predict.krige can then be called to compute
interpolation surfaces and prediction errors.
krige(formula, data=sys.parent(), subset, na.action=na.fail,
covfun, nc=10000, ...)
z ~ loc(x,y)
z is the kriging variable and
x and
y are the spatial locations,
that is,
z[i] is observed at the location (
x[i],y[i]).
The right hand side must contain a call to the function
loc.
A polynomial trend surface is of the form:
z ~ loc(x,y) + x + y + x^2 + y^2
loc function.
A constant term is always fit.
All terms on the right hand side must be entered with a
+ sign.
The
loc call can include arguments
angle and
ratio to correct
for geometric anisotropy; see the
loc help file.
Note that an evaluated
loc object cannot be used in
formula.
....
formula.
formula after any
subset argument has
been used.
The default (with
na.fail) is to create an error
if any missing values are found.
A possible alternative is
na.omit, which deletes observations
that contain one or more missing values.
covfun.
"krige" with components:
loc
function call in
formula.
loc
function call in
formula.
predict.krige
for computing interpolations.
The kriging system is solved using generalized
least squares (see Ripley, 1981).
The polynomial terms are scaled to (-1, 1) internally to avoid
numeric problems; the
coefficients component returned is for these scaled
terms.
This implementation of kriging does not handle multiple observations
at a point.
Methods for objects of class
"krige" include
predict and
print.
Cressie, Noel A. C. (1993).
Statistics for Spatial Data,
Revised Edition.
Wiley, New York.
Ripley, Brian D. (1981).
Spatial Statistics.
Wiley, New York
# krige the Coal Ash data with a quadratic trend in the x direction
# using a spherical covariance function:
kcoal <- krige(coal ~ loc(x, y) + x + x^2, data = coal.ash,
covfun = spher.cov, range = 4.31, sill = 0.14, nugget = 0.89)
# predictions over default 30 x 30 grid
pcoal <- predict(kcoal)
# plot prediction surface
wireframe(fit ~ x * y, data = pcoal,
screen = list(z = 300, x = -60, y = 0), drape = T)