Ethiopia is praying for rain according to a recent report from the Guardian, and ReliefWeb suggests that a lack of rain may be linked to malnutrition in Tchad, as well as reduced crops in Niger. 

The African rainfall deficit appears to be widespread: the Cairo Review reports severe droughts that have been experienced across the Sahel, the Horn of Africa, and Southern Africa in 2011 and 2020. But the picture is also more complicated, as heavy rains have unleashed massive flooding across South Sudan according to the Red Cross.

A similar ambiguity can also be seen in the future prospects for this region. The recent Intergovernmental Panel on Climate Change (IPCC) report (Assessment Report 6, often referred to as ‘AR6’) presents maps showing projected changes in the precipitation, e.g. in Figure SPM.5 and the IPCC Atlas. 

The most recent precipitation projections reveal a remarkable dark green blob covering parts of Sahel and Sahara in addition to the Arabian peninsula, suggesting that this dry region may be blessed with more rainfall in the future (e.g. Figure 1). 

Figure 1. A map from Figure SPM.5 from IPCC AR6 slides, showing percentage change in annual mean precipitation from a historical baseline (1850-1900). These results represent the Coupled Model Intercomparison Project phase 6 (also known as ‘CMIP6’) Global Climate Models (GCMs). 

However, increasing rainfall amounts may indicate a break with previous analysis for the region comprising Sahel and southern Sahara, as historical observations suggests diminishing rather than increasing rainfall in the same region (e.g. NOAA NCEI and CICS-NC Fig 1.7). It also differs from similar projections based on the previous CMIP5 generation of GCMs and the CORDEX-Africa results (both of which are available through the IPCC AR6 Atlas). 

One problem with Sahel, Sahara and the Arabian Peninsula is a lack of observations, a fact that is acknowledged in Figure SPM.3 showing how agricultural and ecological drought tendencies have changed since 1950. Lacking ground data makes it harder to evaluate model simulations and calibrate the projections.

So how to interpret this information? We don’t have much local data to examine, but there is global data, such as the ERA5 reanalyses, that nevertheless may provide some useful analysis, even if it doesn’t reflect the African rainfall with as high an accuracy as we would like. 

ERA5 and rainfall trends 1950-2020. 

The wet African blob in the projections presented in Figure 1 indicates proportional changes in annual rainfall amount over a region where it hardly rains at all. We are talking about one day of rainfall per year or even less, according to the ERA5 reanalysis (see figure below). 

Figure 2. The annual average number of rainy days is of the order of 1 rainy day or less per year over parts of Sahara according to the ERA5 reanalysis (threshold 1 mm/day). 

A 50-year baseline for such a dry region may contain about 50 rainy days or less. Each rainy day is subject to weather’s randomness, and if the number of cases is small enough then they also are insufficient for providing reliable statistics. 

In other words, a 50-year baseline may be unstable if the rain is so rare that a few downpour events can alter the statistics. This is not a problem for regions where it rains frequently, however. 

It’s also tricky to combine different trend estimates from models with different biases and tendencies (models are not perfect). If we compare the proportional trend in rainfall amount from ERA5 and the trend in terms of mm/year, as in Figure 3, then it is clear that regions with very low baselines easily can have high proportional trends.

Figure 3. Trend estimates of the rainfall amount over Africa in terms of percentage change per decade (upper) and mm/year per decade (lower). 

We can estimate the proportional change for each simulation, and then take the mean to estimate the ensemble mean. In this case, models with a lower baseline in regions with really rare rainfall events may carry more weight than those which happened to simulate more random rainfall events in the same period.

It’s also possible to take the average of the trends from each model in terms of mm/year per decade, and then estimate the proportional change based on ensemble mean. Models that overestimate rainfall will then carry more weight. 

Replication of the projected precipitation trends

We can try to replicate the results presented in the IPCC AR6 projections to affirm its conclusion. Below is an ensemble mean trend map for simulated trends over the period 1950-2020 that has a similar wet blob over Sahel, Sahara and the Arabian Peninsula. This analysis was based on 37 different CMIP6 models and a larger ensemble than the IPCC Atlas, which used 33 different models.

Figure 4. A 37 CMIP6 model ensemble mean of the proportional trend for annual rainfall amount simulated over 1950-2020.

The ensemble mean of the proportional trend estimates of the selected CMIP6 simulations in the replicated analysis shows a similar pattern as reported in the IPCC AR6. Hence, my replication also shows that the simulated historical trends for the Sahel and southern Sahara region indicate increasingly wetter conditions with a similar spatial pattern as Figure 1, but with opposite trends to those found in the ERA5 reanalysis shown in Figure 3.

Does the discrepancy between CMIP6 simulations and the ERA5 reanalysis really mean that the models have shortcomings? Not necessarily for really dry regions, such as the Sahara and Sahel, because it only rains very rarely and the rainfall statistics may not be robust over the driest regions.

The ensemble mean is of course estimated from a number of different GCM simulations, all of which are presented in the animated GIF below:

Figure 5. Animated maps of the individual ensemble members that form the basis of the ensemble mean shown in Figure 4.

When we check the trend map of each GCM, it is apparent that they show a wide range of different trend patterns. We see that few of the individual simulations resemble the ensemble mean.  One of them even resembles the ERA5 results. 

So I would argue that the solid wet blob in Figure 1 potentially gives a misleading impression. But it is possible that the future may bring a few more heavy downpour events with even greater amounts of rainfall, increasing the risk of flooding, and that such a tendency wasn’t captured in the CMIP5 and CORDEX-Africa simulations. 

A small increase in the number of rainy days and their duration can give an exaggerated impression of a change in exceptionally dry regions such as Sahara and Sahel. It would perhaps be useful to look at other statistics such as the average number of rainy days per year (or the wet-day frequency) and the mean rainfall intensity, because they may provide more robust results. 

The IPCC Atlas does, however, provide information about the annual maximum one-day and five-day rainfall (the rains in this dry region don’t last many days) as well as consecutive number of dry days, which all are projected to increase in this dry region. But even if they or the annual rainfall were to double throughout this century, Sahara and Sahel will still be very dry.

Another question is whether the ERA5 reanalysis provides a representative picture of past trends. 

In other words, I wouldn’t bank on Sahel and southern Sahara being blessed with more steady rain in the future. We also need to dig deeper into the question whether the simulated change in rainfall over Africa is real and has a physical explanation.