spde {RGeostats}R Documentation

Perform Estimation or Simulations using SPDE technology

Description

Perform Estimation or Simulations using SPDE technology

Usage

spde(dbin = NA, dbout, model, triswitch = "nqQ", nostat = NA, 
     gext=NA, mean=NA, seed = 232131,
     mesh.dbin=TRUE, mesh.dbout=TRUE, cov.extract=NA,
     flag.est= FALSE, flag.std=FALSE, flag.gibbs=FALSE,
     flag.modif=FALSE, nbsimu = 1, ngburn = 50, ngiter = 100, ngint=5,
     verbose = FALSE, mesh=NA,
     Q1rows=NA, Q1cols=NA, Q1vals=NA, Q1nrow=0, Q1ncol=0,
     Q2rows=NA, Q2cols=NA, Q2vals=NA, Q2nrow=0, Q2ncol=0,
     Q3rows=NA, Q3cols=NA, Q3vals=NA, Q3nrow=0, Q3ncol=0,
     A1rows=NA, A1cols=NA, A1vals=NA, A1nrow=0, A1ncol=0,
     A2rows=NA, A2cols=NA, A2vals=NA, A2nrow=0, A2ncol=0,
     A3rows=NA, A3cols=NA, A3vals=NA, A3nrow=0, A3ncol=0,
     radix = "SPDE", modify.target = db.locmod())

Arguments

dbin

The db-class structure containing the input data. In the case of non-conditional Simulations, this file is optional: when absent, we perform non-conditional simulations; when present, we perform conditional simulations. It must obviously be present for estimation. When used, there must be a single variable defined.

dbout

The db-class structure which contains the results.

model

The model-class describing the spatial characteristics.

triswitch

Command line for the internal triangulation step. For more information see meshing. In case of a model with several basic structures (nugget effect excluded), this parameter can be used as an array of properties (one property per basic structure). If the number of properties is smaller than the number of basic structures, the last property is used for the last basic structures.

nostat

List of non-stationary parameters. For details see model.param.define.

gext

The 'dbout' may be dilated by gext. This argument designates an array, with its dimension equal to the dimension of the space and which contains the extension in each direction. If not defined, the 'dbout' is not dilated and the simulated results may suffer some edge effect problems.

mean

Array containing the mean of each variable.

seed

Seed for the random number generation.

mesh.dbin

When TRUE, the location corresponding to the Input Data are systematically included in the meshing

mesh.dbout

When TRUE, the location corresponding to the Output Targets are systematically included in the meshing

cov.extract

List of the ranks of the basic covariance structures to be extracted. For the time being, only the nugget component (if any) can be filtered.

flag.est

When TRUE, the estimation is calculated.

flag.std

When TRUE, the standard deviation of the estimation error is calculated.

flag.gibbs

When TRUE, the iterative Gibbs method is used.

flag.modif

When TRUE, the simulation outcomes are not stored individually. Instead the simulations outcomes are summarized in two output variables, i.e. the mean and standard deviation of dispersion of the simulations, which can be considered respectively as the conditional expectation and the conditional variance.

nbsimu

Number of simulations.

ngburn

Number of burning iterations when the iterative Gibbs method is used as a simulation procedure. During these burning simulations, the intervals are gradually restrained from almost no constraint down to the final constraints.

ngiter

Number of iterations when the iterative Gibbs method is used as a simulation procedure.

ngint

Number of iterations inside the Gibbs sampler iterative algorithm

verbose

Verbose option

mesh

A mesh-class object

Q1nrow, Q1ncol, Q1rows, Q1cols, Q1vals

For the first internal Q matrix,

If Q1rows (first argument) is a sparse matrix (from 'Matrix' package), then the following arguments ('Q1cols', 'Q1vals', 'Q1nrow' and 'Q1ncol') are useless.

Otherwise three arrays (same dimension) which provide respectively the row, the column indices, as well as the value for the corresponding cell element.

Q2nrow, Q2ncol, Q2rows, Q2cols, Q2vals

For the second internal Q matrix,

If Q2rows (first argument) is a sparse matrix (from 'Matrix' package), then the following arguments ('Q2cols', 'Q2vals', 'Q2nrow' and 'Q2ncol') are useless.

Otherwise three arrays (same dimension) which provide respectively the row, the column indices, as well as the value for the corresponding cell element.

Q3nrow, Q3ncol, Q3rows, Q3cols, Q3vals

For the third internal Q matrix,

If Q2rows (first argument) is a sparse matrix (from 'Matrix' package), then the following arguments ('Q2cols', 'Q2vals', 'Q2nrow' and 'Q2ncol') are useless.

Otherwise three arrays (same dimension) which provide respectively the row, the column indices, as well as the value for the corresponding cell element.

A1nrow, A1ncol, A1rows, A1cols, A1vals

For the first internal A interpolation matrix,

If A1rows (first argument) is a sparse matrix (from 'Matrix' package), then the following arguments ('A1cols', 'A1vals', 'A1nrow' and 'A1ncol') are useless.

Otherwise three arrays (same dimension) which provide respectively the row, the column indices, as well as the value for the corresponding cell element.

A2nrow, A2ncol, A2rows, A2cols, A2vals

For the second internal A interpolation matrix,

If A2rows (first argument) is a sparse matrix (from 'Matrix' package), then the following arguments ('A2cols', 'A2vals', 'A2nrow' and 'A2ncol') are useless.

Otherwise three arrays (same dimension) which provide respectively the row, the column indices, as well as the value for the corresponding cell element.

A3nrow, A3ncol, A3rows, A3cols, A3vals

For the third internal A interpolation matrix,

If A3rows (first argument) is a sparse matrix (from 'Matrix' package), then the following arguments ('A3cols', 'A3vals', 'A3nrow' and 'A3ncol') are useless.

Otherwise three arrays (same dimension) which provide respectively the row, the column indices, as well as the value for the corresponding cell element.

radix

Radix of the name given to the variables storing the results in the target Db.

modify.target

Decides whether or not the newly created variables will have their locator defined or not. For more information, see db.locmod.

Details

The Double Quilting can be switched OFF by using the function set.keypair() with the keyword "Flag_Double_Quilt". By default, its value is set to 1.

The keypair mechanism has also been used to transmit the some values calculated internally to the user with the keywords:

Another use of the keypair mechanism is used to introduce a set of faults (that can only be used in the 2-D procedure using triangulation). For that sake, it suffices to use: set.keypair("Intercept_Faults",segments,nseg,2) where the argument 'segments' is a matrix with 2 columns and as many rows as they are fault vertices.

Using set.keypair("Save_Indices",1) will allow saving the indices of the points (target, data, steiner) involved in the establishment of the Q matrix for each one of the parts.

Using set.keypair("Flag_Simu_Chol",0) specifies if either Chebychev (0) or Cholesky (1) algorithm is used for the non conditional simulations.

Using set.keypair("Number_Polynomials_Chebychev",10001), we can define explicitely the number of polynomials used in the Chebychev approximation.

Using set.keypair("Chebychev_Tolerance",1e-6), we can define explicitely the tolerance of the polynomial approximation.

Using get.keypair("B.maxcol.sumabsrow"), we obtain the maximum (over the lines) of the sum of the absolute value of the elements of each row.

Using get.keypair("SPDE_DEBUG"), we set the verbose level enabling DEBUG option.

Value

The output db-class where the following variables have been added:

References

Pereira, M. and Desassis, N. (2019) Efficient simulation of Gaussian Markov Random Fields by Chebychev polynomial approximation. Spatial Statistics (31)

[Package RGeostats version 14.0.10 Index]