Short-term rainfall prediction is crucial in regions like West Africa, where unpredictable climate fluctuations between droughts and heavy rainfall complicate water and resource management. Accurate hourly forecasts can enhance water allocation for agriculture, drinking water, and irrigation, while improving disaster preparedness for floods and droughts. This study introduces a hybrid approach using LSTM networks and TimeGAN to address traditional forecasting limitations. By generating synthetic data with TimeGAN, the model tackles data scarcity and improves prediction robustness. Results show that models trained on synthetic data perform as well as or better than those trained on real data, demonstrating the value of synthetic data augmentation. This approach has significant potential for water management, agriculture, and disaster mitigation, offering scalable solutions to build resilience against extreme weather. The study highlights the effectiveness of combining generative models with recurrent neural networks for accurate, adaptive forecasting systems tailored to local needs.