AIR-QUALITY PREDICTIONS OF THE URBAN AIRSHED MODEL CONTAINING IMPROVED ADVECTION AND CHEMISTRY ALGORITHMS

We have carried out an extensive study on the effect of replacing the advection and chemistry algorithms of the Urban Airshed Model (UAM) wi th more accurate algorithms, using the SCAQS episode of August 26-28, 1987, and in put data supplied by the California Air Resources Board. Replacing the UAM chemistry solver by the implicit-explicit hybrid sol ver (IEH) tends to lower the afternoon ozone concentrations somewhat, the afternoon PAN concentrations more noticeably, and the H2O2 concent rations significantly. IEH is also much less likely to have a converge nce failure than the UAM solver. Replacing the Smolarkiewicz scheme by the forward Euler Taylor Galerkin scheme (FETG) yields a greater impa ct. The replacement increases the local peak NO concentrations and low ers the local peak ozone and CO concentrations significantly. Replacin g both chemistry and advection algorithms lowers the peak ozone, PAN, and H2O2 concentrations quite noticeably. For example, the predicted p eak ozone for August 28 is reduced from 234 to 200 ppb. The complicate d wind field for this episode generates two high-ozone subdomains in t he model-one in the east and one in the northwest. Reducing the NOx em ission by 50% shifts the predicted domainwide peak ozone from the east to the northwest. If we disregard this shift, then the use of more ac curate modules in the UAM leads to a 50% increase in the domainwide pe ak ozone while the original UAM yields a 27% increase. Both the origin al and improved UAM versions give a 20% increase in peak hourly ozone and a 30% increase in peak 8-h ozone in the eastern subdomain, but muc h greater and vastly different increases for each of the peak 1-h and peak 8-h ozone concentrations in the northwestern subdomain. The best available numerical methods should be used in modeling a complex syste m like air quality so that potential hidden errors may be identified a nd the design of air quality control may be more reliable.