Global warming increases the proportion of more damaging heat extremes

Extreme heat events often result in considerable harm to both ecosystems and human populations. Heat extremes arise from diverse processes, resulting in heatwaves with distinct characteristics and therefore potentially strongly varying impacts and trends. Relying on the surface energy balance decomposition of temperature, we categorize terrestrial summer heat extremes from 1979 to 2020 into four types: Sunny-humid (36.5%), Sunny–dry (24.5%), Advective (25.0%), and Adiabatic (14.0%). Sunny-humid and Sunny-dry heat extremes are characterized by high-pressure systems and diminished cloud cover, resulting in heightened solar radiation. However, they diverge concerning soil moisture and latent heat fluxes. Conversely, the latter two types emerge from advective heating due to anomalies in the horizontal wind and adiabatic heating from air subsidence, respectively. Both are correlated with an upsurge in downward longwave radiation. Sunny-dry and Advective heat extremes lead to more detrimental effects on terrestrial ecosystem production (reducing net ecosystem uptake by 0.09 gC/m2/d and decreasing maize yield by 7.6%) and human health (raising the thermal stress index by 8.6 K and increasing human mortality by 3.3%), respectively. State-of-the-art climate models (CMIP6) generally replicate the relative proportions and the geographical distributions of the four types of heatwaves but tend to underestimate the Advective heatwave days. Under a high emission scenario (SSP585), the proportion of Sunny-dry and Advective heat extremes increases by 3.4% and 1.5%, respectively, while Sunny-humid and Adiabatic heatwave days decrease by 3.2% and 1.7%, respectively. This suggests, on top of the already expected increase in heatwaves, additional heat stress on both terrestrial carbon uptake potential and human populations. Our findings underscore the importance of distinction among different types of heat extremes and their impacts, paving the way to develop tailored adaptive.