This is a one-entry function for several spatial prediction and simulation methods, for model objects
of class '>gmSpatialModel. The several methods are chosen by means of pars
objects of the
appropriate class.
# S3 method for gmSpatialModel
predict(object, newdata = NULL, pars = object@parameters, ...)
a complete "gmSpatialModel", containing conditioning data and unconditional model
a collection of locations where a prediction/simulation is desired; this is typically
a sp::SpatialPoints()
, a data.frame or similar of X-Y(-Z) coordinates; or perhaps for gridded data
an object of class sp::GridTopology()
, sp::SpatialGrid()
or sp::SpatialPixels()
parameters describing the method to use, encloded in an object of appropriate class (see below)
further parameters for generic functionality, currently ignored
Depending on the nature of newdata
, the result will be a data container of the same kind,
extended with the predictions or simulations. For instance, if we want to obtain predictions on the
locations of a "SpatialPoints", the result will be a sp::SpatialPointsDataFrame()
; if we want to obtain
simulations on the coordinates provided by a "data.frame", the result will be a DataFrameStack()
with
the spatial coordinates stored as an extra attribute; or if the input for a simulation is a masked grid of class
sp::SpatialPixels()
, the result will be of class sp::SpatialPixelsDataFrame()
which data
slot will be
a DataFrameStack.
Package "gmGeostats" aims at providing a broad series of algorithms for geostatistical prediction
and simulation. All can be accesses through this interface, provided that arguments object
and pars
are of the
appropriate kind. In object
, the most important criterion is the nature of its slot model
. In pars
its class counts: for the creation of informative parameters in the appropriate format and class, a series
of accessory functions are provided as well.
Classical (gaussian-based two-point) geostatistics are obtained if object@model
contains a covariance function,
or a variogram model. Argument pars
can be created with functions such as KrigingNeighbourhood()
,
SequentialSimulation()
, TurningBands()
or CholeskyDecomposition()
to respectively trigger a cokriging, as
sequential Gaussian simulation, a turning bands simulation, or a simulation via Cholesky decomposition.
The kriging neighbourhood can as well be incorporated in the "gmSpatialModel" object
directly, or even be
nested in a "SequentialSimulation" parameter object.
Conversely, to run a multipoint geostatistics algorithm, the first condition is that object@model
contains a
training image. Additionally, pars
must describe the characteristics of the algorithm to use. Currently, only
direct sampling is available: it can be obtained by providing some parameter object created with a call to
DirectSamplingParameters()
. Currently it is also necessary that newdata
is a gridded set of locations.
Other gmSpatialModel:
as.gmSpatialModel()
,
gmSpatialModel-class
,
make.gmCompositionalGaussianSpatialModel()
,
make.gmCompositionalMPSSpatialModel()
,
make.gmMultivariateGaussianSpatialModel()