Revisiting the ‘East African Paradox’: CMIP6 models also fail to simulate observed drying trends in the Horn of Africa Long Rains

The recent 5-season drought in the Horn of Africa, which contributed to food security issues that nearly resulted in a declaration of famine by the UN, has renewed interest in the “East African Paradox” (cf. Rowell et al., 2015): despite observed drying trends in the March-April-May “long” rains, global coupled climate models—whose output is increasingly used to drive hydrological models and inform projections of the socioeconomic risks of climate change in East Africa—project increases in seasonal rainfall totals over both the historical period and throughout future projections in the region. This ‘Paradox’ could arise from low-frequency internal variability causing drying even if long-term trends are wetting or from structural biases in climate models (e.g. simulation of the equatorial Pacific Ocean) that cause spurious trends in model simulations. Large Ensembles, including for SST-forced runs, make differentiating between internal variability and biases in model mean behavior more feasible, and another decade of observational data since the emergence of the ‘Paradox’ helps improve our understanding of historic internal variability.

We use a large multi-model ensemble of opportunity of coupled and SST-forced runs from the latest model generation (CMIP6), spanning the observational record, to revisit the magnitude and causes of the ‘Paradox’. We find that drying trends in the long rains are timescale-dependent and weaker than they were during the peak ‘Paradox’ period. This is mostly well modeled by the SST-forced ensemble, though coupled models continue to have erroneously strong wetting trends. The ‘Paradox’ therefore is reduced to what are the causes of low frequency SST trends and why coupled models cannot reproduce them.  We will discuss if these changes are the result of natural variability temporally masking the forced trend and what the sign of that trend might be.  These results have implications for projections of future climate impacts with a potentially quantifiable range of internal variability providing more actionable information than the deep uncertainty on forced trends introduced by structural model errors.