GeoVarestbootstrap: Update a 'GeoFit' object using parametric bootstrap for std...

In GeoModels: Procedures for Gaussian and Non Gaussian Geostatistical (Large) Data Analysis

Update a GeoFit object using parametric bootstrap for std error estimation.

Description

The procedure updates a GeoFit object using a classical parametric bootstrap. The fitted model is simulated and refitted for each bootstrap replication; the empirical covariance matrix of the successful parameter estimates is used as the bootstrap covariance estimate. A faster score-bootstrap alternative is GeoVarest. Full/Standard likelihood fits are not accepted; obtain their Hessian-based standard errors directly with GeoFit(..., varest = TRUE).

Usage

GeoVarestbootstrap(
  fit, K = 100, sparse = FALSE,
  optimizer = NULL, lower = NULL, upper = NULL,
  method = "cholesky", alpha = 0.95, L = 10000,
  parallel = FALSE, ncores = 6, progress = TRUE,
  seed = NULL, min_success_rate = 0.8
)

Arguments

fit	A fitted object obtained from the GeoFit.
K	The number of simulations in the parametric bootstrap.
sparse	Logical; if TRUE then cholesky decomposition is performed using sparse matrices algorithms (spam packake).
optimizer	The type of optimization algorithm (see GeoFit for details). If NULL then the optimization algorithm of the object fit is chosen.
lower	An optional named list giving the values for the lower bound of the space parameter when the optimizer is L-BFGS-B or nlminb or optimize.
upper	An optional named list giving the values for the upper bound of the space parameter when the optimizer is L-BFGS-B or nlminb or optimize.
method	String; The method of simulation. Default is cholesky. For large data set three options are TB or CE (see the GeoSimapprox) function.
alpha	Numeric; The level of the confidence interval.
L	Numeric; the number of lines in the turning band method.
parallel	Logical; default FALSE. If TRUE then the estimation step is parallelized
ncores	Positive integer or NULL; default 6. With parallel=TRUE, an explicit integer requests that many workers, capped by detected cores and the number of bootstrap jobs. Set ncores=NULL for automatic selection, capped at six workers and normally leaving one detected core free.
progress	Logic; If TRUE then a progress bar is shown.
seed	Optional integer seed for reproducibility of the simulated samples.
min_success_rate	Minimum fraction of successful bootstrap refits required to return variance estimates. The default is 0.8.

Details

For spatio-temporal fits, the original numeric coordt values stored in fit are reused. Irregularly spaced times are supported with method = "cholesky"; approximate simulation methods retain the restrictions documented in GeoSimapprox.

The function updates a GeoFit object by refitting each simulated data set. A sensitivity matrix is not required to estimate bootstrap standard errors and confidence intervals. If fit$sensmat is available, CLIC and CLBIC penalties are also computed. For stochastic nearest-neighbor fits, the realized retained-pair graph from the original fit is reused unchanged across all bootstrap refits.

Parallel multisession workers are started with the package-library paths of the calling R session, including the library containing the loaded GeoModels installation.

For method = "TB" without a copula, parallel workers are persistent and replications are streamed one at a time (simulate, evaluate, release). This keeps peak memory bounded when both L and K are large while still reusing the selected pair graph and static fitting context within each worker. For fitted copula models, method="TB" is simulated through the turning-bands backend of GeoSimCopula; the copula path is currently not streamed replicate-by-replicate. On platforms where future reports forked multicore execution as safe, GeoModels uses it automatically to reduce worker startup and serialization overhead; otherwise it falls back to multisession with the current GeoModels library path propagated explicitly.

Value

Returns an updated object of class GeoFit. The main updated components include stderr, varcov, godambe when invertible, conf.int, pvalues, and estimates. When a compatible sensmat is available, claic/clic, clbic, and clic_penalty are also updated.

Three-dimensional coordinates

For a purely spatial univariate fit with explicit irregular three-dimensional Euclidean coordinates, method = "TB" is supported when no anisopars were used. Method "CE" remains restricted to regular two-dimensional grids. Use method = "cholesky" for bivariate, spatio-temporal, anisotropic, or other unsupported three-dimensional cases.

Author(s)

Moreno Bevilacqua, [e-mail],https://sites.google.com/view/moreno-bevilacqua/home, Víctor Morales Oñate, [e-mail], https://sites.google.com/site/moralesonatevictor/, Christian Caamaño-Carrillo, [e-mail],https://www.researchgate.net/profile/Christian-Caamano

See Also

GeoFit for the fitted objects used as input

Examples

library(GeoModels)

################################################################
###
### Example 1. Test on the parameter
### of a regression model using conditional composite likelihood
###
###############################################################
set.seed(342)
model="Gaussian" 
# Define the spatial-coordinates of the points:
NN=3500
x = runif(NN, 0, 1)
y = runif(NN, 0, 1)
coords = cbind(x,y)
# Parameters
mean=1; mean1=-1.25; # regression parameters
 sill=1 # variance

# matrix covariates
X=cbind(rep(1,nrow(coords)),runif(nrow(coords)))

# model correlation 
corrmodel="Matern"
smooth=0.5;scale=0.1; nugget=0;

# simulation
param=list(smooth=smooth,mean=mean,mean1=mean1,
 sill=sill,scale=scale,nugget=nugget)
data = GeoSim(coordx=coords, corrmodel=corrmodel,
 model=model, param=param,X=X)$data

I=Inf

fixed=list(nugget=nugget,smooth=smooth)
start=list(mean=mean,mean1=mean1,scale=scale,sill=sill)

lower=list(mean=-I,mean1=-I,scale=0,sill=0)
upper=list(mean=I,mean1=I,scale=I,sill=I)
# Maximum pairwise composite-likelihood fitting of the RF:
fit = GeoFit(data=data,coordx=coords,corrmodel=corrmodel, model=model,
 likelihood="Conditional",type="Pairwise",sensitivity=TRUE,
 lower=lower,upper=upper,neighb=3,
 optimizer="nlminb",X=X,
 start=start,fixed=fixed)

unlist(fit$param)


#fit_update=GeoVarestbootstrap(fit,K=100,parallel=TRUE)
#fit_update$stderr
#fit_update$conf.int
#fit_update$pvalues

GeoModels documentation built on Sept. 23, 2026, 5:07 p.m.