Recent studies have demonstrated that the accuracy of numerical weather prediction (NWP) models in solar radiation forecasting can be enhanced by incorporating machine learning (ML) techniques. However, ML models are inherently dependent on historical data, which raises concerns about their ability to adapt to evolving atmospheric conditions over time. In this study, we investigate the temporal robustness and generalizability of ML-based solar radiation forecasting models by utilizing long-term reanalysis datasets. Our objective is to evaluate how well these models maintain performance across different time periods and to identify potential limitations in their adaptability to changing climate patterns.
