Investigating the performance of high-resolution subseasonal precipitation forecasts in support of food insecurity early warning

Abstract Anticipating precipitation extremes across sub-Saharan Africa can help mobilize interventions, trigger anticipatory actions, and promote beneficial actions like water harvesting. Reliable crop model forecasts can help identify when and where food aid interventions can be most beneficial. To date, however, there has been little research evaluating the utility of rainfall forecasts. This study, therefore, assesses the efficacy of the Subseasonal Consortium database (SubC) for use in a regional crop water balance model—the Water Requirement Satisfaction Index (WRSI)—in east Africa. We find that combining two dekads (20 days) of statistically downscaled and bias-corrected SubC precipitation data with climatological information delivers improved estimates of end-of-season conditions over a 17-year test period.
Our results show that SubC forecasts provide a 35-55% reduction in end-of-season WRSI Root Mean Squared Error in 60% of the east African agropastoral areas during the short rains, with the highest accuracy being in areas most vulnerable to inconsistent precipitation timing and quantities. Across the 17 tested seasons, 1999/00 to 2015/16, use of the SubC either improved or did not degrade the accuracy of WRSI prediction compared to a benchmark model for over 70% of the seasons and for 90% of the study region. In general, the improved accuracy provided by two dekads of SubC forecast is nearly equivalent to what can be attained with one dekad of “perfect” forecast (i.e., observation data). In effect, a 20 day forecast provides a 10-day advance in our early warning capabilities. During extreme events, such the 2005/2006 drought in east Africa, the SubC-driven WRSI could provide advanced warning of poor cropping conditions and potential crop failure up to 3 months before the end of the season. Overall, these improvements provide earlier and more accurate estimates of likely seasonal water balance outcomes, and allow for the identification of where interventions may be needed.