Author(s): Hongjie Yu; Yue-Ping Xu; Man Yue Lam; Reza Ahmadian

Linked Author(s): Reza Ahmadian, Man Yue Lam

Keywords: Flood Machine Learning Land Use Change Adaption

Abstract: Effective flood mapping is essential for adapting to natural hazards in the context of climate change. Traditional hydrodynamic models, while accurate, are computationally intensive, limiting their application in real-time flood warning systems. Machine learning (ML) models offer a computationally efficient alternative but require extensive data for training, posing challenges in dynamic geographical conditions. This study employs Model-Agnostic Meta-Learning (MAML) to enhance the adaptability of ML models for flood simulation. Using the Jiao River Basin in China as a case study, we trained a fully connected neural network (FCN) with MAML, utilizing flood data within historical and recent land use. The meta model demonstrated superior performance, achieving a Root Mean Square Error (RMSE) of 0.04m and an Overall Accuracy (OA) of 0.96, compared to the base model's RMSE of 0.07m and OA of 0.91. These findings highlight the potential of MAML in developing robust, adaptive ML models for rapid flood mapping and early warning systems.

DOI:

Year: 2025