Macroecología y ecoinformáticasesgos, errores y predicciones en el modelado de distribuciones

Abstract

In spite of the biases and errors of the open access biodiversity databases we need to take advantage of the occurrences stored in those databases for analyzing the global patterns of biodiversity. Here, we aimed to test which modelling method produces better predictions when calibrated with data samples that have biases and errors. We tested two different methods, a complex one, Maxent, and a simple one, Bioclim. We created a virtual species, sampled its distribution with both, bias and errors, and calibrated the models with those samples. Results indicated that Bioclim produces better predictions than Maxent when calibrated with biased data sets. Bioclim did not overestimate the species� range and it was able to produce accurate predictions even when calibrated with small and biased data samples (25-50 points). However, when wrong occurrences were included in the calibration samples, Bioclim over-predicted the species� range. Our experiments indicated that in that case, Maxent predictions remained robust and provided accurate maps. Thus, if the calibration data samples have just biases, Bioclim provided better maps than Maxent. However, when samples have both, biases and wrong occurrences, Maxent model provided better results than Bioclim