Scholarlink

<p>Accelerated urbanization and intensifying climate change have increased the frequency and spatial extent of flood disasters worldwide. To enhance the capability of flood prevention and mitigation, fast and accurate prediction of flood spatiotemporal evolution is crucial for timely issuing warnings and formulating response strategies. In existing flood research, traditional hydrodynamic models can dynamically simulate the flood evolution process by solving mass and momentum conservation equations, featuring high accuracy and reliability. However, such models have certain limitations in real-time prediction, and their high demand for input data, complex calculation processes, and slow numerical solution speed restrict their applications in flood emergency response. To address the limitations of traditional hydrodynamic models, the Miami River Basin in Florida, USA, is selected as the study area, and an integrated watershed flood modeling and prediction framework (SRR-ConvLSTM) is proposed that combines the Spatial Reduction and Reconstruction algorithm (SRR) and ConvLSTM to rapidly achieve short-term flood prediction. This framework selects representative locations through the spatial reduction algorithm, uses the flood inundation simulation model to offline generate historical flood data, combines the ConvLSTM model to predict floods at representative locations, and finally reconstructs the full flood inundation extent through the spatial reconstruction algorithm. The experimental results show that the SRR-ConvLSTM framework demonstrates excellent accuracy and stability in predicting flood water depth for the next 24&#xa0;h, with MAE and RMSE of 0.110&#xa0;m and 0.219&#xa0;m, respectively. Meanwhile, the reconstructed flood inundation area highly matches the results of the HEC- RAS model, and the computational efficiency is 121 times higher than that of traditional hydrodynamic models. In general, the SRR-ConvLSTM framework combines spatial reduction, historical flood data, ConvLSTM, and spatial reconstruction technology, ensuring both accuracy and computational efficiency while achieving short-term flood prediction.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Fast and Accurate Flood Inundation Dynamics Modeling and Prediction Based on Spatial Reconstruction and ConvLSTM

  • 期刊论文

  • 发表日期：

    2026年6月22日

查看全文
  • Shuang Zhu,
  • Qiaolong Chen,
  • Jun Guo,
  • Xiangang Luo,
  • Yi Liu,
  • Yida Li,
  • Hui Qin

摘要

Accelerated urbanization and intensifying climate change have increased the frequency and spatial extent of flood disasters worldwide. To enhance the capability of flood prevention and mitigation, fast and accurate prediction of flood spatiotemporal evolution is crucial for timely issuing warnings and formulating response strategies. In existing flood research, traditional hydrodynamic models can dynamically simulate the flood evolution process by solving mass and momentum conservation equations, featuring high accuracy and reliability. However, such models have certain limitations in real-time prediction, and their high demand for input data, complex calculation processes, and slow numerical solution speed restrict their applications in flood emergency response. To address the limitations of traditional hydrodynamic models, the Miami River Basin in Florida, USA, is selected as the study area, and an integrated watershed flood modeling and prediction framework (SRR-ConvLSTM) is proposed that combines the Spatial Reduction and Reconstruction algorithm (SRR) and ConvLSTM to rapidly achieve short-term flood prediction. This framework selects representative locations through the spatial reduction algorithm, uses the flood inundation simulation model to offline generate historical flood data, combines the ConvLSTM model to predict floods at representative locations, and finally reconstructs the full flood inundation extent through the spatial reconstruction algorithm. The experimental results show that the SRR-ConvLSTM framework demonstrates excellent accuracy and stability in predicting flood water depth for the next 24 h, with MAE and RMSE of 0.110 m and 0.219 m, respectively. Meanwhile, the reconstructed flood inundation area highly matches the results of the HEC- RAS model, and the computational efficiency is 121 times higher than that of traditional hydrodynamic models. In general, the SRR-ConvLSTM framework combines spatial reduction, historical flood data, ConvLSTM, and spatial reconstruction technology, ensuring both accuracy and computational efficiency while achieving short-term flood prediction.