[Ozone Sensitivity Analysis in Urban Beijing Based on Random Forest].

ABSTRACT

The basis and key step to developing ozone (O3) prevention and control measures is determining the non-linear relationship between O3 and its precursors. Based on online observations of O3, volatile organic compounds (VOCs), nitrogen oxides (NOx), and meteorological elements from April to September 2020 at an urban site in Beijing, we analyzed the pollution characteristics of O3 and its precursors, explored key factors affecting O3 using the random forest (RF) model combined with SHAP values, and explored the O3-VOCs-NOx sensitivity through a multi-scenarios analysis. The results of correlation analysis showed that the hourly concentration of O3 was significantly positively correlated with temperature (T) and negatively correlated with TVOCs and NOx. However, in terms of the daily values, O3 was significantly positively correlated with T, TVOCs, and NOx. The simulated O3 values by the RF model agreed with the measured values. The SHAP values of each characteristic variable were further calculated. The results suggested that T and NOx showed the two highest effects on O3, with positive and negative values, respectively. Based on the average NOx and VOCs on O3 pollution days during the observation period (the base scenario), multi-scenarios with different NOx and VOCs were set up. The RF model was used to calculate O3 under different scenarios and obtain the O3 isopleth (EKMA curve). The results showed that the O3-VOCs-NOx sensitivity in urban areas of Beijing was in the VOCs-limited regime, which was consistent with the results obtained from the observation-based box model(OBM). This indicated that the RF model could be used as a complementary method for O3-VOCs-NOx sensitivity analysis.