Extended Data Fig. 6: Accuracy of original ecosystem-cover predictions.
![Extended Data Fig. 6](https://cdn.statically.io/img/media.springernature.com/full/springer-static/esm/art%3A10.1038%2Fs41586-020-2784-9/MediaObjects/41586_2020_2784_Fig10_ESM.jpg)
a–f, The accuracy of the predictions of the original proportion of each ecosystem type in each planning unit was quantified using the root mean square error (r.m.s.e.). To better understand any heterogeneity in prediction accuracy, we calculated the r.m.s.e. separately for each of the five land-cover classes (forest, grassland, shrubland, wetland and desert) in addition to the overall r.m.s.e. Overall, predictive accuracy was excellent (total r.m.s.e 6.73%, f) with relatively little variation among the five land-cover types: forests, 4.0% (a); grasslands, 1.7% (b); shrublands, 4.3% (c); wetlands, 1.2% (d); and arid areas, 2.6% (e).