Author(s): Ken Watanabe; Masayasu Irie; Makiko Iguchi
Linked Author(s):
Keywords: Multi-point flood forecasting; Hybrid model; Deep learning; Distributed rainfall-runoff model; Missing dat
Abstract: In recent years, flood forecasting technology has become more important due to the increase in severe flood disasters owing to climate change. As real-time water level forecasting methods, hybrid forecasting methods that combine the advantages of rainfall-runoff models and machine learning models have been studied actively. While many previous reports have validated the accuracy of hybrid models for a single point in a watershed, there are few reports on validating the accuracy of multi-point simultaneous water level forecasts assumed in actual systems. In addition, as observation data is often missing in actual real-time forecasting systems, it is important to be able to provide stable forecasts even when some observation data is missing. In this study, the performance of a hybrid method using a distributed rainfall-runoff model (RRI model) and a deep learning model (DNN) was examined in the above situations. We evaluated forecast accuracy for simultaneous multi-point water level prediction and the impact of missing data on forecast accuracy. The results showed that the effect of hybridization on improving forecast accuracy became larger from upstream to downstream, and root mean squared error (RMSE) was improved by up to 10% or more compared to the simple DNN model. Furthermore, the hybrid model showed more stable prediction accuracy in missing data cases, and when trained with masked data simulating missing data, the prediction accuracy degradation due to missing data was limited to 6.9% in RMSE. These results indicate that this hybrid model is suitable for actual real-time forecasting systems.
Year: 2024