| Trend Modelling {RandomFields} | R Documentation |
The coding of trends, in particular multivariate trends, will be described here.
See RFcalc, RMtrend and also the examples below for some insight on the possibilities of trend modelling.
Martin Schlather, schlather@math.uni-mannheim.de, http://ms.math.uni-mannheim.de
RFcalc, RM, RMmodels, RMtrend, RMmodelsMultivariate.
require(geoR)
data(ca20) ## data set from geoR
ca20.df <- as.data.frame(ca20)
head(ca20.df)
RFoptions(coordnames=c("east", "north"), varnames="data")
## covariance model with variance, scale and nugget to be estimated;
## just to abbreviate later on
M <- RMexp(var=NA, scale=NA) + RMnugget(var=NA)
## short definition of a trend using the fact that ca20.df is a
## data.frame
ca20.RFmod02 <- ~ 1 + altitude + M
(ca20.fit02.RF <- RFfit(ca20.RFmod02, data=ca20.df, M=M))
## long definition which also allows for more general constructions
ca20.RFmod02 <- NA + NA*RMcovariate(ca20.df$altitude) + M
(ca20.fit02.RF <- RFfit(ca20.RFmod02, data=ca20.df))
## Note that the following also works.
## Here, the covariance model must be the first summand
ca20.RFmod02 <- M + NA + ca20.df$altitude
print(ca20.fit02.RF <- RFfit(ca20.RFmod02, data=ca20.df))
### The following does NOT work, as R assumes (NA + ca20.df$altitude) + M
### In particular, the model definition gives a warning, and the
### RFfit call gives an error:
(ca20.RFmod02 <- NA + ca20.df$altitude + M)
try(ca20.fit02.RF <- RFfit(ca20.RFmod02, data=ca20.df)) ### error ...
## factors:
ca20.RFmod03 <- ~ 1 + area + M ###
(ca20.fit03.RF <- RFfit(ca20.RFmod03, data=ca20.df, M=M))