`set'

The general form of the set command is

set parameter = value ;

The list of variables that can be `set' [default value between brackets]:

General options

set debug=2;

set debug level to 2 [1; options are listed in Appendix C.4]

set cn_max=1.0e8;

check the condition number of matrices. If it is larger than cn_max, then generate a missing value and report a (near) singularity warning. Matrices checked are $V$ and $F'V^{-1}F$ (or $F'D^{-1}F$). A suitable value for cn_max seems $1/ \sqrt{{\tt DBL\_EPSILON}}$, which is about $10^8$. Condition numbers are estimated using LU factorization [24], and may be an order of magnitude wrong. Condition numbers are reported if debug is set to report covariance matrices. [not set: check for singularity only during variogram model fitting]

set logfile='gstat.log';

set the file name where debug information is written to (see set debug and Appendix C.4) [stdout, debug information is written to the screen]

set mv='MisVal';

define the default missing value string as MisVal [the string NA] (Note that numerical missing values can be defined with mv in a data command)

set output='file';

write ascii output to file (e.g., variogram estimates, predictions and variance at non-gridded locations)

set zero=1.0e-10;

specify the highest value absolote differences in distance and prediction variances may have to be considered equal to zero [10 $\times$ DBL_EPSILON, about $2^{-15}$].

Variogram modelling options

set alpha=45.0;

directional sample (co-) variogram: set direction in $<x,y>$ plane, in positive degrees clockwise from positive y (North) [0.0]

set beta=30.0;

directional sample (co-) variogram: set direction in $z$, in positive degrees up from the $<x,y>$ plane [0.0]

set Cressie=1;

(for sample variogram calculation) use Cressie's square-root variogram estimator [0]

set cutoff=0.5;

set cutoff (max. dist. for sample variogram) at 0.5 [a fixed fraction of the maximum distance, see set fraction]

set dots=1000;

change the number of plotting points at which gstat will let gnuplot switch from plotting points (+) with numbers of points of pairs, to plotting dots without numbers [500]

set fit=1;

fit the variogram model to the experimental variogram, using weighted least squares fit. Values for fit are shown in table 4.2 [0, do not fit]

Table 4.2: values for fit

fit	fit by	weight
1	gstat	$N_j$
2	gstat	$\gamma(h_j)^{-2} N_j$
3	gnuplot	$N_j$
4	gnuplot	$\gamma(h_j)^{-2} N_j$
5	gstat	REML

set fit_limit=1.0e-10;

set fit limit to 1.0e-10 (Appendix A.1) [1.0e-6]

set format='%.3g';

the format used for real values in variograms, e.g. %.3g limits the number of significant digits shown to 3. A valid C-language format string for a double should be used, misspecification may result in unpredictable behaviour. [ %g : use 6 significant digits]

set fraction=0.25;

specify the default cutoff for sample variogram calculation as fraction of the length of the diagonal in the square or block spanning the data locations [0.333]

set gnuplot='mygnuplot';

invoke the program mygnuplot as gnuplot (variogram display) [gnuplot, or wgnuplot for Win32]

set gnuplot35='gpt35';

invoke the program gpt35 (gnuplot version 3.5) for variogram display only [use gnuplot, or the value of set gnuplot]

set gpterm='latex';

set the gnuplot terminal specification and options (a string to follow the gnuplot ``set term'' command). This option will overrule the `postscript' or `gif' settings from the variogram modelling user interface, thus allowing plotting to other graphic file formats and modification of options (see gnuplot documentation). [for gif: 'gif transparent size 480, 360', for POSTSCRIPT: 'postscript eps solid 17'

set intervals=20;

specify the default number of intervals for sample variogram calculation [15]

set iter=20;

use not more than 20 iterations on iterative fit methods [50]

set plot='file';

file defines the file name for gnuplot commands (not set, use temporary files)

set pager='less';

use `less' as pager to be called from the variogram modeling interface [the value of the environment variable PAGER (if set), or else the program more]

set secure=1;

prevent any calls to the functions system(), popen() or remove(), terminate program whenever one of the first two appear (once set, it cannot be set back) [0, not secure]

set sym=1;

force directional sample cross covariance and pseudo cross semivariance to be symmetric [0, asymmetric]

set tol_hor=45.0;

directional sample (co-)variogram: set horizontal tolerance angle in degrees [90.0]

set tol_ver=20.0;

directional sample (co-) variogram: set vertical tolerance angle in degrees [90.0]

set width=0.05;

set lag width to 0.05 (distance interval width for sample variogram) [ cutoff/intervals]

set gls=1;

use generalised least squares residuals instead of the default ordinary least squares (OLS) residuals for sample variograms or covariograms [0, use OLS or WLS residuals]

set zero_dist=1;

determine what happens with variogram estimates at distance zero. Values are 1: include in first interval, 2: omit, 3: calculate separately [1 for variograms, 3 for covariograms]

Prediction or simulation options

set idp=3.5;

set inverse distance power to 3.5 [2]

set nblockdiscr=10;

use regular block discretization with 10 points in each dimension at non-zero block size (note: 10 in 3 dimensions results in 1000 discretizing points) [4, and use Gauss quadrature (see Appendix A.3)]

set nsim=100;

create 100 independent simulations when following a single random path (output maps will get the simulation number attached to their names, therefore short names should be chosen in environments with file name restrictions) [1]

set n_uk=40;

(for conditional simulation only) use universal (or ordinary) kriging instead of simple kriging when the number of data in a kriging setting is greater than or equal to 40. For multivariable prediction the neighbourhood size is summed over all variables, otherwise it is evaluated per variable. Setting n_uk to zero limits use of simple kriging to empty neighbourhoods only [very large: always use simple kriging]

set order=2;

define the action when order relation violations occur during indicator simulation (section 2.6; table 2.1) or indicator kriging (section 2.5). Values are 0: no correction for indicator kriging, assure that estimated probabilities are in [0,1] before simulation; 1: as 0, but also for indicator kriging; 2: rescale the estimated probabilities if their sum is larger than 1; 3: rescale the estimated probabilities so that they sum up to 1; 4: do order relation correction for cumulative indicators (using the upward-downward averaging steps of GSLIB [8]). [0: do only basic order relation violation corrections for indicator simulation]

set quantile=0.25;

when method was set to med, report $p$-quantile of local neighbourhood selection as prediction value, and $(1-p)$-value as prediction variance [0.5: the median]

set rp=0;

follow regular, non-random path during sequential simulation [1, follow a random path]

set seed=1023;

set seed for random number generator [0: seed is read from the internal clock. If possible, microseconds are used. To check this, run gstat a few times with debug set to 2].

set useed=4053341103U;

set seed for random number generator when outside the range of a signed integer; note the `U' at the end of the number [see seed].

set sparse=10;

Use sparse matrix routines for covariance matrix. The number of sparse should be a reasonable estimate of the number of non-zero columns in each row of the covariance matrix $V$. [only available when sparse matrix routines in meschach are linked in; 0: use dense matrices]

set xvalid=1;

turn cross validation on (if prediction is possible) [0, no cross validation]

set zmap=10.0;

set height of mask map(s) to 10 when observations are 3-D. [0.0]