Modeling the impacts of 2D/3D urban structure on PM2.5 at high resolution by combining UAV multispectral/LiDAR measurements and multi-source remote sensing images

Abstract

Understanding the impacts of 2D/3D urban structure on PM2.5 is critical for protecting resident health and sustainable development. However, accurate understanding is limited by the lack of high-resolution PM2.5 simulation in cities with sparse monitoring stations and the simultaneously consideration of different landscape types both in 2D and 3D directions. Using Yinchuan City as a case study, an integrated simulation model combining satellite AOD, NDVI and meteorological data was developed using random forest (RF) algorithm to simulate city-scale PM2.5 variation at 30 m resolution. Then, the most critical 2D/3D urban metrics in altering PM2.5 and their impacts were explored through RF analysis. Finally, optimal urban structure was identified using Bayesian Network and multi-scenario analysis. The results indicated that 1) the established RF model could effectively simulate within-city variation of PM2.5 at 30 m with an R2 of 0.75. Thus, it is feasible to use the abundant temporal information and high-resolution AOD to map city-scale PM2.5 and overcome the limitation of spare monitoring stations. 2) The emission-related metric of distance to polluting enterprise and city center had the largest impacts on PM2.5, followed by NDVI, building height, impervious surface area and shape, and UGS green volume, area, shape and connectivity. Optimized 3D urban structure for reducing PM2.5 was proposed from the three aspects of emission source location, UGS pattern and impervious surface. This study deepens the understanding of how 2D/3D urban structure impact PM2.5 and provides scientific references for optimization possibility of urban air quality in both horizontal and vertical directions.