Do pseudo-absence selection strategies influence species distribution models and their predictions? An information-theoretic approach based on simulated data

Table 1 AIC-based calculations for the various pseudo-absence strategies.

Pseudo-absence Selection strategy	Ranking of true model out of 726 candidate models based on Akaike weights (w_i)	Delta AIC (ΔAIC) of the true model	Akaike weight (w_i) of the true model	Akaike weight (w_i) of the top ranking model
ENFA
100	9	3.604723	0.023729	0.143891
90	42	6.920832	0.004007	0.127529
80	16	3.882684	0.018746	0.130621
70	14	3.845144	0.021403	0.146365
60	13	3.863541	0.021315	0.14711
50	12	3.737772	0.025063843	0.162440578
Bioclim
100	55	9.80294298	0.001365089	0.183588
90	22	3.599437	0.013378	0.080906
80	33	3.027241	0.009334	0.042406
70	38	3.581413	0.007646	0.045825144
60	26	3.146754	0.0108	0.052088
50	33	3.708884	0.008264	0.052791
Random	7	1.474885235	0.034985305	0.073139658
True absences	2	1.130481	0.104319	0.183588

For each pseudo-absence strategy we fit 726 candidate logistic regression models using 131 presence records and 10,000 pseudo-absences. (ΔAIC) values less than 2 indicate considerable support, > 2 but < 10 indicate substantially less, and > 10 indicate essentially no support by the data. Akaike weights (w_i) can be interpreted as the percent of times we would expect a candidate model to be the one most strongly supported by the data if the experiment were repeated on a different sample.

ISSN: 1472-6785