Manual Page for Command >>> SELECT
This command invokes the automatic parameter selection option of iTOUGH2. The
parameters defined in block > PARAMETER are screened according to two selection
criteria (see commands >>>> SENSITIVITY and >>>> CORRELATION)
or will be truncated based on the eigenvalues of the Fisher information matrix
(see commands >>>> EIGENVALUE, or >>>> TRUNCATE).
Only the most sensitive and/or most independent parameters are subjected to the optimization process.
If command >>> SUPER is used, superparameters will be generated and optimized
(see Tonkin and Doherty [WRR, 41, W10412, doi:10.1029/2005WR003995, 2005]).
Only the most sensitive and/or most independent parameters are subjected to the
optimization process. The selection procedure is repeated every few iterations.
Automatic parameter selection allows one to submit a larger set of parameters to the
estimation process. If a parameter is not sensitive enough to be estimated from the available
data, it is automatically removed from the set of parameters being updated. This makes the
inversion faster because fewer parameters have to be perturbed for calculating the Jacobian
matrix (the full Jacobian is only calculated every few iteration when the selection criteria are
reevaluated). The inversion is also more robust. Parameters that are not sensitive or highly
correlated tend to be changed drastically during an iTOUGH2 iteration which may cause
unnecessary numerical difficulties. If they are (temporarily) removed from the parameter set,
they remain at their initial current value.
An additional advantage of using this option is the fact that sensitivities, estimation
uncertainties, and parameter correlations are calculated for all the specified parameters,
regardless of whether they are updated during the optimization.
Due to the non-linearity of the inverse problem at hand, sensitivity coefficients and parameter
correlations constantly change during the optimization. Therefore, the selection criteria must
be reevaluated from time to time (see command >>>> ITERATION (s)), i.e., parameters may be deactivated
and reactivated during the course of an inversion.
>>> use SUPERparameters
>>>> pick superparameters based on : 4 largest EIGENVALUES
>>>> repeat selection every : 3 rd ITERATION
Back to Command Index
Page updated: January 2, 2012