Using Machine Learning to Classify Lightning for Earth Networks Total Lightning Network (ENTLN)

Every year, almost 10,000 people are killed due to lightning strikes. Its unpredictable nature makes it a weather hazard. Lightning is associated with weather events like thunderstorms. Understanding the lightning characteristics of a thunderstorm, such as the intracloud (IC) to cloud-to-ground (CG) ratio, can help in better understanding the maturity of a storm. Hence, there is a need for detecting and predicting lightning flash effectively for forecasting accurate weather information.

With 1600 sensors globally, the Earth Networks Total Lightning Network (ENTLN) is a system which has been operating over a long time in both tracking and detecting lightning with a high detection rate. In 2014, a new multi-parameter classification algorithm was introduced, and results indicate that this introduced significant improvements.

In this study, we look to build in these and develop an innovative machine learning approach to enhance its performance.

We investigate potential improvements from applying different machine learning algorithms to lightning classification. The methodology comprises of training model and using it alongside the ENTLN processor for predicting the type of lightning flash. Ground-truth lightning data acquired from various sources (Lightning Observatory lab in Gainesville (LOG), Langmuir Lab at New Mexico Tech (LL) and Wits University in South Africa) are used for training and validation of the model. The initial results from our study show that the methodology presents an acceptable performance.

The developed work in this study would prove to be subtly useful in improving the performance of the present ENTLN processor via machine learning techniques and contributes to the literature by expanding knowledge of various types of lightning flashes.