Comparison of different hybrid modeling methods to estimate intraurban NO2 concentrations

Abstract

Exposure to air pollution has a significant impact on the health of urban populations, so the improvement of methods that model the concentrations of air pollutants within complex urban areas is important in health studies to adequately asses the exposure of the population. This paper presents several hybrid, high-resolution models to simulate the variability of ambient NO2 concentrations in Seoul, the capital of South Korea. These models combine the Community Multiscale Quality (CMAQ) as a regional photochemical model with a fine scale model of either the California Puff dispersion model (CALPUFF) or the land use regression model (LUR). We compared high-resolution estimates of the spatial NO2 concentration from four different hybrid models, including 1) raw CMAQ-CALPUFF; 2) observation-fused CMAQ-CALPUFF; 3) raw CMAQ-LUR; and 4) observation-fused CMAQ-LUR. We conducted numerical simulations of the NO2 concentrations during the winter season and compared the results with field data obtained from mobile measurements captured from December 2017 to February 2018. The results indicate that observation-fused hybrid models offered improved agreement with the mobile measurements: for the CMAQ-CALPUFF model, statistical bias and error were reduced to about 82% and 57%, respectively by using observation-fused CMAQ. We also found significant differences in the sub grid variability of the NO2 concentrations for the different hybrid models. The predictions obtained with CMAQCALPUFF showed concentrations that were more widely distributed (1.7 and 1.4 times for the 10-90th range, observation-fused case) when compared to the only-CMAQ and CMAQ-LUR predictions, respectively. Our study suggests that a properly evaluated hybrid model can increase the predictive accuracy of air pollutant concen-tration in complex urban areas to improve exposure assessments in health studies.