Mitigating Spatial Differences in Observation Rate of Automated Telemetry Systems

Abstract

Wildlife ecologists are increasingly interested in determining spatial distributions and habitat use of ungulates from locations estimated from both conventional and automated telemetry systems (ATS). If the performance of an ATS causes spatial versus random variation in probability of obtaining an acceptable location (observation rate), analysis of habitat selection is potentially biased. We define observation rate as the percentage of acceptable locations (i.e., those that meet signal strength, signal-to-noise ratios, geometric dilution of precision criteria) of the total locations attempted. An ATS at the Starkey Experimental Forest and Range (Starkey) in Oregon tracks movements of elk (Cervus elaphus), mule deer (Odocoileus hemionus), and cattle. We detected localized variation in observation rate of stationary radiocollars in 1993. Subsequently, we devised a method to estimate observation rate at various spatial scales using animal location data over 4 years (1992-95 ; n = 907,156 location attempts) to determine if the variation was spatial or random. We formulated 5 variants of a general linear model to obtain estimates of spatial variation in observation rate. All 5 models assumed spatially correlated error terms estimated from isotropic semivariograms. Three models included environmental variables as covariates correlated with observation rate. Models then were compared based on mean error, coefficient of determination, and residual plots. Random variation accounted for 47-53%, and spatial variation accounted for 38-45% of the variation in observation rate. One model was selected to demonstrate application of the correction to mitigate spatial bias in observation rate. Our results demonstrate the utility of semivariograms to detect and quantify spatial variation in observation rate of animal locations determined from an ATS.