Historical model biases in monthly high temperature anomalies indicate under-estimation of future temperature extremes

Both the mean climate state and anomalies from the mean determine the impact of extreme events, yet how models represent the latter is largely unexplored. Here, we assess skill of climate models participating in the Coupled Model Intercomparison Project Phase 6 (CMIP6) for predicting monthly high temperature anomalies relative to monthly means over time. Models project that, as the planet warms, these high temperature anomalies will increase in subtropical regions and decrease in high latitudes. Both postive and negative biases remain largely unchanged regionally and seasonally within two historical periods, 1980–2001 and 2002–2023. Globally, models have underestimated the 22-year average monthly high temperature anomalies by 2–3% and 22-year maximum anomalies by 11–12%. If historical biases of models carry forward, the on-average 2100 extreme high temperatures in some regions and months could be greater than current projections by 3 K, and greater than 5 K for the most extreme cases. Models tend to underestimate mean monthly maximum temperature anomalies by 2-3% and extreme anomalies by 11-12% over the period 1980 and 2023, which could lead to temperatures 3 °C to 5 °C higher than currently projected by 2100, according to an analysis of multi-model ensemble.