Environmental Engineering Reference
In-Depth Information
Fig. 2.2 Verification,
calibration and validation as
feedback loops between
different model levels
Feedback Loops
Conceptual
Model
Differential
Equation
Verification
Solution
Calibration
Post
Processing
Validation
Evaluation
The feedback loops within the system of tasks can be related to terms as
verification, calibration and validation, which is visualized in Fig.
2.2
. All these
terms do not have a uniquely definition and thus may be found in slightly different
contexts in the scientific literature. The term
verification
is mostly used in connec-
tion with software testing, which is a crucial step of software development. When
a software code is entered and runs regularly without error messages from
the computer, it is not sure that there are no 'bugs' in the code. In order to
test the correct performance of the computer program, test cases are set up, to
check, if the program delivers the correct answer. Test cases can be based on simple
post-processing, on analytical solutions, on theoretical derivations and on inter-
comparison with results from other codes.
Comparison with test cases, which are generally accepted in the concerned
scientific community (so called benchmarks), is called
benchmarking
. Within the
outlined step concept verification is thus a feedback loop in which it is checked,
whether the computed solution delivers the solution of the differential equation.
Unless it is a simple straight forward computation, the MATLAB
modeler also
has to verify her/his implemented m-code. In case of errors the code
debugging
becomes necessary. How to 'debug' in MATLAB
®
is explained in Sect.
2.7
.
The term
calibration
is used for the procedure of adjusting the model parameters
for a specific application of the code. The term is almost identical to
parameter
estimation
and strongly related to the term
inverse modeling
. Within the step
concept the term means a feedback loop, in which the solution or some entity that
is determined by post-processing is compared with values, obtained from the real
site. In case the check is negative, usually some parameter values have to be adjusted
in order to obtain a good fit. If that does not help, it may become necessary to make
adjustments in the mathematical formulation or even in the conceptual model.
®
