Lightning flash-cells detection using machine learning algorithms
Abstract
The paper presents an analysis of the possibility of applying different machine learning algorithms to the detection of thunderstorm cells. The basis for the presented analysis is exclusively data on lightning strikes collected using a network for the detection of lightning activity. No additional meteorological data or indicators are used, and the thunderstorm cell is observed exclusively from the aspect of thunderstorm activity without investing in the meteorological aspects of this problem. Namely, the problem of thunder cell detection, viewed exclusively from the perspective of using available data on lightning strikes, consists in spatial and temporal cluster finding, which falls under the domain of unsupervised machine learning. The paper analyzes the possibility of applying different algorithms for finding spatio-temporal groups of intense thunderstorm activity. After the positive identification of the cluster, the associated thunderstorm cell is formed by applying the convex hull algorithm. The properties of the convex hull (eg its surface area, center of mass, etc.) are used to identify the thunderstorm cell in space and time and to track and predict its movement.
