In a warming world, temperature extremes are projected to significantly alter global weather patterns, even at a relatively modest 1.5°C warming. Many studies have already observed this change in various regions. However, inconsistencies persist among climate models in predicting the magnitude of these changes, even when controlling for the same level of mean global warming. These inconsistencies are partly attributed to differences in model representation of land-atmosphere feedback mechanisms. Daily maximum temperature (TX), frequently used to assess heatwave intensity, is influenced by various processes such as solar radiation, heat transport, and the exchange of sensible and latent heat fluxes with the surface. The evaporation of surface moisture plays a crucial role in mitigating atmospheric warming and, consequently, TX. TX variability tends to be greater under drier surface conditions compared to wetter ones; soil moisture deficit can amplify TX and heatwaves. Evidence suggests that many current climate models inaccurately simulate soil moisture, potentially exaggerating TX variability. This study investigates this hypothesis, postulating that more realistic models may reveal a more significant amplification of heatwave frequency in coming decades, leading to substantial discrepancies in projected heatwave changes under future warming scenarios. Prior research has indicated a reduction in temperature change uncertainties when using prescribed land heat fluxes and found a systematic increase in interannual variability of summer temperatures in Europe using models that realistically represent variability. This research acknowledges the influence of multiple factors on TX variability, beyond surface heat fluxes, such as dynamics and aerosols; however, it focuses on surface heat fluxes as a primary driver.

Existing research highlights the observed and projected increases in temperature extremes globally. Studies have demonstrated the significant societal impacts of extreme heat events on human health, agriculture, and water resources. The role of soil moisture in modulating surface temperatures has been explored, with evidence suggesting a positive correlation between soil moisture deficit and amplified temperature extremes. However, inconsistencies exist among climate models in accurately representing land-atmosphere interactions and their effects on temperature variability. Previous studies have explored using prescribed land heat fluxes to reduce uncertainties in future temperature projections. Additionally, work has shown a relationship between realistic representation of temperature variability in models and more accurate predictions of future interannual variability. These studies form the basis for this research, which seeks to refine future projections by considering the ability of climate models to accurately simulate present-day temperature variability.

This study employs several key metrics and techniques to analyze temperature extremes and their future projections. The primary index used is the number of days above the 98th percentile (TX98p) of daily maximum temperature (TX) during the warmest season (June-August in the Northern Hemisphere, December-February in the Southern Hemisphere, and year-round in the tropics). This metric indicates the number of extremely hot days relative to a 1995-2005 climatology. To evaluate model accuracy, a metric, ATX, quantifies the historical variability of TX, representing the difference between the mean TX and the 95th percentile of TX. This metric focuses solely on TX variability, disregarding any bias in the mean TX. The study uses ERA5 reanalysis as a global reference dataset. The relationship between ATX and changes in TX98p under different warming scenarios is investigated. The emergent constraint (EC) is applied only where the ATX-TX98p correlation is significant, primarily in tropical and subtropical areas and South and East Asia. To account for observational and model variability uncertainties, the study uses the internal variability of the HAPPI ensemble, forced with observed sea surface temperatures, as an estimate of model internal variability. The difference between ERA5 and the CHIRTS satellite dataset provides an estimate of observational error. Models falling within the combined uncertainty range of ERA5 and HAPPI are considered to realistically represent ATX and selected for future change simulations. Both regional and global constraints are applied and compared. Sensitivity tests are conducted by randomly removing one model from the ensemble to assess robustness. Finally, the physical mechanism linking changes in TX98p and land drying is evaluated through analysis of latent and sensible heat fluxes.

The study reveals a significant negative correlation between the historical variability metric ATX and the projected change in TX98p. Regions with overestimated variance of hot days in present-day models show smaller future changes in TX98p, justifying the use of ATX as an emergent constraint. Applying the EC in tropical, subtropical regions, and South and East Asia leads to projections of larger changes in TX98p than those obtained using unconstrained ensembles. In these regions, the increase in the number of extremely hot days could be up to 50% larger than previously estimated, with potentially twice as many hot days as unconstrained predictions. Models that accurately represent ATX (and thus, humidity feedbacks) tend to project faster warming of hot extremes. These findings remain consistent across different warming targets (+1.5°C, +2°C, and end-of-century). The constrained TX98p signal suggests that the level of increase in the frequency of hot days previously estimated to occur by the end of the century might be reached by 2060. Both regional and global constraints yield similar results, reinforcing the key findings. Analysis of latent and sensible heat fluxes supports the link between land surface humidity and the frequency of hot extremes, with larger temperature variability occurring under drier conditions (weaker latent heat fluxes). Mid-latitude regions show weaker correlations, suggesting other dominant processes beyond soil drying. The study performs sensitivity tests to check for bias, showing the findings are robust.

This study's findings highlight a significant underestimate in the projected frequency of unusually hot days in low-latitude regions due to biases in climate models. The emergent constraint, based on the accurate representation of land-atmosphere humidity exchange, strengthens the results, particularly in tropical regions and South-East Asia. This emphasizes the role of land surface humidity processes in driving changes in hot extremes. While the findings are less clear in other regions, suggesting the dominance of other processes, the consistency of results using regional and global constraints supports the overall conclusion. The study acknowledges its focus on dry-bulb temperature and does not consider humidity's effect on heat stress, a crucial factor in tropical regions. Future work could investigate the wet-bulb temperature signal using a similar methodology.

This research demonstrates that climate models exhibit biases in simulating the difference between hot and average days, leading to underestimations of future hot extremes. The emergent constraint applied significantly improves the projection of hot extremes in tropical and subtropical areas. The findings emphasize the importance of accurately representing land-atmosphere interactions, particularly humidity feedbacks, in climate models to obtain reliable predictions of future temperature extremes. Future research should expand on this work by incorporating humidity into the heat stress metrics and by investigating additional emergent constraints related to other climate variables.

The study acknowledges some limitations. The primary constraint used relates to surface temperature variability and does not explicitly account for humidity's role in heat stress, particularly in tropical regions. Additionally, the robustness of the emergent constraint's signal in certain regions is contingent upon the number of models deemed realistic, potentially limiting generalizability. While the study utilizes sensitivity analysis to address these points, further research focusing on humidity and broader model selection strategies is warranted.