In partnership with the US EPA, NOAA has implemented the early phases of a national air quality forecast capability: as of August, 2007, providing ground-level ozone predictions for the Eastern US and smoke predictions for the contiguous US (CONUS). The capability is being built in stages, based on successful phased development and testing prior to operational implementation. Experimental testing of an expanded CONUS ozone forecast capability is in progress and will be deployed into operations following successful achievement of operational readiness criteria for prediction accuracy and reliability. Targeted deployment of nationwide ozone forecasts within 2 years will be followed by the addition of quantitative particulate matter forecasts, and extended forecast periods - out to day 2 and beyond. Predictions are currently generated with linked weather and air quality models run operationally at NOAA's National Centers for Environmental Prediction (NCEP): for ozone, the NOAA-EPA Community Multiscale Air Quality (CMAQ) model driven by NOAA's operational North American mesoscale weather prediction model (NAM), the Weather Research and Forecasting Nonhydrostatic Mesoscale Model (WRF-NMM); smoke predictions are generated from NOAA's HYbrid Single-Particle Lagrangian Integrated Trajectory (HYSPLIT) model. The capability is an end-to-end system providing predictions of hour-by-hour ground-level smoke and ozone concentrations through the next day at 12km resolution, disseminated on NWS and EPA servers. Ozone predictions are updated twice daily at 6 and 12 UTC; smoke predictions, based on satellite-derived fire location date combined with fire emissions estimated with the US Forest Service's BlueSky capability, are updated each day at 6 UTC. Near-real-time verification is conducted each day: with AIRNow observations for ozone, and satellite-based smoke observations for predicted smoke in the atmospheric column.

Experimental testing of CONUS ozone in 2006 showed significant underprediction for ground-level ozone in parts of western US, especially in California. Analysis of 2006 performance led to the following 2007 improvements in the experimental ozone capability in testing over the CONUS, in addition to WRF-NMM upgrades: 1) closer coupling of the meteorological data between CMAQ and WRF-NMM via revised estimation of boundary-layer height and corrected plume-rise calculations for CMAQ, 2) use of improved boundary-layer dynamics in CMAQ, 3) updates to pollutant emissions databases, and 4) improved treatments of leaf canopy resistance, which result in improved estimates of deposition velocity for several important ozone precursors: NO, NO2 and CO. Retrospective testing showed systematic improvement of predictions for California over the earlier test configuration, while maintaining or improving prediction performance over the rest of the domain. Real-time testing (in progress, as of August 2007) shows good prediction performance over most of the domain, including California. The most notable exceptions are episodes of underprediction for the Los Angeles basin, associated with predictions that are locally distinguished by light onshore winds and dry conditions. Further analysis of performance is in progress to identify effective means to resolve this issue.

The NOAA-EPA team is continuing developmental testing of ground-level aerosol predictions, using a modified CMAQ capability and pollutant data from emissions inventories only. Results show qualitative prediction performance, with seasonal variability: systematic underprediction in summertime, consistent with missing source inputs, and more varied errors in wintertime, characterized by systematic, and sometimes compensating, biases in some species. Research is ongoing to develop capabilities for real-time quantitative ingest of additional emissions sources important for quantitative predictions of particulate matter.

A focus group of state and local air quality forecasters is providing feedback on NOAA test products. Based on experience with air quality conditions in their communities, their input focuses on localized performance, complementing performance evaluations based on domain-wide verification statistics.