Modeling Concepts - Coastal Lagoons: Ecosystem Processes and Modeling for Sustainable Use and Development - page 297

During a sensitivity analysis, all the parameters and input data are varied indi-
vidually, usually by a constant percentage, in order to determine which one causes
the greatest change in the simulated solution. 4a Results are presented as a fraction
or percentage of the baseline conditions or as a ratio of input parameter change to
the simulated dependent variable change. For example, a sensitivity analysis may
show that a 1% variation of a selected parameter will produce a 3% variation in the
steady-state solution variation over the total area.
Sensitivity analysis can be useful in two respects. First, it provides calibration
guidelines by isolating the parameters that are best suited for calibration purposes.
These parameters are those producing significant variations during the analysis.
Should a model solution not be sensitive to some parameter, the value of this
parameter can be taken out of the literature and need not be measured in the modeled
domain.
Second, sensitivity analysis can define the error range in the initial and boundary
conditions. This is called error analysis. 4a It assigns estimates of uncertainty to all
significant parameters, initial conditions, and boundary conditions and determines
the combined uncertainty in the output or dependent variables. Monte Carlo methods
can be used for such a purpose. 4a

6.5.3 C ALIBRATION

Unfortunately, there is no guarantee that, when provided with a carefully prepared
bathymetry, well-planned boundary conditions, and other required input parame-
ters, a model will give good results immediately. In fact, the opposite is often the
case. In several instances, the model even becomes unstable during the first few
simulations. 37
This is why we must first calibrate a model with field observations of dependent
variables measured inside the model domain. The purpose of calibration is to tune
the model so that the differences between the computed and the measured values
are reduced. 37 Some recommendations on data collection programs aimed to model
calibration and validation are given in Chapter 7, Section 7.4.5 and Table 7.1.
The main parameters normally used to calibrate a hydrodynamic model are the
bed resistance, the eddy viscosity, the bathymetry, the boundary conditions, and the
wind friction. Some practical recommendations 37 on the use of these and other param-
eters during calibration are provided below.
The bed resistance can be used to calibrate or stabilize a model solution. This
is especially true in shallow areas where, for example, bed resistance will change
tidal amplitudes and phases significantly. 37
The eddy viscosity is mainly used to stabilize a numerical solution. For example,
increasing the eddy viscosity in an area will smooth out spurious results such as
higher frequent oscillations in water levels or wiggles in the flow field—zigzagging
current vectors—in that modeled area. The eddy viscosity can of course also be used
to calibrate the model. Eddy viscosity changes will only change the tidal amplitude
but not the tidal phase.
If the eddy viscosity is used for calibration in areas of variable water depths,
a Smagorinsky formulation 46,47 relating the eddy viscosity to the (variable) local