Extended Data Fig. 4: Analyses of the latitudinal gradients in the residuals from three nested models.
From: Co-limitation towards lower latitudes shapes global forest diversity gradients
For each model, we show a scatter plot of absolute residual values (blue circles) and a column plot of root-mean-squared error (RMSE) calculated from the training dataset, with 2° bins and vertical axis in reverse order (black columns). Model (A) consisted of all 47 explanatory variables plus latitude. Model (B) contained all the explanatory variables in (A) except for the 21 bioclimatic variables. Model (C) as the base model only contained a zero constant. RMSE values in the upper-right corner of each panel represent the total RMSE, with p-values calculated from the one-sided F-test (see Variance Partitioning for details). p(A) stands for the p-value of all 47 explanatory variables plus latitude, and p(B) stands for the p-value of 21 bioclimatic variables.
