Seasonal precipitation and streamflow forecasting can dramatically reduce vulnerability to climate variability in African Sahel. The region experiences high rainfall variability and a skilful forecast can help populations prepare for above-normal or below-normal rainy seasons. This study is a part of a larger project that aims to develop a streamflow forecasting framework for the Sirba watershed located in Niger and Burkina Faso, West Africa. The first step of the larger project consists of predicting seasonal precipitation on the watershed several months ahead using Sea Surface Temperature (SST) from the Pacific and Atlantic Oceans. The second step (not yet completed) will consist of transforming the forecasted precipitation into streamflow. The July-September precipitation was linked to gridded SST data sets from the Pacific and Atlantic Oceans. Several techniques were used to reduce the dimensionality of the predictors: a) discarding grid points for which the correlation with the predictand is below a certain threshold; b) building a linear model with a grid point SST and discarding it by considering the Nash Sutcliffe efficiency coefficient; c) using principal component analysis or canonical correlation analysis to obtain a small number of meaningful predictors. The prediction skills of the retained predictors are checked using a leave one out cross validation procedure. Results shows that the March-May Pacific Ocean SST are the best predictor for the coming rainy season (r~2=0.378, E_f=0.312), followed by the October-December SST from the Atlantic Ocean (r~2=0.254, E_f = 0.237).
展开▼