Short-Term Rainfall Forecasting Using LSTM, ARIMA, and LSTM Time Series Generator Models

Rainfall forecasting in Bangladesh holds immense significance due to the country's heavy reliance on agriculture and its vulnerability to climate-related disasters. Machine learning algorithms have become increasingly popular for weather forecasting due to their robustness and performance in recent times. In this study, we propose a machine learning model to forecast short-term rainfall in Cumilla, Bangladesh based on historical data. We use daily data to forecast rainfall on a day-to-day basis. Three machine learning models (ARIMA, LSTM, and LSTM Time Series Generator) are used in this study. The models forecast two types of rainfall data, i.e., 7 days and 14 days. Additionally, three seasons are considered for the forecasting, i.e., summer (April), rainy (June), and autumn (August). The performance of the ML models is analyzed using continuous indicators such as RMSE, RSE, and MSLE.