ABSTRACT
Inundation warning systems are a basic and essential tool of the non-structural flood mitigation measures in the western Pacific Rim. However, the multiple and complex disaster factors in inundation prediction and early warning system are a significant challenge for task of accurately forecasting real-time inundation warning. Emergency operations authorities are currently limited in issuing timely and effective inundations warnings based upon the timeliness and prediction capabilities of early warning information. An efficient and accurate river stage-inundation warning model has been developed for more efficiently and accurately predicting the flood and inundation information to enhance decisionmaking and reduce flood damage. This study incorporates the QPESUMS (Quantitative Precipitation Estimation and Segregation Using Multiple Sensors) technique and RNN (Recurrent Neural Networks) for real-time rainfall and river stage forecasting. Conservation of mass was maintained in the river stage forecasting calculation scheme and adjusted to real-time observations so that model forecasts for preceding 1-3 hour intervals of river stage could be more accurately predicted and implemented into early warning systems. The river stage-inundation warning model was verified using available data from the Water Resources Agency, Ministry of Economic Affairs in Taiwan. Five Typhoon events between 2008 and 2009 (Typhoons: Kalmaegi, Fungwong, Sinlaku, Jangmi and Morakot) were used in the model verification and validation process. Simulation results showed that the predicted 1-3 hrs river stage hydrograph forecasts a close to the observed river stage hydrographs (RMSE between 0.26 m and 0.84 m). The modelling approach offers a 2-3 hours advance warning lead time when forecasting river stages which precede the current the current warning level messages.
