I'm running some model of SDM, but I'm not clear if there is ecological difference in data sample or data mining structure to use any of these variations.
The attached references will help you to understand these terms and their implications for your model. Cross-validation, subsampling and bootstrapping are all resampling methods, but they perform resampling differently. In the case of MaxEnt, it seems that cross-validation is the preferred method but there is also some disadvantages that you need to consider.
The attached references will help you to understand these terms and their implications for your model. Cross-validation, subsampling and bootstrapping are all resampling methods, but they perform resampling differently. In the case of MaxEnt, it seems that cross-validation is the preferred method but there is also some disadvantages that you need to consider.
I would say you only can truly account for ecological meaningfulness of your model only while you are choosing your predictor and dependent variables, spatial context, spatial scale, as well as, somewhat analytic algorithm. Even models built on ecologically compromised data can produce statistically sound results. Certainly read papers attached by Luther.
These are three methods of sampling your presence data for multiple runs and don't have any intrinsic ecological significance. The method to use will often be determined by the number of presence values you have for your model. For example if you have a small number of presence data and want to do 100 reps then bootstrapping might be your only option. The ecological value of one over the others comes with tests of the predictive power of your final models.