Long‐Term Trends and Machine Learning‐Based Predictions of Atmospheric Visibility in the Largest Industrial City of South Korea

ABSTRACT Atmospheric visibility serves as a practical indicator of air quality, yet it is often overlooked in environmental evaluation. Therefore, this study investigates long‐term visibility trends in Ulsan, a multi‐industrial city of South Korea, from 2015 to 2020, and explores major influencing factors. Hourly visibility data were analyzed alongside concentrations of criteria air pollutants (PM 2.5 , PM 10 , NO 2 , SO 2 , CO, and O 3 ) and meteorological parameters (temperature, relative humidity, and wind speed). The mean visibility in Ulsan was 16.9 ± 2.04 km, with a slight but statistically significant increase of 0.12 km/year. Correlation and mutual information analyses identified relative humidity and concentrations of PM 2.5 and PM 10 as the dominant drivers of visibility reduction. Visibility was persistently enhanced in all seasons except winter despite reductions in PM 2.5 and PM 10 concentrations. Three machine learning models, including random forest (RF), multiple linear regression (MLR), and multilayer perceptron (MLP), were developed to improve visibility forecasting. Among these, the RF model achieved the highest predictive accuracy ( r 2 = 0.74), outperforming MLR and MLP. The study highlights the need for further investigation into the complex interplay between air quality and meteorological conditions and demonstrates the potential of machine learning for accurate predictions of visibility.