Skip to main content
. 2013 Sep 5;8(9):e74816. doi: 10.1371/journal.pone.0074816

Table 1. Traditional and modified ROC AUC values for ENMs built on GARP and Maxent algorithms.

MODEL TRAINING MODEL TESTING PARAMETER GARP Maxent
On-Diagonal Off-Diagonal Minimum 1.0914 1.2605
Maximum 1.6151 1.7410
Mean 1.4528 1.4403
Standard deviation 0.0884 0.0876
# replicates≤1 0 0
P <0.001 <0.001
Traditional ROC AUC 0.8376 0.886
Off-Diagonal On-Diagonal Minimum 1.0171 1.2317
Maximum 1.6569 1.8229
Mean 1.3137 1.3439
Standard deviation 0.2075 0.1637
# replicates≤1 0 0
P <0.001 <0.001
Traditional ROC AUC 0.8377 0.8770
1980s 2000s Minimum 1.0014 1.1259
Maximum 1.504 1.5011
Mean 1.1025 1.1541
Standard deviation 0.0992 0.0609
# replicates≤1 0 0
P <0.001 <0.001
Traditional ROC AUC 0.7306 0.7483
1980s projected onto 2000s conditions 2000s Minimum 1.4170 1.4348
Maximum 1.5431 1.7221
Mean 1.4377 1.4449
Standard deviation 0.2 0.326
# replicates≤1 0 0
P <0.001 <0.001
Traditional ROC AUC 0.7964 0.8622

Traditional and modified ROC AUC values for GARP and Maxent models trained and tested with independent spatial subsets of WHO occurrence data subsets (first two sections). Performance of models trained with data from WHO (1980s) and applied directly to recent occurrence data, or projected onto environmental conditions of the 2000s is shown in the succeeding sections. Modified ROC AUC values are reported as minimum, maximum, mean, and standard deviation of 1000 random replicates; also provided is the number of bootstrap replicates falling at or below 1.