Fig. 1: Risk of heat mortality for São Paulo, Paris and Bangkok.

Risk of heat mortality for São Paulo (Brasil, a, d, g), Paris (France, b, e, h) and Bangkok (Thailand, c, f, i). a–c Relative risk of mortality relative to the location-specific minimum mortality temperatures reported as best linear unbiased predictions (BLUPs) with 95% confidence interval (shaded area). Vertical dotted lines show the log-linear extrapolation used for projections when future temperatures exceed current temperatures. Dashed vertical lines show present-day 99th percentile temperatures. d–f Impact exceedance frequency curves of annual heat mortality fractions for the observed years (black line, markers denote individual years), as well as the climate of 2000 (warming level of 0.7 °C), the climate of 2020 (warming level of 1.2 °C), 1.5 °C warming and 2 °C warming. The modelled impact exceedance frequency curves are reported as the median value over the five single-model initial-condition large ensembles (SMILEs). g–i Modelled magnitude of the annual heat mortality fraction from a 1-in-100 year season for different warming levels. Results are displayed for each SMILE (bars) including uncertainty estimates depicted with the 95% empirical confidence intervals accounting for the internal climate variability (inner whisker) and imprecision of the exposure-response associations (outer whisker). The black horizontal line denotes the median estimate for each global warming level.