Detection and Prediction of the Rainy Season Onset in West Africa

Abstract

The onset of the rainy season is crucial for rain-fed farming in tropical countries. The timing of the onset can drastically influence the performance of the growing season, which subsequently impacts the food security of rural communities. Despite the high societal importance, the availability of operational forecast of the onset is limited, especially across the African continent. While there is little ambiguity about the overall behaviour of the West African Monsoon (WAM), there are several man-made definitions of its onset. Those definitions can address the onset both regionally and locally, the latter ones being the most commonly used. In this study we tackle the operational prediction of the WAM onset, focusing on Ghana and neighbouring countries. The regional approach is the main focus of both the climatological research and the development of the forecasting algorithm. However, the local perspective has been analysed too, representing the common practise for onset’s forecasting. Firstly, the correlation between the onset and atmospheric variables is investi- gated in the period 1981-2021, using satellite-based rainfall records (CHIRPS dataset) and ERA5 re-analysed wind fields at different pressure levels. Afterwards, the operational prediction of the onset is produced by post-processing operational weather forecasts, issued by the European Center for Medium-range Weather Forecasts (ECMWF). It is found that ECMWF rainfall forecasts (both medium-range and sub-seasonal) are the necessary ingredients of any prediction of the onset. Other atmospheric variables, such as wind fields at 925, 850, and 200 hPa, revealed to bring little to no advantage for the operational forecast of the onset, despite showing a clear correlation with the WAM in the climate. The best forecast performances are obtained with a threshold- based algorithm, which detects the onset imposing conditions on the amount and the temporal distribution of rainfall. Both regional and local onset’s forecast display promising performances in specific areas of the analysed region. Those areas are characterised by high spatial coherence of the rainfall pattern and can also be influenced by the adopted onset’s definition. We conclude that the operational forecast of the onset is feasible with both the regional and the local approach, but only for a portion (about 47%) of the analysed domain. However, predicting the onset is far from straightforward and the obtained forecast are dependent on several semi-arbitrary choices, first and foremost on the chosen onset’s definition. Due to this dependency, it is of utmost importance to include the end-user needs in the development of any rainy season onset’s forecast.