Li et al. (2025) A novel CNN-based method using GNSS tomography and WRF data for regional rainfall prediction

Identification

• Journal: Advances in Space Research
• Year: 2025
• Date: 2025-12-22
• Authors: Longjiang Li, Saeid Haji-Aghajany, Kefei Zhang, Witold Rohm, Xiaoming Wang, Suqin Wu, Haobo Li, Dongsheng Zhao, Minghao ZHANG
• DOI: 10.1016/j.asr.2025.12.067

Research Groups

• School of Environment and Spatial Informatics, China University of Mining and Technology, Xuzhou, China
• Institute of Geodesy and Geoinformatics, Wrocław University of Environmental and Life Sciences, Wrocław, Poland
• College of Geomatics, Shandong University of Science and Technology, Qingdao, China
• Aerospace Information Research Institute, Chinese Academy of Sciences, Beijing, China
• School of Science, RMIT University, Melbourne, Australia

Short Summary

This study introduces a novel Convolutional Neural Network (CNN)-based method for regional rainfall prediction, integrating four-dimensional wet refractivity fields from GNSS tomography with meteorological data from the WRF model. The model achieved a 92.9 % True Positive Rate and a 4.8 % False Discovery Rate, demonstrating strong performance for predicting rainfall events, especially those with mid-to-high intensity.

Objective

• To develop and evaluate a novel Convolutional Neural Network (CNN)-based method for predicting regional rainfall areas and their spatiotemporal evolution, utilizing four-dimensional wet refractivity fields derived from GNSS tomography and meteorological data from the Weather Research and Forecasting (WRF) model.

Study Configuration

• Spatial Scale: Regional, specifically across Poland.
• Temporal Scale: A 300-day period during 2020–2021.

Methodology and Data

• Models used: Convolutional Neural Network (CNN), Weather Research and Forecasting (WRF) model.
• Data sources:
  • GNSS tomography (four-dimensional wet refractivity fields)
  • WRF model (pressure, temperature, cloud water mixing ratio, water vapor mixing ratio)
  • Global Precipitation Measurement (GPM) mission (rainfall observations, serving as target data)

Main Results

• The proposed CNN model achieved a True Positive Rate (TPR) of 92.9 % and a False Discovery Rate (FDR) of 4.8 %.
• The model demonstrated better performance in predicting rainfall events with mid-to-high rainfall rates (>0.1 mm/h) compared to predicting only heavy rainfall events (>5 mm/h).
• Extending the prediction window lengths improved the model's capability to capture rainfall events by reducing false positives, thereby enhancing overall performance.
• The new approach exhibited strong performance for meteorological applications across Poland.

Contributions

• Proposes a novel CNN-based method that integrates four-dimensional wet refractivity fields from GNSS tomography with WRF meteorological data for regional rainfall prediction.
• Addresses the limitation of previous studies by focusing on the prediction of rainfall areas and their spatiotemporal evolution, rather than just temporal variations in Precipitable Water Vapor (PWV).
• Demonstrates high predictive accuracy (92.9 % TPR, 4.8 % FDR) for regional rainfall, particularly for mid-to-high intensity events.
• Investigates the impact of varying rainfall intensity thresholds and prediction window lengths on model performance.

Funding

No specific funding projects, programs, or reference codes were mentioned in the provided text.

Citation

@article{Li2025novel,
  author = {Li, Longjiang and Haji-Aghajany, Saeid and Zhang, Kefei and Rohm, Witold and Wang, Xiaoming and Wu, Suqin and Li, Haobo and Zhao, Dongsheng and ZHANG, Minghao},
  title = {A novel CNN-based method using GNSS tomography and WRF data for regional rainfall prediction},
  journal = {Advances in Space Research},
  year = {2025},
  doi = {10.1016/j.asr.2025.12.067},
  url = {https://doi.org/10.1016/j.asr.2025.12.067}
}