7A.7 Assimilation of fine aerosols using WRF-Chem and an Ensemble Kalman Filter

Pagowski, Mariusz, Georg A. Grell, Judith Berner, Kate Smith, and Ming Hu, National Oceanic and Atmospheric Administration

Because of interactions between meteorology and chemistry future data assimilation systems will likely not only include meteorological but also chemical observations and state. Such systems hold a promise for improved prediction of both weather and chemical composition.

Prediction of atmospheric aerosols is the focus of this presentation. Aerosols not only impact radiation and cloud processes but are also a pivotal contributor to air quality and single most critical factor affecting human mortality due atmospheric pollution.

Current skill of prediction of aerosol composition and concentrations is poor compared to the skill of weather forecasts as demonstrated by evaluation statistics and large discrepancies between results from different global and regional chemical models. Such state of the art originates mainly from

• weaknesses of chemical parameterizations;

• strong sensitivity of the system to uncertain and burdened with large errors emission source estimates;

• limited availability of observations of vertical profiles and aerosol composition and shortcomings of satellite retrieval for assimilation and verification;

• deficiencies in forecasting boundary layer meteorology.

In the presentation, we examine the impact of assimilating MODIS AOD and in-situ measurements on the prediction of fine aerosol concentrations over North America in the summer of 2012. We use an on-line meteorology-chemistry model WRF-Chem and a hybrid data assimilation system which includes the Gridpoint Statistical Intepolation (GSI) and an Ensemble Kalman Filter. Similar to our earlier work we note large initial benefit of data assimilation and relatively quick deterioration of forecast verification scores with time. Simultaneous assimilation of meteorology and chemistry allows us to elucidate both positive and negative effects of regression relationships derived by the Kalman filter. We further seek to improve forecasts by increasing model error using stochastic parameterizations of selected model parameters. We present a comprehensive evaluation of the results, identify deficiencies of the assimilation and an outline future work.