Skip to main content
Dryad logo

Wild Pig Management at the Jack and Laura Dangermond Preserve

Citation

Song, Shuhan; Omasta, Peter; Truong, Benson; Zekanoski, AJ (2021), Wild Pig Management at the Jack and Laura Dangermond Preserve, Dryad, Dataset, https://doi.org/10.25349/D9P32J

Abstract

This dataset was created for the Wild Pig Management at the Jack and Laura Dangermond Preserve, a group project by Bren School of Environmental Science and Management at the University of California, Santa Barbara. The project includes a population study of wild pigs (Sus scrofa) and a cost analysis of three management scenarios. For the wild pig population study, this dataset contains pig count data created by tagging camera trap photos between October 2013 to August 2015 (01_WildPigPopulation_TimelapseData.csv), estimated abundance data by camera trap stations (02_WildPigPopulation_AbundanceByStation.csv), and spatial files of their interpolated distribution in the Preserve (03_WildPigPopulation_Distribution.zip). In the abundance dataset, April Check refers to the camera trap photos taken between 2013-10-23 and 2014-04-22 while September Check refers to the time period of 2014-04-24 to 2014-09-25. For cost analysis, this dataset provides spatial files of four proposed fencing areas (11_CostAnalysis_ProtectedAreasByFencing.zip), the costs of six fencing scenarios for Dangermond Preserve (12_CostAnalysis_DangermondFencingCost.csv), and cost and pig removal data gathered from seven case studies (13_CostAnalysis_CaseStudies.csv). Raw data used to generate the dataset were provided by The Nature Conservancy at Dangermond Preserve. 

Methods

Data collection:

Camera trap photos were collected by WRA and TNC from 2013-2014 using 38 camera trap stations. The data from camera traps were gathered once in April 2014 (April Check) and once in September 2014 (September Check). September Check also contains a few observations from June 2015 to August 2015 which were excluded in the analysis.  Fencing cost data were provided by TNC. Fencing protected areas and footage were measured from spatial files. 

Data processing:

Step 1. Classify camera trap photos into animals, humans, and vehicles in MegaDetector by Microsoft AI for Earth. This step allowed us to quickly pick out animal photos, which were narrowed down from ~400,000 images to ~250,000.

Step 2. Manually tag wild pigs (Sus scrofa) in Timelapse2. Timelapse2 takes the classification output from MegaDetector to filter out animal photos in the image set. From the camera trap photos, we collected data of date, time, temperature, geographic location, and the number of pigs in each image. See 01_WildPigPopulation_TimelapseData.csv for the result data.

Step 3. Estimate the number of pig groups at each camera trap station. We first defined one day as one visit and calculated the observed number of groups for each day at each camera trap from 01_WildPigPopulation_TimelapseData.csv. We used 5 minutes and 30 seconds to divide up groups. Then, we used the N-mixture model (`unmarked` package in R) to find the estimated number of pig groups.

Step 4. Estimate the pig abundance. We calculated a weighted average group size at each camera trap, then times it to the estimated number of pig groups to find the estimated abundance. The result of this step can be found in 02_WildPigPopulation_AbundanceByStation.csv. The overall density of wild pigs was estimated by looking at the extremes of the estimated abundance. We examined the sum of abundance, the maximum abundance per site, and consulted with experts to nail down a density of ~2 pigs/km2.

Step 5. Model wild pig distribution in Dangermond using kriging in R. We first detrended the spatial trend to meet the assumption of constant mean and variance through the study area. We determined the global trend to be removed by comparing first order and second-order polynomial fit. We then looked at the sample experimental variogram plot and fitted it with the Spherical model. Following that, we sent the fitted variogram to kriging interpolation, which used the localized pattern produced by sample data to compute the weights of neighboring pig counts. After kriging, we combined the output with the previously removed global trend to produce the final result of the interpolation. See the TIFF file in 03_WildPigPopulation_Distribution/ folder.

Step 6. Estimate total cost and cost per area protected under five fencing scenarios in the Dangermond Preserve. We first took measurements of the five proposed areas where TNC might install fences in ArcGIS (see shapefile in 11_CostAnalysis_ProtectedAreasByFencing/ folder). Given the measured area protected, we calculated the total costs of fencing, gates, and old fence removal (see 12_CostAnalysis_CasegermondFencingCost.csv).

Step 7. Examine cost efficiency from case studies. We took data of fencing costs, footage, acres, and the number of pigs removed from seven case studies and calculated the cost efficiency corresponding to the density of wild pigs. See 13_CostAnalysis_CaseStudies.csv.

For a more detailed methodology, please refer to Wild Pig Management at the Jack and Laura Dangermond Preserve Report on the website of Bren School or contact Shuhan Song to obtain access to our GitHub repository.

Usage Notes

 Please refer to the README.txt files for usage information.
 

Funding

James S. Bower Foundation