Top-down estimate of black carbon emissions for city clusters using ground observations: a case study in southern Jiangsu, China
We combined a chemistry transport model (the Weather Research and Forecasting and the Models-3 Community Multi-scale Air Quality Model, WRF/CMAQ), a multiple regression model, and available ground observations to optimize black carbon (BC) emissions at monthly, emission sector, and city cluster level. We derived top-down emissions and reduced deviations between simulations and observations for the southern Jiangsu city cluster, a typical developed region of eastern China. Scaled from a high-resolution inventory for 2012 based on changes in activity levels, the BC emissions in southern Jiangsu were calculated at 27.0 Gg yr−1 for 2015 (JS-prior). The annual mean concentration of BC at Xianlin Campus of Nanjing University (NJU, a suburban site) was simulated at 3.4 µg m−3, 11 % lower than the observed 3.8 µg m−3. In contrast, it was simulated at 3.4 µg m−3 at Jiangsu Provincial Academy of Environmental Science (PAES, an urban site), 36 % higher than the observed 2.5 µg m−3. The discrepancies at the two sites implied the uncertainty of the bottom-up inventory of BC emissions. Assuming a near-linear response of BC concentrations to emission changes, we applied a multiple regression model to fit the hourly surface concentrations of BC at the two sites, based on the detailed source contributions to ambient BC levels from brute-force simulation. Constrained with this top-down method, BC emissions were estimated at 13.4 Gg yr−1 (JS-posterior), 50 % smaller than the bottom-up estimate, and stronger seasonal variations were found. Biases between simulations and observations were reduced for most months at the two sites when JS-posterior was applied. At PAES, in particular, the simulated annual mean declined to 2.6 µg m−3 and the annual normalized mean error (NME) decreased from 72.0 % to 57.6 %. However, application of JS-posterior slightly enhanced NMEs in July and October at NJU where simulated concentrations with JS-prior were lower than observations, implying that reduction in total emissions could not correct modeling underestimation. The effects of the observation site, including numbers and spatial representativeness on the top-down estimate, were further quantified. The best modeling performance was obtained when observations of both sites were used with their difference in spatial functions considered in emission constraining. Given the limited BC observation data in the area, therefore, more measurements with better spatiotemporal coverage were recommended for constraining BC emissions effectively. Top-down estimates derived from JS-prior and the Multi-resolution Emission Inventory for China (MEIC) were compared to test the sensitivity of the method to the a priori emission input. The differences in emission levels, spatial distributions, and modeling performances were largely reduced after constraining, implying that the impact of the a priori inventory was limited on the top-down estimate. Sensitivity analysis proved the rationality of the near-linearity assumption between emissions and concentrations, and the impact of wet deposition on the multiple regression model was demonstrated to be moderate through data screening based on simulated wet deposition and satellite-derived precipitation.