nullAUC: Generate null distribution models for effective AUC...

In brshipley/megaSDM: Integrating Dispersal and Time-step Analyses into Species Distribution Models

Generate null distribution models for effective AUC comparison

Description

AUC values rely on both omission rate (false negatives) and commision rate (false positives); however, MaxEnt is a presence-only method, making raw AUC values uninformative for comparing across models and species (Jimenez-Valverde 2012). One way to use AUC values to examine presence-only model predictions is to generate model replicates using randomly-generated occurrence data, evaluating their performance using a subset of the real occurrence data. This function generates null models and calculates the Test AUC values when applied to the subset of real occurrence data, for comparison with the model training on the actual data. This method was developed by Bohl et al. 2019.

Usage

nullAUC(envdata, replicates = 50, bufflist = NA, modelpar)

Arguments

envdata	a SpatRaster or list of raster files corresponding to the area the model will be trained on.
replicates	how many times should the null model be run to get a distribution of AUC values? Default is 50 replicates.
bufflist	(optional) if background points were spatially-constrained, provide the paths to the buffer files (.shp) used.
modelpar	a list of the arguments passed from the MaxEntModel function. These arguments should be exactly the same as the actual model generation for effective comparison. For the method described by Bohl et al. (2019), a list of test-samples must be given (evaluate models on the real data). Otherwise, the null distribution will be evaluated on a subset of the random samples (see Raes & ter Steege (2007))

Value

A .csv file (NullModel_AUC.csv) with the Test AUC values for each replicate of the null model.

brshipley/megaSDM documentation built on Nov. 26, 2024, 6:08 a.m.