PM_2.5 Vulnerability Areas Analysis Using Deep Neural Network: Application to Seoul City

Fine particulate matter (PM_2.5), defined as particles with a diameter of 2.5 micrometers or less, poses significant health risks as it can penetrate alveoli upon inhalation. Research on the vulnerability to PM_2.5 has predominantly focused on indicators derived from the Intergovernmental Panel on Climate Change (IPCC) vulnerability assessments on health in the context of climate change (2007). This study aims to analyze PM_2.5 vulnerability areas using a Deep Neural Network (DNN) model. The research area is set in Seoul, South Korea, characterized by high population density and a basin-like topography that inhibits atmospheric dispersion, thus presenting environmental vulnerability to air pollution. In the DNN model, the dependent variable is defined as the PM_2.5 vulnerability areas, which correspond to currently designated intensive management zones for PM_2.5. For the independent variables, data were collected based on the criteria for the designation of these management zones, including the annual average concentration of PM_2.5, the number of high PM_2.5 concentration days, the annual average concentration of PM₁₀, the number of high PM₁₀ concentration days, demographics of vulnerable populations aged 65 and older and under 15, facilities utilized by these vulnerable groups, sources of particulate matter emissions, and pollution sources from industrial activities, totaling ten datasets. Through a data preprocessing process, the study established a spatial database structured in a 100 m × 100 m grid across Seoul, with the independent variables serving as attribute values. The DNN model was quantitatively evaluated using performance metrics, and based on these results, a vulnerability map for PM_2.5 in Seoul was visualized and analyzed. The findings of this study are anticipated to serve as foundational data for the Ministry of Environment’s policy formulation aimed at data and AI-based environmental management.