Transferability and model evaluation in ecological niche modeling: a comparison of GARP and Maxent

摘要: We compared predictive success in two common algorithms for modeling species' ecological niches, GARP and Maxent, a situation that challenged the to be generalthat is, able predict distributions broad unsampled regions, here termed transferability. The results were strikingly different between algorithmsMaxent models reconstructed overall of species at low thresholds, but higher levels Maxent predictions reflected overfitting input data; models, on other hand, succeeded anticipating most distributional potential, cost increased (apparent, least) commission error. Receiver operating characteristic (ROC) tests weak discerning into areas from those not. Such transferability is clearly novel challenge algorithms, requires qualities than does predicting within densely sampled landscapesin this case, was transferable only very biases gaps data may frequently affect based requiring careful interpretation model results.