Deep learning-based forecasting of daily maximum ozone levels and assessment of socioeconomic and health impacts in South Korea

Abstract

Accurate forecasting of ground-level ozone (O₃) is essential for assessing its public health and socioeconomic impacts. This study evaluates the performance of three deep learning models—Deep Convolutional Neural Networks (Deep-CNN), Long Short-Term Memory (LSTM), and Deep Neural Networks (DNN)—in forecasting daily maximum O₃ concentrations across all 19 provinces of South Korea for a seven-day period. Among the models, Deep-CNN demonstrated superior accuracy on forecast day 1, achieving an Index of Agreement (IOA) of 0.93, outperforming LSTM (IOA = 0.92) and DNN (IOA = 0.86). This improved performance is attributed to Deep-CNN's ability to capture spatial-temporal features relevant to O₃ dynamics. A novel contribution of this study is the integration of high-accuracy O₃ forecasts with province- and gender-specific health and socioeconomic indicators to assess environmental impacts. Pearson's correlation coefficient (r) and Spearman's rank correlation coefficient (ρ), along with their associated p-values, were used to evaluate the strength, direction, and significance of these associations. Significant correlations were found between O₃ and female respiratory mortality (r = 0.53, ρ = 0.42; p = 0.020, 0.024), cardiovascular mortality in both genders, and male employment (r = 0.48, ρ = 0.76; p = 0.039, 0.0002). Female employment showed weaker linear correlation (r = 0.42, p = 0.061), but a strong monotonic trend (ρ = 0.74, p = 0.0003). By linking deep learning-based air quality forecasting with health and socioeconomic outcomes, this study provides critical insights for policymakers aiming to mitigate O₃-related risks and promote health equity across demographic groups.