Improving ESP Probabilistic Forecasts Using Persistence Information

The purpose of this study was to improve the Ensemble Streamflow Prediction (ESP) probabilistic forecasts by combining them with streamflow persistence information. ESP was applied to one-month ahead inflow forecasting for the Chungju multipurpose dam in Korea. This ESP system ran a rainfall-runoff model called TANK in the beginning of a forecast month on a daily basis. The historical rainfall data for the 36-year period from 1966 to 2001 were employed as the meteorological input in order to generate the corresponding 36 inflow scenarios each month. To issue probabilistic forecasts, the range of the generated inflow scenarios was divided into three flow categories and a probability was assigned to each flow category from the fitted probability density function. Using the optimal linear combination (OLC) approach, the ESP forecast was then combined with the persistence and naive forecasts. The persistence forecast represents a conditional probability of a flow category, provided that the inflow category of the previous month is given. The naive forecast assigns a 33.3% probability to each flow category. Split sampling simulation proved that OLC improves the accuracy of ESP forecasting for the non-flood season.