[R-sig-Geo] MSE
Ashton Shortridge
ashton at msu.edu
Mon Dec 14 20:03:19 CET 2009
Hi Pete,
well, if you want to compare different model results at predicted locations
(e.g. IDW vs krging, or ordinary kriging vs. kriging with a trend, or local vs
global kriging, etc etc, you could use some kind of mean squared difference.
Maybe something like this:
kr.thing1 <- krige(stuff)
kr.thing2 <- krige(stuff)
RMS.t1t2 <- sqrt(mean((kr.thing1$var1.pred - kr.thing2$var1.pred)^2))
would do what you want?
Ashton
On Monday 14 December 2009 12:08:58 Pete Larson wrote:
> Hello,
>
> Thank you for your reply. I would like to do model comparison, but the
> cv function is just too slow for now.
>
> Any other ideas?
>
> Thanks,
>
> Pete
>
> Ashton Shortridge wrote:
> > Hi Tom,
> >
> > There isn't any Mean Squared Error from kriging, assuming you are using
> > all the points in the model, because there are no residuals. Kriging
> > 'honors' the data. You can use do cross-validation (defaults to leave one
> > out), but this is pretty intensive and slow, at least in the R version
> > (krige.cv).
> >
> > Gtsat Kriging does return a variance estimate for every prediction point,
> > the square root of which is often called the kriging standard error, but
> > this is a measure of uncertainty at that point, rather than actual error.
> >
> > Or perhaps I've completely misunderstood your question!
> >
> > Ashton
> >
> > On Monday 14 December 2009 06:16:20 Pete Larson wrote:
> >> Hello,
> >>
> >> Is there any easy way to get the MSE from the gstat kriging fucntions?
> >> Or just the residuals at the measurement points?
> >>
> >> Thanks,
> >>
> >> Pete
> >>
> >> Edzer J. Pebesma wrote:
> >>> Dear Tom,
> >>>
> >>> as your grids are spatial grids, you may want to look into the web
> >>> service standards of the OGC (www.opengeospatial.org) for them, WCS
> >>> (for reading) and WCS-T (for writing) in particular. We're
> >>> experimenting with this currently, using the recently open sourced
> >>> rasdaman software (www.rasdaman.org), which uses a data base in the
> >>> back end. I don't think OPeNDAP has special infrastructure for spatial
> >>> grids; CFNetCDF does, but by choosing this you may limit integration
> >>> with other software that does assume spatial data to be served over
> >>> OGC standards. CFNetCDF is one of the possible formats to be served
> >>> over WCS (and GeoTIFF etc) -- if I'm correctly informed.
> >>>
> >>> Bests,
> >>>
> >>> Stefano Casalegno wrote:
> >>>> Dear Tom,
> >>>> my best acknowledgment for the Ecogrid challenging initiative.
> >>>> I guess the main limitation will be "time" and this is related to
> >>>> the size of grids.
> >>>> What is the extent of the grids ?
> >>>> What tool you plan to use for point 1. Overlay some point data and
> >>>> get the values of grids ?
> >>>>
> >>>> On a more theoretical basis,
> >>>> "distribution models should rely upon data that covers the whole
> >>>> extent of the distribution of the biota under examination"
> >>>>
> >>>> Please consider forest data from
> >>>> http://forest.jrc.ec.europa.eu/climate-change
> >>>> as supplementary predictors variables within an European
> >>>> biodiversity and nature conservation application.
> >>>>
> >>>> Cheers
> >>>> Stefano
> >>>>
> >>>>
> >>>> ________________________________
> >>>>
> >>>> Stefano Casalegno, Ph.D.
> >>>>
> >>>> via Greppi 16, 21021 Angera - Italy
> >>>> email: stefano at casalegno.net
> >>>> web: http://www.casalegno.net/stefano
> >>>> ________________________________
> >>>>
> >>>> On Dec 13, 2009, at 6:29 PM, Tomislav Hengl wrote:
> >>>>> Dear R-sig-geo,
> >>>>>
> >>>>> As a part of our project (EcoGRID.nl) we have prepared some 60
> >>>>> thematic grids that we use as auxiliary predictors for species
> >>>>> distribution modeling. At this stage, we would like to put the
> >>>>> gridded maps (50/100 m base resolution) into some efficient
> >>>>> sharable geo-database.
> >>>>>
> >>>>> We will most probably put the data into the NetCDF format (http://
> >>>>> www.unidata.ucar.edu/software/netcdf/) because it can handle any-
> >>>>> dimensional array data, and because it has been in continuous
> >>>>> development and widespread use for many years. NetCDF grids can be
> >>>>> read relatively easy into R using the RNetCDF package (e.g. http://
> >>>>> spatial-analyst.net/DATA/readNCDF.zip). Another alternative is to
> >>>>> use PostGIS WKT Raster format (http://trac.osgeo.org/postgis/wiki/
> >>>>> WKTRaster), but this seems to be still rather experimental (?).
> >>>>>
> >>>>> Once we put the grids into NetCDF format. We plan to install
> >>>>> OPeNDAP (http://www.opendap.org) server on top to make the files
> >>>>> accessible through the web; then Geoserver (http://
> >>>>> www.geoserver.org) or UMN Mapserver (http://mapserver.org) to feed
> >>>>> a WMS from NetCDF files (raster data). Finally, we plan to add a
> >>>>> simple OpenLayers interface on top of that (Geoserver has it built
> >>>>> in) to allow direct browsing of the data and metadata (e.g. such as
> >>>>> this one: http://africamap.harvard.edu/)
> >>>>>
> >>>>> Just to be clear, we want to put the data on a server because we
> >>>>> would like to run a number of operations directly on the server
> >>>>> (via rgdal?):
> >>>>>
> >>>>> 1. Overlay some point data and get the values of grids (without a
> >>>>> need to download the grids locally);
> >>>>> 2. Subset/mask and resample gridded data of interest (for a given
> >>>>> bounding box and proj4 string; again without a need to download the
> >>>>> data locally);
> >>>>> 3. Upscale the grids from 100 m to 250, 500 m and 1 km resolution
> >>>>> (then download the upscaled grids).
> >>>>> 4. Write/upload new grids to the database (e.g. using WebDAV).
> >>>>> 5. Browse the grids (and metadata) via the OpenLayers.
> >>>>>
> >>>>> These are only our wishes of course. We do not know if all this is
> >>>>> really possible with the current software.
> >>>>>
> >>>>> Any examples or comments/suggestions/experiences are welcome
> >>>>> (before we start installing and testing the functionality).
> >>>>>
> >>>>> Thanx!
> >>>>>
> >>>>> Tomislav Hengl and Lourens Veen
> >>>>>
> >>>>> _______________________________________________
> >>>>> R-sig-Geo mailing list
> >>>>> R-sig-Geo at stat.math.ethz.ch
> >>>>> https://stat.ethz.ch/mailman/listinfo/r-sig-geo
> >>>>
> >>>> [[alternative HTML version deleted]]
> >>>>
> >>>> _______________________________________________
> >>>> R-sig-Geo mailing list
> >>>> R-sig-Geo at stat.math.ethz.ch
> >>>> https://stat.ethz.ch/mailman/listinfo/r-sig-geo
> >>>
> >>> _______________________________________________
> >>> R-sig-Geo mailing list
> >>> R-sig-Geo at stat.math.ethz.ch
> >>> https://stat.ethz.ch/mailman/listinfo/r-sig-geo
> >>
> >> _______________________________________________
> >> R-sig-Geo mailing list
> >> R-sig-Geo at stat.math.ethz.ch
> >> https://stat.ethz.ch/mailman/listinfo/r-sig-geo
>
--
Ashton Shortridge
Associate Professor ashton at msu.edu
Dept of Geography http://www.msu.edu/~ashton
235 Geography Building ph (517) 432-3561
Michigan State University fx (517) 432-1671
More information about the R-sig-Geo
mailing list