compiling tile correlations
The function tileCorrelation takes the
data.frame treeData, which includes the tile
column, and returns a summary matrix of correlation data.
The rows are tile numbers. The columns include the mean intra- and
inter-specific correlations for that tile
('intraMu', 'interMu') and the 90% interval
('intra05', 'intra95', 'inter05', 'inter95').
tileCorrelation( tdata, wide, varName = 'diam', FULLCOR = F ){
# tdata requires a tile column, 1, 2, ...
# returns:
# corMat - tile by mean pairwise correlation differences
# covMatrix - species by species mean covariances (not a covariance matrix)
# growth - mean species growthHere is an example for diameter growth for BCI data.
load( 'treeData_diam/BCI.rdata', verbose = T )
wide <- 30
treeData <- minCensus( treeData, xytree, minCens = 4 )
treeData <- getTile( treeData, wide = wide )
bciCor <- tileCorrelation( treeData, wide = wide, varName = 'diam' )
tileData <- bciCor$tileData
covMatrix <- bciCor$covMatrix
corMatrix <- bciCor$corMatrix
comMatrix <- bciCor$comMatrix
xmu <- bciCor$xmu
xse <- bciCor$xse
ntree <- bciCor$ntreeHere is the tiled stem map for BCI:
par( bty='n')
plot( treeData$x, treeData$y, cex = treeData$diam/200,
col = blueGreen(5)[ 2*(1 + treeData$tile %% 2) ],
asp = 1, xlab = '', ylab = '')
Thirty-meter tiles for BCI. Correlations between trees are evaluated at
this scale, defined by parameter wide.
The function tileCorrelation returns several
objects:
| Object | Structure | variables |
|---|---|---|
| tileData | tile by variables | mean intra/inter correlation, n, richness, diversity, density |
| corMatrix | spp by spp | mean pairwise correlation |
| covMatrix | spp by spp | mean pairwise covariance |
| comMatrix | spp by spp | no. of pairwise comparisons |
| xmu | tile by spp | mean response (‘diam’ or ‘fecEstMu’) |
| xse | tile by spp | SE of response |
| ntree | tile by spp | no. of trees |
The value NA is used where there are insufficient
comparisons.
This code gives the proportionate fractional dominance
library( corrplot )
corMatrix[ !is.finite(corMatrix) ] <- 0
covMatrix[ !is.finite(covMatrix) ] <- 0
corrplot( t(corMatrix), is.corr = T, type = 'lower', order = 'hclust', tl.pos = 'n',
addgrid.col = NA)The differences in scale for species makes the covariance matrix hard to plot (most values are white). Here is the matrix of correlations:
Mean pairwise correlations for BCI growth data, with high values (blue) clustered along the diagonal.
This code plots histograms:
par( mfrow=c(1,2), bty='n', omi = c(.2, .1, .1, .1) )
corHist( tileData[,2], ylab = 'Frequency', col = cols[2] )
corHist( tileData[,1], add = T )
legend( 'topright', c('intra','inter'), text.col = cols, bty='n')
title( 'a) Pairwise correlations', font.main = 6 )
corHist( tileData[,1] - tileData[,2], col = 'white', xlim = c(-.05, .2), ylim = c(0, 6) )
abline( v = 0, col = 'white', lwd=3)
abline( v = 0, lty=2, col = 'grey', lwd=3)
corHist( tileData[,1] - tileData[,2], col = cols[3], add = T )
title( 'b) Intra- minus inter-specific', font.main = 6 )
mtext('Correlation', 1, outer=T, line = -1.5, cex = 1.2)Below are histograms for mean correlations between individuals of the same and different species within a neighborhood (tile) at BCI. At right is the histogram of correlation differences.
Histograms of mean tile correlations between tree growth rates for intra- and inter-specific comparisons (left) with their difference (right) for BCI.
The next block uses the function getData to merge sites
by biome. To change this grouping edit your version of the function
groupSites( sites ) in the file
treeCorFunctions.R. Here are comparisons for diameter data
from multiple sites:
wide <- 30
varName <- 'diam'
minCens <- 4
minDiam <- 5
tmp <- getData( varName, wide, minCens, minDiam = minDiam )
tileList <- tmp$tileList
corList <- tmp$corList
covList <- tmp$covList
comList <- tmp$comList
xmu <- tmp$xmu
xse <- tmp$xse
tileList <- tileList[ sapply(tileList, nrow) > 10 ]
dlim <- c(-.2, 1)
breaks <- seq( -1.5, 1.5, by = .05 )
words <- plotSetup(names(tileList))
np <- nrow(words)
xlabels <- ylabels <- F
for( j in 1:np){
ylabels <- F
yaxt <- 'n'
if( j == 1 ){
ylabels <- T
yaxt <- 's'
}
tmp <- mergeCov( tileList[j], covList[j], corList[j], comList[j], cint = .04 )
x <- tmp$density
tl <- tmp$tileList
st <- tmp$stats
corHist( tl[,2], col = cols[2], xlim = dlim, ylim = c(0, 3),
breaks = breaks, xaxt = 'n', yaxt = yaxt )
corHist( tl[,1], col = cols[1], breaks = breaks, add = T )
title( words[j,1], font.main= 6, line = -.5, cex=.9 )
title( words[j,2], font.main= 6, line = -1.5, cex=.9 )
if( j == np )legend( 'right', c('intra','inter'), text.col = cols, bty='n')
corHist( tl[,1] - tl[,2], col = 'white', xlim = dlim, ylim = c(0, 3),
breaks = breaks, xaxt = 'n', yaxt = yaxt )
abline( v = 0, col = 'white', lwd=3)
abline( v = 0, lty=2, col = 'grey', lwd=3)
corHist( tl[,1] - tl[,2], col = cols[3], breaks = breaks, add = T )
ylim <- c(.1,100)
dif <- tl[,1] - tl[,2]
wf <- which( is.finite(dif) )
plot( dif, tl[,'BA'], xaxt = 'n', ylim = ylim, yaxt = 'n',
xlab = '', ylab = '', xlim = dlim, log = 'y', cex=.8,
pch = 16, col = .getColor( cols[3], .4) )
axis( 2, labels = ylabels)
axis( 1, labels = F )
abline( v = 0, lwd = 3, col = 'white' )
abline( v = 0, lwd = 3, lty = 2, col = 'grey' )
if( length(wf) > 5 )reg2plot( dif, tl[,'BA'], col = cols[1], p = .1, ylim = ylim )
ylim <- c(.5,5)
plot( dif, tl[,'diversity'], xaxt = 'n', yaxt = 'n',
xlab = '', ylab = '', ylim = ylim, log = 'y', xlim = dlim, cex=.8,
pch = 16, col = .getColor( cols[3], .3) )
axis( 2, labels = ylabels)
axis( 1, labels = T )
abline( v = 0, lwd = 3, col = 'white' )
abline( v = 0, lwd = 3, lty = 2, col = 'grey' )
if( length(wf) > 5 )reg2plot( dif, tl[,'diversity'], col = cols[1], p = .1, ylim = ylim )
}
mtext( paste0( c( 'Diversity', 'BA (m2/ha)', 'Cor Difference', 'Correlation' ),
collapse = ' '), 2, outer=T, line = 1 )
mtext( 'Correlation', 1, outer = T, line = 3 )
Intra- and inter-specific comparisons (first row) for diameter growth, with the difference histogram (2nd row). The two rows at bottom plot the correlation difference against the tile density and diversity (Shannon entropy).
Same as above with wide = 50 m.
Intra- and inter-specific comparisons (first row) for fecundity, with the difference histogram (2nd row). The two rows at bottom plot the correlation difference against the tile density and diversity (Shannon entropy).
Mean correlation difference by site for diameter growth.
