The increasing variability and scarcity of historical rainfall data pose significant challenges for climate resilience planning, especially in data-constrained regions. This paper proposes a novel generative adversarial network (GAN) framework for the synthesis of high-resolution spatiotemporal rainfall data. Leveraging a hybrid architecture composed of convolutional LSTM blocks and conditioning vectors derived from climatic zones and seasonal metadata, the model captures complex temporal patterns and spatial heterogeneity inherent in long-term rainfall records. Experiments were conducted on a century-long Indian rainfall dataset, with model performance evaluated using RMSE, SSIM, Kullback–Leibler Divergence, Fréchet Inception Distance, and Pearson correlation. The proposed model demonstrated superior realism and structural consistency when compared with baseline models including LSTM, WGAN, and stochastic weather generators. Ablation studies revealed the critical role of adversarial loss and temporal modeling in enhancing fidelity. The framework offers promising utility for generating synthetic rainfall inputs in hydrological simulation, flood risk assessment, and scenario-based adaptation planning. Ethical considerations surrounding uncertainty propagation, transparency, and equitable access to synthetic data are also addressed.