Cyclone prediction using machine learning

Abstract

Forecasting Tropical cyclones (TCs) are the most unsettling problem for meteorologists and amongst the dangerous natural disasters worldwide. One super tropical cyclone can cause up to one thousand deaths, create more than 50 billion US dollars damages in a single event, and have been responsible for more than one lac deaths in recent history. About 7% of global cyclone frequency occurs over the North Indian Ocean, which causes heavy loss of life and property over the region. Accurate insight into the tropical cyclone can help us save nature and human lives, and we aim to develop models for better prediction results. Much advancement has been made in the domain of Tropical cyclones fore casting using satellite data by machine learning(ML) techniques and tested by many meteorologists to study the many problems in TCs. It opens a new window to find solutions for the bottleneck problems in the tropical cyclone forecasting domain, where traditional methods fail to solve these problems. We have tremendous interest in testing these techniques for tropical cyclone forecasting in the North Indian Ocean region. This paper used the best track data since 1970-2020 from IMD and JTWC Meteorolog ical Departments, with key cyclone parameters such as latitude, longitude, minimum central cyclone pressure, and the maximum cyclonic wind inten sity to forecast the cyclone tracks in the North Indian Ocean region. Here the main aim to build an architecture based on the deep learning tech nique(DL) of Recurrent neural networks(RNN) to do the Track forecasts and Intensity forecasts. Keywords: Tropical cyclone; Machine learning;Deep learning: forecasts; Track; Intensity; Longitude; Latitude;Central pressure.