Yan et al. (2025) Time-Extended Bayesian Retrieval of Dual-Polarization Radar Data Enhancing Short-Term Precipitation Forecasts

Identification

• Journal: Remote Sensing
• Year: 2025
• Date: 2025-12-11
• Authors: Jiapeng Yan, Chong Wu, Xingtao Song, Yonglin Chen
• DOI: 10.3390/rs17244003

Research Groups

• State Key Laboratory of Disaster Weather Science and Technology, Chinese Academy of Meteorological Sciences, Beijing, China
• School of Atmospheric Physics, Nanjing University of Information Science and Technology, Nanjing, China

Short Summary

This study developed a time-extended Bayesian retrieval method for dual-polarization radar data to address performance degradation in convective structure retrieval caused by temporal biases in Numerical Weather Prediction (NWP) models. The method significantly improved initial moisture fields and enhanced short-term precipitation forecasts (0–6 hours) by effectively resolving these temporal biases.

Objective

• To address the degradation in convective structure retrieval capability caused by temporal biases in Numerical Weather Prediction (NWP) models when using spatial neighborhood sampling methods with radar data for relative humidity field retrieval.
• To develop a time-extended Bayesian retrieval method and construct a dual-polarization radar data assimilation framework compatible with the China Meteorological Administration Mesoscale Model (CMA-MESO) to enhance short-term precipitation forecasts.

Study Configuration

• Spatial Scale: The study region covered North China (36.5°N to 41.5°N and 113.5°E to 119.5°E). The CMA-MESO model used a horizontal spatial resolution of 1 km with 71 vertical layers, and a model top at 10 hPa. Vertical interpolation of radar data was performed between model layers 15 and 42 at every third layer. A square sampling window with an edge length of 192 km was used, with subsampling at a 6 km resolution.
• Temporal Scale: Assimilation experiments were conducted with a 1-hour cycling frequency. The time-extended Bayesian method incorporated forecast fields from two hours before and two hours after the assimilation time (a total of 5 time steps). Forecasts were evaluated for lead times of 0–6 hours. The case study focused on a heavy rainfall event from 29 July to 1 August 2023, with hourly assimilation from 22:00 UTC on 29 July to 00:00 UTC on 30 July 2023, and forecasts initialized from 00:00 UTC on 30 July 2023.

Methodology and Data

• Models used: China Meteorological Administration Mesoscale Model (CMA-MESO V6.0), ZJU-AERO (Accurate and Efficient Radar Operator designed by Zhejiang University) dual-polarization forward observation operator, and a 1D+3D-Var (one-dimensional retrieval followed by a three-dimensional variational assimilation technique) framework.
• Data sources: Observational data from 20 S-band dual-polarization weather radars across North China, providing Horizontal Reflectivity (ZH), Specific Differential Phase (KDP), and Differential Reflectivity (ZDR). The hourly precipitation fusion product from the China Meteorological Administration Multi-Source Precipitation Analysis System (CMPAS) served as the observational benchmark for precipitation forecast evaluation.

Main Results

• The time-extended Bayesian retrieval method effectively resolved retrieval performance degradation caused by model temporal biases, leading to improved sampling capability compared to traditional spatial-only sampling.
• When applied to a heavy rainfall event in North China in July 2023, the method effectively reduced retrieval errors and enhanced the initial moisture fields in NWP models.
• Subsequent assimilation of retrieved humidity fields significantly enhanced Threat Scores (TS) for 0–6 hour precipitation forecasts and demonstrated a notable improvement in overprediction bias.
• For 0–3 hour precipitation, the "Extend" experiment outperformed the control (CTRL) across all thresholds below 50 mm, with the most pronounced TS improvements at the 10 mm threshold. Bias scores for higher thresholds (20 mm and 50 mm) were significantly closer to the ideal value of 1 compared to CTRL.
• For 3–6 hour forecasts, the "Extend" experiment showed superior TS at the 20 mm threshold (0.336 vs. 0.281 for CTRL) and a markedly reduced 50 mm threshold bias (1.438 vs. 2.067 for CTRL).
• The assimilation of humidity pseudo-observations led to sustained and effective influences on subsequent forecasts, improving the overall structure of echo forecasts and effectively constraining the spatial extent of intense echo regions.

Contributions

• This study introduces a novel time-extended Bayesian retrieval framework that integrates dual-polarization radar parameters with a time-neighborhood sampling strategy, effectively overcoming the limitations of spatial-only sampling under model temporal drift.
• It confirms the value of dual-polarization radar data assimilation based on this time-extended strategy, offering an effective reference for improving initial field construction and short-term forecasting in NWP models.
• The framework provides a feasible solution to overcome the limitations of spatial-only sampling under model temporal drift, with potential for extension to other rapidly evolving weather systems.
• The approach enhances the spatiotemporal consistency of the retrieval process by incorporating time-extended forecast information, leading to more accurate moisture analysis and improved convective precipitation forecasting.

Funding

• National Natural Science Foundation of China (42305156)
• National Key Research and Development Program of China (2023YFC3007501)
• Independent Research Project of National Key Laboratory of Severe Weather (2025QZA05)
• Young Innovation Team on Quality Control and Product Development of New Data Types (NMIC-2024-QN02), National Meteorological Information Center

Citation

@article{Yan2025TimeExtended,
  author = {Yan, Jiapeng and Wu, Chong and Song, Xingtao and Chen, Yonglin},
  title = {Time-Extended Bayesian Retrieval of Dual-Polarization Radar Data Enhancing Short-Term Precipitation Forecasts},
  journal = {Remote Sensing},
  year = {2025},
  doi = {10.3390/rs17244003},
  url = {https://doi.org/10.3390/rs17244003}
}