Machine Learning-based Statistical Prediction of Cyclonic Disturbance Frequency during Post-monsoon over the Bay of Bengal

Abstract

Deadly tropical cyclones (TCs) form quickly over the Bay of Bengal (BoB) during the post-monsoon season (October to December; OND) and cause significant socio-economic damage across India and the neighbouring countries. For better planning to reduce the risks associated with cyclonic activities, seasonal forecasting in advance would be beneficial. The current study has assessed the influences of large-scale dynamic and thermodynamic parameters of the preceding seasons, i.e., June to September, on the post-monsoon cyclonic disturbance (CD) frequency over BoB to develop statistical models for seasonal prediction. Six parameters, including sea surface temperature, sea level pressure, relative humidity at 500 hPa, zonal wind at 200 hPa and 850 hPa, and meridional wind at 850 hPa levels, with significant correlations with the formation of CDs over BoB from 1982–2020 (39 years), were selected as potential predictors. By utilizing the selected predictors, four machine learning (ML) models, namely Principal Component Regression (PCR), Support Vector Regression (SVR), Random Forest (RFR) and Artificial Neural Network (ANN), were built to forecast the frequency of CDs. The RFR model showed relatively better skills in both quantitative and categorical forecasts among the four models. Hence, it can be utilized more reliably for seasonal CD frequency over BoB.