The distribution of PM2.5 concentrations at the neighborhood scale is interfered with not only by spatial morphology, but also by spatial functions, and this effect shows heterogeneity in different urban locations due to differences in spatial dynamics. Taking the area within the Fourth Ring Road of Beijing in 2020 as an example, this paper provides urban planners with an evaluation framework for fine-grained prediction of neighborhood-scale PM2.5 distribution based on the Pix2PixGAN model and the multi-source data with crowd-source data at its core. Through the crowd-source data, the spatial functions and corresponding spatial dynamics in the neighborhood scale can be characterized in detail. Then the framework utilizes Geographically weighted poisson regression (GWPR) to screen spatial functions with significant geographic correlation with PM2.5 distribution and their heterogeneity coefficients. Based on the filtered functions, heterogeneity coefficients, and PM2.5 distribution raster, the dataset of Pix2PixGAN model is composed, and the optimal model with the most accurate PM2.5 distribution prediction is trained and selected. Finally, the paper illustrates how to analyze the qualitative and quantitative impacts of planning proposals on the distribution of PM2.5 through the optimal model, using six planning scenarios as examples.
Read full abstract