A quantitative exploration of the interactions and synergistic driving mechanisms between factors affecting regional air quality based on deep learning

Abstract

Air pollution represents a significant environmental challenge on a global scale. The exploration of factors and mechanisms influencing air quality has garnered considerable attention and produced a substantial body of research. Nonetheless, the absence of effective research tools has resulted in a limited understanding of the nonlinear interactive effects of multiple factors on air quality, as well as their patterns and spatiotemporal variability characteristics. Furthermore, the efficacy of multifactor joint control strategies in practical air pollution management remains inadequate. A novel regional air environmental quality prediction model was developed utilizing a neural network with long short-term memory (LSTM) in deep learning based on an analysis of the spatiotemporal heterogeneity of regional air environmental quality in the Yangtze River Economic Belt (YEB). The model results were attributed and interpreted by integrating the SHapley Additive explanation (SHAP) method, focusing on the possible interactive effects and synergistic driving mechanisms of PM2.5 and O3 as the main pollutants and other factors on the air quality index (AQI) in a quantitative way, respectively. The SHAP interaction values of five factor pairs, namely PM2.5 and temperature, PM2.5 and precipitation, O3 and temperature, O3 and NO2, and O3 and PM2.5, are depicted in dependence scatter plots that exhibit three distinct trajectory shapes, namely linear, U-shaped, and irregular. The linear pattern of the dependence scatter plots for PM2.5 and precipitation in the YEB, midstream, and downstream regions, as well as for O3 and temperature in the upstream region, is evident. The irregular patterns of PM2.5 and precipitation, O3 and NO2, O3 and PM2.5 in the upstream, and PM2.5 and temperature in the downstream are observed to be correspondingly related. Additionally, a U-shaped pattern is exhibited by O3 and temperature in the YEB, midstream and downstream, PM2.5 and temperature in the upstream and midstream, O3 and NO2 in the YEB, midstream and downstream, and O3 and PM2.5 in the YEB. Of the three models, the U-shaped model's identification of inflection points and corresponding quantitative determination of factor thresholds hold greater practical implications for decision-making in air environment management. This pattern suggests that the synergistic effect of factor pairs initially decreases with changes in each factor but subsequently begins to increase after reaching inflection points. It is anticipated that this research will contribute to the advancement of theoretical inquiry into the nonlinear multifactor interaction effect and synergistic driving model of regional air quality. Additionally, it is expected to furnish a valuable foundation for decision-making in the development of joint air pollution prevention and control strategies aimed at addressing compound air pollution issues.