Evaluation of mass balance, 4D-Var, and hybrid approaches to constraining NOx emissions using OMI NO2 remote sensing measurements.

Abstract:

Nitrogen oxides (NO_x) contribute to photochemical smog, tropospheric ozone and aerosol, and human health problems. Remote sensing observations provide a valuable means of constraining emissions of NO_x, and thus improving our ability to use air quality models for further understanding these issues. Traditional top-down estimates have provided important constraints for NO_x emission inventories in China, but are either time-consuming (e.g., 4D-Var) or only crudely represent the influence of atmospheric transport and chemistry (e.g., mass balance). Here, we combine mass balance and 4D-Var approaches, and investigate the improvements in simulated NO_x column density over China. Scaling factors derived from the mass balance approach with OMI observations are first applied to NO_x emissions. In this process, a smoothing kernel is used to account for emissions from adjacent grid cells, and optimized NO_x emissions are derived using maximum likelihood estimation, which weigh top-down and bottom up estimates by their relative errors. This is followed by subsequent inversion using an adjoint-based 4D-Var approach with GEOS-Chem at the 0.5x0.667 degree resolution. We consider the correlations between errors in neighboring grid cells by using off-diagonal terms in scaling factors’ error covariance matrix. An optimal value of the regularization parameter is selected using an L-curve and minimization of total error. We compare the solutions obtained using this hybrid approach with that obtained from standard 4D-Var, as well as to the direct solution from the mass balance approach itself. Differences between these methods in specific grid cells are investigated. We demonstrate the effect of transport and chemistry on the performance of mass balance and 4D-Var, identifying cases where the smoothing kernel and weighting errors in mass balance can cause the scaling factor to be in a direction that occasionally increases residual error. This study shows potential to facilitate decadal-scale NO_x emission inversions.