Fine spatiotemporal prediction of ozone in the Beijing-Tianjin-Hebei based on random forest
Since 2013, China has been focusing on air pollution and has achieved significant results. The concentrations of various air pollutants, including PM2.5, have been continuously decreasing and stabilizing at lower levels. However, in recent years, the concentration of ozone has been continuously increasing, becoming a new challenge for atmospheric environmental governance in China. The Beijing-Tianjin-Hebei region is currently experiencing severe ozone pollution. Predicting the spatiotemporal concentration of ozone in this region can help understand the changing patterns and main influencing factors of ozone. To address these issues, this study employs the random forest method to conduct fine-scale spatiotemporal predictions of the maximum 8-hour ozone concentration in the Beijing-Tianjin-Hebei region for the year 2021 at a 1 km resolution. Using the random forest method, the importance of variables affecting ozone concentration was obtained, and the main factors affecting ozone concentration were analyzed. The results of 1km ozone spatiotemporal distribution were used to analyze the distribution characteristics of ozone in different seasons. Moreover, the study quantitatively calculated the number of days with excessive ozone levels and population exposure. The results showed that: (1) The random forest method accurately predicted the high spatial resolution ozone distribution in the Beijing-Tianjin-Hebei region, demonstrating stable and reliable performance at both annual and seasonal scales. By comparing the prediction accuracy of this study with existing studies, it was found that this study achieved good results in spatial resolution and modeling accuracy, with an R2 of 0.84 and a spatial resolution of 1 km. (2) In the Beijing-Tianjin-Hebei region, the main meteorological factors affecting ozone concentration are temperature and solar radiation intensity, which are consistent with the photochemical reaction principle of ozone. In winter, ozone concentration is greatly affected by wind speed. (3) Ozone concentration shows fluctuating characteristics by season, with summer>spring>autumn>winter. The spatial distribution also shows obvious seasonal characteristics, with spring, summer, and autumn showing high concentrations in the southeast and low concentrations in the northwest, while winter is the opposite. (4) High ozone concentration areas are concentrated in the east and south. The distribution of days exceeding the standard is similar to the spatiotemporal distribution pattern of ozone concentration. Within a 1 km × 1 km range, the highest number of days exceeding the standard throughout the year is 65 days, meaning that 17.8％ of the days in a year have an ozone concentration exceeding 160 days μg/m3. Cities such as Baoding, Beijing, Shijiazhuang, and Zhangjiakou have significant population concentrations in areas with high ozone concentrations, while Langfang has a significant population concentration in areas with low ozone concentrations. In conclusion, the ozone spatiotemporal prediction method based on random forests has high accuracy and can provide valuable insights for high-resolution ozone spatial prediction at regional and urban scales, thereby saving substantial atmospheric monitoring costs. It not only enhances our understanding of ozone pollution dynamics but also offers suggestions for mitigating the adverse impacts of ozone pollution on human health and the environment.
NETL
NSTL
万方数据
