Fig. 2: Variable selection and effects.

a–c Variable selection order determined by recursive feature elimination based on Random Forest and 5-fold cross-validation for leaf N (a), P (b), and N:P (c). The last (and most important) variable to be removed in the recursive feature elimination is plotted at the bottom. The R² indicated by the bar at the bottom of panels a–c is for models with a single predictor (‘ndep’ for leaf N and P, and ‘co2’ for leaf N:P). The R² indicated by the next bar above is for a model with one additional predictor, as indicated by the label along the y-axis. The final selection of variables is indicated by the green bars. Brown bars indicate additional, next most important predictors, but not used for subsequent analyses. d–f Effect magnitudes of the selected variables, measured by the coefficients of normalized fixed effects in LMMs. Only variables for which the t-value in the respective LMM was significant at the 1%-level are shown. ‘ndep’ is nitrogen deposition, ‘tmonthmin’ is the mean temperature of the coldest month, ‘ALSA’ is the aluminum saturation of the soil solution, ‘co2’ is the atmospheric CO₂ concentration of the respective measurement year, ‘elv’ is elevation above sea level, ‘mav’ is the mean daytime vapor pressure deficit, ‘gs_accl’ is the predicted optimal stomatal conductance, ‘ai’ is the aridity index, ‘pmonthmin’ is the precipitation of the driest month, ‘AWC_CLASS’ is the available water storage capacity class, ‘mapgs’ is the mean growing season-total precipitation, ‘map’ is the mean annual precipitation. The remaining variable names are explained in Supplementary Tables 1–2. Source data are provided as a Source Data file.