Variational Inverse Modeling of Regional Nitrogen Oxides Emissions

Inverse modeling methods have been widely used for model performance improvement and parameter estimation. For air quality studies, inverse modeling is often used for emission inversion as emissions are associated with significant amount of uncertainties. NOx emission sources are estimated through a four dimensional variational (4D-Var) inverse modeling approach using satellite and ground-level observations. In the non-temporal set-ups, NO_x emissions are adjusted by assimilating NO_2 columns from satellite (OMI) and NO_2 from ground based observations separately. The results indicate that the average scaling factors vary from 0.41 to 1.74 or from 0.45 to 2.52 when ground observations and OMI observations are used, respectively. The average scaling factors are in the range of 0.43 to 1.79 when both types of observations were used simultaneously. The total amount of emissions increase 3.5% when inversion is based on ground observations only, 17.3% when inversion includes only OMI observations and 13.6% when both observations are used simultaneously. Under the temporal set-up, NOx emissions are adjusted by using OMI and modified surface observations to estimate hourly profiles of emissions at each location. These results indicate that inverting emissions on an hourly basis leads to reduced distance between observations and modeled concentrations. The majority of scaling factors are higher than one. Lightning is one of the more uncertain sources of Nitrogen dioxides. While inverted emissions included point, mobile, biogenic and other sources in the air quality model (CMAQ), lightening as one of the most significant natural sources of NOx was excluded in previous inversion setups. To account for the impact of this source, lightning emissions are parameterized as input for CMAQ model from the observations of Lightning Detection Networks flash rates. We filtered the places and observations that are dominated by anthropogenic sources. NO emission inversion in filtered areas resulted in reduction in bias and root mean square error between vertical column densities calculated by CMAQ and observed by OMI. However, scarcity of data prevents these results from being extended to all lightning events across the domain and episode.