Analysis and forecasting of meteorological drought using PROPHET and SARIMA models deploying machine learning technique for southwestern region of Bangladesh

The southwestern Bangladesh is predominantly agrarian region, and recurrence of drought impacts directly on crops, reducing farmers' incomes and threatening national food security. This study analyzes and forecasts meteorological droughts for southwestern region using Probabilistic Forecasting with Structural Time Series (PROPHET) and Seasonal Autoregressive Integrated Moving Average (SARIMA) models. PROPHET excels at capturing long-term, non-linear trends, while SARIMA efficiently models seasonal variations. The essential climatic variables (e.g., temperature, rainfall, soil moisture) from Jashore, Jhenaidah, Kushtia, and Chuadanga were collected from 1994 to 2018 to enhance drought prediction accuracy up to 2050. In both cases, the calculation was performed by deploying the Machine Learning Techniques. Both models showed that drought intensity varies spatiotemporally and that frequent drought occurrences are common in all districts. Almost 50 % of the projected years (2019–2050) will be considered drought years (≥6 dry months in a year). 2049 and 2050 are nearly dry years in all districts. Considering the total months and years of drought, Chuadanga is followed by the Jashore, Jhenaidah, and Kushtia districts. Moreover, a strong correlation was found between the predicted and observed drought, whereas the R2 values were 0.83, 0.75, 0.88, and 0.76 for Jashore, Jhenaidah, Kushtia, and Chuadanga, respectively. Hence, these models would provide robust forecasts, helping to identify increasing drought severity in the region. Model validation using key performance metrics demonstrates their reliability in supporting water resource management. The findings underscore the importance of proactive drought mitigation strategies and suggest that future research should incorporate additional climate variables to improve prediction accuracy.