Integrating dynamic environmental predictors and species occurrences: Toward true dynamic species distribution models

While biological distributions are not static and change/evolve through space and time, nonstationarity of climatic and land-use conditions is frequently neglected in species distribution models. Even recent techniques accounting for spatiotemporal variation of species occurrence basically consider the environmental predictors as static; specifically, in most studies using species distribution models, predictor values are averaged over a 50- or 30-year time period. This could lead to a strong bias due to monthly/annual variation between the climatic conditions in which species' locations were recorded and those used to develop species distribution models or even a complete mismatch if locations have been recorded more recently. Moreover, the impact of land-use change has only recently begun to be fully explored in species distribution models, but again without considering year-specific values. Excluding dynamic climate and land-use predictors could provide misleading estimation of species distribution. In recent years, however, open-access spatially explicit databases that provide high-resolution monthly and annual variation in climate (for the period 1901-2016) and land-use (for the period 1992-2015) conditions at a global scale have become available. Combining species locations collected in a given month of a given year with the relative climatic and land-use predictors derived from these datasets would thus lead to the development of true dynamic species distribution models (D-SDMs), improving predictive accuracy and avoiding mismatch between species locations and predictor variables. Thus, we strongly encourage modelers to develop D-SDMs using month- and year-specific climatic data as well as year-specific land-use data that match the period in which species data were collected.