You are on page 1of 20

Disturbance and Population Structure on the Shifting Mosaic Landscape

Author(s): James S. Clark


Source: Ecology, Vol. 72, No. 3 (Jun., 1991), pp. 1119-1137
Published by: Wiley
Stable URL: http://www.jstor.org/stable/1940610
Accessed: 25-06-2016 23:58 UTC

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at
http://about.jstor.org/terms

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.

Wiley is collaborating with JSTOR to digitize, preserve and extend access to Ecology

This content downloaded from 128.122.230.132 on Sat, 25 Jun 2016 23:58:28 UTC
All use subject to http://about.jstor.org/terms
Ecology, 72(3), 199 1, pp. 1119-1137
? 1991 by the Ecological Society of America

DISTURBANCE AND POPULATION STRUCTURE ON THE


SHIFTING MOSAIC LANDSCAPE'

JAMES S. CLARK
Department of Botany, University of Georgia, Athens, Georgia 30602 USA,2
The Biological Survey, New York State Museum, Albany, New York 12230 USA, and
Department of Ecology and Behavioral Biology, University of Minnesota, Minneapolis, Minnesota 55455 USA

Abstract. A stochastic model of plant population dynamics is developed and analyzed


to determine how density and age structure depend on thinning rates and disturbance
regimes. Probability distributions of age and density are derived from the distribution of
regeneration niches on a landscape and the thinning rates of cohorts on patches created by
adult mortality or larger disturbances. The theory is then extended to different types of
disturbances that operate at different scales and are interdependent, such as treefalls that
only become important as the early-successional trees that initially colonize an area affected
by a larger scale disturbance become mature.
In general, landscapes that provide frequent regeneration niches support high-density
young stands. Density distributions are negatively skewed. Decreasing frequency of regen-
eration niches results in lower mean density, higher variance, and increased (less negative)
skewness. When regeneration niches are rare, density is low, variance is low, skewness is
positive, and the age classes are highly variable. In more complex cases, regeneration niches
may depend on the time since the last large disturbance; for example, canopy gaps can
become more frequent as postfire cohorts become senescent. Then age class distributions
on a given disturbed area become increasingly platykurtic with time; skewness is negative
for early-successional species and positive for late-successional species. The exponential
distribution of age classes commonly observed in late-successional species is the asymptotic
result of a more general distribution that depends on time since the last disturbance. Across
a landscape that supports these interdependent disturbance processes, infrequent distur-
bances result in density distributions for late-successional species with large mean, negative
skewness, and low variance. An early-successional species is present at low density with
positive skewness and low variance. Density distributions for both species types are more
platykurtic and have higher variance when disturbance frequency - thinning rate. Frequent
disturbances produce higher densities of early-successional species with positive skewness
and lower densities of late-successional species. The landscape distributions of age classes
are J-shaped for both species. With frequent disturbance, age distributions are leptokurtic
for both species, but more so for late-successional species. Age class distributions are
increasingly platykurtic with less frequent disturbance. The "intermediate" disturbance
frequency that maximizes the probability of being reproductively mature at the time of
the next disturbance event (Clark 1991a) is also that which maximizes the density of
reproductive individuals on this shifting mosaic landscape.
Key words: age structure; demography; density; disturbance; patch; perennials; plant populations;
regeneration niche; shifting mosaic; stochastic process.

INTRODUCTION lem of applying to plants the population models de-


veloped for mobile organisms, such as animals or
The development of an analytical theory of plant
plankton in chemostats, has long been recognized
demography has been frustrated by problems of com-
(Schaffer and Leigh 1976, Pacala 1989). Plant recruit-
plexity related to spatial heterogeneity in population
ment, growth, and mortality respond to extremely lo-
structure and limiting resources (Schaffer and Leigh
calized resource levels rather than to landscape aver-
1976, Pacala and Silander 1985, Pacala 1988), asym-
ages. There is also an uncertainty associated with the
metric competition (large individuals control a dispro-
timing of recruitment events, which occur episodically,
portionate share of the resource base) (Harper 1977,
even in "undisturbed" assemblages (Harper 1977, Spurr
Weiner 1985, 1986), and time-dependent parameters
and Barnes 1980). Different species require different
such as birth and death rates (Clark 1991 a). The prob-
types of disturbances or "regeneration niches" (sensu
Grubb 1977), and it is likely that the pattern of these
' Manuscript received 22 December 1988; revised 15 Feb-
events has important implications for population dy-
ruary 1990; accepted 10 August 1990; final version received
14 September 1990. namics at the metapopulation scale (Comins and Noble
2 Present address. 1985, Warner and Chesson 1985, Hastings and Wolin

This content downloaded from 128.122.230.132 on Sat, 25 Jun 2016 23:58:28 UTC
All use subject to http://about.jstor.org/terms
1120 JAMES S. CLARK Ecology, Vol. 72, No. 3

1989, Pacala 1989). Local dynamics are further com- essary for analysis and understanding. Population dy-
plicated by larger scale "disturbances" that cause adult namics of perennial plants are necessarily stochastic
mortality, permit regeneration, and synchronize pop- because populations often consist of cohorts that date
ulation dynamics across landscapes (Shugart 1984, from the occurrence of a regeneration niche (Grubb
Turner et al. 1989). Thus analytical treatment of plant 1977, Harper 1977, Spurr and Barnes 1980). I focus
demography is restricted largely to annuals (reviewed on the distribution of regeneration niches, which occur
in Watkinson 1986 and Pacala 1989) or to dynamics stochastically in time and space and can be quantified
within single generations of perennials (Hara 1984, in natural assemblages, and on subsequent thinning on
1985, Clark 1990). those patches as individuals increase in size. The model
Lack of a tractable model that applies to the de- structure is similar to that contained in gap models
mography of perennial plants makes it unclear how (Botkin et al. 1972, Shugart 1984) in assuming a mosaic
population structure depends on local dynamics and of different-aged patches.
disturbance regimes. It appears likely, for example, that The products of this theory are probability distri-
the way in which disturbance probability changes with butions of patch age classes, of cohort densities, and
plant age should influence population dynamics (Pick- of plant age classes. Patch age classes are an important
ett and White 1985, Clark 1989, Cohen and Levin component of population structure, because cohorts
1991). Recent field studies have focused on determin- may pass through a predictable sequence of stages that
ing how this probability changes over time (Johnson determine many ecosystem attributes, such as leaf area
1979, Romme 1982, Johnson and Van Wagner 1985, (Waring and Schlesinger 1985), primary production
Foster 1988, Baker 1989, Clark 1989), and there has (Sprugel 1984, 1985), nutrient cycling (Vitousek and
been some speculation concerning effects on popula- Reiners 1975, Pastor and Post 1986), and gap area
tion structure (Johnson 1979, Clark 1989). Such ques- (Oliver 1981, Clark 1991 b). Plant density approaches
tions might be addressed with forest "gap" models a point equilibrium nowhere on this shifting mosaic
(Botkin et al. 1972, Shugart 1984) and the appropriate landscape, so distributions of densities are a useful
experimental design. But there exists no analyzable means for summarizing population size. Age structure
model to investigate the consequences of disturbance depends on distributions of patches and their densities;
regimes for structure at the scale of the metapopulation. without knowledge of the mosaic structure, it is im-
Specifically, how should densities and age classes be possible to represent age structure in a fashion that
distributed in the "shifting mosaic" landscape that de- permits analysis of the factors that produce it. Results
velops in forests that may rarely experience large ex- have application to questions of landscape patterns of
ogenous disturbances (Watt 1947, Bormann and Li- disturbance, succession, life history, and spatial and
kens 1979)? How does this population structure evolve temporal heterogeneity.
following larger disturbances? How do dynamics with-
THEORETICAL DEVELOPMENT
in local cohorts influence metapopulation structure?
What type of structure is to be expected in landscapes The landscape consists of a mosaic of patches, each
ranging from those that experience infrequent distur- supporting a cohort of plants that became established
bances to those characterized by frequent disturbance? a yr ago. Disturbances serve as regeneration niches
What constitutes an "intermediate disturbance re- (sensu Grubb 1977); they occur episodically, resulting
gime" (Connell 1978, Huston 1979) for a shifting mo- in adult mortality and the initiation of a new cohort
saic landscape, and what are its implications? How of small individuals at high density, with patch age a
does the structure of a population change at local and being reset to zero. The model does not depend on
regional scales en route to regional dominance or ex- assumptions regarding why disturbances occur (e.g.,
tinction of a species? The answers to these questions whether they are considered "allogenic" vs. "autogen-
represent an important step toward understanding per- ic" [Runkle 1985]). Between disturbances on any given
sistence and species coexistence, for it is the age and patch, plants increase in size, and density x(a) decreas-
density structure of a population that determines es at rate dx(a)/da. The structure of this metapopula-
whether a species can persist. Moreover, interpreting tion is summarized by distributions of (a) patch age
effects of fluctuating environments and climate change classes w(a), (b) cohort densities J(x), and (c) plant age
requires a theory that establishes the population struc- classes fj(a).
ture to be expected in the absence of these sources of Disturbances can occur at two spatial scales (Fig. 1).
variability. Type r disturbances are sufficiently small that the pop-
Here I present a stochastic model of plant population ulation that becomes established on the patch can be
dynamics to describe and analyze this mosaic structure. treated as a single cohort. Type r disturbances are nest-
The model contains somewhat more complexity than ed within a landscape that may experience a second
is typically included in analytical population models disturbance process s that affects larger areas of age t.
to accommodate the inherently stochastic nature of A type s patch contains many type r patches, each of
plant populations and the mix of spatial and temporal age a ' t (Fig. 1). An example would be a landscape
scales. I nonetheless strive to retain the simplicity nec- that experiences a gap-phase process r and fire regime

This content downloaded from 128.122.230.132 on Sat, 25 Jun 2016 23:58:28 UTC
All use subject to http://about.jstor.org/terms
June 1991 PLANT DENSITY AND AGE STRUCTURE 1121

type-r disturbance create regeneration opportunities. Because patches are


type-s disturbance out of phase, the landscape is a mosaic of different-
aged patches.
5) Disturbance can be described as a stationary re-
newal process (Clark 1989).
Assumption (1) is partially relaxed in a later section,
The shifting mosaic with large-scale disturbances, where
type r patches are nested within type s patches. As-
sumption (2), an assumption my model shares with
E0 t t 0 gap simulation models, simplifies the analysis. I discuss
t=0 t=0 how the decrease in propagule availability at low den-
sities is likely to affect model predictions. Assumption
(3) is a reasonable way to simplify local dynamics in
view of the ways in which thinning rates compensate
for initial density (Harper 1977). Thinning rate de-
pends on growth of the individual plants within the
cohort (Harper 1977, Aikman and Watkinson 1980,
Norberg 1988). Although densities of seeds and seed-
lings are highly variable, densities tend to converge to
this thinning equation from a range of initial densities
as a result of the response of thinning rate to crowding
a =0
(Harper 1977, Westoby 1984, Clark 1990, 1991 b). As-
sumption (5) is a stationarity assumption, which says
that the distribution of disturbance events does not
lype-r disturbance lypes disturbance depend on the time at which we observe the process.
This assumption is partially relaxed in the section, The
shifting mosaic with large-scale disturbances, where type
r disturbances depend on the elapsed time since the
last type s disturbance.
Temporal domain
The shifting mosaic structure (the type r process)
FIG. 1. Model of a shifting mosaic population that ex-
periences two superimposed and interdependent disturbance Metapopulation structure across this landscape de-
processes. A small-scale type r process (e.g., canopy gap for-
pends on the rate of thinning within each patch and
mation) is nested within a larger type s process (e.g., fire
on the age distribution of patches. Per capita mortality
occurrence) (above). Type s disturbances synchronize type r
patches. On any given patch, recruitment is episodic, followed rates in crowded even-aged plant cohorts tend to be
by thinning (center). The occurrence of disturbances that allow nearly constant following canopy closure and before
recruitment occurs stochastically with probability that may
increase with time since the last disturbance (below).

s. In this case, process r describes a shifting mosaic gap


process that develops with time t since the last fire s.
I begin with the simplest case by solving for the
structure of this mnetapopulation that is composed of
a mosaic of patches created by a type r (e.g., "gap")
40
disturbance process. The analysis explores how a sin-
c ~~crown area 7+ > aM203
gle-species metapopulation is influenced by local and
regional processes. I then generalize this result to in-
clude a disturbance process s that synchronizes type r 100 - - + 0
patches across broader portions of a landscape for an 0 30 60 90 120 150

early- and late-successional species. Several assump- Age (a)


tions are used to simplify the analysis:
FIG. 2. The relationship between exclusive crown area
1) The location of a given patch relative to others projection A and density x in a Picea stand. Data points are
on the landscape does not influence its dynamics. indicated by squares. Stand densities move horizontally to-
2) Dispersal is high and seed production is uninflu- ward the thinning line until the canopy closes, at which point
enced by local cohort dynamics. densities follow the exponential model, until plants approach
maximum size. Dashed lines are fitted values to the model
3) Thinning within a patch is deterministic and
from Clark (1990). The crown-area projection A is predicted
driven by plant growth. from the same parameter values using the relation 1 /x(dx/
4) Disturbances destroy existing adults, and they da) = - 1/A(dA/da) (Clark 1991 b).

This content downloaded from 128.122.230.132 on Sat, 25 Jun 2016 23:58:28 UTC
All use subject to http://about.jstor.org/terms
1122 JAMES S. CLARK Ecology, Vol. 72, No. 3

K () 100 (d)

id2 -l 50

C', 0 ~~~~~~~~~~~~~~~~~~~~(e)

(b)

_ D E[Ar]

F- 0 40 80 0 2 0

0.2

o(C) ((n x

E[Ar]

0.0

0 0.0- I
0 40 80 0 11

Age (a) Density (x(a))


FIG. 3. Relationships used to derive the distribution of densities across a shifting mosaic landscape having constant
disturbance probability that increases with age: (a) survivorship of the local cohort; (b) the rate of change in density; (c) the
probability that a randomly selected patch is of age a; (d) the inverse of (a), i.e., the age of a patch when density is x; (e)
distribution w, (a) from (c) solved in terms of density x; (f) the probability that a randomly selected patch supports den-
sity x.

plants approach maximum size (Harper 1977, Peet and the landscape occupied by age class a, termed the "re-
Christensen 1980), because of the effect of individual currence time." For a process that began in the distant
plant growth on population thinning (Clark 1990). As past, the distribution of recurrence times tends to
growth rates of plants decline, so too do density-de-
pendent mortality rates (Fig. 2). At this time when wr (a) = [S(a) (2)
density-dependent mortality is decreasing, density-in-
dependent mortality is on the rise, and canopy gaps (Cox 1962), where Sr(a) describes the survivor function
begin to appear. The extent to which density-depen- for the disturbance regime, and E[Aj is the time A,
dent vs. density-independent factors dominate in older expected to elapse between disturbances (Fig. 3c). Sr(a)
cohorts, however, will vary with species. I therefore is the probability that a patch will survive at least a yr
make the simplifying assumption that a local cohort without another disturbance and can be estimated from
thins at per capita rate Xp, disturbance history data (Johnson and Van Wagner
1985, Clark 1989).
x(a) = Ke- Pa, (1)
Distribution of cohort densities. -The distribution of
where the coefficient K has units of density and incor- cohort densities can be derived using the age distri-
porates the plant-height: crown-breadth ratio of the bution of patches together with the fact that the thin-
species (Norberg 1988, Clark 1990; Fig. 3a). ning Eq. 1 establishes a relationship between patch age
Age distribution of patches. -The age distribution of and stand density. Let the age at which a patch supports
patches is equivalent to the age distribution of cohorts, density x(a) be a(x) (Fig. 3d). Then the recurrence time
because we have thus far assumed that plants become can be solved in terms of density through the inversion
established immediately following patch formation, i.e., or[a(x)]= Xr(X) (Fig. 3e). The second ingredient needed
a disturbance. This distribution is the proportion of for a distribution of densities is the proportion of time

This content downloaded from 128.122.230.132 on Sat, 25 Jun 2016 23:58:28 UTC
All use subject to http://about.jstor.org/terms
June 1991 PLANT DENSITY AND AGE STRUCTURE 1123

a(x) during which density actually is x, given by the cv[X]


reciprocal of the thinning rate

da (X + Xp) 1 ____- r (8)


A(X) = dx(a) (3) X, L2XA + Xr (Xr + X,)2j

where X is an observed value of x. The variability of


(Appendix I). The distribution of densities across the patch densities is maximized at intermediate distur-
landscape then is the product of wr(x) and A(x): bance frequencies, as can be demonstrated from the
variance
f;(x) = r(X) zA(x) (4)
XrK2 Xr2K2
(see Fig. 3f and Appendix I).
Var[X] = 2XA + Xr (Xa + Xr )2
For example, consider a species that regenerates im-
mediately following patch formation and thins accord- Setting a Var[X]/8Xr = 0 yields
ing to Eq. 1. Further assume that patch formation (dis-
turbance) has constant probability Xr, with survivor '\* (XP +Xr* )3
function Sr(a) = e-Ara and expected interval between
Xr* =(2XP + Xr *)2 (9)
the formation of new patches E[Ar] = 1/Xr. From Eq. which contains a root Xr* where variance is maximized
2, the distribution of recurrence times,
(Fig. 5).
Distribution of cohort age classes. -The age class dis-
w,(a) = Xre-Aa, (5)
tribution contains two components, the age distribu-
tion of patches Wr(a) and the plant density on patches
is exponential. Solving for a in Eq. 1 yields
of age a,

ln In K
Lx(a)1
fp(a) A)r(a)x(a) (10)
a(x) =-
Wr(a)x(a) da

which is substituted in Eq. 5 to give


fp(a) is the probability that a randomly selected plant
is of age a. For the exponential recurrence time given
W(x) = Xrexp[-Xra(x)] = X- by Eq. 5, the representation of age class a is solved as
\K

The time spent at a given density is determined from


fp(a) = (Xr + Xp)exp[-a(Xr + Xp)], ( 11)
the derivative of Eq. 1, which is used in Eq. 3 to give which is a simple exponential distribution with param-
A(x) = [Xpx(a)]- -'. The appropriate substitutions in Eq. eter (Xr + XP) (Fig. 4a-c). This distribution has expec-
4 yield the probability density function tation E[AP] = (Xr + Xp)-I and coefficient of variation
cv(AP) = 1, where AP is the age of a plant selected
randomly from this metapopulation.
0 Xr (_) O < X < K
Local and regional effects on landscape structure. -
JPX \K/
f(x) = (6) The structure on this mosaic population, which is char-
acterized by the simplest possible stochastic distur-
10 otherwise bance process (i.e., constant probability), displays dy-
namics that will prove robust over the more complex
f(x) is the frequency of density x (Fig. 3f). That Eq. 6 treatments that follow. It is therefore worth considering
is a probability density function is shown by integrating some of these dynamics at this stage.
rK
Increasing thinning rate XP has the effect of decreasing
over (0, K), ff(x) dx = 1. This distribution is increas-
the average age (Fig. 6b) and density (Fig. 6a) of the
metapopulation and the less obvious consequence of
ing for Xr > Xp (Fig. 4a), uniform for Xr = X, (Fig. 4b),
increasing the coefficient of variation on densities (Fig.
and decreasing for Xr < Xp (Fig. 4c). The mth moment
of x is 6c). At the scale of a single cohort, lower densities are
characteristic of older stands (Fig. 3a). At the meta-
population scale, however, high thinning rates cause
E[Xm] = XrK
low densities to be associated with young stands (Fig.
mXP + Xr'
6a, b). Despite the fact that thinning rate has no effect
having expectation on the distribution of patch age classes, the average
tree is younger.
E[X] rK ( 7) Although the model assumes deterministic local dy-
XP + Xr
namics, it is worth noting that higher thinning rates do
and coefficient of variation not necessarily increase variability within individual

This content downloaded from 128.122.230.132 on Sat, 25 Jun 2016 23:58:28 UTC
All use subject to http://about.jstor.org/terms
1124 JAMES S. CLARK Ecology, Vol. 72, No. 3

(a1) (b) (c)


0.06-

> 0.03-

0.00 "-;
0 50 100 0 50 100 0 50 100

0.2- Density (x) )

20.1

0.0
Plant age (a)
0.10

4- 0.05

U-

.- 0 20 40 60 0 20 40 60 0 20 40 60
Patch age (a)

>b (d) (e) (f)


Z 0.06-
0
._

o < 0 03 <:

a- 0.00-1
0 50 100 0 50 100 0 50 100
Density (x)

0.1

0.0

Plant age (a)


0.10_

0051

0.00
0 20 40 60 0 20 40 60 0 20 40 60
Patch age (a )

10 20 50

Expected interval,E[Ar]
FIG. 4. Distributions of density fix) and cohort age classes f,(a) on a shifting mosaic landscape having six different
disturbance regimes, characterized by patch age classes wr(a). The distributions of density are given by Eq. 8. Figure shows
examples using two different shape parameter values, c,(c, = 1, above, and c, = 3, below) and three values of the expected
interval, E[Aj]: E[AjI = 10, left, E[AjI = 20, center, E[Aj] = 50, right. The thinning rate used here is XP = 0.05.

This content downloaded from 128.122.230.132 on Sat, 25 Jun 2016 23:58:28 UTC
All use subject to http://about.jstor.org/terms
June 1991 PLANT DENSITY AND AGE STRUCTURE 1125

40 pected age. Density variance is maximized at inter-

E[X] mediate disturbance frequencies (Fig. 5). In general,


higher thinning rates XAP increase variability across the
landscape, while more frequent regeneration oppor-
7 -- - - - --- oX] tunities (higher Xr) render the metapopulation more
uniform.
Changing disturbance probability. -These analytical
c 20 / results for constant disturbance probability aid under-
standing of the more complex structure that arises from
more flexible (and realistic) disturbance distributions.
2Ip These flexible distributions are analyzed here to estab-
lish what implications, if any, age-dependent distur-
bance probability holds for population structure.
Changing disturbance probability is accomplished by
0 I I a power transformation of the time scale using a Wei-
0.0 0.1 02
bull distribution Sr(a) = exp[-(Xra)cr]. The correspond-
ing distribution of patch age classes is
Disturbance frequency (2r)
1
FIG. 5. The maximum density variance at "intermediate" Wr(a) = exp[-(Xra)cri (12)
disturbance frequencies X,. Expected density E[X] increases E[Ar]
with X,, as young, dense stands account for a greater propor- where Cr is a dimensionless shape parameter, E[Ar] =
tion of the metapopulation. Coefficient of variation simul-
taneously decreases, however, as the population becomes more r( 1/c, + 1)/Xr is the expectation of the Weibull distri-
homogeneous, resulting in an intermediate maximum for bution, and r() is the gamma function. The moments
standard deviation on density a[fX. of Wr(a) are derived in Appendix II. Increasing Cr de-
scribes a probability of disturbance that increases with
time since the last disturbance. Solving for a(x) from
cohorts, because that variability depends on the degree Eq. 12 and making the appropriate substitutions in Eq.
of density compensation (Clark 1991b). If within-co- 4 results in the distribution of cohort densities
hort thinning rates were density and age independent
with rate parameter Xp, for example, then higher mor- f(W X exp t-IX nt-2
tality rate would indeed cause increased variability as
described by the coefficient of variation for this bi- 0 < X K. (13)

nomial process, exp[ 2p] /1K (1 - e-APa). If density Eq. 6 is a special case of Eq. 13, where Cr = 1. The age
distribution from Eq. 10 can be written as
compensation is high, however, thinning rates are re-
sponsive to random variability in local mortality (Fig.
2), and variance in cohort densities is constrained. The fp(a) = exp[-(a - a] (14)
binomial case does not hold, because mortality risk of f exp[-(Xra)cr- Xa] da
plants depends on whether neighboring plants have
survived. Indeed, given that XP is positively related to For Cr = 1, this expression is equivalent to Eq. 11.
plant growth rate and that higher growth rates increase Although we cannot solve for fp(a), nor for the mo-
density compensation (Clark 1991b), then high thin- ments of f(x), the structure represented by these dis-
ning rate is likely to attend decreased variability within tributions can be analyzed to determine how local dy-
cohorts, until density compensation declines late in namics and the landscape disturbance regime affect
stand development. Thus, increased thinning rates do local and metapopulation structure. The mode of den-
not necessarily increase within-patch variability, al- sity distribution Eq. 13 occurs where df(x)/dx = 0,
though they do increase among-patch variability. given by

At the metapopulation scale, the parameter of the


exponential age distribution is not equal to the mor- {KexpL-I) Cr1) IX 0 C
f ,(X Crl(cr-

tality rate, but is greater than the mortality rate in an


amount equal to the disturbance frequency (Eq. 11). X= Cr =1 Xp > Xr
Although thinning rate XP drives expected density and
age in the same direction (i.e., 8E[X]/8XP and 8E[Ap]/ l K C =1Xp < Xr
aP are both negative for all x and a, respectively) (Fig.
6), the frequency of regeneration opportunities X, drives
(15)
expected density and age in opposite directions (Fig.
4a-c). More frequent regeneration opportunities in- For X , Xr, this mode occurs near K/2. The rate at
crease the expected density while decreasing the ex- which local populations thin (XP) influences the regional

This content downloaded from 128.122.230.132 on Sat, 25 Jun 2016 23:58:28 UTC
All use subject to http://about.jstor.org/terms
1126 JAMES S. CLARK Ecology, Vol. 72, No. 3

(a) Density (x) (b) Age (a)

loo- 20-

802\ ~~~~E[X] I E[Ap]

40-

(c) (d)
2- 2-
CV[X] CV[Ap]
.n 1.5 1.5

o I /I -- .....

U 0.5- 0.5-

U 0.0 0.2 0.0 0.1 0.2

Mortality rate Ap
FIG. 6. Effect of local mortality rate on metapopulation structure. Both density (a) and expected age (b) decrease with
increasing mortality rate. The coefficient of variation on density increases (c), however, while that for age is unaffected by
mortality rate (d).

density structure according to summarized by parameters X, and cr. As in the previous


exponential case, increased disturbance frequency de-
8f(x) _ cAc, - 1ex(Ac) creases the expected patch age and variance, but the
aXP XAE[Ajx coefficient of variation on patch ages is independent of
the rate at which these patches are formed X, (Appendix
where A = (X,/Xp)ln(K/x). Increasing thinning rate de-
II). For a given disturbance rate constant (i.e., constant
creases the expected density, having a positive effect
r), increased c, has the effect of reducing the variance
on the representation of lower density cohorts and a
on patch ages, despite the fact that patches are still out
negative effect on higher density cohorts. This sensi-
of phase. Compare for example wr(a) distributions in
tivity to thinning rate is greatest at intermediate den-
Fig. 4b vs. e. This effect on patch ages results, in turn,
sities xP, where
in distributions of densities having higher modal den-
sities, higher mean, and lower variance. Increasing c,
X = K exp( cr-/cr) (16) decreases expected plant age when disturbance is fre-
quent (Fig. 4a vs. d) and vice versa (Fig. 4c vs. f).
The densities most sensitive to thinning rate XP, when Increased c, decreases variability in age structure, and
Xp is high (low), are rare and greater (less) than the it reduces kurtosis. Thus, reducing the variance on patch
modal density for the landscape. When thinning rates ages has effects that are different from either of the
are in the range of the disturbance frequency, local effects observed by changing thinning rates or distur-
population dynamics have their strongest influence on bance frequency alone.
the densities that are among those most commonly
The shifting mosaic with large-scale
observed at the metapopulation level and little effect
disturbances
on densities that are infrequent (very low or very high).
All else being equal, the landscape density of the meta- In order to generalize population dynamics to in-
population will be lower and the coefficient of variation clude large-scale type s disturbances, it is necessary to
higher for greater Xp. As with the exponential case, simplify the shifting mosaic structure and to incor-
kurtosis is lowest and skewness O when XP X r. porate the dependency of type r disturbances on time
The effects of the regional disturbance regime are t since the last type s disturbance (Fig. 1). In the last

This content downloaded from 128.122.230.132 on Sat, 25 Jun 2016 23:58:28 UTC
All use subject to http://about.jstor.org/terms
June 1991 PLANT DENSITY AND AGE STRUCTURE 1127

(a) 20-( a ) 20- ~~~~species 2

-) 10 Id
10

0-

0 20 40

Time (t)

time since fire (t)


( 0b3) 5 15 25 35 45

0 20 40 0 20 40 0 20 40 0 20 40 0 20 40

Cohort age (a)


FIG. 7. Contrasting density and age structure of early- (1) and late- (2) successional species as a function of time since the
last large (type s) disturbance. (a) Density of species 1 increases, then decreases following disturbance, while that of species
2 increases. (b) The variance on age increases for both species over time, but distributions are skewed in opposite directions.
As time since the last large disturbance becomes remote, the age distribution of species 2 approaches the limiting exponential
distribution, given by the broken line.

section, I claimed that the patch age distribution was rt

given by the survivor function divided by the expected Xk(t) = K m,(t - a)S,(a)Sk(a) da. (17)
interval in Eq. 2. In fact, Eq. 2 is a special case of a
time-dependent process By direct analogy to the type r case (Eq. 4), the distri-
bution of densities across a landscape of type s patches
w,(a, t) = m,(t - a)Sr(t), is

where mj(a) is the renewal density (Cox 1962), the f(Xk) = WS(Xk) A (Xk)A
expected number of disturbances to occur on a land-
scape a yr following the last type s disturbance. As the where WS(xk) is the recurrence density for the type s

time since the last type s disturbance becomes remote, process. The age class distribution on a single type s

m,(t - a) tends to I/E[ArI, yielding Eq. 2. The renewal patch changes with time t,

density is related to S,(a) as Mr(t -a)Sr(a)Sk(a)


fk(a, t)= , (18)
d
(t) =: - dt Si (t), m,(t -a)S,(a)Sk(a) da

where -dSi/dt lit is the probability that the ith dis- and that for a landscape of type s patches at elapsed
turbance occurs on (t, t + At) and So(t) = SQ) (e.g., time T since the type s process began is
Cox 1962). This relationship can be used together with
the assumption that type s disturbances are much larger
fTw,(t)fk.(a, t) dt
To

than type r disturbances (so much so that type r dy-


namics can be treated deterministically) to derive and gkT(a) T t (19)
analyze population structure across this larger land- fT J sw(t)fk(a, t) da dt
scape. The recruitment rate 1k(t) of species k on a type
s patch is proportional to the renewal density, I subsequently use this approach in analytical and nu-
merical examples to demonstrate effects of local (thin-
k(t) = Kmr(t)-
ning and small disturbances) and regional (large dis-
The density of species k on a single type r patch at time turbances) processes on metapopulation structure.
t following occurrence of the most recent type s dis- An analytical example. -Consider the simplest case
turbance is of constant probability of type r and type s distur-

This content downloaded from 128.122.230.132 on Sat, 25 Jun 2016 23:58:28 UTC
All use subject to http://about.jstor.org/terms
1128 JAMES S. CLARK Ecology, Vol. 72, No. 3

0.4 0.4 024

3 .2- . 0.2-

0.000_ .
0 10 20 0 10 20 0 10 20
Density (x)

0.3 0.1E 0.2

C
C
0.2-
f 0.1 - .. , 0.1
Plant age (a) 0.2- 0.2-
o 1

c 02 0.0
Plantc age 0.0
(a 0.2 0) 0.0

O 0 4 20 40 0 20 40
.0

D A .A .2>_OZn L

C 0 10 20 0 10 20 0 10 20
Density (x )
0.3 0.3- 0.3-

0.21 Plant age (a) 0.2] .2

s0.1- t 0.1- 0.1-

0.0 0.0 0.0


0 20 40 0 20 40 0 20 40
Patch age (a)
10 20 50

Expected interval, E[Ts]


FIG. 8. Dynamics of two species across a landscape where large disturbances are distributed as co(t) and where dynamics
on any given patch are as shown in Fig. 7. Distributions of densities fix), cohort age classes g(a), and patch age classes wo(t)
are shown for two values of C5 and four different expected intervals between disturbances E[TS]. X,* and X2* (upper right-
hand graph) represent the upper integration limit for density distributions, analogous to K in Eq. 6.

bances. The age distribution of type s patches is giv- with corresponding rate equation
en by

dXk = KXr - Xk (XP + Xr).


Us(t) = X~e-'~,,

and type r disturbances are occurring at constant rate Solving for time t in terms of density and substituting
mr(t) = Xr. The density of species k at time t is solved in w,(t) gives a recurrence density
from Eq. 17 as

Xk(t) = r [1 - e-t(P+Xr] WS(Xk) = \ 1 - X A(XP? X|)1T7\


X~p ? Xr

This content downloaded from 128.122.230.132 on Sat, 25 Jun 2016 23:58:28 UTC
All use subject to http://about.jstor.org/terms
June 1991 PLANT DENSITY AND AGE STRUCTURE 1129

0.4 terministic type r dynamics. It is the distribution of


densities across type s patches. The age distribution on
0.2 a single type s patch is solved from Eq. 18 as

0.0
0 10 20 fk(a, t) (= ? Xr)e-at(xp+Xr) 0 < a < t
X1* X2*
This result tends to the type r case (Eq. 11) as type s
disturbances become rare,

0.3-
0X31 lim fk(a, t) = (XA + X )ea(AP+xr)
t woo0

This limiting result also applies to cases of more com-


plex recruitment patterns, provided that recruitment
rate tends to some nonzero value with time since the
0.2-
last type s disturbance (Fig. 7). I made some analytical
headway with the landscape age distribution (Eq. 19),
0.1 -

X (a +X~e a ( +X r) jT e Xstd
0.0- gkMa =SP + Xr)ea(p- F es dt d,
0 2b 40 1 - e'~~~s j0 1 ~~ - P Xr

which tends to

0.2- I eo east dt
gk-(a) = Xs(Xp + Xr)e-a(Xp+xr) eO it- tcit
i1-it1
as T becomes large.
A numerical example.-The assumptions of con-
0 10 20
stant disturbance probability can be relaxed to accom-
modate recruitment rates and small-scale disturbances
0.3-
that depend on large (type s) disturbances. I use the
example of fire for this large disturbance, but the model
0.2- 1
could be applied to other disturbances, such as large
blowdowns. To illustrate the effect of these interacting
disturbance processes, I consider two species that differ
in the time t at which they begin to colonize a site
0.0,
0.2- following the most recent type s disturbance. These
species are an early-successional species 1 and a late-
0.1 - successional species 2. Species 1 finds regeneration
F.IG.8 oniud niches following some large catastrophic disturbance
that results in mortality of many adults, whereas spe-
* ( ) 20 406 cies 2 finds increasingly more regeneration niches with
time since this catastrophic disturbance. This second
100 species might be a gap colonizer, because canopy gaps
FIG. 8. Continued. tend to be rare until early-successional species attain
maximum size. If the recruitment rates for species k
is given by 3k(t), where k is either 1 or 2, then these
trends in recruitment rates are described by dfj(t)/dt
< 0 and df2(t)/dt - 0. To limit the number of sub-
scripts, I represent the thinning and recruitment rates
of both species by single parameters XP and K, respec-
The product of this quantity and the reciprocal of dXk/ tively.
dt (see Eq. 3) yields the probability density function In this example, I assume that the distribution of
type s patch ages can be described by a Weibull dis-
tribution with recurrence time
f(Xk) = X+ [1 Xk ( X

s(t)= E[T] exp[-(Xst)cs], (20)


O?Xk? KXr
Thi d + X,
where E[Ts] is time expected to elapse between type s
This distribution differs from Eq. 6 in assuming de- disturbances, and Ts is an observed interval between

This content downloaded from 128.122.230.132 on Sat, 25 Jun 2016 23:58:28 UTC
All use subject to http://about.jstor.org/terms
1130 JAMES S. CLARK Ecology, Vol. 72, No. 3

disturbances. The Weibull distribution has been shown recruitment of young age classes continues. When dis-
to fit disturbance history data (Johnson 1979, Baker turbances are infrequent (A, << X,) the landscape dis-
1989, Clark 1989). The density of regeneration niches tribution of age class of the late-successional species
for the early-successional species 1 decreases with time. approaches an exponential distribution with parameter
For simplicity I assume this decrease is exponential, X, (Fig. 8).
being proportional to
DISCUSSION
$,(t) = e-Pt. One of the most popular concepts of forest dynamics
Let the increase in regeneration niches for species 2 be to emerge in recent years is that of the "shifting mo-
its complement: saic" (Watt 1947, Bormann and Likens 1979, Peet
1981, Shugart 1984). The so-called old-growth forest
:2(t) e-Pt.
consists of a patchwork of small cohorts, each having
For species 1, the effects of regional disturbance re- a unique history that depends on the episodic occur-
gime, represented by XA and cs, are qualitatively similar rence of disturbances. Processes exogeneous to the stand
at this scale to those of X, and c, at the finer scale are not required to introduce heterogeneity, because
(compare density distributions in Fig. 4 with those for variability is produced by the plants themselves. On
species 1 in Fig. 8). For species 2, the effects are op- any given piece of ground the dynamics consist of re-
posite. Increasing intervals between large disturbances cruitment phases followed by periods of thinning. These
have the effect of shifting the mode of the early-suc- local dynamics are produced by simulation models of
cessional species 1 to lower densities and the late-suc- small patches that clearly show this two-phase process
cessional species 2 toward high density (Fig. 8). As the (e.g., Shugart 1984:Fig. 5.4). Only when the output
density of the early-successional species decreases, the from a large number of patches is averaged do recruit-
distribution becomes more platykurtic and subse- ment and mortality appear somewhat constant.
quently more leptokurtic, as it becomes extinct in most In this paper, I derived the structure of a "shifting
stands. The density distribution of the late-successional mosaic" metapopulation, and I examined the effects
species also becomes initially more platykurtic, as the of local population dynamics and of regional distur-
mode shifts toward higher densities, and it then be- bance regimes on that structure. The local dynamics
comes more leptokurtic and negatively skewed. Both are represented by changing age and density, which
species exhibit the maximum variation in densities at depend on growth rates of individual plants within
intermediate times since the last large disturbance. small cohorts. The regional disturbance regime is de-
Age distributions are J-shaped for both species (Fig. scribed by a distribution of events that result in adult
8). These landscape age class distributions are a com- mortality and the initiation of new cohorts. The struc-
posite of the age class distributions (Eq. 18) of the ture of this metapopulation is summarized by distri-
individual sites (Fig. 7b), with the relative contribution butions of cohort (patch) age classes, plant age classes,
of the distributions at different times since fire deter- and cohort densities. This highly idealized class of
mined by the recurrence time of the fire regime. The models serves to isolate several important relation-
age distributions of both species have low expectation ships.
when disturbances are frequent, because all patches are
of young age, and thus cohorts are necessarily young. Effects of local dynamics
Species 1 always has a component of older individuals, The link between growth of plants and cohort mor-
because it begins to reproduce sooner after the large tality rates (Clark 1991a) can be used together with
disturbance. For the early-successional species, results results presented here to demonstrate implications of
demonstrate the widely observed J-shaped distribu- local dynamics for metapopulation structure. The fast-
tions that result from landscape surveys actually com- er plants grow, the more rapidly they thin. This effect
posed of local distributions having modes >0 (Hough of growth rate on mortality is obvious in all forestry
1932, West et al. 1981; Fig. 7b). Age class distributions yield tables (reviewed in Clark 1991 b), and it is com-
g, (a) for this species approximate the age distributions monly observed in populations of herbaceous plants
w,(t) of the patches themselves to an extent determined (reviewed in Harper 1977). This negative effect of
by the degree to which recruitment can continue with growth rate on density has effects at the metapopula-
time since the last disturbance. The age distribution of tion level, influencing not only expected values, but
species 1 becomes increasingly uniform where distur- also population heterogeneity. The directions and mag-
bance is infrequent, as total density declines. This is nitudes of these growth-rate effects depend on the scale
because young patches account for a smaller proportion at which the population is observed.
of the total landscape. These results agree with the Growth rates may have opposing effects on within-
regional syntheses of West et al. (1981), where com- vs. among-cohort variability. High growth rate and
posite distributions of older stands were increasingly attendant high thinning rate results in low expected age
platykurtic and positively skewed. and density of this shifting mosaic metapopulation (Fig.
Species 2 maintains the J-shaped character because 6a, b). At the same time, the coefficient of variation

This content downloaded from 128.122.230.132 on Sat, 25 Jun 2016 23:58:28 UTC
All use subject to http://about.jstor.org/terms
June 1991 PLANT DENSITY AND AGE STRUCTURE 1131

on plant age is unchanged (Fig. 6d), but that of density that will not reach reproductive maturity before the
is increased (Fig. 6c). The faster plants grow, however, next disturbance occurs (see below, Intermediate dis-
the greater may be the adjustment of density-depen- turbance and the link to life history theory). The simple
dent mortality rates to stochastic variability in mor- distributions presented here likely overrepresent ju-
tality rates (Clark 1991 b). It is therefore possible that venile classes for species that are rare. A partial anti-
higher growth rates result in more predictable thinning dote to the high variability of juvenile classes would
curves. The implication is that increased growth rate be to ignore them entirely, as is done in gap simulation
can increase local homogeneity while at the same time models, and simply focus on structure of the remaining
rendering the metapopulation more heterogeneous. population.
Where large-scale disturbances episodically synchro-
nize patch dynamics across large areas, higher thinning
Increasing disturbance probability with age
rates imply lower density and lower age for early-suc-
cessional and late-successional species (Fig. 8). Results presented here support earlier speculation
(e.g., Johnson 1979) that the shape of the disturbance
distribution (represented by parameter c) has impor-
The effect of disturbance regime
tant effects on population structure. The simplest case
The general patterns in age structure across a gra- of constant disturbance probability has been advocated
dient from high to low regeneration-niche frequency on the basis of some empirical studies (Van Wagner
range from high-density young stands to a mix of den- 1978) and is assumed in most stochastic population
sities and age classes to low-density stands dominated models.
by older individuals. Constant disturbance probability It appears in forests, however, that the probability
results in an exponential age structure with parameter both of gaps (Kohyama 1987, Suzuki and Tsukahara
(Xr + X,) (Fig. 4a-c). Disturbance frequency "substi- 1987, Foster 1988) and of fire (Heinselman 1973, John-
tutes" for mortality rate in the age distribution (Eq. son 1979, Romme 1982, Baker 1989, Clark 1989) often
1 1), and more frequent disturbance has an effect similar increases with time since the last such event. This in-
to higher mortality rate. This represents an important creasing probability is described here by parameter Cr
result, because exponential age distributions appear to or cs > 1. As a result of the more precipitous rise in
commonly arise in real forests (e.g., Hough 1932, Leak disturbance probability that occurs with large Cr, dis-
1965, 1975, Van Wagner 1978), and the parameter of turbances tend to occur with higher probabilities at a
the distribution is shown here to be the sum of two certain age, and the variance on disturbance intervals
potentially very different sources of mortality, the declines. For example, stands of some early-succes-
"thinning rate," which is largely a consequence of com- sional tree species tend to fall apart at senescence, with
petition, and the "disturbance frequency," which can gap area increasing rather suddenly as growth rates of
contain a component that is exogeneous to the local trees decline. The population effect of increased Cr de-
stand dynamics and which may affect much larger ar- pends on the relationship between thinning rate (XP)
eas. Moreover, while one component (Xr) of this pa- and expected disturbance interval (E[ArI) and on the
rameter tends to be associated with recruitment, the scale at which the population is observed. In fact, the
other (X,) is not. influence of time-dependent disturbance probability
For density distributions, however, more frequent on age structure changes qualitatively from species
disturbance has effects opposite of those produced by having low thinning rate (XP << 1 /E[ArI) to those having
increased thinning rate; as disturbance becomes more high thinning rate (XP > 1/E[ArI). A clustering of high
frequent, density increases, and the metapopulation disturbance probabilities at particular ages (i.e., Cr >
becomes more homogeneous (Fig. 4). Infrequent re- 1) within a shifting mosaic metapopulation has the
generation niches (e.g., Fig. 4c) produce distributions effects of decreasing the expected age and increasing
of densities with a mode near zero, positive skewness, the expected density for populations characterized by
and low variance. Many plants become old and se- low thinning rates relative to the expected disturbance
nescent before the next regeneration niche becomes interval (compare Fig. 4a and d). This response occurs
available. Where mortality and regeneration-niche time despite the fact that the expected disturbance interval
scales are roughly equivalent (Fig. 4b), the mode occurs is held constant. However, the coefficient of variation
at intermediate densities, skewness may approach zero, decreases for both age and density.
and variance is high. These stands consist of both young At high thinning rates (XP > 1/E[ArI, e.g., 50-yr in-
and old cohorts at high and low densities, respectively. tervals in Fig. 4), increasing parameter Cr increases both
Where regeneration niches occur frequently (Fig. 4a), the expected age and density (compare Fig. 4c and f),
densities are high, skewness is negative, and variance but the coefficients of disturbance intervals within
is again low. Where the agent responsible for the re- patches (increasing cr) results in decreased variance on
generation niche also destroys mature individuals, as patch age classes, which in turn decreases the coefficient
is often the case, these frequent disturbances result in of variation on plant densities and on plant age classes
a transient phase of young individuals at high density of long- and short-lived species. Despite the fact that

This content downloaded from 128.122.230.132 on Sat, 25 Jun 2016 23:58:28 UTC
All use subject to http://about.jstor.org/terms
1132 JAMES S. CLARK Ecology, Vol. 72, No. 3

patches are out of phase with one another, the increas- observation of exponential age-class distributions in
ing predictability of patch age produces younger and "late-successional" species (Leak 1965, 1975) repre-
denser metapopulations of populations having low sents the limiting distribution of a more general rela-
thinning rates, older and denser metapopulations of tionship that accommodates recruitment and mortality
populations having high thinning rates, and more ho- from both sources. Both species display maximum
mogeneous metapopulations regardless of thinning rate. variance on stand densities at intermediate disturbance
frequencies (Fig. 8, see next section).
Superimposed and interdependent The effects of variance in the intervals on large-scale
disturbance processes
disturbances (compare c, = 1 vs. c, = 3 in Fig. 8) also
On many landscapes, several types of disturbances depend on thinning rate, in addition to whether a spe-
that provide regeneration niches for different types of cies is early vs. late successional. If such disturbances
species may operate at vastly different scales. An ob- are rare, this variance has little effect on population
vious example is fire, which may occur over a spatial structure (e.g., E[T5] = 100, c, = 1 vs. c, = 3 in Fig. 8).
scale that is large, and treefalls, which occur on the For long-lived species, or when such disturbances are
same piece of ground, albeit at different times (e.g., more frequent (e.g., E[Tj] = 10 in Fig. 8), however,
Spies and Franklin 1989). Moreover, the occurrence decreasing the variance on expected disturbance inter-
of treefalls themselves depends on occurrence of fire, val decreases the expected age and density of late-suc-
because only when trees that regenerate following fire cessional species, and it increases the density and de-
become large will treefall produce the canopy gaps re- creases the age of early-successional species. The
quired for regeneration by a gap species (Peet and coefficient of variation on densities is decreased, while
Christensen 1980, Oliver 1981, Peet 1981). Early-suc- the variance is maximized at intermediate frequencies
cessional species, such as Betula papyrifera, Pinus for both species.
banksiana, P. resinosa, P. contorta, and Populus spp., It is therefore important to consider the time-de-
may colonize soon after such a disturbance at high pendence of disturbance probability for disturbances
density and subsequently thin. Other species, such as that occur frequently, and it is of diminishing impor-
Acer saccharum, Tilia americana, and Fagus grandi- tance for disturbances that are rare relative to the time
folia, germinate beneath a closed canopy and maintain scale of the thinning process. Examples of disturbances
low metabolic rates until a gap in the canopy results that occur with sufficient frequency that this time-de-
in increased resources that allow "recruitment" to the pendence is important include gap processes and fire
canopy. Different sizes and types of gaps may favor in some temperate forests. Thus, the relative time scales
different species (Connell 1979, Runkle 1982, Brokaw for cohort thinning and disturbance frequency, and the
1985, Canham 1988). Although this description is temporal correlation in the mosaic landscape both have
oversimplified, it captures a general pattern that has important consequences for population structure. The
been observed so consistently that it serves as a basis effects of both of these considerations are complex and,
for prevailing concepts of "succession," "regeneration in some cases, unexpected. An understanding of the
niche," and "gap-phase replacement." representation of density and age classes and of meta-
On a large disturbed area where occurrence of re- population heterogeneity requires knowledge of the time
generation niches depends on time since this large dis- development of cohort densities and changing distur-
turbance, the age distributions of early- and late-suc- bance probability at several temporal and spatial scales.
cessional species are skewed in opposite directions (Fig.
Intermediate disturbance and the link to
7b). Across a landscape of such large disturbances,
however, both distributions are positively skewed, with
life history theory

the early-successional species having a longer "tail" In a companion paper (Clark 1991 a) I showed that
(Fig. 8). The structure of an early-successional popu- the probability that a plant will be reproductively ma-
lation responds to an increase in disturbance frequency ture at the time of the next recruitment opportunity s
in a fashion qualitatively similar to a gap species on a is maximized at "intermediate" disturbance frequen-
simple shifting mosaic (Fig. 4): density increases and cies. This intermediate frequency is one having expec-
age decreases. Increasing time between large distur- tation that is greater than the maturation time a, and
bances results in a metapopulation containing few old less than the longevity a2, which is correlated with
and low-density stands of early-successional species. maturation time (Loehle 1988), i.e., a2 = aao . Because
The age class distribution approaches that of the dis- many tree species that occur in temperate forests share
turbance regime itself, with the regional population similar life histories, this intermediate disturbance fre-
exhibiting high spatial variability (compare g(a) and quency is also expected to maximize species diversity.
W,(t) distributions in Fig. 8). On any given disturbed Given the focus on population structure taken here (as
area, a late-successional species eventually assumes a opposed to life history optimization in Clark 1991 a),
limiting exponential distribution with the parameter it is reasonable to ask whether such an "intermediate"
equal to the thinning rate plus the frequency with which disturbance frequency also exists that maximizes re-
canopy gaps occur (Eq. 11; Fig. 7b). Thus, the empirical productive potential from the standpoint of population

This content downloaded from 128.122.230.132 on Sat, 25 Jun 2016 23:58:28 UTC
All use subject to http://about.jstor.org/terms
June 1991 PLANT DENSITY AND AGE STRUCTURE 1133

1.0- density of reproductive individuals suggests (1) an in-


ternal consistency of the theory of population dynamics
E[x]
on a mosaic landscape (i.e., the individual perspective
0.8-
does not yield results different from a population per-
spective), (2) a close link between optimal life history
0.6 reproductive fraction from an individual perspective and the mosaic struc-
0 06 ture of a metapopulation, and (3) a specific definition
for the "intermediate disturbance hypothesis" as one
0.4- that maximizes both the reproductive potential of in-
dividual plants and the reproductive portion of the
metapopulation. As disturbance becomes less frequent,
0.2- Xr
the reproductive fraction of the population increases,
but more local populations go extinct (Fig. 9); the pop-
0.0- ulation consists of increasingly larger and older indi-
0 40 80 120 160 200 viduals within lower density stands. At the other end
of the spectrum are populations nearing extinction be-
Expected interval
cause mortality agents occur too frequently. These
FIG. 9. Comparison of predictions for the optimal life stands consist of individuals that are too young to re-
history from Clark (199la) with that for maximum density
produce, and they occur at high density only because
of reproductive individuals from theory presented here. The
reproductive fraction xr (solid curve) is the product of ex- they are young and thus small. Thus, the high-density
pected density E[x] and the fraction of the population that is stage that occurs when disturbance is frequent is a tran-
of reproductive age (broken curves). This product is equal to sient phenomenon, because the seed source rapidly di-
the optimal maturation time r from Clark (1991a), when
minishes with this loss of reproductive individuals from
density is measured in units of K, i.e., r = X,/K. The maturation
the population.
time is a, = 30 and thinning rate is XP = 0.01.
The maximum among-cohort variability in density
classes also occurs when disturbance frequency is in

structure. Instead of maximizing the reproductive the range of thinning rate (Figs. 4, 5, 8). This conclusion

probability ? discussed in Clark (1991 a), one candidate applies to the simple shifting mosaic (Fig. 4), and to
both early- and late-successional species on the more
for maximization here is the density of reproductive
individuals Xr, which includes all individuals of the age complex landscape that experiences large-scale distur-

interval (a, iaa). This density can be derived as the bances, despite the fact that disturbance frequency is

product of expected density E[X] and the proportion driving the density distribution of early- and late-suc-
cessional species in opposite directions (Fig. 8). Thus,
of the population that is of reproductive age,
at a range of spatial scales, the approximate equiva-
raa2
lence of these time scales maximizes variability, and
xr = E[x] J fp(a) da. it coincides with the relationship where persistence is
predicted to be most likely. The turnover of species
For the exponential case, expected density is given by along disturbance gradients (e.g., Huston 1979, Tilman
Eq. 7. The fraction of the population that is reproduc- 1988) derive at least in part from this relationship be-
tively mature is derived form Eq. 11 as tween time scales. These relationships are important
Zaa2
in view of the frequent regional changes in forest tree
composition that have occurred in temperate zones just
J fp(a) da = exp[-(Xr + Xp)a1]
over the last 10 000 yr (Wright 1974, Davis 1981, Pay-
ette and Filion 1985, Jacobson et al. 1987).
-exp[-(Xr + Xp)aal]. (24)
The density of reproductive individuals is thus Population dynamics as a mosaic

There is often a concern in population theory that

pXrK {exp[-(Xr + Xp)a1] - exp[-(Xr + Xp)aa1j}. deterministic results might be modified in some qual-
Xp + Xr
itative way upon the introduction of random or time-
This result is proportional to s for the same disturbance dependent variability in demographic parameters such
regime, given by Eq. 10 in Clark (199 la), i.e., as birth or death rates. A number of studies show that
this type of variability may produce unexpected results
Xr = K?,
in some models (Abrams 1984, Chesson 1984), while
and so is maximized at the same disturbance frequenc, in others the main effect of limited stochasticity is to
(Fig. 9). simply replace a fixed-point equilibrium with a prob-
The fact that the disturbance regime that maximize ability distribution having expectation in the neigh-
reproductive probability is also that which producer borhood of that point equilibrium (May 1973, Turrelli
the age and density structure containing the maximum 1981). For many applications, it is reasonable to ignore

This content downloaded from 128.122.230.132 on Sat, 25 Jun 2016 23:58:28 UTC
All use subject to http://about.jstor.org/terms
1134 JAMES S. CLARK Ecology, Vol. 72, No. 3

this random component of birth and death rates, be- erate the distribution of regeneration niches from the
cause the effect of variation appears obvious or because assemblage itself, with the occurrence of each type of
variation in the system under consideration is deemed regeneration niche influenced by the size and density
insignificant (it may be small in magnitude, or depen- of each species within the assemblage. I view this ap-
dencies may be weak). proach as less useful at the level of analytical models
The type of stochasticity treated here cannot so easily because of its complexity, the lack of information on
be ignored as it represents the basic demography that direct and higher order effects of species combinations
characterizes many populations of perennial plants. This on the disturbance regime, and the problem of under-
variation is that which results from the uncertainty standing how the assemblage interacts with exogenous
associated with the timing of recruitment events. Even factors to affect the distributions of disturbances such
if all environmental variables could be held constant, as blowdowns, fires, and landslides. Moreover, the gap
a survey of stem densities and age classes across any process in temperate forests may be relatively insen-
landscape would generate a frequency distribution, be- sitive to species composition (see Runkle 1985, Clark
cause of the discrete nature of individual plants, be- 1991 a). This approach is also more applicable to real
cause plants are constantly increasing in size, and be- situations, because the distributions of disturbances
cause increasing individual size for a population of and plant survivorship can be quantified in forests
sessile organisms necessitates decreasing density with without understanding all of the dependencies of that
time. As plants approach maximum size, growth slows disturbance occurrence on the physical environment
and density is reduced to a point where canopy gaps and biotic factors. These simple analytical results pro-
begin to appear. It is at this point that recruitment rate vide a perspective for complementary numerical mod-
increases. Variability is produced by the population els that assume a similar structure but contain more
itself The result is a metapopulation that is a collection complexity.
of cohorts (Watt 1947, Bormann and Likens 1979),
each possessing a unique history and thus structure.
ACKNOWLEDGMENTS
In order to address this process, I have treated the
For their helpful discussions and/or reviews of the paper I
metapopulation as a mosaic of local populations. Plant
thank P. Abrams, H. Caswell, P. Chesson, P. Munholland, S.
densities, competitors, and resources do not approach Pickett, D. Royall, D. Tilman, H. E. Wright, and three anon-
a constant density anywhere on a landscape, recruit- ymous reviewers. This research was supported by a fellowship
ment is confined to short periods, and a long history from the Graduate School of the University of Minnesota,
and NSF grants BSR-8715251 and BSR-8818355. Limno-
of observation implicates the importance of the dis-
logical Research Center contribution number 377 and New
tribution of regeneration niches on landscapes (e.g., York State Education contribution number 608.
Grubb 1977). Taller plants preempt more light and
larger root systems may permit control of a dispro- LITERATURE CITED
portionate amount of belowground nutrients. The es- Abrams, P. A. 1984. Variability in resource consumption
tablishment of a seedling is most strongly dependent rates and the coexistence of competing species. Theoretical
on the occurrence of "safe sites" (Harper 1977) or "re- Population Biology 25:106-124.
Aikman, D. P., and A. R. Watkinson. 1980. A model for
generation niches" (e.g., Watt 1947, Grubb 1977). "The
growth and self-thinning in even-aged monocultures of
vast majority of tree species are dependent either on
plants. Annals of Botany 45:419-427.
fire, flooding, or windthrow to provide suitable seed- Baker, W. L. 1989. Effect of scale and spatial heterogeneity
beds for their establishment" (Spurr and Barnes 1980: on fire-interval distributions. Canadian Journal of Forest
71). Different species utilize different regeneration Research 19:700-706.
Bormann, F. H., and G. E. Likens. 1979. Pattern and process
niches, and it is the availability of regeneration niches
in a forested ecosystem. Springer-Verlag, New York, New
that poses an important limiting factor on a popula- York, USA.
tion's success. Botkin, D. F., J. F. Janak, and J. R. Wallis. 1972. Some
Competition influences the probability of surviving ecological consequences of a computer model of forest
growth. Journal of Ecology 60:849-872.
until that next recruitment opportunity occurs. A strong
Brokaw, N. V. 1985. Treefalls, regrowth, and community
competitor can insure local extinction of a weaker com-
structure in tropical forests. Pages 53-69 in S. T. A. Pickett
petitor for the same regeneration niche by increasing and P. S. White, editors. The ecology of natural disturbance
that competitor's mortality rate to an extent where it and patch dynamics. Academic Press, New York, New York,
does not survive to reproductive maturity or until the USA.
Canham, C. D. 1988. Growth and canopy architecture of
next recruitment opportunity. Thus, theory presented
shade-tolerant trees: response to canopy gaps. Ecology 69:
here places competition within the context of survi- 789-795.
vorship of established cohorts within patches of finite Chesson, P. L. 1984. The storage effect in stochastic pop-
area. ulation models. Lecture Notes in Biomathematics 54:76-
89.
Finally, the occurrence of these opportunities for re-
Clark, J. S. 1989. Ecological disturbance as a renewal pro-
cruitment (gaps) often depends on the dynamics of the
cess: theory and application to fire history. Oikos 56:17-
overstory, so it might also be viewed as one component 30.
of competition. A more complete model might gen- 1990. Integration of ecological levels: individual

This content downloaded from 128.122.230.132 on Sat, 25 Jun 2016 23:58:28 UTC
All use subject to http://about.jstor.org/terms
June 1991 PLANT DENSITY AND AGE STRUCTURE 1135

plant growth, population mortality, and ecosystem process. Loehle, C. 1988. Tree life history strategies: the role of de-
Journal of Ecology 78:275-299. fenses. Canadian Journal of Forest Research 18:209-222.
1991 a. Disturbance and tree life history on the shift- May, R. M. 1973. Stability and complexity in model eco-
ing mosaic landscape. Ecology 72:1102-1118. systems. Princeton University Press, Princeton, New Jer-
* 1991 b. Relationships between individual plant sey, USA.
growth and the dynamics of populations and ecosystems. Norberg, A. 1988. Theory of growth geometry of plants and
In D. DeAngelis and L. Gross, editors. Populations, com- self-thinning of plant populations: geometric similarity,
munities, and ecosystems: an individual perspective. Rout- elastic similarity, and different growth modes of plant parts.
ledge, Chapman, and Hall, New York, New York, USA, in American Naturalist 131:220-256.
press. Oliver, C. D. 1981. Forest development in North America
Cohen, D., and S. A. Levin. 1991. Dispersal in patchy en- following major disturbance. Forest Ecology and Manage-
vironments: the effects of temporal and spatial structure. ment 3:153-168.
Theoretical Population Biology, in press. Pacala, S. W. 1988. Competitive equivalence: the coevo-
Comins, H. N., and I. R. Noble. 1985. Dispersal, variability, lutionary consequences of sedentary habit. American Nat-
and transient niches: species coexistence in a uniformly uralist 132:576-593.
variable environment. American Naturalist 126:706-723. 1989. Plant population dynamic theory. Pages 54-
Connell, J. H. 1978. Diversity in tropical rain forests and 67 in J. Roughgarden, R. M. May, and S. A. Levin, editors.
coral reefs. Science 199:1302-1310. Perspectives in ecological theory. Princeton University Press,
. 1979. Tropical rain forests and coral reefs as open Princeton, New Jersey, USA.
non-equilibrium systems. Pages 141-163 in R. M. Ander- Pacala, S. W., and J. A. Silander. 1985. Neighborhood mod-
son, B. 0. Turner, and L. R. Turner, editors. Population els of plant population dynamics. I. Single-species models
dynamics. Blackwell, Oxford, England. of annuals. American Naturalist 125:385-411.
Cox, D. R. 1962. Renewal theory. Methuen, London, En- Pastor, J., and W. M. Post. 1986. Influence of climate, soil
gland. moisture, and succession on forest carbon and nitrogen
Davis, M. B. 1981. Quaternary history and the stability of cycles. Biogeochemistry 2:3-27.
forest communities. Pages 132-153 in D. C. West, H. H. Payette, S., and L. Filion. 1985. White spruce expansion at
Shugart, and D. B. Botkin, editors. Forest succession: con- the tree line and recent climatic change. Canadian Journal
cepts and application. Springer-Verlag, New York, New of Forest Research 15:241-251.
York, USA. Peet, R. K. 1981. Changes in biomass and production during
Foster, D. R. 1988. Species and stand response to cata- secondary succession. Pages 324-338 in D. C. West, H. H.
strophic wind in central New England, U.S.A. Journal of Shugart, and D. B. Botkin, editors. Forest succession: con-
Ecology 76:135-151. cepts and application. Springer-Verlag, New York, New
Grubb, P. J. 1977. The maintenance of species richness in York, USA.
plant communities. The importance of the regeneration Peet, R. K., and N. L. Christensen. 1980. Succession: a
niche. Biological Review 52:107-145. population process. Vegetatio 43:131-140.
Hara, T. 1984. Dynamics of stand structure in plant mono- Pickett, S. T. A., and P. S. White. 1985. Patch dynamics: a
cultures. Journal of Theoretical Biology 110:223-239. synthesis. Pages 371-384 in S. T. A. Pickett and P. S. White,
1985. A model for mortality in a self-thinning plant editors. The ecology of natural disturbance and patch dy-
population. Annals of Botany 55:667-674. namics. Academic Press, Orlando, Florida, USA.
Harper, J. L. 1977. Population biology of plants. Academic Romme, W. H. 1982. Fire and landscape diversity in sub-
Press, London, England. alpine forests of Yellowstone National Park. Ecological
Hastings, A., and C. L. Wolin. 1989. Within-patch dynamics Monographs 52:199-221.
in a metapopulation. Ecology 70:1261-1266. Runkle, J. R. 1982. Patterns of disturbance in some old-
Heinselman, M. L. 1973. Fire in the virgin forests of the growth mesic forests of eastern North America. Ecology 63:
Boundary Waters Canoe Area, Minnesota, Quaternary Re- 1533-1546.
search 3:329-382. . 1985. Disturbance regimes in temperate forests.
Hough, A. F. 1932. Some diameter distributions in forest Pages 17-34 in S. T. A. Pickett and P. S. White, editors.
stands of northwestern Pennsylvania. Journal of Forestry The ecology of natural disturbance and patch dynamics.
30:933-943. Academic Press, Orlando, Florida, USA.
Huston, M. A. 1979. A general hypothesis of species di-
Schaffer, W. M., and E. G. Leigh. 1976. The prospective
versity. American Naturalist 113:81-101.
role of mathematical theory in plant ecology. Systematic
Jacobson, G. L., T. Webb, and E. C. Grimm. 1987. Patterns
Botany 1:209-232.
and rates of vegetation change during the deglaciation of
Shugart, H. H. A. 1984. Theory of forest dynamics: the
eastern North America. Pages 277-288 in W. F. Ruddiman
ecological implications of forest succession models. Spring-
and H. E. Wright, editors. North America and adjacent
er-Verlag, New York, New York, USA.
oceans during the last deglaciation. The geology of North
America, K-3. Geological Society of America, Boulder,
Spies, T. A., and J. F. Franklin. 1989. Gap characteristics
and vegetation response in coniferous forests of the Pacific
Colorado, USA.
Northwest. Ecology 70:543-545.
Johnson, E. A. 1979. Fire recurrence in the subarctic and
its implications for vegetation composition. Canadian Jour- Sprugel, D. G. 1984. Density, biomass, productivity, and
nal of Botany 57:1374-1379. nutrient-cycling changes during stand development in wave-
Johnson, E. A., and C. E. Van Wagner. 1985. The theory regenerated balsam fir forests. Ecological Monographs 54:
165-186.
and, use of two fire history models. Canadian Journal of
Forest Research 15:214-220. 1985. Natural disturbance and ecosystem energet-
Kohyama, T. 1987. Stand dynamics in a primary warm- ics. Pages 335-352 in S. T. A. Pickett and P. S. White,
temperate rain forest analyzed by the diffusion equation. editors. The ecology of natural disturbance and patch dy-
Botanical Magazine (Tokyo) 100:305-317. namics. Academic Press, New York, New York, USA.
Leak, W. B. 1965. The J-shaped probability distribution. Spurr, S. H., and B. V. Barnes. 1980. Forest ecology. Wiley,
Forest Science 11:405-409. New York, New York, USA.
. 1975. Age distribution in virgin red spruce and Suzuki, E., and Tsukahara, J. 1987. Age structure and re-
northern hardwoods. Ecology 56:1451-1454. generation of old growth Cryptomeria japonica forests on

This content downloaded from 128.122.230.132 on Sat, 25 Jun 2016 23:58:28 UTC
All use subject to http://about.jstor.org/terms
1136 JAMES S. CLARK Ecologyv Vol. 72, No. 3

Yakushima Island. Botanical Magazine (Tokyo) 100:223- Warner, R. R., and P. L. Chesson. 1985. Coexistence me-
241. diated by recruitment fluctuations: a field to the storage
Tilman, D. 1988. Plant strategies and the dynamics and effect. American Naturalist 125:769-787.
structure of plant communities. Princeton University Press, Watkinson, A. R. 1986. Plant population dynamics. Pages
Princeton, New Jersey, USA. 137-184 in M. J. Crawley, editor. Plant ecology. Blackwell
Turelli, M. 1981. Niche overlap and invasion of competitors Scientific, Oxford, England.
in random environments. I. Models without demographic Watt, A. S. 1947. Pattern and process in the plant com-
stochasticity. Theoretical Population Biology 20:1-56. munity. Journal of Ecology 35:1-22.
Turner, M. G., V. H. Dale, and R. H. Gardner. 1989. Pre- Weiner, J. 1985. Size hierarchies in experimental popula-
dicting across scales: theory development and testing. Land- tions of annual plants. Ecology 66:743-752.
scape Ecology 3:245-252. 1986. How competition for light and nutrients af-
Van Wagner, C. E. 1978. Age-class distribution and the fects size variability in Ipomoea tricolor populations. Ecol-
forest fire cycle. Canadian Journal of Forest Research 8: ogy 67:1425-1427.
220-227. West, D. C., H. H. Shugart, and J. W. Ranney. 1981. Pop-
Vitousek, P. M., and W. A. Reiners. 1975. Ecosystem suc- ulation structure of forests over a large area. Forest Science
cession and nutrient retention: a hypothesis. BioScience 25: 27:701-710.
376-381. Westoby, M. 1984. The self-thinning rule. Advances in Eco-
Waring, R. H., and W. H. Schlesinger. 1985. Forest eco- logical Research 14:167-225.
systems: concepts and management. Academic Press, Or- Wright, H. E. 1974. Landscape development, forest fires,
lando, Florida, USA. and wilderness management. Science 186:487-495.

APPENDIX I

The effect of thinning rate on the distribution of densities


Consider a cohort of plants that are rapidly growing (and
thus, rapidly thinning) soon after a regeneration niche be- U(x) = Vla(x)] a(u - 1) - a(u)
comes available (a = 1). The high thinning rate dictates that v - (v - 1)
the population actually spends a small fraction of a year at a The quotient in Eq. 1. 1 is the relative proportion of the age
given density x. In contrast, at some later time a > 1 plants interval (v - 1, v) at which density is between u - 1 and u
are thinning less rapidly (Fig. 1). The "fraction" of time a(x) (Fig. A. 1). During the remainder of the interval (v - 1, v)
actually spent at a given density is much greater in an amount density is not between u - 1 and u. Expanding a(x) in a
determined by the thinning rate. This result is demonstrated Taylor series about (v - 1) and ignoring all terms of order
as follows: -2, we can write
Define discrete probability density functions U and V such
that
a(u) = (v - 1) + d [U - x(v - 1)] + O(X2),
dx
U(x) = Pr[u - 1 x(a) < u]
and

and da
a(u - 1) = (v - 1) + da [(u - 1) - x(v - 1)] + O(X2),
Vla(x)] = Pr[v - 1 a(x) < v],
where o(x2) represents all terms of order -2. Substituting in
Eq. 1. 1 yields
where u and v are values of density and age, respectively (Fig.
A. 1). U is the probability that a randomly selected patch da u - 1 - u
U(X) = Vja(x)] v-(v-1
supports density x. V is the probability that a randomly se-
lected patch is a yr old. We wish to solve for U(x) and to
establish its relationship to the continuous distribution fix). = V(a)
U(x) is the product of two events, the probability that a
random patch is of an age a(x), i.e., that age at which the
patch supports density x, and the portion of age a(x) at which Given that w(a) is the limiting distribution for V(a), then i\(x)
density actually is x, = I da/dx I and fix) is the continuous analogue of U(x).

This content downloaded from 128.122.230.132 on Sat, 25 Jun 2016 23:58:28 UTC
All use subject to http://about.jstor.org/terms
June 1991 PLANT DENSITY AND AGE STRUCTURE 1137

APPENDIX II
x(v-1)
Derivation of moments for the distribution of patch ages
U
Given the distribution of patch ages (recurrence times)
::::..
LI-i..... .
::::x a..).a
.......)..
.-]) v
Bl I:::: ..:::::::.
wr(a) = / exp[-(Xa)j],
F(1 /c 1 1)
......... .........:::::::: ::: :.::::::::.
........... ................. ::.. :........:....- .....
::::::::::::::: ........... ...::: ..::: ................... .

the mth moment of wr(a) is written as


........ .... ...., 11 - ...... :-0: . ?
x(v) . .. , ...'-.EE-. I .,,-.---.
. . . ... . . . . . . . . . . . .

Exam] =J amzr(a) dal


v-1 a(u) a(u - 1)v

r it amexp[-(Xra)1'] da.
::- i v :::::::::::~:Ez
The substitution u = (XAa)' yields

_ _ _ _ _ OC (M fi)

FIG. A. 1. Geometric interpretation.... of. th contribution


-cro Ic I) u c e-u du.
XjmcrF(l/c + 1) J

The integral expression is now a gamma function with pa-


enced in Appendix I.~ ~ ~~..........I.............
rameter . Using this fact, together with the property
c

r(a + 1) = ar(a), the mth moment is written as

? - V
rV
zM + I0

Age a Xr-ro 1c)


FIG. A. 1. Geometric interpretation of the contribution of yielding expectation
thinning rate to the distribution of cohort densities. The co-
hort thins at a rate that is approximated by the slope of the E[Ar r (2/c)
shaded triangle. Labeled values of density and age are refer- Xrr(i/c)'
enced in Appendix I. and coefficient of variation

rJ /( c) - F2(2/c)
cv[Ar] = r(2/c) F(31c[ - F(l/c) J

By use of well-known properties of the gamma function, it


can be shown that these results collapse to the exponential
for c = 1, i.e., E[ArI = I/Xr and cv[Ar] = 1, and that c = 2

decreases this expectation to E[Ar] =


XrVgr

This content downloaded from 128.122.230.132 on Sat, 25 Jun 2016 23:58:28 UTC
All use subject to http://about.jstor.org/terms

You might also like