Modelling occupancy of an imperilled stream fish at multiple scales while accounting for imperfect detection: implications for conservation

1. Predictive models of species distribution are useful tools to identify habitats of imperilled species for protection, inventory and restoration. Critical aspects of such models include the influence of scale, uncertainties associated with imperfect detection and spatial autocorrelation and transferability of model predictions. 2. We addressed these issues in developing occupancy models of the imperilled eastern sand darter (Ammocrypta pellucida) based on surveys of the Grand and Thames Rivers, Ontario, Canada. 3. Eastern sand darter detection probabilities were remarkably different between streams, but factors affecting site occupancy were similar. The proportion of sand and fine gravel was most important, but water clarity and biotic indices also received support in additive models. Accounting for spatial autocorrelation reduced the effect of important covariates. 4. Occupancy was more closely related to substratum at the site level than factors at broader scales (reach and valley segment), further emphasising the substratum specificity of this species. 5. Almost all of the top-ranked site and reach occupancy models had good predictive performance based on assessments of transferability. These models indicate that three formerly occupied Ontario catchments have a high probability of supporting the species and deserve consideration for repatriation. 6. Our methods demonstrate how a comprehensive approach to occupancy modelling can be used to help guide recovery efforts for imperilled species.