np.fitgeo: Fit a nonparametric geostatistical model

In npsp: Nonparametric Spatial Statistics

Fit a nonparametric geostatistical model

Description

Fits a nonparametric (isotropic) geostatistical model (jointly estimates the trend and the variogram) by calling locpol, np.svariso.corr (or np.svariso ) and fitsvar.sb.iso iteratively. At each iteration, the trend estimation bandwith is updated by a call to h.cv.

Usage

np.fitgeo(x, ...)

## Default S3 method:
np.fitgeo(
  x,
  y,
  nbin = NULL,
  iter = 2,
  h = NULL,
  tol = 0.05,
  set.NA = FALSE,
  h.svar = NULL,
  corr.svar = iter > 0,
  maxlag = NULL,
  nlags = NULL,
  dk = 0,
  svm.resid = FALSE,
  hat.bin = corr.svar,
  warn = FALSE,
  plot = FALSE,
  window = NULL,
  ...
)

## S3 method for class 'locpol.bin'
np.fitgeo(
  x,
  svm,
  iter = 1,
  tol = 0.05,
  h.svar = svm$esv$locpol$h,
  dk = 0,
  corr.svar = TRUE,
  svm.resid = FALSE,
  hat.bin = corr.svar,
  warn = FALSE,
  plot = FALSE,
  ...
)

## S3 method for class 'fitgeo'
np.fitgeo(
  x,
  iter = 1,
  tol = 0.05,
  h.svar = x$svm$esv$locpol$h,
  dk = x$svm$par$dk,
  corr.svar = TRUE,
  svm.resid = FALSE,
  hat.bin = corr.svar,
  warn = FALSE,
  plot = FALSE,
  ...
)

Arguments

x	a (data) object used to select a method.
...	further arguments passed to h.cv (trend bandwidth selection parameters).
y	vector of data (response variable).
nbin	vector with the number of bins on each dimension.
iter	maximum number of iterations (of the whole algorithm).
h	initial bandwidth matrix for trend estimation (final bandwidth if iter = 1).
tol	relative convergence tolerance (semivariogram).
set.NA	logical. If TRUE, sets the bin averages corresponding to cells without data to NA.
h.svar	bandwidth matrix for variogram estimation.
corr.svar	logical; if TRUE (default), a bias-corrected semivariogram estimate is computed (see np.svariso.corr). If FALSE the (uncorrected) residual variogram is computed (the traditional approach in geostatistics).
maxlag	maximum lag. Defaults to 55% of largest lag.
nlags	number of lags. Defaults to 101.
dk	dimension of the Shapiro-Botha variogram model (see fitsvar.sb.iso).
svm.resid	logical; if TRUE, the fitted (uncorrected) residual semivariogram model is computed and returned (this parameter has no effect when corr.svar = FALSE).
hat.bin	logical; if TRUE, the hat matrix of the binned data is returned.
warn	logical; sets the handling of warning messages in bandwidth selection (h.cv).
plot	logical; if TRUE, semivariogram estimates obtained at each iteration are plotted.
window	spatial window (values outside this window will be masked), currently an sp-object of class extending SpatialPolygons.
svm	(fitted) variogram model (object of class fitsvar or svarmod).

Details

Currently, only isotropic semivariogram estimation is supported.

If parameter h is not specified, h.cv is called with the default values (modified CV) to set it. If parameter h.svar is not specified, is set to 1.5*h.cv.svar.bin()$h.

Setting corr.svar = TRUE may be very slow (and memory demanding) when the number of data is large (note also that the bias in the residual variogram decreases when the sample size increases).

Value

Returns an object of class fitgeo (extends np.geo). A locpol.bin object with the additional (some optional) 3 components:

svm	fitted variogram model (object of class fitsvar).
svm0	(if requested) fitted residual variogram model (object of class fitsvar).
residuals	model residuals.

See Also

locpol, fitsvar.sb.iso, np.svar, np.svariso.corr, np.geo.

Examples

geomod <- np.fitgeo(aquifer[,1:2], aquifer$head, svm.resid = TRUE)
plot(geomod)


# Uncorrected variogram estimator
geomod0 <- np.fitgeo(aquifer[,1:2], aquifer$head, iter = 0, corr.svar = FALSE)
plot(geomod0)

# Additional iteration with bias-corrected variogram estimator
geomod1 <- np.fitgeo(geomod0, corr.svar = TRUE, svm.resid = TRUE)
plot(geomod1)

npsp documentation built on May 29, 2024, 5:31 a.m.