Fig. 3: Factors influencing the spatial variability in live biomass carbon changes during 2016-2022 over northern ecosystems.
From: Large live biomass carbon losses from droughts in the northern temperate ecosystems during 2016-2022
a Relative importance of the variables explaining the spatial variability of biomass carbon changes in temperate biomes. b–e Partial dependence plots of the top four variables, namely, tree cover (b), surface solar radiation changes (ΔRadiation) (c), soil nitrogen (d) and soil moisture changes (ΔSM) (e). The Y-axis represents the SHAP value for the corresponding predictor (X-axis), where positive values contribute to higher predicted Δbiomass values, and negative values to lower ones. The red line indicates the median SHAP values across 50 bins of x-axis variables, plotted only for bins containing more than 20 samples explained at least 10 times by SHAP. The partial dependence plots illustrate how individual variables contribute to the predicted Δbiomass, capturing both their direct effects and interactions with other variables (Methods). f Similar to (a), but for boreal biomes. g–j Partial dependence plots of the top four explanatory variables in boreal biomes, namely, ΔRadiation (g), non-fire-induced forest losses changes (h), forest loss due to stand-replacing wildfires (i) and vapor pressure deficit changes (ΔVPD) ( j). Details on these variables are given in Supplementary Table 2. Source data are provided as a Source Data file.
