Volume 66

Robust Trends and Seasonal Attribution of Global Solar Radiation at Multi?Stations in Long-Term Monthly Series: A GAM-Based Assessment of Meteorological Drivers Xuexue NIE, Changying Xiang

Abstract

To support urban energy planning and data-driven assessment, this paper proposes a comprehensive framework for robust trend analysis and seasonal meteorological attribution of multi-stations, long-term, monthly global solar radiation (GSR) data, elucidating the seasonal heterogeneity. Firstly, we conduct anomaly detection to ensure data robustness. Subsequently, long?term trends are analysed using the Mann–Kendall test and Sen’s slope estimator, while the same calendar?month variability is quantified by the Coefficient of Variation (CoV). The CoV results indicate lower variability at the urban GSR station compared to the coastal GSR station, with both stations displaying similar seasonal patterns — higher variability from November to May and reduced variability from June to October. After that, we apply Generalised Additive Model (GAM) P-splines to model the Pooled Anomaly (PA) and Seasonaldisaggregation and evaluate the performance of the two models using leave-one-year-out (LOYO) cross?validation. The PA model attains a Q2-LOYO value of 0.72, whereas the Season model performs best in spring ( Q2-LOYO= 0.83), with relatively lower performance observed during summer and autumn compared to the PA model. The attribute results indicate cloud cover and temperature are both key factors: Cloud cover exerts a significantly negative influence on GSR, with the strongest inhibitory effect observed during winter, whereas temperature demonstrates the highest positive sensitivity in autumn. The impacts of these variables in spring and summer are comparable in magnitude. Relative humidity is unstable in summer, making its independent effect hard to identify, while a significant negative trend in other seasons. Rainfall consistently exhibits a negative association across all seasons, with the most substantial suppressive effect observed in winter. Air pressure and wind speed generally serve as marginal contributors to GSR variability.

Keywords GSR, Meteorological, Robustness, Long-term Trend, Seasonal, Machine Learning, CoV, GAM