资源环境科技发展态势分析平台

We evaluate two inverse modeling methods by conducting inversion experiments using the GEOS-Chem chemical transport model and its adjoint. We simulate synthetic NH3 column density as observed by the Cross-track Infrared Sounder over North America to test the ability of the iterative finite difference mass balance (IFDMB) and the four-dimensional variational assimilation (4D-Var) methods to recover known NH3 emissions. Comparing to the more rigorous 4D-Var method, the IFDMB approach requires 3-4 times lower computational cost and yields similar or smaller errors (12-17% vs 17-26%) in the top-down inventories at 2 degrees x 2.5 degrees resolution. These errors in IFDMB-derived emission estimates are amplified (53-62%) if compared to the assumed true emissions at 0.25 degrees x 0.3125 degrees resolution. When directly conducting inversions at 0.25 degrees x 0.3125 degrees, the IFDMB consistently exhibits larger errors (44-69% vs 30-45%) than the 4D-Var approach. Analysis of simulated differences in NH3 columns and in NH3 emissions suggests stronger misalignments at the finer resolution, since the local column is more strongly influenced by spatial smearing from neighboring grids. Adjoint calculations indicate that the number of adjacent grids needed to account for most (>65%) of the emission contributions to the local columnar NH3 abundance over an NH3 source site increases from similar to 1 at 2 degrees x 2.5 degrees to similar to 10 at 0.25 degrees x 0.3125 degrees, leading to increased errors especially in IFDMB. Applying inversion results at 2 degrees x 2.5 degrees to update the a priori emissions at 0.25 degrees x 0.3125 degrees could improve the accuracy of IFDMB inversions and reduce the computational cost of 4D-Var.