This research aims to enhance the accuracy of solar irradiance estimations by combining satellite data and on-site measurements. Advanced deep learning models, including recurrent neural network (RNN), long short-term memory (LSTM), and gated recurrent units (GRU), were utilized to achieve this goal. The results yielded promising outcomes, with the root mean square error ranging from 96 to 97 W/m2 across various models. Notably, the RNN model demonstrated exceptional performance with the lowest relative root mean square error (rRMSE) at 36.22%. This indicates that the RNN model can effectively identify intricate connections within the input data, thereby generating precise global horizontal irradiance estimates. These findings carry significant implications for renewable energy applications as they underscore the potential of deep learning models in refining solar irradiance estimation processes.