Cao et al. (2025) Enhancing short-term PWV prediction through GNSS and ERA5 data fusion

Identification

• Journal: Atmospheric Research
• Year: 2025
• Date: 2025-11-28
• Authors: Yuxuan Cao, Jun Tang, Haojun Li, Yibin Yao, Liang Zhang, Chaoqian Xu
• DOI: 10.1016/j.atmosres.2025.108663

Research Groups

• Faculty of Land Resources Engineering, Kunming University of Science and Technology, Kunming, China
• School of Geodesy and Geomatics, Wuhan University, Wuhan, China

Short Summary

This study developed a multi-source data fusion model combining Global Navigation Satellite System (GNSS) and ERA5 precipitable water vapor (PWV) to enhance short-term, high-accuracy, and high-spatial-resolution PWV predictions, demonstrating significant improvements in prediction accuracy using Transformer and Long Short-Term Memory (LSTM) neural networks.

Objective

• To construct a high-accuracy, high-spatial-resolution continuous real-time precipitable water vapor (PWV) prediction model by fusing multi-source PWV data (GNSS and ERA5) to overcome limitations of traditional single data source predictions.

Study Configuration

• Spatial Scale: High spatial resolution (objective); relies on GNSS ground station distribution and ERA5 global reanalysis.
• Temporal Scale: Short-term, real-time, continuous monitoring using a sliding window technique.

Methodology and Data

• Models used: Transformer neural network, Long Short-Term Memory (LSTM) neural network, with incorporated feature engineering.
• Data sources: Global Navigation Satellite System (GNSS) precipitable water vapor (PWV), ERA5 (fifth generation of the European Centre for Medium-Range Weather Forecasts) PWV.

Main Results

• The fused PWV data significantly improved prediction accuracy compared to sole ERA5 PWV.
• For Transformer predictions, the root mean square error (RMSE) decreased from 1.596 mm to 1.253 mm (a 21.49 % reduction), and the correlation coefficient (R) increased from 0.967 to 0.979 (a 1.24 % improvement).
• For LSTM predictions, the RMSE decreased from 1.601 mm to 1.3 mm (an 18.83 % reduction), and R increased from 0.967 to 0.979 (a 1.24 % improvement).
• The Transformer model demonstrated superior performance over the LSTM model for short-term PWV predictions.

Contributions

• Developed a novel multi-source data fusion approach for PWV prediction, combining GNSS and ERA5 data to achieve higher accuracy and spatial resolution than traditional single-source methods.
• Integrated feature engineering into Transformer and LSTM models, enhancing their capability for time series PWV prediction.
• Demonstrated the effectiveness of fused PWV data and the superior performance of the Transformer model for real-time, short-term PWV monitoring.

Funding

• Not specified in the provided paper text.

Citation

@article{Cao2025Enhancing,
  author = {Cao, Yuxuan and Tang, Jun and Li, Haojun and Yao, Yibin and Zhang, Liang and Xu, Chaoqian},
  title = {Enhancing short-term PWV prediction through GNSS and ERA5 data fusion},
  journal = {Atmospheric Research},
  year = {2025},
  doi = {10.1016/j.atmosres.2025.108663},
  url = {https://doi.org/10.1016/j.atmosres.2025.108663}
}