paleotree (version 3.4.7)

communityEcology: Miscellaneous Functions for Community Ecology

Description

This is just a small collection of miscellaneous functions that may be useful, primarily for community ecology analyses, particularly for paleoecological data. They are here mainly for pedagogical reasons (i.e. for students) as they don't appear to be available in other ecology-focused packages.

Usage

pairwiseSpearmanRho(
  x,
  dropAbsent = "bothAbsent",
  asDistance = FALSE,
  diag = NULL,
  upper = NULL,
  na.rm = FALSE
)
HurlbertPIE(x, nAnalyze = Inf)

Value

pairwiseSpearmanRho will return either a full matrix (the default) or (if asDistance is true, a distance matrix, with only the lower triangle shown (by default). See details.

HurlbertPIE returns a named vector of PIE values for the input data.

Arguments

x

The community abundance matrix. Taxonomic units are assumed to be the columns and sites (samples) are assumed to be the rows, for both functions. The abundances can be absolute counts of specimens for particular taxa in each sample, or it can be proportional (relative) abundances, where all taxon abundances at a site are divided by the total number of specimens collected at that site. For function pairwiseSpearmanRho, the input x should be a matrix with two dimensions. For HurlbertPIE, a vector (an object with only one dimension) will be treated as if it was matrix with a single row and number of columns equal to its length.

dropAbsent

Should absent taxa be dropped? Must be one of either: dropAbsent = 'bothAbsent' (drop taxa absent in both sites for a given pairwise comparison), dropAbsent = 'eitherAbsent' (drop taxa absent in either site), or dropAbsent = 'noDrop' (drop none of the taxa absent in either site). The default dropAbsent = 'bothAbsent' is recommended; see examples.

asDistance

Should the rho coefficients be rescaled on a scale similar to dissimilarity metrics, i.e. bounded 0 to 1, with 1 representing maximum dissimilarity (i.e. a Spearman rho correlation of -1)? (Note that dissimilarity = (1 - rho) / 2 )

diag

Should the diagonal of the output distance matrix be included?

upper

Should the upper triangle of the output distance matrix be included?

na.rm

Should taxa listed with NA values be dropped from a pair-wise site comparison? If na.rm = FALSE, the returned value for that site pair will be NA if NAs are present.

nAnalyze

Allows users to select that PIE be calculated only on the nAnalyze most-abundant taxa in a site sample. nAnalyze must be a vector of length = 1, consisting of a whole-number value greater than 1. By default, nAnalyze = Inf so all taxa are accepted. Note that if there are less taxa in a sample than nAnalyze, the number present will be used.

Author

David W. Bapst

Details

pairwiseSpearmanRho returns Spearman rho correlation coefficients based on the rank abundances of taxa (columns) within sites (rows) from the input matrix, by internally wrapping the function cor.test. It allows for various options that automatically allow for dropping taxa not shared between two sites (the default), as well as several other options. This allows the rho coefficient to behave like the Bray-Curtis distance, in that it is not affected by the number of taxa absent in both sites.

pairwiseSpearmanRho can also rescale the rho coefficients with (1-rho)/2 to provide a measure similar to a dissimilarity metric, bounded between 0 and 1. This function was written so several arguments would be in a similar format to the vegan library function vegdist. If used to obtain rho rescaled as a dissimilarity, the default output will be the lower triangle of a distance matrix object, just as is returned by default by vegdist. This behavior can be modified via the arguments for including the diagonal and upper triangle of the matrix. Otherwise, a full matrix is returned (by default) if the asDistance argument is not enabled.

HurlbertPIE provides the 'Probability of Interspecific Encounter' metric for relative community abundance data, a commonly used metric for evenness of community abundance data based on derivations in Hurlbert (1971). An optional argument allows users to apply Hurlbert's PIE to only a subselection of the most abundant taxa.

References

Hurlbert, S. H. 1971. The nonconcept of species diversity: a critique and alternative parameters. Ecology 52(4):577-586.

See Also

twoWayEcologyCluster; example dataset: kanto

Examples

# let's load some example data:
# a classic dataset collected by Satoshi & Okido from the Kanto region

data(kanto)

rhoBothAbsent <- pairwiseSpearmanRho(kanto,dropAbsent = "bothAbsent")

#other dropping options
rhoEitherAbsent <- pairwiseSpearmanRho(kanto,dropAbsent = "eitherAbsent")
rhoNoDrop <- pairwiseSpearmanRho(kanto,dropAbsent = "noDrop")

#compare
layout(1:3)
lim <- c(-1,1)
plot(rhoBothAbsent, rhoEitherAbsent, xlim = lim, ylim = lim)
	abline(0,1)
plot(rhoBothAbsent, rhoNoDrop, xlim = lim, ylim = lim)
	abline(0,1)
plot(rhoEitherAbsent, rhoNoDrop, xlim = lim, ylim = lim)
	abline(0,1)
layout(1)

#using dropAbsent = "eitherAbsent" reduces the number of taxa so much that
	# the number of taxa present drops too low to be useful
#dropping none of the taxa restricts the rho measures to high coefficients
	# due to the many shared 0s for absent taxa

#############

# Try the rho coefficients as a rescaled dissimilarity
rhoDist <- pairwiseSpearmanRho(kanto,asDistance = TRUE,dropAbsent = "bothAbsent")

# What happens if we use these in typical distance matrix based analyses?

# Cluster analysis
clustRes <- hclust(rhoDist)
plot(clustRes)

# Principle Coordinates Analysis
pcoRes <- pcoa(rhoDist,correction = "lingoes")
scores <- pcoRes$vectors
#plot the PCO
plot(scores,type = "n")
text(labels = rownames(kanto),scores[,1],scores[,2],cex = 0.5)

##################################

# measuring evenness with Hurlbert's PIE

kantoPIE <- HurlbertPIE(kanto)

#histogram
hist(kantoPIE)
#evenness of the kanto data is fairly high

#barplot
parX <- par(mar = c(7,5,3,3))
barplot(kantoPIE,las = 3,cex.names = 0.7,
	ylab = "Hurlbert's PIE",ylim = c(0.5,1),xpd = FALSE)
par(parX)

#and we can see that the Tower has extremely low unevenness
	#...overly high abundance of ghosts?

# NOTE it doesn't matter whether we use absolute abundances
# or proportional (relative) abundances
kantoProp<-t(apply(kanto,1,function(x) x/sum(x)))
kantoPropPIE <- HurlbertPIE(kantoProp)
identical(kantoPIE,kantoPropPIE)

#let's look at evenness of 5 most abundant taxa
kantoPIE_5 <- HurlbertPIE(kanto,nAnalyze = 5)

#barplot
parX <- par(mar = c(7,5,3,3))
barplot(kantoPIE_5,las = 3,cex.names = 0.7,
	ylab = "Hurlbert's PIE for 5 most abundant taxa",ylim = c(0.5,1),xpd = FALSE)
par(parX)