wine grape growers can expect as high as $4,000 profit per acre vineyard land in full production. Many communities are fostering continued vineyard development and marketing regional wines to increase local tourism, another import element of California’s economy. Therefore growing quality wine grapes is an important component of California agriculture.

Photo 1. Example of new hillside vineyard in Sonoma County

Upland areas, historically considered marginal agricultural land, are increasingly targeted for vineyard development (Photo 1). In some areas, the conversion from woodlands and forestland to vineyard is extensive. For example, Santa Barbara County Planning Department reported that the amount of vineyard has doubled to 8,000 hectares since 1996 leading to the loss of over 2,000 oak trees, a larger number than all rural development and subdivisions were responsible for removing in the previous ten years. Even in lesser-known wine grape growing areas such as San Luis Obispo County, an estimated 800 hectares of rangeland is being converted to vineyard a year. In Sonoma County, we have estimated that at least 660 hectares of dense oak woodland were lost to vineyard development between 1990 and 1997. Sonoma County has also seen a 31.5% growth in vineyard area in that time period (4720 hectares), for a total of 19,707 ha (see figure 1 for the location of Sonoma County).

Vineyard owners are coming under increasing scrutiny from urban neighbors, the environmental community, and government agencies concerned about the effect of vineyards on natural resources. For example, riparian vegetation clearing, wetland conversion, endangered species, hillside erosion, native tree removal, and impeded wildlife migration corridors have become contentious issues facing vineyardists in California. The expansion of vineyards especially on hillsides resulting in changing view-sheds, deforestation, and watershed degradation has spurred a flurry of farmer neighbor conflicts across coastal California. Complaints against continued vineyard development have been repeatedly aired in local newspapers. In response to citizens concerns, local regulatory policies are rapidly evolving to prevent hillside erosion and protect natural resources. In particular, Napa and Sonoma Counties instigated hillside agricultural development restrictions to prevent soil loss and protect stream corridors. These local ordinances often require farmers to register new vineyards and represent some of the first restrictions to agricultural development in California. Multi-stakeholder committees have developed the guidelines for these regulations. This represents a method of policy development growing throughout the country to settle environmental issues. In Santa Barbara County, a narrowly defeated oak woodland protection ballot initiative lead to a renewed interest in finding a compromise between agricultural and environmental interest groups. A set of guidelines to avoid large-scale oak tree removal and to maintain viable oak habitats was developed by a multi-stakeholder committee in Santa Barbara County.

Objectives

The goal of this project was to provide an assessment of where vineyards have been planted, identify other areas suitable for development, and to address the natural resource and environmental policies that are associated with agricultural expansion in Sonoma County. The specific objectives of our research were to 1) integrate vineyard maps with physical and habitat data for Sonoma County into a GIS to assess patterns of vineyard development; 2) make a model of areas suitable for future vineyard development; 3) use the GIS and to evaluate proposed and adopted regulations; and 4) evaluate the risk of habitat loss and fragmentation.

Methods

GIS Development

We were able to build vineyard expansion models for Sonoma County because vineyards had been mapped by the Sonoma County Grape Growers Association and digitized by Circuit Rider Productions, Inc (CRP). Two sources were used for vineyard mapping: 1990 non-orthorectified aerial photographs, and maps provided by the grape growers for the 1990 to 1997 time period. This map shows vineyards for most of Sonoma County except for the coast, including the Alexander Valley, Russian River Valley, Chalk Hill, Dry Creek Valley, Knights Valley, Los Carneros, Sonoma Mountain, and Sonoma Valley appellation areas (figure 2). The vineyard location data was provided as AutoCAD DXF files and converted to an Arc/Info GIS layer by us for further analysis. We labeled each vineyard polygon with a pre- or post-1990 establishment date based upon the source data.

This data was then incorporated into a geographic information system (GIS) and integrated with other GIS data that captures the physiographic and climatological factors that might affect where vineyards are established. The primary physiographic variables we analyzed were slope, aspect, and elevation. These variables influence a vineyard's microclimate. We also analyzed other available variables from our GIS that might be influential on where a vineyard is established. These were distance from perennial streams for post-1990 vineyards, proximity of post-1990 vineyard to a pre-existing (pre-1990) vineyard, distance from roads for post-1990 vineyards, existing land-use, designated appellation region (also known as viticultural area), and existing vegetation. Interviews with viticulture experts confirmed that these variables could be influential.

The GIS data collected to analyze these variables came from a variety of sources. The slope and aspect layers were derived from 30-meter USGS Digital Elevation Models (DEM) that were downloaded from the California Geographical Survey website at the California State University at Northridge (http://geogdata.csun.edu/). The streams and roads layers were 1:100,000 TIGER94 files also downloaded from the CSU-Northridge web site. Proximity to existing vineyards was calculated by using a data layer of Sonoma County's vineyards described above. We used the California Department of Conservation's Farmland Mapping and Monitoring Project (FMMP) "farmland map" of Sonoma County to show land-use. In addition to showing the locations of farmland and grazing land, this layer also includes the extent of urban areas in the County. We used data from the 1990 version of the FMMP. Mean annual precipitation was obtained from the Teale Data Center under a site license with the University of California - Berkeley. Appellation region was obtained from CRP along with the vineyard locations layer. We used a California Department of Forestry hardwood vegetation layer based on 1990 satellite imagery.

The 1972 Sonoma County soil survey, created by the Soil Conservation Service, predecessor of the Natural Conservation Service, was not available in a GIS format, so we digitized the data completely for the Alexander Valley and Dry Creek appellation areas as an Arc/Info coverage. We are also in the process of digitizing additional soils data for the Russian River Valley and Sonoma Valley appellation areas.

Model of Suitable Areas for Future Vineyard Expansion

Logistic regression analysis is appropriate when the response variable is binary (e.g. 1 = post-1990 vineyard, 0 = not post-1990 vineyard) and when independent variables are categorical or continuous. For this project, logistic regression allowed us to calculate the probability that a given unit of land is suitable for vineyard development based on its physical characteristics and recent expansion patterns.

The data explored for this model included distance to nearest urban area, 1990 California Department of Forestry "hardwood pixel" layer, pre-1990 and post-1990 vineyard locations, elevation, slope, aspect, agricultural and urban landuse from the 1990 Farmland Mapping and Monitoring Project, distance to roads, distance to existing (pre-1990 vineyards), distance to perennial streams, precipitation, and public land. All layers were converted to grids with a resolution of 1 ha (100 m x 100 m). All 1 ha grids were created using a reference grid of the same size so that grids could be easily overlaid. Each hectare represented a single observation for statistical analysis and had a corresponding value for each of the data layers.

We eliminated land not available to vineyard development so it would not be considered in our model. This land consisted of public land, urban areas, areas already in vineyard, elevations greater than 800 meters, slopes greater than 50 percent (26.6 degrees) and water bodies. The cutoffs of 800 meters and 50 percent were chosen because the Sonoma County vineyard layer had no vineyards above 800 meters, and a new county ordinance prohibits development on slopes greater than 50 percent.

In order to test the ability of models to predict vineyard expansion, spatially distinct post-1990 vineyards were split into 2 sets, a model building set and a model testing set. Random sampling of these spatially distinct vineyards was stratified by appellation area and size class. Vineyards were grouped into 3 size classes: large (> 100 ha), medium (<100 ha and < 10 ha), and small (<10 ha). One-half of the vineyards within each appellation-size class were randomly assigned to the build set and the other half to the test set. Stratification by appellation region ensured that vineyards in each set were well distributed throughout the county. Stratification by size ensured that vineyard size distribution was similar for each set, resulting in similar numbers of vineyard observations. The build set had a total of 2233 post-1990 vineyard observations, while the test set had a total of 2132. In order to randomly select non-vineyard observations, a 100 m grid of random values was created using the RANDOM command in GRID. From this grid, another grid was derived that contained these values only for 1990 available cells that did not become vineyard. The desired number of non-vineyard cells was selected by taking that number of cells with the largest random number assignments. All model testing was conducted using a single data set with a 1:1 ratio of non-vineyard to vineyard observations. Non-vineyard cells were selected using the random grid from above.

Site characteristics (e.g. vineyard or non-vineyard, elevation, slope, distance from urban areas, etc.) were extracted for each model building cell using the SAMPLE command in grid. The resulting ASCII file was imported into EXCEL for data manipulation and then into SAS for analysis. Each cell and its related characteristics were treated as an independent observation in statistical analyses.

The most complex logistic model included 35 explanatory variables. These variables represented elevation, slope, aspect (north, northeast, east, southeast, south, southwest, northwest, flat; west was used as baseline), vegetation (blue oak woodland, coast live oak woodland, montane hardwood, unclassified hardwood, conifer, shrub, urban, other; grass was used as baseline), land-use (primary farmland, unique farmland, farmland of local importance, farmland of state importance, other; grazing land was used as baseline), distance to nearest perennial stream, distance to nearest road, distance to nearest pre-1990 vineyard, and distance to nearest 1990 urban area. Eight of the variables were interaction terms: aspect*slope (north*slope, northeast*slope, east*slope, southeast*slope, south*slope, southwest*slope, northwest*slope; west*slope used as baseline) and elevation*slope. Aspect*slope takes into account changes in the effect of aspect on vineyard suitability as slope changes, and elevation*slope takes into account changes in the effect of elevation on vineyard suitability as slope changes.

Using the most complex model as a starting point, categorical variables were combined using backward stepwise regression and information theory in the following manner. Each logistic regression model received a Akaike Information Criteria (AIC) value based on its ability to describe the input data (see Burnham and Anderson, 1998). The lower the AIC value, the closer to "truth" the model is believed to be (Akaike, 1973). In performing a backward stepwise regression, we repeatedly selected the model with one less variable that had the lowest AIC value. For example, since we started with a model with 35 variables, the next step was to build multiple models with 34 variables. In doing so, we chose to remove a variable (making it part of the baseline) or combine 2 variables into a single variable (e.g. blue oak woodland and coast live oak woodland may be combined into a single variable that represents oak woodland). The 34-variable model that had the lowest AIC value was chosen as the best model with 34 variables. Multiple models with 33 variables were then built using the best 34-variable model as a reference. The best 33-variable model was selected using AIC values. This process continued until no model with a lower AIC value could be found. Out of all models built, the model with the lowest AIC value was chosen as the final model. Continuous variables were not removed at any point in this process.

In order to determine the classification accuracy of the final model, the equation from the model was used to calculate the probability value for each cell in the model testing sample. The 2132 (the actual number of post-1990 vineyard observations) cells with the highest probabilities were classified as vineyard, and the 2156 (the actual number of non-vineyard observations) cells with the lowest probabilities were classified as non-vineyard. For each cell, the classification was compared with whether or not the cell actually became vineyard. Classification accuracy was calculated as

(number of cells classified correctly/total number of cells classified) * 100%

A probability surface was created for the entire county by applying the equation from the final model to each cell. This was done using map algebra in GRID.

Policy Analysis

In March 1998 we were contacted by a multi-stakeholder committee charged with developing a regulation to restrict some hillside vineyard development practices to protect natural resources. The committee restricted their objectives to concerns over soil erosion and water quality. During this process we provided both the farmers and the environmentalists on this committee with GIS analysis that allowed them to compare the total vineyard acerage effected by various regulatory options. For example, which vineyards and grape growing areas would be effected if vineyards were limited to 30% slope as compared to 50%. In May of 1999, the Sonoma County Board of Supervisors adopted an ordinance that regulates vineyard planting and replanting. The ordinance requires than people planting or replanting vineyards register with the county agricultural commissioner, and submit an erosion control plan on steep slopes. The slope categories are 0-15% (Level I), 15-30% (Level II), 30-50% (Level III), and greater than 50% (Level IV) for non-highly-erodible soils, and are 0-10% (Level I), 10-15% (Level II), 15-50% (Level III), and greater than 50% (Level IV) for certain designated highly erodible soils. Level I vineyard plantings and replantings only require registration with the agricultural commissioner and use of the county's normal erosion control measures. Level II and III vineyards plantings require registration and an approved erosion control plan. Level IV vineyard plantings are not allowed under the ordinance, unless greater than 50% areas form less than 7.5% of the planting, and are not near the edge of the planting.

We mapped the four slope classes from the ordinance using the 30-meter USGS DEM. This slope map (Figure 5) was intersected with the current vineyard map and areas identified from our model as suitable for future vineyard development. The number of vineyard acres that fall in each of the slope classes was calculated for all vineyards developed prior to 1990. We also did this analysis for 4720 hectares of the most suitable vineyard sites according to our model. This amount was used because that is the amount of vineyard that was planted from 1990-1997 according to the available data. The same analysis was also done for all the acerage that had a probability score of greater than 0.5 for our suitability model which totaled to 64,298 hectares. As the soils data were not finished for the entire county, we treated all areas as being non-highly erodible soils.

Habitat Fragmentation Analysis

In order to identify where vineyard expansion may lead to forest fragmentation in the future, we identified changes in habitat connectivity if vineyards were developed in areas with a suitability level of p>0.5 using the logistic regression model. We calculated core areas of oak woodland using the "core.aml" program from Shawn Saving and Greg Greenwood of the California Department of Forestry (CDF). This program uses a GRID layer of a county's vegetation to calculate the core areas of oak woodland. A core area was defined as at least 100 hectares in size, at least 50 meters (two 25 meter pixels) away from urban areas, and within 25 meters (1 pixel) of other oak woodland pixels. These land-cover classes were available in the "CDF hardwood pixel" vegetation layer based on 1990 Landsat Thematic Mapper imagery that we had previously obtained from CDF. The core variables are user-definable, but we used those suggested by Shawn Saving.

Using the core program, we calculated the core oak woodland at the "current" time period, which reflected the 1990 date of the source imagery. We then took our p>0.5 development scenario from the logistic regression model and converted all of the greater than 0.5 development probability areas that were in oak woodland to the "grass" class, which is what most vineyards had been classified as by the CDF. We ran the core program with this reclassified layer and obtained the number and size of the core areas under a possible vineyard expansion scenario and noted areas where development would fragment large forest patches and make smaller forest patches disappear.

Results

GIS Analysis of Sonoma Counties Vineyards

The results from this research project currently provides an assessment of where vineyards are in Sonoma County and addresses the related natural resource issues and local policies that are associated with hillside agricultural expansion. Vineyards were mapped by a local grape growers association from 1990 non-orthorectified aerial photographs and more recent vineyard developments were added by the growers themselves (Figure 2). This map now shows vineyards for all parts of Sonoma County, including the Alexander Valley, Russian River Valley, Chalk Hill, Dry Creek Valley, Knights Valley, Los Carneros, Sonoma Mountain, and Sonoma Valley appellation areas. Using this data we calculated that approximately 4,720 hectares of new vineyards were planted from 1990 through 1997. This makes a total of at least 19,707 hectares in 1997, 20% more than were reported in the 1998 County Crop Report.

This data was then incorporated into a geographic information system (GIS) and integrated with several other GIS data layers including physiographic data such as slope, aspect, and elevation and vegetation maps. This allowed us to demonstrate that upland areas are increasingly targeted for vineyard development. For example, in Sonoma County we estimate that 25% of the vineyards developed since 1990 were on slopes greater than 10 degrees and 42% are above 100 meters in elevation (Figure 3a & 3b). For comparison, less than 6% of the vineyards established prior to 1990 were on slopes greater than 10 degrees and only 18% were above 100 meters (Figure 3a & 3b).

Intersecting the vineyard layer with the 1990 CDF hardwood pixel vegetation layer allowed us to estimate the amount of hardwood forest that was lost to agricultural conversion from 1990 to 1997. The CDF layer shows that 660 hectares of vineyard developed between 1990 to 1997 were in hardwood forest in 1990.

Logistic Regression Model

The results from our statistical analysis demonstrate which variables are correlated with where vineyards are on the landscape and allowed us to map suitable areas for vineyard planting in Sonoma County based on past trends of agricultural development

The 17-variable logistic regression model we used was:

Log odds = 1.1795 - 0.00210*elevation - 0.0383*slope + 0.000088*(elevation*slope) + 0.3938*(east + southeast + south + north) + 0.6663*(northeast + northwest) - 0.3914*(flat) - 0.0170*(southeast*slope + northwest*slope) + 0.6065*(local farmland + prime farmland) + 1.1862*(unique farmland + statewide farmland) - 0.7926*(other use) - 1.1239*(montane hardwood + potential hardwood + coast live oak woodland) - 0.4428*(conifer) - 0.9140*(urban cdf)+ 0.000281*(distance to 1990 urban areas) + 0.00021*(distance to perennial streams) - 0.00082*(distance to pre-1990 vineyards) - 0.00008*(distance to roads)

The probability surface resulting from the selected logistic regression model is presented in figure 4. Using half of the new vineyards data set to test the model, we calculated the accuracy of the logistic regression model to be 77%.

For the logistic regression model, the following variables decrease the probability of a cell being suitable for vineyard: increasing elevation, increasing slope, and increasing distance from existing vineyards. The following variables increased the probability of a cell being suitable for vineyard: increasing distance to nearest perennial stream, increasing distance from urban areas, if the area was in grass, shrub, blue oak woodland, or "other" vegetation in the 1990 CDF layer; if farmland was already present in 1990, if the slopes were facing northwest or northeast. Blue oak woodland, grass, shrub, or "other vegetation" were the most likely vegetation types to be found suitable for vineyard development, followed by conifer, and then urban. Montane hardwood, unclassified hardwood, and coast live oak woodland were the least likely to be developed. Unique farmland and farmland of state importance were the most likely land use types to be developed, followed by prime farmland and farmland of local importance, and then grazing land. "Other land" was the least likely land use type to be developed. Northeast and northwest facing slopes had the strongest relationship with vineyard development. East, southeast, south, and north facing slopes had the next highest probability of being developed, followed by northwest, southwest, and west facing slopes. Flat areas (defined as slopes with no discernible aspect in the digital elevation model) had the weakest relationship with vineyard development. There was also a weak relationship between vineyards and steep northwest and southeast facing slopes.

The lack of suitability in the coastal areas is most likely due to the heavy weighting of the "distance to existing vineyards" variable in the logistic regression model. As no vineyards were mapped on the coast (because the mapping project excluded this area), these areas were far from existing vineyards, so the logistic regression model predicted these areas as unsuitable.

Policy Results

We were interested to see how the new Sonoma County ordinance might affect current and future vineyard expansion. We mapped the four vineyard regulatory tiers (Level I, II, III, and IV) (Figure 5) and intersected these areas with the current vineyard map and areas identified from our model as suitable for future vineyard development. This allowed us to assess the number of acres that fall into each tier based on 1) vineyards developed prior to 1997; 2) vineyards that are most likely to be developed if the same amount of vineyard is developed in the future that was developed between 1990-1997 (4,720 hectares); 3) suitable vineyard areas with p>0.5 from the model (64,298 hectares). Figure 6 reveals the distribution of these acres in the different ordinance tiers. As the soils data were not finished for the entire county, we treated all areas as being non-highly erodible soils for this County-wide analysis.

As we had digitized the soils for two of the appellation regions, we were able to analyze the potential impact of the vineyard ordinance for both highly erodible soils and regular soils in these two regions. To demonstrate the importance of the soil type information, we mapped one of the largest private properties under single ownership in Sonoma County, located at the northwest edge of the Alexander Valley appellation area (see figure 7). For this area using slope and soil type information, we found that 442 hectares or 5.8% were in the Level I slope category, 1254 hectares or 16.4% were Level II, 3484 hectares or 45.5% were Level III, and 2481 or 32.4% were Level IV. While almost a third of the ranch is restricted from vineyard development, over 6,200 are still available, including the areas that were historically in vineyards but which had been removed in the 1950s.

Consequences for Habitat Fragmentation

Using the 1990 CDF vegetation layer, we calculated that there were 72 core areas of oak woodland, totaling 88,279 ha, of which 7 were larger than 4,000 ha and 36 were between 100 ha (minimum) and 200 ha. We chose a greater than 0.5 probablility level from the logistic regression model as our "better than even chance for vineyard development scenario" for the fragmentation analysis. This level included a large amount of vineyard (93,587 ha, or 23% of Sonoma County's area). After changing all areas in the logistic regression model with a probability greater than 0.5 into vineyard, we calculated that there were 76 core oak woodland areas, totaling 75,814 ha (a decline of 14.1% or 12,465 ha in area), with only 4 being greater than 4,000 ha and 32 in the 100 to 200 ha size. The increased count of core areas was due to the larger core areas being split into several smaller ones. An example of the potential fragmentation of oak woodlands is shown for the Jimtown area in the northeast part of Sonoma County in figure 8. Four areas of change in oak woodland connectivity or extent are shown with labels A, B, C, and D. Labels A and D show new core fragments that had previously been connected to a larger core area of oak woodland, and B and C are former core areas that disappear because of potential vineyard development.

Discussion, Benefits, and Dissemination of Findings

The GIS and model developed through this research project has resulted in a useful planning tool for policy makers and to assist conservation objectives. The initial GIS analysis quantified changes in planting acres and location, an important step toward documenting Sonoma County's changing landscape. The logistic regression model provided us with a probability surface that allows us to identify which combination of variables were important for determining a suitable vineyard site. These results are being used by the County Agricultural Commission to determine the effects of the new policies on their oversight activities.

The methods used allowed us to do more than identify areas of potential habitat loss. By comparing the forest patches before and after conversion of areas suitable for vineyard development we are able to identify priority sites that if protected could prevent fragmenting the largest remaining forested areas in Sonoma County. We are currently working with the Sonoma County Agricultural Preservation and Open Space District on updating their conservation easement acquisition plan so that these areas at high risk from forest loss and fragmentation could be protected from vineyard development.

In addition to the ongoing applications of this research, we have presented this project to multiple grape grower workshops in Sonoma, Napa, and San Luis Obispo Counties. We are in the process of developing a method for distributing the digital data from this project to interested members of the Sonoma County Grape Growers Association. Most recently, we have been working with 6 staff members at the Santa Rosa Press Democrat newspaper to provide direction for a series of stories on the impact of expanding vineyards to agricultural and natural diversity in the County. This series will certainly be an important part of the ongoing dialog on land-use in Sonoma County. We recently presented a comparison between two modeling methods for developing a vineyard expansion model at the ESRI users conference. A summary of this work can be found in the conference proceedings. The GISdeveloped as part of this project has allowed us to serve a diverse clientele including planners, farmers, and environmentalists.

Future improvements and applications

We need to make several important improvements to our mapping, modeling, and outreach effort on this topic. In addition to testing the importance of soil type as mapped by the USDA/NRCS which is currently being done in our lab, we need to expand the accuracy and geographic extent of our vineyard mapping to the north coast by using satellite imagery. We are eager to begin the next phase of this research which will include improving the forecasting of vineyard expansion by adding how economic variables effect the pattern and rate of vineyard planting depending on various market scenarios and other important variables such as water availability.

We also are also interested in incorporating vineyard development data into a cumulative watershed impact assessment since hillside farming has consequences for soil erosion and water use both of which can effect salmon habitat. This type of research will increase our ability to help make agriculture sustainable at the landscape scale.

Acknowledgments

This project was funded by the UC Sustainable Agriculture Research and Education Program. We would also like to thank Emily Heaton, Colin Brooks, and Dave Newburn for GIS analysis and model development. This work would not have been possible without the Sonoma County Grape Growers Association who shared their vineyard data with us, and Circuit Rider Productions Inc. for creating the vineyard data layer. Thanks also to Tim Pudoff of the Sonoma County Information Systems Department for sharing his Sonoma County GIS layers, some of which appear in the maps in this paper. Thanks to Shawn Saving and Greg Greenwood for their "core.aml." Finally, thanks to all the other cooperators in this project whom we have not named here.

Figure 1: Location of Sonoma County in California

Figure 2: Location of Sonoma County in California

Figure 2: Vineyard data set for Sonoma County

Figure 3a

Figur 3a

Figure 3b

Figure 4: Logistic regression model

Figure 5: Areas of Sonoma County divided into Levels I-IV

Figure 6: Vineyard area in each ordinance tier for 3 mapped data sets

Figure 7: Examples of soils dta and vineyard ordinance

Figure 8: Core analysis pre and post vineyard build out model using p>0.5 from the suitability model.

References

Akaike, H. 1973. Information Theory and an Extension of the Maximum Likelihood Principle. International Symposium on Information Theory (2nd edition), Budapest: pp. 267-281.

Burnham, K.P., and D.R. Anderson. 1998. Model Selection and Inference: a Practical Information - Theoretic Approach. New York, New York: Springer-Verlag New York, Inc. 353 pp.