This project involves developing a sophisticated machine learning model for the classification of lightning events based on electromagnetic waveform data. Working in Georgia Tech's Low Frequency Radio Lab, I'm creating a system that can accurately distinguish between Intracloud and Cloud-to-Ground lightning strikes with over 95% accuracy. This research contributes to advancing atmospheric science understanding and has applications in weather prediction and safety systems.
Developed ML model achieving >95% accuracy in distinguishing between Intracloud vs Cloud-to-Ground lightning events
Built data pipeline processing electromagnetic waveform data from 100,000,000+ annual lightning strikes
Authoring graduate-level research paper using novel methodologies to advance atmospheric science research
Contributing to atmospheric science research with potential applications in weather prediction and safety systems
The project utilizes advanced machine learning techniques for signal processing and classification. The system processes complex electromagnetic waveform data, extracting relevant features for lightning event classification. The model architecture is designed to handle large-scale datasets while maintaining high accuracy and computational efficiency.
One of the primary challenges was handling the massive scale of data from over 100 million annual lightning strikes while maintaining processing efficiency. I addressed this by developing an optimized data pipeline that can efficiently process and classify electromagnetic waveforms. Another challenge was achieving the high accuracy required for research-grade results, which required careful feature engineering and model optimization to distinguish between subtle differences in lightning event signatures.
The model successfully achieves over 95% accuracy in lightning classification, meeting the stringent requirements for research applications. This represents a significant advancement over previous classification methods, which have traditionally relied on Low Frequency (LF) data rather than the much more challenging Very Low Frequency (VLF) data that my model processes. While past models have used only manually selected features extracted from waveforms, my innovative approach feeds raw waveform samples directly into a Convolutional Neural Network alongside other engineered features, allowing the model to learn complex patterns that traditional feature extraction methods might miss.
This work is contributing to a graduate-level research paper that will advance our understanding of atmospheric phenomena. The methodologies developed in this project have potential applications in weather prediction systems, lightning safety protocols, and atmospheric research. The project demonstrates the practical application of machine learning in atmospheric science and contributes to the broader field of electromagnetic signal analysis, particularly in the challenging VLF domain where signal characteristics are more subtle and require sophisticated analysis techniques.