g.setproj, r.proj, rant about s.surf.**

[Timothy Keitt]

I think you're referring to the situation where one measures the
distance to the n nearest neighbors and then applies some statistics
to determine over-dispersion or under-dispersion, etc.  I'm interested
in resampling a lattice where I have a coordinate pair and want to
find the pixel which best represents those coordinates (which may not
point exactly at the center of any pixel, but lies somewhere between
four or more neighbors).  I think choosing the pixel with the minimal
euclidean distance should work without distorting the pattern of the
image.  I'd be interested though in alternative methods.  Thanks for
the input.

Tim

>Nearest neighbor analysis as first introduced by Hopkins and Skellman
>(1954) is a technique devised to study spacing of plant individuals (i.e.
>the pattern of points in a plane; see Pielou 1969).  A major assumption of
>this technique is that each individual can be represented by a point in
>space surrounded by some circle of radius r, which implies geometric
>regularity.  For some individuals this technique seems to work fine 
>although Simberloff (1978) pointed out that nearest neighbor techniques
>tend to overestimate the degree of spatial regularity in ant-lion
>populations.  Because patches of interest are more typically 
>convoluted in nature, difficulties arise when trying to apply nearest 
>neighbor analysis.  For example, if we are interested in the spatial 
>pattern of some resource A we might ask how patches of A are distributed 
>across a landscape.  If A is found in regularly shaped patches then 
>nearest neighbor analysis might prove appropriate.  However, if A is 
>distributed in highly convoluted patches or in linear patches we face an 
>inherent problem of where to place our point of measure.  Do we 
>arbitraily select some point in the patch or try to determine the 
>geometric center?  If the point of measure is arbitrary then I can 
>imagine 2 patches, one on either side of a third with equivalent nearest 
>neighbor values but which in reality are very different from one another.
>             
>
>    
>
>Brad Robbins						Dept. of Biology
>brobbins at chuma.cas.usf.edu				Univ. of South Florida
>
>
>On 30 Aug 1994, Timothy Keitt wrote:
>
>> 
>> >
>> >Have you given any thought to how you might deal with the inherent 
>> >problems of nearest neighbor techniques when dealing with non-point data?
>> >
>> 
>> No.  Could you elaborate?
>> 
>> T.
>> 
>> 
>> 
>> 
>


================================================================
| Timothy H. Keitt                       Landscape and Systems |      
| The University of New Mexico                         Ecology |
| Department of Biology                                        |
| Albuquerque NM 87131              Open URL -> ftp://algodones|
| tkeitt at algodones.unm.edu      .unm.edu/pub/tkeitt/tkeitt.html|
================================================================