Doubt regarding ENMeval and projections.

[Jhan Carlos Carrillo Restrepo]

Hi All. I hope this message finds you well.

In a recent conversation (https://groups.google.com/g/maxent/c/ITcKVLQx9TY), it has been suggested that one should use the selected model directly by ENMeval without building another model using the selected parameters (e.g. in Maxent). However, if I remain with the ENMeval model, how do I make projections to another geographical area? This question is because ENMeval doesn't include a function to make projections (in case it does, please excuse my ignorance). I've tried doing this with the predict function. However, when I compare the results of the projections generated with predict on ENMeval results and the projections generated rebuilding the model (in Maxent), these two are different. Is there a problem with the "predict" function when making projections that were generated with ENMeval? or Viceversa.

Another question, divided into two parts: I have some species with a good number of occurrences, but other species have very few points (10 points). Therefore, what is recommended in the literature is that I should use other algorithms than Maxent. I think BIOCLIM is a good option, but my first question is what do you suggest? Which algorithm has been found to work well with few occurrences and can generate good projections?
In your experience, what is the best R package (Biomod2, sdm, Wallace) to explore and complement all ENMeval functions and generate a complete and good workflow? With which package can I generate a script where I model multiple species and make it reproducible?

Thank you very much for taking the time to read this message. Excuse me if my doubts are very naive, I just want to know what I'm doing and do it right.

Best, Jhan.

[Purpleberry Fira]

Hi, i get the same problem with ENMeval. but then i tried to update the package and also use the latest maxent.

--
You received this message because you are subscribed to the Google Groups "Maxent" group.
To unsubscribe from this group and stop receiving emails from it, send an email to maxent+un...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/maxent/165aa7fa-da22-4640-8f24-f742a9c115f0n%40googlegroups.com.

[Jamie M. Kass]

To respond to the first question, if you are making a model with the exact same background points and settings (feature classes and reg. multipliers) using the Maxent Java software and with ENMeval using algorithm = "maxent.jar", the output for both should be identical. Using predict() for either with the same transformation type (i.e., cloglog, etc.) should return the same raster prediction.

Check out the paper below for strategies on modeling ranges for species with small sample sizes. Wallace is a good application to use if you want reproducibility because you can download the code used to run your analysis at the end, and the back-end uses ENMeval.

Shcheglovitova, M., & Anderson, R. P. (2013). Estimating optimal complexity for ecological niche models: a jackknife approach for species with small sample sizes. Ecological Modelling, 269, 9-17.

Breiner, F. T., Guisan, A., Bergamini, A., & Nobis, M. P. (2015). Overcoming limitations of modelling rare species by using ensembles of small models. Methods in Ecology and Evolution, 6(10), 1210-1218.

Jamie