Current land use is a poor predictor of hellbender occurrence: why assumptions matter when predicting distributions of data‐deficient species

Abstract

AbstractAimUnderstanding species distributions is fundamental to effective conservation planning. Data deficiency is common among rare and imperiled species and poses challenges for conservation planning because status assessments become reliant on scant data that can introduce bias. We used occupancy modelling to evaluate support for commonly accepted, but previously untested, hypotheses regarding factors that drive the occurrence of an imperiled and data‐deficient amphibian, the eastern hellbender (Cryptobranchus alleganiensis). We investigated the potential for mismatch between areas likely to be identified as having high conservation priority based on the common assumption that hellbender occurrence corresponds to areas of high forest cover and those identified by well‐informed models.LocationSouth‐west Virginia, USA.MethodsWe conducted triplicate surveys to detect C. alleganiensis in 49 stream reaches stratified across a land use gradient and two major drainages. We used a Bayesian multimodel framework to investigate factors associated with C. alleganiensis occupancy. We used the best‐performing models to predict probability of occupancy at the scale of a 50‐m stream reach throughout our study area and identify areas most likely to be occupied.ResultsOccurrence of C. alleganiensis was explained primarily by differences in underlying geology and topography (i.e. physiography) and negative effects of agriculture were only modestly supported. Best‐performing models suggested ~ 35% of our study area was occupied. Our findings suggest that predictions from models informed by presence‐only data and current land use would likely underestimate C. alleganiensis occupancy by as much as one‐third and incorrectly classify over half the currently occupied area to be of little importance to the species.Main conclusionOur study highlights the potential danger of assuming that the distribution of data‐deficient species can be approximated using untested, but commonly accepted, species–habitat associations.