Source: http://www.ngwa.org/publication/abstracts/Morrissey.html
Editorial printed in the November-December issue of GROUNDWATER,
Vol. 38, No. 6, pg 801.
Ground Water Modeling and Automated Computer Design
by Dan Morrissey, RPG
The development and wide availability of graphical user interface (GUI)
processors for ground water flow and solute transport modeling has
vastly increased the efficiency of the modeling process. However, this
technology has created new problems for ground water scientists. These
problems fall into three general categories: software reliability, model
misuse, and training. The software, being relatively new and not subject
to extensive peer review, may contain a variety of deficiencies. The
mechanics of modeling have been simplified to a point where
inexperienced scientists are misusing complex flow and transport models.
Training in ground water modeling may be too focused on the use of GUI
software and not enough on essential ground water modeling concepts.
As a hydrologist with 25 years of experience in ground water modeling
and teaching of modeling, I remember the dark ages of digital modeling
when card readers and mainframe computers were standard equipment.
Although some computerized mapping of data was done, much of it was
rudimentary and analysis was done with homegrown software. Training
concentrated on building and understanding the inner workings of models.
When personal computers and spreadsheet software became widely
available, data input and analysis became more streamlined. However,
scientists were still frustrated by the problem of creating properly
formatted data sets and the need to export data to other software to
facilitate analysis.
The introduction of GUI-based processing software has eliminated these
problems. Scientists who have experienced the dark ages especially
welcome this technology. Complex arrays of data can now be created
quickly from external databases. Boundary conditions, wells, rivers, and
other hydrogeologic features can be specified easily with
Òpoint-and-clickÓ methods. Geostatistical routines embedded in GUI
software allow automatic interpolation and extrapolation of sparse data.
Automated calibration routines are easy to implement. Input arrays and
modeling results can be visualized instantly in a variety of
perspectives and enhanced with color animation.
The new GUI software was an inevitable step in the development of
modeling technology and is here to stay. The capabilities of this new
software impress all but the most hardened techno-grouches. Scientists
can now use GUI-processors with powerful personal computers to create
models that are larger and more complex than ever. However, experience
has shown me that there are some serious pitfalls that may be
encountered, especially by inexperienced users.
A serious potential problem involves the quality of the software.
Problems with software are not unique to the ground water modeling
community. Experience has shown me that GUI preprocessors are not
subject to the same level of peer review as the underlying core models.
In one specific case I detected an error in a widely used preprocessor
that caused incorrect values of data to be put into the modeling
program. This was an error that could have been detected only by a user
who was experienced in the use of this specific ground water model and
who was curious enough to carefully review the model input and output
data. This problem occurred a few years ago when the software was
relatively new, but the experience showed me that users are essentially
the reviewers.
GUI software makes the mechanics of ground water modeling easy for
inexperienced users and isolates the modeler from the core models. They
simplify data processing by using default values for certain model
options and, in some cases, reducing the full functionality of the core
models. Although error checking in GUI software is helpful, it cannot
check for all possible errors and may lead to a false sense of security.
Automatic creation of data sets coupled with instant color display of
results may also create a false sense of security and foster a
Òpush-a-button-and-out-pops-a-modelÓ syndrome. The average present day
GUI-based modeler may be unaware of the structure and organization of
core data input files and model output listings. If these files are not
understood, or if the user is not aware of their existence, they will
not be reviewed to ensure that they are correct or that modeling results
are reasonable.
Training in the use of GUI software should be preceded by training in
the concepts of ground water modeling and in the use of the core flow or
transport models. There seems to be a tendency for recent students to
believe that training in the use of GUI software is actually training in
modeling. This would be equivalent to calling training in the use of a
word processor as training in journalism. Those of us who use GUI
software should have enough basic understanding of the models we use
that we can be aware of the effect that the use of specific GUI software
may have on our modeling analysis.
Developers of GUI software have a responsibility to provide some level
of quality control, to be experienced in the use of specific models for
which they develop their software, and to provide decent documentation
of their products. If the GUI does internal calculations that use
different algorithms than the original model, or if there are default
values embedded in the GUI, they must be carefully documented for users.
If a GUI limits the functionality of a model, the user should be made
aware of it. The GUI should make it easy to check core model input and
output files for accuracy. When errors in a GUI are detected, they
should be quickly brought to the attention of all users. As the line
between the GUI software and the core models becomes more blurred, and
the model becomes the GUI, there will be a need to make the codes for
GUI software more available for external review.
In the end, the ultimate responsibility for a ground water modeling
analysis is the practicing hydrologist. If GUIs are used as black boxes
and users become hypnotized by slick new technology the quality of our
modeling analyses will suffer.
____________________
Mr. Dan Morrissey
McDonald Morrissey Associates, Inc.
280 Pleasant Street
Concord, NH 03301
603-228-2280
Fax 603-228-2112