[4-6]

WRF-LETKF application for flood forecasting on Kyushu torrential rainfall

Tomoki Ushiyama, Takahiro Sayama, and Yoichi Iwami (Public Works Research Institute)

Severe rainfall often causes flood disasters in Japan, and its number is increasing nowadays. However, it is still hard to predict rainfall quantitatively by Numerical Weather Prediction (NWP) especially for small scale rainfall systems. Regional ensemble prediction system (REPS) would be a capable tool for this purpose. We examine the feasibility of REPS in rainfall prediction for flood forecasting that occurred in Kyushu, Japan, on July 2012, which caused a severe flood disaster by record breaking rainfall of 507 mm/day. We employed double nesting WRF-LETKF (Weather Research and Forecasting- Local Ensemble Transform Kalman Filter) with 15km/3km resolution. The predicted rainfall was applied into a distributed hydrological model for flood forecasting. Some of the ensemble members in 33 hour forecasts of 51-member REPS predicted similar rainfall pattern as observed, but their amounts were still underestimated and the locations were at several 10 km away. Hydrological simulations using this predicted rainfall could not reproduce the flood. We suppose one of the reasons of failure prediction is displacement of the rainfall area. Then, we displaced rainfall area to be the maximum rainfall in the river basin and applied it into the hydrological model. Some of the members eventually forecasted flood possibility.