| fit.variogram.reml {gstat} | R Documentation |
Fit Variogram Sills to Data, using REML (only for direct variograms; not for cross variograms)
fit.variogram.reml(formula, locations, data, model, debug.level = 1, set, degree = 0)
formula |
formula defining the response vector and (possible)
regressors; in case of absence of regressors, use e.g. z~1 |
locations |
spatial data locations; a formula with the coordinate variables in the right hand (dependent variable) side. |
data |
data frame where the names in formula and locations are to be found |
model |
variogram model to be fitted, output of vgm |
debug.level |
debug level; set to 65 to see the iteration trace and log likelihood |
set |
additional options that can be set; use set=list(iter=100)
to set the max. number of iterations to 100. |
degree |
order of trend surface in the location, between 0 and 3 |
an object of class "variogram.model"; see fit.variogram
This implementation only uses REML fitting of sill parameters. For each
iteration, an n x n matrix is inverted, with $n$ the number of
observations, so for large data sets this method becomes
demanding. I guess there is much more to likelihood variogram fitting in
package geoR, and probably also in nlme.
Edzer J. Pebesma
Christensen, R. Linear models for multivariate, Time Series, and Spatial Data, Springer, NY, 1991.
Kitanidis, P., Minimum-Variance Quadratic Estimation of Covariances of Regionalized Variables, Mathematical Geology 17 (2), 195–208, 1985
data(meuse) fit.variogram.reml(log(zinc)~1, ~x+y, meuse, model = vgm(1, "Sph", 900,1))