Gap-Filling of Hourly Carbon Monoxide Observations in Air Korea Stations Using Meterological Data and Machine Learning

Carbon monoxide (CO) is a major air pollutant primarily emitted in urban areas with high traffic volumes and has significant impacts on air quality. CO concentration changes are determined by complex factors such as meteorological conditions, emission sources, and photochemical reactions. Although accurate time series analysis is essential, obtaining a complete dataset is challenging due to equipment malfunctions and data collection issues, especially in high-resolution hourly data. Such incomplete time series data with missing values can decrease the accuracy of air quality prediction models and reduce the reliability of short and long-term atmospheric pattern analyses. To address these challenges, this study developed a machine learning-based gap-filling model for CO concentrations on a nationwide scale in South Korea. The research utilized hourly data from 2020 to 2023, using national air pollution monitoring network data provided by Air Korea and meteorological variables from the Automated Synoptic Observing System (ASOS). To select key input variables, correlation analysis between CO and other air pollutants was conducted, revealing the highest correlation with nitrogen dioxide (NO₂), a known co-pollutant (cc=0.60). This strong relationship exists because both substances share common emission sources such as traffic and fossil fuel combustion, and exhibit similar diffusion patterns due to changes in the atmospheric boundary layer. Particulate matter (PM₂.₅) also showed a significant correlation with CO (cc=0.51) and was included as a key input variable. Further correlation analyses between CO and meteorological variables were conducted, and statistically significant parameters were integrated into the dataset to enhance the quality of data matching. Using this constructed match-up data, machine learning-based gap-filling for CO concentration was performed. The reliability of the model was evaluated through validation using actual measurements from 2024. The AI-based gap-filling model developed in this study overcomes the limitations of existing spatial interpolation techniques such as kriging, enabling the acquisition of complete and continuous CO time series data. This is expected to contribute to various atmospheric environmental research areas including air quality modeling, pollution source management policy development, and understanding of atmospheric chemistry processes. Furthermore, this research methodology can be applied to reconstruct missing values for other air pollutants, potentially aiding in the establishment of a comprehensive air pollution monitoring system.