Example of universal kriging with R/gstat in GRASS

Thomas:

Please do:

sessionInfo()

on both machines, and see if any of the package versions are different.

It often also helps to name function arguments explicitly, to be sure that 
they are being understood correctly inside the functions. Depending on the 
class of the data= arugment, as far as I recall different versions of the 
gstat package treat the locations= argument differently.

Roger


On Fri, 12 May 2006, Thomas Adams wrote:

  

Roger,

I have made considerable progress in what I was trying to do with gstat 
& Universal Kriging using R. I was fortunate enough to find a detailed 
example of what I was trying to do: http://spatial-analyst.net/RKguide.php.

I exported from GRASS my elevation grid, "gtopo30.dem", and temperature 
point file, "temps.txt", to ascii files to assure I followed a parallel 
path with my data.

So, with my data, following Tomislav Hengl's example, I have:

 > temps<-read.delim("temps.txt",sep=" ")
 > summary(temps)

cat lon lat z T name
Min. : 1.00 Min. :-90.05 Min. :35.82 Min. : 99.0 Min. :61.00 AGC : 1
1st Qu.: 31.75 1st Qu.:-86.63 1st Qu.:38.27 1st Qu.: 197.0 1st Qu.:64.00 
AID : 1
Median : 60.50 Median :-84.06 Median :40.09 Median : 255.5 Median :66.00 
ALN : 1
Mean : 60.37 Mean :-84.14 Mean :39.86 Mean : 314.2 Mean :66.83 AOH : 1
3rd Qu.: 89.25 3rd Qu.:-81.47 3rd Qu.:41.58 3rd Qu.: 360.0 3rd Qu.:69.00 
AOO : 1
Max. :118.00 Max. :-78.32 Max. :42.49 Max. :1155.0 Max. :73.00 ARB : 1
(Other):110
X Y
Min. :-341460 Min. :-342057
1st Qu.: -53847 1st Qu.: -71919
Median : 163999 Median : 119784
Mean : 154474 Mean : 99916
3rd Qu.: 371950 3rd Qu.: 282632
Max. : 632335 Max. : 398186

 > library(sp)
 > dem<-read.asciigrid("gtopo30.dem")
 > class(dem)

[1] "SpatialGridDataFrame"
attr(,"package")
[1] "sp"

 > image(dem)
 > points(Y ~ X, data=temps)
 > class(temps)

[1] "data.frame"

 > coordinates(temps)=~X+Y
 > dem.ov=overlay(dem,temps)
 > summary(dem.ov)

Object of class SpatialPointsDataFrame
Coordinates:
min max
X -341459.8 632334.6
Y -342056.9 398185.9
Is projected: NA
proj4string : [NA]
Number of points: 116
Data attributes:
gtopo30.dem
Min. : 115.3
1st Qu.: 196.9
Median : 245.4
Mean : 306.9
3rd Qu.: 331.0
Max. :1064.5

 > temps$gtopo30.dem=dem.ov$gtopo30.dem
 > library(lattice)
 > plot(T~gtopo30.dem, as.data.frame(temps))
 > abline(lm(T~gtopo30.dem, as.data.frame(temps)))
 > library(gstat)

 > vgm <- vgm(psill=8,model="Exp",range=600000,nugget=3.8)
 > vgm_temps_r<-fit.variogram(variogram(T~gtopo30.dem,temps), model=vgm)
 > plot(variogram(T~gtopo30.dem,temps),main = "fitted by gstat")
 > temps_uk<-krige(T~gtopo30.dem,temps,dem, vgm_temps_r)

[using universal kriging]

 > library(lattice)
 > trellis.par.set(sp.theme())
 > spplot(temps_uk,"var1.pred", main="Universal kriging predictions 

TEMPERATURE")


Which works perfectly on my Macintosh running Mac OS X 10.4 and using R 
2.2.1. (see attachment, temperatures in deg. F) However, following the 
*identical* steps with the identical data on Linux, at the step:

temps_uk<-krige(T~gtopo30.dem,temps,dem, vgm_temps_r)

I get the error:

Error in eval(expr, envir, enclos) : object "gtopo30.dem" not found

This has me baffled; any thoughts? I could send you my files if you 
would like to see what happens for you?

Regards,
Tom

BTW, the grid spacing on my DEM is coarse (9 km) and I will probably do 
my final analyses at 1-km.


Roger Bivand wrote:
    

On Fri, 28 Apr 2006, Thomas Adams wrote:

  
      

Roger,

Your suggestion:

fullgrid(dem) <- FALSE

did turn dem into class type SpatialGridDataFrame, but when I tried:

z <- predict(UK_fit,newdata=dem)

I got an error:

Error in model.frame(... :
invalid variable type.

I think I should restate the problem:

I have a file 'temps' which has class SpatialPointsDataFrame read from 
GRASS 6.1, that looks like:

coordinates cat x y z temp name
1 (-341460, -2154.42) 1 -90.05 38.90 166 63 ALN
2 (-198769, 301388) 2 -88.47 41.77 215 67 ARR
3 (-334899, -40321) 3 -89.95 38.55 140 66 BLV
4 (-240028, 163910) 4 -88.92 40.48 268 69 BMI
5 (-187957, 114806) 5 -88.27 40.04 229 64 CMI
6 (-351730, -37305.9) 6 -90.15 38.57 126 65 CPS
7 (-242424, 98244.7) 7 -88.92 39.87 204 66 DEC
8 (-179844, 315889) 8 -88.24 41.91 232 69 DPA
9 (-136093, -24538.2) 9 -87.61 38.76 131 68 LWV
10 (-278964, -126152) 10 -89.25 37.78 125 66 MDH
11 (-140792, 302011) 11 -87.75 41.79 187 73 MDW
12 (-364737, 274189) 12 -90.51 41.45 180 73 MLI
13 (-190503, 54493.9) 13 -88.28 39.48 219 64 MTO

and I have a a file 'dem' which has class SpatialGridDataFrame which 
just consists of grid of elevation values read from GRASS 6.1 using 
dem<-readFLOAT6sp(). (Sorry, I know I'm repeating myself).

What I want to do is to use the grid of elevation values ('dem') as a 
proxy in the spatial interpolation of the 'temp' values in my 'temps' 
file that are located at the coordinates in parentheses(). Notice that 
the temps file also has 'z' values of elevations. So, is this what you 
already understood? Converting 'dem' to a SpatialPixelsDataFrame seemed 
to only leave me with the grid locations and not the elevation values ? 
is this right.
    
        

What does:

summary(dem) 

say before and after doing

fullgrid(dem) <- FALSE?

Afterwards it should be a SpatialPixelsDataFrame with 

names(dem)

being "z". Saying summary(dem) will give you an idea of what is inside, 
str() should too.

Roger

PS. This is usually a one-off thing, once it works, you note down how, and 
then it just does from then on.


  
      

Thanks again for your help!

Regards,
Tom


Roger Bivand wrote:
    
        

On Fri, 28 Apr 2006, Thomas Adams wrote:

  
      
          

Roger,

This got me further along, but I am encountering a problem with:

z <- predict(UK_fit, newdata=BMcD_SPx)

The gstat step works for me, where I have:

UK_fit<-gstat(formula=temps$temp~dem,data=temps,model=efitted)

temps has class SpatialPointsDataFrame:

              coordinates cat      x     y    z temp name
1     (-341460, -2154.42)   1 -90.05 38.90  166   63  ALN
2       (-198769, 301388)   2 -88.47 41.77  215   67  ARR
3       (-334899, -40321)   3 -89.95 38.55  140   66  BLV
4       (-240028, 163910)   4 -88.92 40.48  268   69  BMI
5       (-187957, 114806)   5 -88.27 40.04  229   64  CMI
6     (-351730, -37305.9)   6 -90.15 38.57  126   65  CPS
7      (-242424, 98244.7)   7 -88.92 39.87  204   66  DEC
8       (-179844, 315889)   8 -88.24 41.91  232   69  DPA
9     (-136093, -24538.2)   9 -87.61 38.76  131   68  LWV
10     (-278964, -126152)  10 -89.25 37.78  125   66  MDH
11      (-140792, 302011)  11 -87.75 41.79  187   73  MDW
12      (-364737, 274189)  12 -90.51 41.45  180   73  MLI
13     (-190503, 54493.9)  13 -88.28 39.48  219   64  MTO

and dem has class SpatialGridDataFrame and just consists of grid values.
    
        
            

I think 

fullgrid(dem) <- FALSE

should make a SpatialPixelsDataFrame, but you'll have to make sure the 
name of the dem variable is the same as in the formula.

Roger

  
      
          

I tried to create a SpatialPixelsDataFrame for predict(), but with (for 
example):

m = SpatialPixelsDataFrame(points=meuse.grid[c("x","y")],data=meuse.grid)

I have nothing like meuse.grid, so this does not work. I can use 
image(dem), which produces a plot of elevation values. My point is that 
meuse.grid and my dem files have very different structures.

I'm not sure where to go to from here.

Regards,
Tom


Roger Bivand wrote:
    
        
            

On Thu, 27 Apr 2006, Thomas Adams wrote:

  
      
          
              

List:

I can not seem to work out the syntax for using R/gstat within a GRASS 
6.1 session to do universal kriging. I have a DEM (elevation data on a 
grid) and point data for temperature; theoretically, the temperatures 
should relate to elevation. So, I am trying to spatially interpolate the 
temperature data based on the elevations at the grid points. How do I 
setup the gstat command in R/gstat (and using spgrass6, of course)? I 
have no trouble reading in my elevation data (DEM) from GRASS and I have 
no problem doing ordinary kriging of my temperature data using 
GRASS/R/gstat.
    
        
            
                

What do the data look like? Do you have temperature and elevation at the
observation points and elevation over the grid? If temperature is the 
variable for which you want to interpolate, then the formula argument in 
the gstat() function would be temp ~ elev, data=pointsdata (if a 
SpatialPointsDataFrame no need for location= ~ x + y). Then the predict() 
step would need a SpatialGridDataFrame object as newdata, with elev as 
(one of) the columns in the data slot.

An example for the Meuse bank data in Burrough and McDonnell:

cvgm <- variogram(Zn ~ Fldf, data=BMcD, width=100, cutoff=1000)
uefitted <- fit.variogram(cvgm, vgm(psill=1, model="Exp", range=100, 
  nugget=1))
UK_fit <- gstat(id="UK_fit", formula = Zn ~ Fldf, data = BMcD, 
  model=uefitted)
z <- predict(UK_fit, newdata=BMcD_SPx)

where BMcD_SPx is a SpatialPixelsDataFrame (the grid has ragged edges) 
with flood frequencies in Fldf (actually a factor, but works neatly).

Hope this helps,

Roger

  
      
          
              

Regards,
Tom