Li et al. (2026) Deep learning-based precipitation nowcasting integrating radar echoes and rain gauge data

Identification

• Journal: Journal of Hydrology
• Year: 2026
• Date: 2026-08-09
• Authors: Leijing Li, Jianzhu Li, Yongtao Wang, Chongju Shang, Zhixia Li, Jin Liu
• DOI: 10.1016/j.jhydrol.2026.136210

Research Groups

• State Key Laboratory of Hydraulic Engineering Simulation and Safety, Tianjin University, China
• Guizhou Water Conservation Science and Research Institute, China
• Hebei Xingtai Meteorological Bureau, China

Short Summary

This study presents a deep learning-based precipitation nowcasting model integrating radar echoes and rain gauge data to improve the accuracy of short-term heavy rainfall forecasts.

Objective

• Investigate the feasibility of combining radar echoes with rain gauge-constrained physically guided priors for enhanced precipitation nowcasting accuracy.

Study Configuration

• Spatial Scale: Small spatial scales, focusing on local storm events within the Liulin watershed.
• Temporal Scale: Short-term forecasting (1-hour and 2-hour lead times) for heavy rainfall events.

Methodology and Data

• Models used:
  • SAAR-UNet: Radar echo extrapolation model utilizing an ASPP module, residual convolution, and spatial attention mechanism.
  • RST-RainNet: Quantitative precipitation estimation model combining radar echo sequences with gauge-constrained QPE priors and performing residual-aware fusion for precipitation nowcasting.
• Data sources: Weather radar data, rain gauge observations.

Main Results

• SAAR-UNet achieved the most robust performance for strong echo extrapolation among comparative experiments against attention-based and convolutional U-Net variants.
• RST-RainNet consistently demonstrated the highest correlation coefficients and lowest overall errors at both 1-hour and 2-hour lead times, accurately reproducing temporal evolution and spatial structure of rainfall.

Contributions

• This study contributes to improving precipitation nowcasting accuracy by integrating radar echoes with rain gauge-constrained physically guided priors.
• The proposed models (SAAR-UNet and RST-RainNet) show enhanced performance in predicting short-term heavy rainfall events, providing valuable insights for hydrological modeling and flood simulations.

Funding

• This research was supported by the State Key Laboratory of Hydraulic Engineering Simulation and Safety, Tianjin University (Grant No. 2022SKLHES).
• The authors also acknowledge funding from the Guizhou Water Conservation Science and Research Institute (Grant No. GWCSRI-2023).

Citation

@article{Li2026Deep,
  author = {Li, Leijing and Li, Jianzhu and Wang, Yongtao and Shang, Chongju and Li, Zhixia and Liu, Jin},
  title = {Deep learning-based precipitation nowcasting integrating radar echoes and rain gauge data},
  journal = {Journal of Hydrology},
  year = {2026},
  doi = {10.1016/j.jhydrol.2026.136210},
  url = {https://doi.org/10.1016/j.jhydrol.2026.136210}
}