Ecological Archives A020-007-A1

Manuel Colunga-Garcia, Roger A. Magarey, Robert A. Haack, Stuart H. Gage, and Jiaquo Qi. 2010. Enhancing early detection of exotic pests in agricultural and forest ecosystems using an urban-gradient framework. Ecological Applications 20:303–310.

Appendix A. Description of data sources and computation of the urban gradient for agricultural and forest ecosystems.


To develop an urban gradient (UG), we first constructed an urban influence index based on two assumptions regarding exotic pest invasions. First is the assumption that the influence of factors that facilitate exotic pest introductions is greater in urban areas receiving larger tonnage amounts of imported products that are known to harbor invasive species. Second is the assumption that the intensity of these factors is stronger for agricultural and forest ecosystems located closer to urban areas compared with areas more distant. The distance between agricultural lands and forest lands and adjacent urban areas was used as a proxy variable for the first assumption. The estimated tonnage of imported products (see Colunga-Garcia et al. 2009) was used as a proxy variable for the second assumption. Both variables were integrated into the following equation:


where UIi is an urban influence index for the ith county, k is the number of urban areas adjacent to the ithcounty, Dj is the shortest distance between the ith county agricultural land/forest land and thejth adjacent urban area and Tj is the estimated tonnage of imported products that arrived into the jth adjacent urban area. Separate computations of the UI index were conducted for agricultural and forest ecosystems in all counties within the contiguous United States. We used counties as the unit of computation because this is the scale at which most pest and crop data are available.

The first step in the computation of the urban gradient was to estimate the land cover centroids for agricultural and forest land within each county (Fig. A1a). To this end, we determined the location of each county’s agricultural land, by using the combined landcover classes of hay/pasture and cultivated crops provided in the 2001 National Land-Cover Database (Homer et al. 2007).

   FIG. A1. General approach to estimate an index of urban influence.

For the location of forest land, we used the combined landcover classes of deciduous, evergreen, and mixed forest provided in the same landcover source. For each landcover type (agricultural or forest) we calculated its percent coverage per 900 m2 and built a raster map with 10 categories with each representing 10% of the cropland. This raster map was overlaid on a U.S. county map and the resulting raster was converted to a multi-polygon map. The landcover centroid in each county (Fig. A2) was calculated from the individual agricultural or forest land polygons within a county using the agricultural land percentage values as a weighting factor.

   FIG. A2. Example of county centroids for agricultural and forest landcover in southern California.

The second step was to determine which urban areas were adjacent to each class of county land cover (Fig. A1b). We downloaded the 2000 version of the United States cartographic file for urban areas and updated its attribute information to reflect changes made after 2000 (US Census Bureau 2002a,b). Only urban areas located within the contiguous United States that intersected roads with commercial truck flow were considered (see next section). To determine which urban areas were adjacent to a county’s agricultural land or forest land we intersected the voronoi cells of urban areas (Fig. A5) with the two selected county landcover types. Urban areas whose voronoi polygons were intersected by a county landcover were considered adjacent to that county. The third step was to estimate the shortest distance between agricultural or forest land and adjacent urban areas (Fig. A1c). A cartographic boundary file of cities/towns for the contiguous United States (i.e., incorporated places) was downloaded and overlaid on the urban areas cartographic file. Then, using the resulting city/urban area map, we calculated the centroid of all cities in each county. Finally, for each urban area, we estimated the distances between the centroids of all cities and the centroid of the county’s agricultural land or forest land and the shortest distance was selected (see Fig. A5).The fourth step was to allocate the tonnage of selected imports among the individual urban areas (Fig. A1d) (Colunga-Garcia et al. 2009). Data on imported commodities were obtained from the Freight Analysis Framework (FAF) (Federal Highway Administration 2006). For agricultural ecosystems we focused our analysis on the FAF imports for cereal grains and other agricultural products. For forest ecosystems we focused our analyses on three categories of imports that have commonly been associated with wood-infesting insects: (a) wood products themselves, and the wood packaging material associated with imports of (b) non-metallic mineral products (including marble and ceramic tiles), and (c) machinery. The final step in the calculation of UI (Fig. A1e) was to use the distance measurements and imported tonnage values in Eq. A.1.

After computing all UI values, we used these values to sort all counties in order from lowest to highest UI. We computed each county’s contribution to the total U.S. agricultural or forest land. Agricultural land area per county was estimated using the combined landcover classes of hay/pasture and cultivated crops provided in the 2001 National Land-Cover Data Forest land area was estimated using the combined landcover classes of deciduous, evergreen, and mixed forest provided in the same landcover source. Next we divided each county percentage by 10, and grouped the counties in 20 levels (0–0.5, 0.51–1.0, etc.). The result was two urban gradients (UG), one each for agricultural and forest ecosystems, both ranging from 0.5–10.0, where each UG level encompassed 5% the U.S. agricultural land (agricultural ecosystems) or forest land (forest ecosystems) (Fig. A3).

   FIG. A3. Percentage of agricultural (left) and forest (right) land distribution across the urban gradient.



To determine urban adjacency we used the spatial properties of the voronoi diagrams (Okabe et al. 2000). A voronoi diagram is a set of polygons that allocate a planar space among a set of features, such that any point inside a polygon is closest to the feature it contains. Using the space allocation tools of ArcGIS® (ESRI 2004) we built a voronoi diagram map for all urban areas (Fig. A4a). This tool rasterized the cartographic file of urban areas and calculated, for each urban area, all the cells that were near to them based on the least accumulative cost over a "cost" raster.

   FIG. A4. (a) US urban areas and their resulting Voronoi cells after applying a cost land allocation algorithm. (b) The same procedure but applied only to large urban areas.

The "cost" raster (0.5 × 0.5 km resolution) was based on a rasterized version of the cartographic file of the Freight Analysis Framework (FAF) network (Alam et al. 2007, Federal Highway Administration 2007) (Fig. A5).

   FIG. A5. Freight Analysis Framework network raster.

To create the range of cost values in the raster we obtained the inverse of the truck flow for 2002 (data was associated to the cartographic file) the most recent year in which FAF data were available. Urban areas whose Voronoi polygons were intersected by a county (i.e., a county’s agricultural land or forest land) were considered adjacent to that county. This approach, however, presented a problem for counties surrounded by several smaller urban areas. Large urban areas (i.e., greater than 50,000 people) located just beyond smaller urban areas might not be able to "contact" a county even though they were close enough to affect it. To address this issue, we built a second voronoi diagram map for large urban areas only (Fig. A4b). If an urban area’s voronoi cell in this second map intersected a county, then that urban area was also considered adjacent to a particular county. The criterion to use the 50,000 people as a threshold to define large urban areas was arbitrary from an ecological perspective. It was applied, however, because the U.S. Census Bureau uses this figure as the threshold to distinguish between urban clusters and urbanized areas (U.S. Census Bureau 2001).

To illustrate the above procedure, we present as an example the case of the agricultural landcover for Barry Co., Michigan (Fig. A6). This county’s landcover is intersected by the voronoi cells of 8 urban areas (Fig A6, left column). A ninth urban area is added as a result of the county being intersected by one large urban area (i.e., greater than 50,000 people) (Fig. A6, right column). For the calculation of the urban influence index (Appendix A, Equation 1) these nine urban areas will be considered as adjacent to Barry Co. Distance measurements for the urban influence index are taken between the landcover centroid of Barry Co. and the closest city (to the landcover centroid) in each urban area as shown by the blue arrows. Notice that the urban areas selected with this approach represent only a "sample" of all urban areas that may actually influence a county. This is why the urban influence equation averages rather than sums the individual effects of the selected urban areas (Tonnage/Distance).

   FIG. A6. Determining adjacent urban areas to the agricultural landcover of Barry Co. Michigan, USA.


Alam, M., E. Fekpe, and M. Majed. 2007. FAF2 freight traffic analysis.

Colunga-Garcia, M., R. A. Haack, and A. O. Adelaja. 2009. Freight transportation and the potential for invasions of exotic insects in urban and periurban forests of the United States. Journal of Economic Entomology 102:237–246.

ESRI. 2004. Using ArcGIS Spatial Analyst. ESRI Publications, Redlands, California, USA.

Federal Highway Administration. 2006. Freight analysis framework (FAF) version 2.2, user guide.

Federal Highway Administration. 2007. FAF2 highway link and truck data and documentation: 2002 and 2035.

Homer, C., J. Dewitz, J. Fry, M. Coan, N. Hossain, C. Larson, N. Herold, A. McKerrow, J. N. VanDriel, and J. Wickham. 2007. Completion of the 2001 National Land Cover Database for the Conterminous United States. Photogrammetric Engineering and Remote Sensing 73:337–341.

Okabe A, B. Boots, K. Sugihara, and S. Nok Chiu. 2000. Spatial tessellations: concepts and applications of voronoi diagrams. John Wiley and Sons, New York, New York, USA.

U.S. Census Bureau. 2001. Urban areas criteria for Census 2000. Federal Register 66(60):17018–17033. Wednesday, March 28, 2001.

U.S. Census Bureau. 2002a. Qualifying urban areas for Census 2000; correction. Federal Register 67(164):54630–54631. Friday, August 23.

U.S. Census Bureau. 2002b. Qualifying urban areas for Census 2000; correction. Federal Register 67 (224):70045. Wednesday, November 20.

[Back to A020-007]