An Ensemble Learning Approach for Drought Analysis and Forecasting in Central Bangladesh

Drought is a significant natural disaster in Bangladesh, characterized by prolonged periods of reduced rainfall. The study utilizes meteorological data, including monthly maximum and minimum temperatures, rainfall, humidity, wind speed, sunshine duration, and cloud cover, obtained from the Bangladesh Agricultural Research Council (BARC). This data is employed to compute the standardized precipitation index (SPI), the standardized precipitation evapotranspiration index (SPEI), and the Z-score index (ZSI). The focus of the study is to evaluate the accuracy of machine learning (ML) algorithms in predicting meteorological droughts in central region of Bangladesh from 1990 to 2022, using data from seven meteorological stations. SPI and SPEI were mapped within a GIS environment to assess drought hazards and identify key social and physical vulnerability factors. The random forest (RF) model demonstrated high performance in predicting SPI and SPEI, achieving an accuracy of 93.8%–94.0%, precision of 90.9%–92.7%, recall of 89.5%–92.0%, and F1-scores of 90.3%–92.0%. Its error metrics included MAE (0.055–0.068), MSE (0.0032–0.0052), RMSE (0.056–0.072), and R² (0.914–0.965) across an 80% training and 20% testing split. Additionally, the autoregressive integrated moving average (ARIMA) model was used to predict future SPI and SPEI values for 2023–2030, helping to identify the timing of both short-term and long-term droughts. The study revealed that integrating ARIMA with ML algorithms improved forecasting accuracy, achieving over 92.0% accuracy in predicting SPI and SPEI, thereby significantly enhancing drought prediction capabilities.