We employed the Random Forest algorithm to integrate climatic and non-climatic drivers from 2000-2020 as explanatory variables. These drivers include global meteorological factors (precipitation, temperature, potential evapotranspiration, soil moisture), leaf area index, land cover, soil physicochemical properties (particle composition, pH), and topographic features (slope, elevation). We then generated a global ER gridded dataset with a spatial resolution of 0.5°×0.5°. The 10-fold cross-validation R2 and root mean square error (RMSE) values of the RF model is 0.55 and 15 t ha-1 yr-1, respectively.