On the parallelization of a comprehensive regional-scale air quality model

Rick D. Saylor, Ryan I. Fernandes

Research output: Contribution to journalArticlepeer-review

9 Scopus citations


The complexity and nonlinearity of atmospheric flow dynamics, coupled with the nonlinear and multiphase character of atmospheric chemistry, make air quality modeling a particularly challenging problem. During the last two decades, comprehensive regional-scale air quality models have been developed which provide relatively thorough descriptions of the physics and chemistry of the atmosphere; however, a common weakness of these models is the coarse horizontal spatial resolution that must be employed to allow for reasonable simulation turnaround times. Typical horizontal grid spacings of 20-120 km do not provide adequate resolution for many important atmospheric phenomena such as clouds and smaller-scale turbulence. One way to achieve finer grid resolutions and maintain reasonable simulation turnaround times is to use parallelization. This work reports on the parallelization of the STEM-II regional-scale acid deposition and photochemical oxidant model on an IBM 3090-600J mainframe using IBM Parallel FORTRAN. A parallel speed-up of 4.65 (using five processors) has been achieved by parallelizing the code's transport and atmospheric chemistry calculations employing a locally one-dimensional time-splitting algorithm. A parallelized fraction of 0.98 has been estimated from multiple processor timings, which according to Amdahl's law would allow us to approach an ultimate speed-up of near 50 on similarly configured massively parallel machines.

Original languageBritish English
Pages (from-to)625-631
Number of pages7
JournalAtmospheric Environment Part A, General Topics
Issue number4
StatePublished - Mar 1993


  • Atmospheric chemistry
  • convection-diffusion equation
  • locally one-dimensional
  • time-splitting speed-up


Dive into the research topics of 'On the parallelization of a comprehensive regional-scale air quality model'. Together they form a unique fingerprint.

Cite this