Table 2.

Variable importance for how each parameter type influences the forecast error of predicted deforestation rates, as well as statistical tests (one-way ANOVA and Tukey’s HSD post-hoc) comparing forecast errors between the different levels of each parameter, for n = 2794 jurisdictions

Parameter	Variable Importance		Statistical tests
	Mean	SE	One-Way ANOVA	Tukey’s HSD post-hoc
Projection approach	4.90	0.12	F(6,19424) = 366, p < 0.01	All pairs p < 0.01, except between linear models global, global_s, regional, and regional_s (p = 1.0)
Forest dataset	2.28	0.16	F(4,12646) = 63, p < 0.01	All pairs p < 0.05, except between the hansen15-hansen30 (p = 0.70), hansen30-hansen60 (p = 0.07), and hansen60-esacci (p = 0.76)
Historical reference years	0.82	0.42	ANOVA F(5,16600) = 4.2, p < 0.01	5–10, 6–10, 7–10, 8–10 (p < 0.05), all other pairs p > 0.05

^aBold indicates statistically-significant values