Introduction

Precipitation variability and change in Africa have profound implications for water resources, agriculture, and ecosystems, impacting millions reliant on rain-fed livelihoods1,2,3,4. Extreme events, including prolonged droughts and intense rainfall, further strain already vulnerable communities and infrastructure5,6,7,8. For example, studies have shown that a + 3 °C warmer climate would result in food deficits (due to shifts in rainfall and higher evaporative demand) that would impact over 1.5 billion Africans, especially in Eastern and Western Africa, and result in an ever higher reliance on food imports to meet caloric demand9,10. Thus, understanding precipitation responses to global warming is essential for informing effective adaptation strategies across Africa.

Previous studies have highlighted significant regional variability in precipitation projections across Africa at the end of the 21st century, with some regions expected to see increased rainfall and others likely to experience reductions11,12. In Western Africa, a zonal contrast is projected over the Sahel, with increased precipitation in the central Sahel and decreased rainfall in the western Sahel11,13,14,15,16,17,18,19,20,21. This pattern results from atmospheric warming, which strengthens low-level moisture flux and shifts the monsoon northward, enhancing central Sahel rainfall13,22. Meanwhile, rising sea surface temperatures increase tropospheric stability and land-based subsidence, reducing rainfall in the western Sahel13,22,23,24. Southern Africa is projected to experience significant drying, especially during the austral summer, due to the poleward shift of westerlies in a warming climate11,25,26,27,28,29,30. This shift strengthens the subtropical high-pressure belt, reducing frontal rainfall over Southern Africa26,30,31,32. Eastern Africa is projected to experience increased precipitation, with the short rains (October–December) strengthening more than the long rains (March-May) due to anomalous circulation patterns enhancing moisture convergence29,33,34,35,36. These include a stronger continental low over Southern Africa, a strengthened subtropical high over the South Indian Ocean, and a weakened eastern Sahara subtropical high. Additionally, warming sea surface temperatures over the western Indian Ocean weaken the Walker circulation, further amplifying short-duration rainfall events37,38,39. Meanwhile, Central Africa’s precipitation projections remain inconsistent due to model uncertainties in convection processes and land-atmosphere interactions40. A weakened northern African Easterly Jet (AEJ) is linked to enhanced moisture convergence in models projecting increased rainfall, whereas a stronger AEJ reduces moisture convergence in models predicting drying35,41,42. These findings highlight the spatial heterogeneity of climate change impacts on African rainfall and the significant uncertainties in future projections.

Despite extensive efforts to project future precipitation changes across Africa, considerable uncertainty remains, and its sources are largely unknown. Additionally, the regional mechanisms driving these changes are not fully understood, particularly the relative contributions of dynamic and thermodynamic processes, which have yet to be quantified. This study addresses these knowledge gaps by systematically investigating not only how regional precipitation would respond to global warming but also the sources of uncertainty in precipitation projections across Africa as well as the mechanisms driving these regional precipitation changes. Using CMIP6 simulations under the SSP2-4.5 and SSP5-8.5 scenarios, this study explores how intermodel variability in projections can be attributed to dynamic or thermodynamic processes.

Results

Projected changes in precipitation

Projected changes in summer mean precipitation for each African region (May to September in the Northern Hemisphere and November to March in the Southern Hemisphere), as simulated by 40 CMIP6 models and their ensemble mean (EnsMean), are presented in Fig. 1 under the SSP5-8.5 and SSP2-4.5 scenarios. Although EnsMean shows an overall increasing trend throughout the 21st century for most regions, regional differences exist. In both scenarios, regions exhibiting a significant increase include Western Africa (WAF; Fig. 1a), North Central Africa (NCAF; Fig. 1c), North Eastern Africa (NEAF; Fig. 1d), South Central Africa (SCAF; Fig. 1e), and South Eastern Africa (SEAF; Fig. 1f), with a stronger increase generally projected under the SSP5-8.5 scenario (Figs. 1 and S2). These wetter conditions are consistent with findings from previous studies11,12.

Fig. 1: Projected regional changes in summer precipitation under the SSP5-8.5 and SSP2-4.5 scenarios.
figure 1

a–i Area-averaged time series of projected change in summer precipitation (%) for the nine African regions during the twenty-first century. Black, blue, and red curves represent the CMIP6 EnsMean for the historical, SSP2-4.5, and SSP5-8.5 scenarios, respectively, and the shaded areas around each curve represent the range of the historical and projected changes across all models. The curves are obtained by taking the difference of each future year with respect to the average from the historical period (1965–2014) and then taking the 10-year running average afterward. j The CMIP6 EnsMean spatial summer rainfall change under SSP5-8.5, with corresponding SSP2-4.5 maps available in the Supplement (Fig. S2). Summer refers to May to September for the Northern Hemisphere and November to March for the Southern Hemisphere. The stippled area in (j) indicates grid points where at least 70% of the models agree with the sign of EnsMean.

Full size image

Relative to the historical period, CMIP6 EnsMean precipitation under the SSP5-8.5 scenario is projected to increase by up to 7% in WAF, 75% in the Sahara (SAH), 15% in NCAF, 25% in NEAF, 17% in SCAF, and 24% in SEAF by the end of the 21st century (Fig. 1a–i). Although less intense at the annual scale, the increase is robust, with at least 70% of the models consistently agreeing on the sign of change across most grid points (Fig. S2). In WAF, while the area-averaged rainfall increases, spatial distribution is not homogeneous. For instance, the western part of WAF is projected to experience a decrease in summer rainfall under the SSP5-8.5 scenario, consistent with previous studies suggesting a drier western Sahel and a wetter central Sahel18,19,20,21. These opposing changes result in a relatively constant WAF-averaged EnsMean precipitation from the mid to late 21st century (Fig. 1a). SCAF and SEAF are projected to experience the highest increase in precipitation in all of Africa, reaching 1 mm/day (2 mm/day) under the SSP2-4.5 (SSP5-8.5) scenario (Fig. 1e–f). The area-averaged analysis also suggests a decrease in precipitation over West Southern Africa (WSAF; Fig. 1g), with a robust decline projected to reach −0.25 mm/day (−2.5%) under SSP2-4.5 and −0.5 mm/day (-5%) under SSP5-8.5 by the end of the 21st century (Fig. 1g). In East Southern Africa (ESAF) and Madagascar (MDG), the area-averaged projected summer rainfall change remains near zero in both scenarios due to inconsistencies in spatial distribution (Figs. 1h–i, S2e–f). Specifically, over ESAF, a decrease is projected in the central region, while the northern and southeastern areas experience an increase. Similarly, in MDG, rainfall is projected to increase in the south but decline in the north. Despite these regional variations, a slight overall increase is projected in both regions under the SSP5-8.5 scenario (Fig. 1h–i).

It is worth noting that, across both scenarios, the robustness of the results presented herein is not sensitive to the length of the historical and projected periods. When shorter (30-year) historical and projected periods in both the near and far future are considered, similar patterns of projected annual and summer precipitation changes are observed, though with slight differences in magnitude (Figs. S3 and S4). We also note that convective precipitation dominates the total precipitation presented here during the historical period. Specifically, convective precipitation accounts for more than 70% of total precipitation across most of Africa, except in parts of northeastern and western coastal SAH, where convective and large-scale precipitation contribute nearly equally (Fig. 2a, d). In the second half of the 21st century, under both scenarios, the contribution of convective precipitation is projected to increase over SAH and slightly over ESAF, while decreasing over WAF and CAF (Figs. 2c and S5c). Conversely, the influence of large-scale precipitation is expected to increase in regions where convective precipitation declines (Figs. 2f and S5f).

Fig. 2: Percentage contribution of convective and large-scale precipitation to total summer rainfall.
figure 2

The contribution (%) of summer (a, b) convective precipitation and (d, e) large-scale (stratiform) precipitation to the total summer precipitation (stratiform + convective) from CMIP6 EnsMean in the historical and SSP5-8.5 scenarios. Panels (c) and (f) are the corresponding projected changes. Summer refers to May to September for the Northern Hemisphere and November to March for the Southern Hemisphere. The stippled area in (c) and (f) indicates grid points where at least 70% of the models agree with the sign of EnsMean.

Full size image

To evaluate potential signals lost in the 5-month climatological summer mean precipitation across African regions, Figs. S6 and S7 illustrate the spatiotemporal seasonality of precipitation and precipitation change, respectively. While no significant shifts in precipitation seasonality are projected across African regions (Fig. S6), an intensification of precipitation is expected during the late summer months in WAF, NCAF, NEAF, ESAF, and MDG. In SEAF, precipitation is projected to increase consistently across all five summer months, whereas WSAF is anticipated to experience a decline over the same period (Fig. S7). These regional change patterns become more pronounced under the SSP5-8.5 scenario.

Sources of uncertainty in projected precipitation changes

The model spread in Fig. 1a–i, along with the intermodel standard deviation (STD) of projected summer precipitation (Fig. S2g, h), highlights the uncertainty in precipitation projections. Many regions exhibit considerable model spread, with WAF showing the highest and SAH the lowest, indicating greater uncertainty in WAF and relatively lower uncertainty in SAH (Fig S2g–h). Additionally, the model spread is consistently larger under the SSP5-8.5 scenario compared to the SSP2-4.5 scenario (Figs. 1a–i, S2g–h). To assess the sources of uncertainty in projected summer precipitation, we quantify the contributions of model uncertainty, scenario uncertainty, and internal variability to the total variance in decadal-mean precipitation projections at different lead times throughout the 21st century (Fig. 3). Our results indicate that model uncertainty (blue shading) is the dominant source of projection uncertainty across all African regions. Specifically, model uncertainty increases with lead time and accounts for over 85% of the total variance in mean summer precipitation projections by the end of the 21st century (Fig. 3a, i). Given that convective precipitation is the primary contributor to total precipitation across Africa (Figs. 2 and S5), we attribute this uncertainty to differences and/or misrepresentation of subgrid-scale processes, especially deep convection, cloud microphysics, and boundary layer turbulence—all of which control key aspects of tropical rainfall generation, such as the initiation and organization of convective storms, cloud–radiation interactions, and vertical moisture transport43,44,45. However, while the contribution of large-scale/stratiform precipitation—represented in the model’s dynamical core through large-scale vertical motion and detailed microphysics—is relatively smaller, its influence on total precipitation and projection uncertainty is becoming increasingly significant.

Fig. 3: Evolution of uncertainty components.
figure 3

The fraction of total variance in mean summer precipitation projections explained by the three components of total uncertainty (unit: %) is shown for (a–i) the nine African regions. Blue, green, and orange shading represent model, scenario, and internal uncertainty, respectively.

Full size image

In contrast to model uncertainty, the influence of internal variability (orange shading) diminishes rapidly as lead times increase, playing a key role only in near-term projections. By the end of the century, internal variability contributes around 10% of the total variance in WSAF, ESAF, and MDG, with smaller contributions (<5%) in other regions. Scenario uncertainty remains relatively small across most regions, becoming more significant only toward the end of the century. Notably, it has no impact in WAF (Fig. 3a) and is negligible in WSAF, ESAF, and MDG (Fig. 3g–i). The influence of scenario uncertainty is most pronounced (around 10%) in NEAF, SCAF, and SEAF (Fig. 3d–f).

Dynamic and thermodynamic drivers of projected precipitation changes

We diagnose the moisture budget equations (MBEs) described in Eqs. 1–5 to reveal the relative contributions of thermodynamic (atmospheric moisture changes) and dynamic (atmospheric circulation changes) processes to the projected summer precipitation change over African regions. As outlined in Eqs. 1–4, we analyze the projected changes in each term of the MBE and present the regional area-averaged values in Fig. 4. In both scenarios, the projected wetter conditions over WAF, NCAF, SCAF, NEAF, and SEAF are primarily associated with enhanced vertical moisture advection \((-\langle \omega {\partial }_{p}q{\rangle}^{\prime})\), dominated by the robust contribution from the vertical thermodynamic term (\(-\langle \bar{\omega }{\partial }_{p}{q}^{\prime} \rangle ,\) vertical TH) (Figs. 3, S8, S9, and S10). This contribution is stronger under the SSP5-8.5 scenario, consistent with the greater projected increase in precipitation. The dominant contribution of the vertical TH term is shaped by subgrid-scale parameterization choices (e.g., deep and shallow convection), which govern convective precipitation. Meanwhile, uncertainties related to subgrid-scale parameterization, including deep and shallow convection treatment, contribute to differences in the magnitude and spatial distribution of enhanced moisture advection. It is worth noting that the increase in precipitation in WAF and NCAF (Fig. 4b, c) is partially offset by the contribution of the vertical dynamic term (\(-\langle \omega^{\prime} {\partial }_{p}\bar{q}\rangle\), vertical DY), while the horizontal thermodynamic term (\(-\langle \bar{{V}_{h}}{\nabla }_{h}{q}^{\prime} \rangle\), horizontal TH) plays a role in counteracting the increase in precipitation in NEAF. Similarly, the slight and highly uncertain increase in precipitation projected in ESAF and MDG (Fig. 4h–i) is driven by the vertical TH term, with a partial offset from the vertical DY term. In SAH (Fig. 4a), the projected wetter conditions are influenced by evaporation (E′), the horizontal dynamic term (\(-\langle {{V}_{h}}^{{\prime} }{\nabla }_{h}\bar{q}\), horizontal DY), vertical DY, and non-linear (NL) terms, with the horizontal DY term being the dominant contributor. However, the projected decrease in the horizontal TH term offsets the expected precipitation increase. In WSAF (Fig. 4g), the projected drier conditions are primarily driven by a reduction in all MBE terms except for the vertical TH term, which has a strong opposite sign. Here, the vertical DY term is the dominant contributor to the decrease in precipitation. These area-averaged results align well with the spatial patterns presented in Figs. S9 and S10, highlighting the relative spatial homogeneity of projected MBE terms over different African regions.

Fig. 4: Projected area-averaged regional changes in summer moisture budget terms (mm/day).
figure 4

The changes in moisture budget terms in the long term (2050–2099, under the SSP2-4.5 and SSP5–8.5 scenarios) averaged over (a–i) the nine regions, relative to 1965–2014. The bars represent the CMIP6 EnsMean, while the vertical line denotes the plus/minus one-time STD based on the spread across all CMIP6 models.

Full size image

In general, the mechanisms driving the changes in summer precipitation can be broadly categorized as follows: (a) regions with projected wetter conditions driven by the vertical TH term (WAF, NCAF, SCAF, NEAF, SEAF, ESAF, and MDG), (b) a region with projected drier conditions driven by the vertical DY term (WSAF), and (c) a region with projected wetter conditions driven by the horizontal DY term (SAH). For regions under (a), despite the negative changes in vertical velocity anomalies \(\left({\omega}^{\prime} \right)\), as shown in Fig. 5c, d, projected summer precipitation increases under both scenarios. This increase is mainly driven by significant moisture increases (q′) (Fig. 5e, f). Following the diagnosis of moisture using Eq. 6, in both scenarios, the increase in the reconstructed moisture (which is consistent with the actual moisture) is largely driven by temperature in all regions, with strong agreement across models (Fig. 6a, b).

Fig. 5: Projected changes in summer near-surface air temperature, vertical velocity at 500 hPa, and vertically integrated specific humidity.
figure 5

Future changes (2050–2099 relative to 1965–2014) of (a, b) near-surface air temperature (unit: K), (c, d) vertical velocity at 500 hPa (unit: Pa/s), and (e–f) vertically integrated specific humidity (unit: kg/m2) in summer projected by CMIP6 models under the SSP2-4.5 and SSP5-8.5 scenarios. Dots indicate areas where at least 70% of the selected models agree on the sign of the EnsMean changes.

Full size image
Fig. 6: The response of specific humidity to changes in temperature and relative humidity.
figure 6

The changes in specific humidity due to changes in temperature and relative humidity (averaged over 1000 – 850 hPa) are shown for the CMIP6 EnsMean under the (a) SSP2-4.5 and (b) SSP5-8.5 scenarios. The prime symbol (ʹ) indicates the change calculated as future (2050–2099) minus historical (1965–2014). QʹActual refers to the model-derived Qʹ, while QʹRecon represents the reconstructed Q' using Equation 6. QʹT denotes the Qʹ response to Tʹ at climatological RH, and QʹRH denotes the Qʹ response to RHʹ at climatological T. The vertical line denotes the plus/minus one-time STD based on the spread across all CMIP6 models.

Full size image

Moreover, consistent with projected increases in moisture, moisture flux convergence is expected to strengthen significantly over WAF, NCAF, and SCAF, driven mainly by stronger low-level westerlies transporting additional moisture from the Atlantic Ocean (Fig. 7a, b). Similarly, enhanced advection from the Indian Ocean is expected to increase moisture flux convergence over SEAF, ESAF, and MDG (Fig. 7c, d). Further analysis of the vertical moisture transport profile across the regions (Figs. S11-S12) indicates that the projected increase in total moisture convergence occurs between 1000 and 850 hPa, with meridional moisture convergence being the primary contributor in WAF, NCAF, NEAF, SCAF, and SEAF. In contrast, in MDG and ESAF, the projected increase in low-level total moisture flux divergence—mainly driven by the zonal component in MDG and both zonal and meridional components in ESAF—partially accounts for the slight increase in precipitation in these regions.

Fig. 7: Projected changes in low-level (1000–850 hPa) moisture flux divergence and transport.
figure 7

Future changes (2050–2099 relative to 1965–2014) in May–September (MJJAS) and November–March (NDJFM) vertically integrated low-level (1000–850 hPa) moisture flux divergence (shading; unit: 10−5 kg.m−2.s−1) and moisture transport (vectors) under the (a–c) SSP2-4.5 and (b–d) SSP5-8.5 scenarios.

Full size image

In the region under (b) (i.e., WSAF), although near-surface temperatures are projected to rise (Fig. 5a, b), a weakening of the ascending branch of the Hadley circulation between 20 °S and 10 °S occurs from November to March (Fig. S13e, f). This weakening reduces the strength of vertical velocity (Fig. 5c, d), suppressing convection and resulting in decreased precipitation. For the region under (c) (i.e., SAH), a weakening of the descending branch of the Northern Hemisphere Hadley circulation is projected under both scenarios (Fig. S13b, c). This weakening is more pronounced under the SSP5-8.5 scenario and is associated with a projected increase in near-surface temperatures (Fig. 5a, b). Combined with increased low-level moisture convergence, this favors enhanced convection and increased precipitation in the region.

Intermodel uncertainty in projected precipitation changes: roles of dynamic and thermodynamic processes

Earlier, we emphasized the noticeable intermodel spread in projected summer precipitation changes across African regions. Here, we analyze the intermodel relationship between the projected changes in precipitation and the moisture budget terms in order to identify the potential source of intermodel uncertainty. The results are summarized in Figs. 8a and S14a for each region under the SSP5-8.5 and SSP2-4.5 scenarios, respectively. Although the vertical TH term predominantly drives the EnsMean projected precipitation increase over WAF, NCAF, SCAF, NEAF, SEAF, ESAF, and MDG (Fig. 4), its correlation with precipitation changes is largely negative and statistically insignificant (at 95% confidence level) across most regions (Fig. 8a). This shows that, while the vertical TH term consistently emerges as a dominant contributor to the mean precipitation increase (as shown in the previous section), it does not account for the intermodel spread. Likewise, the correlation between the horizontal TH term and the projected precipitation increase is also negative and statistically insignificant across most regions. In contrast, the relationships between precipitation and the vertical DY and NL terms are strongly positive and statistically significant (especially the vertical DY term) across all regions (Fig. 8a). This suggests that the spread between models comes from the vertical DY processes, implying that models with stronger vertical circulation weakening will yield a smaller increase in precipitation in most regions (Fig. S15).

Fig. 8: Intermodel relationship between changes in precipitation, global mean surface air temperature (GMSAT), and equilibrium climate sensitivity (ECS) with moisture budget terms.
figure 8

Correlation between (a) precipitation changes, (b) GMSAT changes, and (c) ECS and the change in moisture budget terms across all regions under the SSP5-8.5 scenario. The projected changes are for 2050–2099 relative to 1965–2014. The values represent correlation, with the star beside some correlation values indicating their significance (p-values < 0.05).

Full size image

Previous studies have shown that intermodel projection uncertainties can be partially influenced by variations in global mean warming across models46,47,48. To explore whether the intermodel uncertainty in projected summer precipitation and moisture budget terms is linked to the spread in global mean surface air temperature (GMSAT) changes, we assess their correlations (Figs. 8b and S14b). Results indicate that, in both scenarios, the vertical TH term is positively and significantly related to GMSAT changes in most regions, except SAH, where a negative correlation is observed. This suggests that models projecting a warmer climate also tend to simulate a stronger vertical TH term (Fig. S16). Over SAH, the NL term exhibits the strongest positive and most significant relationship with GMSAT changes. Since GMSAT is closely linked to climate sensitivity, we also analyze the relationship between moisture budget terms and equilibrium climate sensitivity (ECS). Our results reveal that the vertical TH term strongly depends on ECS, showing positive and significant correlations across most regions except SAH (Figs. 8c, S14c, and S17). This highlights the critical role of intermodel ECS spread in driving uncertainty in the vertical TH term. Given the variability in climate sensitivity among the CMIP6 models49, these differences contribute to variations in the projected vertical TH term across the models.

Discussion

This study presents a comprehensive assessment of the dynamic and thermodynamic processes driving projected changes in mean precipitation across Africa alongside the sources of projection uncertainty using an ensemble of 40 CMIP6 models. Our results reveal an overall increasing trend through the 21st century, with regional variations. Under the SSP5-8.5 and SSP2-4.5 scenarios, significant increases are projected in WAF, NCAF, NEAF, SCAF, and SEAF, with stronger increases under SSP5-8.5. By the late 21st century, precipitation is expected to rise by up to 75% in SAH, 25% in NEAF, and 24% in SEAF. The increase is robust, with at least 70% of models agreeing on the direction of the ensemble mean change. It should be noted that WAF exhibits spatial variability, with decreasing (increasing) projected changes in the western (central-eastern) portion. SCAF and SEAF show the highest increases, while WSAF is projected to experience a decrease. ESAF and MDG display near-zero area-averaged changes due to mixed regional patterns, with ESAF seeing decreases in central areas and increases in the north and southeast, while MDG shows increases in the south and decreases in the north. It is important to highlight that convective precipitation makes up over 70% of the total precipitation across Africa.

A key contribution of this study is the quantification of uncertainty sources in African precipitation projections. We found that model uncertainty, primarily driven by subgrid-scale parameterizations, dominates across all regions, accounting for over 85% of total variance by the late 21st century. Internal variability plays a significant role in near-term projections but declines over time, contributing approximately 10% over WSAF, ESAF, and MDG by the end of the 21st century. On the other hand, scenario uncertainty remains relatively small in most regions but becomes slightly distinct at the end of the century, particularly in NEAF, SCAF, and SEAF.

To understand the mechanisms driving precipitation changes, a moisture budget analysis was conducted to distinguish between thermodynamic and dynamic contributions. Our results suggest that vertical thermodynamic effects, associated with increased atmospheric moisture, play a key role in regions projected to become wetter (WAF, NCAF, NEAF, SCAF, and SEAF). In contrast, vertical dynamic changes, associated with a projected decrease in ascending motion, largely explain drying trends in WSAF. However, across Africa, the interplay between atmospheric mechanisms varies spatially and is often modulated by regional-scale processes such as changes in the African Easterly Jet and subtropical high-pressure systems1,18,28,38,39,41,42,50,51.

We note that distinct processes govern projected precipitation changes and intermodel spread. Precipitation increases are mainly driven by vertical thermodynamic (TH) processes, reflecting the robust influence of increased atmospheric moisture in a warmer climate. In contrast, intermodel uncertainty is more strongly associated with vertical dynamical (DY) processes, including variations in circulation and convective responses. Furthermore, we found that intermodel uncertainty in African hydroclimate projections is also influenced by variations in global mean surface air temperature (GMSAT), with a positive correlation between TH and GMSAT in most regions, except SAH. Additionally, variability in equilibrium climate sensitivity (ECS) across CMIP6 models contributes to uncertainties in the vertical TH response, underscoring the role of climate sensitivity in precipitation projection uncertainty.

Overall, while substantial uncertainty remains in the projection of African precipitation due to intermodel spread, the spatial patterns and sign of change are largely consistent across models and emission scenarios. This coherence offers valuable guidance for adaptation planning. Nevertheless, decision-makers are advised to pursue flexible adaptation strategies that accommodate a possible range of precipitation changes. We advocate for continued investment in climate model development and improved process-level understanding, which are critical for reducing uncertainties and enhancing the robustness of climate projections for informed decision-making.

Methods

Study area

This study focuses on eight African regions as defined by the latest IPCC-AR6 climate reference regions52: Sahara (SAH), Western Africa (WAF), Central Africa (CAF), North Eastern Africa (NEAF), South Eastern Africa (SEAF), East Southern Africa (ESAF), West Southern Africa (WSAF), and Madagascar (MDG). To account for distinct seasonal precipitation cycles, we further subdivided CAF into northern (NCAF) and southern (SCAF) regions (Fig. S1).

Data

We use output from 40 CMIP6 historical (1965–2014) and future (2050–2099) projections (see Table S1 for more details) under Shared Socioeconomic Pathways (SSP) 2-4.5 and 5-8.5, representing moderate and very high greenhouse-gas (GHG) emissions scenarios. Rather than the commonly used 30-year period found in the literature, we adopt a 50-year averaging period for both historical and future projections to enhance the statistical robustness of our results. This choice is particularly important for regional precipitation over Africa, which is characterized by strong internal variability and episodic extremes. A longer averaging window reduces the influence of interannual and decadal variability, provides a more reliable estimate of forced climate responses under scenarios such as SSP5-8.5, and is consistent with established uncertainty quantification frameworks53,54 that rely on long-term means to distinguish forced signals from internal variability. Monthly precipitation, evaporation, 2-m temperature, zonal wind, meridional wind, vertical velocity, and specific humidity from the first member are employed for each model. To assess the performance of model outputs (Fig. S1), the following observational datasets are utilized: the Climate Hazards group Infrared Precipitation with Stations55, with a grid spacing of 0.05°; the Global Precipitation Climatology Center full data daily V201856 (GPCC_FDD), at a spatial resolution of 1°; the unified gauge-based analysis of global daily precipitation57 (CPC), with a horizontal resolution of 0.5°; and the NOAA NCEP CPC FEWS African Precipitation Climatology from the Famine Early Warning System (ARC2), available at a grid resolution of 0.1° (ftp.cpc.ncep.noaa.gov/fews/AFR_CLIM/ARC2/).

To enable consistent comparisons across all datasets and generate the multi-model ensemble mean (hereafter referred to as “EnsMean”), we regrid all data to a spatial resolution of 2.81° × 2.81° using first-order conservative remapping, based on the lowest model resolution. This approach (rather than upscaling coarse data) is commonly used in multi-model analyses4,58,59,60 because it avoids interpolation artifacts that could bias statistical comparisons between models. The analyses focus on the extended summer months—May to September (November to March) for the Northern (Southern) Hemisphere—covering the African continent. These are the rainy months in the regions of interest (Fig. S1a–i).

Uncertainty and moisture budget analysis

Previous studies have shown that uncertainty in climate projections arises from three main sources: internal variability, model errors, and unknown future trajectories of greenhouse gas and aerosol emissions47,53,61,62,63,64,65. In this study, we assess the uncertainty in precipitation projections across Africa by applying the method developed by refs. 53,64. Following the approach described in ref. 47 (see Eqs. 1–6 in their supporting information), we also compute the contribution of each of these sources of uncertainty to the total variance in decadal-mean projections.

Furthermore, to investigate the underlying physical mechanisms of precipitation changes across Africa, we diagnose the moisture budget equation following the approach described by refs. 46,47,65:

$${P}^{{\prime} }={E}^{{\prime} }-{{{{\partial }_{t}{{\langle }}q{{\rangle }}}^{{\prime} }-{{\langle }}V}_{h}{\nabla }_{h}q{{\rangle }}}^{{\prime} }-{{{\langle }}\omega {\partial }_{p}q{{\rangle }}}^{{\prime} }+{{Res}}^{{\prime} }$$
(1)

The variables P, E, q, ω, and \({V}_{h}\) represent precipitation, evaporation, specific humidity, vertical velocity, and horizontal velocity, respectively. Res refers to the residual term. Primes represent the climatological difference between the future and the historical, while angle brackets (“\(\langle \rangle\)”) signify the mass vertical integration throughout the troposphere, as shown in Eq. 2.

$${{\langle }}X{{\rangle }}=\frac{1}{g}{\int }_{{P}_{s}}^{{P}_{T}}{X}\,{dp}$$
(2)

On the seasonal mean timescale, the time derivative of specific humidity, \({{\partial }_{t}\langle q\rangle }^{{\prime} }\), is negligible as it is significantly smaller than the other terms. \(-{\langle V}_{h}{\nabla }_{h}q\rangle\) represents vertically integrated horizontal moisture advection, and \(-\langle \omega {\partial }_{p}q\rangle\) denotes vertically integrated vertical moisture advection. According to Eq. 1, the terms \(-{{\langle V}_{h}{\nabla }_{h}q\rangle }^{{\prime} }\) and \(-{\langle \omega {\partial }_{p}q\rangle }^{{\prime} }\) can be further broken down as follows:

$${-{{\langle }}{V}_{h}{\nabla }_{h}q{{\rangle }}}^{{\prime} }=-{{\langle }}{{V}_{h}}^{{\prime} }{\nabla }_{h}\bar{q}{{\rangle }}-{{\langle }}\overline{{V}_{h}}{\nabla }_{h}{q}^{{\prime} }{{\rangle }}-{{\langle }}{{V}_{h}}^{{\prime} }{\nabla }_{h}{q}^{{\prime} }{{\rangle }}$$
(3)
$${-{{\langle }}\omega {\partial }_{p}q{{\rangle }}}^{{\prime} }=-{{\langle }}{\omega }^{{\prime} }{\partial }_{p}\bar{q}{{\rangle }}-{{\langle }}\bar{\omega }{\partial }_{p}{q}^{{\prime} }{{\rangle }}-{{\langle }}{\omega }^{{\prime} }{\partial }_{p}{q}^{{\prime} }{{\rangle}}$$
(4)

The overbar represents the baseline climatology. The terms \(-\langle {{V}_{h}}^{{\prime} }\nabla \bar{q}\rangle\) and \(-\langle {\omega }^{{\prime} }{\partial }_{p}\bar{q}\rangle\) correspond to the dynamic (DY) components, describing changes in horizontal and vertical moisture advection, respectively, driven by circulation changes. In contrast, \(-\langle \overline{{V}_{h}}\nabla {q}^{{\prime} }\rangle\) and \(-\langle \bar{\omega }{\partial }_{p}{q}^{{\prime} }\rangle\) are the thermodynamic (TH) components capturing the influence of water vapor changes associated with temperature changes. The horizontal and vertical non-linear terms, represented by \(-\langle {{V}_{h}}^{{\prime} }{\partial }_{p}{q}^{{\prime} }\rangle\) and \(-\langle {\omega }^{{\prime} }{\partial }_{p}{q}^{{\prime} }\rangle\), respectively, are collectively denoted as NL. Consequently, the overall precipitation changes can be expressed as:

$${P}^{{\prime} }={E}^{{\prime} }-{{\langle }}{{V}_{h}}^{{\prime} }{\nabla }_{h}\bar{q}{{\rangle }}-{{\langle }}\overline{{V}_{h}}{\nabla }_{h}{q}^{{\prime} }{{\rangle }}-{{\langle }}{\omega }^{{\prime} }{\partial }_{p}\bar{q}{{\rangle }}-{{\langle }}\bar{\omega }{\partial }_{p}{q}^{{\prime} }{{\rangle }}+{NL}+{{Res}}^{{\prime} }$$
(5)

To better understand the factor(s) responsible for changes in specific humidity (q), we decompose these changes into contributions from temperature (T) and relative humidity (RH), following the approach of ref. 66. Since specific humidity depends on temperature (°C) and relative humidity (%), it can be expressed as:

$$q=\frac{6.1094\frac{{M_{w}}}{{M_{d}}}\frac{{RH}}{100}exp \left(\frac{17.625T}{T+243.04}\right)}{p-6.1094\left(1-\frac{{M_{w}}}{{M_{d}}}\right)\frac{{RH}}{100}exp \left(\frac{17.625T}{T+243.04}\right)}$$
(6)

Here, \({M}_{w}\) represents the molar mass of water vapor (\({M}_{w}\) = 18.01528 g/mol), while Md denotes the molar mass of dry air (Md = 28.9634 g/mol). Utilizing Eq. 6, we analyze how specific humidity (q′) responds to changes in temperature (T′) under constant relative humidity (i.e., RH at climatology; \(\overline{{RH}}\)), as well as its response to changes in relative humidity (RH’) when temperature remains fixed (i.e., T at climatology; \(\bar{T}\)).