Real-time estimation of PM2.5 concentrations at high spatial resolution in Busan by fusing observational data with chemical transport model outputs

Abstract

PM2.5 pollution has attracted a lot of attention recently as the adverse effects on health it can cause are being recognised. Ambient PM2.5 data are usually obtained from two sources, the first source is observations collected from air quality monitoring networks, unfortunately, this data tends to be quite sparse due to the lack of coverage by these monitoring networks. The second source is the outputs from numerical modelling, having high spatial resolution at the level of grid cell, unfortunately with this approach little information on the uncertainty of those predicted values is available. In this study, we tackled these problems by developing a data fusion approach to predict PM2.5 concentrations on an hourly basis with high spatial resolution, this was done by regressing the observed data with the 1 km2 CMAQ-derived spatial coefficients after obtaining site-pairs from a seasonal cluster analysis. When applied to two extensive observational datasets (25 AQM sites and 21 mobile AQM sites) over a period from January 2019 to June 2020 these fusion-driven hourly PM2.5 predictions substantially outperformed (R = 0.80, RMSE = 8.23) the raw numerical model simulations (R = 0.55, RMSE = 12.91). Adopting an experimental bias-correction process further improved the prediction performance (R = 0.84, RMSE = 7.59). Applying this fusion approach to real-time PM2.5 predictions for the same 25 individual air quality monitoring sites over five months (July to November 2020) showed the excellent predictive power (R = 0.82, RMSE = 8.74) of our approach.