Comparando las estimaciones de selección de hábitat mediante modelos de distribución de especies y step selection functions

Recently, two methods of habitat selection have gained more relevance in the ecological and animal distribution literature: step selection functions (SSF) and MaxEnt. Despite their similarity these models are hardly ever used in the same context. The former is usually associated with studies based in movement ecology, and the latter is connected to species distribution modeling. Motivated by the difficulty in estimating habitat preferences using SSF, I compared the accuracy of predictions from both models based on movement data. As a case study, I used jaguar movement data from 5 countries in Latin America and created SSF and MaxEnt models based on climatic data and land use available from WorldClim and satellite imagery. I compared the accuracy of both types of models using the “Area Under Curve” (AUC) metric, on a separate subset of data. SSF models presented an average AUC of 0.5482 ± 0.0217 in comparison with 0.7205 ± 0.0142 of their MaxEnt equivalents. I believe those differences are partially caused by the convergence difficulties of SSF and its use of conditional logistic regression. Consequently, I recommend the use of MaxEnt in predictive modelling, such as the ones needed in reserve and corridor design.