Accounting for spatial heterogeneity in visual obstruction in line-transect distance sampling of gopher tortoises

Line-transect distance sampling (LTDS) surveys are commonly used to estimate abundance of animals or objects. In terrestrial LTDS surveys of gopher tortoise (Gopherus polyphemus) burrows, the presence of ground-level vegetation substantially decreases detection of burrows of all sizes, but no field or analytical methods exist to control for spatially heterogeneous vegetation obstruction as a source of variation in detection. We propose the addition of a simple measurement of ground-level vegetation that serves as a covariate for the detection function. We present a Bayesian hierarchical model in which covariates burrow width and nearby vegetation height help to account for detection bias and improve precision of estimated density. We investigate the performance of this covariate by simulation and by using real LTDS data collected before and after application of prescribed fire. We collected data in 2018 at the Jones Center at Ichauway in Newton, Georgia, USA. Across all simulations, our model including both covariates produced the most accurate density point estimates of any of the models tested. For our case study, our Bayesian model with vegetation covariates tended to produce similar estimates of density before and after burns. Our study indicates that any level of spatial variation in vegetation obstruction decreases detection of burrows and may lead to underestimation in population size (<= 68%) and proportion of individuals with small burrow sizes (<= 32%) when not considered during analysis. Our work is extensible to other terrestrial sampling efforts where systematic measurement of a spatially distributed obstructing feature is feasible during the LTDS survey.