Application of time-lagged ensemble approach with auto-regressive processors to reduce uncertainties in peak discharge and timing

Abstract

In spite of the popularity of ensemble streamflow predictions (ESP), data necessary to produce these predictions are still limited in many countries. This study developed an ensemble flood forecasting methodology for an urban catchment in Korea where only deterministic forecasts of three-hour rainfall accumulations are provided twice a day. The forecasted ensembles of rainfall were created using a time-lagged ensemble approach and used in a rainfall-runoff model to generate ensemble flood forecasts. An auto-regressive (AR) processor was then applied to the input and output ensembles to improve the accuracy of the flood forecasts. The Jungrang catchment was selected to evaluate the performance of the proposed methodology. This is one of the urban areas vulnerable to flooding. From 1991 to 2010, there were 154 fatalities and 5,278 injured victims resulting from flooding. An accuracy evaluation using observations from 2002 to 2009 found that the time-lagged ensemble approach alone produced significant bias but the AR processor reduced the relative error percentage of the peak discharge from 60% to 10% and also decreased the peak timing error from more than 10 h to less than 3 h, on average. The proposed methodology is easy and inexpensive to implement with the existing products and models and thus can be immediately activated until a new product for forecasted meteorological ensembles is officially issued in Korea.

Keywords

• Auto-regressive processor
• Error correction
• Flood forecasting
• Korea
• Time-lagged ensemble