Effects of different variable sets on the potential distribution of fish species in the Amazon Basin

Estimating species' potential distribution is one of the main objectives of macroecology, especially when sampling biases can affect knowledge on how environmental variables affect species distribution. Ecological niche models estimate species' environmental niches from different variables and their occurrences. Using the presence-only data from eight Amazonian fish species, which inhabit rivers and streams, we aimed to (a) explore the effect of different sets variables on the spatial distributions of target species and (b) evaluate the predictive responses of MaxEnt to sets of variables with different degrees of complexity. MaxEnt has high flexibility in relation to the input data and its performance is influenced by a moderate number of adjustable parameters, allowing for high precision results when balancing underestimation and overestimation errors. We used environmental predictors in MaxEnt the principal components of climatic, topographic and edaphic variables as inputs. The combination of topographic and edaphic variables produced more precise and spatially restricted distribution ranges for all species when compared to those generated with climatic variables. All models reached high AUC values, especially for stream species. Modelled range sizes were broader for the river species, suggesting different tolerance thresholds and habitat preferences when compared to stream species. The complexity of the different variables sets did not affect MaxEnt's prediction capacity. However, for stream species, MaxEnt showed a greater predictive power. This work increases the knowledge with regards to the influence of different environmental predictors on the spatial patterns of the distribution of Amazonian fish.