Rain Discrimination with Machine Learning Classifiers for Opportunistic Rain Detection System Using Satellite...

Area of Science:

• Environmental Science
• Meteorology
• Remote Sensing

Background:

• Climate change intensifies extreme weather events, necessitating advanced disaster risk management systems.
• Real-time monitoring and forecasting are vital for early warnings and hazard mitigation.
• Oblique earth-space links (OELs) offer a promising method for real-time rainfall detection.

Purpose of the Study:

• To address the challenge of classifying rainy and non-rainy periods using OEL data.
• To compare the performance of different machine learning (ML) classifiers for rainfall event detection.
• To evaluate the effectiveness of neural networks (NN) against support vector machines (SVM), random forest (RF), and decision trees (DT).

Main Methods:

• Utilized data from eighteen rain events, correlating satellite-to-earth link quality with tipping bucket rain gauge (TBRG) measurements.
• Preprocessed TBRG data and extracted feature sets (6 and 12 features) from microwave link data.
• Applied and compared four ML classifiers: SVM, NN, RF, and DT.

Main Results:

• The neural network (NN) classifier demonstrated superior performance in distinguishing between rainy and non-rainy periods.
• Performance was evaluated across various data arrangements and feature set sizes.
• NN consistently outperformed SVM, RF, and DT in the classification task.

Conclusions:

• Neural networks are highly effective for classifying rainfall events using OEL data.
• This research contributes to improving real-time rainfall monitoring systems for disaster preparedness.
• Accurate classification of rainfall periods is a key step towards reliable rainfall intensity estimation via OELs.