Vegan: an introduction to ordination

Jari Oksanen
processed with vegan 2.5-7 in R Under development (unstable) (2020-11-27 r79520) on
November 27, 2020

Abstract
The document describes typical, simple work pathways of vegetation
ordination. Unconstrained ordination uses as examples detrended corre-
spondence analysis and non-metric multidimensional scaling, and shows
how to interpret their results by fitting environmental vectors and factors
or smooth environmental surfaces to the graph. The basic plotting com-
mand, and more advanced plotting commands for congested plots are also
discussed, as well as adding items such as ellipses, convex hulls, and other
items for classes. The constrained ordination uses constrained (canonical)
correspondence analysis as an example. It is first shown how a model is
defined, then the document discusses model building and significance tests
of the whole analysis, single constraints and axes.

Contents
1 Ordination 2
1.1 Detrended correspondence analysis . . . . . . . . . . . . . . . . . 2
1.2 Non-metric multidimensional scaling . . . . . . . . . . . . . . . . 2

2 Ordination graphics 4
2.1 Cluttered plots . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2.2 Adding items to ordination plots . . . . . . . . . . . . . . . . . . 5

3 Fitting environmental variables 6

4 Constrained ordination 7
4.1 Significance tests . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
4.2 Conditioned or partial ordination . . . . . . . . . . . . . . . . . . 11

Vegan is a package for community ecologists. This documents explains how the
commonly used ordination methods can be performed in vegan. The document
only is a very basic introduction. The current document only describes a small
part of all vegan functions. For most functions, the canonical references are
the vegan help pages, and some of the most important additional functions are
listed at this document.

1
1 Ordination
The vegan package contains all common ordination methods: Principal compo-
nent analysis (function rda, or prcomp in the base R), correspondence analysis
(cca), detrended correspondence analysis (decorana) and a wrapper for non-
metric multidimensional scaling (metaMDS). Functions rda and cca mainly are
designed for constrained ordination, and will be discussed later. In this chapter
I describe functions decorana and metaMDS.

1.1 Detrended correspondence analysis

Detrended correspondence analysis (dca) is done like this:
> library(vegan)
> data(dune)
> ord <- decorana(dune)
This saves ordination results in ord:
> ord
Call:
decorana(veg = dune)

Detrended correspondence analysis with 26 segments.

Rescaling of axes with 4 iterations.

DCA1 DCA2 DCA3 DCA4

Eigenvalues 0.5117 0.3036 0.12125 0.14267
Decorana values 0.5360 0.2869 0.08136 0.04814
Axis lengths 3.7004 3.1166 1.30055 1.47888

The display of results is very brief: only eigenvalues and used options are listed.
Actual ordination results are not shown, but you can see them with command
summary(ord), or extract the scores with command scores. The plot function
also automatically knows how to access the scores.

1.2 Non-metric multidimensional scaling

Function metaMDS is a bit special case. The actual ordination is performed by
function vegan function monoMDS (or alternatively using isoMDS of the MASS
package). Function metaMDS is a wrapper to perform non-metric multidimen-
sional scaling (nmds) like recommended in community ordination: it uses ade-
quate dissimilarity measures (function vegdist), then it runs nmds several times
with random starting configurations, compares results (function procrustes),
and stops after finding twice a similar minimum stress solution. Finally it scales
and rotates the solution, and adds species scores to the configuration as weighted
averages (function wascores):
> ord <- metaMDS(dune)
Run 0 stress 0.1192678
Run 1 stress 0.1192678
... Procrustes: rmse 3.740558e-05 max resid 0.0001132411
... Similar to previous best

2
Run 2 stress 0.1183186
... New best solution
... Procrustes: rmse 0.02027054 max resid 0.06496303
Run 3 stress 0.1192678
Run 4 stress 0.1192679
Run 5 stress 0.2192919
Run 6 stress 0.1192678
Run 7 stress 0.1183186
... Procrustes: rmse 4.798313e-07 max resid 1.503447e-06
... Similar to previous best
Run 8 stress 0.1192679
Run 9 stress 0.1183186
... New best solution
... Procrustes: rmse 5.688499e-06 max resid 1.842467e-05
... Similar to previous best
Run 10 stress 0.1183186
... Procrustes: rmse 2.091255e-06 max resid 4.85836e-06
... Similar to previous best
Run 11 stress 0.1922241
Run 12 stress 0.1183186
... New best solution
... Procrustes: rmse 1.737353e-06 max resid 4.863911e-06
... Similar to previous best
Run 13 stress 0.1183186
... Procrustes: rmse 2.319046e-05 max resid 6.636297e-05
... Similar to previous best
Run 14 stress 0.1192678
Run 15 stress 0.1809578
Run 16 stress 0.1183186
... Procrustes: rmse 3.221461e-06 max resid 1.07411e-05
... Similar to previous best
Run 17 stress 0.1812933
Run 18 stress 0.1183186
... Procrustes: rmse 2.613774e-06 max resid 6.008369e-06
... Similar to previous best
Run 19 stress 0.1192678
Run 20 stress 0.1808911
*** Solution reached
> ord
Call:
metaMDS(comm = dune)

global Multidimensional Scaling using monoMDS

Data: dune
Distance: bray

Dimensions: 2
Stress: 0.1183186
Stress type 1, weak ties
Two convergent solutions found after 20 tries
Scaling: centring, PC rotation, halfchange scaling
Species: expanded scores based on ‘dune’

3
 + +
1.5
+
1.0

+
+
+
0.5
NMDS2

+
+
+
+ +
+ +
+ +
0.0

+ +
+
++ +
+
+
−0.5

+ +
+

+
+
−1.0

−1.0 −0.5 0.0 0.5 1.0 1.5 Figure 1: Default ordination

NMDS1 plot.

2 Ordination graphics
Ordination is nothing but a way of drawing graphs, and it is best to inspect
ordinations only graphically (which also implies that they should not be taken
too seriously).
All ordination results of vegan can be displayed with a plot command (Fig.
1):
> plot(ord)

Default plot command uses either black circles for sites and red pluses for
species, or black and red text for sites and species, resp. The choices depend on
the number of items in the plot and ordination method. You can override the
default choice by setting type = "p" for points, or type = "t" for text. For a
better control of ordination graphics you can first draw an empty plot (type =
"n") and then add species and sites separately using points or text functions.
In this way you can combine points and text, and you can select colours and
character sizes freely (Fig. 2):
> plot(ord, type = "n")
> points(ord, display = "sites", cex = 0.8, pch=21, col="red", bg="yellow")
> text(ord, display = "spec", cex=0.7, col="blue")
All vegan ordination methods have a specific plot function. In addition,
vegan has an alternative plotting function ordiplot that also knows many non-
vegan ordination methods, such as prcomp, cmdscale and isoMDS. All vegan
plot functions return invisibly an ordiplot object, so that you can use ordiplot
support functions with the results (points, text, identify).
Function ordirgl (requires rgl package) provides dynamic three-dimensional
graphics that can be spun around or zoomed into with your mouse. Function or-
diplot3d (requires package scatterplot3d) displays simple three-dimensional
scatterplots.

4
 Airaprae Empenigr
1.5
Hyporadi
1.0

Salirepe

Anthodor
Comapalu
Vicilath
0.5
NMDS2

Callcusp
Planlanc
Scorautu
Bracruta Eleopalu
Trifprat Ranuflam
Achimill Trifrepe
0.0

Rumeacet Juncarti
Sagiproc
Bellpere
Lolipere
Poaprat
Agrostol
Bromhord
−0.5

Poatriv
Alopgeni
Juncbufo
Elymrepe

Cirsarve
Chenalbu
Figure 2: A more colourful ordi-
−1.0

−1.0 −0.5 0.0 0.5 1.0 1.5 nation plot where sites are points,
NMDS1 and species are text.

2.1 Cluttered plots

Ordination plots are often congested: there is a large number of sites and species,
and it may be impossible to display all clearly. In particular, two or more species
may have identical scores and are plotted over each other. Vegan does not have
(yet?) automatic tools for clean plotting in these cases, but here some methods
you can try:
• Zoom into graph setting axis limits xlim and ylim. You must typically
set both, because vegan will maintain equal aspect ratio of axes.
• Use points and add label only to some points with identify command.

• Use select argument in ordination text and points functions to only

show the specified items.
• Use ordilabel function that uses opaque background to the text: some
text labels will be covered, but the uppermost are readable.

• Use automatic orditorp function that uses text only if this can be done
without overwriting previous labels, but points in other cases.
• Use automatic ordipointlabel function that uses both points and text
labels, and tries to optimize the location of the text to avoid overwriting.

• Use interactive orditkplot function that draws both points and labels for
ordination scores, and allows you to drag labels to better positions. You
can export the results of the edited graph to encapsulated postscript, pdf,
png or jpeg files, or copy directly to encapsulated postscript, or return the
edited positions to R for further processing.

2.2 Adding items to ordination plots

Vegan has a group of functions for adding information about classification or
grouping of points onto ordination diagrams. Function ordihull adds convex

5
 1.0

NM
0.5
NMDS2

0.0

BF

HF
−0.5

SF

Figure 3: Convex hull, ellipsoid

hull, standard error ellipse and a
−0.5 0.0 0.5 1.0 spider web diagram for Manage-
NMDS1 ment levels in ordination.

hulls, ordiellipse adds ellipses enclosing all points in the group (ellipsoid
hulls) or ellipses of standard deviation, standard error or confidence areas, and
ordispider combines items to their centroid (Fig. 3):
> data(dune.env)
> attach(dune.env)
> plot(ord, disp="sites", type="n")
> ordihull(ord, Management, col=1:4, lwd=3)
> ordiellipse(ord, Management, col=1:4, kind = "ehull", lwd=3)
> ordiellipse(ord, Management, col=1:4, draw="polygon")
> ordispider(ord, Management, col=1:4, label = TRUE)
> points(ord, disp="sites", pch=21, col="red", bg="yellow", cex=1.3)

In addition, you can overlay a cluster dendrogram from hclust using ordicluster
or a minimum spanning tree from spantree with its lines function. Segmented
arrows can be added with ordiarrows, lines with ordisegments and regular
grids with ordigrid.

3 Fitting environmental variables

Vegan provides two functions for fitting environmental variables onto ordina-
tion:
• envfit fits vectors of continuous variables and centroids of levels of class
variables (defined as factor in R). The arrow shows the direction of the
(increasing) gradient, and the length of the arrow is proportional to the
correlation between the variable and the ordination.

• ordisurf (which requires package mgcv) fits smooth surfaces for continu-
ous variables onto ordination using thinplate splines with cross-validatory
selection of smoothness.
Function envfit can be called with a formula interface, and it optionally
can assess the “significance” of the variables using permutation tests:

6
> ord.fit <- envfit(ord ~ A1 + Management, data=dune.env, perm=999)
> ord.fit
***VECTORS

NMDS1 NMDS2 r2 Pr(>r)

A1 0.96474 0.26322 0.3649 0.023 *
---
Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1
Permutation: free
Number of permutations: 999

***FACTORS:

Centroids:
NMDS1 NMDS2
ManagementBF -0.4534 -0.0102
ManagementHF -0.2636 -0.1282
ManagementNM 0.2958 0.5790
ManagementSF 0.1506 -0.4670

Goodness of fit:
r2 Pr(>r)
Management 0.4134 0.005 **
---
Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1
Permutation: free
Number of permutations: 999

The result can be drawn directly or added to an ordination diagram (Fig. 4):
> plot(ord, dis="site")
> plot(ord.fit)

Function ordisurf directly adds a fitted surface onto ordination, but it

returns the result of the fitted thinplate spline gam (Fig. 4):
> ordisurf(ord, A1, add=TRUE)
Family: gaussian
Link function: identity

Formula:
y ~ s(x1, x2, k = 10, bs = "tp", fx = FALSE)

Estimated degrees of freedom:

1.59 total = 2.59

REML score: 41.58727

4 Constrained ordination
Vegan has three methods of constrained ordination: constrained or “canonical”
correspondence analysis (function cca), redundancy analysis (function rda) and
distance-based redundancy analysis (function capscale). All these functions

7
 1.0

4.5
ManagementNM
0.5

A1
NMDS2

3.5

6.5
0.0

ManagementBF
ManagementHF

6
5.5
−0.5

ManagementSF
5

Figure 4: Fitted vector and

4

smooth surface for the thickness

−0.5 0.0 0.5 1.0 of A1 horizon (A1, in cm), and
NMDS1 centroids of Management levels.

can have a conditioning term that is “partialled out”. I only demonstrate cca,
but all functions accept similar commands and can be used in the same way.
The preferred way is to use formula interface, where the left hand side
gives the community data frame and the right hand side lists the constraining
variables:
> ord <- cca(dune ~ A1 + Management, data=dune.env)
> ord
Call: cca(formula = dune ~ A1 + Management, data = dune.env)

Inertia Proportion Rank

Total 2.1153 1.0000
Constrained 0.7798 0.3686 4
Unconstrained 1.3355 0.6314 15
Inertia is scaled Chi-square

Eigenvalues for constrained axes:

CCA1 CCA2 CCA3 CCA4
0.3187 0.2372 0.1322 0.0917

Eigenvalues for unconstrained axes:

CA1 CA2 CA3 CA4 CA5 CA6 CA7 CA8 CA9 CA10
0.3620 0.2029 0.1527 0.1345 0.1110 0.0800 0.0767 0.0553 0.0444 0.0415
CA11 CA12 CA13 CA14 CA15
0.0317 0.0178 0.0116 0.0087 0.0047
The results can be plotted with (Fig. 5):
> plot(ord)
There are three groups of items: sites, species and centroids (and biplot arrows)
of environmental variables. All these can be added individually to an empty
plot, and all previously explained tricks of controlling graphics still apply.
It is not recommended to perform constrained ordination with all environ-
mental variables you happen to have: adding the number of constraints means
slacker constraint, and you finally end up with solution similar to unconstrained

8
 17

19
2

18 11 10

ManagementBF
1

EmpenigrHyporadi Vicilath 7 1
Airaprae
Salirepe 6 5
Achimill 2
Anthodor Bromhord
Planlanc
Lolipere
ScorautuManagementHF
Poaprat
CCA2

Bellpere Trifprat
Trifrepe
ManagementNM
0

Bracruta Rumeacet
Elymrepe
Juncarti Sagiproc Poatriv

Juncbufo4 3
Ranuflam 9
Callcusp Alopgeni
Cirsarve
Agrostol
Eleopalu
ManagementSF
−1

20 8

12
Chenalbu
13

Comapalu
−2

A1
14 15
16

−3 −2 −1 0 1 2 Figure 5: Default plot from con-

CCA1 strained correspondence analysis.

ordination. In that case it is better to use unconstrained ordination with envi-

ronmental fitting. However, if you really want to do so, it is possible with the
following shortcut in formula:
> cca(dune ~ ., data=dune.env)
Call: cca(formula = dune ~ A1 + Moisture + Management + Use +
Manure, data = dune.env)

Inertia Proportion Rank

Total 2.1153 1.0000
Constrained 1.5032 0.7106 12
Unconstrained 0.6121 0.2894 7
Inertia is scaled Chi-square
Some constraints were aliased because they were collinear (redundant)

Eigenvalues for constrained axes:

CCA1 CCA2 CCA3 CCA4 CCA5 CCA6 CCA7 CCA8 CCA9 CCA10
0.4671 0.3410 0.1761 0.1532 0.0953 0.0703 0.0589 0.0499 0.0318 0.0260
CCA11 CCA12
0.0228 0.0108

Eigenvalues for unconstrained axes:

CA1 CA2 CA3 CA4 CA5 CA6 CA7
0.27237 0.10876 0.08975 0.06305 0.03489 0.02529 0.01798

4.1 Significance tests

vegan provides permutation tests for the significance of constraints. The test
mimics standard analysis of variance function (anova), and the default test
analyses all constraints simultaneously:
> anova(ord)
Permutation test for cca under reduced model
Permutation: free
Number of permutations: 999

9
Model: cca(formula = dune ~ A1 + Management, data = dune.env)
Df ChiSquare F Pr(>F)
Model 4 0.77978 2.1896 0.001 ***
Residual 15 1.33549
---
Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1
The function actually used was anova.cca, but you do not need to give its name
in full, because R automatically chooses the correct anova variant for the result
of constrained ordination.
It is also possible to analyse terms separately:
> anova(ord, by="term", permutations=199)
Permutation test for cca under reduced model
Terms added sequentially (first to last)
Permutation: free
Number of permutations: 199

Model: cca(formula = dune ~ A1 + Management, data = dune.env)

Df ChiSquare F Pr(>F)
A1 1 0.22476 2.5245 0.005 **
Management 3 0.55502 2.0780 0.005 **
Residual 15 1.33549
---
Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1
This test is sequential: the terms are analysed in the order they happen to be
in the model. You can also analyse significances of marginal effects (“Type III
effects”):
> anova(ord, by="mar", permutations=199)
Permutation test for cca under reduced model
Marginal effects of terms
Permutation: free
Number of permutations: 199

Model: cca(formula = dune ~ A1 + Management, data = dune.env)

Df ChiSquare F Pr(>F)
A1 1 0.17594 1.9761 0.065 .
Management 3 0.55502 2.0780 0.005 **
Residual 15 1.33549
---
Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1
Moreover, it is possible to analyse significance of each axis:
> anova(ord, by="axis", permutations=499)
Permutation test for cca under reduced model
Forward tests for axes
Permutation: free
Number of permutations: 499

Model: cca(formula = dune ~ A1 + Management, data = dune.env)

Df ChiSquare F Pr(>F)

10
CCA1 1 0.31875 3.5801 0.020 *
CCA2 1 0.23718 2.6640 0.074 .
CCA3 1 0.13217 1.4845 0.326
CCA4 1 0.09168 1.0297 0.430
Residual 15 1.33549
---
Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1

4.2 Conditioned or partial ordination

All constrained ordination methods can have terms that are partialled out from
the analysis before constraints:
> ord <- cca(dune ~ A1 + Management + Condition(Moisture), data=dune.env)
> ord
Call: cca(formula = dune ~ A1 + Management +
Condition(Moisture), data = dune.env)

Inertia Proportion Rank

Total 2.1153 1.0000
Conditional 0.6283 0.2970 3
Constrained 0.5109 0.2415 4
Unconstrained 0.9761 0.4615 12
Inertia is scaled Chi-square

Eigenvalues for constrained axes:

CCA1 CCA2 CCA3 CCA4
0.24932 0.12090 0.08160 0.05904

Eigenvalues for unconstrained axes:

CA1 CA2 CA3 CA4 CA5 CA6 CA7 CA8 CA9
0.30637 0.13191 0.11516 0.10947 0.07724 0.07575 0.04871 0.03758 0.03106
CA10 CA11 CA12
0.02102 0.01254 0.00928

This partials out the effect of Moisture before analysing the effects of A1 and
Management. This also influences the significances of the terms:
> anova(ord, by="term", permutations=499)
Permutation test for cca under reduced model
Terms added sequentially (first to last)
Permutation: free
Number of permutations: 499

Model: cca(formula = dune ~ A1 + Management + Condition(Moisture), data = dune.env)

Df ChiSquare F Pr(>F)
A1 1 0.11543 1.4190 0.13
Management 3 0.39543 1.6205 0.02 *
Residual 12 0.97610
---
Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1

If we had a designed experiment, we may wish to restrict the permutations so

that the observations only are permuted within levels of Moisture. Restricted

11
permutation is based on the powerful permute package. Function how() can
be used to define permutation schemes. In the following, we set the levels with
plots argument:
> how <- how(nperm=499, plots = Plots(strata=dune.env$Moisture))
> anova(ord, by="term", permutations = how)
Permutation test for cca under reduced model
Terms added sequentially (first to last)
Plots: dune.env$Moisture, plot permutation: none
Permutation: free
Number of permutations: 499

Model: cca(formula = dune ~ A1 + Management + Condition(Moisture), data = dune.env)

Df ChiSquare F Pr(>F)
A1 1 0.11543 1.4190 0.266
Management 3 0.39543 1.6205 0.004 **
Residual 12 0.97610
---
Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1

12