Fig. 2: Contribution of environmental variables to edge effects \((\frac{\Delta {{\bf{AGB}}}}{\Delta {\boldsymbol{D}}})\).

a, SHAP summary plot showing the contribution of each environmental variable to predicted edge effects across grid cells. Variables are ranked by their mean absolute SHAP value (|SHAP|), with the most influential variables listed at the top. The x axis indicates the SHAP value (that is, contribution to prediction) and each dot represents a local (grid-cell level) SHAP value. The overall distribution of points illustrates the global importance of each variable. b,c, SHAP dependence plots for MAT (b) and agricultural land cover (c). The x axis shows the variable value and the y axis shows its corresponding SHAP value. Colour shading reflects the density of data points, with lighter colours indicating higher density. d,e, SHAP dependence plot for MAP, coloured by MAT (d) and agriculture (e) to illustrate interaction effects between variables. The red lines in b–e represent LOESS-smoothed trends. f, Dominant environmental variable by biome. Each grid cell is coloured according to the SHAP value of the single most important variable, defined by the highest mean |SHAP|, from biome-specific XGBoost models for tropical/subtropical, temperate and boreal forests. Only the top variable per biome is shown; for full variable sets and biome-specific results, see Methods and Extended Data Fig. 4.