Loading ...
Sorry, an error occurred while loading the content.

[ai-geostats] 3D kriging with GSLIB

Expand Messages
  • Luke Spadavecchia
    Hello all, I am currently trying to get a 3D kriging to run using the KT3D program from GSLIB. I have managed to run a 2D version of this with no problems, but
    Message 1 of 1 , Feb 8, 2006
    • 0 Attachment
      Hello all,

      I am currently trying to get a 3D kriging to run using the KT3D program
      from GSLIB. I have managed to run a 2D version of this with no problems,
      but I just can't seem to work out the 3D implementation. My data is a
      set of time vectors from 113 weather stations across central Oregon.

      I have successfully run space-only 2D runs, using Monthly or Annual
      averages for each of the stations. Details of the 2D run can be seen on
      my website:

      http://www.geos.ed.ac.uk/homes/s0198247/Geostats.html

      Please forgive anything which is apocriphal/stupid on this page
      (feedback very welcome).

      The variogram was developed in space by taking the average of the
      spatial variograms for each monthly averaged dataset (average of 48
      variograms), as calculated by GeoR, and incorporating an elevation
      trend. The temporal variogram was calculated from the "flattened" and
      averaged data (altitude trend removed). I have managed to derive an
      anisotropic variogram which will describe the spatiotemporal variation,
      using the techniques suggested in the GSLIB manual - zonal anisotropy is
      handled by setting some structures very long or short ranges to mask
      their effects in certain directions. However, I think perhaps it is this
      which is causing my problem: When I come to run a 3D implementation,
      most of my grid is no-data (-999). The debug file says something like
      singular matrix for these points? Should I try to model the variogram as
      a seperable space-time system rather than a single 3D ellipsoid? Any
      help with this would be hugely appreciated.

      Here are some relevant bits of my parameter file, for those who are
      interested:

      Grid Definition:

      98 562.6186 1.0 -nx,xmn,xsiz
      86 4903.722 1.0 -ny,ymn,ysiz
      1827 1.0 1.0 -nz,zmn,zsiz

      My coordinate system is spatial in the x and y directions (UTM zone 10,
      measured in km), but temporal in the vertical (z) direction. The time
      step is measured in days. My grid is therefore 98 x 86 = 8,428 square
      kilometres, by 1827 days = 15,397,956 grid squares.

      Search ellipsoid:

      250.0 250.0 30.0 -maximum search radii
      0.0 0.0 0.0 -angles for search ellipsoid

      Am I right in thinking this means x_radius = 250 km, y_radius = 250 km,
      z_radius = 30 days?

      Variogram:

      3 0.0 -nst, nugget effect
      1 5.5 0.0 0.0 0.0 -it,cc,ang1,ang2,ang3
      3000.0 3000.0 5.8 -a_hmax, a_hmin, a_vert
      1 2.2 0.0 0.0 0.0 -it,cc,ang1,ang2,ang3
      0.001 0.001 10000.0 -a_hmax, a_hmin, a_vert
      2 1.0 0.0 0.0 0.0 -it,cc,ang1,ang2,ang3
      68.0 68.0 16.8 -a_hmax, a_hmin, a_vert

      Here I have no nugget effect, as the temporal variogram doesn't display
      a nugget effect. I attempt to impose an anisotropic nugget effect with
      the vary short scale structure. The teporal variogram has a much higher
      sill than the spatial variogram, so I impose this using very long ranges
      on the first spherical structure for the spatial dimensions. Is this
      appropriate? Should I specify these in order of max to min range, and
      use rotations (eg for the second structure ang2 = 90, and ranges 10000,
      0.001, 0.001)?

      Many thanks

      Luke Spadavecchia
    Your message has been successfully submitted and would be delivered to recipients shortly.