ABSTRACT

For facing the increasing occurrence of extreme droughts in the future, the long-term reservoir inflow forecasting is an essential tool for the authorities of water resources management to estimate the reservoir water storage for 1 ~ 6 months ahead and support to decide the suitable drought relief measures when the predicted water storage is insufficient. Based on the Memorandum of Understanding on Cooperation between the Water Resources Agency (WRA) and the Central Weather Administration (CWA), the CWA uses the raw forecasting data of ECMWF SEAS5 to produce the downscaled and corrected monthly rainfall forecasts for 1 ~ 6 months ahead for the major reservoir catchments in Taiwan. These downscaled and corrected monthly rainfall forecasts are then provided to the WRA for executing reservoir inflow forecasting. The monthly rainfall forecasts are further converted to daily rainfall forecasts by the k-nearest neighbors algorithm, which are then input into the modified HBV hydrological model to simulate the inflows for each reservoir. The study uses the historical downscaled and corrected rainfall forecasts to conduct 1 ~ 6 monthsahead flow forecasting once each month and evaluates the flow forecasting accuracy for Shimen Reservoir, Deji Reservoir, Zengwen Reservoir, and Gaoping Creek Weir. The analysis results show that using the downscaled and corrected rainfall forecasting data of ECMWF SEAS5 for reservoir inflow forecasting has higher forecasting performance than the traditional method adopting Q80 and Q90 (the 80th and 90th percentiles of the flow duration curve, respectively) as the forecasts.

Keywords：ECMWF SEAS5, HBV hydrological model, Seasonal rainfall forecasting, Inflow forecasting