Introduction

Heatwaves (HWs) are broadly defined as prolonged periods of excessively hot weather during which daytime (daily maximum) and/or nighttime (daily minimum) temperatures persistently exceed a locally defined threshold due to anomalous anticyclonic circulations, synoptic conditions, land-atmosphere interactions, and/or soil moisture deficits1,2,3. Due to the global increase in temperature1,4,5, HW events have increased significantly in recent decades6,7,8, leading to increasing research focus on understanding the variability and trends of HW characteristics as well as the underlying mechanisms1,2,8,9,10,11,12,13,14. HWs accompanied by high humidity pose significant risks to human health by modulating human thermal stress and exacerbating heat-related illnesses4,15,16, especially among vulnerable populations such as the elderly, infants17,18,19,20, and people with underlying health conditions21. Additionally, HWs can physically damage energy infrastructure by placing further strain on the system through increased cooling demand15,22 and adversely affecting ecosystems and agricultural productivity6,23,24. The complex, multi-dimensional nature of HWs and their driving mechanisms involve interactions across multiple spatiotemporal scales2. Therefore, understanding the processes that trigger HWs, amplify their severity, and extend their duration is crucial, particularly under both natural and anthropogenic climatic influences.

Previous studies have investigated HW characteristics over a wide range of spatiotemporal scales, including event-specific, regional, and global levels1,4,6,7,8,9,10,11,12,13,14,16,18,22. For instance, the HW in western Europe during the summer of 2003 made it the hottest summer in 500 years25, resulting in approximately 40,000 mortalities and US$10 billion in economic losses10. Similarly, the HW in eastern Europe and Russia in 2010 caused 55,000 fatalities due to a record-breaking temperature increase of about 8.4 °C, which persisted for several weeks from mid-July to August, leading to US$15 billion in agricultural losses25,26,27. Both the 2003 and 2010 HWs were attributed to persistent blocking anticyclones centered over Russia and western Europe, which resulted in reduced cloudiness and warm air advection coupled with feedback from land-atmosphere interactions triggered by soil moisture deficits10,25,27. Additionally, during the HW in western Europe in 2019, daily maximum temperatures in northeastern France and the Benelux region exceeded climatological norms by nearly 20 °C12. This anomaly was even more pronounced than the maximum temperature deviations recorded in the same regions during the summer of 2003, which were approximately 16 °C above normal values.

More recently, studies from North America, Asia, and Africa have documented the increasing impact of HW events. For instance, Weinberger et al.28 attributed 0.44% of all mortalities in the United States to extreme heat, corresponding to about 5600 deaths annually. Similarly, Lucarin et al.14 reported over 1000 attributable deaths from the HW in western North America in the summer of 2021, alongside damages amounting to several billion US dollars. In Australia, Coates et al.29 estimated at least 473 heat-related deaths, coupled with multibillion-dollar economic losses. In sub-Saharan Africa, the availability of comprehensive data on heat-related mortality and economic losses remains limited due to low-quality datasets30. Although efforts to enhance meteorological data quality have been made31, studies continue to reveal rising health risks associated with extreme heat, particularly in urban centers18. Nevertheless, projections suggest that heat-related mortality in Nigeria, for example, could increase significantly, rising from 9000 to 23,000 annual deaths (and potentially up to 43,000 based on extreme scenarios) by the end of the 21st century32. Additionally, southern Africa has reported substantial economic losses linked to extreme HWs and drought33.

Studies have revealed common underlying mechanisms driving HWs, which can be broadly categorized into large-scale circulation systems and land-atmosphere interactions. Large-scale circulation systems are predominantly characterized by high-pressure features resulting from persistent anomalous anticyclonic conditions that create persistent blocking patterns2,10,11,12,14,33,34. For example, García-Herrera et al.10 attributed the 2003 European HW to consecutive episodes of intense anticyclonic anomalies with strong meridional airflow components. Similarly, Lucarin et al.14 reported that the 2021 western North America HW was caused by an intense and persistent omega block, a large-scale mid-latitude weather pattern characterized by warm air advection, subsidence, and anomalous clear-sky insolation35. In South Africa, record-breaking HWs during the 2015/2016 austral summer were initiated by a high-pressure system over Namibia and Botswana, with further intensification attributed to the 2015/2016 ENSO events. Additionally, Tedesco et al.34 and Bonne et al.11 identified an anticyclonic ridge as the trigger for the 2012 HW over Greenland. These anticyclonic episodes establish persistent blocking patterns that trap warm air near the surface, amplifying HW intensity and duration. On the other hand, the land-atmosphere feedback mechanism arises from surface radiation flux imbalance due to depleted soil moisture, lack of precipitation, elevated heat loads, terrain characteristics, or alteration of local surface albedo through human activities, such as land cover changes, urbanization, and other land surface–altering activities2,12,13,33,36,37,38. These factors often suppress evaporation, thereby increasing the available energy, which is primarily redirected to enhance sensible heat flux2. This process creates a positive feedback loop where rising air temperatures further dry the soil, amplifying heat stress2,12. This phenomenon is particularly pronounced in transitional zones—regions between dry and wet climates—where soil moisture is naturally limited2,36. Depleted soil moisture significantly increases temperature variability and intensifies HW intensity and duration12 due to persistent dry spells, reduced precipitation, or antecedent drought conditions39. Other factors, such as vegetation and climate change, also intensify thermal stress by altering land surface properties15. Moreover, studies have highlighted distinctions between the mechanisms driving daytime and nighttime HWs. Thomas et al.23 found that daytime HWs are triggered by dry conditions, reduced cloud cover, and increased sensible heat flux. In contrast, enhanced nighttime HWs are often associated with large-scale circulation-induced dry conditions and warm air advection12,40. Notably, the mechanisms driving nighttime HWs tend to be more region-specific, whereas daytime HW mechanisms are often widespread23.

The physical understanding of observed changes in extreme climate over Africa remains limited due to sparse observational data, strong climate variability, challenges in capturing local-scale processes, limited regional climate modeling efforts, and inadequate computing infrastructure41,42,43,44. Considering the increasing impact of climate change on extreme weather events, particularly on HW events due to human-enhanced global warming, substantial evidence highlights the human imprint on recent HW occurrences. Factors such as greenhouse gas emissions, land-use changes, and urbanization have intensified and/or prolonged HW events in recent decades4,6,9,13,15,24,45,46,47.

Unlike other regions of the world, across Africa a comprehensive understanding of HW characteristics and underlying mechanisms remains limited. Existing studies have focused primarily on HW trends8,48,49, hotspots18,40, and impacts33,46,50. However, only a few studies have examined HW mechanisms in specific subregions, such as southern Africa33,51 and West Africa52. A comprehensive investigation of HW mechanisms across the entire African continent is lacking. Furthermore, many studies rely on traditional timescales, such as analyzing HWs over fixed three- or five-month summer periods in the Northern or Southern Hemisphere. While this approach is common, it has significant limitations for Africa. The climatologically warmest months can vary significantly across grid points within both hemispheres, as different regions within the same hemisphere receive maximum solar radiation at different times of the year. Consequently, these areas experience peak warming during varying weeks or months, underscoring the need for more adaptive and regionally representative methodologies, with consequences for accurately representing the underlying mechanisms responsible for HW events2. Therefore, this study aims to address the inadequacy of traditional timescales and investigate the mechanisms that intensify HWs across Africa. We propose an innovative methodology for identifying timescales that reflect periods of maximum temperature at each grid point across the continent (see Fig. 1, and methods). Our results reveal two distinct periods over the entire African continent: an initial period of no significant change/trend in HWs and a recent period of increasing trends across Africa. We further investigate the mechanisms driving the HWs during these periods with a focus on the possible roles of natural internal variability and anthropogenic influences.

Fig. 1: Grid-based season selection for capturing the hottest local months.
figure 1

Comparison of (a) the traditional boreal/austral summer (fixed seasonal approach) with (b) the grid-based climatologically hottest three months using monthly CRU near-surface temperature from 1965 to 2014.

Full size image

Results

Trends in heatwave characteristics during the historical and recent periods

While greenhouse gas concentrations have risen steadily—particularly since the 1950s—global aerosol emissions peaked around the 1970s and have subsequently declined, especially in developed regions, due to the implementation of air quality regulations53,54,55,56. These opposing trends, which are projected to persist throughout the 21st century, are expected to significantly modulate HW characteristics by enhancing warming through the attenuation of aerosol-induced cooling. To evaluate whether similar trends are evident over Africa, we used output from the Community Earth System Model version 2 Large Ensemble project (CESM2-LE). This large ensemble was specifically designed to distinguish externally forced responses from internal variability57, making it particularly well suited for studying regional climate variability and extremes, including HWs58,59. To detect shifts in key climate drivers, we applied the Pettitt test—a non-parametric method that detects statistically significant shifts in time series without assuming a specific distribution—to the weighted area-averaged annual ensemble mean of CESM2-LE all-forcing (hereafter “ALLF”) of sulfate aerosol burden (SO₄), black carbon burden, CO₂ burden, and near-surface temperature (see “Methods”)60,61. Based on ALLF time series, our results reveal statistically significant change points occurring in 1977 for SO₄, 1982 for both black carbon and CO₂ burdens, and 1985 for near-surface temperature (Fig. S1; similar variation is evident in ERA5 near-surface temperature time series, but figure not shown). Based on these breakpoints, we partitioned subsequent analysis into two distinct periods: a “historical period” (1950–1979), characterized by a relatively stable near-surface temperature trend resulting from compensatory effects—cooling from increasing SO₄ burden and slight warming from rising black carbon and CO₂ concentrations; and a “recent period” (1985–2014), marked by a pronounced warming signal as SO₄ levels stabilized while black carbon and CO₂ burdens continued to increase.

To assess the fidelity of simulated HWs, we compared ALLF with the fifth-generation ECMWF reanalysis (ERA5), which is widely used for studying HWs across Africa40,49,62,63,64. We first assessed the spatial trends of near-surface maximum (TASmax) and minimum (TASmin) temperatures during the historical and recent periods (Fig. 2). During the historical period, ERA5 exhibits a statistically significant and pronounced warming trend in both TASmin and TASmax over eastern Africa and parts of the Sahel and Sahara (Fig. 2a, c). In contrast, ALLF fails to capture this pattern, instead exhibiting a spatially uniform, weak, and statistically non-robust temperature trend (Fig. 2b, d). This discrepancy likely results from ensemble averaging, where divergent regional trends across individual members—ranging from warming to cooling—are smoothed out (Figs. S2 and S3). To enhance confidence in our findings, we emphasize results only at grid points where at least 70% of ensemble members agree on the sign of the trend in the ensemble mean. This threshold ensures that the identified signals reflect a high degree of ensemble consensus, thereby reducing the influence of outlier members and partially mitigating the averaging effect inherent in the ensemble mean65,66. In contrast to the historical period, both ERA5 and ALLF show widespread warming across much of Africa during the recent period (Fig. 2e–h). Although some localized cooling is evident in ERA5 and in some individual all-forcing ensemble members (Figs. S4 and S5), ALLF exhibits a consistent and robust warming signal over the continent.

Fig. 2: Spatial trend of near-surface maximum and minimum temperature in ERA5 and CESM2-LE all-forcing ensemble mean (ALLF).
figure 2

The trend in the near-surface minimum (a,b,e,f) and maximum (c,d,g,h) temperature during the historical (1950–1979; a–d) and recent period (1985–2014; e–h) for ERA5 (a,c,e,g) and ALLF (b,d,f,h). Stippling indicates grid points where the trend in ERA5 is statistically significant (p values < 0.05), and hatching indicates grid points where at least 70% of the ensemble members agree on the sign of the trend in the ensemble mean.

Full size image

Having assessed trends in minimum and maximum temperatures during the historical and recent periods, we now investigate changes in HW characteristics to better understand the evolving nature of extreme heat across Africa (Fig. 3, Fig. S6–S7). During the historical period (1950–1979), trends in compound, daytime, and nighttime HW metrics—including frequency, duration, and intensity—are generally weak and statistically significant only over a few grid points, with the most notable signals observed for daytime HWs in southern Africa in ERA5 (Figs. 3a1–3; S6a1–3 and S7a1–3). ALLF similarly exhibits weak and non-robust trends, as fewer than 70% of ensemble members agree on the sign of the ALLF trend for any HW metric or type (Figs. 3b1–3; S6b1–3 and S7b1–3). This lack of a robust signal in ALLF largely reflects the competing influences of aerosol-induced cooling and modest greenhouse gas–driven warming. To disentangle the effects of different external drivers, we analyze simulations from the Single-Forcing Large Ensemble Project within CESM2-LE, which isolate the influence of individual forcings—such as aerosols, greenhouse gases, biomass burning, and everything else—by holding other forcings fixed at preindustrial levels67. The aerosol forcing ensemble mean (AER) exhibits a robust and spatially coherent negative trend in compound HWs across much of Africa, with area-averaged values of –0.08 HWs season−1 decade−1 (frequency, Fig. 3c1), –0.2 days HW−1 decade−1 (duration, Fig. S6c1), and –0.53 °C HW−1 decade−1 (intensity, Fig. S7c1), highlighting the cooling influence of aerosols on extreme heat characteristics. Daytime and nighttime HWs under AER also show negative trends, although with greater spatial heterogeneity. Northern Africa exhibits weak or non-robust changes, while robust pronounced declines in nighttime HW metrics appear over the Congo Basin, with intensity trends reaching −0.7 °C HW−1 decade−1 (Figs. 3c2, c3; S6c2, c3 and S7c2, c3).

Fig. 3: Historical and recent trends in HW frequency for ERA5 and CESM2 forcings.
figure 3

The trends in compound (a1–f1, a4–f4), daytime (a2–f2, a5–f5), and nighttime (a3–f3, a6–f6) HW frequency ((HWs season−1) decade−1) for the historical (1950–1979) and recent period (1985–2014) for ERA5 (a1–6), all-forcing (b1–6; ALLF), aerosol forcing (c1–6; AER), greenhouse gas forcing (d1–6; GHG), biomass burning (e1–6; BMB), and everything else (f1–6; EE). Stippling in (a1–6) indicates grid points where the trend in ERA5 is statistically significant (p values < 0.05), and hatching in (b1–f6) indicates grid points where at least 70% of the ensemble members agree on the sign of the trend in the ensemble mean.

Full size image

Notably, nighttime HWs under AER exhibit stronger negative area-averaged trends in frequency (–0.14 vs. –0.11 HWs season−1 decade−1) and duration (–0.24 vs. –0.22 days HW−1 decade−1) compared to daytime HWs, though slightly weaker trends in intensity (–0.18 vs. –0.22 °C HW−1 decade−1). In contrast, the ensemble mean of the greenhouse gas–only forcing (GHG) displays modestly positive trends in HW metrics across the continent (Figs. 3d1–3; S6d1–3 and S7d1–3). Meanwhile, HW trends under biomass burning (BMB) and everything else (EE) forcing are negligible and non-robust across Africa (Figs. 3e1–f3; S6e1–f3 and S7e1–f3).

Unlike the historical period, the recent period (1985–2014) is marked by a more widespread statistically significant warming trend in compound HWs across parts of Africa in ERA5, with increases in frequency (0.14 HWs season−1 decade−1; Fig. 3a4), duration (0.33 days HW−1 decade−1; Fig. S6a4), and intensity (0.78 °C HW−1 decade−1; Fig. S7a4). Daytime HWs show a spatial pattern similar to compound HWs but with slightly lower intensity (Figs. 3a5; S6a5 and S7a5). Nighttime HWs exhibit even stronger increases, particularly in frequency (0.19 vs. 0.15 HWs season−1 decade−1) and duration (0.35 vs. 0.30 days HW−1 decade−1), compared to daytime HWs (Figs. 3a5, a6 and S6a5, a6). These ERA5-observed trends are broadly captured by ALLF, which mirrors the magnitude and spatial pattern of the intensification, especially the stronger increase in nighttime HW frequency and duration relative to daytime HWs (Figs. 3b4–6 and S6b4–6). Trends in ALLF are robust across most of Africa, except in the Congo Basin for compound HWs and in parts of the northern Sahara for daytime HWs. Moreover, HW trends under GHG-only forcing closely resemble those in ALLF (Figs. 3d4–6; S6d4–6 and S7d4–6), indicating that the recent intensification of HWs is associated primarily with increasing GHG concentrations. In contrast, HW trends under AER, BMB, and EE forcings remain weak and largely non-robust (Figs. 3, S6, S7c4–6, e4–6, and f4–6), reinforcing the dominant role of GHG forcing in recent HW changes over Africa.

Overall, we note that CESM2-LE, depending on the grid point and HW metric, either underrepresents or overrepresents the full extent of anthropogenic intensification of HWs over Africa. Specifically, by assessing whether ERA5 HW trends lie within the 50-member all-forcing ensemble spread—or not—for each grid point during both the historical and recent periods (Fig. 4), we found that during the historical period, CESM2-LE reasonably captures the spatial trends in compound HWs across Africa, although localized overestimations and underestimations are evident (Fig. 4a–c). The model tends to overestimate daytime HW trends over southern Africa (Fig. 4d–f), while generally underestimating nighttime HW trends across the continent (Fig. 4g–i). In the recent period, however, CESM2-LE exhibits a tendency to overestimate compound HW trends over much of Africa, except in the Congo Basin, where it underestimates the trends (Fig. 4j–l). Daytime HWs are underestimated in parts of the northern Sahara but overestimated in western Africa (Fig. 4m–o). Similarly, nighttime HW trends are mostly overestimated, with notable exceptions—such as regions around Nigeria, South Sudan, and Mozambique—where the model underestimates the observed trends (Fig. 4p–r).

Fig. 4: Model overestimation/underestimation of HW metrics relative to ERA5.
figure 4

Overestimation/underestimation of HW metrics in CESM2-LE all-forcing members compared to ERA5 during the historical (1950–1979, a–i) and recent (1985–2014, j–r) periods for compound (a–c, j–l), daytime (d–f, m–o), and nighttime (g–i, p–r) HWs. Dark red (dark blue) dots indicate where the model underestimates (overestimates) the trend in HW metrics compared to ERA5. Light blue and light red dots represent trends within the distribution's tails (below the 25th percentile or above the 75th percentile, respectively). White areas denote regions with no significant model-observation discrepancies.

Full size image

Relationship between near-surface temperature and heatwave metrics

We note that changes in HW metrics for compound, daytime, and nighttime HWs are closely related to changes in near-surface air temperature (∆TAS) characteristics (Fig. 5). Our results reveal statistically significant correlations between ∆TAS characteristics (mean, minimum, and maximum) and changes in HW metrics, particularly for HW frequency (Fig. 5a–c). Specifically, statistically significant strong positive correlations are observed for ∆HW frequency across all three HW types—compound (r = 0.76), daytime (r = 0.85), and nighttime (r = 0.75). Similarly, ∆HW intensity also exhibits a strong positive correlation with ∆TAS for compound HWs (r = 0.73), but the relationship is relatively weaker for daytime (r = 0.42) and nighttime (r = 0.14) events. In contrast, the correlation between ∆HW duration and ∆TAS is weak and statistically insignificant (r < 0.4), suggesting that rising near-surface temperatures do not necessarily lead to prolonged HWs and highlighting the potential role of other physical processes in modulating HW duration.

Fig. 5: Positive correlation between change in HW metrics and temperature.
figure 5

Relationship between changes in a compound, b daytime, and c nighttime HW frequency (HWs season−1), duration (days HW−1), and intensity (°C HW−1); and changes in daily mean, maximum, and minimum near-surface temperatures (°C), respectively. Changes are computed as “recent period (1985–2014)” minus “historical period (1950–1979)”. The solid line indicates the trend, with shaded areas representing the 95% confidence interval. For each HW metric, the correlation coefficient (denoted as "r") and p value (denoted as "p") are provided in the corresponding label. d–f displays the spatial patterns of the change in corresponding temperature, and hatching indicates grid points where at least 70% of the ensemble members agree on the sign of the trend in the ensemble mean.

Full size image

Contribution of Internal Variability to HW Trends

Uncertainty in climate projections arises primarily from emissions scenarios, model limitations, and internal climate variability (IV)68. We leverage the large ensemble of historical simulations from a single climate model to isolate IV, enabling a clearer separation of natural and anthropogenic influences on HW changes. To quantify the role of IV in HW changes, we assessed its relative contribution during the historical and recent periods (Fig. 6). During the historical period, our results show IV dominating all three HW metrics and types, accounting for 80% or more of observed changes across Africa (Fig. 6a–i). In contrast, during the recent period, the contribution of IV to HW trends decreased substantially across much of the continent (Fig. 6j–r), dropping to as low as 30% for compound and nighttime HWs, signaling a stronger anthropogenic influence. Despite this decline, IV remains a significant driver of daytime HW metrics in northern as well as southeastern Africa (Fig. 6m–o). These regions in daytime HWs correspond to areas (Libya, northern Algeria, and Egypt) where ALLF exhibits weak and non-robust HW trends, indicating that these areas are still largely influenced by internal climate variability.

Fig. 6: Relative contribution of IV to HW metrics.
figure 6

The relative contribution of IV (%) to HW frequency, duration, and intensity for compound (a–c, j–l), daytime (d–f, m–o), and nighttime (g–i, p–r) HWs during the historical (1950–1979, a–i) and recent (1985–2014, j–s) periods.

Full size image

Physical Mechanisms Associated with HW Changes

Identifying and understanding the physical mechanisms underlying changes in HWs, as well as the influence of individual external forcings, is essential for increasing confidence in future projections2. Given that observed changes in HW metrics across all HW types are strongly associated with variations in TAS, we investigate the physical drivers of recent warming in near-surface temperature—and, by extension, the associated HW trends. Previous studies have shown that surface air temperature changes are influenced by a combination of physical processes, including but not limited to regional and large-scale circulation, and surface energy exchanges—encompassing both radiative and turbulent heat fluxes69,70,71,72,73.

As earlier reported, the historical ALLF trends in HW metrics across all HW types exhibited weak and non-robust signals due to the opposing effect of aerosol-induced cooling and modest GHG-induced warming (Figs. 3b1–3; S6b1–3 and S7b1–3). This pronounced historical aerosol cooling (Figs. 3c1–3; S6c1–3 and S7c1–3) is largely due to increased sulfate aerosol emissions (≥1.7 kg m−2 decade−1 in the historical period vs. ≤ −1.2 kg m−2 decade−1 in the recent period; Fig. S8a, b), which likely enhanced atmospheric concentration of cloud condensation nuclei (CCN), promoting cloud formation and increasing cloud albedo (Fig. S9a–d). A similar cooling trend in HW metrics between 1950 and 1975 has also been reported over the United States for daytime and compound HWs, and over Mexico for nighttime HWs23.

In contrast, recent intensification of all HW types corresponds with widespread positive TAS anomalies across Africa, with warming hotspots (>1 °C) north of 17°N (Fig. 5d–f). This spatial pattern aligns with increased HW metrics, notably for compound HWs (Figs. 3b4; S6b4 and S7b4). The recent GHG-driven warming, amplified by rising GHGs, black carbon aerosols, and increased atmospheric water vapor, is accompanied by enhanced clear-sky longwave radiation and linked to reduced cloud cover, particularly mid-to-high-level clouds. Specifically, compared to the historical period, clear-sky downwelling longwave radiation and black carbon aerosols have increased by approximately 10 W m−2 (Figs. 7a, i,  S10a, i and S11a, i) and 3 kg m−2 decade−1 (Figs. S1b and S8f), respectively, mainly due to GHG forcing. This increase in radiative forcing corresponds to decreased cloud cover (Fig. S9e–h), partially due to a reduction in sulfate aerosols (≥3 kg m−2 decade−1), especially over the Sahara (Fig. S8a, b). It is worth noting that in contrast to clear-sky longwave radiation, which is strongly influenced by GHGs, changes in clear-sky shortwave radiation generally act to cool the surface (Fig. 7b, j). This cooling is primarily linked to increased shortwave-absorbing black carbon and declining SO₄ aerosols in the recent period compared to the historical period (Figs. S1 and S8). Black carbon directly interacts with and absorbs shortwave radiation, reducing the amount that reaches the surface and thereby potentially cooling it. However, while these shortwave effects partially offset the warming induced by longwave radiation, their overall contribution to recent trends in surface air temperature and associated HWs across Africa appears smaller than that of longwave forcing.

Fig. 7: Physical mechanisms associated with increasing trends in recent HW characteristics.
figure 7

Change in downwelling longwave radiation under clear sky (a, e, i; W m−2 decade−1); downwelling shortwave radiation under clear sky (b, f, j; W m−2 decade−1); total precipitable water (c, g, k; kg m−2 decade−1); geopotential height at 500 hPa (d, h, l; m decade−1) in the recent period (1985–2014) compared to the historical period (1950–1979) for all forcing (a–d; ALLF), aerosol forcing (e–h; AER), and greenhouse gas forcing (i–l; GHG). Hatching indicates grid points where at least 70% of the ensemble members agree on the sign of the change in the ensemble mean.

Full size image

Additionally, atmospheric water vapor has also intensified relative to the historical period due to the increasing greenhouse effect (Figs. 7c, k; S10c, k and S11c, k), further exacerbating warming. Large-scale circulation changes have also contributed to the recent warming through strengthened mid-tropospheric anticyclonic systems, as indicated by rising 500 hPa geopotential heights (Figs. 7d, l; S10d, l and S11d, l), which promote persistent atmospheric blocking and trap warm air near the surface2. Furthermore, intensified surface energy fluxes—general enhancement of both sensible and latent heat—have contributed significantly to recent warming, with sensible heat flux dominating in western and southern Africa, and latent heat flux being more prominent in the Congo Basin and East Africa due to the abundant moisture in the region (Fig. S12b, d). Overall, our results show that the recent intensification of HWs is associated with increased near-surface warming. This warming results from a cascade of interactions and feedbacks—including enhanced clear-sky longwave radiation and surface heat fluxes, reduced cloud cover due to lower cloud-aiding aerosol concentrations, increased emissions of GHGs and absorbing aerosols, intensified atmospheric water vapor, and strengthened anticyclonic circulation.

Discussion

This study investigates key metrics—frequency, duration, and intensity—of three heatwave (HW) types (daytime, nighttime, and compound) across Africa using CESM2 Large Ensemble (CESM2-LE) all-forcing simulations and single-forcing experiments to robustly quantify the relative roles of internal variability (IV) and external forcing. The single-forcing experiments are further used to isolate the individual contributions of anthropogenic aerosols and greenhouse gas (GHG) emissions. Given that Africa spans both hemispheres, peak temperatures vary geographically. To address this, we developed an innovative approach that accounts for regional seasonality by identifying the climatologically hottest three consecutive months at each grid point, harmonizing summer-season definitions across the continent. We analyze and contrast two distinct historical periods: (1) 1950–1979, characterized by a relatively stable near-surface temperature trend due to compensating effects—cooling from increased sulfate (SO₄) aerosol burden and slight warming from rising black carbon and CO₂ concentrations; and (2) 1985–2014, marked by pronounced warming following SO₄ emissions stabilization and continued increases in black carbon and CO₂. Our result reveals that anthropogenic forcing is central to explaining observed HW trends, though its influence differs between periods. During 1950–1979, the ensemble mean of all-forcing (ALLF) simulations shows weak and non-robust HW trends, largely due to aerosol-induced cooling counteracting modest GHG-driven warming. Enhanced SO₄ emissions likely increased atmospheric cloud condensation nuclei (CCN), promoted cloud formation, and enhanced cloud albedo, thereby suppressing surface74,75,76. Moreover, internal climate variability accounted for over 80% of the variance in all HW types during this period. In contrast, during 1985–2014, declining SO₄ emissions and rising GHG concentrations led to widespread warming, circulation adjustments, and robust positive trends in all HW metrics. Compound, daytime, and nighttime HWs became significantly more frequent, longer lasting, and more intense across most of Africa. This warming is primarily GHG-driven and amplified by increased clear-sky longwave radiation associated with reduced cloud cover, elevated GHG and black carbon emissions, and enhanced large-scale anticyclonic circulation and atmospheric moisture, with IV contributing only ~30%.

Our findings highlight that recent increases in HW metrics across Africa are driven primarily by anthropogenic activities—most notably GHG emissions—which have intensified and prolonged heat extremes. Additional contributions from land-use changes such as deforestation, urbanization, and agriculture may further enhance the greenhouse effect4,6,46,77. These human-induced changes not only increase the likelihood of extreme heat events but also exacerbate societal challenges, including water scarcity, crop failure, heat-related illnesses, elevated energy demand, and adverse economic and public health outcomes. The observed robust upward trend in HWs provides critical insights for informing targeted adaptation and mitigation strategies. Strengthening multi-hazard early warning systems, improving heat-risk literacy, and co-developing locally tailored adaptation strategies with vulnerable populations can substantially reduce heatwave exposure and improve preparedness. Although the CESM2-LE historical simulations end in 2014, recent data from ERA5 (Fig. S13) indicate that HWs have continued to intensify through 2024. While CESM2-LE provides a valuable framework for characterizing internal variability, future work should compare these results with other large ensemble datasets to identify robust signals and assess the influence of inter-model differences on projected HW changes.

Methods

Data

In this study, we used daily minimum and maximum temperatures from the CESM Large Ensemble version 2 (CESM2-LE). CESM2-LE comprises 115 ensemble members (including 65 single-forcing members and 50 all-forcing members that have consistent CMIP6 protocols for biomass burning) generated using the Community Atmosphere Model version 6 (CAM6). We chose CESM2-LE because it provides a substantial number of simulations using the same model protocol but with different initial conditions and forcing scenarios. This large ensemble approach allows us to effectively isolate the role of internal climate variability versus external forcings. Unlike the previous CESM1 “all but one” approach, CESM2 employs the “only” approach for forcing experiments. This design allows researchers to separately analyze oceanic and atmospheric contributions to the spread of ensembles, as well as the effects of Atlantic Meridional Overturning Circulation (AMOC) initial conditions on the broader Earth system67. The simulation spans the historical period and extends into the future (under SSP3-7.0) from 1850 to 2100, following the same protocols as CMIP678. For our analysis, we used all 50 members from all-forcing (ALLF), 20 members from the anthropogenic aerosol (AER), and 15 members each from the greenhouse gases (GHG), biomass burning (BMB), and everything else (EE) forcing scenarios57,67,79. In the AER, BMB, and GHG (EE) experiments, the rest of the forcings (AER, BMB, and GHG levels) are fixed at 1850-year levels.

Additionally, Climate Research Unit (CRU) monthly mean near-surface temperature data is used to identify the grid-based hottest season (further discussed in the next section). We also used daily maximum and minimum temperature (computed from hourly data) from ECMWF Reanalysis v5 (ERA5) data—a global reanalysis product from the European Centre for Medium-Range Weather Forecasts (ECMWF), to evaluate model performance. ERA5 incorporates significant improvements over ERA-Interim, including updated physics, data assimilation techniques, and increased spatial resolution (31 km)80.

Heatwave analysis

Most global studies, including those over Africa, use a fixed three or five boreal (austral) summer months in the Northern (Southern) Hemisphere for HW analysis51,81,82,83,84. However, because maximum solar radiation occurs at different times of the year in various regions of both hemispheres, the use of a fixed seasonal month for HW analysis has limitations. These months may not correspond to the periods when different regions receive their peak radiation, especially considering the annual north–south migration of solar radiation. Therefore, this study proposes a new methodology for selecting a temporal scale that reflects the climatologically warmest months across grid points. We calculated the hottest three consecutive months at each grid based on long-term climatology (1965–2014) using CRU TS v4.08. The results presented in Fig. 1 reveal significant spatial variation in the hottest three months across regions. Notably, even along the same latitude, the peak season can shift, emphasizing the limitation of relying on fixed seasonal definitions—particularly in equatorial and lower-latitude regions. This new methodology enhances our ability to capture grid-specific, regional, and temporal variations in HW activity, thereby supporting more targeted and effective climate adaptation strategies. The identified months were then used to filter the dataset to analyze HW characteristics and mechanisms. The following variables from CESM2 were analyzed for the hottest months to identify the underlying physical mechanisms modulating HW characteristics: sulfate aerosol burden (SO4), black carbon burden, total cloud fraction, low cloud fraction, mid-level cloud fraction, high cloud fraction, sensible heat flux, latent heat flux, downwelling shortwave and longwave under clear/all-sky conditions, geopotential height, total precipitable water, and daily mean, minimum, and maximum near-surface temperatures.

The study investigated recent changes in HWs due to anthropogenic influence, focusing on the period from 1950 to 2014. To identify the period during which a potential statistically significant change point occurred, we employed the Pettitt test60,61, a non-parametric method with a significance level set at p value < 0.05. Prior to applying the Pettitt test, we computed the normalized hottest 3-month area-weighted mean over all of Africa for key variables (SO₄ aerosol burden, black carbon burden, CO₂ burden, and near-surface temperature) in ALLF. Based on the change points, we divided the study period into two distinct periods: the “historical period” (1950–1979), characterized by rising SO₄ and CO₂ levels with relatively stable near-surface temperatures, and the “recent period,” marked by stabilized SO₄, continuously increasing CO₂ levels, and pronounced near-surface warming (Fig. S1).

Heatwave definition and description

In this study, following García-Martínez and Bollasina4, daytime (nighttime) HWs are defined as periods of at least three consecutive days where TASmax (TASmin) temperatures exceed the 90th percentile threshold, while TASmin (TASmax) remains below the threshold (Eqs. 1 and 2). The threshold is calculated using a 15-day window centered on each calendar day for the entire study period from 1950 to 2014, and it is calculated for individual members separately.

$${{{\rm{Daytime\; HW}}}},\,{{{{\rm{HW}}}}}_{{day}} = \, {{{{\rm{TAS}}}}}_{\max } > {{{{\rm{TAS}}}}}_{\max -{threshold}}\, \&\\ {{{{\rm{TAS}}}}}_{\min } < {{TAS}}_{\min -{threshold}}$$
(1)
$${{{\rm{Nighttime\; HW}}}},\,{{{{\rm{HW}}}}}_{{night}} = {{{{\rm{TAS}}}}}_{\max } < {{TAS}}_{\max -{threshold}}\, \& \\ {{{{\rm{TAS}}}}}_{\min } > {{{{\rm{TAS}}}}}_{\min -{threshold}}$$
(2)

For compound HWs, both the daily minimum and maximum temperature must exceed the threshold, as defined in Eq. 3.

$${{{\rm{Compound\; HW}}}},\,{{{{\rm{HW}}}}}_{{compound}} = {{{{\rm{TAS}}}}}_{\max } > {{{{\rm{TAS}}}}}_{\max -{threshold}}\, \& \\ {{{{\rm{TAS}}}}}_{\min } > {{{{\rm{TAS}}}}}_{\min -{threshold}}$$
(3)

HWs are further characterized by several metrics, including frequency, duration, and intensity. Here, frequency is defined as the total number of HW events within a season, duration as the average number of HW days per event, and intensity as the averaged sum of temperatures exceeding the 90th percentile threshold during HW events over the season (Eqs. 4–6). The metrics are computed for each member/observation separately before computing the ensemble mean, and robustness is defined as grid points where at least 70% of models agree on the sign of the ensemble mean change65,66.

$${{{{\rm{HWI}}}}}_{{day}}=\,{\sum }_{{{{\rm{t}}}}=1}^{{{{\rm{t}}}}={{{\rm{duration}}}}}\left[{{{{\rm{TAS}}}}}_{\max }\left(t\right)-{{{{\rm{TAS}}}}}_{\max -{threshold}}\right]$$
(4)
$${{{{\rm{HWI}}}}}_{{night}}=\,{\sum }_{{{{\rm{t}}}}=1}^{{{{\rm{t}}}}={{{\rm{duration}}}}}\left[{{{{\rm{TAS}}}}}_{\min }\left(t\right)-{{{{\rm{TAS}}}}}_{\min -{threshold}}\right]$$
(5)
$${{{{\rm{HWI}}}}}_{{compound}} = \,{\sum }_{{{{\rm{t}}}}=1}^{{{{\rm{t}}}}={{{\rm{duration}}}}}\left[\left[{{{{\rm{TAS}}}}}_{\max }\left(t\right)-{{{{\rm{TAS}}}}}_{\max -{threshold}}\right] \right.\\ \left.+\left[{{{{\rm{TAS}}}}}_{\min }\left(t\right)-{{{{\rm{TAS}}}}}_{\min -{threshold}}\right]\right]$$
(6)

Quantification of internal variability

The Earth’s climate is influenced by both natural internal variability and anthropogenic climate change85. To assess the relative contribution of internal variability (IV) to changes in HW occurrence, we employ a signal-to-noise ratio approach, following Deser et al.85,86, Nath et al.87,88, and García-Martínez and Bollasina4. This is calculated as a percentage and denoted as IV:

$${{{\rm{IV}}}}=\left(\frac{\sigma }{{{{\rm{abs}}}}\left({\upmu} \right)+\sigma }\right)* 100$$
(7)

where \(\sigma\) and \(\upmu\) represent the standard deviation and ensemble mean of HW trends across 50 members of CESM2-LE all-forcing, respectively. In this context, the standard deviation indicates the spread of ensemble members, providing an estimate of IV. Meanwhile, the ensemble mean represents the externally forced component. Therefore, the relative contribution of the forced component as a percentage can be derived from \(100-{{{\rm{IV}}}}\).