A novel ensemble Kalman inversion with bias correction and its application to estimating CO emission over China

Inverse modeling of air-pollution emission with ensemble Kalman filter (EnKF) typically uses the sampling covariance matrix of the ensemble to represent the model error covariance matrix. Due to limited computational resources, pseudo smooth random perturbation fields are added to the priori estimations to produce the finite-size ensemble samples. However, errors in models and data are often systematic rather than random, such as model errors caused by bias from meteorological data, which are not well represented by random noise. These systematic errors can generally lead to the overestimation of forecast ensemble covariance and eventually lead to overestimation of the emissions.

In this study, we proposed a bias correction method to remove the systematic error from meteorological data using sensitivity analysis combined with in-situ observations. In this new method, bias and sensitivity of key meteorological variables are used to quantify model errors caused by meteorological simulation data. A case study of inverse modeling of carbon monoxide (CO) in China showed that the bias correction method can remove the biases from the Weather Research and Forecasting model and largely improve the estimate of CO emissions.