Ensemble flood simulation for a small dam catchment in Japan using nonhydrostatic model rainfalls – Part 2: Flood forecasting using 1600-member 4D-EnVar-predicted rainfalls

Abstract

Abstract. This paper is a continuation of the authors' previous paper (Part 1) on the feasibility of ensemble flood forecasting for a small dam catchment (Kasahori dam; approx. 70 km2) in Niigata, Japan, using a distributed rainfall–runoff model and rainfall ensemble forecasts. The ensemble forecasts were given by an advanced four-dimensional, variational-ensemble assimilation system using the Japan Meteorological Agency nonhydrostatic model (4D-EnVar-NHM). A noteworthy feature of this system was the use of a very large number of ensemble members (1600), which yielded a significant improvement in the rainfall forecast compared to Part 1. The ensemble flood forecasting using the 1600 rainfalls succeeded in indicating the necessity of emergency flood operation with the occurrence probability and enough lead time (e.g., 12 h) with regard to an extreme event. A new method for dynamical selection of the best ensemble member based on the Bayesian reasoning with different evaluation periods is proposed. As the result, it is recognized that the selection based on Nash–Sutcliffe efficiency (NSE) does not provide an exact discharge forecast with several hours lead time, but it can provide some trend in the near future.