**Syntax**

` >>> WORTH (PARAMETER/PREDICTION) (DETERMINANT/TRACE) (SET) (METRIC: imetric) (PERCENT)`

**Parent Command**

` >> OPTION `

**Subcommand**

` -
`

**Description**

This command makes iTOUGH2 perform a data-worth analysis. The significance of an observation for an inversion (or prediction) is assessed by evaluating how much the parameter uncertainty (keyword `PARAMETER`) or prediction uncertainty (keyword `PREDICTION`) is increased when adding potential (see command `>>>> POTENTIAL`) or removing existing calibration data points or data sets. This impact is measured either by the trace (keyword `TRACE`; default) or determinant (keyword `DETERMINANT`) of the respective covariance matrix. If keyword `SET` is present, the data-worth analysis is performed for entire data sets, i.e., not individual data points. The value of adding or removing prior information about parameters is also examined (see command `>>>> VARIATION`).

If one or multiple predictions are specified (see command `>>>> PREDICTION`), the data-worth analysis is performed based on the trace of the prediction covariance matrix; if no such predictions are specified, the data-worth analysis can only be based on either the trace or the determinant of the estimation covariance matrix.

The data-worth analysis is performed whenever the Jacobian matrix and estimation or prediction covariance matrices are evaluated (i.e., also for a local sensitivity analysis and derivative-based optimization). In these cases, use command `>>> WORTH` as a subcommand of `>> OUTPUT` to select printout options.

The data-worth analysis requires *n*+1 or 2*n*+1 simulation runs to calculate the Jacobian matrix using forward or centered finite differences, respectively. The data-worth analysis can be computationally costly if the significance of many potential observations is evaluated (note, however, that the number of predictions has essentially no effect on computational cost).

The metric used to evaluate data worth can be selected using keyword `METRIC`, followed by an integer that identifies the function (see separate manual).

The data worth can be printed directly or as a percentage (keyword `PERCENT`) of the total data worth of all actual or potential calibration points.

Background information about data-worth analysis can be found in Finsterle (*Water Resour. Res., 51*(12), 9904-9924, doi:10.1002/2015WR017445, 2015).

**Example**

` > COMPUTATION
>> OPTION
>>> perform a data-WORTH analysis for entire data SETs based on TRACE of covariance matrix
<<<
<<`

**See Also**

` >>>> POTENTIAL | >>>> PREDICTION | >>>> VARIATION | >>> WORTH (ou) `