Table 1. Logistic regression model and parameters.

i		θi	θi (standardized)
0	Intercept	4.16	1.80
1	Sand	2.38 e−02	6.08 e-01
2	Clay	−1.88 e−01	1.10
3	Density	−5.99	−9.09 e−02
4	Clay/density	1.83 e−01	4.79 e−01
5	MSI	−7.05	−1.09
6	MAR	2.90 e−03	2.29

Model fitting was done on 50,000 samples (1 km spatial resolution). All parameters have P values < 1e−04. Forest is predicted where the log odds of forest occurrence (see Methods for details). The κ-agreement index with data is 0.69 (substantial agreement). See Supplementary Fig. 2 and Supplementary Table 1 for alternative prediction models.