This paper may be of the interest of some friends working with this species in South America. This species is really making a mess in many places, and I am glad people are researching on this issue.

To me the paper have several pros and cons in regard to the methods.

Pros:

1) Training the model on a study region made with a Minimum Convex Polygon of localities likely exclude areas that may contain suitable climate for the species, but the species does not inhabit because of historical reasons (barriers) or biotic interactions. This approach seems to be better than using the (sometime huge) rectangles...

2) They tuned the regularization multipliers


Cons:

3) They only used AUC/ROC to evaluate models. This mean of evaluation is not sufficient to evaluate model performance.

4) THE MOST IMPORTANT: The biological question of this paper relay on model transferability, and the authors did not evaluated model transferability at all (according to the paper). As with any other study in which transferability is not evaluated, I would take the conclusions of this paper with a grain of salt.