Variable selection through BIOMOD package

[Gafarou AGOUNDE]

Hi everyone. Hope you are doing fine.

I’m currently running the SDM for a plant species of West Africa through the Biomod2 package. However, I have a concern about variable selection or the variables I should retain for my target species. Indeed, I used 29 variables globally (17 bioclimatic and 12 soil nutrients). So, after running the code (1 and 2) I had in my Rstudio console the table1. Thus, my goal was to retain the 5 most contributing variables by algorithm based on the greatest values. By doing this, I retained bio12, bio7, bio3, bio2, and bio10 (see blue column table1). However, when I opened the HTML file of the MaxEnt in my directory, I noticed the most important variables color green (Table1) were not the same as those found in (Table2, green color). Thus, I’m very confused………..!

Honestly, after several read the tutorial of Biomod2, I didn't understand what represents values or scores for each variable in (Table1).

So, my questions are:

1-         Are the values in table1 the scores, percentage, or AIC?

2-         What’s the difference between the two tables?

3-         Should  I base only on table 1 for variable selection? If no...

4-         What is the best method to select variables from the Biomod2 package?

 

For me, variables with values toward 1  are the most important for my target species

Code (1 and 2):## check variable importance
(Crossopterix_models_var_import <- get_variables_importance(Crossopterix _models))

## make the mean of variable importance by algorithm
apply(Crossopterix _models_var_import, c(1,2), mean)

 

Please, find attached the two tables

Table 1: Table from the code

Table2 : HTML of the MaxEnt

[Purpleberry Fira]

Hi,

I also doing sdm for plant species. I also confused at first about using the VIF method alone to select variables. but i found this research by Feng et al. 2019, they discuss bout collinearity in enm. Maxent is capable of regulating redundant variable contributions, so collinearity in variables have less effect on maxent model. So bcoz of previous literature such as Elith et al and Phillips et al. 2011 suggest that we remove high correlated variables, what i did was analyse first the variable important and permutation important of the variables. from the high listed variables importance then we compare with the list of variables selected by VIF procedure or any algorithm procedure related. So the intercept variables from each list should be use in the model to create a niche model that is meaningful to plant species.  Finally from the slection u made, u can try do the VIF procedure again. In my study my final selected model all passed the VIF < 10.

--
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/70141304-3d92-4a85-9b3f-d88d46d2522an%40googlegroups.com.

[Gafarou AGOUNDE]

Thanks Dear Fira for your contribution.

So, could you please share with me the script you used for VIF?

Thanks...!

Management of Wildlife, Protected Areas and Rangeland | MSc, University of Abomey-Calavi, Benin republic. Phone number:(+229)96692934

To view this discussion on the web visit https://groups.google.com/d/msgid/maxent/CAMFn2sC%3DDWswCWZvALSUhY0Sp3T6oh4C9UMhHkZ7anFQ9o_3wg%40mail.gmail.com.

[Purpleberry Fira]

i followed the SDM package by babak naimi for VIF procedure. it is really easy. sorry i cant get the code right now im on phone. 

To view this discussion on the web visit https://groups.google.com/d/msgid/maxent/CAJj42dPEdA6mWhD1-jvV%3D7VMT%3DbGWmJPXyUJms6azuY9Urrcig%40mail.gmail.com.