Epistemic uncertainty in predicting shorebird biogeography affected by sea-level rise

[Display omitted] ► Sources of epistemic uncertainty are analyzed. ► The land-cover strongly affects the uncertainty of species distribution predictions. ► Care has to be posed on the selection of environmental covariates ► MaxEnt results to be the best model. ► A similarity index is introduced for assessing model-fit. Accurate spatio-temporal predictions of land-cover are fundamentally important for assessing geomorphological and ecological patterns and processes. This study quantifies the epistemic uncertainty in the species distribution modeling, which is generated by spatio-temporal gaps between the biogeographical data, model selection and model complexity. Epistemic uncertainty is generally given by the sum of subjective and objective uncertainty. The subjective uncertainty generated by the modeler-choice in the manipulation of the environmental variables was analyzed. The Snowy Plover in Florida (Charadrius alexandrinus nivosus, SP), a residential shorebird whose geographic range is extended along the Panhandle-Big Bend-Peninsula Gulf coast was considered as case-study. The first fundamental step for studying the species distribution and how it will be affected by climate change is to obtain an accurate description of the shorebird coastal habitat. The land-cover was translated into ecosystem classes using a land-cover model that predicts the evolution of coastal ecosystems affected by sea-level rise scenarios. The best land-cover map decreased the objective uncertainty (intrinsically present in data or models) in representing the spatial structure of the coastal ecosystem, reduced the temporal gaps with the occurrence data, and diminished the subjective uncertainty due to the conversion from land-cover to model-classes. Multimodeling was performed to reduce the uncertainty in the prediction of the species distribution related to model uncertainty. The best representation of the species distribution was performed by MaxEnt. The area under the receiver operating characteristic curve (AUC), the omission/commission test, the similarity index of the response curves, and the jackknife test were used simultaneously as indicators of the predictability of each species distribution model. The availability of updated high-resolution biogeoclimatological data was proven to be necessary in order to properly predict the species ranges for conservation purposes.