You are on page 1of 22

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/2515171

Controlling Chaos in Ecology: From Deterministic to Individual-based Models

Article · March 2001


Source: CiteSeer

CITATIONS READS

5 33

4 authors, including:

Javier G. P. Gamarra Marta Ginovart


Food and Agriculture Organization of the United Nations Universitat Politècnica de Catalunya
49 PUBLICATIONS   1,067 CITATIONS    107 PUBLICATIONS   649 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Global Forest Biodiversity Initiative View project

IBMs (Individual-based models) using thermodynamic behavior-rules for the individuals. View project

All content following this page was uploaded by Javier G. P. Gamarra on 15 January 2013.

The user has requested enhancement of the downloaded file.


Bulletin of Mathematical Biology (1999) 61, 1187–1207
Article No. bulm.1999.0141
Available online at http://www.idealibrary.com on

Controlling Chaos in Ecology: From Deterministic to


Individual-based Models
RICARD V. SOLÉ
Complex Systems Research Group,
Department of Physics FEN, Universitat Politécnica de Catalunya,
Campus Nord B5, 08034 Barcelona,
Spain
E-mail: ricard@complex.upc.es

JAVIER G. P. GAMARRA
Complex Systems Research Group,
Department of Physics FEN, Universitat Politécnica de Catalunya,
Campus Nord B5, 08034 Barcelona,
Spain
and Forest Technology Center of Catalonia,
Pujada del Seminari s/n,
25280 Solsona,
Spain
E-mail: javier@complex.upc.es

MARTA GINOVART AND DANIEL LÓPEZ


Escola Superior d’Agricultura de Barcelona,
Urgell 187, 08036 Barcelona,
Spain

The possibility of chaos control in biological systems has been stimulated by recent
advances in the study of heart and brain tissue dynamics. More recently, some au-
thors have conjectured that such a method might be applied to population dynamics
and even play a nontrivial evolutionary role in ecology. In this paper we explore
this idea by means of both mathematical and individual-based simulation models.
Because of the intrinsic noise linked to individual behavior, controlling a noisy sys-
tem becomes more difficult but, as shown here, it is a feasible task allowed to be
experimentally tested.
c 1999 Society for Mathematical Biology

1. I NTRODUCTION

Complex dynamical patterns are an inherent part of both real and model ecosys-
tems and have been a matter of extensive studies in recent years [Hassell et al.
(1976); Godfray and Blythe (1990); Tilman and Wedin (1991); Bascompte and
Solé (1995) and references therein]. Chaos is a particularly relevant example of
0092-8240/99/061187 + 21 $30.00/0 c 1999 Society for Mathematical Biology

1188 R. V. Solé et al.

these patterns and it can be obtained even from very simple nonlinear models as
a result of intrinsic nonlinearities (May, 1974, 1976; May and Oster, 1976; Schaf-
fer and Kot, 1986). As an example, let us consider the following one-dimensional
map:   
Xt
X t+1 = X t exp r 1 − (1)
K
describing single-species population dynamics with nonoverlapped generations
where r is a constant representing the population growth rate (under no shortance
of resources) and K is the carrying capacity of the population (maximum popula-
tion attainable by the system). This is the well-known Ricker model (Ricker, 1954)
but other uniparametric, smooth maps with a single maximum on their interval of
definition display similar features (in particular a period-doubling scenario leading
to chaos).
Although the evidence for this type of dynamics in nature is still controversial
[see Berryman and Millstein (1989); Lomnicki (1989); Nisbet et al. (1989); Rohani
and Earn (1997)], recent experimental studies involving populations of flour beetle
Tribolium castaneum (Constantino et al., 1995, 1997; Dennis et al., 1997) have
confirmed the theoretical predictions even when stochastic terms were included
in the model. By means of changes in demographic parameters, these authors
obtained a broad variation in the nature of population fluctuations by showing that
experimental data were consistent with the expected model predictions (including
steady points, oscillations and chaos).
A different possibility for showing the existence of chaos in real populations is
provided through the theory of chaos control (Ott et al., 1990; Ditto et al., 1995). It
has been shown that, under some constraints, a chaotic system may be ‘controlled’:
unstable periodic orbits can be stabilized by means of small, periodic perturbations
of the system parameters (Ott et al., 1990) or states (Güemez and Matı́as, 1993;
Parthasarathy and Sinha, 1995). In fact, the very nature of chaotic dynamics, with
its sensitive dependence on initial conditions and an infinite number of unstable
periodic orbits simultaneously embedded in phase space, makes feasible the pos-
sibility of chaos control by moving and keeping the system’s trajectory close to
one of the (unstable) orbits, artificially stabilizing one of them. The flexibility of
this dynamics is opposed to the case where the fixed point is stable. In such a
case, moving to another periodic orbit needs a very large alteration of the system.
Further work explored the possibility of chaos control in spatiotemporal dynamics,
both in physics and biology (Astakhov et al., 1995; Doebeli and Ruxton, 1997).
Biological applications include the stabilization of chaos in neural systems (Schiff
et al., 1994; Solé and Menéndez de la Prida, 1995), chemical reactions (Petrov et
al., 1993) and cardiac tissue (Garfinkel et al., 1992).
Some ecological and evolutionary consequences have already been pointed out.
Doebeli (1993) hypothesized about the possibility of natural self-control in a pop-
ulation under unstable Ricker dynamics, as a way to keep the population at equi-
librium in a form of K selection. Besides, Doebeli and Ruxton (1997) have argued
Controlling Chaos in Ecology 1189

about the possibility of chaos avoidance in some populations located in areas where
external factors such as climate may have seasonal changes that drive the dynamics
from a chaotic situation to more stable dynamics with small fluctuations, defining
this external seasonality as a natural way in which populations can be stabilized.
In this sense some studies on mammal populations, such as the classical ones on
lynx fur returns or vole populations in zones closer to the poles, whose dynamics
is clearly cyclical, could be affected by this kind of control.
Deterministic models of populations with chaos control are well known in the
literature [see, e.g., McCallum (1992); Stone (1993)]. However, it is fairly well
known that the dynamics in real ecosystems may be seriously affected when noise
is at work. In this context, Allen et al. (1993) showed that local population noise
is amplified in the presence of chaos, due to the resulting asynchronous behavior
among populations [see also Solé and Gamarra (1998)]. However, there remains
an open question: might a population under a chaotic regime be stabilized in the
presence of internal noise? If so, which are the ecological consequences for real
ecosystems? Our initial hypothesis is that populations with global mixing (popula-
tions located in a sufficiently small area to allow dispersal to every point in the area
with the same probability, or, inversely, populations with large enough dispersal
capacity to all points) are not affected in the presence of noise by asynchronici-
ties large enough to avoid the correlating effect of control in the population. Thus
the techniques for controlling chaotic dynamics would be able to overcome the
problem of noise and consequently, the robustness of control would be a further
argument supporting the existence of instabilities due to the nonlinear nature of
ecosystems.
Here we will consider the applicability of some of the techniques for control-
ling chaos in order to test the existence of deterministic chaos in ecosystems and
the possibilities for new experimental manipulations of laboratory populations will
be discussed. The presence of noise in the dynamics and two individual-oriented
approaches will shed light onto the robustness of control under stochastic environ-
ments (as those found in microecosystems).

2. M ETHODS OF S TABILIZATION IN D ISCRETE S YSTEMS

2.1. Parametric (internal) perturbations. Pioneering work on controlling un-


stable orbits was early made by Ott et al. (1990) (OGY method). The methodology
of control needs a previous knowledge of the trajectory of the orbit to stabilize, or
at least the location of the fixed points in a Poincaré section. The method has been
successfully applied in physics (i.e., chaotic oscillators) (Ditto et al., 1990), but it is
too difficult to implement in ecological systems: the need for previous knowledge
of the system’s trajectory implies the use of very long time series and ecological
data sets available are still scarce and short.
A more plausible protocol involves the use of periodic perturbations on the popu-
1190 R. V. Solé et al.

lation growth rate [see Doebeli (1993); Stone (1993)]. Let us start with the single-
species population model described by (1), where X t+1 = X t f µ (X t ), such that
f µ (X ) = exp[r (1 − X )] and thus lim X →∞ f µ (X ) = 0. Let us assume that r > rc ,
so we are at the chaotic domain of the parameter space. This state can be charac-
terized through the Lyapunov exponent, defined as λ L = hln |∂x f µ (xi )|i, where the
average is taken over the entire system’s dynamics, i.e.,: {x1 , . . . , xn }. Since we
are in the chaotic domain, a fast and intuitive method of control is to perturb the
parameter r so that we move to r < rc . That is, if we consider model (1) involving
control each p generations:
X t+1 = f µ0 (X t0 )

where X t0 = X t−( p−1) and µ0 = r + γ where γ < 0 is a constant rate accounting


for seasonal internal causes of population decline due to density dependence. This
procedure, illustrated in Fig. 1(a) and (b), where the dynamics is biperiodic for
the Ricker’s map, ensures that the slope of the curve | f 0 (µ, xt∗ )| < 1, satisfying
the conditions for stability [see May (1976); Stone (1993)], where xt∗ is the fixed
point, i.e., the intersection of the xt+1 = xt line and the function f (µ, xt ), then a
period-doubling reversal appears as a direct consequence of moving far below rc
(Stone, 1993). In case we apply a perturbation every two generations, the Ricker’s
map must be represented in a xt+2 − xt graph (second-iterate map), where xt+2 =
f µ ( f µ0 (xt )), and the minimum number of orbits attainable is 1n = p = 2, due to
the existence of two stable fixed points X 1∗ and X 2∗ .

2.2. State (external) perturbations. Two different approaches may be treated.


Both add an external feedback to the population every p iterations and both are
very easy to apply and have a more meaningful ecological understanding to ac-
knowledge the role that dispersal might play among metapopulations in order to
avoid chaos and consequently the probability of extinction.
The first one, developed by Güemez and Matı́as (1993) (GM or proportional
feedback method) involves stabilizing one orbit of the many unstable periodic ones
of a chaotic system by perturbing the system with periodic pulses, which are pro-
portional to the state the system presents. For some values of intensity and pe-
riodicity of the perturbation the orbit is stabilized. The protocol should be used
to control systems where we do not know the parameters at work, but where we
can take measurements over the variables, i.e., the population size of biomass of a
species.
Control is exerted through the application of the feedback to the population x
every p iterations:
X t + 1 = f µ (X t0 ) ∗ (1 + γ )

where, again, X t0 = X t−( p−1) , the parameter µ is kept constant and γ represents
the strength of the pulse. This type of control may be interpreted as a seasonal
or periodic density-dependent immigration (γ > 0) or emigration (γ < 0) term
Controlling Chaos in Ecology 1191

3 (a ) 3 (b)
γ=0 γ=1

2 2

N(t)
N(t)

1 1

0 0
0 1 2 3 0 1 2 3
N (t – 1)
N (t – 1)
3 (c) 4 (d)
γ=1
γ = 0.15
3
2
N(t)

N(t)

1
1

0 0
0 1 2 3 0 1 2 3 4
N (t – 1) N (t – 1)

Figure 1. Graphical interpretation of some of the techniques of control,


where some perturbation γ is added to either a parameter or a state variable
of the system in the Ricker map, such that r 0 = r (1 − γ ). The graph is
shown for r = 3 and Tcont = 1 in all cases. As Y increases, the slope
(derivative) of the map at the fixed point, i.e., that point where the curve
crosses the N (t) = N (t − 1) line, moves towards the curve, getting values
d F(N ) < 1, where F(N ) is the corresponding Ricker map. (a) Chaotic
dN
regime, no control. (b) Parametric control: r 0 = r (1 − γ ). (c) and (d) are
both external methods GM and PS respectively.

when individuals tend to be clumped or territorial, respectively, thus adding a mul-


tiplicative pulse to the population size. Figure 1(c) represents the effect of the
GM control method on the first-iterate map for a period of control p = 1 with
strength γ = 0.15. We can represent the r − γ phase space showing the con-
tour lines separating the areas corresponding to different stable periodic orbits, as
in Fig. 2(a). In the first return map the sensitivity of the system to the control
feedback is stronger for negative values of γ , giving raise to smaller population
sizes by smoothing the single hump of the map return. For γ > 0, the fixed point
gets higher values, shifts towards the rightmost part of the return map and, under
1192 R. V. Solé et al.

some values, stabilizes the orbit for larger population sizes. A similar protocol
rooting in the target of the perturbation was first developed by McCallum (1992).
The idea grew from assuming a chaotic population following a Ricker’s dynamics
and adding an immigrant population as a structural perturbation every time step
to observe that unstable orbits became periodic under certain values of immigra-
tion γ . Stone (1993) further extended this work to demonstrate that immigration
opened the possibility of period-doubling reversals. This method was later stated
as the method of constant feedback control (Parthasarathy and Sinha, 1995) (PS
hereafter). Such state perturbation consists of applying a constant pulse γ every p
generations in the form
X t+1 = f µ (X t0 ) + γ

with X t0 = X t−( p−1) and the parameter µ kept constant. Let us consider again
that we have r ≈ rc . The return map in Fig. 1(d) shows that the addition of a
positive γ moves up the map and increases the value of the nontrivial fixed point
by displacing it towards the plateau region, therefore changing the value of the first
derivative such that | f µ0 (x)| < 1. The phase-space diagram in Fig. 2(b) shows that
if γ < 0 then there will be only a few possible values under which the orbit visit
periodic windows. Control may appear in this method with very small values in
1γ , so the nature of the orbits appearing in this system seems to be more sensible
to perturbations at high r values. There is also an area stated as the escape region,
generally under negative values of gamma, where emigration is so strong, that the
resulting return map presents a displacement of the unstable trivial fixed point from
zero to X t∗ → 0, so when the system visits values of X t below this fixed point, the
population infallibly becomes extinct.

3. S TABILIZATION IN C ONTINUOUS T IME

Perturbation of unstable orbits in order to stabilize the dynamics is also possible


for the case of populations following continuous kinetics and systems with several
dimensions. The extension of the previous perturbation methods to a continuous
model is straightforward and has been shown to work successfully. In this case,
since the nonlinear equations are integrated by means of a characteristic δt-time
discretization, control will be applied each pδt steps in the simulation.
We have addressed the problem of controlling chaos in a typical ecological sys-
tem, such as a model plankton community, where apparently internally driven ir-
regular oscillations may occur, as has been observed in some experimental mi-
croecosystems (Ringelberg, 1977). This realistic model, a three-dimensional ODE
model of plankton dynamics, showing a single food web with a series of interac-
tions able to produce chaotic dynamics in some of the trophic levels, was made by
Scheffer (1991). He developed a model with three trophic levels that involved
planktonic algae (A), herbivorous zooplankton (Z ) and carnivorous zooplank-
Controlling Chaos in Ecology 1193

0.25
>4 >4
0 4

Strength of control
>4 2
1
–0.25

–0.5

–0.75 1

–1
0 1 2 3 4

2.75

2
Strength of control

1.25 1

0.5 2 4 >4

–0.25
escape
escape
–1
0 1 2 3 4
Growth rate

Figure 2. r − γ phase-space diagrams for (a) GM and (b) PS control


methods characterizing regions of p = 1, 2, 4, > 4 periodic orbits and an
escape region, characterized numerically. In both, Tcont = 1.

ton (C) and the differential equations describing the dynamics are

dA
= r A − c A2 − g Z
dt
dZ
= g Z ce − z mort Z − pc
dt
dC
= pCce − cmort C (2)
dt

where r > 0 is the algal relative growth rate, c > 0 is a competition coefficient
for algae, g ≥ 0 and p ≥ 0 are respectively a grazing and a predation rate both
being Monod saturated functions and thus identifying a type II functional response
for herbivory and predation. ce ∈ (0, 1) is the conversion efficiency from food to
biomass and z mort > 0 and cmort > 0 are mortality rates. In this three-trophic level
model, strange attractors may appear as a consequence of increasing r by a slight
amount.
1194 R. V. Solé et al.

In order to test whether control methods may work we ran simulations under
the same parameter values that Scheffer (1991) set to get strange attractors in his
model. We used the proportional (GM) and constant feedback perturbations in
each of the three trophic levels appearing in equation (2). Graphical results appear
in Fig. 3. Through several values of p and γ , control may appear in the form
of a wide variety of dynamical behaviors, from stable nodes [Fig. 3(a)], to stable
focus [Fig. 3(d)] and limit cycles [Fig. 3(b), (c), (e), (f)]. There is a special case
[Fig. 3(e)], where using the GM control method the orbit wraps around the strange
attractor. In such case, the regular dynamics continues several iterations after the
control routine was stopped, maintaining the system in the stable area.
The control induced in this system also contributes to produce a period-doubling
reversal [sensu Stone (1993)]. The main point for the control of chaos in these
n-dimensional continuous systems is that a small perturbation applied on one of
the trophic levels, is able to propagate very rapidly over the whole system by its
effect on the interactions. As far as chaotic dynamics is flexible enough, the whole
ecosystem changes the trajectories of its components towards more regular ones.
Now the obvious question is whether or not these control methods will be suc-
cessful in controlling real populations. A first criticism comes from the lack of
noise in our models. Noise can strongly modify our previous expectations. Since
the standard control techniques require deterministic dynamics in order to guaran-
tee that the orbit will visit the desired domains where control is applied, the intro-
duction of noise can prevent the dynamics from reaching such domains. Addition-
ally, some authors have suggested that the effect of individual, microscopic fluc-
tuations might introduce fundamental effects eventually breaking down a macro-
scopic description [Fox and Keitzer, 1990; see however Peeters and Nicolis, 1992].
An additional point concerns the translation of these theoretical models to realis-
tic, small-sized and intrinsically noisy laboratory populations. Since the control
method is expected to be performed by adding given amounts of individuals to the
population, and because of the unavoidable random factors arising from individual
complexities, there is in principle no guarantee that an efficient control is going to
be reached. In the next two sections we explore this problem by using two differ-
ent individual-based models of both discrete and continuous dynamics involving
reasonable population sizes in the range of those used in experimental systems.

4. C ONTROL IN A D ISCRETE S TOCHASTIC IBM

The study of the dynamics of nonoverlaping generations (i.e., generations where


reproductive individuals and offspring do not overlap in time) has already been
widely carried mainly in terms of one-dimensional nonlinear deterministic models
(May, 1974; Hassell et al., 1976). These are able to generate a wide spectrum
of behaviors (from stable to chaotic), by simply changing one parameter in the
underlying map (usually the growth rate).
Controlling Chaos in Ecology 1195

GM PS
0.8 0.8
(a) (b)

0.6 0.6

0.4 0.4
C

0.2 0.2

0.0 0.0
0.0 1.0 2.0 3.0 0.0 1.0 2.0 3.0
Z Z
0.8 0.8
(c) (d)
0.6 0.6

0.4 0.4
C

0.2 0.2

0.0 0.0
0.0 2.0 4.0 6.0 0.0 2.0 4.0 6.0
A A
2.0 2.0
(e) (f)
1.5 1.5

1.0 1.0
Z

0.5 0.5

0.0 0.0
0.0 2.0 4.0 6.0 0.0 2.0 4.0 6.0
A A

Figure 3. Attractors reconstructed for the planktonic system. Both GM


and PS control methods have been used. Control is exerted on algae [(a)
and (b)], herbivorous zooplankton [(c) and (d)] and carnivorous zooplank-
ton [(e) and (f)]. Dashed lines represent the strange attractor and solid lines
represent the trajectory under control. Parameters used are: (a) p = 1,
γ = −0.1; (b) p = 3, γ = 0.3; (c) p = 3, γ = −0.05; (d) p = 3,
γ = 0.01; (e) p = 3, γ = −0.2; (f) p = 3, γ = 0.001. The rest of
parameters are the same as in Scheffer (1991).
1196 R. V. Solé et al.

Some problems may appear in the use of these models, such as the fact that the
growth rates are considered as fixed, and the interactions taking place among the
individuals. In this section a more realistic simulation includes the presence of
noise by considering the effects of population size and random variability in the
growth and mortality rates. As far as we know, this is the first fully individual-
based counterpart of the standard mathematical models involving single-species
discrete dynamics. In this IBM model, a detailed description of single individuals
is introduced, necessary for a complete description of the within-generation dy-
namics. The overall dynamics as described by the population values from genera-
tion to generation displays the expected pattern observed from one-hump models
of nonoverlapped generations.

4.1. Model description. The basic methodology of our system comprises several
characteristics to have in mind:
1. Discrete space and time.
2. A group of N individuals each with its own physiological properties, such
as biomass or age, all of them dependent on time.
3. A group of cells in a lattice, with a certain quantity of energy available on
each cell which may change with time and interact with the individuals.
4. Periodic boundary conditions.
In our system we consider individuals moving, feeding on energy, reproducing
and dying. For computational purposes, individuals and energy are placed in a
two-dimensional L × L lattice, although the global mixing makes the spatial di-
mension irrelevant. The basic temporal scale under these interactions corresponds
to the within-generation dynamics. At the end of each generation each individ-
ual reproduces and dies and the next generation starts with the resulting offspring
feeding on a newly updated energy population, i.e., the system behaves as a forced
chemostat. Thus, new generations of individuals (grazers) are conditioned by the
final results of the within-generations evolution (the full flow diagram is provided
in Fig. 4). This protocol is repeated a certain number of iterations.
Again, our system develops as a well-mixed system, where the only local inter-
actions account for competition for resources among individuals in a cell (which
may lay the basic rules for density dependence), but there is global movement of
individuals and offspring.
At the beginning of each generation there is a quantity E of energy particles ran-
domly distributed over the whole lattice. Within each generation, there is neither
energy input nor energy movement across the system. During the evolution of the
generation, each individual (level 1 in Fig. 4) consumes a random fraction of quan-
tity E(µ, σ ) of the energy available in the cell with certain dissipation rate, the
individuals grow and randomly move to other cells looking for more energy, until
the end of the generation is accomplished (i.e., when individuals have exploited
all the energy in the system). When this happens (level 2 in Fig. 4), we have to
Controlling Chaos in Ecology 1197

INITIAL BLOCK
START

DESCRIPTION OF INDIVIDUALS
AND ENVIRONMENT

INITIAL CONFIGURATION
(INDIVIDUALS AND ENVIRONMENTAL
CONDITIONS)

ITERATION STEP MAIN LOOP

EVALUATION OF CONDITIONS
FOR THE END OF GENERATION

LIST OF ACTIVE
INDIVIDUALS

FOR EACH INDIVIDUAL


-MOVEMENT
-NOURISHMENT
SEARCH AND CONSUMPTION
-METABOLISM
LEVEL 1 GROWTH AND DISSIPATION

LEVEL 2 NO
EMPTY LIST?

YES

NO END OF
GENERATION?

YES

-REPRODUCTION
-NEW GENERATION
-DEATH

VARIABLES DATA ANALYSIS


UPDATING

NUTRIENT NO FINAL
RE-INITIALIZATION CONDITIONS?

YES

END

Figure 4. Schematic representation of the processes taking place in the


stochastic Ricker IBM.
1198 R. V. Solé et al.

compare the individual biomass, which increment is proportional to the quantity of


resource consumed, with the fixed minimum mass for reproduction. The grazers
that achieve this biomass reproduce some fixed quantity of offspring and die and
the resulting offspring is redistributed over the whole. The newborns are the ones
that start the next generation. Environmental conditions such as the quantity of en-
ergy available only will be modified by direct consumption during the generation
time. A fixed amount of energy is then updated at the beginning of the next gener-
ation. The system is closed, since there is no energy input in the within-generation
time. The list of parameters used in the model is shown in Table 1. Using this
model, we have found all the basic features known from the analysis of the Ricker
map equation, from periodic doublings to intermittency. The analysis of the cor-
relation dimension for the dynamics shows that it is one-dimensional with fractal
dimension less than one for chaotic attractors.

Table 1. Main parameters used in the stochastic Ricker-map IBM.


Parameter Definition Default value Effect when increased

E Energy input 2 × 106 Larger amplitude


Tcont Control period — PDBa
γ Perturbation size — Larger amplitude
ω Number of offsprings/individual 5 Larger amplitude, PDBa
δ Dissipation rate at consumption 0.3 PHBb
Mb Mass at birth 250 PHBb
Mrep Minimum mass of reproduction 103 PDBa
a PDB: Period-doubling bifurcations. b PHB: Period-halving bifurcations.

Again we introduce additive external perturbations in order to stabilize the dy-


namics in our Ricker map. Some results are given in Fig. 5. From top to bottom in
this figure we show: (a) the system with no control, at the chaotic regime, (b) two-
periodic controlled orbit (control is applied from generation n = 300 to n = 700)
and (c) single-point controlled attractor. In all these examples we use a system
with a maximum size N = 2000 with default values as those indicated in Table 1.
The two sets of control values are γ = 150 and Tcont = 2 for the periodic con-
trolled orbit and γ = 400 and Tcont = 1 for the steady state. Extensive numerical
simulations show that this control method is highly effective and in fact we can
plot a bifurcation diagram using γ as a parameter and show that a period-halving
scenario is obtained.
Once again, the system may be controlled under certain regimes of perturbation.
In our discrete system, the period of control is not affected by the generation-
time, but control is more effective as Tcont becomes smaller. The perturbation size
does not need to be large (i.e., γ ≈ L×L10
), notwithstanding that the system has to
overcome the stochastic fluctuations affecting the dynamics.

4.2. Lotka–Volterra predator–prey individual-based model. In this section we


introduce the basic rules describing a Lotka–Volterra (LV) model with microscopic
Controlling Chaos in Ecology 1199

1500 1500

1200 1200

900 900

600 600

300 300

1500 1500

1200 1200
Population size

900 900
N (t + 1)

600 600

300 300

1500 1500

1200 1200

900 900

600 600

300 300

0 0
0 500 1000 0 300 600 900 1200 1500
Generations N (t)

Figure 5. First iterate map of the output of a simulation on the stochastic


Ricker IBM with additive (PS) control with control periods Tcont = 1, 2.
Empty and full circles represent the chaotic and control regimes, respec-
tively. In both simulations dd = 2 and γ = 25. The other parameters are
defined in Table 1.

interactions among its elements. Again, one of the reasons to use an individual-
based model stems from the implicit randomness of its individual interactions, now
among different species, setting explicitly the stochastic nature of the system (Jud-
son, 1994; Wilson, 1998). A similar LV model was introduced by Ricker involving
a rather sophisticated set of individual properties (Solé and Valls, 1992). Here we
restrict ourselves to a much more simple set of interactions. Let us note, however,
that the basic properties of the bifurcations are the same as those observed in the
1200 R. V. Solé et al.

original model. The continuous nature of the model roots from the time-coupling
of reproduction among its individuals. The exploration of this model is justified
since a single-species population as the one described in the previous section is
unlikely to be available. Two-species microecosystems are well known from Huf-
faker’s classic studies.
Due to the fact that we are not intending to explore the spatial correlations, but
instead the stochastic nature of the system and the posibilities for control under the
presence of noise, we do not explore the spatially extended version of the model
but a well-mixed model where interactions among individuals occur as a periodi-
cally forced chemostat, where the concentration of substrate or energy enters the
system periodically or, more specifically, sinusoidally, causing the whole system
to oscillate repeteadly with some characteristic period and amplitude, controlled
primarily by the period and amplitude of the input energy, and thus creating a time
dependency in the energy input (Pimm, 1991; Kot et al., 1992; Rinaldi et al., 1993).
This forced predator–prey model is much more easy to display chaotic dynamics
and remain stabilized. If a three-species ecosystem is used then the top predator is
likely to become extinct, unless very high populations are used.
For simplicity, we consider our physical system as formed by a number of N
noninteracting sites (where no spatial structure exists). This obviously introduces
a limitation into the prey population, which, because of the finiteness of the system
will grow logistically. Let N 1 (t), N 2 (t) be the number of prey and predators at a
given step t. Below, we describe the set of rules used.

Energy release. Prey feed on some type of substrate which we will call energy,
which is randomly introduced into the cells. Let E i (t) be the amount of energy
at the ith site (i = 1, . . . , N ). In our model E i (t) ∈ 0, 1 depending on whether
the site is energy-free (0) or not (1). In a first step in which each cell can receive
energy with a probability
 
1 t
p(t) = 1 + sin ∈ [0, 1]
2 T

where T is a parameter introducing the period of the energy fluctuations, i.e., P =


T

. As expected, high values of T approach the behavior towards an unforced
system, where the system behaves as a typical predator–prey LV system, giving
rise to a simple limit cycle. If a site is such that E i (t) = 0, it becomes E i (t +1) = 1
with probability p(t), independently of the prey and predator states in the cell.

Interactions. The following step involves interactions among preys and energy
and among predators and preys. Let Si1 (t) and Si2 (t) be the states of the ith site for
prey and predator, respectively, i.e.,

j 1 i-site occupied
Si (t) =
0 i-site free of j
Controlling Chaos in Ecology 1201

for both prey ( j = 1) and predator ( j = 2). Then the following rules are applied
n = N times (this defines our timescale: for each t there are N updatings). We
choose a site 1 ≤ i ≤ N at random.

1. If Si1 (t) = 0 and Si2 (t) = 0 then we choose the next site.
2. Death. Prey and predators die with certain probability m 1 ≤ m 2 . If there is
no death then:
3. If Si1 (t) = 1 and Si2 (t) = 0 (only prey present) two possible outcomes are
possible:

(a) If E i (t) = 0, then no reproduction is allowed to occur and the prey


moves to a new free site k (such that Sk1 (t) = 0) which is chosen at
random.
(b) Feeding and reproduction of preys. If E i (t) = 1 then feeding occurs
(E i (t) = 0) and reproduction is possible with probability r 1 ∈ [0, 1].
The parent individual remains in the same position, whilst the offspring
moves to another randomly chosen site.

4. If Si1 (t) = 0 and Si2 (t) = 1 (only predator present) then the predator moves
to another randomly chosen site looking for a prey [as in rule 3(a)].
5. If Si1 (t) − 1 and Si2 (t) = 1 (prey and predator present). Movement, feeding
and reproduction occur as in the former case, but the predator feeds on prey
before energy is offset from the cell.

Table 2 shows the parameter set used in the model. The onset of control in this
chaotic three-species system has been attached through the addition of external
perturbations by adding γ preys each Tcont time (Fig. 6). Control is also possible
by adding predators to the system, although the parameter set is somewhat differ-
ent. Hastings and Powell (1991) and McCann et al. (1998) state that chaos may
arise whenever the period of one oscillation (i.e., the cycles followed by the preda-
tor population, which are longer than that of the prey due to the longer lifetimes
of predators) is not some multiple of the other frequency (i.e., that of the prey).
Interestingly, our results suggest that, in order to achieve better control, such as
limit cycles or even stable points with some random fluctuactions, it is necessary
for the system to have a control period that is some multiple of the period of energy
fluctuation, as may be seen in Fig. 6. Our conjecture is that by doing this, we get to
couple the oscillations in all three systems, so in some way we are able to ‘unforce’
the system, avoiding asynchronization and chaos in all trophic levels. By adding
large-sized perturbations, we accelerate the dynamics of the system by rapidly in-
creasing the prey population to its upper bound in the attractor. As a result the
attractor is wrapped around by the limit cycle (see Fig. 6).
1202 R. V. Solé et al.

Table 2. Parameters and default values used in the LV IBM.


Parameter Definition Default value Effect when increased
L×L System size 30 × 30 Improves control
T Period of energy sine wave 40 Larger cycles
Tcont Control period 20 Large amplitude, PDBa
γ Perturbation size — Large amplitude
m1 Death rate of prey 0.05 Extinction likely
m2 Death rate of predator 0.2 Fixed point
r1 Feeding-reproduction rate of prey 0.95 Short-term correlations
r2 Feeding-reproduction rate of predator 0.85 Extinction likely
d1 Interspecific DDb factor of prey 40 Improves control
d2 Interspecific DDb factor of predator 40 Improves control
a PDB: Period-doubling bifurcations. b DD: Density-dependence.

5. D ISCUSSION

Although not fully demonstrated yet, the existence of chaotic dynamics in real
ecosystems has been a matter of extensive studies over recent years. It opens
new daring questions in biology, due to its intrinsically unpredictable behavior and
counterintuitive results. Even more important, it is a useful platform from which
new and fascinating testable designs may arise. One step ahead, the possibilities
for chaos control help us to unfold the apparently hidden nature of chaos, pulling
out the reversibility of one of the main characteristics of chaos: the presence of
period-doubling bifurcations (Stone, 1993).
Techniques for stabilizing the infinity of unstable periodic orbits appearing in the
presence of chaos are based mainly in two categories: the addition of perturbations
to the parameters acting over the dynamics (Ott et al., 1990) and to the dynamic
variables at play (Güemez and Matı́as, 1993; Parthasarathy and Sinha, 1995). The
first one, more difficult to apply in real systems, needs some previous knowledge of
the dynamics of the system and gets less effective as the system gets more complex.
The second one is simpler, although without some knowledge of the system needs
some trials before being sure about the effect of positive or negative perturbations
are needed.
The construction of individual-based population models may help us to under-
stand the role of noise in the characterization of the attractors and the effectiveness
of control techniques in the presence of stochastic fluctuations. So far as we have
not used locally interacting individuals, but global mixing of populations, we have
assumed that each cell of the lattice is independent of another one, where each cell
shares the same probabilities of dispersal to every cell of the lattice. Application of
control works well for both systems studied, the LV predator–prey forced chemo-
stat and the Ricker individual-based model. For the constant feedback method,
Controlling Chaos in Ecology 1203

A B
(a) 400 400

300 300

Predators
Predators

200 200

100 100

0 0
0 100 200 300 400 0 100 200 300 400
400
Preys C 400
Preys D

300 300
Predators

200 Predators 200

100 100

0 0
0 100 200 300 400 0 100 200 300 400
Preys Preys

A B
(b) 400 400

300 300
Preys
Preys

200 200

100 100

0 0
300 600 900 1200 1500 300 600 900 1200 1500
Number of iterations Number of iterations
C D
400 400

300 300
Preys

Preys

200 200

100 100

0 0
300 600 900 1200 1500 0 100 200 300 400
Number of iterations Number of iterations

Figure 6. (a) Numerical output of several simulations of the stochastic LV


3D IBM. Attractors defined by the control PS method are represented by
solid lines. (A) T = 10, Tcont = 20, γ = 120. (B) T = 40, Tcont = 20,
γ = 200. (C) T = 5, Tcont = 5, γ = 180. (D) T = 5, Tcont = 5, γ = 80.
The other parameters are defined in Table 2. (b) The same as in Fig. 6(a),
but representing the evolution of preys in time.
1204 R. V. Solé et al.

used for both IBMs, the perturbation consists of the addition of individuals in ran-
dom sites chosen from a uniform distribution, which have direct effects on the intra-
and interspecific density dependence of the populations. Thus, under unstable dy-
namics, the system tends to visit extreme values. The nature of the fluctuations
relies on the following description: if a population has depleted the resources due
to its initially high density, its intraspecific competition for resources available, will
have been strong enough to increase its mortality, so the population will achieve
very low values at the end and the next step, when the available resource has been
renewed, very low initial populations will start to consume it. This process, if fast
enough, will be able to decouple the resource and the population dynamics in con-
tinuous systems. If, when resources have been depleted, we add some individuals
(i.e., we add an structural perturbation) to the already scarce population, higher
offspring individuals will start consuming the available resource when renewed, so
basically the effect of the perturbation has been to increase the number of visits
over less extreme values and, in some cases, to achieve an artificially created point
equilibrium or will couple both subsystems if the requirements of conmensurabil-
ity in continuous systems (such as the planktonic or the LV IBM described) are
fulfilled. This requirement for control is that the period applied for control should
be some multiple of the period of the chemostat, as some authors have previously
suggested (Hastings and Powell, 1991; McCann et al., 1998). Moreover, its ef-
fectiveness depends on the size and frequency of the perturbation applied, as in
discrete systems. For planktonic microecosystems, simple models of competition
among algae have defined strange attractors, though the set of parameters needed
for it is highly restricted to certain values. Outside this range (i.e., some parameters
are modified) the system behaves more or less periodically, so the appearance of
chaos in the real system does not seem likely. However, if the model is able to
generate chaos for some set of parameters, deterministic chaotic dynamics might
in fact be implicit in the basis of the dynamics, and that shift would be certainly
defining a parametric perturbation to leave the chaotic regime.
The relationship with real systems does not come up from a direct interpretation
of the results, surely because of the complexity inherent in such systems. However
some basic assumptions can be applied from our models. Basically, in populations
undergoing global mixing, i.e., small populations where the size of its geographic
range is closed and sufficiently small to ensure that dispersal is uniform over the
whole range, a systems manager would need to add some individuals [as McCal-
lum (1992) suggested] or remove them (which is equivalent to the assumption of
negative values of γ ) from the system. In biological control, if some nonlinear
dynamics is at play, and a given population needs to be over a certain threshold in
order to produce considerable damage on crops, control techniques may be used
to maintain the population under this threshold, even in the presence of stochastic
external events, such as short episodes of high temperatures.
The presence of spatial correlations and/or pink or blue noise would increase
temporal correlations and would need some further study. The presence of chaos
Controlling Chaos in Ecology 1205

in some metapopulation models has already been observed to contribute to spatial


asynchrony among the populations by increasing the noise of its local components
(Allen et al., 1993; Solé and Gamarra, 1998). Thus, spatial synchronization (by
means of periodic pulses) seems to be a possible alternative for controlling the pres-
ence of chaos and outbreaks in some populations. Currently an artificial change in
the parameters of the systems (such as the growth rate) seems a fairly impossible
task, techniques of control based on perturbations over the dynamic variables seem
to be the only plausible alternative.
Some final remarks need to be emphasized. The characterization of an individual-
based model following a Ricker map is, as far as we know, the first trial to model a
simple system of nonoverlapping generations by means of the coupling of contin-
uous and discrete events working at different timescales, giving rise to a perfectly
drawn Ricker dynamics with white noise. The application of control over this sys-
tem seems sufficiently plausible to have in mind that chaos control is effectively
allowed in some real ecosystems.

ACKNOWLEDGEMENTS

The authors thank D. Alonso, J. Bascompte, B. Luque and I. Salazar-Ciudad for


interesting discussions and suggestions. This work has been partially supported by
grant DGYCIT PB97-0693.

R EFERENCES

Allen, J. C., W. M. Schaffer and D. Rosko (1993). Chaos reduces species extinction by
amplifying local population noise. Nature 364, 229–232.
Astakhov, V. V., V. S. Ansihehenko and A. V. Shabunin (1995). Controlling spatiotemporal
chaos in a chain of coupled logistic maps. IEEE Trans. Circuits Syst. 1 42, 352–357.
Bascompte, J. and R. V. Solé (1995). Rethinking complexity: modelling spatiotemporal
dynamics in ecology. Trends Ecol. Evol. 10, 361–366.
Berryman, A. A. and J. A. Millstein (1989). Are ecological systems chaotic—and if not,
why not? Trends Ecol. Evol. 4, 26–28.
Constantino, R. F., J. M. Cushing, B. Dennis and R. A. Desharnais (1995). Experimentally
induced transitions in the dynamic behaviour of insect populations. Nature 375, 227–
230.
Constantino, R. F., R. A. Desharnais, J. M. Cushing and B. Dennis (1997). Chaotic dynam-
ics in an insect populations. Science 275, 389–391.
Dennis, B., R. A. Desharnais, J. M. Cushing and R. F. Constantino (1997). Transitions in
population dynamics: equilibria to periodic cycles to aperiodic cycles. J. Anim. Ecol.
66, 704–729.
Ditto, W. L., S. N. Rauseo and M. L. Spano (1990). Experimental control of chaos. Phys.
Rev. Lett. 65, 3211–3214.
1206 R. V. Solé et al.

Ditto, W. L., M. L. Spano and J. F. Lindner (1995). Techniques for the control of chaos.
Physica D 86, 198–211.
Doebeli, M. (1993). The evolutionary advantage of controlled chaos. Proc. R. Soc. Lond.
B 254, 281–285.
Doebeli, M. and G. D. Ruxton (1997). Controlling spatiotemporal chaos in metapopula-
tions with long-range dispersal. Bull. Math. Biol. 59, 497–515.
Fox, R. F. and J. E. Keitzer (1990). Effect of molecular fluctuations on the description of
chaos by microvariable equations. Phys. Rev. Lett. 64, 249–251.
Garfinkel, A., M. L. Spano, W. L. Ditto and J. N. Weiss (1992). Controlling cardiac chaos.
Science 257, 1230–1235.
Godfray, H. C. J. and S. P. Blythe (1990). Complex dynamics in multispecies communities.
Phil. Trans. R. Soc. Lond. B 330, 221–233.
Güemez, J. and M. A. Matı́as (1993). Control of chaos in uni-dimensional maps. Phys.
Lett. A 181, 29–32.
Hassell, M. P., J. H. Lawton and R. M. May (1976). Patterns of dynamical behavior in
single-species populations. J. Anim. Ecol. 45, 471–486.
Hastings, A. (1993). Complex interactions between dispersal and dynamics, lessons from
coupled logistic equations. Ecology 74, 1362–1372.
Hastings, A. and T. Powell (1991). Chaos in a three-species food chain. Ecology 72, 896–
903.
Judson, O. P. (1994). The rise of the individual-based model in ecology. Trends Ecol. Evol.
9, 9–14.
Kot, M., G. S. Sayler and T. W. Schultz (1992). Complex dynamics in a model microbial
system. Bull. Math. Biol. 54, 619–648.
Lomnicki, A. (1989). Avoiding chaos. Trends Ecol. Evol. 4, 239.
May, R. M. (1974). Biological populations with nonoverlapping generations: stable points,
stable cycles and chaos. Science 186, 645–647.
May, R. M. (1976). Simple mathematical models with very complicated dynamics. Nature
261, 459–467.
May, R. M. and G. F. Oster (1976). Bifurcations and dynamic complexity in simple eco-
logical models. Am. Nat. 110, 573–599.
McCallum, H. I. (1992). Effects of inmigration on chaotic population dynamics. J. Theor.
Biol. 154, 277–284.
McCann, K., A. Hastings and G. R. Huxel (1998). Weak trophic interactions and the bal-
ance of nature. Nature 395, 794–798.
Nisbet, R., S. Blythe, B. Gurney, H. Metz and K. Stokes (1989). Avoiding chaos. Trends
Ecol. Evol. 4, 238–239.
Ott, E., C. Grebogi and J. A. Yorke (1990). Controlling chaos. Phys. Rev. Lett. 64, 1196–
1199.
Parthasarathy, S. and S. Sinha (1995). Controlling chaos in unidimensional maps using
constant feedback. Phys. Rev. E 51, 6239–6242.
Peeters, P. and G. Nicolis (1992). Intrinsic fluctuations in chaotic dynamics. Physica A 188,
426–435.
Controlling Chaos in Ecology 1207

Petrov, V., V. Gáspár, J. Masere and K. Showalter (1993). Controlling chaos in the
Belousov–Zhabotinsky reaction. Nature 361, 240–243.
Pimm, S. L. (1991). The Balance of Nature?: Ecological Issues in the Conservation of
Species and Communities, Chicago: University of Chicago Press.
Ricker, W. E. (1954). Stock and recruitment. Fish. Res. Board. Canada 11, 559–623.
Rinaldi, S., S. Muratori and Y. Kuznetsov (1993). Multiple attractors, catastrophes and
chaos in seasonally perturbed predator–prey communities. Bull. Math. Biol. 55, 15–35.
Ringelberg, J. (1977). Properties of an aquatic microecosystem. II. Steady-state phenom-
ena in the autotrophic subsystem. Helgoländer Meeresunters 30, 134–143.
Rohani, P. and D. J. D. Earn (1997). Chaos in a cup of flour. Trends Ecol. Evol. 12, 171.
Schaffer, W. M. (1984). Stretching and folding in lynx for returns: evidence for a strange
attractor in nature? Am. Nat. 124, 798–820.
Schaffer, W. M. and M. Kot (1986). Chaos in ecological systems: the coals that Newcas-
tle forgot. Trends Ecol. Evol. 1, 58–63.
Scheffer, M. (1991). Should we expect strange attractors behind plankton dynamics-and if
so, should we bother? J. Plankton. Res. 13, 1291–1305.
Schiff, S. J., K. Jerger, D. H. Duong, T. Chang, M. L. Spano and W. L. Ditto (1994).
Controlling chaos in the brain. Nature 370, 615.
Solé, R. V. and J. G. P. Gamarra (1998). Chaos, dispersal and extinction in coupled ecosys-
tems. J. Theor. Biol. 193, 539–541.
Solé, R. V. and L. Menéndez de la Prida (1995). Controlling chaos in discrete neural net-
works. Phys. Lett. A 199, 65–69.
Solé, R. V. and J. Valls (1992). Nonlinear phenomena and chaos in a Monte Carlo simu-
lated microbial ecosystem. Bull. Math. Biol. 54, 939–955.
Stone, L. (1993). Period-doubling reversals and chaos in simple ecological models. Nature
365, 617–620.
Tilman, D. and D. Wedin (1991). Oscillations and chaos in the dynamics of a peren-
nial grass. Nature 353, 653–655.
Wilson, W. G. (1998). Resolving discrepancies between deterministic population models
and individual-based simulations. Am. Nat. 151, 116–134.

Received 2 May 1999 and accepted 12 July 1999

View publication stats

You might also like