An aggregation and episode selection scheme for EPA's Models-3 CMAQ

In support of studies mandated by the 1990 Clean Air Act Amendments, the Models-3 Community Multi-scale Air Quality (CMAQ) model is used by EPA Program Offices to estimate deposition and air concentrations associated with specified levels of emissions. These studies require CMAQ-based distributional estimates of ozone, acidic deposition, fine particulate matter (PM2.5) and visibility, on seasonal and annual time frames. Unfortunately, it is not computationally feasible to execute CMAQ over such extended time periods. Therefore, in practice CMAQ must be executed for a finite number of episodes or "events," which are selected to represent a wide variety of meteorological classes. A statistical procedure called aggregation must then be applied to the outputs derived from CMAQ simulations in order to derive the required seasonal and annual estimates.

The meteorological classes or "clusters" were identified and characterized using Ward's method of clustering analysis on the 700 mb wind fields obtained from the NCEP/NCAR reanalysis project. Nine years of data (1984 -1992) were clustered for an area that covered the contiguous United States and surrounding areas. Alternative schemes were compared using relative efficiencies and meteorological considerations until an optimal scheme was defined that includes 20 clusters (five per season). From these homogeneous clusters, a stratified sample of 40 events was then selected using a systematic sampling technique. These 40 events will then be aggregated into the desired seasonal or annual time frames.

A preliminary evaluation of the aggregation and episode selection scheme has been performed which compared aggregated estimates of the extinction coefficient (bext) to the actual bext observed at over 200 stations nationwide. The bext was selected as the evaluative parameter for two reasons. First, it serves well as a surrogate for PM2.5, for which little observational data exist. Second, of all of the air quality parameters simulated by CMAQ, this visibility parameter has one of the most spatially and temporally comprehensive observational data sets available. The evaluation revealed good representation in that a majority of the stations recorded mean aggregated bext falling withing ± 10% of the observed mean bext for the entire nine-year period.

Ongoing work is investigating the ability of the scheme to replicate bext on finer temporal and spatial scales in order to accommodate various applications of CMAQ. For instance will aggregated estimates of bext for a particularly anomalous year (such as 1988) still fall within ± 10% of the observed means or will the estimates deteriorate? Likewise, will this approach, which was developed on a continental scale, perform as well at smaller scales? These concerns will be addressed as specific CMAQ simulations are planned and performed.