Data from: Methods to account for incomplete viewsheds in distance sampling
Abstract
Conventional distance sampling is a logistically feasible method for estimating the densities of unmarked animals. The probability density function (PDF) of the sampling area specifies the expected proportion of the population that occurs at each distance from the observer and is a fundamental component of distance sampling models. Current approaches set this PDF either to equal probability at each distance, for line transects, or an increasing probability for point transects (because sampling area increases with radial distance from a point). Geographic Information Systems allow measurements of the area viewable from a given location (i.e., the viewshed), the structure of which may not reflect theoretical PDFs for either line (rectangular) or point (circular) transects. We simulated three datasets to test how variation in the viewshed structure affects estimates of detection probability, abundance, and density. We then implemented a novel application of Bayesian distance sampling models to test the magnitude of parameter bias recovered by accounting for incomplete viewsheds. Lastly, we compared parameter estimates from Bayesian hierarchical models that used either traditional or custom PDFs to analyze a dataset of 95 county-level spotlight surveys of white-tailed deer (Odocoileus virginianus) in Iowa, USA. For empirical data, viewable sampling area decreased with distance at an average rate of 3% every 100 m (range from 1–7% among counties). Our model correction decreased county-level density estimates by an average of 18% (range from 13–27% among counties), which depended on how sharply visibility declined. We suggest incomplete viewsheds be handled by considering the expected distribution of animals inside and outside of the viewshed. More generally, we show that customizing a PDF to more accurately reflect the study system improves density estimates and offers flexibility when the distribution of animals from the observer deviates from traditional assumptions.
https://doi.org/10.5061/dryad.pvmcvdnwx
Description of the data and file structure
Deer were sampled with nocturnal spotlight surveys conducted along gravel roads in Iowa, USA.
Files and variables
File: df6.csv
Description: Dataset with each row representing a county, which is the level of site in this study.
Variables
- row_id: Unique identifier for rows.
- county: County that the transect was conducted in. This is the level of site.
- dummy_county: Dummy variable for site id.
- mean_group_size: Mean number of deer per group/detection.
- total_area: Number of 30x30 m pixels within the viewshed.
- box_area: Number of 30x30 m pixels within the rectangular sample area.
File: df5.csv
Description: Dataset with each row representing a detected group of deer.
Variables
- row_id: Unique identifier for rows.
- dist_to_road: Distance between the detected group of deer and the nearest road.
- deer_count: Number of deer per detection.
- county: Name of the county of the sample. This is the site.
- binary_viewshed: Specifies whether the deer was located within the estimated viewshed.
- percent_area_0to80: Percent of the viewshed within the first distance bin.
- percent_area_80to160: Percent of the viewshed within the second distance bin.
- percent_area_160to240: Percent of the viewshed within the third distance bin.
- percent_area_240to320: Percent of the viewshed within the fourth distance bin.
- percent_area_320to400: Percent of the viewshed within the fifth distance bin.
- total_area: Number of 30x30 m pixels within the viewshed.
- box_area: Number of 30x30 m pixels within the rectangular study area.
- dummy_county: Dummy variable for site.
Code/software
We provide two R scripts to analyze this data: one for the simulation study and one for the empirical study.
Deer were sampled with nocturnal spotlight surveys along gravel roads in Iowa, USA.