This project has been carried out jointly with the Department of Environmental
Chemistry of MEP-TNO (TNO Institute for Environment, Energy and Process
Innovation) and was part of the
TASC Project HPCN for Environmental
Applications'.
This project was sponsored by the Stichting HPCN with financial
support from the Ministry of Economic Affairs.
The project started in January 1996 and ended December 1999.
Project Description:
The subject of study are the numerical algorithms and implementation
of LOTOS, a dispersion model of MEP-TNO for 'LOng Term Ozone Simulation'.
The current 4-layer LOTOS model is used for a variety of environmental
studies related to air pollution.
The aim of the project is to significantly enhance the suitability,
the accuracy, and the computational efficiency of LOTOS.
Hereby, the development of a three-dimensional, parallel LOTOS is one of
the main issues. Part of the project is the description of the new 3D model
(more). But the focus lies on numerical algorithms and
parallelization.
Among others, attention will be given to fast stiff ODE
solvers for the atmospheric chemistry problem,
advection schemes, schemes for solving turbulent diffusion and stiff chemistry
coupled, operator
splitting and local refinement or zooming techniques.
To have a handy tool to compare the suitability of various computer platforms
for an Air Quality Model (AQM) like LOTOS we developed a benchmark code for
a 3D prototype of an atmospheric dispersion model (see description in
html or
postscript).
This prototype contains the pocesses that
are relevant from the numerical and from the computational point of view.
The numerical algorithms used are similar to the ones we will actually
use in LOTOS. The implementation of the code is based on the experiments
we did on various platforms.
With this benchmark code we measured the performance of the various computer
architecture paradigms. We also looked at the I/O performance of the Cray T3E
for an off-line model (see the description of the
NCF/CRG 1997/1998 project).
The results of our evaluation are not surprising: For real computer speed
one should use a dedicated shared memory vector/parallel architecture or
a distributed memory architecture in case of memory constraints.
Both are expensive. Much cheaper and somewhat competitive is a
number of coupled workstations.
Our I/O experiments showed that, in contrast to the expectations raised
by the advertising slogan `the Cray T3E has a scalable I/O architecture',
I/O does not scale on the T3E.
With respect to the implementation of the full 3D LOTOS model we draw the
following conclusions from the experiments:
- To avoid divergence of different implementations aimed at different
computer platforms it is highly recommendable to have one implementation
of LOTOS. The experiments with our benchmark code on various platforms
show that this is possible without loosing efficiency.
- I/O experiments on the Cray T3E show that
- The necessary time to read the input data for LOTOS will be small compared
to the computational time.
- On the other hand, the output can have a significant influence on the
`through-put' time if one really wants to write all concentrations at every
time step to file, which is for instance the case when an AQM is coupled to
a powerful visualization / steering tool.
- The (almost) portable implementation of parallel asynchronous I/O
will be an efficient choice on all architectures and will result in negligible
I/O times on `shared nothing' architectures where every
processor has its own memory and its own disk.
People involved:
Papers:
[25]
(3D LOTOS model),
[23,
28,
31,
33]
(numerical aspects), and
[20,
29]
(implementation, performance)
from the
full list of publications in the
MAS1.1:
Atmospheric Flow and Transport Problems project overview.
LOTOS is a three-dimensional Eulerian regional air quality model.
The starting-point of these atmospheric transport/chemistry
models is the solution of the continuity
equation that describes the change in concentration of a chemical species in
the air as a result of transport, emissions, chemistry and some other processes.
The LOTOS model is driven by analyzed
meteorological data and by an emission data base and is thought to be
embedded in a larger model from which concentration values outside the
LOTOS domain can be extracted.
It is developed to simulate various chemical and physical processes in the
troposphere / tropopause on a regional scale (~ 1000 km).
The physical domain of the model is part of a shell around the earth. In the
vertical direction hybrid coordinates are used: the lower boundary is given
by the orography of the earth and the upper boundary by a surface having
equal pressure. For reasons of computational simplicity this irregular
physical domain (as well as the governing PDE system) is transformed into a
rectangular computational domain.
Go to the
TASC Project HPCN for Environmental
Applications' project overview, to the CWI
MAS1.1:
Atmospheric Flow and Transport Problems project overview.
or to the
CWI home page.
Joke Blom
(gollum@cwi.nl)
Last update 00/01/20.