krige() function with SpatialPointsDataFrame in R

Question

I would like to make some kriging in a belgian territory.

I got a data set called clean.donnees.Ni after removing outliers and NAvalues.

I first tried simply with this, but got this warning as result :

Ni.krig <- krige(formula = Ni~1,
 data = clean.donnees.Ni,
 locations = ~x+y,
 newdata = krig.grid, 
 model = Ni.vg.fit)
Warnings : Covariance matrix singular at location [50686, 156893,0]: skipping

I found on other topics that I probably got points at the same location. (R gstat krige() - Covariance matrix singular at location [5.88,47.4,0]: skipping)

They advice to use this to remove points on the same location:

clean.donnees.Ni <- clean.donnees.Ni[-zerodist(clean.donnees.Ni)[,1],]

When I did this, I finally thought I could make krige() work, but instead it nows returns me this error :

Ni.krig <- krige(formula = Ni~1,
 data = clean.donnees.Ni,
 locations = ~x+y,
 newdata = krig.grid, 
 model = Ni.vg.fit)
Error in `coordinates<-`(`*tmp*`, value = locations) : 
 setting coordinates cannot be done on Spatial objects, where they have already been set

Because I had to transform clean.donnees.Ni in a SpatialPointsDataFrame, locations can't read ~x+y anymore.

Does anyone knows how to makes kriging work ?

Answer 1

If your data is a spatial data object then you give it to the locations argument - you only do locations = ~x+y if you have a non-spatial data frame with coordinates in those columns. So I think:

coordinates(krig.grid) <- ~x+y
Ni.krig <- krige(formula = Ni~1,
 locations = clean.donnees.Ni,
 newdata = krig.grid, 
 model = Ni.vg.fit)

will work for you.

I can reproduce your error with a test dataset, and make it work:

> d = data.frame(x=runif(100), y=runif(100), Z=rnorm(100))
> coordinates(d)=~x+y
> 
> k = krige(formula = Z~1, data=d, locations=~x+y, newdata=d)
Error in `coordinates<-`(`*tmp*`, value = locations) : 
 setting coordinates cannot be done on Spatial objects, where they have already been set
> 
> k = krige(formula = Z~1, locations=d, newdata=d)
[inverse distance weighted interpolation]

From the help:

locations: object of class ‘Spatial’ or ‘sf’, or (deprecated) formula
 defines the spatial data locations (coordinates) such as
 ‘~x+y’
 data: data frame: should contain the dependent variable,
 independent variables, and coordinates, should be missing if
 locations contains data.

• Jouline: I tried but now I get this as error: Error in h(simpleError(msg, call)) : error in evaluating the argument 'obj' in selecting a method for function 'coordinates': trying to get slot "bbox" from an object (class "data.table") that is not an S4 object I don't really understand why it doesn't work : > typeof(clean.donnees.Ni) [1] "S4"
• Spacedman: Somewhere you've got a data.table object. Krig.grid perhaps if its not your data points. Edit your Q and show summary and/or class of your inputs.
• Jouline: Hi, you were right, my krig.grid had to be also an S4 object ! I still got troubles to plot this, because I've never worked with S4 object. Maybe you can help me ?
• Spacedman: If you've got more problems, ask a new question, and give as much detail as possible on the data you are feeding the functions - show their summary or print them, or create sample data from things we can all use. Tag it "r" and "kriging" and I'll see it!