On 06/09/11 01:13, Alper ALTINOK wrote:
I appreciate for the advices, let me explain the situation a bit more with
another example;
Say, points are representing sampled fields for pests. Imagine you are making
samplings from tomato fields to reveal possible clusters (hotspots) of pests,
in a region, say 100km by 100km. Last year, you sampled 50 fields from that
region, and this year 50 fields sampled. You want to compare these two years by
means of cluster locations, but there is a problem; some tomato field locations
are same with previous year, but some are changed to (nearby fields) due to
crop-rotation, farmers' preferences etc., so these point sets do not overlap
perfectly. The question is, can you compare (and comment on) pest clusters
while point locations different among these point sets? (or, will it be
statistically correct?)
What I was hoping is, to find a way to tell, how (dis)similar the point
locations of year1 and year2 are? In other words, when compared, whether the
points of these two sampling sets are statistically significantly distributed
against each other, or not. If not, I think comparison of two point sets will
become possible. This is why I am looking for cross nearest neighor function.
Well I did my best to explain the situation, hope this clarifies the issue,
many thanks for the comments.
If I understand you correctly, the ``points'' of your pattern consist of
these 50 fields.
Each year each field may be classified as having an infestation of pests
or not. What
this gives you is a binary random field on an irregular "grid" of
points. That field is evolving
over time, and is/has been observed at a number (10?) of regularly
spaced times.
You are interested in knowing whether the locations of the ``1's''
(infestations) in your binary
random field (at each observation time) are dependent upon the locations
of the 1's at the
previous observation time.
The fly in the ointment is that sometimes some of the 50 points in your
irregular grid
change *a little bit* due to crop rotation etc. This seems to me to be
an aspect of the
problem that is very hard to model. Off the top of my head I would
approach it by
replacing any points that change by the *centroid* of their various
manifestations.
This I think, would be a perfectly reasonable approximation provided
that the amount
(distance) by which the points change location is *small* compared with
distances
between genuinely distinct points.
Given that this approximation is acceptable, how can you go about
answering the question
of interest? You are *not* dealing with a point process, so point
process techniques are
not of any use, although some of the machinery in spatstat would
probably be of use in
effecting the calculations that you need to make.
I don't believe that there are any pre-programmed techniques in spatial
statistical
analysis to handle the problem. Others may correct me if I am wrong
about this.
If I am correct then you will have to invent (and program up) your own
technique.
I have a few ideas about how you might go about this, but I will refrain
from making
any further suggestions at this time.
cheers,
Rolf Turner
_______________________________________________
R-sig-Geo mailing list
R-sig-Geo@r-project.org
https://stat.ethz.ch/mailman/listinfo/r-sig-geo
