underscore.txt

This file documents modifications and additions to the set of files
generated when OUTNAM of the OBS input file (USGS Open-File Report 00-184,
page 27) is not NONE.  These files are identified by the underscore ("_")
character as the first character of the filename extension.

--------------------------------------------------------------------------

Version 1.1 of Modflow-2000 added support for generation of an output file
with the extension _pc when OUTNAM is not 'NONE', the Parameter-Estimation
Process is active, and parameter estimation converges.  This file contains a
record for each estimated parameter, listing: parameter name, LN, optimized
parameter value, BL, BU, lower 95-percent confidence limit on the parameter
value, and upper 95-percent confidence limit on the parameter value.

--------------------------------------------------------------------------

Version 1.12 of Modflow-2000 includes a modification to the _os file.
A fifth column has been added to the _os output file, which was originally
described in table 5 on page 28 of USGS Open-File Report 00-184.  The fifth
column lists the observation time, in model time units, from the beginning
of the simulation.  For prior information, -1.0 is written.  The code
modification for ADV assumes that all ADV observations are in the first
stress period.

--------------------------------------------------------------------------

Version 1.12 also added support for a new optional setting in the OBS
Process input file is available.  If "ALLFILES" is specified following the
ISCALS value, and if the value specified for OUTNAM is not "NONE", up to
three additional output files ("_wt", "_su", and "_mc") are generated with
the filename base defined by OUTNAM:

(1) OUTNAM._wt:

When OUTNAM in the OBS input file is not "NONE", and "ALLFILES" is specified
after ISCALS, a new file is generated, identified by the extension "_wt".
This file contains observation weights, or the full weight matrix if any of
the observation types is entered with a variance-covariance matrix (WTQ in
the Observation Process input instructions of USGS Open-File Report 00-184).
The first line of this file indicates whether the rest of the file contains
a vector of weights or a correlated-weight matrix.  In either case, the
ordering of weights is such that weights on all head observations are listed
first, followed by weights on observations of the Drain, River, General Head
Boundary, and Streamflow Routing (STR) Packages; observations of
constant-head flow (Basic Package); and observations of the Drain-Return and
Advective-Transport Packages.

If no package has observations for which a variance-covariance matrix has
been entered, the first line contains the word "VECTOR", the second line
contains ND (the number of observations), and the following ND lines contain
the weight calculated for each observation, with one weight (one
observation) per line.

If any package has observations for which a variance-covariance matrix has
been entered, the matrix is written to the _wt file in the compressed format
described in the file "compressedmatrix.txt" in the doc directory.  The
first line contains the word "COMPRESSEDMATRIX".  The second line contains
NNZ (the number of non-zero entries in the full weight matrix for all
observations) and the two dimensions of the full weight matrix (always ND
and ND, since it is a square matrix).  The following lines contain the
position and value of the NNZ non-zero values of the ND-by-ND
correlated-weight matrix.  The position is an integer indicating the element
number where each non-zero entry belongs in the full matrix, assuming a
column-major storage scheme.  Example: assume a simulation uses 3 head
observations, 4 (correlated) observations using the Drain Package, and 2
(correlated) observations of the River Package.  This order is the same
order in which observations are printed in other "_" output files.  The full
weight matrix (noting that the matrix is symmetrical) would be a 9-by-9
matrix and might look like this:

  11.0   0.0   0.0   0.0   0.0   0.0   0.0   0.0   0.0
   0.0  22.0   0.0   0.0   0.0   0.0   0.0   0.0   0.0
   0.0   0.0  33.0   0.0   0.0   0.0   0.0   0.0   0.0
   0.0   0.0   0.0  44.0  0.54  0.64  0.74   0.0   0.0
   0.0   0.0   0.0  0.54  55.0  0.65  0.75   0.0   0.0
   0.0   0.0   0.0  0.64  0.65  66.0  0.76   0.0   0.0
   0.0   0.0   0.0  0.74  0.75  0.76  77.0   0.0   0.0
   0.0   0.0   0.0   0.0   0.0   0.0   0.0  88.0  0.98
   0.0   0.0   0.0   0.0   0.0   0.0   0.0  0.98  99.0

For this simulation, the _wt file would look like this:

COMPRESSEDMATRIX
             23              9              9
                   1        11.00000
                  11        22.00000
                  21        33.00000
                  31        44.00000
                  32        0.540000
                  33        0.640000
                  34        0.740000
                  40        0.540000
                  41        55.00000
                  42        0.650000
                  43        0.750000
                  49        0.640000
                  50        0.650000
                  51        66.00000
                  52        0.760000
                  58        0.740000
                  59        0.750000
                  60        0.760000
                  61        77.00000
                  71        88.00000
                  72        0.980000
                  80        0.980000
                  81        99.00000

(2) OUTNAM._su:

When the OBS and SEN Processes are active, OUTNAM in the OBS input file is
not "NONE", and "ALLFILES" is specified after ISCALS, a new file is
generated, identified by the extension "_su".  This file contains the
sensitivity of the simulated equivalent of each observation to each
parameter.  The first line of the file contains column names, where the
first two columns, named "OBSERVATION" and "SYMBOL", hold the user-supplied
observation names and plot symbols, respectively.  The remaining column
names identify the parameters for which ISENS > 0 in the SEN input file.
Following the line of column names is a line for each observation,
containing the observation name, plot symbol, and simulated-equivalent
sensitivity for that observation relative to each parameter for which ISENS
> 0.

(3) OUTNAM._mc:

When the PES Process is active, OUTNAM in the OBS input file is
not "NONE", and "ALLFILES" is specified after ISCALS, a new file is
generated, identified by the extension "_mc".  This file contains the matrix
of parameter correlations.  The first line of this file lists the names of
all parameters for which ISENS > 0 in the SEN input file.

--------------------------------------------------------------------------

Richard Yager of USGS in Ithaca, NY, kindly supplied the following
information related to the _rc and _rb files produced by RESAN-2000:

     The "_rc" file lists Cook's D, a measure of the influence of the
     observation on the set of estimated parameter values, along with the
     observation name and plot symbol. The "_rb" file lists the name and
     plot symbol of each observation along with the set of DFBETAS, a
     measure of the influence of the observation on each parameter value.
     The derivation of Cook's D is given in Cook and Weisburg (1982) and
     the derivation of DFBETAS is given in Belsley and others (1980). The
     application of both influence statistics to nonlinear regression of
     ground-water flow modeling is discussed by Yager (1998).

     Belsley, D.A., Kuh, E., and Welsch, R.E., 1980, Regression
     diagnostics: identifying influential data and sources of colinearity,
     John Wiley, New York.

     Cook, R.D., and Weisburg, S., 1982, Residuals and influence in
     regression, Monographs on Statistics and Applied Probability, v. 128,
     Chapman and Hall, New York.

     Yager, R.M., 1998, Detecting influential observations in nonlinear
     regression modeling of groundwater flow: Water Resources Research,
     v. 34, no. 7, p. 1623-1633.
