THE STUDY OF TROPICAL CYCLONE CHARACTERISTICS USING MATHEMATICAL AND MACHINE LEARNING TECHNIQUES

Abstract

According to the 2030 Agenda for Sustainable Development, disaster risk reduction is essen tial to social and economic development. Storms and floods are among the most impactful natural disasters, contributing significantly to their frequency, the number of affected in dividuals, and economic losses. Storms and floods in coastal areas are mainly caused by tropical cyclones in tropical and subtropical parts of the globe. Approximately 0.8 million people lost their lives, and a financial loss of US$ 1407.6 billion was caused by tropical cyclones in the last 50 years. The atmospheric and oceanic causative components are multi dimensional in nature and have a complicated non-linear connection, making any estimation effort linked to tropical cyclones (TCs) difficult. Estimating the TC’s radius of maximum wind, intensity, and track is the primary focus of study on tropical cyclones. The major ity of the operational models that are now in use are statistical and numerical in nature. The numerical techniques need a lot of work and time. Complex non-linear interactions between several causative elements with geographical and temporal dimensions cannot be captured by statistical approaches because they are too simplistic. Numerous deep-learning research has been published recently that successfully address several kinds of estimation issues related to tropical cyclones. This research work tries to answer various estimation problems related to a TC. Starting from the radius of maximum wind (RMW) of a TC over the North Indian Ocean, a region with frequent and intense TC activity, the first work proposes a model to estimate the RMW vi Monu using historical data and mathematical correlations between the latitude coordinate of the TC center, estimated pressure drop at the center up to 12 hectopascals, and RMW. The accuracy of the approach is determined using statistical metrics, including error percentage, t-test, and root mean square error (RMSE) compared to existing methods. An ensemble machine learning model has been developed, further refining RMW estimation, taking in put from existing methods, and targeting the data provided by the India Meteorological Department. In order to address the issue of accurately estimating the track of TC, the second work proposes a neural network method. The neural network takes the result of three traditional methods and targets the data provided by the IMD. It is trained using 56 TCs and tested on 6 TCs from 2014 to 2024. In this work, we didn’t use any satellite images. The accuracy of the approach is determined using statistical metrics, including error percentage and RMSE. The third works explore the Satellite Consensus (SATCON) algorithm for estimating TC intensity using infrared and microwave sensor-based images, analyzing the performance for pre-monsoon and post-monsoon TC as well as for intensity based TC categories. In the fourth work, we develop a neural network model to estimate the intensity of TC. As discussed in the second work, satellite images weren’t used; instead, results from three state-of-the-art methods were used as the input, with the targeting of the data provided by the IMD. The first work is a traditional method for estimating the RMW over the NIO. It achieves a mean absolute error percentage ranging from approximately 6% to 32%, whereas the other studies have a mean absolute error range of 13% to 128% concerning IMD best track data. Following this, our ensemble machine learning model has an RMSE of 10.63 nautical miles and an error percentage of 17.00 %, which are lower than other methods. In the second work, our neural network model achieves an RMSE of 0.14 knots and an error percentage of 0.41%,lower than the alternative methods. In the third work, we demonstrate that SATCON is more effective in the post-monsoon across the West Pacific basin than in the pre-monsoon. Also, the ability of the algorithm to estimate intensity is determined to be rather excellent for mid-range TCs. In the fourth work, the RMSE of 5.71 knots and an error percentage of 10.07%, lower than existing methods.