Jackknife on variables/environmental predictors

[Simon Tarr]

Hello, is it possible to return information on jackknife of variable importance within R?

If one were to run a model using the MaxEnt GUI, you can tick "Do jackknife to measure variable importantance". I'd like to replicate this functionality within R. Ideally I'd like to use ENMEval to find the "best" model, along with the variable importance measures for that model. Is this possible?

Thanks!

[Cass Kalinski]

Simon,

the @results dataframe in the ENMevaluate object returns the variable importance for each model tested in ENMeval.

Cass

[Not to be confused with THE Kass!]

[Simon Tarr]

HI Cass, thanks for the reply.

I'm unsure exatly how the variable importance shown in @results relates to the training gain bar charts from the MaxEnt GUI when ticking "Do jackknife to measure variable importance".

The variable importance shown in @results is just the percentage contribution/perumtation importance- it doesn't give me any information on the absolute change in model gain when the explanatory variable is included/excluded. Further, if I run a model using dismo::maxent and check @results the list returned is much longer and gives me the absolute gain changes from the variable jackknifing process (i.e. it'll show gain with the variable and without the varaible), enabling me to plot this myself within R. 

I don't think ENMEval returns this information by default because I don't think it's jackknifing the predictor variables, although I could be wrong. If it's not run by default but there is an option to jackknife the variables, do you know how to do this?

I wanted this variable jackknifing information via ENMEval if possible because I have many species to analyse. I can take the optimal model identifed by ENMeval and pass this information manually to dismo::maxent to get the information that I want but it would be a really time consuming process. Especially for the feature classes: ENMeval returns a single character (e.g. L for linear) whereas dismo::maxent expects a string (e.g. args=c("linear=TRUE")). I'd probably have to write a script to convert all the single characters into strings.

Thanks

Simon

[Lays]

Hey, simon. Did you ever find out a way to jackknife the variables using enmeval? I am going through the same question and saw your messages here. 

Thanks

Lays