A supercomputing solution for air quality permitting
By Dr Ka-Hing Yau and Prof. Jesse L.Thé
A contour diagram of the ground level concentration
of SO2 predicted by CALPUFF for
a one-day simulation. Two distinct plumes
emerge from the stack because the wind
shifts its direction during the day.
Canada has been enjoying
increasing activity in mining
and exploration, particularly
among the resource-rich
provinces such as British Columbia
and Alberta. However, these are also
home to vulnerable national parks and
scenic areas.
In the US the national parks are
often classified as Class I area under
the provision of Clean Air Act. Haze
formation due to pollutants and acid
rain are of particular concerns in
national parks. New projects in the
neighboring areas are subject to stringent
environmental impact assessment.
This usually involves regional
studies of visibility analysis and deposition
of sulfuric or nitrous chemicals.
In the US, regulators are suggesting
more frequently that request applicants
utilize CALPUFF to perform Class I
area impact analyses, long range transport
of pollutants, visibility studies for
BART, and impact evaluation of
sources close to the ocean.
These studies involve visibility
analysis, shoreline fumigation, and
calculation of the deposition of acidic
species. CALPUFF is also a preferred
long-range model in many Canadian
provinces including B.C. and Alberta.
However, the execution of CALPUFF
requires considerable expertise and
resources. Modelers performing these
analyses may require external support
on several fronts, from preparation of
the input data to the execution of the
model.
CALPUFF is a generalized non-steady-
state air quality system for regulatory
use. It was originally developed
by Earth Tech Inc. under a contract
with the California Air Resources
Board (CARB). The US EPA has proposed
the CALPUFF modeling system
as a Guideline model for regulatory
applications involving long distance
transport and on a case-by-case basis
for near-field applications where non-steady-
state effects are significant. The
latter applies to situations where factors
such as spatial variability in the
meteorological fields, calm winds,
fumigation, recirculation or stagnation,
and terrain or coastal effects are
important.
As a case in point, we present our
experience with a reputable client in
Kuwait. The project is a typical air
quality modeling exercise for a thermal
power plant facility. The client
needs to use refined CALPUFF modeling
in a manner similar to Class I
area analysis. This assessment begins
with first evaluating the impacts from
the existing sources. Subsequently, one
must estimate the increase in pollution
levels due to the new sources.
Furthermore, various emission rate
scenarios are inspected to assess air
quality as the new power plant expands
its output in subsequent stages and
uses various grades of fuel quality.
Each calculation must be performed
over a 5-year period of meteorological
data.
Preparation of meteorological data
is often the first hurdle for modelers.
Many international users outside of the
US (including Canada) often face particular
difficulty in obtaining conventional
data for their home countries.
We needed to employ meteorological
data generated by MM5, an advanced
numerical weather prediction (NWP)
model. The resources required for executing
the MM5 model are beyond the
scope of this article. However, it suffices
to say that 10 to 100 times more
expertise and computer resources are
required compared with common regulatory
models such as AERMOD or
CALPUFF when used with normal
meteorological data.
For the Kuwait project we executed
the prognostic model over the entire
country at a resolution of 12 km. To
complete the preparation of model
input, we also included satellite base
maps, global terrain data, and land use
data prepared by us with a resolution
of 1 km.
The execution of the CALPUFF
model can be very slow because it has
to trace numerous puffs in the computational domain. A single straight 5-
year CALPUFF simulation of the
client’s project required 80 days of
CPU time on a single high-end PC.
The whole project, with various emissions
scenarios, would take 336 days
for completion over a bank of 14 highend
Xeon workstations in our company.
To avoid any delay in the project,
we developed an effective computing
solution.
Lakes Environmental software
offers a supercomputing service based
on grid computing. The system distributes
many small jobs running in the
background of every computer and
workstation available in the company.
The straight CALPUFF execution was
divided into a series of short monthly
calculations for each single source.
These small jobs were conveniently
fed into the PC cluster. This resulted in
a quick turnaround and less disturbance
to coworkers, who had contributed
their desktops for the computing
pool. The partial impacts from
each source were later added up to
yield the total impact.
The above method works because puffs
are non-interacting in the
CALPUFF model. Moreover, the
incremental impacts of the new
sources were obtained right away.
There was no need to repeat the calculations
for the old sources. Further
simplification was obtained by grouping
identical sources to eliminate
duplication. In addition, scaling of
source strength was employed to estimate
impact changes with respect to
the fuel sulfur concentration. The latter
is an approximation as the chemical
formulation in CALPUFF is nonlinear.
Nevertheless, it turned out to be an
excellent approximation for the area
of concern.
Lakes Environmental Software
completed all the necessary calculations
in one month, instead of one year.
