Ideally, you could make these models in R, here is a tutorial for the AUC: https://www.datanalytics.com/2016/03/29/el-auc-es-la-probabilidad-de-que/

A basic form is the basic way to calculate AUC is by using an average of a series of trapezoidal approximations of your area.The AUC is equal to the probability that a random sample of positive observation has a prediction of the probability (of being positive) greater than that of a random sample of negative observation.

You may find a partial answer in Duque-Lazo et al. 2016

Dear Matthew,

Are you using R statistical software or the stand-alone MaxEnt application? If the former, the solution is far easy. All you need are

1) measured predictors in the independent dataset

2) doing a prediction using the predictors and the trained MaxEnt model

3) observed values of the dependent variable in the independent dataset

4) compare the observed values to the predicted values using package 'ROCR', i.e. calculating AUC.

If you need more help, let us know!

HTH,

Ákos

Thank you everyone for helping to answer my question.

Ákos Bede-Fazekas I used the standalone MaxEnt application to generate the model. Can I upload the MaxEnt output into R to calculate the AUC scores? Or does the application have to be ran through R first?

Thank you,

Matthew

Dear Matthew,

Although I've never used this function and its documentation does not mention files created by the stand-alone MaxEnt application, but you should try load.model() from package 'maxent'.

https://www.rdocumentation.org/packages/maxent/versions/1.3.3.1/topics/load.model

HTH,

Ákos

Dear Mathew,

For accuracy assessment i would recommend partial ROC instead of AUC which you can do through online ENMTools ( shiny.conabio.gob.mx:3838/nichetoolb2/ ) (Peterson et al., 2008).

See

Peterson, A. T., M. Papeş, and J. Soberón. 2008. Rethinking receiver operating characteristic analysis applications in ecological niche modeling. Ecological Modelling 2131:63–72.

Thank You.

Uzma

This may help. Here is some R code to perform the test. I am assuming you have your data in a directory in your computer. This calculations follow Peterson et al. 2008.

# Installing and loading packages

if(!require(devtools)){

install.packages("devtools")

}

if(!require(kuenm)){

devtools::install_github("marlonecobos/kuenm")

}

if(!require(raster)){

install.packages("raster")

}

library(kuenm)

library(raster)

# your model

model