[ai-geostats] Re: [R-sig-Geo] Software (R, free matlab toolboxes etc etc...)
- Koen Hufkens wrote:
>Hi list(s),Hi Koen,
>I'm looking for (free) geostatistical software that works with images
>and not as much on point data.
>I've been playing around with multiresolution wavelet decompositions and
>I want to collect some statistical information of these images
>(semivariance analysis, autocorrelation, Moran's I index, etc etc...).
>For the time being I used R, the wavethresh package and the spdep
>package. This latter was not optimal because the neighbour list
>constructed by spdep becomes a little to much for my limited computer
>system. So an additional R package would be great, but not necessary.
>Other (free) alternatives would be great.
>So what packages toolboxes etc. are not point based, and can handle
>images? Any idea's, suggestions?
R package gstat (or the gstat stand-alone program) has two
optimizations built in for calculating semivariograms, the first for
large data sets, the second for gridded data:
1. it doesn't store the full pairwise distance matrix, but accumulates
semivariance and number of point pairs,
2. for gridded data, it derives distances from row/column index, instead
of recalculating them for each point, and it only evaluates point pairs
for which we know they need evaluation.
The first makes the calculation of semivariances possible within limited
memory, the second one makes the behaviour of the computation
closer to O(n), instead of O(n*n) for the case where you ignore
grid topology---this makes the computation scalable and feasable
for larger grids.
The documentation may not be all too clear on this, and perhaps
the user interface can be improved: not many people need this;
contact me on the gstat-info list for details where needed.