Download Model-Independent Parameter Estimation Watermark

Transcript
The PEST Control File
4-5
one observation. See Chapter 6 for further details.
If PEST is run in regularisation mode you must ensure that the PEST control file contains a
“regularisation” section. You must also ensure that there are at least two observation groups,
one of which is named “regul”. See Chapters 7 and 8 for further details.
Note that if PESTMODE is supplied as “estimation” there is no need to include a “predictive
analysis” section or a “regularisation” section in the PEST control file.
NPAR
This is the total number of parameters used for the current PEST case, including adjustable,
fixed and tied parameters; NPAR must be supplied as an integer.
NOBS
This integer variable represents the total number of observations used in the current case.
Note that, when counting the number of observations to evaluate NOBS, any dummy
observations (see Chapter 3) that may be referenced in one or a number of instruction files are
ignored.
NPARGP
This is the number of parameter groups; parameter groups are discussed in detail below.
NPARGP is an integer variable.
NPRIOR
NPRIOR, another integer variable, is the number of articles of prior information that you wish
to include in the parameter estimation process. If there are no articles of prior information,
NPRIOR must be zero.
In general, you should ensure that the number of adjustable parameters is less than or equal to
the number of observations for which there are non-zero weights plus the number of articles
of prior information for which there are non-zero weights. If this is not the case the PEST
“normal” matrix of equation 2.16 will not be positive definite (in fact it will be singular) and
a unique solution to the parameter estimation problem will not achievable. Nevertheless
PEST will probably still determine a parameter vector which minimises the objective
function, using the Marquardt parameter to make the normal matrix positive definite; see
equation 2.23. However, a parameter vector determined in this way is not unique;
furthermore, PEST will be unable to determine the parameter covariance matrix and to
calculate parameter uncertainty levels because the Marquardt lambda is not added to the
normal matrix prior to the determination of these quantities.
NOBSGP
NOBSGP, another integer variable, is the number of observation groups used in the current
case. Each observation and each prior information equation must be assigned to an
observation group (they can all be assigned to the same group if desired). When PEST