Comparative Analysis of SPI Index for Drought Conditions in North-West Bangladesh: A Study of CMIP6 Model Data and Machine Learning-Based Predictions

Rising global temperatures have been linked to changes in rainfall patterns and an increase in extreme rainfall-related weather events worldwide. Because of its fluctuating precipitation, Bangladesh, a nation susceptible to natural catastrophes, has experienced and will continue to confront more catastrophic calamities. Among these hazards, drought is a more concerning issue for an agriculture-dependent country like Bangladesh. This study has addressed this issue and analyzed the drought condition for the historical period (1985–2014) and near future (2025–2054) by estimating the SPI index in the north-west region of Bangladesh. In this study, an investigation has been done for future projections under various scenarios, such as SSP-245 and SSP-370, using seven suitable Coupled Model Intercomparisons Project 6 (CMIP6). Also, the SPI index has been predicted using a feed-forward backpropagation algorithm in an artificial neural network (ANN). This study has compared the results from two analyses (7 CMIP6 models) and machine-learning-based predictive output. For this study, the drought index was determined to be the Standard Precipitation Index (SPI) on three timescales: three months, six months, and twelve months. For the analysis, a three-layer artificial neural network model was used. In order to determine the most accurate predictive model for the SPI, this model was trained utilizing the SPI timescales with varying lag durations. The correlation coefficient indicated a high accuracy range (75%–85%) in predictive values, demonstrating the model's effectiveness. Additionally, the comparison of observed versus predicted curves for the SPI index across the three timescales also revealed similar trends. The SPI index, derived from 7 CMIP6 models, shows that in the near future, drought events for SSP-370 scenarios are more frequent than SSP-245 scenarios. For the historical period, Chirps precipitation data has been used along with CMIP6 model data, and it has also shown an increasing trend in drought frequencies with time. This study has analyzed the historical and future drought conditions, which can benefit policymakers by improving infrastructure for giving early warning to farmers and taking necessary precautions to protect the losses due to drought.