Impacts of Different Background Error Statistics on 3-Dimensional Variational (3D-Var) Data Assimilation for PM2.5 Predictions and Analysis

Abstract

Data assimilation is the technique that statistically combines background with observation data, and products best estimate of the physical state which is also known as the analysis field. The key component of data assimilation is constructing a correctly specified background error covariance (BEC) matrix. The representative methods for estimating background error covariance matrices are the National Meteorological Center (NMC) method and the ensemble method. The NMC method defines the background error by using two model simulations with different initial times but the models verify at the same time. On the other hand, the ensemble method assumes the mean of ensemble members as the true state, and defines the background error as the difference between each member and the mean of ensemble members. In this study, to understand how the estimating method of BEC affects 3-dimensional variational data assimilation, chemical transport modeling simulation was performed using the WRF-CMAQ system during the period of the Korea-United States Air Quality Study (KORUS-AQ) campaign. Also, this study compares the result of data assimilation with the background error covariance matrices by using the NMC method and the ensemble method. For estimating background error with the ensemble method, the ensemble members were determined using various input data—including emission inventories, meteorological initial and boundary conditions, and planetary boundary layer schemes—which contribute to the uncertainties of the CMAQ model simulations. After generation of BEC matrices, both the analysis and prediction fields—produced by assimilating ground-based PM2.5 observations for every 3 hours—were compared based on the performance of predictions.