Urban Air Quality Shifts in China: Application of Additive Model and Transfer Learning to Major Cities.

Abstract

The impact of reduced human activity on air quality in seven major Chinese cities was investigated by utilizing datasets of air pollutants and meteorological conditions from 2016 to 2021. A Generalized Additive Model (GAM) was developed to predict air quality during reduced-activity periods and rigorously validated against ground station measurements, achieving an R² of 0.85-0.93. Predictions were compared to the observed pollutant reductions (e.g., NO₂ declined by 34% in 2020 vs. 2019), confirming model reliability. Transfer learning further refined the accuracy, reducing RMSE by 32-44% across pollutants when benchmarked against real-world data. Notable NO₂ declines were observed in Beijing (42%), Changchun (38%), and Wuhan (36%), primarily due to decreased vehicular traffic and industrial activity. Despite occasional anomalies caused by localized events such as fireworks (Beijing, February 2020) and agricultural burning (Changchun, April 2020), our findings highlight the strong influence of human activity reductions on urban air quality. These results offer valuable insights for designing long-term pollution mitigation strategies and urban air quality policies.