# Triplots of the parsimonious RDA (with fitted site scores)
par(mfrow = c(1, 2))
# Scaling 1
triplot.rda(spe.rda.pars,
site.sc = "lc",
scaling = 1,
cex.char2 = 0.8,
pos.env = 2,
mult.spe = 0.9,
mult.arrow = 0.92,
mar.percent = 0.01
)
# Scaling 2 : see accompanying material
Since there is now a third significant canonical axis, you could plot other
combinations: axes 1 and 3, and axes 2 and 3.
This triplot indeed presents the same structures as the one produced with all
explanatory variables (Fig. 6.1a). The sites and species show the same relationships.
The three selected explanatory variables are sufficient to reveal the major features of
the data.
6.3.2.7 Environmental Reconstruction: Projecting New Sites in an RDA
to Estimate the Values of Explanatory Variables
An RDA model is generally used to interpret the structure of the response data as
explained by the set of independent variables. But if the model is built from species
that act as bioindicators of environmental conditions represented by the explanatory
variables and if the model explains a fairly large amount of variance, then RDA can
be applied in an opposite way, i.e. to estimate the values of (quantitative)
explanatory variables on the basis of the abundances of the species. This is sometimes also called calibration or bioindication. It can be computed with function
calibrate() of package vegan.
As an illustration, let us imagine two new sites along the Doubs River where
fishes were captured and counted, but no environmental variable was measured. In
this example, the new sites are made of the (rounded) mean abundances of the 27 fish
species in the first 15 sites (new site 1) and of the last 14 sites (new site 2). We now
want to use our parsimonious RDA result to estimate the values of variables ele,
oxy and bod on the basis of the fish abundances of the new sites.
232
6 Canonical Ordination
par(mfrow = c(1, 2))
# Scaling 1
triplot.rda(spe.rda.pars,
site.sc = "lc",
scaling = 1,
cex.char2 = 0.8,
pos.env = 2,
mult.spe = 0.9,
mult.arrow = 0.92,
mar.percent = 0.01
)
# Scaling 2 : see accompanying material
Since there is now a third significant canonical axis, you could plot other
combinations: axes 1 and 3, and axes 2 and 3.
This triplot indeed presents the same structures as the one produced with all
explanatory variables (Fig. 6.1a). The sites and species show the same relationships.
The three selected explanatory variables are sufficient to reveal the major features of
the data.
6.3.2.7 Environmental Reconstruction: Projecting New Sites in an RDA
to Estimate the Values of Explanatory Variables
An RDA model is generally used to interpret the structure of the response data as
explained by the set of independent variables. But if the model is built from species
that act as bioindicators of environmental conditions represented by the explanatory
variables and if the model explains a fairly large amount of variance, then RDA can
be applied in an opposite way, i.e. to estimate the values of (quantitative)
explanatory variables on the basis of the abundances of the species. This is sometimes also called calibration or bioindication. It can be computed with function
calibrate() of package vegan.
As an illustration, let us imagine two new sites along the Doubs River where
fishes were captured and counted, but no environmental variable was measured. In
this example, the new sites are made of the (rounded) mean abundances of the 27 fish
species in the first 15 sites (new site 1) and of the last 14 sites (new site 2). We now
want to use our parsimonious RDA result to estimate the values of variables ele,
oxy and bod on the basis of the fish abundances of the new sites.
232
6 Canonical Ordination