Accurate short-term forecasting of solar irradiance and wind speed is pivotal for renewable energy scheduling and grid stability in India, where climatic diversity complicates resource prediction. This paper presents a hybrid deep learning framework that combines convolutional neural networks (CNNs) and long short-term memory (LSTM) layers within a transfer learning (TL) formulation to improve forecast accuracy in data-scarce regions. Hourly meteorological variables from the NASA POWER database are used, with Jaipur (2019–2024) serving as the source site and Chennai (2023–2024) as the target site. The model is first pre-trained on the Jaipur dataset and then fine-tuned on Chennai data by freezing lower-level layers and updating higher-level layers. Experimental results demonstrate strong cross-regional generalization, achieving a daytime solar mean absolute percentage error (MAPE) of 10.34% and a wind MAPE of 5.53% on the Chennai test set. Compared with a day-ahead persistence baseline, the proposed method reduces daytime solar MAPE by 59% and wind MAPE by 83%, indicating that transfer learning can substantially reduce forecasting uncertainty when local training data are limited.

Keywords- Solar irradiance forecasting, wind speed forecasting, transfer learning, CNN–LSTM, NASA POWER, India.