| Title |
Comparative Study of Machine Learning Models and Distributed Runoff Models for Predicting Flood Water Level |
| Authors |
(Tasuku Kubo) ; (Takeo Okazaki) |
| DOI |
https://doi.org/10.5573/IEIESPC.2023.12.3.215 |
| Keywords |
Flood forecasting; Parameter optimization; Machine learning; Physics-based model; Underestimation error |
| Abstract |
Conventional flood forecasting methods can be roughly classified as physics-based models and data-oriented models, which both require parameter optimization. In parameter optimization, a search is generally done to minimize the magnitude of overall errors. However, under- and overestimation errors are not equivalent in flood forecasting since underestimation of the water level leads to delays in decision making. We propose a risk-aware forecasting method that uses a weighted loss function. We applied the proposed method to both physics-based models and machine learning models and compared the prediction results to clarify the difference in the prediction results according to the base model used. The results show that the model optimized by the weighted loss function reduced the underestimation error while maintaining the overall error. |