International Association of Vegetation Science
Assembly and Response Rules: Two Goals for Predictive Community Ecology
Author(s): Paul A. Keddy
Source:
Journal of Vegetation Science,
Vol. 3, No. 2 (Apr., 1992), pp. 157-164
Published by: Blackwell Publishing
Stable URL: http://www.jstor.org/stable/3235676 .
Accessed: 18/04/2011 17:09
Your use of the JSTOR archive indicates your acceptance of JSTOR's Terms and Conditions of Use, available at .
http://www.jstor.org/page/info/about/policies/terms.jsp. JSTOR's Terms and Conditions of Use provides, in part, that unless
you have obtained prior permission, you may not download an entire issue of a journal or multiple copies of articles, and you
may use content in the JSTOR archive only for your personal, non-commercial use.
Please contact the publisher regarding any further use of this work. Publisher contact information may be obtained at .
http://www.jstor.org/action/showPublisher?publisherCode=black. .
Each copy of any part of a JSTOR transmission must contain the same copyright notice that appears on the screen or printed
page of such transmission.
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of
content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms
of scholarship. For more information about JSTOR, please contact support@jstor.org.
Blackwell Publishing and International Association of Vegetation Science are collaborating with JSTOR to
digitize, preserve and extend access to Journal of Vegetation Science.
http://www.jstor.org
Journal
of
Vegetation
Science 3:
157-164,
1992
?
IAVS;
Opulus
Press
Uppsala.
Printed in
Sweden
157
Assembly
and
response
rules:
two
goals
for
predictive
community
ecology
Keddy,
Paul
A.
Department
of
Biology,
University
of
Ottawa,
Ottawa,
Ontario,
Canada,
KIN
6N5;
Tel.
+1
613 564
3447;
Fax +1 613
564
5014
Abstract.
Assembly
rules
provide
one
possible
unifying
frame-
work
for
community
ecology.
Given
a
species
pool,
and
measured
traits for
each
species,
the
objective
is to
specify
which
traits
(and
therefore
which subset of
species)
will
occur
in a
particular
environment. Because
the
problem
primarily
involves
traits
and
environments,
answers should be
general-
izable to
systems
with
very
different
taxonomic
composition.
In this
context,
the
environment
functions like a
filter
(or
sieve)
removing
all
species
lacking
specified
combinations of
traits. In
this
way,
assembly
rules
are a
community
level
analogue
of
natural
selection.
Response
rules
follow a
similar
process
except
that
they
transform
a vector
of
species
abun-
dances to
a new
vector
using
the
same
information.
Examples
already
exist
from a
range
of
habitats,
scales,
and
kinds of
organisms.
Keywords:
Community
ecology;
Fire;
Flooding;
Island;
Natu-
ral
selection;
Prediction;
Model;
Species
pool;
Trait.
Introduction
Given the
world's
growing
environmental
problems,
the need for
accurate
predictive
models for
ecological
communities has
never been
greater.
However,
there
appears
to be
little
consensus
upon
how
they
will
be
developed.
This
lack of
agreement
has
been
described
by
Lewontin
(1974)
as
the
agony
of
community
ecol-
ogy.
Indeed,
a
major
criticism of
community
ecology
is
that
it is still a
soft
science
dealing
primarily
with
description
of
plant
and animal
associations
rather than
a hard
science
making
accurate
predictions
about
speci-
fied
state
variables. The
transition to a hard
science is
not
only important
for the
growth
of
the
discipline,
but is
essential to
guide
political
decision
making
about envi-
ronmental
issues.
Whether at the
global
scale
(e.g.
cli-
mate
change)
or
the local scale
(e.g. pollution
of a
single
lake)
the
questions
remain
the same: can
we
predict
the
future states
of
communities? The
ability
to answer
such
questions
rather than
speculate
about them
is essential.
In
the
last
decade,
it
appears,
a
major
transition
has
occurred. In
a
range
of
studies
(e.g.
van
der
Valk
1981;
Haefner
1978,
1981;
Box
1981;
Nobel
&
Slatyer
1980)
designed
for
different
purposes,
in
different
communi-
ties,
in
different
parts
of
the
world,
one
can
find
a
consistent
set
of
features
which focus
upon
accurate
ecological
predictions.
It
appears
that
when
community
ecologists
set
the
goal
of
prediction
rather
than
descrip-
tion,
that
decision
in
and of
itself leads
to
certain
recur-
ring
research
strategies
and
outcomes.
These
can
be
called
'assembly
rules' and
'response
rules'
(Keddy
1989).
My
objectives
here
are to
draw
attention to
these
developments,
illustrate the
progress
which
has
already
been
made,
formalize
the
procedure
further,
and
suggest
some
future
possible
developments
for the
discipline.
Context
Before
proceeding,
we
need
to
briefly
consider
the
context
of
the
problem,
the
scale
and
level of
organiza-
tion
at which
community
ecology
operates,
and the
state
variables
which it
explores.
It is
easy
to
confuse
the
levels of
population
and
community
ecology,
perhaps
because
they
have
similar
historical
roots
(McIntosh
1985)
and
because
there
are
population
ecologists
who
suggest
that
the
community
level
of
organization
can
only
be
studied
by
examining
the
component
population
(e.g.
Harper
1977,
1982).
However,
in
theory
and
in
practice
there
are
differ-
ences between
these
levels.
Community
ecologists
seek
to
predict
the
properties
of
aggregations
of
populations,
just
as
population
ecologists
wish to
predict
the
proper-
ties
of
aggregations
of
individuals
(Table
1).
There
are
many possible
properties
of
aggregations
of
popula-
tions;
to date
community
ecologists
have
concentrated
on
a few
such as
species
richness,
biomass,
diversity
and
life
form.
But
in
spite
of the
many possibilities,
much
of
traditional
community
ecology
is
a
description
of
a
community
in
terms of
its
component
populations
-
a
Keddy,
P. A.
species
list
for different
habitats.
In other cases abun-
dances are also
known,
so that the
description
is a vector
of
species
abundances.
These do not
begin
to exhaust
the
many
possible
state variables which
might
be used to
describe the
community
level of
organization,
but since
we have
so much information
on vectors of
species
abundances,
these serve as the
starting point
for assem-
bly
and
response
rules.
What follows
is an
explicitly 'top
down' research
strategy.
That
is,
it
begins
with
specified properties
of
the
community
level
of
organization,
and asks
about the
minimum level of
knowledge
necessary
from lower
levels of
organization
to
predict
the
community
level
properties.
Prigogene
&
Stengers
(1984)
and
Allen &
Starr
(1982;
see also Allen & Hoekstra
1990)
have
explored
the difficulties
in
relating properties
at one
level of
organization
to those
of
another,
and there
are
compelling
arguments
for
why
reductionistic research
strategies
will not work
for certain
problems
in
ecology
(Wimsatt
1982;
Rigler
1982).
None-the-less,
what fol-
lows is a reductionist
approach
in that it relates the
properties
of both
populations
and
individuals to the
community
level. At the
same
time,
however,
it
empha-
sizes that the
community
level
perspective
can
guide
the
selection of variables at these
lower levels of
organiza-
tion,
rather than the reverse.
Assembly
rules
The
objective
of
assembly
rules is to
predict
which
subset of the
total
species
pool
for a
given region
will
occur
in a
specified
habitat. It
basically
is a
problem
of
deleting
those
species
unsuited to
a
specified
set of
environmental
conditions
(Fig.
1).
A
first
objective
would
be
simply
to
predict
the
presence
or absence of
species
in
a habitat.
The second
objective
would
be to
predict
species
pool
S1
S2
S3
Sn
m
I
community
b.
environmental
--
filter
~
Si
I
S4
Sn-o_c
m
Fig.
1.
Assembly
rules
specify
which subset of
species
in the
total
pool
(left)
would tolerate
specified
conditions and
form a
community
(right).
abundance as well as
presence.
The
process
of
constructing
communities from
spe-
cies
pools
is in
many
ways analogous
to the
processes
of
evolution
through
natural selection. At the heart of our
understanding
of evolution is the
process
of natural
selection.
Habitats serve as filters for
genotypes,
with
the least suited
genotypes being
filtered
out,
and the best
suited
surviving
to
reproduce.
In
the case of
assembly
rules,
habitats are
again
serving
as
filters.
However,
in
this
case,
the filters
operate
on traits and eliminate those
sets of traits which are
unsuitable to that environment.
The
species
which
comprise
the
community
are those
which survive the filter.
An
early attempt
at
'assembly
rules' was carried
out
by
Diamond
(1975)
who used
descriptive
data
(lists
of
bird
species present
on
islands)
to
generate
rules
about
species
composition
on islands
of different size. There
are two
major
criticisms
of his
approach.
First,
the rules
were
only descriptions
of
the data rather than actual
predictions.
Second,
the assumed
mechanism was com-
petition.
The harsh
criticism of this work
(Connor
&
Simberloff
1979)
and a
long
list of
exchanges
in the
Table 1.
Comparison
of three levels of
organization
in
ecological
research.
State variables
State variable
Organizing concepts
measured
predicted
Community
traits
species
present
assembly
rules
ecology
environment guilds
present
response
rules
biomass
diversity
Population
birth rates
population
size
life
history
evolution
ecology
death rates
age
classes
population regulation
immigration
emigration
Population
gene
flow
breeding system
evolution
genetics
heterozygosity
mode of
reproduction
reproductive
allocation
inbreeding
158
-
Assembly
and
response
rules
-
literature thereafter
(e.g.
Grant &
Abbott
1980;
Dia-
mond &
Gilpin
1982;
Gilpin
& Diamond
1982;
Wright
& Biehl
1982;
Simberloff
1983,
1984)
seem to have
detracted from
recognition
that the
goal
itself was laud-
able
(Keddy
1989).
Is there another
approach
which
focuses
on the
goal
rather than the
methodology?
Assembly
rules
might
be
developed
as follows. We
begin
with a total
species
list for an area of
landscape
-
say
a
bird check list or a
plant species
list. We also
collect
systematic
data on the traits of these
species;
these traits could
include
morphological, physiological
or
ecological
features. We then
specify
a
particular
set
of environmental
conditions. Our
objective
is then to
devise a series of rules that will
predict
which subset of
those
species
will be found under the
specified
set of
environmental conditions.
The best
way
to visualize this
is as a
process
of deletion where
the
environment acts
as
a filter
removing species
which lack traits for
persisting
under a
particular
set of conditions.
Assembly
rules therefore
require
two initial data sets
for
ecological
communities:
a
species pool,
and a ma-
trix
giving
the traits of
species
in this
pool (Fig.
2).
'Assembly
rules' then
specify
which
particular
subset of
traits
(and
therefore
species possessing
them)
will be
filtered out. More
precisely,
in
the situation where we
have
knowledge
of
n
traits for each
species
in
the
pool,
we are
looking
for a
procedure
to
specify
whether or not
certain traits
(or
sets
of
them)
will
permit
a
species
to
persist
under a defined set
of environmental
conditions.
A
general
(but
obviously
over-simplified) approach
might
be
as follows: for a
specified
habitat,
we can
try
to
find a series of coefficients
using
t traits to
assign
a value
of
p
to each
species.
That
is,
for each
species
in
the
pool
atl
+
bt2
+
ct3+
dt4
+...
+
nt4
=p
species pool
S1
S2
S3
Sn
trait matrix
tll
t12
t13
t1j
t21'" .
t23
"
23
*
,
tni
tnj
_
community
S1
deletion
*
rules
*
Sn-oc
Fig.
2.
The
general
form
of
assembly
rules. Two data sets are
needed. The total
species pool,
and a
series of traits
for each
species.
Deletion rules then determine which traits
(and
there-
fore which subset of
species)
form the
community.
in
water level.
A
key
element of van der Valk's model
was the
recognition
that
only
one
major
trait was neces-
sary
to
predict regeneration:
whether or not a
species
could
germinate
under water.
By measuring only
this
one trait
on
all
species
one can
predict
which
part
of the
species pool
will occur under either set of conditions
(Fig.
3).
The
particular appeal
of this model lies
in
the
simplicity
of the trait matrix and
resulting equations.
For a flooded
wetland,
the
equation
is
simply:
at,
=p
where
tl
is %
germination
under flooded conditions.
If
p
>
0,
the
species
will be
present
in
the
vegetation.
(1)
If
p
>
Pcrit,
the
species
will be
present
in
the habitat.
If
p
<
Pcri,
it will not.
In
this
way,
a
species
list for that
habitat
might
be assembled. The exact
procedures
for
doing
this most
effectively
need further work. Two
promising examples
are the
expert systems approach
(Noble 1987)
and
ecosystem assembly
grammar
(Haefner
1978,
1981).
Analogous
methods
might
be used to sort
species
into
expected categories
of abundance. Exam-
ples
of this
already
exist,
and it
may
be easiest to
picture
the
general approach by considering
three of them.
flooded
I
'I I ~
aquatic
vegetation
I
(n
species)
I
wetland
plant
_
species pool
(s species)
I
^I
mud flat
vegetation
Ia
^
^
~(s-n
species)
I
emergent
Assembly
rules
for
wetlands
An
early
attempt
to
predict species
composition
in
wetlands
in
this
manner is found
in
van der Valk
(1981).
Species
in
prairie
wetlands must
periodically regenerate
from buried
seeds.
The
problem
was to
predict species
composition
in
these wetlands after a
specified
change
Fig.
3.
An
example
of
assembly
rules
from
vegetation
cycles
in
prairie
marshes
(van
der Valk
1981).
The
species pool
(left)
can
yield
either
aquatic vegetation
or mud flat
vegetation
depending upon
water levels for
germination.
Ability
to
ger-
minate under water is the sole trait which must be measured
to
make this
prediction.
159
Keddy,
P. A.
Species pool
trait matrix
filter
--------------------------------------.---
delete
species
unable
to
germinate
on mud
filter
------------------------
---------- ------
delete
species
where
adults lack
aerenchyma
filter
----------------------
----
.. .. .
delete
weak
competitors
|community|
Fig.
4. Several
sequential
deletion rules can be likened to
filters which
progressively
reduce the
subset
of
species
which
will form a
community.
Assembly
rules
for
birds on islands
A
second
example
is found
in
the work of Haefner
(1975,
1981)
who
developed
a series of rules for
pre-
dicting
the
species
composition
of
foliage-gleaning pas-
serine birds on small coastal islands. His
predictor
vari-
ables were measured characteristics of the islands. His
objective
was to "construct an
algorithm
such
that,
given
an
arbitrary species pool
and an
arbitrary
collec-
tion of environmental
factors,
the
output
of the
algo-
rithm is a list of
species
associated with the environ-
ment". This was
done
through
an
ecosystem assembly
grammar
(Haefner 1978).
In
this case the trait matrix
consisted of
published
habitat
requirements
for
the
spe-
cies concerned.
Based
upon knowledge
of an island's
habitat features
(e.g.
tree
size),
Haefner was able to
predict species composition
on the islands with
surpris-
ing accuracy.
A
potential
criticism of this
work is that
the trait matrix describes habitat
characteristics rather
than the traits of
organisms
themselves.
Assembly
rules
for
world
vegetation
types
A
final
example
can be
found
in
the work of Box
(1981)
who
used methods similar to Haefner to
predict
world
vegetation types. Again,
known information about
the environmental conditions tolerated
by
different
plant
growth
forms
was
used to
filter out
plant types
in
the
pool
until
only
a subset was left.
Box, however,
then
applied
a second set of rules which
essentially
ranked the
remaining
subset of
species
according
to their relative
competitive
abilities.
Filtering
out all
except
the domi-
nants left a
second,
smaller set of
plant types,
which
corresponded
well with
existing
world
vegetation types.
In
its
simplest
application, only
one rule
might
need
to
be
applied
to
assemble
a
community.
However,
as
Box's work
illustrates,
a series of
sequential
deletion
rules
may
be
necessary. Fig.
4
shows a
hypothetical
example:
three
progressive
filters which determine
spe-
cies
composition
in
a wetland
which
is
allowed to
dry
and then flood
permanently.
First,
the subset of
species
which
initially grows
is that which can
generate
on mud.
Once the site
is
reflooded,
species lacking aerenchyma
are deleted from the
foregoing
subset.
Finally,
the re-
maining species grow
and
interact,
and
only
those with
strong competitive
ability
persist.
These three filters
therefore determine final
species composition,
and three
traits
(germination
requirements,
flood tolerance and
competitive ability)
must
be measured to
predict
the
species
which will
persist.
Note that the third filter is
imposed
by competition
from other
species.
In other
systems
or
environments,
traits
conferring
resistance to
predation may
require
inclusion. Thus this
procedure
may
include
traits
conferring
resistance to either abiotic
or biotic
components
of the environment.
Response
rules
Response
rules
grow
out of
assembly
rules.
They
specify
how an initial vector of
species composition
will
respond
when an environmental factor is
changed;
[Lewontin
(1974)
has
called this 'transformation
rules'].
Examples
would
include:
how will
prairie vegetation
respond
to fire or
grazing?
How
will
bird communities
respond
to forest clearance? How will stream inverte-
brates
respond
to siltation? There are two
ways
in
which
response
rules
would
differ
in
form from
assembly
rules.
First,
one
begins
with a subset of
species already
present,
and must
predict
how these
will
respond
to the
perturbation
(deletion rules).
Second,
one must re-ex-
amine the
species pool
and trait matrix for
species likely
to
replace
those
presently occurring (Fig.
5) (addition
rules).
Again,
this could be done
by specifying
coeffi-
cients for measured traits and critical
p
values as
in
equation
1.
Qualitative
examples
can
again
be found.
Fire in
vegetation
Noble
&
Slatyer
(1980)
have described
general ap-
proaches
to
predicting
the
response
of
plant
communi-
ties to
perturbation by
fire. The two
stages
of
response
rules are
clearly
illustrated
in
their work.
First,
a fire removes certain
species
from the
vegeta-
tion. These
species
can be
predicted
from
knowledge
of
160
