Fig. 2: Role of temperature variance in driving the spatial heterogeneity of future hot extremes.

Summer season averaged temperature variance in historical simulations (1981–2000, a) and the difference of that between future (2081-2100) and historical simulations (1981–2000) in the high-resolution Community Earth System Model (HR-CESM) (b). Grid points with a negative correlation between hot extreme change and historical temperature variance are shaded by plus markers in (a). c Scatter plot between anomalous hot extreme total duration changes (future minus historical) and anomalous historical temperature variance (historical) in the regions highlighted by plus markers in (a). d Scatter plot between anomalous hot extreme total duration changes (future minus historical) and anomalous temperature variance changes (future minus historical) in the regions highlighted by plus markers in (a). The anomalous values plotted in (c) and (d) are computed as the absolute values minus the global mean. e Changes in T90p (the proportion of temperatures exceeding the 90th percentile) relative to different standard deviation levels (from 1σ to 4σ) and mean temperature shifts based on idealized Gaussian distributions. f Temperature probability distribution functions (PDFs) for high-response regions identified by K-means clustering in historical (blue solid line) and future (red solid line) simulations in HR-CESM. The red dashed line represents the shifted PDF corresponding to the mean temperature change. Bar plots show region-averaged total duration and temperature variance for historical (blue) and future (red) periods, with corresponding values and fold increases labeled above the bars. g Same as (f) but for low-response regions.