A Novel Hybrid Framework for Annual and Quarterly Rainfall Prediction Combining CNN and LSTM Networks

Accurate rainfall prediction is crucial for agriculture, water resource management, and disaster preparedness. This study explores the application of machine learning techniques, including RNN, LSTM and GRU models, to predict annual and quarterly rainfall using historical rainfall data. Additionally, a novel hybrid approach combining CNN with LSTM is proposed to enhance prediction accuracy. The CNN layers extract high-level features from monthly rainfall data, while the LSTM layers capture temporal dependencies, making it suitable for time-series predictions. The models were evaluated on a dataset comprising rainfall data from 1901 to 2015, normalized for consistency. Predictions were made for annual and quarterly rainfall totals, leveraging monthly rainfall as input features. Comparative analysis revealed that the CNN-LSTM model outperformed individual RNN, GRU, and standalone LSTM models in terms of mean absolute error (MAE) and overall prediction accuracy. The CNN-LSTM model’s ability to effectively combine feature extraction and sequence modeling contributed significantly to its superior performance.