Soil heat extremes can outpace air temperature extremes

Extreme heat events cause substantial societal and ecological impacts and have been increasing in both intensity and frequency globally and regionally. Yet, the processes governing heat extremes—particularly land–atmosphere coupling—remain incompletely understood, leading to uncertainty in projections. Soil temperature directly links soil moisture and near-surface air temperature and strongly influences hydrology and the terrestrial carbon cycle, but its direct role in feedbacks has received less attention than soil moisture alone. This study investigates how soil hot extremes over Europe have evolved in recent decades relative to air hot extremes, using in situ observations, remote sensing, and reanalysis, and explores the role of soil temperature in intensifying hot spells via the soil moisture–temperature feedback under future warming.

Prior work highlights that land conditions can intensify and propagate heatwaves through diabatic heating, especially under persistent high-pressure systems, where soil moisture deficits reduce evaporation, increase sensible heat flux, and warm the lower atmosphere. Observations in China and Germany have reported higher soil than air warming rates and differing trends in maximum soil versus air temperatures. The coupling strength between air and soil temperatures depends on land cover, aerodynamic conductance, soil water content, and soil properties, which vary spatially. Climate models represent land–atmosphere coupling with varying fidelity, contributing to uncertainty in projections of heat extremes.

Data sources: Subdaily air and soil temperatures (<10 cm depth) from FLUXNET2015, ICOS, ARPAV, DWD, and Météo France were aggregated to daily maxima (days with any missing subdaily data were set to missing). Satellite skin temperature (CM SAF Meteosat, 1991–2015) and daily gridded air temperatures (E-OBS) were used alongside ERA5Land hourly reanalysis (2 m air temperature; 0–7 cm soil temperature and moisture; 1996–2021). Additional ICOS variables (radiation, fluxes, soil water content) were used to assess flux trends during the hottest week at selected stations. Extreme indices: Two indices were computed. (1) TX7d: mean of daily maximum temperatures during the hottest week per year (intensity metric). (2) TX90p: percentage of hot days per summer month, where hot days are those with daily maximum temperature above the calendar-day 90th percentile (5-day centered window) for a base period (first 10 years of each series; 1996–2005 for gridded products). Annual TX90p was the average over JJA. Quality control and coverage: TX7d was computed for years with no more than 20 consecutive missing days between Apr 1–Sep 30; sensitivity tests with stricter criteria yielded similar conclusions. TX90p months with <10 missing days were used (similar results with <5). Trends were assessed only where time series had >10 years of data. Trend detection used the Mann–Kendall test (significance at 90%) and Sen’s slope estimator. Stations included in trend analyses: TX7d at 118 stations (total 160 air–soil pairs across datasets: 6 FLUXNET, 11 ICOS, 14 ARPAV, 40 DWD, 47 Météo France). TX90p at 103 stations (154 pairs: 11 ICOS, 14 ARPAV, 49 DWD, 29 Météo France). Spatial station density is highest in Germany, France, and Italy. ERA5Land analyses were repeated for 1970–2021 to test period sensitivity. Comparisons: Differences between soil and air trends were summarized as (a) Abs. difference: absolute soil minus absolute air slopes (where positive values indicate faster change in soil extremes), and (b) Incr. difference: soil minus air slopes conditioned on both being positive (where positive values indicate faster increase in soil extremes). Future projections: Daily maximum air temperature (2 m) and 6-hourly soil temperature (5 cm) from five CMIP6 models (MIROC-ES2L, MIROC6, MPI-ESM1.2-LR, MPI-ESM1.2-HR/MR, EC-Earth3) were used for historical and SSP5-8.5 simulations (first realizations). Daily maximum soil temperatures were derived from subdaily outputs. Hot days were identified via TX90p based on air temperatures (JJA). For each model and grid cell, the percentage of hot days with Tsoil(max) > Tair(max) was computed, representing days when soil releases sensible heat to the atmosphere during hot spells. To avoid differing climate sensitivities, analyses were framed at warming levels of 1.5, 2.0, and 3.0 °C (estimated per model). Multimodel means were mapped after interpolating to the coarsest grid (MIROC-ES2L). A similar analysis using daily mean temperatures provided consistent results with higher percentages.

• Observations (1996–2021) show positive trends in both air and soil TX7d across Central Europe, with greater spatial variability in soil. Across stations, 66% (absolute trends) and 65% (only positive trends) indicate soil hot extremes changing/increasing faster than air.
• Quantitatively, station data indicate soil hot extremes are intensifying 0.7 °C per decade faster than air hot extremes on average over Central Europe and are increasing in frequency at roughly double the rate of air extremes (TX90p).
• Remote sensing (CM SAF + E-OBS) corroborates faster increases in soil-related hot extremes in many regions; over forests, satellite skin temperature reflects canopy rather than topsoil, sometimes showing the opposite signal.
• ERA5Land shows broadly similar spatial patterns of TX7d trends in air and soil, with regions (eastern Germany, western Poland, central-eastern Europe) where soil TX7d increases faster than air by >0.5 °C per decade. Extending ERA5Land to 1970–2021 yields similar conclusions.
• Frequency trends (TX90p): A larger fraction of sites/areas show soil frequency trends exceeding air (e.g., in situ Soil > Air: ~77–78% for absolute/positive-trend comparisons).
• Mechanistic evidence: During the hottest week per year, stations like DE-Tha exhibit constrained soil moisture and latent heat flux, with increasing sensible and ground heat fluxes, and rising TX7d in both air and soil—consistent with a soil moisture–temperature feedback. ERA5Land shows strong negative correlations between soil moisture and soil temperature and strong positive relationships between soil and air temperatures during hot extremes.
• Future projections (CMIP6, SSP5-8.5): The percentage of hot days when maximum soil temperature exceeds maximum air temperature increases over the 21st century, especially in Mediterranean and central-eastern Europe. Under higher warming levels, central/eastern Europe sees larger increases; differences between 3.0 °C and 1.5 °C warming exceed 8% of hot days regionally. By late century, in eastern Europe all five models project >10% greater occurrence than in 1990; in western Europe, four of five models project >10% increases. Using daily mean temperatures indicates even larger and more consistent increases, especially in eastern Europe.

The analysis demonstrates that soil hot extremes have intensified and become more frequent faster than air hot extremes over Europe, particularly in Central Europe. This discrepancy is consistent with the soil moisture–temperature feedback: under dry, warm conditions, reduced soil moisture limits latent cooling, so more net radiation warms the soil, which then releases sensible heat, raising near-surface air temperatures and increasing vapour pressure deficit and evaporative demand—further drying soils and reinforcing heat. Spatial heterogeneity in land cover, rooting depth, soil texture, and hydrology modulates air–soil thermal coupling, explaining regional differences. Remote sensing and reanalysis complement in situ evidence, though canopy effects over forests and model structures affect magnitudes. Projections indicate that as warming proceeds, the probability of days when soil reinforces hot spells rises, enhancing the potential severity and persistence of heatwaves, especially in central-eastern Europe. These findings imply that relying solely on air temperature underestimates risks to hydrology, agriculture, and biogeochemical processes that are more sensitive to soil temperature extremes.

This study provides multi-source evidence that soil heat extremes are outpacing air temperature extremes in intensity and frequency across Europe, with an average additional intensification of about 0.7 °C per decade for soil extremes and roughly doubled frequency increases over Central Europe. It identifies soil temperature as a key element in the soil moisture–temperature feedback that can intensify and propagate hot spells. Projections suggest a growing share of hot days during which soils are warmer than air, implying stronger land–atmosphere coupling and greater heat risk under higher warming levels. Implications include the need to incorporate maximum soil temperatures into impact and risk assessments for agriculture and ecosystem functioning, and to improve representation of soil thermal and moisture processes in climate models. Future work should disentangle contributions from soil water loss, land cover change, and land management, expand observational networks within soils, and better characterize depth-dependent soil thermal dynamics.

• Observational coverage is uneven across Europe; station-based spatial averages are dominated by regions with higher station density (Germany, France, Italy).
• Flux and ancillary measurements are limited, restricting detailed process analyses to a few ICOS sites (e.g., DE-Tha).
• Satellite skin temperatures over forests represent canopy rather than topsoil, complicating soil temperature inference.
• ERA5Land soil temperatures derive from a model, with associated structural uncertainties.
• Soil depth effects are difficult to generalize due to limited and heterogeneous depth-specific observations; no clear depth–trend relationship was found.
• Trend detection uses a 90% significance threshold; missing data handling, while tested, may still affect local estimates.
• Future analysis includes only five CMIP6 models providing subdaily soil temperatures; models differ substantially in land/soil process representation, affecting magnitude though not the direction of projected changes.
• Definitions of hot extremes (TX7d, TX90p) and base periods can influence numerical values, though sensitivity tests indicate conclusions are robust.