Dear Minghao,

Have a look https://github.com/marlonecobos/kuenm

https://peerj.com/articles/6281/

I hope this will help you to resolve your problem.

Dear Minghao,

If you use R and its package ENMeval, then you should train and evaluate your model through the function ENMevaluate(). Don't use the stand-alone maxent.jar.

HTH,

Ákos

Ákos Bede-Fazekas thanks for your reply, "Don't use the stand-alone Maxent.jar." it may be difficult for me, because I need to use the categorical variable like soil classification data as a factor. So using the stand-alone Maxent.jar may be the easiest way for me to get my goal. So it really makes me be stuck in trouble.

Imtiaz Husain thanks for your reply, this R package may aim to select the best performance of models, I have used it before but I find some warnings when I get in some the categorical variables as factors like soil classification data and so on. So thanks for your reply and inspiration again.

Pietro Milanesi thanks for your reply, maybe it is not easy to find a way to evaluate the models of stand-alone Maxent.jar. And I have read this paper about AIC, but what is different is I don't want to directly evaluate the predictive model by AIC or other evaluations, what I want to do is using those evaluations indirectly evaluate the effect of different variables I adding to the models. My goal of using Maxent model is that I want to find the indirect or direct effect between abiotic and biotic variables by comparing models adding different variables. So those evaluations are good ways to convert models to data which can be used in data analysis. And thanks for your reply again.

read in .asc via raster function in raster package. use extract function from raster package to read predictions at presence&background locations. observations are simply c(rep(1, Np), rep(0, Nbg)).

u can use this to derive the treshold(s) you want and generate the confusion matrix(/-ces) to be evaluated by TSS, SEDI etc.

Regardin categorical rasters: no you do not need them. I are to say almost never. they are based on underlying variables such as soil pH, moisture etc which will be more useful as continuous variables.

Rainer Ferdinand Wunderlich thanks for your reply, it is very inspiring, but can you talk more details and steps? Or may you recommend some books or papers about it? And I agree with you about soil, I try to improve my model analysis. Thanks for your reply again.

I suggest getting a grip of R by doing a introductory tutorial and readimg the vignette of the raster package. You can also look at the dismo or maxnet packages of enmeval looks too complex.

Rainer Ferdinand Wunderlich Thanks very much!

Thanks for using the package! ENMevaluate() can incorporate categorical variables into the model by specifying their names as a vector to the "categoricals" argument. AICc is only useful when calculated over multiple model prediction rasters, and the model with delta AICc = 0 represents the best performing model (though delta AICc

If you prefer more of a point and click interface (with R running the show behind the scenes), you can try out the Software for Assisted Habitat Modeling (Morisette et al 2013) package for the VisTrails program.

Morisette, J. T., Jarnevich, C. S., Holcombe, T. R., Talbert, C. B., Ignizio, D., Talbert, M. K., ... & Young, N. E. (2013). VisTrails SAHM: visualization and workflow management for species habitat modeling. Ecography, 36(2), 129-135.

Yes, new GUI tools exist to run these models, evaluate, and visualize their outputs. Another option is the Wallace ecological modeling app: https://wallaceecomod.github.io/. Wallace uses ENMeval in the background. There are others great GUI tools too for niche modeling, like ntbox and shinyBIOMOD.

Peder Engelstad thanks very much! I will read this package and try to do it. Thanks again!

Jamie M Kass thanks! Although I am a new user of R, I get a lot from the package ENMveal. I will do some tries by your advices, thanks again!

Dear all, please inform if there is any available source of workflow (from the scratch) of ENMeval, begining from loading of raster variables and species occurrences.

Thanks

Prakash

Here's the latest ENMeval vignette, which starts from scratch: https://jamiemkass.github.io/ENMeval/articles/ENMeval-2.0.0-vignette.html

Thanks Jamie, hoping to cite you soon

Hello Jamie M Kass , have any example that to project to climate change scenery using user select best model?

If your future climate layers have the same names as the layers used to train your model, you can use dismo::predict on the model object for your chosen model (you can get these objects from e@models, where e is the ENMevaluation object).