Expanding ensembles of species present-day and future climatic suitability to consider the limitations of species occurrence data

Outputs of species distribution models (SDMs) are widely used as indicators of climate conditions favorable for species occurrence. When using these outputs to inform planning and decision making, it is essential that the uncertainties associated with the projections of present-day and future climatic suitability are carefully considered. Climate change assessments routinely consider the uncertainty introduced into SDM outputs by differences in future climate projections, and other uncertainty sources, such as the choice of the threshold to convert simulated probabilities to binary climatically suitable areas, are also oftentimes considered. However, the uncertainty introduced by the limitations of the species occurrence data used in the SDM calibration is rarely evaluated. These limitations, which include location error, sampling bias, and species misidentification, may reduce the utility of SDM outputs in conservation research and practice. Using understory bamboo species in southwest China as examples, here we demonstrate that species occurrences obtained using remote sensing offer an additional dataset for calibrating SDMs that, in conjunction with conventional observations and employing an ensemble approach of outputs from multiple models, provide an estimate of the uncertainty introduced by the species occurrence data. A biweekly time series of the satellite-based Wide Dynamic Range Vegetation Index (WDRI) was employed to estimate bamboo occurrence based on phenological signatures of the bamboo species and their overstory canopies. Using Maxent, a popular modeling framework, present-day and projected future climatic suitability were assessed separately for conventional species presence observations from the Fourth National Giant Panda Survey and for the remotely-sensed presence estimates. The ensemble of model outputs suggests that the uncertainty introduced by the species occurrence data, along with the interaction with other sources of uncertainty, may be as substantial as the uncertainty introduced by the use of different climate scenarios or by the threshold used to estimate binary climatically suitable areas. Ignoring the uncertainty introduced by the limitations of the species occurrences may compromise the interpretation of SDM outputs and reduce their usefulness for conservation planning. Remote sensing is a largely untapped resource for assessing uncertainty in SDM simulations.