top of page

Assessing Habitat Suitability for Vermont's Bobcats

   Throughout much of the 20th century,  bobcats were widely hunted and trapped in Vermont and New England.  Up until 1971, bobcats were among several predators in Vermont with a bounty offered for their hunting. Now, due to the removal of the bounty and gradual increases in forested lands and bobcat prey populations, there are an estimated 2500-3500 bobcats in Vermont. However, although populations have recovered well, recent concerns over loss of forested lands and rapid increase in developed have lead to increased interest in Vermont bobcat populations (Mod3: Suitability/Risk). 

   My goal was to create a GIS habitat layer for bobcats in Vermont. The  Vermont Agency of Natural Resources already has layers for wintering habitat for deer and other game animals, and but one has yet to be developed for the bobcat. A GIS layer for bobcat wintering areas would be a helpful resource for conservation and wildlife management efforts.


    Since bobcats inhabit forest, wetlands, and shrub areas, and tend to avoid agricultural land, roads, and developed areas, land cover type is an important criteria for bobcat habitat suitability (Vermont Fish and Wildlife). To constrain the study area to the extent of suitable land cover type, I reclassified a 2011 National Land Cover Dataset from 16 categories to two : favorable and unfavorable for bobcat habitat.

   Bobcats also tend to avoid roads, so I used distance from roads as another constraint. I converted a Vermont roads vector layer (from VCGI) to raster data, and ran the ArcGIS Euclidean Distance tool to get a layer of continuous data with distances from road features. I reclassified that data to separate all of the land in Chittendon and Addison county within 100 meters of a road. I combined the land cover and road-distance constraints into one layer with boolean values: zeros for water, developed land, agricultural land, and a 100 meter buffer around roads, and ones for everything else.

     Next, I used a prey distribution data (derived from iNaturalist) to indicate higher and lower suitability within the constraint layer. Abundance of prey has repeatedly been found to be correlated to bobcat distribution, so it made sense to use it as a factor indicating habitat suitability (Conner et al, 2001; Knick, 1990). I first clipped the prey distribution data to the extent of the study area, to limit the range of values to the relevant data. Then, I grouped values into five classes and rescaled the data using the Raster Calculator tool to divide by the maximum value and multiply by five, to create a range of values from 0-5. Again using Raster Calculator, I multiplied the constraints layer and the rescaled prey distribution layer to create an HSI with values ranging from 0 - 4.8, with higher values representing greater habitat suitability.

   To assess the accuracy of my HSI, I compared the habitat suitability layer to data on reported bobcat sightings in Chittendon and Addison county, Vermont (compiled from multiple sources). Some pre-processing was required in order to ready the data for regression analysis. In order to get the bobcat sighting data points into a continuous raster layer, I used ArcGIS's Kriging tool. Kriging is a method of spatial interpolation, or a way to estimate value of unknown points based on known values. Kriging determines the value of unknown points by considering the value of nearby known points as well as their distance from the unknown points. The interpolated values are a product of the weighted values of neighboring known points. The result is a continuous raster layer based on known vector points, in this case, a continuous layer of bobcat distribution based on the number and spatial arrangement of recorded sightings. Once I had continuous data layer for both the HSI and known bobcat distribution, I fit that data to a 2500 square meter grid with ArcGIS's Zonal Statistics tool. From there, I converted the values to integers and then converted the layers from raster data to vector polygons containing the summarized data values. Then, in order to get the HSI values and bobcat sighting values into the same layer, I performed Spatial Join to join the bobcat sighting counts to the HSI polygons.

I built a Habitat Suitability Index (HSI) for bobcats in Vermont based on three criteria: land cover type, proximity to roads, and distribution of prey animals (deer and . Then, I compared my HSI to bobcat sightings in Chittendon and Addison county to assess the accuracy of my suitability index.


Join Layer
Target Layer
Target Layer
Scatterplot of bobcat counts vs. HSI values from OLS

    Next, I performed an Ordinary Least Squares regression (OLS) to explore the relationship between the HSI values and bobcat counts. With OLS, I found a weak positive relationship (R squared value of 0.05) that was statistically significant (p value of 0.002; p values < 0.01 are considered statistically significant) .The scatterplot at right shows bobcat counts plotted on HSI values, and the gridded distribution of the points reflects the conversion of values to integers that I performed to prepare data for spatial join.


I also ran a Moran's I Spatial Autocorrelation test on the data to determine the degree to which the data had a spatial component (visible clustering of the bobcat sighting points suggested there would be one). I got a Moran's I index of 0.47 with a p-value of 0.00 , which suggests that the spatial component of the data is strong.

   Given the results of the spatial autocorrelation test, it made sense to run a regression that takes into account the geographic element of the data. I ran a Geographically Weighted Regression (GWR) on the results of the spatial join, and found a slightly higher R squared value of 0.09. GWR fits an equation for every point in a study area, and so allows the relationship to vary over space. The map of the R-squared values (a measure of the effectiveness of my HSI in predicting bobcat sightings) across the study area, at left, shows areas in orange and red where the HSI was slightly more effective.


    A visual comparison of the locations of bobcat sightings data and the HSI I developed provides insight into the weak relationship found in the regression analysis. When creating my habitat suitability index, I used land cover type as a binary constraint and classified agricultural lands as unsuitable habitat for bobcats. However, recently logged areas, farms, and uncultivated agricultural land provide food and cover for the bobcat's prey species (VTF&W). Since prey abundance is an important indicator of bobcat range, as previously noted, suitable habitat for bobcat prey animals could be expected to attract bobcat populations.

   Recent analysis of the bobcat sightings data and land cover type by Lindsay Dreiss shows a strong relationship between field land cover and bobcat sightings. Improving upon the HSI I created would mean changing the land cover constraint classification to include agricultural land as suitable habitat. 

bottom of page