Fig. 2: Relative importance of the potential predictors and conditional regression plots for important predictors.

a, b Relative importance of the potential predictors for SOC density (SOCD) based on linear model analysis (a) and random forest analysis (b). c–f Conditional regression plots with mean annual temperature (MAT) (c) temperature seasonality (d) precipitation seasonality (e), and urban greenness index (UGI) (f). Different colours represent different predictor groups. The variable importance shown in Fig. 2a is based on the sum of the Akaike weights derived from model selection using corrected Akaike information criterion. The cut-off is set at 0.8 (grey dashed line in a) to differentiate among the important predictors. The importance shown in Fig. 2b is based on Mean Decrease Gini of random forest models. The black solid lines in Fig. 1c–f indicate the conditional regression fit. The shaded areas in Fig. 1c–f represent the 95% confidence intervals. Only data with reported information on vegetation type were used for the analysis (n = 282) and an additional analysis was also conducted using all data (n = 420) (Supplementary Fig. 5). MAP mean annual precipitation, GDPP GDP per capita, PD population density, UHI urban heat island index.