Future air quality predictions using deep learning algorithms

Future air quality under a changing climate is a critical concern due to the substantial risks to public health and the environment. Accurately predicting these changes is essential for developing effective strategies, yet the inherent uncertainties present a challenge. Traditional approaches rely on global and regional climate-chemistry models, but these methods are computationally intensive at high spatiotemporal resolutions and often contain biases. To address this issue,recent studies have explored deep learning (DL) techniques for bias correction and downscaling, offering the potential to mitigate computational costs and enhance accuracy compared to traditional methods. We used the WRF-GC model with Shared Socioeconomic Pathways (SSPs) emissions and observational data to train and evaluate various deep learning models for 2015-2023. We then apply these models to project future air quality, including ozone and particulate matter concentrations at 138 sites across the Korean Peninsula. This will provide crucial insights into the health impacts of future changes and, more importantly, support the formulation of effective adaptation policies.