r.resample, r.coarsen

[Jianping Xu]

Michael Shapiro wrote:

> GRASS only does nearest neighbor resampling, you're right.  Would you
> consider contributing r.coarsen?
> 

I'd like to polish the source code and then send it to CERL. The current 
r.coarsen is a ragged modification of r.neighbors, with variable names not
appropriate to r.coarsen... It was an ad hoc solution for our need in a 
project.

Please note that my original post mentioned the bilinear resampling and cubic
convolution resampling. R.coarsen, too, does NOT include these two options. 
I merely borrowed the neighborhood operation methods from r.neighbors for 
aggregation. I believe a combination of r.coarsen and any of the resampling
methods - nearest neoghbor, bilinear, convolution - can solve most aggregation
problems. I may or may not try to code the other two resampling options 
depending on my schedule and possible contributions to done by others.

I received more than dozen requests for the code since I post. I thank all
of you for your interest in the module, and look forward to hear your 
comments and bug reports.

Jianping Xu
jianp at ocean.rutgers.edu

> In <9306112059.AA13489 at dune.rutgers.edu> jianp at dune.rutgers.edu (Jianping Xu) writes:
> 
> >First, it seems to me, GRASS has only one resampling option, nearest neighbor.
> >It does not have bilinear and cubic convolution options.
> 
> >Second, if you want to aggregate images/raster maps with a very fine resolution 
> >(e.g. 10, 20, 30 meters) to a very coarse resolution (e.g. 80, 1000 meters), 
> >you would not get a result you expected because of the nearest neighbor method 
> >(and even all the above three). R.resample picks up information by skipping 
> >many raster cells, yet assigns collected values back to those skipped portions 
> >in the resultant map.
> 
> >We have a simple raster program, "r.coarsen", which aggregate the input map
> >to coarser resolutions, but allowing various aggregation methods - mean,
> >mode, max, min, ... - taken from the standard "r.neighbors" program. Unlike
> >r.neighbors which moves the working window (not the region!) one cell a time,
> >r.coarsen moves the window one full window a time. Send us a mail privately,
> >I'll provide you the code.
> 
> 
> >*****
> >Jianping Xu					Email:	jianp at ocean.rutgers.edu
> >Center for Remote Sensing and Spatial Analysis	Phone:	(908) 932-9631
> >Rutgers University				Fax:	(908) 932-8644
> >New Brunswick, NJ 08903-0231
> 
> -- 
> Michael Shapiro               shapiro at zorro.cecer.army.mil
> U.S. Army CERL                (217) 373-7277
> P.O. Box 9005 
> Champaign, Ill. 61826-9005
>