[R-sig-Geo] MEM : Spatial structures detection issues

[Roger Bivand]

On Sat, 30 Nov 2013, David Bauman wrote:

> Hi everyone,
>
> I am facing a problem I really do not know how to resolve about
> detecting significant spatial structures in a region I am studying.
>
> The study is about the Miombo forest (wooded savanna). I am working in
> a 10 ha permanent forest plot, where all trees are mapped and
> identified. I sampled 24 more or less regulately scattered plots of 25
> x 25 m over the 10 ha. In each plot of 25 x 25 m I got 5 soil samples
> on which I will measure a battery of variables (the explanatory
> variables).
> What I want to study is the bêta-diversity of the ectomycorrhizal
> community, so that in each plot I got about 15 pieces of root
> containing some ectomycorrhizal fungus on it (response variables).
>
> My purpose is to explain the spatial variation of the ectomycorrhizal
> diversity thanks to the soil variables I will have measured.
>
> To do so, I want to use a PCNM (or MEM).
>
> My problem is that :
>
> All the 24 plots are considered as points, so that the euclidean
> distance between them is overestimated. Two adjacent, contiguous,
> plots are considered to be 25 m far from each other, while the
> distance should be 0, since they "touch" each other.

Well, how are you representing the support of the plots? If you are only 
providing a central point rather than a polygon, you are getting what you 
asked for. If you are providing polygon boundaries, but using the 
coordinates() method to return the central points, the same follows. If 
tou need distances between polygon objects, use gDistance() in rgeos (if 
your positional data are projected).

It always helps if questioners show their reasoning by including a small 
example with code, as the problem may result from an unfortunate choice of 
functions in the workflow. Also always include the output of sessionInfo() 
as functions may behave differently in different versions of R and 
packages.

Hope this clarifies,

Roger

> This leads me to the problem that the truncation threshold distance is
> to high to allow the RDA of the response dataframe on the spatial PCNM
> variables to detect a significant linear link between both matrices
> (function anova.cca()).
>
> So the question is : How can I do so that the area of my plots are
> taken into account, and they are not considered as points anymore ?
> This would lower the truncation distance and probably allow me to
> detect spatial structures that, I really think, do exist.
>
> I hope someone with some experience and good ideas will be able to help.
>
> If you want some more concrete information about the study, do not
> hesitate to ask it to me.
>
> Thanks a lot,
>
> David Bauman
>
> _______________________________________________
> R-sig-Geo mailing list
> R-sig-Geo at r-project.org
> https://stat.ethz.ch/mailman/listinfo/r-sig-geo
>

-- 
Roger Bivand
Department of Economics, Norwegian School of Economics,
Helleveien 30, N-5045 Bergen, Norway.
voice: +47 55 95 93 55; fax +47 55 95 95 43
e-mail: Roger.Bivand at nhh.no