Updates and evaluation of the Community Multiscale Air Quality (CMAQ) Model –2004 release

An updated version of the Community Multiscale Air Quality (CMAQ) model is scheduled for public release in August of 2004. This version contains additional scientific updates as well as significant improvements in computational efficiency. The 2004 version of the CMAQ model (CMAQ04) has been run using meteorology simulated by the Fifth Generation PSU/NCAR Mesoscale Model (MM5) for the entire year of 2001 at 36 km grid resolution over the continental U.S. for the purpose of evaluation against ambient monitor networks for aerosol and gaseous constituents.

Scientific upgrades have focused mainly on aerosol processes, with improvements to aerosol dynamics and representation of size distributions in the log-normal modal scheme. There have also been updates in cloud processes, vertical diffusion, and dry deposition. However, the biggest changes from the September 2003 release are related to computational efficiency. The aerosol coagulation solver has been replaced and the partitioning of secondary organic aerosols has been improved. These changes more than double the speed of the aerosol computations and improved overall model efficiency by about 25 %.

Operator splitting time stepping has also been revised. Previously, the operator splitting time-step was set by the minimum horizontal Courant condition limited time-step over all vertical layers with imposed maximum and minimum values. At coarser resolution, this scheme overly restricts the model synchronization time-step due to high wind speeds in the upper troposphere. In the new version, the synchronization time-step is set according to the Courant time-step only in the lower layers (up to about 2.4 km).

CMAQ04 has been optimized for parallel systems. Parallel efficiency has been greatly improved for this version through modification of the I/O system. CMAQ has been run on a variety of parallel systems, including Linux with anywhere from 2 to 32 processors and IBM eServer using up to 64 processors. For example, our recent full year 2001 simulation took 5 days on a 4-cpu Linux system

The 2001 run has been evaluated against aerosol component measurements from the STN, CASTNet, and IMPROVE networks. Comparisons were also made with ozone, CO, SO2, and PM2.5 data from the AIRS network. Various statistics stratified in space and time as well as spatial analysis plots will be presented. We have also performed extensive evaluation of the meteorology data simulated by MM5 using surface networks, radiosonde soundings, and radar profilers.

Disclaimer - Although this work was reviewed by EPA and approved for publication, it may not necessarily reflect official Agency policy