Verification of rainfall forecasts for the Vaal Dam catchment for the summer rainfall seasons of 1994 to 1998

Abstract

Rainfall forecasts compiled by the South African Weather Service (SAWS) are used daily by agriculture, industry, sportsmen and the general public. Because of the importance of the rainfall forecast, it is of considerable interest to know how reliable these forecasts are. The SAWS evaluates the rainfall forecasts issued by the Central Forecasting Office (CFO) on a daily basis. A hit score is determined in each of 19 rainfall districts throughout South Africa. This renders a result of approximately 60% correct forecasts for all 19 districts. This paper investigates the accuracy of the predicted rainfall percentages or rainfall classes. The 24h rainfall forecasts are verified for the summer months (October to March) of 1994 to 1998 for the Vaal Dam catchment. The rainfall distribution classes were chosen to represent no-rain, isolated rain, scattered rain and widespread rain. Results indicate that the percentage of correct forecasts is less than the values obtained by using the SAWS's hit score method. The predicted widespread rain class is less than 25% correct. The tendency exists to forecast scattered rain whenever any rain event is predicted. A rain event is seldom missed (3%), but some rain is often forecast (45%) for days, which then remain dry days. A more comprehensive rainfall forecast evaluation system is recommended.


Keywords: Rainfall forecasts, Verification, Rainfall distribution classes, Vaal Dam catchment


(WaterSA: 2003 29(2): 195-200)