top of page
This site was designed with the
.com
website builder. Create your website today.Start Now

Evaluating Bobcat Habitats in Vermont

Updating the Bobcat Habitat Suitability Index (HSI) for Vermont
Vermont Bobcat Populations

Although they are rarely seen by humans, bobcats are native to the Vermont wilderness. While their numbers have decreased significantly since their peak in the early 20th century, bobcats are well distributed and common in Vermont today (Vermont Fish and Wildlife). However, due to increasing development of land and forestry, habitats suitable for Vermont bobcats are becoming less and less available. Therefore, for this project, I will be attempting to find suitable habitats for bobcats in Addison and Chittenden counties in Vermont by creating a suitability model, specifically a Habitat Suitability Index (HSI).

​

An HSI is an index of habitat suitability as a function of environmental variables that are weighted varying amounts depending on their impact on the given species. There are two kinds of criteria: 1) constraints, which are variables that are evaluated in the binary, so each pixel is given a yes or no, and 2) factors, which are continuous or categorical variables that have a varying impact on the species from positive to negative. After creating an HSI with my data, I will be comparing this HSI to bobcat sightings in the state using a regression analysis to evaluate the effectiveness of my HSI at predicting where bobcat actually live in the two counties.

​

Bobcat in Olympic National Park, Washington, from National Geographic.

http://cdn.natgeotv.com.au/factsheets/thumbnails/bobcat-fierce.jpg?v=18&azure=false&scale=both&width=1600&height=900&mode=crop

Data and Methods

The first step in creating an HSI is deciding which variables will be used for the evaluation. For my analysis, I focused on habitat suitability factors based on research by the Vermont Fish and Wildlife Department. The Vermont FWD found that bobcats used "forest, scrub/shrub habitat, and wetlands or stream buffers" for habitat and did not use agriculture, developed areas, and roads, so I used land cover, roads, and streams as factors in my analysis. They also found that bobcats cannot hunt prey if snow is deeper than 8 inches, so I used snow depth data for my study area. Finally, I used prey and competitor density data that would tell me where bobcats like to go to hunt, and where they will stay away from due to high concentrations of competitors. Thus, the data used in this analysis comes from a range of sources:

​

​

The following data was used to define the study area and comparison:

1. Data Processing and Re-scaling

Using this data, I could create a habitat suitability index (HSI) for bobcat populations in Vermont. The data for each variable had to be manipulated differently to create a uniform scale with which to perform the analysis. For my analysis, I decided to use a scale from

0 - 5, with 5 being the most ideal conditions to support habitats for bobcats. The 0-5 scale was used for factors, while I gave a 0 or 1 to constraints. My variables are listed below with my classification scheme from 0-5, whether the variable is a factor or constraint, and the methods I used to standardize each using ArcGIS:

​

  • Snow depth (constraint) - I reclassified the snow depth data to differentiate between areas that had more and less than 8 inches of snow. The areas with less than 8 were given a 1, meaning they are suitable for bobcats, while the other areas were given a 0, meaning they are not suitable for bobcats.

    • 0 - 8 inches (1)

    • 8+ inches (0) - not suitable for bobcats

  • Roads density (factor) - Using ArcGIS, I created a layer of the "density" of roads at each pixel using the Line Density tool, which calculates a magnitude-per-unit area from roads that fall within a radius around each pixel. I re-scaled this density layer to have a value between 0, highest density of roads, and 5, lowest density (most suitable for bobcats) by dividing each pixel's value by the maximum value, and then multiplying by 5. This value was then subtracted from 5 to invert the scale.

    • 0, high --> 5, low

  • Streams (factor) - First, I used the Euclidean Distance tool in ArcGIS to calculate the distance each pixel is away from a stream. Next, I reclassified this data in 5 classes between 1 and 5, as shown below. Pixels closer to streams have higher values, meaning they are more suitable for bobcats.

    • 0 - 500 meters (5)

    • 500 - 1000 meters (4)

    • 1000 - 2000 meters (3)

    • 2000 - 5000 meters (2)

    • 5000+ (1)

  • Prey density (factor) - I rescaled the prey density layer onto a scale from 0 - 5 by dividing each pixel's value by the maximum value, and then multiplying by 5.

    • 0, low --> 5, high

  • Competitor density (factor) - I re-scaled the competitor density layer onto a scale from 0, highest density, to 5, the lowest density meaning most suitable for bobcats, by dividing each pixel's value by the maximum value, and then multiplying by 5. This value was then subtracted from 5 to invert the scale.

    • 0, high --> 5, low

  • Land cover (factor) - I reclassified the value for each specific land type to a value between 0, unsuitable for bobcats, and 5, most suitable for bobcats.

    • Forested, shrub, wetland (5)

    • Developed, agriculture (1)

    • Water (0)

2. Assigning Weights

Once each factor was re-scaled between 0 and 5, and the constraint was 0 or 1, I could assign weights to each variable based on how influential it was on determining bobcat habitats. There has been a number of studies looking at the impact of different land covers/uses, streams and roads, as well as competitor and prey density, on bobcat habitats. Donovan et al (2011) find that certain land uses  have a significant impact on bobcat habitats, especially developed land and forest/wetland areas. Poessel et al (2014) investigated the effect of roads on movement and home ranges of bobcats, finding that while roads were often included in the home ranges of bobcats, there was increased mortality risk to these populations. Finally, Reed et al. (2017) also found the importance of certain land uses on restricting bobcat habitats like developed and agriculture land, while forest, shrub/scrub and wetland were positive indicators. Vermont Fish and Wildlife also highlight the importance of riparian buffers to bobcats for uninterrupted travel. Since much of the literature focused on land cover/use, streams, and roads, these were given the highest weight in my analysis. Prey and competitor density, since they were not discussed but are important, were given less weight. Here are the weights I chose for each variable, which must add up to 1:

​

  1. Land Cover - 0.40

  2. Distance to Streams - 0.20

  3. Roads density - 0.20

  4. Prey density - 0.10

  5. Competitor density - 0.10

  6. Snow cover was used as a constraint, meaning I restricted my suitability areas to places where snow depth was under 8 inches

​

​

With these weights, I used the Weighted Sum tool in ArcGIS, which allows the user to input multiple variables, with given weights, and outputs the final result. Next, I overlayed this weighted later with the snow cover constraint to restrict the habitats to areas with under 8 inches of snow. The result of this was a Habitat Suitability Index for Addison and Chittenden county (above).

3. Spatial Interpolation

The bobcat sighting data from iNaturalist was in the form of discrete points that held values for the number of sightings each dot represented. I have summarized that data in the map below (left). However, to assess the accuracy of my HSI, I needed to have an estimation for bobcat sightings at every signal point in the two counties. This is where spatial interpolation comes in. Spatial Interpolation is a technique that  estimates surface values at unsampled points based on known surface values in the area. There are many different forms of spatial interpolation that are better for different kinds of data. For this analysis, I used a technique called Kriging, which considers the value and distance of the sample point from the unknown point and uses weights based on the nearest neighbors, but reduces weights for neighbors that are close together. The benefits of Kriging are that it tailors the calculations to the data by analyzing points using weighted average estimation to minimize biases. The results of Kriging are below (right).

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

​

 

To compare my bobcat sighting layer with my HSI, I needed to have standardized areas to assess. Thus, I used a grid of 2500 square-meter squares across the extent of my study area to average each layer within. Using a tool called Zonal Statistics, I found the mean value of each layer within each 2500-square meter zone. After converting the values to integers (a requirement of converting the data to polygons), I transformed the continuous data to discrete polygons, then joined the layers together so I could perform a regression analysis.

​

4. Regression Analysis

To complete my analysis and assess the accuracy of the habitat suitability layer, I ran a regression analysis in ArcGIS to look at how well the HSI explains the spatial variation in bobcat sightings. I used the OLS Regression tool to run the initial regression, which analyzed how much of bobcat sightings can be explained by the HSI. I also used a Geographically Weighted Regression (GWR), which is a regression that takes into account the spatial relationships and variations between the data. This is a local model in that it fits the regression equation to every feature in the study area to calculate a coefficient and R-squared values for each individual feature. Shi et al (2006) find that GWR is a much better indicator of variable influence than OLS when doing analysis of species distribution and habitats.

​

Results

The GWR, on the other hand, gave me insight into how the index helped explain the sightings differently across the study area. The maps to the right show the coefficient values and the R-squared values for each feature in the study area. Based on the GWR coefficient map, the index explained the sightings very well across space, especially in north-central Addison County and parts of eastern Chittenden counties. The wide swaths of negative correlation along the eastern border is where there was greater than 8 inches of snow. The model did not work well there since there were in fact some sightings in this area.

​

The R-squared map shows that the model performed very well in the same areas of the two counties, and much less so across large parts of the central parts of the counties. The GWR is a valuable insight into how the model worked across space, much more in detail than the OLS.

​

Overall, my two regressions showed that the HSI clearly had a positive relationship with the sightings, yet this relationship varied across space, especially bad where the index indicated no bobcats and in fact there were a few sightings.

​

My OLS regression analysis output had several interesting results:

​

  1. The coefficient value for the HSI was 0.075862, which was not given a star showing significance from ArcGIS, but instead demonstrated a positive relationship between the index and the sightings. This means that as the index values increase in a certain area, the number of sightings tended to increase.  The coefficient represents the strength and type of relationship between the variables.

  2. The p-value for the HSI was 0.007945*, which means that the index variable had a significant relationship with the number of sightings. Generally, a p-value below 0.05 with an asterisk demonstrates significance.

  3. The Multiple R-squared value for the HSI was 0.007924, which is not a significant value. R-squared is measured on a scale from -1 to 1 and shows how much variation in the sightings can be explained by the HSI.

​

Overall, the results of the OLS regression analysis show that my HSI has a somewhat significant positive relationship with the interpolated bobcat sightings. Based on this, it is safe to say that my habitat suitability layer is a solid baseline for understanding where potential habitats for bobcats could be located.

​

Limitations and Sources of Error

Due to the limitations of this approach, there are a number of potential source of error that may have impacted the results:

​

  1. The determination of the weights, while done with input from existing literature, was done without significant quantitative analysis or deep knowledge of the topic. With more time and research into bobcat habitats, the weights could have been assigned with more certainty and this may have helped the model perform better.

  2. Similarly, the assignment of values from 0 to 5 for different variables was done without quantitative evidence. Specifically, the distances from the streams were given classifications that are mostly arbitrary and may have affected the analysis. The need to classify data inherently will make the results of the analysis less accurate,

  3. Another possible source of inaccuracy was aggregating the data into the grid zones. While the large size of 2500 square meters helped with processing time, a smaller grid size would have made the analysis more exact.

Management Implications

Based on my analysis, it is clear that there is significant bobcat populations in Vermont that require specific types of areas to survive. Modeling habitat suitability indices for species like bobcats is an effective way of managing the populations and looking at how the patterns change over time. States like Vermont must take initiative to address the environmental conditions impacting bobcats.

​

The Vermont Fish and Wildlife Department completed a study of bobcat populations in February 2017 that offered insights into habitat loss and potential threats. These insights include the increase in developed lands and forest loss outpacing population growth in Vermont. The study found that the average home range size was 27 square miles for males, and 8.8 square miles for females. Tracking bobcat movements in a similar study area to my analysis garnered results that showed the habits of bobcats and what kinds of environmental conditions they prefer.

Adult bobcat perched on a tree branch.

http://www.wildlifemanagementpro.com/wp-content/uploads/2010/09/bobcat-facts-pictures-adult-01.jpg

​

I would recommend several future steps Vermont could take to improve the management of bobcat populations in Vermont:

​

  1. Infrastructure improvement projects to implement riparian buffers along streams to allow bobcats and other species un-fragmented habitats and corridors of movement. In addition, maintain un-fragemented habitats with land covers that are suitable for bobcats, including forest, wetlands, scrub/shrub (Vermont Fish and Wildlife).

  2. Increase awareness of popular bobcat crossings along major roads and possible construction of vegetated animal bridges over major roadways that could allow animals to cross without danger of cars. To minimize need for this, focus the creation of habitats on areas away from major roads.

  3. Investment in research and tracking studies that can continually offer new data and habitat models for conservation experts to use.

Sources

Donovan, T.M., Freeman, M.D., Howard, A., Royar, K., Abouelezz, H., Mickey, R. 2011. Quantifying home range habitat requirements for bobcats (Lynx rufus) in Vermont, USA. Biol. Conserv., 144 (2011), pp. 2799-2809

​

Poessel, Sharon A., Burdett Christopher L., Boydston, Erin E., Lyren, Lisa M. , Alonso, Robert S., Fisher, Robert N., Crooks, Kevin R. Roads influence movement and home ranges of a fragmentation-sensitive carnivore, the bobcat, in an urban landscape, In Biological Conservation, Volume 180, 2014, Pages 224-232, ISSN 0006-3207.

​

Reed, Gregory C., Litvaitis, John A., Ellingwood, Mark, Tate, Patrick, Broman, Derick J.A., Sirén, Alexej P.K., Carroll, Rory P. Describing habitat suitability of bobcats (Lynx rufus) using several sources of information obtained at multiple spatial scales, In Mammalian Biology - Zeitschrift für Säugetierkunde, Volume 82, 2017, Pages 17-26, ISSN 1616-5047, https://doi.org/10.1016/j.mambio.2016.10.002.

​

Shi, Haijin, Laurent, Edward J., LeBouton, Joseph, Racevskis, Laila, Hall, Kimberly R., Donovan, Michael , Doepker, Robert V., Walters, Michael B., Lupi, Frank, Liu, Jianguo. Local spatial modeling of white-tailed deer distribution, In Ecological Modelling, Volume 190, Issues 1–2, 2006, Pages 171-189, ISSN 0304-3800.

​

Vermont Fish and Wildlife Department. Bobcat Study in Vermont, February 2017. http://www.vtfishandwildlife.com/UserFiles/Servers/Server_73079/File/Hunt/trapping/bobcat-study.pdf

​

Vermont Fish and Wildlife Department. Eastern Bobcat Fact Sheet, August 2014. http://www.vtfishandwildlife.com/common/pages/DisplayFile.aspx?itemId=227899

bottom of page