Variation in Sahel Rainfall and Drought since 1950: Nature and Predictability

The Sahel is one of the most drought-prone regions in the world. This paper focuses on climate sources of drought, and addresses interannual variability as well as the substantial decadal-to-multidecadal variability (MDV) that is a prominent feature of the region's climate. The Sahel region of West Africa is a transition zone between equatorial climates and landscape to the south, and desert to the north. The climatology of the region is dominated by dry conditions for most of the year, with a single peak in rainfall during boreal summer. We analyze a range of station and merged station/satellite precipitation datasets, as well as reanalysis products, sea surface temperature (SST) data, and General Circulation Model (GCM) output to investigate and illustrate aspects of rainfall and drought variation and predictability. The decline in rainfall from the wet decades of the 1950s and 60s to the dry decades of the 1970s and 80s has been well documented. In recent decades, a moderate recovery has emerged, with average seasonal totals during 1994-2010 significantly higher than those during 1970-1993. These fluctuations have expression in large-scale SST fluctuations in all ocean basins, primarily contrasting the South Atlantic / Indian Ocean versus the North Atlantic, including a connection to the Atlantic Multidecadal Oscillation, with cold phase favoring drier conditions in the Sahel.

We have evaluated the changing character of low seasonal rainfall total event frequencies in the Sahel region 1950-2010, for the three periods: 1994-2010 (partial recovery, period 3), 1950-1969 (period 1, very wet), 1970-1993 (period 2, very dry). Sensitivity is assessed for different choices of years and sub-regions (over the domain 18W to 30E, 10N to 20N). Changes in such event frequencies serve as a proxy for many impacts on society, and are especially relevant to the feasibility of the emerging drought management tool of index insurance. A change in event frequency impacts directly payout frequency. We quantify changes in event frequencies over time, highlighting the role of changes in mean, variance and distribution shape. We discuss how updating climate normals in real-time can damp the bias in event frequency outcomes (when estimating the risk of an event in year i, based on data prior to year i). Such information has many risk management applications, as well as to support the feasibility of index insurance in the presence of a changing climate. At small scales (station or small-scale gridded data), we highlight how seasonal rainfall totals in the Sahel tend to have a systematic positive skew, and that application of the skew-normal distribution is a promising approach to specify/predict the impact of MDV on moderate to severe (such as 1 in 20 year) drought event frequencies. These findings also have potential implication and application for downscaling of seasonal rainfall totals in general in the Sahel.

The specification of the statistical characteristics of natural climate variation (as discussed above, interannual variability and MDV) can be used to define a stochastic statistical simulation model that can generate multiple realizations of climate, to assess the risk of drought frequency changes under specified scenarios (as discussed in Siebert and Ward, 2011, J. Appl. Meteor. Climatol., 50, 560-578). This may provide complementary information to that achievable from other sources of information on the MDV timescale (e.g., monitoring the current observed state of key MDV SST patterns, decadal prediction from GCMs and global change scenarios from GCMs). On the interannual timescale, distinct SST patterns (mostly tropical Pacific and tropical Atlantic) also drive a substantial fraction of the rainfall variation, bringing a degree of predictability to interannual drought variation. This is seen in both empirical prediction models using observed SST, and in GCM predictions. For the latter, the application of a model output statistics approach using the GCM's regional low-level circulation is often the most effective way to generate prediction information about Sahel rainfall.

The character of the rainfall changes through the three epoch phases (from wet to dry and returning to relatively wet), though not symmetric in all aspects, does nonetheless show some significant symmetric features. A recent finding is that the percentage change in rainfall tends to be smallest at the beginning of the rainy season, gradually increasing to a maximum in October. This is true for both the decline in rainfall (from period 1 to period 2) and for the increase in rainfall (from period 2 to period 3). This may point to a role of regional land-surface processes, amplifying the expression of SST-forcing as the season progresses from June to October. We therefore explore how the low-frequency MDV fluctuations in rainfall have particularly strong expression and impacts at the end of the rainy season, with implications for management strategies in general for the low-frequency (MDV) drought fluctuations in the region.