Automating Farmland Preservation Using Geospatial Technologies: Prospects and Policy Implications

James R. Myers, Center for Remote Sensing and Spatial Analysis and Department of Geography, Rutgers University, jimm@crssa.rutgers.edu

David L. Tulloch, Center for Remote Sensing and Spatial Analysis and Department of Landscape Architecture, Rutgers University, dtulloch@crssa.rutgers.edu

John Hasse, Department of Geography and Anthropology, Rowan University, hasse@rowan.edu

Abstract

In order to test the capacity of geospatial technologies to assist in the ranking and assessment of farms for preservation, an attempt was made to automate the farmland preservation ranking scheme used by Hunterdon County, NJ.  Significant barriers to this automation were encountered, including data availability and complexity of the automating the ranking criteria.  Despite these difficulties, however, the automation was successful.  Once the automation was completed, several examples of geospatially enabled analyses and extensions using the automation were developed.  The powerful analytical capabilities evidenced by these examples suggest that the effort expended to automate the farmland ranking is worthwhile despite the difficulties encountered.  The techniques and extensions developed here are likely to be applicable to other situations involving the identification and selection of land for preservation, provided the necessary technical expertise and base data layers are available.

Introduction

New Jersey has an aggressive open space and farmland preservation program.  Since enabling legislation was passed in 1983, over 86,000 acres of farmland has been preserved through the farmland preservation program administered by the state.  The pace of acquisition has increased yearly increasing since inception, and will likely continue to do so.  In 1999, the Garden State Preservation Trust Act became law in New Jersey, establishing a stable funding source for open space and farmland preservation.  The act has provided the funding necessary to pursue the ambitious goal of preserving an additional one million acres of open space by 2010, half of which will be farmland.

With a guaranteed annual appropriation of $98 million for 10 years and the authorization to issue up to $1 billion in bonds, the county and state agencies responsible for land preservation have been saddled with the responsibility to spend significant amounts of money quickly but effectively.  The additional pressure the million acre goal places on the existing county and state farmland preservation programs will require the agencies that administer those programs to explore techniques that maximize the efficiency of the preservation process. Geospatial technologies hold great promise to reduce the time, effort and variability involved with many preservation program tasks.

GIS and Farmland Preservation

The use of geospatial technologies in the analysis and process of farmland preservation has been growing.  The USDA’s Land Evaluation and Site Assesment (LESA) tool (Pease and Coughlin 1996) models physical and cultural factors of a parcel of land using a weighting and scoring system.  While variations of the model are being developed to analyze zoning changes and evaluate parcels for conservation easements (Land Information Bulletin 2000a), LESA has also been directly applied to farmland preservation in Pennsylvania (Land Information Bulletin 2000b).  Significant work in developing tools for farmland preservation has also been done in Wisconsin (Jackson-Smith and Bukovac 1998, Johnson and Jacobs 1994).

Case Study

Located in east-central New Jersey, about 1 hour’s drive from New York City (Figure 1), Hunterdon County is relatively rural.  However, the county has experienced significant suburban development in the past 20 years.  From 1990 to 2000, the population grew 10% to 110,000 people (US Census Bureau 2000).  Because of the leapfrog and large-lot nature of the development, the impact on the county’s 279,943 acres has been larger than mere population growth suggests.  Between 1985 and 1995, over 12,000 acres of the county were developed and more than 9,000 acres of agricultural land were converted to other uses (Center for Remote Sensing and Spatial Analysis 2000).  These figures represent a 33% increase in developed land and a 7.5% decrease in agricultural land.  Hunterdon County was chosen a case study not only because of the development pressure on farmland.  The county also has an active farmland preservation program and a publicly available GIS database that includes parcel mapping.

Hunterdon County's primary tool for long-term farmland preservation is the purchase of development rights.   The farmland preservation program is coordinated by at the state level but administered by the county.  Although they must conform to general state guidelines, it is the responsibility of the county to determine the criteria they use to determine the farms for which they purchase the development rights.  Active since 1983, the county farmland preservation program has permanently preserved 8,137 acres of farmland as of December 2001 (Hunterdon County 2002).

Automation of Scoring Criteria

The Hunterdon CADB has developed a five criteria-scoring system to rank farms for preservation that is based on soils, boundaries and buffers, local commitment, size and density, and farm and family characteristics (Table 1). Each applicant’s property is then assigned quantitative scores for each of the five criteria.  The scoring regime used in this research was current as of August, 2001. 

Table 1.  The five ranking criteria.

Soils - In order to concentrate preservation efforts on the most productive farmland, soils are the most heavily weighted criterion.  In particular, this category emphasizes areas designated as prime farmland soils, with a lesser emphasis on soils of statewide, unique, or local importance, as defined by the Natural Resource Conservation Service (NRCS) (Formula 1). A farm can receive up to 30 points based on the quality of soils present.  A farm containing exclusively prime soils receives all 30 points, while a farm containing locally important soils would only get 10 points.

To automate the soils criteria, the relevant soil categories – prime, statewide, unique and local – were extracted from the NRCS soils coverage, coded with their corresponding point values of 30, 20, 25 and 10 and converted into an ARC/INFO grid.  After creating a summary table describing the area of each type of soil in each parcel, the total soil points for each parcel were calculated using Formula 1 and recorded in a field in the parcel coverage. 

Boundaries and Buffers - Because the land uses adjacent to a property affect its preservation value, the county allots up to 20 points to parcels depending on the surrounding land uses (Table 2).  The more conducive a neighboring land use to continued agricultural viability, the more points a parcel receives for being adjacent to that land use.  The points are allocated based on the percentage of the perimeter occupied by each type of land use. 

Table 2.  Land uses utilized in boundaries and buffers criterion.

This criterion proved to be the most difficult to automate.  The varied land uses listed in Table 2 were not available from a single source and needed to be acquired, modified or generated separately.   More problematically, many of the “adjacent” land uses were in fact separated from a parcel by a road break in the parcel coverage.  To account for this, all of the layers containing the data used in calculating this criteria were buffered by 75 ft.  Each of the layers was then converted to a grid, assigned a value equal to the points it contributes and then combined into a maximum-value grid.  Because they are low value but supersede other land uses, residential and high-use parks were then overlain on this grid to produce the final layer used.  To calculate the value of each parcel’s perimeter, a gridded boundary was generated inside each parcel.  This boundary was overlaid with the cumulative land use layer to produce a perimeter value layer.  The perimeter’s value for each parcel was determined by using a zonal summation and was included as a field in the parcel coverage.

Local commitment - Each of the 26 municipalities in Hunterdon County has instituted a unique set of regulations through zoning and other land use powers that either reinforce or undermine long-term agricultural viability.  The county has assigned a score to each municipality based these regulations and their presumed impact (Table 3).  The maximum awarded to a municipality is 22 points, the least 0.  Each parcel is assigned the score of its municipality using a look-up table.

Table 3. Factors contributing to local commitment score.

Size and density - In order to foster the long-term viability of agriculture, the county seeks to preserve large farms and contiguous areas of preserved farmland.  They therefore award parcels points based on their size and the density of preserved farmland surrounding them.  Parcels twice the county average farm size or lager receive 12 points, smaller parcels proportionately fewer points.  Parcels receive 1 point for each preserved farm and 2 points for each concurrent applicant within 0.5 miles, up to a maximum of 12 points.

Size points were assigned to parcels based on the following formula (Formula 2):

To calculate the density score, permanently preserved parcels and concurrent applications into ARC/INFO grids.  Performing A 0.8 km neighborhood diversity analysis was then d on each of these grids this provided output grids in which each cell contained the total number of different preserved farms or concurrent applications within the search radius.  The maximum value of each of these grids within each parcel was determined using a zonal summary analysis.  These maximum values were converted into the density score for each parcel using the following formula (Formula 3):

Farm and family characteristics - The farm and family criterion uses four measures to award up to 10 points for farm-specific conditions which promote agricultural viability.  These measures are percentage of land actively cropped or grazed, soil conservation measures, good farm management practices, and on-site farming investments.  Because they are derived from farm specific actions of farmers and land owners, they can not be automated using geospatial technologies.

Assessment

To determine the total score for each parcel, the scores for each individual criterion were summed and this total score was stored in a new column in the parcel coverage.  Once this was done, every parcel in the county was scored as though its owner had applied for preservation (Figure 2). 

This represents a significant advancement over the manual scoring technique, in which the parcels of 20 to 30 applicants were scored once or twice a year.  To understand how comparable this automated procedure was to the manual ranking performed by county staff, the manually calculated scores of the 1999 applicants were compared to the scores generated by the automated procedure.  Table 4 shows the summary statistics of differences between the manual and automated scores for the ranking criteria, soils, size and boundaries and buffers.  These statistics indicated a high degree of correspondence between the manually and automatically calculated scores, especially considering the high number of possible points for each criterion.  Although such a correspondence is desirable, it is important to note that any divergence between the manual and automated scores doesn’t necessarily indicate that the automation is inaccurate.  Indeed, because the automation is far more replicable than the manual calculation, such divergences may indicate problems with the manual scoring.  This is especially true for complicated measures such as the boundaries and buffers criterion.  Conversely, the lack of divergences evidenced here suggests that the manual scoring has been carried out accurately, at least in 1999.

Table 4. Summary statistics of differences between manual and automated ranking for selected criteria.

Extensions

The capacity to rank all parcels in the county simultaneously allows for several interesting extensions of the automation. These include targeting farms for preservation, assessing the effectiveness of current policies and procedures and exploring the impact of policy changes.

Targeting – Having a county-wide ranking allows for the easy identification of high ranking farms which are not yet preserved.  For example, Figure 3 shows in blue unpreserved parcels that fall within agricultural development areas which have a score greater than or equal to the mean score of all preserved farms. 

Because the parcel coverage can be linked to a database containing the names and addresses of the current owners, county employees can contact the owners of important unpreserved farms in order to solicit their involvement in the program.  If successful, this solicitation could have a significant impact on the success of the program by creating or strengthening core areas of agricultural preservation. 

Assessment of Current Policy – The county-wide nature of the automation provides a tool to assess the impact of currently implemented policies and procedures on the ranking and preservation process.  The contribution of each criterion to a parcel’s total score could be mapped separately.  This would allow for such analyses as the correlation of parcel size to soil score.  Moving beyond the scoring criteria, the automation could be used to assess other policy elements, such as the comprehensiveness of current agricultural development areas (ADAs). ADAs are areas of the county that have been determined to have the capacity to support agriculture in a profitable way. Only farms within ADAs are eligble for preservation. As Figure 3 indicates, only a few parcels with a score greater than the mean of preserved parcels lay outside of current agricultural development areas.  This suggests that current ADAs adequately capture most of the quality farmland that the county is interesting in preserving.  Of course, the county may wish to explore the character of those highly ranked parcels and extend the ADAs to include them if they are good candidates for farmland preservation.

Assessing Policy Changes – In addition to examining the effects of current policy and procedures on the preservation process, the automation can be used to explore the impact of policy changes as well.  One obvious example of this type of analysis is examining the impact on the relative ranking of parcels caused by adjusting the current criteria, such as halving the importance soils by reducing its weight from 30 to 15 points.  The impact of entirely new criteria could also be explored.  For example, the effects of adding a criterion that gave points to parcel that fall within areas of general open space, as opposed to farmland, preservation interest could be assessed. 

The possibilities for new criteria are, of course, limitless in both their variety and complexity.  One particularly powerful approach enabled by the use of geospatial technologies is the inclusion of criteria derived from spatial analysis and modeling.  As an example of the utility of this approach, an econometric model of landscape conversion from agriculture to other uses was developed and integrated into the automated ranking scheme.  Using a logit-based econometrics model (see Parks et al. 2000 for an example of the methodology), a number of variables were tested for their significance in explaining the conversion of agricultural land in Hunterdon County.  Five thousand random points were generated in the county, and the land use at those points was determined in 1986 and 1995.  Those points that were agricultural in 1986 were used to develop the econometrics model.  Three separate models were created.  The first described the probability of agricultural land changing to developed land, the second agricultural land to forest and the third agricultural land remaining as agricultural land.  The variables used were census block population density in 1990, mean census block housing value in 1990, whether the point was in a sewer service area or not, the size of the tax parcel the point occurred in, the area of developed land within ½ mile of the point, the distance to the nearest county, state or federal road, and the travel time to the eastern and southern gateways of the county.   The output from the econometric modeling provides a coefficient, sign and statistical significance for each variable.  By combining this output with the variables in a raster-based GIS, probability surfaces can be created depicting the likelihood of any point remaining in agriculture or converting to forest or developed land. 

The information generated by the econometric modeling can be integrated into the farmland preservation automation in a variety of ways.  One source of tension that exists in any preservation program that purchases land or development rights centers around whether land imminently threatened with development should be targeted.  Such land is likely to be expensive, and may become surrounded by development if preserved.  More land could be preserved for the same amount of money in areas with less development pressure.  Econometric models, such as the one developed here, that predict which agricultural areas are likely to become developed can assist in determining the impact on preservation if imminently threatened or less threatened areas are favored.  Figure 4 shows an example of this type of assessment. 

In essence, this map represents the winners and losers if imminently threatened areas are favored.  The parcels shown in green gain in relative rank after the addition of a 15-point criterion that gives additional points to those parcels likely be developed as determined by the econometric model.  Those in red lose relative rank.  A map showing the impact of favoring areas less likely to be developed would be the inverse of this map. 

Conclusions

This project was conducted in part as a test of the feasibility of integrating geospatial technologies into the farmland preservation ranking and decision making process in New Jersey.  The comparison between the automated and manual ranking results suggests that the automation is acceptably accurate.  More importantly, if the same base data and techniques are used, it will undoubtedly be more repeatable than a manual ranking carried by a number of staff members.  This replicability may significantly increase the trust both the participating farmers and taxpayers have in the results of the ranking. 

There are, of course, limitations to the automation.  Because farmland preservation in New Jersey (and in much of the US) is parcel-based, a high-quality digital parcel map is required for the automation to be carried out.  This is an expensive data layer, and consequently many areas do not yet have one available.  The other data needed to carry out the automation must also be available or easily derived from existing sources.  Additionally, the significant technical knowledge and resources available to the researchers during the course of this project may be unavailable to local government staff members attempting their own automation.

The potential benefits derived from the automation appear to be significant enough to warrant the effort required to overcome the difficulties.  Once the automation is complete there are a myriad of extensions possible. The ability to assess the impact of current and potential ranking criteria on preservation may alone be worth the effort.  The integration of the econometrics model with the farmland preservation ranking scheme is only one example of the potential for insightful and innovative analyses made possible by the integrating capabilities of geospatial technologies.  Other applications include modeling future preservation landscapes, understanding the potential environmental consequences of particular preservation ranking schemes and exploring the interplay and cumulative impact of farmland and other open space preservation programs.

Acknowledgements

This project has been made possible in part by funding through the Rural Development section of the USDA National Research Initiative Competitive Grants Program.  It also would not have been possible without spatial data from the Hunterdon County Geographic Information Systems Division and farmland preservation policy information from the staff of the Hunterdon County Agriculture Development Board.

References

Center for Remote Sensing and Spatial Analsyis, Rutgers University.  http://www.crssa.rutgers.edu/projects/lc.  Accessed: May 29, 2002.

Hunterdon County.  http://www.co.hunterdon.nj.us/cadb.htm.  Accessed: May 29, 2002.

Jackson-Smith, D., Bukovac, J., 1998. Limitations of Agricultural Land Use Planning Tools in Rural Wisconsin. paper presentation. Annual Meeting of the American Collegiate Schools of Planning (ACSP); November 1-5; Atlanta, GA.

Johnson, S.E. and Jacobs, H.M., 1994. Public Education for Growth Management: Lessons from Wisconsin's Farmland Preservation Program. J. Soil Water Conserv., 49: 333.

Land Information Bulletin, 2000a.  Farmland Preservation and GIS: A Model for Deriving Farmland Priority Zones.  Technical Paper No. 3.  University of Wisconsin-Madison Land Information and Computer Graphics Facility, Madison, WI.

Land Information Bulletin, 2000b.  Farmland Protection and GIS:  GIS Interface Helps Pennsylvania Counties Prioritize Farmland for Preservation.  National Consortium for Rural Geospatial Innovations, Chesapeake, Pennsylvania State University, University Park, PA.

Parks, P.J, Hardie, I.W., Tedder, C.A and D.N. Wear.  2000.  Using resource economics to anticipate forest land use change in the US mid-Atlantic region.  Environmental Monitoring and Assessment 63: 175 – 185.

Pease, J.R. and Coughlin, R.E., 1996. Land Evaluation and Site Assessment: A Guidebook for Rating Agricultural Lands, 2nd Edition. United States Department of Agriculture, Washington DC.

US Census Bureau 2000. http://www.census.gov/.  Accessed: September 30, 2001.