raster correlations

[Bruce K. Wylie (303) 490 8336]

I used r.mask to identify sites (sites file=> vector=> raster for mask)
Then is used r.covar with the -r  option to get correlations using
only the pixels corresponding to site locations. 

Some geostatistic functions such as semivarigrams and kriging would
be nice additions to GRASS.

Regards,
Bruce Wylie

----- Begin message from MX%"grassu-list at max.cecer.army.mil"  7-Dec-92


I'm also interested in these kinds of correlations. I've used
"r.coin" to tabulate coincidence of archaeological sites and
aspects of their environment. This does produce Chi-square
results, but is rather limited in its applications. There are
two problems with my kind of data:
- one is that there are very few sites with respect to the
extent of the area (38 sites for 110 square km or 250,000 cells),
severely limiting the kind of statistical analysis you can use.
- the other is the spatial interrelation of my data (and yours?).
Most statistical tests for correlation proceed from the assump-
tion that the data are independent. Work is in progress to
develop stat tests that do reckon with spatial interrelation, but
GRASS does not provide any of this yet.

If you find the information you need and are able to develop a
useful test of the correlation between two or more raster maps,
could you email your solution to me?

Martijn at scanner.frw.uva.nl
>
> Please help! I thought I had read grassu-mail on how someone
> had used r.mapcalc to do correlations between raster files
> (like the correlation coefficients you can get from doing
> x vs y scatter diagrams). I now cannot find this mail message
> anywhere. Does it sound familiar to anyone? I have a lot of
> rasters of chemical parameters and am interested in seeing how
> well high and low areas of one item correlate to highs and lows
> of another. Any suggestions will be appreciated--
>
> stewalt at ncrds.er.usgs.gov
>



----- End forwarded message