Anthropogenically-driven increases in the risks of summertime compound hot extremes

The study addresses how compound summertime hot extremes—events combining hot days and hot nights within the same 24 hours—have changed historically, what physical mechanisms drive these changes, how much is attributable to anthropogenic forcings, and how such events will evolve and affect populations in the future. While individual hot days and hot nights are well studied, combined daytime–nighttime extremes are less understood despite being more detrimental to human health and natural systems due to reduced nighttime recovery. The authors propose a bivariate classification of hot extremes (independent hot days, independent hot nights, compound hot extremes) to reassess observations, perform detection–attribution analyses, and generate constrained projections, thereby improving risk assessment and informing adaptation and mitigation strategies focused on compound heat.

Prior work documents increasing frequency and intensity of hot extremes globally with strong anthropogenic contributions. Studies have often used daily mean temperatures or univariate daytime/nighttime metrics, which may overlook compounding effects that exacerbate health and ecological impacts. Literature indicates meteorological differences between compound and single-component extremes and emphasizes their heightened impacts on human health, agriculture, and ecosystems. Suggested drivers include circulation anomalies (anticyclones), land–atmosphere coupling (especially soil moisture deficits), and general warming. Detection–attribution frameworks and CMIP5-based projections are established tools, while compound event risk frameworks highlight the need to analyze co-occurring extremes. This study builds on these strands by explicitly defining compound hot extremes and quantifying their drivers and future risks.

Data: Daily Tmax and Tmin observations from HadGHCND (3.75° × 2.5° grids) for Northern Hemisphere land (1960–2012), with sensitivity tests using Berkeley Earth daily Tmax/Tmin regridded to the same grid and mask. Climate model outputs: CMIP5 historical and RCP4.5/RCP8.5 projections from models with at least three ensemble members with daily Tmax/Tmin. Model data were bilinearly regridded to 3.75° × 2.5° and masked by observation availability. Population projections used spatially explicit SSP scenarios at 1/8° resolution aggregated to climate grids. Definitions: Summer is June–August. For each calendar day, hot day/night thresholds are the historical (1960–2012) 90th percentiles computed from 15-day windows (±7 days) across years, capturing intra-seasonal acclimatization. Three nonoverlapping types are defined: compound hot extreme (hot day followed by hot night within 24 h), independent hot day (hot day without a following hot night), independent hot night (hot night without a preceding hot day). Frequency is the number of qualifying days; intensity is the exceedance above thresholds (°C), emphasizing heat above high backgrounds. Area-weighted hemispheric means were computed. Trend estimation: Trends in frequency (days decade−1) and intensity (°C decade−1) computed using Theil–Sen estimator with 90% confidence intervals; Mann–Kendall test (α = 0.05) for grid-scale significance. Relative changes (% decade−1) also computed vs. 1961–1990 means. Attribution of mean vs variability: General warming signals estimated by fitting a second-order polynomial to summer-mean Tmax/Tmin (1960–2012) at each grid, then removed from daily series. Recomputed trends on residuals represent effects of changing variability (interannual, seasonal cycle, intraseasonal, diurnal). Contributions of mean shift vs variability inferred from differences, with 5–95% uncertainty via resampling grids (100,000 draws). Parameter contributions of location/scale/shape of temperature distributions were also examined. Physical drivers: Examined links between changes in compound extremes and (i) circulation trends via sea-level pressure and 500 hPa geopotential height trends; and (ii) land–atmosphere coupling via correlations between detrended precipitation and Tmax/Tmin (negative correlations indicating stronger soil moisture–temperature coupling). Detection and attribution: Optimal fingerprinting with observed changes Y modeled as sum of scaled fingerprints X plus internal variability ε. Fingerprints constructed from MME mean responses to external forcings: ALL, NAT; ANT = ALL − NAT; three-signal analysis separates GHG and other anthropogenic (OANT, dominated by aerosols/land-use) from NAT. Series preprocessed as nonoverlapping 3-year means (18 samples, 1960–2012). Internal variability covariance estimated from preindustrial control runs using regularization; residual consistency tests applied. Scaling factors’ 5–95% ranges obtained using independent control chunks. Attributable trends computed as scaled MME trends with uncertainty propagation. Constrained projections: Historical (1960–2012) anomalies reconstructed by summing optimally scaled MME responses to GHG, OANT, and NAT; post-2012 projections (RCP4.5, RCP8.5) scaled by ANT scaling factor. Series adjusted to match observed 1960–2012 mean. Global warming levels (1.5, 2, 4 °C) determined from decadal-average MME global mean surface air temperature anomalies relative to 1861–1890. Population exposure: Exposure per decade computed as the spatial average of grid-level products of decadal-average event frequency (from raw, unconstrained projections) and total population (from SSPs) at that grid. Two integrated scenarios presented: RCP4.5–SSP1 and RCP8.5–SSP3.

• Historical increases: Northern Hemisphere (NH) land average frequency and intensity of summertime compound hot extremes increased significantly over 1960–2012 by 1.03 days decade−1 (90% CI: 0.82–1.26) and 0.28 °C decade−1 (90% CI: 0.23–0.33), respectively.
• Dominant driver: Increases are primarily due to summer-mean warming (location shift), with changing temperature variability playing a secondary role. This is consistent in observations and models.
• Spatial patterns: Largest frequency increases observed in the southern United States, Northwest/Southeast Canada, Western/Southern Europe, Mongolia, and Southeast China; strongest intensification in the Southwest US, Northern/Southeast Canada, and large parts of Eurasia.
• Physical mechanisms: Regions with stronger nocturnal land–atmosphere coupling (negative correlation between Tmin and precipitation) exhibit larger frequency increases in compound extremes (significant relationship). Increases in anticyclonic conditions are noted but their dynamical contribution to compound extreme increases is weaker than theoretically expected after accounting for general warming and potential reanalysis biases; joint effects with nocturnal coupling likely enhance daytime–nighttime coupling in hotspots.
• Detection and attribution: External forcings are detected in observed changes. Anthropogenic GHGs dominate the attributable increases. Best estimates of GHG-attributable trends (1960–2012): frequency +1.18 days decade−1 (5–95% UR: 0.96–1.41) and intensity +0.28 °C decade−1 (0.22–0.34). OANT exerts a cooling offset: frequency −0.09 days decade−1 (−0.20 to 0.03) and intensity −0.02 °C decade−1 (−0.04 to −0.01). NAT contributions are detected but much smaller than GHGs. Models slightly underestimate GHG responses for compound extremes (scaling factors slightly >1).
• Projections (constrained): Compound hot extremes show the largest future increases among types. Relative to 2012, under RCP4.5 the NH-average frequency of compound extremes increases about fourfold by 2100 (from 8.3 to 32.0 days per summer). Under RCP8.5, about three-quarters of summer days (~69 days) become compound extremes before 2100, an >8-fold increase; intensity increases roughly threefold. The compound type becomes the dominant hot extreme type across NH lands after ~2030.
• Relation to global warming levels: Compared to a 1.5 °C world, 2 °C warming adds ~5 days of compound extremes and ~0.5 °C in intensity on average over NH lands. Non-mitigated 4–6 °C warming adds ~40–60 days and ~4–6 °C in intensity relative to 1.5 °C. Intensity increases quasi-linearly with GMST.
• Population exposure: NH population exposure to compound hot extremes is projected to increase from 19.5 billion person-days (2010s) to 74.0 (2090s) under RCP4.5–SSP1 (~4×), and to 172.2 under RCP8.5–SSP3 (>8×). Spatial hotspots include eastern US, western Europe, western Asia, and eastern China. After 2030, compound extremes dominate population exposure among hot extreme types.
• Model calibration benefit: Applying scaling factors improves agreement between simulated and observed historical changes and yields larger constrained projections for compound events by century’s end under RCP8.5 (frequency ~13% and intensity ~8% higher than uncalibrated MME means).

By explicitly defining and analyzing compound hot extremes, the study demonstrates that the observed and projected increases in these events are driven predominantly by anthropogenic warming, with greenhouse gases as the principal forcing. This addresses the research gap on combined daytime–nighttime extremes and clarifies that general warming, rather than variability changes or circulation alone, governs long-term trends at hemispheric scales. Mechanistic analyses indicate that nocturnal land–atmosphere coupling and, to a lesser degree, increased anticyclonic conditions help explain regional heterogeneity and the stronger increases in compound events compared to decoupled hot days or nights. The constrained projections and exposure estimates reveal that without strong mitigation, compound heat will become the most common and impactful form of summertime heat hazard across NH lands, substantially escalating risks to health, agriculture, ecosystems, and infrastructure. These findings underscore the urgency of mitigation to limit global warming and targeted adaptation strategies that account for compounded daytime–nighttime heat, especially in urban and densely populated regions.

The study provides the first comprehensive detection, attribution, and projection of summertime compound hot extremes across Northern Hemisphere lands using a bivariate classification. It shows robust historical increases in frequency and intensity primarily driven by general warming attributable to anthropogenic GHGs. Observation-calibrated projections indicate that compound extremes will dominate summertime heat hazards this century, with large increases in both hazard and population exposure, especially under high-emission scenarios. The work advances risk assessment by quantifying compound events and highlights the need for policy and planning to prioritize compound heat risk. Future research should extend analyses to the Southern Hemisphere as data quality improves, refine regional projections by better representing land–atmosphere coupling and synoptic dynamics, integrate vulnerability to translate exposure into risk, and reduce model spread through improved process understanding and observational constraints.

• Spatial coverage: Limited high-quality daily observations over much of the Southern Hemisphere preclude quasi-global analysis; even with Berkeley Earth, daily data quality/availability vary regionally and temporally.
• Data quality sensitivity: Detection–attribution–projection results can be sensitive to observational homogenization and reanalysis biases, especially at regional scales.
• Model/process representation: Models fail to reproduce some regional trends (e.g., reductions in independent hot days in southern Canada and central-eastern China), potentially due to misrepresentation of irrigation, land use, and aerosols, and smoothing of internal variability by MME averaging.
• Attribution uncertainties: Dynamical contributions from anticyclonic changes are weaker than theory suggests after accounting for warming; reanalysis and methodological uncertainties remain.
• Projection spread: Increasing intermember/intermodel spread for compound extremes, linked to divergent precipitation projections and land–atmosphere feedback uncertainties, especially regionally.
• Constrained projections: Calibration assumes current biases persist into the future and cannot capture unforeseen future-specific errors (e.g., abrupt circulation changes). Grid-scale constrained projections are not performed, necessitating use of raw hazards for exposure estimates.
• Exposure estimates: Likely lower bounds due to using unconstrained hazard projections and incomplete land coverage (e.g., exclusion of some populous regions like India in the analysis mask).