ENMeval and Dismo tips on cross-validation

[giuseppe D'ambra]

Hi guys i am developing maxent SDM for different species, i would like to gain some suggestion from you as i am new in the world of species distribution modelling, maxent and R. I red some papers which highly suggest in order to build a solid one to make a species specific tuning of the model based on different factors such as number of occurrence and so on. Hence, in order to maximize model predictive perfomance and minimize overfitting i am doing a pre-assessment of the model using the package ENMeval which i found very useful and very easy to implement. Assuming that i am using as Method for partitioning data "jackknife leave-one-out" for small sample size (eg<25) and "block" for all the other (in the paper of ENMeval, this partition data method has been suggested as among the best to transfer then the model ), Do you think is right if  once i found the best model which will have the best combination of AUC, ΔAICc and ORmtp, i will use that combination of features and Rmvalues to run the model again in dismo, using the same kind  and bins of partition data and using additional arguments in order to perform a cross-validation? The idea is that i want to assess the robustness of my model on multiple run, is this a good practice in doing so or am i completely wrong?

If not, there is a way to assess the robustness of the model using the best combination of feature class and Rm values?

Furthermore, can you suggest any parameters to assess the transferability of my model in time(under climate change scenario)? is Boyce the best index in this case? 

Sorry for the multiple questions, any suggestion is really appreciated. 

Many thanks