Inverse Modeling of BC and CO Sources in WRF-Chem

Abstract:

Traditional inverse modeling approaches that are applied to estimate sources of chemically- and/or radiatively-active constituents rely heavily on model-data error covariances that are typically prescribed and assumed to have diagonal (uncorrelated) structures. In most cases, this type of error representation do not fully account for errors in atmospheric chemistry models, especially with regards to uncertainties in transport and removal/loss mechanisms. Here, we introduce an ensemble Bayesian synthesis inversion framework for estimating black carbon and carbon monoxide sources in a coupled weather-chemistry model, WRF-Chem. We use a quasi Monte Carlo approach in generating a small ensemble of WRF-Chem configuration, through multi-physics representation and perturbations in a) emissions for both CO and BC, b) removal rates for BC, and c) initial and boundary conditions for both meteorology and chemistry. We then carry out ensemble tagged tracer (basis functions) simulations using slightly different ‘model configuration’ for each ensemble member to: a) calculate the model response functions, and b) estimate the associated error covariances based on the ensemble statistics. These simulations are conducted using an ensemble-based data assimilation software package, the Data Assimilation Research Testbed (DART). Conventional meteorological observations are assimilated in WRF-Chem/DART to constrain model meteorological fields. We then carry out a suite of multi-species (BC and CO) Bayesian synthesis inversions using a combination of: 1) ground-based (EPAAQS CO and IMPROVE BC) and satellite measurements (MOPITT CO, MODIS AOD, OMI Near UV AOD) and 2) ensemble-mean basis and response functions together with their associated ensemble error covariances derived from these simulations. We present our initial results of this work particularly a comparison of a posteriori diagnostics using this approach to that of a single WRF-Chem inverse analysis experiment. Finally, we introduce the need for a TransCom-like chemical data assimilation/inverse modeling systems inter-comparison for CO and BC.