Causes of Uncertainty in Projections of East African Rainfall Change

Large uncertainties remain in CMIP5 projections of rainfall change over East Africa, ranging from a modest decline to substantial increases over the coming decades. Similar uncertainties are apparent elsewhere on the African continent, and indeed in other tropical regions, and represent a significant problem for current strategic planning decisions that aim to develop systems that will be resilient to future climate.

A critical question in climate science is whether this model projection uncertainty can be understood, and therefore potentially reduced?

We first define sub-regions of East Africa, for each of the Long Rains and Short Rains seasons, for which projected rainfall anomalies are spatially coherent within individual climate models. Importantly, these differ from regions for which interannual anomalies are spatially coherent, demonstrating a different balance of climate mechanisms between the different time scales.

We then combine and compare a number of experimental and analysis techniques to begin to understand which mechanisms most differ between models and therefore drive projection uncertainty. These are multi-model ensembles of (a) coupled model experiments with an abrupt CO2 increase, (b) atmosphere-only experiments with uniform and pattern SST increases, (c) RCP projections to which we apply a technique to decompose rainfall into two dynamic and two thermodynamic components, and (d) CORDEX regional model projections using a novel separation of variance components.

For the Short Rains, this shows that modelling uncertainty in projected seasonal rainfall change is primarily due to uncertainties in the regional response to both the uniform and pattern components of SST warming (but not uncertainties in the warming itself), and a minimal direct CO2 impact. These manifest as spatial shifts in convection driven by uncertain regional dynamical mechanisms, rather than planetary-scale (dynamic and thermodynamic) processes, and geographically reflect uncertainties in both African and remote mechanisms.

For the Long Rains, results are similar, except that the direct CO2 impact is further reduced, and key regional uncertainties are primarily located beyond Africa, and then communicated to eastern Africa via uncertain changes in lower tropospheric moisture content. Further analysis shows this latter result derives from the behaviour of just two outlying models, which must be further understood to enable an assessment of their reliability. Consequent implications for stakeholders will be that either uncertainty can be significantly reduced by removal of implausible outliers, or that these outliers may still need to be considered as a severe worst case scenario.