Impact of three dimensional data assimilation on high resolution weather and pollution forecasting in the Los Angeles basin

Meteorological modeling is needed in air-quality and emergency-response applications as well as for weather forecasting. We have implemented a real-time limited-area mesoscale data assimilation and forecast system based on the NCAR/Penn State Mesoscale Model 5 (MM5) and the NCAR 3-Dimensional Variational Analysis (3DVAR) data assimilation system. We discuss results of our initial investigation into the impact of data assimilation on the quality of our forecasts. The system is configured to run automatically in near real-time to generate daily hourly forecasts over the Los Angeles basin. The forecasts extend 36 hours from the 12Z (5PST) initialization time. We run two domains at 15km and 5km grid spacing. The 3DVAR system performs data assimilation on both domains. It uses output from the North American Meso Model (NAM) (formerly known as the Eta model) as its background/first guess field and assimilates observations from a number of sources, including conventional and satellite based observations from the Air Force Weather Agency operational database, local observations from Bureau of Land Management (BLM) Remote Automated Weather Stations (RAWS), the South Coast Air Quality Management District (AQMD) monitoring stations, and the NOAA Forecast Systems Laboratory (FSL) Boundary Layer Profile (BLP) sites. The MM5 forecast model is initialized with output from the 3DVAR system and the Navy operational sea surface temperature analysis. Lateral boundary conditions are derived from the NAM data. The system is run using a limited cycling method in which a 6-hour forecast from the 00Z cycle initializes a 06Z model run that in turn produces a 6-hour forecast to initialize the primary MM5/3DVar run at 12Z. We conducted offline parallel runs during the month of September of 2004 that used no data assimilation and compared the skill of these forecasts to those of our operational run in order to evaluate the impact of data assimilation on forecast quality.