Rainfall and streamflow forecasting for operational water management in the Incomati Basin, Southern Africa

When future water situation is unclear to water managers and dam operators, problems on water allocation among competing users may arise due to unforeseen water shortages. Thus, rainfall and streamflow forecasting may help to manage the uncertainty. Forecasting estimates how much water is likely to fall as rainfall and flow passing a specific river location during a particular time based on recent meteorological and catchment characteristics. Incomati River Basin in Southern African is almost in a closing status; thus rainfall and streamflow forecasting are significant and relevant. The main objective of this research was to study the scope of rainfall and streamflow forecasting in the Incomati Basin by reviewing forecasting methods, by examining the needs and usefulness of forecasting, and by testing promising methods for rainfall and streamflow forecasting. The study used statistical methods of correlation and regression analyses. Also to explore the needs and uses, interviews and discussions with water managers, dam operators, water users and other stakeholders were carried out. Data used include Sea Surface Temperature (SST), El Niño Southern Oscillation (ENSO), air temperature, precipitation and streamflow. The study was inthree stages; field survey and interviews, correlation and regression analyses, and rainfall and streamflow forecasting. The correlation analysis covered selected climatic and streamflow stations across whole Incomati Basin. However, models for rainfall and streamflow forecasting were tested for the Kaap sub-catchment located in the Crocodile catchment. Forecasting equations per method per time were generated. The forecasts are presented in terciles of aboveaverage, average and below average. The forecasts were verified using standard forecast verification scores such as probability of detection (POD), probability of false detection (POFD), false alarm rate (FAR) and accuracy. The survey results showed that about 97% of water stakeholders need and use rainfall forecasts. The major uses were personal plans like travelling (29%) and dressing (23%). The usefulness in water sector was reported for water allocation (23%), farming (11%) and flood monitoring (9%).On the other hand only 5% showed the need for streamflow forecasting. The rainfall forecastswere presented to have medium to high benefits. Generally, users affirmed the accuracy and benefits of weather forecasts and had no major concerns on the impacts of wrong forecasts. However, respondents indicated the need to improve the accuracy and accessibility of the forecast. Likewise, water managers expressed the need for both rainfall and flow forecasts but indicated that they face hindrances due to financial and human resource constraints. This shows that there is a need to strengthen water related forecasts and the consequent uses in the basin. For effective water management, forecasts should be user specific and the collaboration between meteorology and hydrology is encouraged. The results of the correlation analysis indicated that, there is a significant correlation between SST, ENSO and rainfall and streamflow in magnitude and in direction. This correlation can be found at the monthly, seasonal and annual scale in one-one level and in lag times. The correlation of ENSO to rainfall over the basin (0.207), as well as ENSO to streamflow in the basin (0.256) was found to be positive (significant at 95% level), which means the increase of ENSO can influence the increase of rainfall and streamflow and vice versa. The SST to rainfall (-0.294), and SST to streamflow (-0.239) correlation was negative (also significant at 95% level) denoting the increase of SST may influence the decrease of rainfall and streamflow and vice versa. There was a significant positive correlation between rainfall and streamflow (0.487 significant at 99% level).Rainfall forecasting was done by using six methods at monthly and three methods at seasonal time resolutions. The ENSO and SST are found to be significant predictors. The monthly rainfall was best forecasted by methods using the previous season's SST and rainfall; and previous month's SST, ENSO and P; Pt= f (SSTts-1, Pts-1) and Pt= f (SSTt-1, ENSOt-1, Pt-1). In seasonal,both ENSO and SST worked equally well; Pts= f (ENSOts-1, Pts-1) and Pts= f (SSTts-1, Pts-1). Monthly streamflow was forecasted by eight methods. The best forecast was found using theprevious month's SST, ENSO, P and Q; Qt= f (SSTt-1, ENSOt-1, Pt-1, Qt-1). Five methods forecasted seasonal streamflow, the best were found to be similar to that of monthly model; Qts= f (SSTts-1, ENSOts-1, Pts-1, Qts-1) and Qt= f (SSTts-1, Pts-1, Qts-1). In general, the tested methods, including the best ones, indicated higher skills (better forecasts) in predicting rainfall and flow in the upper and lower terciles than in the average tercile. The best forecasts for rainfall were found in the wet season while streamflow forecasts were best in low flow periods. The forecasts showed good quality and have value for the users in the studied sub-catchment. The forecastmethods developed could be recommended for operational application across the Incomati Basin, after careful verification for other sub-catchments