TABLE 6.

Performances of correction methods, reported as average of the five top models, for the real datasets (RD), in Qilianshan National Park (QLSNP) and Qomolangma National Nature Reserve (QMLNR). MTSS = Maximum training sensitivity plus specificity logistic threshold; MTSS om = omission rate for MTSS threshold. D represents Schoener's D (Schoener, 1968) niche overlap index, calculated for each ensemble and dataset with respect to FR_RAW ensemble of models (Full Random dataset, FR)

Dataset	Models	N	AUC	AUCdiff	MTSS	MTSS om	D	ΔDgeo
QLSNP
RD	QLSNP_RD_RAW	393	0,858	0,008	0,398	0,271	0,810	0,000
	QLSNP_RD_SR1200	219	0,837	0,009	0,415	0,272	0,841	0,166
	QLSNP_RD_SR2400	158	0,819	0,009	0,424	0,290	0,858	0,254
	QLSNP_RD_SR4800	103	0,766	0,023	0,443	0,421	0,892	0,433
	QLSNP_RD_SR9600	65	0,756	0,034	0,453	0,407	0,921	0,584
	QLSNP_RD_GK1200	393	0,860	0,009	0,417	0,268	0,822	0,066
	QLSNP_RD_GK2400	393	0,862	0,009	0,429	0,262	0,835	0,131
	QLSNP_RD_GK4800	393	0,863	0,009	0,448	0,255	0,860	0,267
	QLSNP_RD_GK9600	393	0,855	0,009	0,471	0,290	0,892	0,433
QMLNR
RD	QMLNR_RD_RAW	220	0,972	0,001	0,194	0,085	0,572	0,000
	QMLNR_RD_SR1200	79	0,958	0,004	0,188	0,126	0,667	0,221
	QMLNR_RD_SR2400	49	0,943	0,007	0,212	0,115	0,780	0,486
	QMLNR_RD_SR4800	28	0,905	0,015	0,331	0,190	0,844	0,636
	QMLNR_RD_SR9600	19	0,904	0,015	0,316	0,150	0,759	0,437
	QMLNR_RD_GK1200	220	0,971	0,001	0,263	0,085	0,646	0,173
	QMLNR_RD_GK2400	220	0,971	0,001	0,321	0,085	0,699	0,297
	QMLNR_RD_GK4800	220	0,971	0,002	0,432	0,082	0,763	0,445
	QMLNR_RD_GK9600	220	0,968	0,002	0,501	0,081	0,768	0,457

ΔD_geo calculated using D overlap with FR_RAW ensemble of models (ΔD_geo=(D_corrected‐D_biased)/(1‐D_biased); Fourcade et al., 2014). Values in bold represent improvement or equal performance with respect to the raw models. The model selected as the best correction (maximizing D, with positive ΔD_geo while reducing MTSS omission rate) is reported in bold and italic.