Model Output Statistics to improve severe storms prediction over Western Sahel

Abstract

The recent increasing trend in the severity of floods, especially across West Africa has been due largely to severe storm outbreaks which are complex and difficult to predict. Most of the weather forecasting centers in Africa now use numerical weather prediction (NWP) model outputs to predict the occurrence and severity of these storms. However, several studies have shown that NWP models and their forecasts are subject to errors and biases because of the complex atmospheric uncertainties and the currently limited knowledge of the mathematical formulation of the atmospheric physics and dynamics. The incorporation of statistical techniques is therefore useful and has indeed become a necessary component of improving NWP model products. Model Output Statistics (MOS), for example, use large multiple regression equations to provide a statistical relationship between the forecast output of NWP models and the observed variables. This study investigates the implementation of the MOS equations generated for Western Sahel (WS), required to correct forecast biases and errors from the 20 km × 20 km resolution Limited Area Model over Africa (Africa LAM) developed by the United Kingdom Meteorological Office (UK Met Office). Daily observed rainfall from January 2005 to December 2006, of 36 selected meteorological stations regionally distributed across the WS as well as the T + 24 h rainfall forecasts from the Africa LAM over the same period were retrieved and analyzed. Results indicated about 76% improvement to the original Africa LAM rainfall forecasts by the MOS method over the southern region of the WS during July–August–September—the period when severe storm activities are highly probable. Results also showed a consistently smaller root mean squared error (RMSE) values from the MOS-corrected rainfall forecasts when compared with the RMSEs from the original Africa LAM rainfall forecasts for all the seasons and regions of the WS.