Kriging and Inverse Distance Interpolation using GSTAT

InterpolationGstat performs inverse distance or kriging interpolation using gstat.

Description

Kriging and inverse distance are popular interpolation methods, especially in earth sciences. There are some routines already available on matlab but are severely limited by matlabs memory constraints. By using gstat to handle interpolation and variogram modelling, very large datasets are easily handled.

USAGE: Requires gstat.exe (freeware) to run (www.gstat.org).

% INVERSE DISTANCE:
% INPUTS:
% x :: sample locations in x
% y :: sample locations in y
% z :: sample data
% res :: resolution of interpolation grid
% radius :: maximum search radius for interpolation
% method = 'invdist'
% uselog :: 1 if z values should be log transformed, 0 if not
% name :: name of output file without extension. e.g. 'predictions'.
% (file is given extension '.grd')
% useloc (OPTIONAL) :: if 1, will use file 'locations.dat' if it exist for
% interpolation locations. If 0, will prompt user to choose whether to use
% locations.dat (if it exists) or rewrite. Useful when calling the
% interpolation routine in a loop.
% S is not required for inverse distance
%
% KRIGING:
% INPUTS:
% x,y,z,res,radius :: as for inverse distance
% method = 'kriging'
% uselog :: as for inverse distance
% name :: as for inverse distance
% S :: Structure containing variogram information as output
% by the variogramfit function.
% useloc (OPTIONAL):: as for inverse distance
%
% OUTPUTS:
% xi :: the prediction locations in x
% yi :: the prediction locations in y
% zi :: the predicted result
% pcolor can be used to visualise. e.g. pcolor(xi,yi,zi); shading interp;

For variogram modelling, use SampleVarioGstat.m to create the experimental variogram. The excellent variogramfit by Wolfgram Schwanghart should be used to fit the experimental variogram. File ID: #25948.

Look at the readme for tips on kriging and inverse distance interpolation, and help InterpolationGstat and help SampleVarioGstat for correct usage.

NOTE:
This routine is intended to make it easy to perform kriging or inverse distance interpolation. It is not intended as a full interface to all the capabilities of gstat. As such only 2D ordinary kriging, inverse distance and experimental variogram modelling are supported, with only key parameters being offered as options.
 

Acknowledgements

The author wishes to acknowledge the following in the creation of this submission:
variogramfit

• MATLAB release: MATLAB 7.9 (2009b)

[Ian Howat]

 

[Sean de]

I don't have gstat; by looking at the code this appears to be a file worthy of much more than 1 star.

[James Ramm]

Gstat is freely available from gstat.org. Its a small executable - just save it to the same directory as InterpolationGstat and you are set.

[Wolfgang Schwanghart]

Hi James, thanks for submitting this function and for the credits! However, I think the functions still need some polishing, since they throw errors.

E.g.
****************
[xi,yi,zi] = InterpolationGstat(x,y,z,0.01,0.5,'kriging',0,'predictions',S,0);
??? Undefined function or variable "use".

Error in ==> InterpolationGstat at 96
if strcmp(use,'n')

*******************
[xi,yi,zi] = InterpolationGstat(x,y,z,0.01,0.5,'kriging',0,'predictions',S,1);
Writing Data File
Invoking Gstat...
gstat: Win32/MinGW version 2.4.1 (12 March 2003)
Copyright (C) 1992, 2003 Edzer J. Pebesma
using Marsaglia's random number generator
data(Krig): SampleData.dat (Ascii table file)
attribute: col[3] [x:] x_1 : [ -1.998, 1.998]
n: 1000 [y:] y_2 : [ -1.999, 1.998]
sample mean: 0.073122 sample std.: 2.40105
[using ordinary kriging]
gstat: read failed on file `locations'
ans =

     7

??? Error using ==> fgetl at 44
Invalid file identifier. Use fopen to generate a valid file identifier.

Error in ==> ImportSurferGrd at 15
code=fgetl(grdfile); % Reads surfer code 'DSAA'

Error in ==> InterpolationGstat at 178
zi =ImportSurferGrd([name,'.grd']);

****************

In addition, the interpolationgstat seems to work only, if your current folder is the folder where gstat.exe is located. Can this restriction be avoided?

I think, it is a great idea to call gstat from within matlab. At their current state, however, your functions still need some editing to work. Looking forward to an update, Wolfgang

[James Ramm]

Hi Wolfgang,
Thanks for letting me know the errors.
I just uploaded a corrected version (should appear tomorrow, april 22nd)
Tested with 'peaks' data sets for both inverse distance and kriging.
Please let me know about any other errors.
Gstat still needs to be in your working directory. I will make this more flexible in the near future.

[Wolfgang Schwanghart]

Hi James,
thanks for the update. There is still a minor error, that could be easily avoided. If you call InterpolationGstat for the first time and set useloc to true, than an error message tells me that File does not exist. Calling it with useloc = false is fine then. Calling it with three output arguments and useloc = true again, returns an error again.

Error in ==> InterpolationGstat at 61
if nargin<8

??? Output argument "xi" (and maybe others) not assigned during call to
"U:\WDmatlab\toolboxes\InterpGstat\InterpolationGstat.m>InterpolationGstat".

Yet, otherwise, the function works well now. The only thing I would like to see in the future is that InterpolationGstat returns the kriging variance as well. I have included it in my kriging function on the FEX because I think it is one of the most important features of kriging.

Best regards, Wolfgang