Machine-Learning–Based Forecasting of an Extreme Precipitation event over Bangladesh

Abstract

The increasing frequency of intense precipitation events in Bangladesh is intensifying both climatic and socioeconomic challenges. Recent 2022 flooding in the northeastern region, triggered by extreme precipitation, exemplifies the nation’s heightened vulnerability to climate-induced disasters and their economic repercussions. Enhancing early warning systems for severe weather events is therefore essential to mitigate the associated risks and reduce potential losses. In recent years, Machine Learning (ML) techniques have gained prominence in climate forecasting owing to their computational efficiency and ability to capture complex nonlinear relationships among atmospheric variables. In this study, the forecasting capabilities of the Random Forest Classifier (RFC) and Artificial Neural Network (ANN) models have been evaluated for predicting extreme precipitation events over Bangladesh with a 24-hour lead time. The analysis also aimed to identify the most influential predictors exhibiting strong correlations with the target variable, next-day precipitation. The findings indicate that Geopotential Height at 1000 hPa, Mean Surface Direct Shortwave Radiation Flux, and Relative Humidity at 500 hPa and 800 hPa are the most significant predictors of next-day precipitation. Furthermore, the comparison of model performance demonstrates that both the RFC and ANN models yield comparable levels of predictive accuracy.