一种基于集合最优插值的排放源快速反演方法

A Fast Emission Inversion Scheme Based on Ensemble Optimal Interpolation

基于集合卡尔曼滤波的源反演方法是估计排放源、提高空气质量模拟和预报精度的有效方法。为构建排放源与污染物浓度之间的误差协方差矩阵,该方法通常需要运行几十次大气化学传输模式。庞大的计算量限制了该方法的应用,使其无法为实时预报系统快速更新排放源。本研究发展了一种基于集合最优插值的排放源反演方法。该方法使用历史集合数据构建误差协方差矩阵,仅需一次常规的空气质量模拟便可根据观测模拟差异反演排放源,从而显著降低计算量。本文使用该方法同化2015年1月全国1107个地面站点观测的CO小时浓度数据,结合2014年1月的历史集合数据集,估计2015年1月全国15 km分辨率的CO排放源。该方案反演的全国CO排放总量仅比使用2015年1月集合数据集的反演量高1%,表明历史时段与反演时段的气象条件差异对月均CO排放的影响有限。使用历史集合数据集更新的排放源再次模拟可将全国349个独立验证站点的平均低估从0.74 mg m^(−3)降至0.01 mg m^(−3),均方根误差降低18%,表明该方法可快速更新排放源并降低其不确定性。

The emission inversion based on the ensemble Kalman filter(EnKF)is an effective method for estimating emissions and improving air quality modeling and forecasting.However,to construct the error covariance matrix between the emissions and pollutant concentrations,this method requires running the chemical transport model tens of times,which is computationally prohibitive and limits its application in updating the emissions for a real-time forecasting system.This study develops an emission inversion method based on the ensemble optimal interpolation(EnOI).The proposed method calculates the error covariance matrix from historical ensemble data and requires only a routine air quality simulation run for emission inversion from the contrasts between the observations and simulations,thereby greatly reducing the computational cost.The proposed method is applied to assimilating hourly surface observations of CO concentrations at 1107 sites over China in January 2015.During the experiment,CO emissions in January 2015 are estimated at a 15-km horizontal resolution using the historical ensemble dataset for January 2014.The total CO emission in China estimated by this scheme is only 1%higher than using an ensemble dataset for January 2015,indicating that the differences in meteorological conditions between the historical and estimated periods have a limited impact on the inversely estimated monthly CO emission.Simulations with the updated emissions reveal a decrease in the downward bias of average CO concentrations at 349 independent validation sites from 0.74 mg m^(−3) to 0.01 mg m^(−3) and a reduction of the root-mean-square error by 18%.The results suggest that the proposed method can be used as a fast emission updating scheme to lessen the uncertainties in the emission inventories.