Fig. 2: Relative importance of daily maximum (T_max) and minimum temperatures (T_min) in interannual variations of saturation vapor pressure (SVP) and actual vapor pressure (AVP) from 1980 to 2023.

a Spatial distribution of relative importance of T_max in interannual SVP, identified using the Random Forest (RF) regression model defined in Eq. (4) with the Shapley Additive Explanations (SHAP) framework. Inset shows boxplots of relative importance (%) across stations (solid boxes) and ERA5-Land grid points (hollow boxes). Pie chart shows the percentage of land area where T_max importance exceeds 50% (red) or is below 50% (blue), based on ERA5-Land data. b Spatial distribution of the dominant driver of interannual AVP changes, identified using the RF regression model defined in Eq. (5) with the SHAP framework, based on ERA5-Land data. Pie chart shows the percentage of land area where AVP is dominantly driven by T_max, T_min, or soil moisture (SM), with either a positive (+) or negative (−) influence. c and d spatial distribution of ridge regression (RR) coefficients of interannual AVP with respect to T_max (c) and T_min (d), derived from the RR model defined in Eq. (5). Insets show boxplots of RR coefficients across observation stations (solid boxes) and grid points (hollow boxes). Pie charts show the percentage of land area with positive (red), negative (blue), and non-significant (light grey) RR coefficients based on ERA5-Land data. “Non-significant” refers to cases where none of the coefficients in the RR model are statistically significant. Stations (4.36%) and areas (1.45%) with non-significant RR coefficients are masked in light grey and excluded from the analysis. All variables in the regressions are detrended and standardized annual averages. In all boxplots, the height of each box represents the interquartile, with the thick black line indicating the median, and the edges denoting the first and third quartiles. Whiskers extend to the 2.5th and 97.5th percentiles.