Ueda et al. (2025) Building a Generalized Pre-Training Model to Predict River Water-Level from Radar Rainfall

⚠️ Warning: This summary was generated from the abstract only, as the full text was not available.

Identification

• Journal: Water
• Year: 2025
• Date: 2025-12-05
• Authors: Futo Ueda, Hiroto TANOUCHI, Nobuyuki EGUSA, Takuya Yoshihiro
• DOI: 10.3390/w17243449

Research Groups

Not specified in the provided text.

Short Summary

This paper develops a generalized deep learning model for river water-level prediction applicable to multiple Japanese rivers by pre-training with inundation data from all Class-A rivers, demonstrating higher accuracy and broader applicability compared to pre-training with only similar rivers.

Objective

• To construct a generalized deep learning model for river water-level prediction commonly applicable to multiple Japanese rivers, overcoming the limitations of previous river-specific model building approaches.

Study Configuration

• Spatial Scale: Multiple Japanese Class-A rivers (major river systems managed by the government).
• Temporal Scale: Not specified in the provided text.

Methodology and Data

• Models used: Deep learning, incorporating a transfer learning approach with a newly defined flow distance matrix.
• Data sources: Radar rainfall data, inundation data from all Japanese Class-A rivers.

Main Results

• Pre-training the deep learning model using inundation data from all Japanese Class-A rivers yields higher water-level prediction accuracy across multiple rivers with varying conditions compared to pre-training using only similar rivers.
• This approach enables the construction of a generalized river water-level prediction model applicable to a wide range of rivers, eliminating the need for laborious pre-selection of similar rivers for training.

Contributions

• Development of a generalized deep learning model for river water-level prediction that is commonly applicable to a wide range of rivers, addressing the previous limitation of laborious river-specific model construction.
• Demonstration that pre-training with comprehensive regional data (all Class-A rivers) significantly improves prediction accuracy and generalizability across diverse river conditions compared to pre-training with only similar rivers.

Funding

Not specified in the provided text.

Citation

@article{Ueda2025Building,
  author = {Ueda, Futo and TANOUCHI, Hiroto and EGUSA, Nobuyuki and Yoshihiro, Takuya},
  title = {Building a Generalized Pre-Training Model to Predict River Water-Level from Radar Rainfall},
  journal = {Water},
  year = {2025},
  doi = {10.3390/w17243449},
  url = {https://doi.org/10.3390/w17243449}
}