Accounting for residual heterogeneity in double-observer sightability models decreases bias in burro abundance estimates

Abstract

Feral burros ( Equus asinus ) and horses ( E. ferus caballus ) inhabiting public land in the western United States are intended to be managed at population levels established to promote a thriving, natural ecological balance. Double-observer sightability (M DS ) models, which use detection records from multiple observers and sighting covariates, perform well for estimating feral horse abundances, but their effectiveness for use in burro populations is less understood. These M DS models help minimize detection bias, yet bias can be further reduced with models that account for unmodeled variation, or residual heterogeneity, in detection probability. In populations containing radio-marked individuals, residual heterogeneity can be estimated with M DS models by including a covariate that corresponds to the marked status of a group (M H models). Another approach is to use information from detections missed by both observers to account for the characteristics that make groups more or less likely to be detected, or recaptured, by the second observer (M R models). We used aerial survey data from 3 burro populations (Sinbad Herd Management Area, UT [2016–2018], Lake Pleasant Herd Management Area, AZ [2017], and Fort Irwin National Training Center, CA [2016–2017]) to develop M DS models applicable for feral burros in the southwestern United States. Our objectives were to quantify precision and bias of standard M DS surveys for feral burros and to examine which model type for incorporating residual heterogeneity (M H or M R ) would result in the least-biased estimates of burro populations relative to the minimum number known alive (MNKA) within the Sinbad Herd Management Area. Standard M DS model estimates achieved a mean coefficient of variation of 0.08, while underestimating MNKA by an average of 27.1%. Accounting for residual heterogeneity through recapture probability in M R models resulted in estimates closer to MNKA than M H models (9.5% vs. 16.5% less than MNKA). Our results indicate that M DS models can achieve precise enough estimates to monitor feral burro populations, but they routinely produce negatively biased estimates. We encourage the use of radio-collars to reduce bias in future burro surveys by accounting for residual heterogeneity through M R models.