Potential role of urban forest in removing PM_2.5: A case study in Seoul by deep learning with satellite data

PM_2.5 has been considered as a crucial issue in densely populated cities because of its adverse effects on human health. As a way to mitigate this problem on city scale, urban forest is recently suggested as a solution due to its potential to absorb and adhere PM_2.5. However, limited observations in the forest prevent a full understanding in the role of forest on urban PM_2.5 concentration. In this paper, we present an application of deep learning using satellite data to clarify the role of forest with respect to the PM_2.5 concentration mitigation in the city. Based on PM_2.5 data collected from the ground-based observation stations, two PM_2.5 prediction models are constructed through the deep learning approach. Satellite-derived aerosol optical depth from the GOCI and MODIS are used as the main predictors of two models. Both models predict spatial distribution throughout Seoul, Korea (R² of 0.61 and 0.78). Particularly, the western parts of Seoul tend to have higher PM_2.5 concentrations than the other parts, whereas mid-outskirts have noticeably low concentrations. A comparison of spatial distributions of PM_2.5 between observations and predictions show that the existing observation network only reflects 15%–60% of entire characteristics of Seoul. Utilizing city-wide PM_2.5 estimations, a comparison of estimated PM_2.5 between urban and forest regions has been performed. The results show that the estimated PM_2.5 concentrations in the forest regions are less than those in urban regions by up to 16.4 μg m⁻³. In contrast to urban regions, the average of PM_2.5 concentrations in the forest regions is likely below the daily mean WHO PM_2.5 outdoor standard. Our study suggests that urban forest could be a potential way to improve urban air quality with a specific focus on PM_2.5. In addition, the deep learning approach used in this study can be applied to other cities where obtaining observation measurements is difficult.