Predicting fish species distribution in estuaries: Influence of species’ ecology in model accuracy

Current threats to biodiversity, combined with limited data availability, have made for species distribution models (SDMs) to be increasingly used due to their ability to predict species’ potential distribution, by relating species occurrence with environmental estimates. Often used in ecology, conservation biology and environmental management, SDMs have been informing conservation strategies, and thus it is becoming crucial to understand how trustworthy their predictions are. Uncertainty in model predictions is expected, but knowing the origin of prediction errors may help reducing it. Indeed, uncertainty may be related not only with data quality and the modelling algorithm used, but also with species ecological characteristics. To investigate whether the performance of SDM’s may vary with species’ ecological characteristics, distribution models for 21 fish species occurring in estuaries from the Portuguese coast were examined. These models were built at two distinct spatial resolutions and seven environmental explanatory variables were used as predictors. SDMs’ accuracy was assessed with the area under the curve (AUC) of receiver operating characteristics (ROC) plots, sensitivity and specificity. Relationships between each measure of accuracy and species ecological characteristics were then examined. SDMs of the fish species presented small differences between the considered scales, and predictors as latitude, temperature and salinity were often selected at both scales. Measures of model accuracy presented differences between species and scales, but generally higher accuracy was obtained at smaller spatial scales. Among the ecological traits tested, species feeding mode and estuarine use functional groups were the most influential on the performance of distribution models. Habitat tolerance (number of habitat types frequented), species abundance, body size and spawning period also showed some effect. This analyses will contribute to distinguish, based on species ecological characteristics, between species whose occurrence is reliably or less reliably predicted by distribution models. Fish species assigned to the benthivores feeding group, and using estuaries opportunistically (either as feeding or nursery grounds) seem to be more accurately predicted by SDM’s.