Using Machine Learning Methods to Develop an Algorithm for Recognizing a Risk of Waterspout Occurrence off the Black Sea Coast of Russia

Abstract

Every year about 50 waterspouts occur over the sea off the Black Sea coast of Russia. Over the past few years, the cases of waterspouts have occurred in the immediate vicinity of the coast with their subsequent destruction. The vortex destruction is often accompanied by short-term wind strengthening up to storm levels. The present study solves the problem of nowcasting the Black Sea waterspouts (building a detailed forecast of their formation for the next 2–6 hours) using machine learning methods. Learning by precedents is considered based on the labeled dataset of the radar characteristics of convective systems with and without waterspouts, models for classifying systems in terms of the risk of waterspout occurrence are constructed. The testing of the models showed that it is fundamentally possible to use them to diagnose systems with already formed waterspouts, as well as to identify the risk of waterspouts in advance (within two hours).