LKrigSAR {LatticeKrig} | R Documentation |
Method that creates the spatial autoregressive (SAR) matrix.
Description
Using the information in LKinfo create the SAR matrix for a given level of the multi-resolution.
Usage
LKrigSAR(object, ...)
## Default S3 method:
LKrigSAR(object, ...)
## S3 method for class 'LKInterval'
LKrigSAR(object, Level, ...)
## S3 method for class 'LKRectangle'
LKrigSAR(object, Level, ...)
## S3 method for class 'LKBox'
LKrigSAR(object, Level, ...)
## S3 method for class 'LKRing'
LKrigSAR(object, Level, ...)
## S3 method for class 'LKCylinder'
LKrigSAR(object, Level, ...)
## S3 method for class 'LKSphere'
LKrigSAR(object, Level, ...)
Arguments
object |
An LKinfo object. |
Level |
The level of the multi-resolution. |
... |
Any additional arguments to pass to this method. |
Details
The model for the Gaussian Markov Random field, c, at a given level is
B c = e,
where B is the SAR matrix computed by this method, and e are uncorrelated N(0,1). The precision matrix for this level is
Q= t(B)%*% B
and so the covariance matrix for c is the inverse of Q:
solve( Q)= solve(B)%*% t( solve(B))
Value
A matrix in the sparse matrix format, spind
, with dimensions given by the
number of lattice points at Level
.
Because this construction is geometry dependent the default version of
this method just returns an error message.
Author(s)
Doug Nychka
See Also
Examples
x<- cbind( c(0,1))
LKinfo<- LKrigSetup(x,LKGeometry="LKInterval",
nlevel=3, NC=3, a.wght=5, alpha=c(1,.5,.2) )
B<- LKrigSAR( LKinfo, Level=2)
B<-spind2full(B)
image.plot( B)
LKinfo<- LKrigSetup(cbind( c(0,360), c(0,1)) ,LKGeometry="LKRing",
nlevel=1, NC=3, a.wght=5, alpha=1)
B<- LKrigSAR( LKinfo, Level=1)
B<-spind2full(B)
image.plot( B)