This research introduces an innovative approach that integrates attention mechanisms into recurrent neural networks (RNNs) to enhance the modeling of temporal dependencies for wind speed downscaling. Global Climate Models (GCMs) typically produce outputs with a broad spatial resolution, which is inadequate for detailed local-scale renewable energy assessments. Focusing on Pune city, this study utilizes attention-augmented Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU) models to address these issues. The proposed methods significantly improve prediction accuracy and provide interpretability through visual attention weights, facilitating the identification of crucial historical data. The results indicate that these models outperform traditional RNN models, highlighting the suitability of attention mechanisms for precise wind resource forecasting.