Sorghum yield prediction from seasonal rainfall forecasts in Burkina Faso

The high variability of rainfall, from interannual to multi-decadal time scales, has serious impacts on food security in the West African Sahel. At five locations in Burkina Faso, we explore the potential to improve model-based prediction of sorghum yields at a range of lead-times by incorporating seasonal rainfall forecasts. Analyses considered empirical and dynamic rainfall forecasts, two methods (regression and stochastic disaggregation) for linking rainfall forecasts with crop simulation, three levels of production technology and four forecast dates (15 May, June, July and August) based on predictors observed from the preceding month, for the period of available data (1957–1998). Accuracy of yield forecasts generally decreased with lead-time. Relative to forecasts based solely on monitored weather and historic climatology, incorporating rainfall forecasts resulted in modest improvements to yield forecasts made in May or June. The benefit from seasonal rainfall forecasts tended to increase with northern latitude. Statistical and dynamic rainfall forecast systems captured much of the multi-decadal variation apparent in historic rainfall and in yields simulated with observed rainfall. This multi-decadal component of rainfall variability accounts for a portion of the apparent predictability of sorghum yields. Correlation between point-scale crop yield simulations and district-scale production statistics (1984–1998) was weakly positive late in the season, and suggest that a dynamic crop model (SARRA-H) has potential to contribute to regional yield prediction beyond what the best linear regression can provide from seasonal rainfall or its predictors. We discuss avenues for further improving crop yield forecasts during the growing season.