Long-term dynamics and machine learning-based mapping of night sky brightness: snowglow, urbanization, and topography effects at major astronomical observatories in Türkiye

Anthropogenic alterations to the nocturnal environment pose a fundamental effect to the optical viability of ground-based astronomical observatories. This study presents a comprehensive spatiotemporal assessment of light pollution dynamics affecting Türkiye’s premier research facilities, the Eastern Anatolia Observatory (DAG) in Erzurum and the TÜBİTAK National Observatory (TUG) in Antalya within the Türkiye National Observatories, situated in contrasting highland and coastal topographies. To bridge the scale gap between the remote sensing and ground-based site testing, we developed a traditional regression and a hybrid machine learning regression Kriging framework. These approaches synthesized high-resolution, high-quality radiance data from the Suomi-NPP VIIRS with extensive in-situ sky quality meter measurements. A rigorous methodological comparison revealed that traditional geostatistical methods (Ordinary/Universal Kriging) suffered from smoothing effects and overfitting in complex terrains. Furthermore, while advanced algorithms like support vector regression achieved high statistical scores, they produced physically implausible extrapolations in data-sparse regions. In contrast, the proposed random forest regression Kriging model proved superior, achieving robust validation accuracy (R2 = 0.74 and 0.80) while maintaining strict physical consistency with natural sky brightness limits. Temporal analyses in the study demonstrated that the high-altitude Erzurum plateau exhibited a distinct seasonal cycle with the snowglow effect, where surface snow cover increased human-induced skyglow by > 1.5 times. In contrast, the Antalya coast exhibited a monotonous deterioration independent of seasonality, triggered by rapid tourism-oriented urbanization.