In the case of regular sampling designs, the number of dbMEM with positive
Moran’s I is close to n/2. The figures show that these variables are cosines and that
they range from broadest to finest scales. As mentioned above, this does not imply
that only periodical structures can be modelled by dbMEM analysis, however. Even
short-range spatial correlation can be modelled by fine-scaled dbMEM variables.
This topic will be addressed later.
7.4.2.3 dbMEM Analysis of the Mite Data
dbMEM analysis is not restricted to regular sampling designs. The drawback with
irregular designs is that the dbMEM variables lose the regularity of their shapes,
making the assessment of scale more difficult at times.
Now it is time to apply dbMEM analysis to the Hellinger-transformed oribatid
mite dataset. In the code below, as in the examples above, dbMEM variables are
constructed using function dbmem() of package adespatial
2 .
## Step 1. Construct the matrix of dbMEM variables
mite.dbmem.tmp <- dbmem(mite.xy, silent = FALSE)
mite.dbmem <- as.data.frame(mite.dbmem.tmp)
# Truncation distance used above:
(thr <- give.thresh(dist(mite.xy)))
# Display and count the eigenvalues
attributes(mite.dbmem.tmp)$values
length(attributes(mite.dbmem.tmp)$values)
Hint The argument silent=FALSE of function dbmem() allows the function to
display the truncation level onscreen; this level is a little bit larger than the
largest distance among sites that keeps all sites connected in a minimum spanning
tree. Here we computed this distance separately by means of function
give.thresh() of package adespatial.
As one can see, there are 22 dbMEM eigenvectors with positive spatial correlation. Prior to our RDA, we will apply forward selection with the Blanchet et al.
(2008a) double stopping criterion.
2 To compute “classical” PCNM eigenfunctions, users can apply function pcnm() of the vegan
package. However, that function does not provide the values of the Moran’s I indices to identify the
eigenvectors modelling positive spatial correlation. These would have to be computed separately.
320
7 Spatial Analysis of Ecological Data
Moran’s I is close to n/2. The figures show that these variables are cosines and that
they range from broadest to finest scales. As mentioned above, this does not imply
that only periodical structures can be modelled by dbMEM analysis, however. Even
short-range spatial correlation can be modelled by fine-scaled dbMEM variables.
This topic will be addressed later.
7.4.2.3 dbMEM Analysis of the Mite Data
dbMEM analysis is not restricted to regular sampling designs. The drawback with
irregular designs is that the dbMEM variables lose the regularity of their shapes,
making the assessment of scale more difficult at times.
Now it is time to apply dbMEM analysis to the Hellinger-transformed oribatid
mite dataset. In the code below, as in the examples above, dbMEM variables are
constructed using function dbmem() of package adespatial
2 .
## Step 1. Construct the matrix of dbMEM variables
mite.dbmem.tmp <- dbmem(mite.xy, silent = FALSE)
mite.dbmem <- as.data.frame(mite.dbmem.tmp)
# Truncation distance used above:
(thr <- give.thresh(dist(mite.xy)))
# Display and count the eigenvalues
attributes(mite.dbmem.tmp)$values
length(attributes(mite.dbmem.tmp)$values)
Hint The argument silent=FALSE of function dbmem() allows the function to
display the truncation level onscreen; this level is a little bit larger than the
largest distance among sites that keeps all sites connected in a minimum spanning
tree. Here we computed this distance separately by means of function
give.thresh() of package adespatial.
As one can see, there are 22 dbMEM eigenvectors with positive spatial correlation. Prior to our RDA, we will apply forward selection with the Blanchet et al.
(2008a) double stopping criterion.
2 To compute “classical” PCNM eigenfunctions, users can apply function pcnm() of the vegan
package. However, that function does not provide the values of the Moran’s I indices to identify the
eigenvectors modelling positive spatial correlation. These would have to be computed separately.
320
7 Spatial Analysis of Ecological Data