Better sensitivity analyses
The sensitivities calculated by programs, such as MODFLOWP and UCODE, can be
used as described in the Methods and Guidelines Report and textbook by Hill and Tiedeman (2007)
to show clearly such
things as the amount of information provided for the estimation of defined
parameters by available data, and the likely importance of each parameter to
predictions of interest. For example,
in figure 1 below, composite scaled sensitivities show that the available
data provide the most information about parameters T2, T3, and RCH1;
associated parameter correlation coefficients also would need to be
considered in an actual application, but are not shown here. In figure 2
below, one-percent scaled sensitivites are used to show that the prediction
of concern, in this case the directional components of advective travel, are
all influenced most by parameters T3 and T4. Thus, for this illustrative
example, the estimation of T4 would be of concern because the available
information provide little information about T4, while it is important to
the predictions of interest. A next step would be to use similar methods to
determine whether additional quantities could be measured to provide
additional information about parameter T4.
Figure 1: Composite scaled sensitivities. Black bars indicate parameters
estimated by regression; gray bars indicate parameters that were not
estimated by regression. The values shown are for illustrative purposes only
and are not associated with an actual synthetic or field problem.
Figure 2: One-percent scaled sensitivities for the directional components of
advective travel, derived using the ADV program. The bars show the range and
average of values calculated for particles started in adjacent cells. Large
absolute values are assoiciated with parameters that are important to the
calculation of advective travel in the noted direction.
Methods and Guidelines Report
ADV program
|