Climate change induced by global warming has altered precipitation patterns across India. In this study, we introduce a novel methodology for forecasting rainfall in a state by utilizing precipitation data from other states that experienced earlier rainfall within the same year. The model employed integrates Long Short-Term Memory (LSTM) networks, vector autoregression (VAR), and Extreme Value Theory (EVT) to capture rainfall patterns across various states of India. These components of the model are used for specific purposes: LSTM captures long-term patterns over time; VAR facilitates the integration of rainfall data from multiple states; and EVT focuses on predicting extreme precipitation events. This methodology involves analyzing precipitation data from states that experienced rainfall earlier in the year than the target state. The combined LSTM–VAR–EVT model utilizes both the precipitation data of the target state and rainfall data from the surrounding states to generate a prediction for the target state. This integrated approach significantly enhances the forecast accuracy compared with the application of any single model in isolation. Consequently, this method is particularly valuable for agriculture, water-resource management, and flood mitigation in regions where precipitation patterns follow similar temporal sequences across states.