Advective, adiabatic and diabatic contributions to heat extremes simulated with the Community Earth System Model version 2

Abstract. Do heat extremes in climate model simulations form for the right physical reasons? Addressing this question is essential to further our confidence in heat extreme projections, to pinpoint regionally varying model biases, and to enable model improvements regarding heat extremes. Here, we perform a detailed process-based evaluation of CMIP6-type simulations with the Community Earth System Model version 2 (CESM2) regarding heat extremes and employ a Lagrangian approach to quantify advective, adiabatic, and diabatic contributions to near-surface temperature anomalies (T′) during heat extremes. Heat extremes are identified at each grid point and year as the day with the largest daily mean two-meter temperature (hereafter termed TX1day events). Comparison of CESM2 results with results of an analogous analysis in ERA5 reanalyses for the time period 1980–2020 reveals that, qualitatively and considering continental-scale variations, near-surface T′ during TX1day events in CESM2 form in a physically similar way as in ERA5: Advective contributions dominate in storm track regions, diabatic contributions dominate over tropical and subtropical land regions, and adiabatic warming contributes significantly to heat extremes over subtropical oceans and extratropical land regions. However, quantitatively and at regional scales, there are considerable differences: CESM2 overestimates the magnitude of near-surface T′ during TX1day events in numerous regions (in the global average the TX1day T′ magnitudes are 3.70 K and 3.21 K in CESM2 and ERA5, respectively). These differences are related to larger advective contributions to TX1day events in CESM2 compared to ERA5. That is, biases in the magnitude of simulated TX1day events appear to be related to circulation differences associated with TX1day events in CESM2 as opposed to ERA5. Furthermore, over land, CESM2 systematically overestimates the diabatic contribution to near-surface T′ during TX1day events (4.61 K in CESM2 vs. 2.48 K in ERA5), and underestimates the adiabatic contribution (1.69 K in CESM2 vs. 3.45 K in ERA5). Biases in these contributions to TX1day T′ are much larger than the biases in the TX1day T′ magnitude, and, consequently, the magnitude of CESM2 TX1day events is often “right for the partly wrong physical reasons”. Composite analyses for TX1day events in selected regions suggest that biases in the land-atmosphere coupling, in particular an erroneous partitioning between sensible and latent heat fluxes, is partly responsible for the overestimation of diabatic contributions in CESM2. We argue that such a detailed quantitative understanding of the differences in the physical processes behind simulated and observed heat extremes is highly relevant for assessing and improving the robustness of heat extreme projections. Our results thus call for analogous investigations with other state-of-the-art models.