Spatiotemporal trends and machine learning-based prediction of temperature variability during the T. Aman rice-growing season in Bangladesh.

Climate change poses significant risks to food security, especially in agriculture-dependent countries like Bangladesh. This study analyzes temperature trends from 1961 to 2023 using data from the Bangladesh Meteorological Department (BMD) across three climatic regions: Barind, Coastal, and Haor. The Mann-Kendall test revealed statistically significant warming trends in both maximum and minimum temperatures, with the most pronounced increase in the Haor region. Moran's analysis detected clear spatial clustering of high-risk zones, with Barind districts facing severe maximum temperature risks (> 40 °C) and Sylhet showing heightened minimum temperature risks. The MLP model achieved the lowest errors across ecosystems, with MSEs of 0.82 (Barind), 1.47 (Coastal), and 1.50 (Haor) for maximum temperature and with MSEs of 0.48 (Barind), 0.44 (Coastal), and 0.48 (Haor) for minimum temperature, outperforming SVM, CNN, LSTM, ANN, RF, and Ensemble models. This is the first region-specific application of machine learning models along with Mann-Kendall trend analysis, Moran's I spatial statistics for rice production in Bangladesh which provides a multidimensional framework that is rarely applied in Bangladesh. These findings underscore the urgent need for region-specific climate adaptation strategies, as rising temperatures threaten rice production and agricultural resilience.