Deep Learning for Opportunistic Rain Estimation via Satellite Microwave Links

Area of Science:

• Meteorology and Remote Sensing
• Machine Learning Applications
• Telecommunications Engineering

Background:

• Accurate precipitation measurement is vital for managing flood and drought risks, but traditional tools face limitations.
• Satellite-to-Earth microwave links (SMLs) offer a novel approach for precipitation estimation, with machine learning (ML) showing promise.
• Direct rainfall estimation from raw signal-to-noise ratio (SNR) data using deep learning (DL) is an underexplored area.

Purpose of the Study:

• To investigate and develop ML models for real-time rainfall detection and estimation using SNR data from SMLs.
• To evaluate the performance of various ML approaches, including DL and gradient boosting machine (GBM), against traditional methods.
• To provide a reliable solution for precipitation estimation via Earth-satellite microwave links, enhancing monitoring capabilities.

Main Methods:

• Utilized a year-long dataset of SNR measurements from interactive satellite receivers, paired with disdrometer and rain-gauge data.
• Developed and evaluated a range of ML models, including deep learning and gradient boosting machine, for rainfall estimation.
• Implemented ensemble approaches for both rainfall detection and cumulative precipitation estimation.

Main Results:

• Developed real-time ML models for rainfall detection and estimation using downlink SNR signals.
• Achieved a 46% reduction in root mean squared error (RMSE) for event-based cumulative precipitation predictions compared to state-of-the-art power-law models.
• Demonstrated the reliability of ML models for precipitation estimation using Earth-satellite microwave links.

Conclusions:

• Machine learning, particularly deep learning, offers a powerful and accurate method for estimating precipitation from satellite microwave link SNR data.
• The developed ML models provide a significant improvement over traditional methods, enhancing the potential for effective precipitation monitoring and disaster risk management.
• This research highlights the value of opportunistic data from communication signals for scientific applications in environmental monitoring.