Ye et al. (2026) Improving precipitation nowcasting via multiphysical parameter fusion in radar echo extrapolation

Identification

• Journal: Journal of Hydrology
• Year: 2026
• Date: 2026-01-13
• Authors: Yuankang Ye, Feng Gao, Shaoqing Zhang, Chang Liu
• DOI: 10.1016/j.jhydrol.2026.134947

Research Groups

• College of Intelligent Systems Science and Engineering, Harbin Engineering University, Harbin, China
• Qingdao Innovation and Development Center, Harbin Engineering University, Qingdao, China
• Key Laboratory of Physical Oceanography, Ministry of Education/Institute for Advanced Ocean Study/Frontiers Science Center for Deep Ocean Multispheres and Earth System (DOMES), Ocean University of China, Qingdao, China
• College of Oceanic and Atmospheric Sciences, Ocean University of China, Qingdao, China
• College of Ocean Science and Engineering, Shandong University of Science and Technology, Qingdao, China

Short Summary

This study introduces a Physics-Informed Multimodal Echo Extrapolation (PIEE) deep learning model that integrates radar reflectivity with four additional physical polarization parameters to significantly improve precipitation nowcasting accuracy, particularly for high-intensity echoes.

Objective

• To improve precipitation nowcasting accuracy by capturing complex dynamic evolutionary features of hydrometeor particles through the integration of multimodal radar physical parameters into a deep learning framework, thereby breaking the conventional unimodal data paradigm.

Study Configuration

• Spatial Scale: High-resolution, catchment scales.
• Temporal Scale: Short-term (nowcasting, typically minutes to a few hours).

Methodology and Data

• Models used: Physics-Informed Multimodal Echo Extrapolation (PIEE) neural network, featuring a three-stage structure: a multimodal encoder with a dual-branch attention-based fusion strategy, a novel gated spatiotemporal self-attention module, and a decoding stage.
• Data sources: Real multimodal radar echo dataset, comprising radar echo reflectivity and four additional physical polarization parameters of hydrometeor particles.

Main Results

• The proposed PIEE model demonstrates superior performance in precipitation nowcasting compared to existing fusion strategies and unimodal baseline architectures.
• Fusion of multiple physical parameters leads to significant improvements across all evaluated metrics.
• Specifically, for high echo intensity regions (≥ 40 dBZ), the model achieved improvements of up to 24.2% in the Critical Success Index (CSI) and 20.3% in the Heidke Skill Score (HSS).
• Systematic ablation experiments quantified the individual contributions of different physical parameters and their combinations to extrapolation accuracy.

Contributions

• Proposes a novel physics-informed multimodal radar echo extrapolation model (PIEE) for enhanced precipitation nowcasting.
• Integrates five physically meaningful radar polarization parameters as direct model inputs, moving beyond the conventional unimodal reflectivity data approach.
• Develops a dual-branch encoder with an attention-based feature fusion strategy to effectively capture intra- and inter-modal dynamics.
• Offers a unified framework for high-resolution nowcasting, highlighting the potential of physics-informed, multimodal deep learning for improving short-term hydrological prediction and risk management.

Funding

Not explicitly mentioned in the provided text.

Citation

@article{Ye2026Improving,
  author = {Ye, Yuankang and Gao, Feng and Zhang, Shaoqing and Liu, Chang},
  title = {Improving precipitation nowcasting via multiphysical parameter fusion in radar echo extrapolation},
  journal = {Journal of Hydrology},
  year = {2026},
  doi = {10.1016/j.jhydrol.2026.134947},
  url = {https://doi.org/10.1016/j.jhydrol.2026.134947}
}