Using Bayesian optimization method and FLEXPART dispersion model to evaluate CO emission in East China based on three-year measurements at high altitude mountain site

The rapid economic growth and energy consumption in East China has resulted in serious regional-scale and trans-boundary air pollutions. Variation of carbon monoxide (CO) is of great interest as a restriction factor for pollutants related with incomplete combustion processes. This study attempted to evaluate CO emission in East China using analytical inverse modeling and nearly three-year measurements (2005-2007) at Mt. Huashan (34°28'52"N, 110°04'36"E, a.s.l.1577 m). Inversion methodology combined FLEXPART_WRF (Version 6.2) with Bayesian optimization iterative algorithm. Regarding calculations, five days backward simulations of air mass movement were used to determine the Source-Receptor Relationship (SRR) between emissions at the source regions and observation at receptor. And then SRR, a priori CO emission (INTEX-B) and corresponding uncertainties (70%) were fed into inversion algorithm to optimize the CO emission by minimizing the mismatch between simulated and observed CO concentrations. To reduce the number of unknowns (a posterior), we used variable-resolution grid setting with high spatial resolution in the vicinity of observation site and coarse resolution for the regions some distance away due to decrease of SRR values. Inversion result showed that simulation of CO mixing ratio with the a posterior information was evidently improved, reducing root-mean square (RMS) of differences between observation and simulation by 30%. A posterior indicated that CO emission in East China was probably 10%, 25% and 6% of underestimated by emission inventory for spring, autumn and winter time. Spatial distribution of a posterior indicated that North China Plain (30°N-40°N, 110°E-120°E) significantly accounted for the total increase of CO emission in the East China; Nevertheless, the CO emission from South China region (20°N-30°N, 100°E-120°E) might be overestimated ~10% annually. Uncertainties analysis with repeated random samplings approach demonstrated that inversion result was relatively sensitive to the emission from the regions where city and industries were concentrated, highlighting the great variability of CO emission from anthropogenic sources.