Evaluation of multi-model rainfall forecasts for the national 7-day ensemble streamflow forecasting service

The Bureau of Meteorology is planning to launch its upgraded national 7-day ensemble streamflow forecast service in mid-2019. This service will fulfil the growing demand for ensemble streamflow forecasting and bring a greater level of accuracy and reliability when compared to the existing 'deterministic service'. It will enable end-users to make more confident water management decisions by considering the likelihood of different streamflow forecast outcomes within a risk assessment framework. Ensemble streamflow forecasts are generated using post-processed multi-model Numerical Weather Prediction (NWP) ensemble rainfall forecasts. These streamflow forecasts are subsequently post-processed using a comprehensive multi-stage error correction model. The forecasting system evaluation was undertaken by comparing rainfall and streamflow forecast performance using four different NWP rainfall products for 26 catchments located in various hydro-climatic regions across Australia. It reveals that the post-processing reduces bias and improves the reliability of rainfall forecasts as well as the corresponding streamflow forecasts. Streamflow post-processing further improves the accuracy and reliability of forecasts significantly at shorter lead-times and the impact declines with the lead-time. Overall, the use of rainfall forecasts of the European Centre for Medium-Range Weather Forecasts (ECMWF) presents relatively better accuracy and reliability of streamflow forecasts at the catchment scale in many occasions compared to the other three products evaluated in this study.