\begin{center}
{\bf Abstract}
\end{center}

Air quality models have become basic scientific tools used in
formulating policy decisions that involve tremendous capital
resources.  The air quality of one region is affected by the emissions
and transport of pollutants from nearby regions.  This necessitates
the use of regional scale models, which cover large areas -- on the
order of 10$^6$ km$^2$, and are computationally intensive.  The next
generation air quality models will include more complex chemistry and
aerosol physics, and will employ sub-grid scales to treat atmospheric
turbulence and other processes.  These advancements in the models will
significantly increase their computational demands.  One way to
maintain a reasonable turnaround time will be to use massively
parallel computational architectures.  An important issue in the
parallel applications of these models is the portability of the model
across various architectures.  The urban and regional multiscale (URM)
model has been parallelized and applied to various computational
environments.  Significant decreases in elapsed computational time are
observed when the model is applied to a cluster of workstations.
Portability issues of the application are analyzed with regard to
architecture-independent code, good performance on different
architectures, and a uniform run-time environment.