资源环境科技发展态势分析平台

For the first time, the Nested Air Quality Predicting Modeling System (NAQPMS) with 5 km horizontal resolution was used to simulate the temporal and spatial variations of PM2.5 for 3 years (from 2013 to 2015) over China. The simulated concentrations of PM2.5 were evaluated against two datasets (U.S. consulates and CNEMC) measured in five megacities: Beijing, Shanghai, Guangzhou, Chengdu and Shenyang. Before model evaluation, reliability and consistency of the two datasets were verified. The results showed that PM2.5 data from U.S. consulates had good agreement with the data from CNEMC sites. The two datasets had high consistency across the five mega cities, with most of the data points distributed between the 25th and 75th percentiles. The model accurately reproduced the spatial and temporal variations of PM2.5 concentration, with FAC2 values over 80% in the five cities. The high-resolution model had good performance in all seasons in Shanghai, with RMSEs 18.3-24.9 mu g/m(3). In Guangzhou, although it was underestimated in March and October, there was also good agreement between the model and PM2.5 observations, with RMSEs 16.2-20.6 mu g/m(3). In Beijing and Shenyang, the two northern cities, the simulated PM2.5 concentration was underestimated in some months, but RMSEs were still as low as 38.1-46.0 mu g/m(3) and 24.2-43.1 mu g/m(3), respectively. In Chengdu, some simulated values were overestimated in summer, but RMSEs at different sites were uniformly distributed (35.2-40.5 mu g/m(3)). On the whole, > 90% of the simulated values met the model performance criteria in all of five megacities.

Model evaluation indicated that the uncertainties in the model results were from several aspects. Underestimated emissions and the Inaccurate simulation of the adverse weather conditions led to the lower estimation of PM2.5 concentrations in Beijing, especially in winter. Conversely, the overestimated PM2.5 concentration in Chengdu implied that the current emission inventory might not consider the emission reduction measures, and the insufficient wet deposition in model also contributed to such phenomenon. Generally, despite the diversity of topography, demography and emissions, we showed that a 5 km resolution model could accurately capture the spatial and temporal variations of PM2.5 concentration.