Concerning the Indicator Kriging, there is nothing special available in RGeostats ... as this is an (almost) standard kriging.
However, it may be interesting to note the following information, close to this topic:
[*] How to convert a continuous variable into an indicator?
Obviously this operation can be performed with a simple transformation hooked to the db.add() function. But a specific function exists which does the job... and is very convenient when you wish to create several indicators. This is the function db.indicator().
An quick example:
- Code: Select all
# Simulate a continuous variable on a regular 2-D grid
db = db.create(nx=c(100,100))
model = model.create(vartype="cubic",range=30,sill=2)
db = simtub(,db,model)
plot(db)
# Create a set of classes
limits = limits.create(mini=c(NA,-2,-1,0,1,2),maxi=c(-2,-1,0,1,2,NA))
limits
# Use these limits to create the indicators within each class
db = db.indicator(db,limits)
db
Note that, instead of an indicator variable per class, you wish to calculate a single categorical variable which contains the rank of the class, please use db.limits() instead.
[*] Which model should I use for kriging ?
Back to Geostatistical course. Not all basic structures are suitable to characterize an indicator variable. Basically, smooth basic structures should be avoided (Cubic, Gaussian, ...).
However there is no specific option in the model fitting procedures in RGeostats which would prevent using forbidden basic structures. This is the user's responsibility.
This limitation also holds in the case the user would like to perform a simultaneous estimation of different indicator variables through a Cokriging.
[*] Do post-processing function exist ?
No. It should be designed and programmed if needed.