[GRASS-stats] gstat error

[Roger Bivand]

On Sun, 26 Apr 2009, Edzer Pebesma wrote:

> Jarek has sent me the data off-line, and it turns out that the data are
> three-dimensional, but lack variability in the third dimension:

This smells like an installed GRASS/OGR plugin, which if autodetected 
generates a spurious third coordinate (probably to match ESRIs usual 
idiocies?). I added an argument to turn this off in spgrass6 CVS last week 
(and corresponding pass-through to rgdal on CVS). So if Jarek updates to 
latest spgrass6 and rgdal from CVS (different sourceforge projects), it 
should help, or do readVECT6(..., plugin=FALSE).

Doing summary(p) would reveal the problem.

Roger

>
>> summary(p)
> Object of class SpatialPointsDataFrame
> Coordinates:
>              min     max
> coords.x1 3560158 3575935
> coords.x2 5636873 5653748
> coords.x3       0       0
> Is projected: TRUE
> proj4string :
> [+proj=stere +lat_0=51.67083333333333 +lon_0=16.67222222222222
> +k=0.999800 +x_0=3703000 +y_0=5627000 +no_defs +a=6378245 +rf=298.3
> +towgs84=33.4,-146.6,-76.3,-0.359,-0.053,0.844,-0.84 +to_meter=1]
> Number of points: 200
> Data attributes:
> ...
>
> Using the "degree=1" way of trend analysis, a first order linear trend
> is computed in the coordinates, meaning in z as well, and this is the
> cause of failure. Ignoring z, one can e.g. do
>
> p$x=coordinates(p)[,1]
> p$y=coordinates(p)[,2]
> r$x=coordinates(r)[,1]
> r$y=coordinates(r)[,2]
> pg=gstat(id="p_1", formula=rzedna~x+y,data=p)
> p_1=predict(pg,r)
> spplot(p_1)
>
> of course, no standardization of coordinates takes place here, so this
> approach will fail for high order trends.
> --
> Edzer
>
> Jarek Jasiewicz wrote:
>>
>>
>> Edzer Pebesma pisze:
>>> Jarek Jasiewicz wrote:
>>>
>>>> Hi
>>>> during trend calculation I recive following error:
>>>>
>>>> R
>>>> library(spgrass6)
>>>> library(gstat)
>>>>
>>>> raster=readRAST6("rzedna")
>>>> p=readVECT6("punkty")
>>>> p at proj4string=raster at proj4string
>>>>
>>>> pg_1=gstat(id="p_1", formula=rzedna~1,data=p, degree=1)
>>>> p_1=predict(pg_1,raster)
>>>> [ordinary or weighted least squares prediction]
>>>>
>>>> "solve.c", line 87: singular matrix in function Usolve()
>>>> Error in predict.gstat(pg_1, raster) : Usolve
>>>>
>>> There are several situations why this can happen that are hard to
>>> distinguish from the point where the error happens. The most likely in
>>> your case (first order trend surface estimation) is that your data are,
>>> geographically, on a line, i.e. one-dimensionally. Or similary, that you
>>> have less than 3 observations. Is that the case?
>>> --
>>> Edzer
>>>
>>
>> Thanks for answer,  but none of that case. Tt is regular data file
>> with 200 random observations generated with v.random in GRASS and
>> stored as SpatialPointDataFrame.
>> before I  posted to the list  I had tried to reproduce it on my
>> archive data with archive scripts (which worked  in older R
>> compilation) and it ALWAYS fail (on every data I tested).
>>
>> In general, the possible problem on my side coud be that I
>> (experientally) installed R from CRAN ubuntu repository instead to
>> compile from source. The ubuntu CRAN-R packages do not include lapack.
>> This is the only difference I noticed.
>>
>> If you need I can send you sample R file (is small, no more than 2MB)
>>
>> Jarek
>>
>>>> my system information:
>>>>
>>>> Ubuntu 8.04
>>>> R 2.8.1
>>>> gstat 0.9.59
>>>> the error have appeared first time on gstat 0.9.49
>>>>
>>>> Errors apears on all my data, even these, where trend was calculated
>>>> some times ago. Commads coming from old data
>>>>
>>>> The error is also reproducible on spearfish dataset with elevation.dem
>>>> and random points generated by v.random
>>>>
>>>> Jarek
>>>> _______________________________________________
>>>> grass-stats mailing list
>>>> grass-stats at lists.osgeo.org
>>>> http://lists.osgeo.org/mailman/listinfo/grass-stats
>>>>
>>>
>>>
>> _______________________________________________
>> grass-stats mailing list
>> grass-stats at lists.osgeo.org
>> http://lists.osgeo.org/mailman/listinfo/grass-stats
>
>

-- 
Roger Bivand
Economic Geography Section, Department of Economics, Norwegian School of
Economics and Business Administration, Helleveien 30, N-5045 Bergen,
Norway. voice: +47 55 95 93 55; fax +47 55 95 95 43
e-mail: Roger.Bivand at nhh.no