African CMIP6 Ensembling — Choosing Climate Models for Regional Adaptation Work

In one paragraph

When you read a future climate projection from a CGIAR Climate Data Hub climate-rationale tool, you’re looking at an average across multiple climate models. We do not use a single model, and we do not blindly average all available models. Instead, we use regionally-tuned subsets — guided by the peer-reviewed literature — so that the projections you see for Ethiopia are based on models that perform well over the Horn of Africa, and the projections for Ghana are based on models that capture the West African monsoon. This page explains what we mean by an ensemble, why we tune by region, which models we use and don’t use, and what’s coming next from CMIP7 and CORDEX-Africa.

What this page is

This page is the public reference behind one specific methodological choice we make for you: which set of climate models we use when we show you a future-projection chart in any of the Climate Data Hub’s climate-rationale tools. The choice was first applied in the AAA Adaptation Atlas’s Build a Climate Rationale notebook, and the reasoning below applies to any African climate-rationale work using NEX-GDDP-CMIP6.

It’s written for the climate-rationale writer — someone preparing a Green Climate Fund (GCF) concept note, an Adaptation Fund proposal, a national NAP submission, or an internal scenario brief — who needs to understand and defensibly cite the methodology behind the numbers they’re seeing. You don’t need a PhD in climate science. You do need to know enough to answer a reviewer who asks “why these models, not others?”

We assume you know what greenhouse-gas emissions are, what climate change is, and roughly what “an SSP scenario” means. Everything else, we define here.

If you remember three things

1. Climate projections are always a range of futures, not a single line. We show you the average across multiple models and how much they disagree.
2. Our default ensemble (AFR-13) drops three of the 18 available models for documented scientific reasons — they’re either too sensitive (one runs too hot) or not well-validated for African climate.
3. Some regions (especially East Africa’s long rains) have a known structural issue where models disagree with observations. We flag this; we don’t try to hide it.

Why not use “the” climate model?

There is no “the” climate model. There are dozens of climate models maintained by research institutes around the world — NOAA’s GFDL, NCAR’s CESM, Météo-France’s IPSL, Germany’s MPI-ESM, the UK Met Office’s UKESM, Japan’s MRI, and many more. Each one is a different team’s best attempt to simulate the global climate system, with different choices about how to represent clouds, ocean mixing, atmospheric chemistry, sea ice, vegetation, and a thousand other processes.

These models disagree with each other. Run the same emissions scenario through ten different climate models and you’ll get ten different projections — sometimes with similar overall direction (everyone warms) but very different magnitudes (some warm twice as much as others). For precipitation, the disagreement is often even larger — some models project wetting over a region where others project drying.

This disagreement is not a bug. It reflects genuine uncertainty in how we represent the climate system. The honest scientific answer to “what will Ethiopia’s rainfall look like in 2050?” is not a single number but a distribution: a most-likely value with a spread around it that captures the range of plausible futures.

Figure 2. No single climate model is best everywhere over Africa. This Taylor diagram (adapted from Samuel et al. 2025) shows how 18 climate models compare against CHIRPS rainfall observations across five African sub-regions. Models that cluster near the reference point are "more skilful"; models far from it have larger biases. The clustering pattern differs between regions — so a model that's good for West Africa may be middling for the Horn of Africa, and vice versa.

The standard way to communicate this honestly is with an ensemble: a collection of models, all run under the same scenario, treated as samples from the space of “plausible futures.” The ensemble mean is your central estimate; the ensemble spread (e.g. ±1 standard deviation) is your uncertainty.

So the question becomes: which models go into the ensemble?

The naïve answer — “all of them, equal weight” — has problems. Some models have known systematic flaws over Africa. Some have implausibly high or low climate sensitivities. And including them at equal weight gives them as much voice as the well-validated ones, which inflates the apparent uncertainty in ways that don’t actually reflect the science.

Our answer: use a regionally-tuned subset, guided by the peer-reviewed evaluation literature.

The 18 NEX-GDDP-CMIP6 models

Our projections come from NASA’s NEX-GDDP-CMIP6 dataset — a statistically-downscaled version of the global CMIP6 model archive, served at 0.25° resolution (about 25 km grid spacing), which is fine enough to be useful at the country and admin-1 scale for Africa.

The Atlas currently ingests 18 climate models from NEX-GDDP-CMIP6. (NASA’s full NEX-GDDP-CMIP6 catalogue is broader — the 2022 Thrasher et al. Scientific Data paper documents 35 models — but the Atlas’s pipeline was built around the original 18-model release.) Some well-known CMIP6 models — UKESM1-0-LL (UK Met Office), HadGEM3-GC31-MM (also UK Met Office), GFDL-CM4 (NOAA) — are not in any NEX-GDDP-CMIP6 release, simply because NASA didn’t downscale them. That constrains what the Atlas can offer.

The full 18-model pool we can draw on:

The NEX-GDDP-CMIP6 18-model pool

ACCESS-CM2 · ACCESS-ESM1-5 · CanESM5 · CMCC-ESM2 · EC-Earth3 · EC-Earth3-Veg-LR · GFDL-ESM4 · INM-CM4-8 · INM-CM5-0 · IPSL-CM6A-LR · KACE-1-0-G · MIROC6 · MPI-ESM1-2-HR · MPI-ESM1-2-LR · MRI-ESM2-0 · NorESM2-LM · NorESM2-MM · TaiESM1

All models are the r1i1p1f1 realisation (the first physical ensemble member, used as the standard reference run).

These 18 are not all equally well-suited to African climate. Our job is to pick the defensible subset for each region, document the reasoning, and let you override the default when you need to.

What “African-tuned” means here

We offer three main ensemble options, plus per-region tuning:

Ensemble options

• AFR-13 — recommended default. 13 of 18 models. Drops the three most-extreme climate sensitivities (one too hot, two too cold) and two models with limited African validation. Use this unless you have a specific reason to override.
• AFR-8 — high-consensus subset. 8 of 18 models. The “consistently top-performing across most African regions” set, drawn from the evaluation literature. Smaller spread, tighter signal.
• FULL-18 — NASA NEX-GDDP-CMIP6 default. All 18 models. Use this if you want maximum uncertainty range, are doing sensitivity analysis, or want comparability with studies that use the full set.
• Per-region “Auto” mode. Selects a region-specific subset based on the country you’ve chosen. The chart shows you which subset is active.

Why per-region tuning?

The clear finding in the literature (Samuel et al. 2025, the most directly applicable evaluation paper for NEX-GDDP-CMIP6 over sub-Saharan Africa) is that no single model consistently outperforms others across all African sub-regions. A model that captures the West African monsoon may be middling at the Horn of Africa long rains; a model that’s good at Southern African subtropical circulation may struggle with the Sahel.

So we use AR6 reference regions (the standard sub-regions defined by Iturbide et al. 2020 for the IPCC Sixth Assessment Report) to organise the choices. When you select a country in a climate-rationale tool that uses this methodology, the underlying ensemble adapts to the region the country sits in.

Figure 1. The AR6 reference regions for Africa, used by the IPCC and adopted here as the organising structure for regionally-tuned climate-model ensembles. Reference: Iturbide et al. 2020.

The regional sub-ensembles

For each AR6 sub-region in sub-Saharan Africa, we select a sub-ensemble whose composition is supported by the regional evaluation literature.

Western Africa (AFR-WAF, 8 models) — drops models that consistently underperform on the West African monsoon (notably the ACCESS family, which has a well-documented Sahel cold bias). Keeps the IPSL family, EC-Earth3 variants, and the GFDL / NorESM family, which the regional literature flags as robust performers over the Sahel and Guinea coast. References: Faye & Akinsanola 2022 (CMIP6 extreme precipitation over West Africa); Makinde et al. 2026 (West African Westerly Jet representation).

Central Africa (AFR-CAF, 8 models) — limited region-specific evaluation literature compared to West and East Africa. We default to the “consistent African performers” subset (same composition as AFR-WAF). Users in Central Africa should treat projections with somewhat higher epistemic uncertainty than other regions; the model-evaluation evidence base is thinner. Reference: Samuel et al. 2025 (continental).

East Africa and Horn of Africa (AFR-EAF, 7 models) — the Park et al. 2023 analysis of long-rains and short-rains representation flags ACCESS-ESM1-5, EC-Earth3-Veg, MRI-ESM2-0, IPSL-CM6A-LR, and MPI-ESM1-2-HR as the best CMIP6 models for the Greater Horn of Africa. Two of their top picks (UKESM1-0-LL and HadGEM3-GC31-MM) are not in NEX-GDDP-CMIP6, so our AFR-EAF subset is a constrained version of their recommendation. See the East African Paradox caveat below — it applies to ALL these models.

Southern Africa — Western (AFR-WSAF, 8 models), and Eastern (AFR-ESAF, 8 models) — the ACCESS family (developed in Australia, tuned for Southern Hemisphere subtropical dynamics) performs notably well over the Kalahari–Namib axis and the eastern escarpment systems. AFR-WSAF and AFR-ESAF both include the ACCESS pair plus the core African-performers set. References: Pinto et al. 2018 (Southern Africa CORDEX/CMIP5 process-based evaluation); Lim Kam Sian et al. 2021 (multi-decadal variability and future changes in precipitation over Southern Africa).

Madagascar and small islands (AFR-MDG, 13 models) — Madagascar is a special case. At 0.25° resolution, the NEX-GDDP-CMIP6 models cannot fully resolve Madagascar’s island climate — the rain-shadow effect of the eastern escarpment, the cyclone tracks, the windward / leeward asymmetry. We use the broader AFR-13 subset for Madagascar to avoid artificially narrowing the uncertainty when the underlying data is fundamentally limited by resolution. Read this as: for Madagascar, treat projection ranges with extra epistemic caution; consult dynamically-downscaled (CORDEX-Africa) products where available.

Figure 5. Which models are in which subset. Our default (AFR-13) is a column to compare against; regional subsets typically differ by 1–3 models from AFR-13. Original to this methodology (Adaptation Atlas team).

How this affects your reading

When you look at a future-projection chart for, say, Kenya, the underlying ensemble is not the same as for Ghana. The chart caption tells you which subset is active. If you’re writing a multi-country proposal and you want to enforce comparability, switch to AFR-13 (continental) or FULL-18 (NASA default) in the ensemble selector — your choice is remembered while you compare countries.

Why we exclude one model called CanESM5

Among the 18 NEX-GDDP-CMIP6 models, one of them — CanESM5, the Canadian Earth System Model version 5 — has a property that has prompted special treatment from the IPCC and from much of the climate-impact research community: it is very sensitive to greenhouse-gas forcing.

A model’s equilibrium climate sensitivity (ECS) is the warming it would eventually reach if atmospheric CO₂ were doubled and the climate were allowed to reach a new steady state. ECS is a fundamental property of how a climate model represents the strength of climate feedbacks (water vapour, clouds, ice-albedo, etc.).

The IPCC AR6 assessed the very likely range for the real climate’s ECS as 2.0 °C to 5.0 °C (Forster et al. 2021, AR6 WGI Chapter 7). The narrower likely range is 2.5 °C to 4.0 °C. These are the ranges supported by combining model evidence with observational constraints (temperature record, ocean heat content, paleoclimate data).

CanESM5’s ECS is 5.62 °C. That’s outside the AR6 very likely range — i.e. the IPCC’s assessment is that the real climate is very unlikely to be this sensitive.

Figure 3. The 18 NEX-GDDP-CMIP6 models' equilibrium climate sensitivities, with the AR6 very likely (2.0–5.0 °C) and likely (2.5–4.0 °C) ranges overlaid. CanESM5 sits outside the very-likely range — it is hotter than the IPCC assessment supports. The INM family (Russian) sits at the cold edge. Our AFR-13 default removes all three.

The implication, set out clearly in Hausfather et al. 2022 in Nature, is that climate-impact studies that include CanESM5 at equal weight with other models risk overstating the warming and the associated impacts — because the model is producing scenarios that the underlying physical-science evidence does not support. The same paper notes that high-sensitivity models “do a poor job of reproducing historical temperatures over time,” which is a separate problem: even setting aside the future, CanESM5 doesn’t match the past.

The IPCC AR6 itself handles this with what it calls constrained projections — combining the raw CMIP6 ensemble with observational constraints that effectively down-weight the hot models. The AFR-13 default is a simpler version of the same operation: rather than down-weighting CanESM5, we exclude it.

For symmetry, the AFR-13 default also drops two Russian models (INM-CM4-8 and INM-CM5-0), which sit at the very cold end of the ECS distribution (~2.0 °C). They have the opposite problem — they may under-project warming. Removing both extremes gives a more interpretable central projection.

How this affects your reading

If you’re using the AFR-13 default, your projections are slightly cooler than they would be with FULL-18 — typically by ~0.2–0.5 °C in end-of-century temperature anomaly, depending on the region. This is intentional — it brings the central estimate into closer alignment with the AR6 constrained projections. If you need to communicate the full uncertainty range (including the hot-model upper tail), switch to FULL-18 in the ensemble selector. Be honest about the trade-off in your proposal: AFR-13 is centred on the assessed-most-likely range; FULL-18 captures the entire model spread, including the tails the IPCC down-weights.

There is a respectable counter-argument — captured in a follow-up Nature commentary by Tebaldi et al. (2022) — that excluding hot models loses important information about the upper-tail uncertainty. We honour this argument by keeping FULL-18 always available as an alternative. The choice between AFR-13 (best central estimate) and FULL-18 (full range, including tails) is yours; we default to the central estimate but do not hide the alternative.

The East African Paradox — a known issue over Ethiopia, Kenya, Somalia

This is the most important honest caveat in this entire page.

Over the Horn of Africa — Ethiopia, Somalia, Kenya, Djibouti, Eritrea, parts of South Sudan, and northern Tanzania and Uganda — the observational rainfall record (CHIRPS, ERA5, and other gridded products) shows a multi-decadal drying trend in the long-rains season (March–April–May). At the same time, CMIP6 climate models project that the long rains should be getting wetter, not drier, under continued greenhouse-gas forcing.

This mismatch is known as the East African Paradox, and it is persistent: it appeared in CMIP3 (mid-2000s), it appeared in CMIP5 (early 2010s), and the most recent evaluation (Schwarzwald et al. 2024 in Journal of Climate) confirms it persists in CMIP6 as well.

Figure 4. The East African Paradox. Left: observed (CHIRPS) March–May rainfall trend over the Horn of Africa shows pronounced drying since the 1980s. Right: CMIP6 ensemble-mean projections show wetting. Adapted with permission from Schwarzwald et al. 2024 (placeholder — to be commissioned).

The current scientific understanding (Park et al. 2023 in Climate Dynamics) is that the bias originates in CMIP6’s representation of equatorial Pacific sea surface temperatures, which propagates through the global Walker Circulation to affect East African rainfall. It is a structural issue across the CMIP6 ensemble — selecting a “better” subset does not resolve it.

What this means in practice for any climate-rationale tool that uses CMIP6 over the Horn of Africa:

• The observational “recent changes” view is correct — the drying trend is real and visible. We are not the ones overstating it.
• The future-projection view for MAM (long rains) over the named countries should be treated with extra caution — the models project wetting, but the observed trend, mechanism studies, and recent crop / livelihood evidence are all consistent with continued or worsening drying.
• No CMIP6 subset (AFR-13, AFR-EAF, AFR-8, FULL-18) resolves the paradox. This is a known limitation of the underlying science, not a choice we have made.

Affected countries — the East African Paradox

The East African Paradox affects long-rains (MAM) projections specifically over:

Ethiopia · Somalia · Kenya · Djibouti · Eritrea · South Sudan · Northern Tanzania · Northern Uganda

For these countries’ MAM rainfall in a CMIP6 future-projection view, the direction of change may be wrong, not just the magnitude. The honest narrative for a climate rationale here is: “The observational record shows substantial drying; climate models project wetting; the scientific consensus is that the models are biased over this region; adaptation planning should account for continued drying as a credible scenario.”

This is, frankly, the most important sentence on this entire page for anyone writing a proposal for these countries.

For seasons other than MAM, and for variables other than rainfall (temperature projections, for instance, are not affected by the paradox), the future-projection view remains reliable in this region.

How this affects your reading

A short practical checklist for using these CMIP6-based projections in a climate rationale:

1. Note which ensemble is active. The chart caption tells you. For most users, the default (AFR-13 or the regional “Auto” mode) is the right choice.
2. Treat the ensemble mean as your central estimate; treat the ensemble spread as your uncertainty. Cite both — “+1.8 °C by 2050 under SSP2-4.5, with a model spread of ±0.6 °C” reads honestly.
3. For very-high-warming scenarios, switch to FULL-18 to include the hot-model upper tail. This is conservative — useful for stress-test framing in adaptation planning (“if we plan for X, but the upper-tail is Y, here’s our exposure”).
4. For East African MAM rainfall projections, always state the paradox caveat explicitly. Cite Schwarzwald et al. 2024 or Park et al. 2023. Frame your adaptation logic around the observational drying, not the modelled wetting.
5. For Madagascar and small-island states, treat the model spread as a floor on uncertainty, not a faithful representation of it. The 0.25° resolution underestimates the island-scale heterogeneity. Cross-reference CORDEX-Africa products where available.
6. For cross-country comparisons, use AFR-13 (continental) — comparing two countries under different regional subsets gets methodologically muddled.

CMIP7 and CORDEX-Africa — what’s coming

The methodology described above uses CMIP6 (specifically NEX-GDDP-CMIP6). Two next-generation efforts will reshape what’s possible for African projections over the next 1–3 years.

CMIP7 Fast Track

The Coupled Model Intercomparison Project Phase 7 (CMIP7) is the next generation of the global climate model archive that feeds IPCC AR7. Unlike CMIP6, which was structured around a large catalogue of experiments, CMIP7 introduces a Fast Track design — a tighter, more standardised set of core experiments intended to support AR7 directly (WCRP-CMIP CMIP7 Fast Track page).

Key things to watch for:

• Scenarios: The new ScenarioMIP-CMIP7 paper (Geoscientific Model Dev. 2026) defines an updated set of socioeconomic and emissions scenarios for the new generation. Expect changes in how the SSP-RCP scenario framework is presented.
• Timeline: CMIP7 simulations are expected to begin in mid-2026; first results will appear in peer-reviewed literature through 2026–2027. AR7 working-group deliverables are expected in the second half of 2026.
• Implications: as CMIP7 results stabilise, we will likely add a CMIP7 ensemble alongside the current CMIP6 view. The transition is unlikely to be a flag-day cut-over — both will coexist for a period to support proposal authors who cited CMIP6-based projections in earlier work.

CORDEX-Africa CMIP6 downscaling

CORDEX (the Coordinated Regional Climate Downscaling Experiment) is the WMO-WCRP initiative that produces dynamically-downscaled regional climate model runs at higher resolution than the global GCMs. For Africa, CORDEX-CORE provides a coordinated set of regional model simulations at ~0.22° resolution (about 22 km) — finer than NEX-GDDP-CMIP6’s statistical downscaling.

The CORDEX-CMIP6 archive is currently active and growing (CORDEX project page; CORDEX-CMIP6 archiving specs v3 released 2025/2026). Once mature, CORDEX-CMIP6 products will offer:

• Higher native resolution — meaningfully better for mountain effects, coastal dynamics, and small-country detail
• Dynamical (rather than statistical) downscaling — physically self-consistent fields, especially for extremes
• Coordinated multi-RCM ensembles per domain, supporting the same kind of regional-tuning approach we describe here

Implications: in the medium term, expect CORDEX-Africa products to be integrated alongside (or, for some use cases, in place of) the NEX-GDDP-CMIP6 baseline. The methodology described on this page — regional sub-ensembles, hot-model exclusion, paradox caveats — will translate broadly, with region-specific adjustments per the CORDEX evaluation literature.

Figure 6. Roadmap of climate-model generations and their expected availability for African adaptation planning. Sources: WCRP CMIP7 Fast Track; CORDEX.

Voices from regional centres

This section is open for contributions

We’re seeking 2–3 paragraph contributions from National Meteorological Services, regional research centres (ICPAC, ACMAD, SADC-CSC, AGRHYMET), and CGIAR partners with regional climate expertise — explaining how your institution uses CMIP6 for the region you cover, which models you favour, and where your perspective differs from the defaults we describe here. The goal of this section is to surface regional expertise that doesn’t always reach the global literature.

To contribute: click Edit this page in the footer to open the source markdown on GitHub, or contact the Climate Data Hub team via the address at the bottom of this page.

Placeholder — partner contributions will be added here as they arrive. Initial outreach in progress for Kenya Meteorological Department / ICPAC, South African Weather Service, Ethiopian National Meteorology Agency, Ghana Meteorological Agency, and Météo Madagascar. Each contribution is reviewed by the Climate Data Hub team and credited to the contributing institution.

Glossary

AR6 — the IPCC’s Sixth Assessment Report (Working Group I on the physical science was published 2021). The current reference cycle for climate-science assessments.

Climate sensitivity (ECS) — the warming a climate model reaches at equilibrium for a doubling of atmospheric CO₂. Used as a fundamental property of how strongly a model represents climate feedbacks. The real climate’s ECS is assessed by AR6 to be 2.5–4.0 °C (likely range) or 2.0–5.0 °C (very-likely range).

CMIP6 — Coupled Model Intercomparison Project Phase 6. The current generation of coordinated global climate model experiments that feed IPCC AR6.

CMIP7 — the next phase, in progress; simulations begin 2026.

CORDEX — Coordinated Regional Climate Downscaling Experiment. WCRP-led effort to produce regional climate model runs at higher resolution than the global GCMs.

Ensemble — a collection of climate models, all run under the same scenario, treated as samples from the space of plausible futures. The ensemble mean is the central estimate; the ensemble spread is the model-disagreement uncertainty.

GCM — General (or Global) Circulation Model. The mathematical models used to simulate the global climate system.

NEX-GDDP-CMIP6 — NASA’s downscaled product based on the CMIP6 archive. 18 models, statistically downscaled to 0.25° resolution. Our primary projection data source.

RCM — Regional Climate Model. A finer-resolution model that takes a global model as input and produces a higher-resolution regional simulation. The basis of CORDEX.

r1i1p1f1 — the standard “first physical ensemble member” of a CMIP6 model run. We use this realisation across all 18 models for consistency.

SSP scenario — Shared Socioeconomic Pathway. A future emissions and socioeconomic scenario; SSP2-4.5 is “middle-of-the-road”; SSP5-8.5 is “high-end fossil-fuelled development”; SSP1-2.6 is “sustainability”.

Walker Circulation — the equatorial Pacific east-west atmospheric circulation that connects sea-surface temperature patterns to remote rainfall. The mechanism implicated in the East African Paradox.

Further reading

If you read only six things to orient yourself before going deeper:

• Hausfather et al. 2022 — Recognize the ‘hot model’ problem — the Nature commentary that reshaped how CMIP6 ensembles are used for impacts.
• Iturbide et al. 2020 — AR6 reference regions — the regions our sub-ensembles are organised around.
• Samuel et al. 2025 — NEX-GDDP-CMIP6 evaluation for sub-Saharan Africa — the evaluation paper that anchors the regional sub-ensemble choices.
• Park et al. 2023 — CMIP6 Horn of Africa biases — the regional study behind the East African paradox caveat.
• IPCC AR6 Interactive Atlas — explore AR6 multi-model projections by region without ingesting raw data.
• WCRP CMIP7 Fast Track — what’s coming after CMIP6.

The full reference list — including regional evaluations, paradox literature, forward-look pointers, and data sources — follows below.

References

Foundation references

• Forster, P. et al. (2021). “The Earth’s Energy Budget, Climate Feedbacks, and Climate Sensitivity.” In AR6 WGI Chapter 7. https://www.ipcc.ch/report/ar6/wg1/chapter/7/
• Hausfather, Z., Marvel, K., Schmidt, G. A., Nielsen-Gammon, J. W., Zelinka, M. (2022). “Climate simulations: recognize the ‘hot model’ problem.” Nature 605, 26–29. https://doi.org/10.1038/d41586-022-01192-2
• Iturbide, M. et al. (2020). “An update of IPCC climate reference regions for subcontinental analysis of climate model data.” Earth System Science Data 12, 2959–2970. https://doi.org/10.5194/essd-12-2959-2020
• Samuel, S., Mengistu Tsidu, G., Dosio, A., Mphale, K. (2025). “Assessment of Historical and Future Mean and Extreme Precipitation Over Sub-Saharan Africa Using NEX-GDDP-CMIP6: Part I — Evaluation of Historical Simulation.” International Journal of Climatology. https://doi.org/10.1002/joc.8672
• Bloch-Johnson, J., Rugenstein, M., Gregory, J., Cael, B. B. & Andrews, T. (2022). “Climate impact assessments should not discount ‘hot’ models.” Nature 608, 667. https://doi.org/10.1038/d41586-022-02241-6

East African Paradox

• Park, S., Sniderman, J. M. K., Frierson, D. M. W., McKinnon, K. A. (2023). “Understanding CMIP6 biases in the representation of the Greater Horn of Africa long and short rains.” Climate Dynamics. https://doi.org/10.1007/s00382-022-06622-5
• Schwarzwald, K., Brönnimann, S., Vellinga, M., Schmid, M. (2024). “Revisiting the ‘East African Paradox’: CMIP6 Models Also Struggle to Reproduce Strong Observed Long Rain Drying Trends.” Journal of Climate 37(24). https://journals.ametsoc.org/view/journals/clim/37/24/JCLI-D-24-0225.1.xml
• Wainwright, C. M., Marsham, J. H., Black, E., Allan, R. P. (2019). “‘Eastern African Paradox’ rainfall decline due to shorter not less intense Long Rains.” npj Climate and Atmospheric Science 2, 34. https://doi.org/10.1038/s41612-019-0091-7

Regional evaluations

• Faye, A. & Akinsanola, A. A. (2022). “Evaluation of extreme precipitation indices over West Africa in CMIP6 models.” Climate Dynamics 58, 925–939. https://doi.org/10.1007/s00382-021-05942-2
• Makinde, A. A. et al. (2024). West African Westerly Jet representation. https://doi.org/10.1002/joc.70371
• Pinto, I., Jack, C., Hewitson, B. (2018). “Process-based model evaluation and projections over Southern Africa from CORDEX and CMIP5 models.” International Journal of Climatology 38(11), 4251–4264. https://doi.org/10.1002/joc.5666
• Ayugi, B. et al. (2021). East Africa CMIP6 future extremes. Water. https://doi.org/10.3390/w13172358

Forward-look

• WCRP-CMIP CMIP7 Fast Track. https://wcrp-cmip.org/cmip-phases/cmip7/fast-track/
• ScenarioMIP for CMIP7 (Geoscientific Model Development 2026). https://gmd.copernicus.org/articles/19/2627/2026/
• CORDEX project. https://cordex.org/
• CORDEX 2025 domain activities. https://cordex.org/domains/domain-activities-2/domain-activities-2025/

Data sources

• NASA NEX-GDDP-CMIP6 product page. https://www.nccs.nasa.gov/services/data-collections/land-based-products/nex-gddp-cmip6
• AR6 Interactive Atlas. https://interactive-atlas.ipcc.ch/
• CHIRPS (rainfall observational dataset). https://www.chc.ucsb.edu/data/chirps

---

Page metadata

Version: 0.1 (draft, 2026-05-28). This page will be refreshed (i) when CMIP7 results stabilise, (ii) when significant new African-evaluation papers appear, and (iii) when partner contributions to the regional perspectives section land. Old versions are retained for citation stability.

Origin: This methodology was first applied in the AAA Adaptation Atlas’s Build a Climate Rationale notebook and is now published as a Climate Data Hub reference for use across CGIAR’s African climate-rationale work.

Citing this page: Adaptation Atlas team (2026). “African CMIP6 Ensembling — Choosing Climate Models for Regional Adaptation Work.” CGIAR Climate Data Hub Wiki.

Contact: Pete Steward, Alliance Bioversity-CIAT, p.steward@cgiar.org