[R-sig-Geo] Moran's I for grid data?

[Sandra Burmeier]

Hi,

I have some questions concerning the calculation of Moran’s I for 
grid-based data.

I’m a plant ecologist analysing the small-scale composition of 
populations of a certain plant species. That is, I’m intending to 
compare the spatial distribution of adult plants, juvenile plants, 
seedlings and seeds in the seed bank. I have sampled several populations 
in 1x4-m-plots which were subdivided by a 20x20-cm-grid, and my data are 
aggregated values for the grid cells (e.g. 5 adult plants in cell 1, 12 
plants in cell 2 and so on).
The questions I would like to answer are the following:
(1) Do adults, juveniles etc. are aggregated within the plots 
(visualisation of the data in a simple map strongly suggests they are)?
(2) Is the level of aggregation (the average size of the clumps) the 
same for different life stages, i.e. do adults and juvenile clump on the 
same scale?

My plan was to calculate Moran’s I values (to answer question 1) and 
draw individual spatial correlograms for the different life stages and 
then compare the graphs (to answer question 2) – but is that appropriate 
for my grid-based data?

In case it is, I have some more questions. So far, I’ve used the package 
spdep, calculated Moran’s I using the function moran.test (using 
knn-objects with the nearest 2, 3 and 4 neighbours – any guidelines here 
on how many make sense?) and plotted the results using sp.correlogram:
/xy.A2.07 <-cbind(A2.07$x,A2.07$y)/
/xy.A2.07 <-matrix(xy.A2.07,ncol=2)/
/xy.A2.07.knn <- knearneigh(xy.A2.07, k=2) # I’ve also tried k=3 and k=4/
/xy.A2.07.nb <- knn2nb(xy.A2.07.knn)/
/xy.A2.07.lw <- nb2listw(xy.A2.07.nb,style="B")/
/moran.test(A2.07$Ara_bl,xy.A2.07.lw) /
/corr.A2.07.bl <- sp.correlogram(xy.A2.07.nb, A2.07$Ara_bl, method = 
"I", order=6, zero.policy=TRUE, style="B")/
/print.spcor(corr.A2.07.bl, "bonferroni")/
/plot.spcor(corr.A2.07.bl)/

Is there any possibility to define the lag-distances used by 
sp.correlogram? Or at least to find out what they are (in terms of the 
map units used, in my case centimetres)?

I am new to both spatial statistics and R, so these questions may be 
rather trivial - my apologies if they are. However, I am really 
wondering whether my approach is appropriate for the kind of data I’m 
dealing with and, if it isn’t, which other method would be suitable.

I would greatly appreciate any help!

Many thanks,
Sandra

-- 
Sandra Burmeier
PhD-Student
Justus-Liebig-University Giessen
Division of Landscape Ecology and Landscape Planning
Heinrich-Buff-Ring 26-32
35392 Giessen
Germany
E-mail: sandra.burmeier at umwelt.uni-giessen.de
http://www.uni-giessen.de/landscape