the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 10 Mar 2026
Historical and Projected Changes in Temperature–Precipitation Compound Hot and Dry Extremes across Africa Based on CMIP5 and CMIP6 Ensemble Simulations
Abstract. Compound climate extremes pose disproportionate risks to societies and ecosystems, yet their evolution across Africa remains poorly constrained. Here we present the first continent-wide, multi-generation assessment of historical and projected temperature–precipitation compound hot–dry extremes across Africa using CMIP5 and CMIP6 multi-model ensembles under the full range of emission scenarios. Observations indicate continent-wide warming of 0.21 °C decade-1; during 1980–2020, accompanied by spatially heterogeneous precipitation trends, establishing a baseline of increasing compound vulnerability. Despite projected increases in precipitation over much of tropical Africa, both CMIP5 and CMIP6 consistently project strong intensification of compound hot–dry extremes, driven primarily by accelerated warming. By mid-century (2035–2065), compound hot–dry frequency increases across all regions, with southern Africa experiencing 0.33–0.43 hot–dry months yr-1; in CMIP6. By end-century (2070–2100), frequencies in western and eastern southern Africa reach 0.36–0.62 months yr-1; under high-emission scenarios, representing more than a doubling relative to low-emission pathways. Compound event severity intensifies nonlinearly: cumulative event magnitude exceeds 10–13 σ in southern Africa and Madagascar, while mean event duration lengthens from ∼1–1.5 months historically to 4–6 months under high emissions. CMIP6 systematically projects stronger increases in compound frequency, magnitude, and duration than CMIP5, reflecting enhanced land–atmosphere coupling and higher climate sensitivity. Although strong mitigation substantially limits these increases, compound hot–dry extremes intensify even under low-emission pathways. These results demonstrate that compound climate risk may escalate regardless of mean precipitation trends, underscoring the urgency of compound-aware adaptation strategies and the substantial mitigation benefits of limiting future climate impacts.
- Preprint
(2888 KB) - Metadata XML
-
Supplement
(2504 KB) - BibTeX
- EndNote
Status: final response (author comments only)
- RC1: 'Comment on egusphere-2026-721', Anonymous Referee #1, 29 Jun 2026
-
RC2: 'Comment on egusphere-2026-721', Anonymous Referee #2, 01 Aug 2026
Referee report on:
Historical and Projected Changes in Temperature-Precipitation Compound Hot and Dry Extremes across Africa Based on CMIP5 and CMIP6 Ensemble Simulations
(egusphere-2026-721)
Overall assessment:
The work is a well-motivated and timely study on a continental-scale analysis that fills a gap within the literature (systematic multi-generation, multi-scenario compound hot-dry assessment for Africa). The writing is generally clear, and the figures are informative. However, the baseline-period inconsistency (major comment 1), the ensemble/percentile computation ambiguity (major comment 2,3,4), and the lack of compound-metric validation (major comment 5) should be addressed before the paper is suitable for publication. I recommend major revisions.
Major comments
- There are some baseline-period inconsistencies throughout the paper. For example, in Section 2.2, we know that the historical period for CMIP5 ends in 2005 and CMIP6 ends in 2014. At the beginning of the section, the authors define CRU dataset as the “historical climate observations”, but “historical” should only be used for the historical simulations. Maybe calling the CRU only the “climate observations”. Then in section 2.4 and 2.5, the authors define the historical baseline as 1975-2005. However, the caption of Figure 7 (and by extension Figure 8) states changes are computed "relative to 1986–2014 baseline" (line ~687). This is either a typo or reveals that a different baseline was actually used for the regression analysis than for the frequency/magnitude/duration analysis. Please reconcile these numbers throughout the manuscript, since this affects the reproducibility of a central result. For example, in Section 2.2 the authors could define all baselines used in the manuscript (to compare with observations, and to make the scenario analysis). This would improve dramatically the readability and clarity of the paper.
- It is not clear from Section 2.5 whether the 90th/10th percentile thresholds and the compound-event flags are computed (a) on the multi-model ensemble-mean time series, or (b) on each model and then aggregated as ensmeble. This may be important because averaging across models before computing percentiles artificially damps the variance of the resulting series (since unforced internal variability is largely incoherent across models), which can substantially bias both the thresholds and the estimated compound-event frequency/duration. Please state explicitly which approach was used, and ideally show a sensitivity test (e.g., compare ensemble-mean-based vs. per-model-then-averaged compound frequencies for at least one region).
- Focusing on the same issue, the 90th/10th percentile thresholds for each calendar month are estimated from only 31 annual values (1975–2005), which makes estimating thresholds intrinsically noisy, especially in regions with high interannual variability (e.g., Sahel, East Africa). Please report/discuss the sampling uncertainty on P90 and P10 (e.g., via bootstrap resampling), or justify why a 31-year window is enough. This matters more here than in typical single-model studies because compound frequencies of 0.02–0.05 months/year in northeast Africa are extremely close to zero and thus sensitive to threshold noise.
- Still on this issue, ensemble spread/uncertainty is lacking until the conclusions. The Conclusion (line 760) acknowledges that "multi-model ensemble means mask inter-model spread," but this is a first-order caveat for a paper whose central analysis is regional "hotspot" identification for adaptation prioritization. I'd recommend reporting model agreement on sign of change (e.g., % of models agreeing) or an inter-model standard deviation/IQR for the key frequency/magnitude/duration metrics in at least one summary figure or table, rather than relegating this to a closing sentence.
- Model evaluation in Section 3.1/Fig. 1 addresses only the univariate temperature and precipitation trends. The compound hot–dry metric (frequency, magnitude, duration) is never validated against CRU-derived compound events for the baseline period. Given that compound risk depends non-linearly on the joint tail behaviour of two variables (and their dependence structure), reproducing marginal trends does not guarantee the models reproduce the observed compound frequency. Adding a historical-period comparison (CRU-derived vs. CMIP5/CMIP6 compound frequency, 1980–2020 or 1975–2005) would substantially strengthen confidence in the projections.
- Compound heat–drought hazards typically occur on daily-to-weekly timescales that can be smoothed out or missed entirely when only monthly means are used. A single anomalously hot week embedded in an otherwise average month does not register as a "compound month." Please justify the choice of monthly resolution more explicitly (data availability, computational tractability, comparability with Manning et al., 2019 who used summer daily data in a different context) and discuss this as a limitation with more specificity. The current Limitations paragraph (lines 760–764) mentions fixed thresholds but not the temporal-resolution issue.
- In Fig7 and Fig9, each regression panel has only n = 8 data points (one per African subregion). With n=8, r² values of 0.47–0.71 and reported significance should be interpreted cautiously. Some panels lack any significance marker (e.g., Fig. 7b, Fig. 8b–h) while conclusions in the text ("robust inverse relationship...”) read as more general than the evidence supports. The sentence: "a robust inverse relationship emerges", may be too strong if several individual scenario panels are not significant. Also note that pooling very different climate regimes (hyper-arid Sahara with humid Congo Basin) into a single linear fit mixes the cause of drying (different physical mechanisms in each region), and a warning on this heterogeneity on the text would help.
- Lines 463–464 state: "... CMIP5 projects slightly higher peak frequencies in WSAF under RCP8.5 (0.58 months per year) than CMIP6 under SSP5-8.5 (0.62 months per year), though the difference is modest." This is contradictory? 0.58 < 0.62, so CMIP6 is actually higher, not CMIP5. Please correct the text.
- Tables S1/S2 (listing the CMIP5/CMIP6 models used) are referenced but their content (number of models, resolutions) is never summarized in the main text. Since ensemble size/composition strongly affects the robustness of the "ensemble mean," please add a sentence (e.g., "CMIP5 ensemble comprises N models... CMIP6 comprises M models...") to the main text, not only the supplement.
Minor comments
Line 504: “both CMIP5 and CMIP6 projects substantial increases” – change to “project”.
Line 134: "This study encompasses the entire African continent and is analysed using...", this is a strange construction (a "study" is not "analysed"). Consider "We analyse the entire African continent using the IPCC AR6 reference regions framework..."
Section 2.5 (lines ~236–248): the sentence describing the ~1% independence probability and threshold rationale is repeated almost entirely immediately before and after Equation 1. Please remove the duplication.
Fig2: Consider using the same range for yy-axis for CMIP5 and CMIP6 graphs, to simplify comparison.
Include a map with region boundaries showing the 8 regions with their acronyms (WAF, CAF, NEAF, SEAF, WSAF, ESAF, SAH, MDG), so any reader can automatically understand where are the regions.
Fig1 and Fig2: the authors consider the continental AFR, but in the remaining manuscript don’t use it again in figures (although it is present in text). For example bar charts in Fig3 to Fig6 for AFR?
Equation 5 sums standardized T and P anomalies with implicit equal weighting. Please briefly justify this choice, since equal weighting of the two is not obvious to combine hazards with different physical units and different impact sensitivities.
Please explicitly state whether CRU-based thresholds are used to evaluate the compound-event methodology in the historical period, and if the 1975–2005 "compound baseline" and the 1980–2020 "trend-detection baseline" (Section 2.3) are the same underlying data subset or different. Currently two different windows appear in the Methods for two different purposes, which could confuse readers on a first read. A short clarifying sentence distinguishing "trend-detection period" from "percentile-threshold baseline period" would help. (See major comment 1).
Consider softening the sentence "the first systematic analysis..." / "the first continent-wide, multi-generation assessment" (Abstract, Introduction, Conclusion), which is repeated three times. While the claim seems reasonable, such superlatives recur so often that it reads as promotional. Consider stating it once, clearly, and let the results speak in the other instances.
Citation: https://doi.org/10.5194/egusphere-2026-721-RC2
Viewed
| HTML | XML | Total | Supplement | BibTeX | EndNote | |
|---|---|---|---|---|---|---|
| 685 | 402 | 71 | 1,158 | 152 | 71 | 103 |
- HTML: 685
- PDF: 402
- XML: 71
- Total: 1,158
- Supplement: 152
- BibTeX: 71
- EndNote: 103
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
This work assesses current and projected trends in hot-dry compound events across the African continent, utilizing CMIP5 and CMIP6 climate model ensembles. The evaluation covers the full set of RCP and SSP scenarios, respectively, across three main periods: historical (1975-2005), near-future (2035-2065), and end-of-century (2070-2100). The authors find an overall increase in the frequency, duration, and magnitude of compound events, primarily driven by a homogenized temperature increase across the continent. Spatial patterns, however, are largely determined by precipitation variability. The systematic analysis reveals regional differences in seasonality and scenario dependence. These results contribute to the understanding of compound hot-dry extremes for Africa, a region underrepresented in current studies. Therefore, I find this manuscript suitable for publication in NHESS. Nevertheless, some points require clarification before acceptance. Please find my detailed comments in the attached document.