Gan et al. (2024) Machine-learning downscaling of GPM satellite precipitation products in mountainous regions: A case study in Chongqing

Identification

• Journal: Atmospheric Research
• Year: 2024
• Date: 2024-09-20
• Authors: Yushi Gan, Yuechen Li, Lihong Wang, Long Zhao, Lei Fan, Haichao Xu, Z. Q. Yin
• DOI: 10.1016/j.atmosres.2024.107698

Research Groups

• Chongqing Engineering Research Center for Remote Sensing Big Data Application, Southwest University
• Chongqing Jinfo Mountain National Field Scientific Observation and Research Station for Karst Ecosystem, Southwest University
• School of Geographical Sciences, Southwest University
• Key Laboratory of Monitoring, Evaluation, and Early Warning of Territorial Spatial Planning Implementation, Ministry of Natural Resources
• China Institute of Water Resources and Hydropower Research

Short Summary

This study applies a "calibration then downscaling" approach using machine learning to improve the spatial resolution of GPM daily precipitation products in the mountainous region of Chongqing, identifying LSTM as the most effective algorithm.

Objective

• To enhance the spatial resolution of GPM satellite precipitation products in complex mountainous terrain and evaluate the performance of different machine learning algorithms in reducing accuracy loss during downscaling.

Study Configuration

• Spatial Scale: Chongqing, China (mountainous region)
• Temporal Scale: Daily precipitation

Methodology and Data

• Models used: Random Forest (RF), Extreme Gradient Boosting (XGBoost), and Long Short-Term Memory (LSTM)
• Data sources: GPM (Global Precipitation Measurement) satellite precipitation products and ground-based observation station data

Main Results

• Calibration: Station-calibrated GPM precipitation products demonstrated higher accuracy and closer alignment with station measurements than the original GPM products.
• Resolution Enhancement: The spatial resolution was successfully improved from 0.1° to 0.01° (~1 km).
• Algorithm Performance: LSTM effectively enhanced spatial resolution without compromising accuracy, whereas RF and XGBoost suffered significant accuracy loss during the downscaling process.
• Spatial Distribution: LSTM results exhibited the greatest spatial continuity and most closely aligned with the actual characteristics of precipitation distribution in the study area.

Contributions

• Provides a high-quality precipitation data generation scheme specifically tailored for mountainous regions characterized by complex terrain and sparse ground station coverage.

Funding

• Not specified in the provided text.

Citation

@article{Gan2024Machinelearning,
  author = {Gan, Yushi and Li, Yuechen and Wang, Lihong and Zhao, Long and Fan, Lei and Xu, Haichao and Yin, Z. Q.},
  title = {Machine-learning downscaling of GPM satellite precipitation products in mountainous regions: A case study in Chongqing},
  journal = {Atmospheric Research},
  year = {2024},
  doi = {10.1016/j.atmosres.2024.107698},
  url = {https://doi.org/10.1016/j.atmosres.2024.107698}
}