You are on page 1of 14

Lars Crawford, Thomas Yu 3/3/14 Food Web Dynamics: Chaos and Complexity Introduction: Ecological food webs are

extremely complex but common systems. In order to try and grasp the dynamics of such a system, one must start simply. The Lotka-Volterra model attempted this with a simple two species food chain whereby a predator species consumed a prey species. Such a model was an excellent introduction to the topic but unrealistic in its simplicity. The model demonstrated that with only two species in the food chain, only two long term results/steady states could be expected: 1) oscillations around a limit cycle corresponding to fluctuations in predator/prey population sizes, and 2) extinction of both species. To expand on this model and answer questions about how population dynamics become more complex as additional species are incorporated, Hastings and Powell (1991) generated a new continuous time model including a third species that sits above the other two in a feed-forward manner: X Y Z. Also incorporated in the model are non-linear functional responses that relate the trophic levels to one another. Though a third species, Z, was included, the model is still considerably unrealistic in its simplicity, yet it begins to hint at the underlying complexity inherent in natural food webs. With this additional species, the dynamics of all three populations once again reach similar long term states: a steady state or limit cycle. However, the limit cycle dynamics are considerably more

complex with many peaks and troughs for populations X and Y that sit on top of larger oscillations dependent on Z. Moreover, the model becomes exquisitely sensitive to initial conditions: only a slight change, even by merely 0.01, cause the solutions to diverge after about 100 time steps. This result gives rise to the concept of chaotic dynamics whereby slight changes in initial conditions result in drastically different end results and demonstrates that such dynamics are likely present in natural environments where food webs are orders of magnitude more complex. Modelling: Through the extension of the Lotka-Volterra model to three species and by incorporating a Holling type II saturating functional responses, Hastings and Powell designed these model differential equations that describe the population dynamics of the respective species:

( ( ) ( )

) ( )

( )

with the functional response: ( )

In these differential equations there are a total of 10 constants: R0, K0, R0, C1, C2, D1, D2, A1, B1 A2, B2. R0 is the intrinsic growth rate and K0 is the carrying capacity of species X; C1 and C2

are conversion rates of prey to predator for species Y and Z, respectively; D 1 and D2 are death rates for species Y and Z, respectively; and A1, B1, A2 and B2 are saturation parameters for the type II response. Parameter B is actually A*h, where h is the amount of time it takes to consume the prey and A is the amount of time it takes to attack/find/kill the prey. To limit the number of parameters, Hastings & Powell non-dimensionalized the equations:

By substituting these new variables into the first system of equations, the new system of equations becomes:

( ( ) ( )

( ) ( )

with the new functional response: ( )

Hastings and Powell carefully selected the following parameter values in an attempt to ensure biological plausibility in their model: a1 = 5.0, b1=3.0, a2=0.1, b2=2.0, d1=0.4, d2=0.01. The argument for such values is as follows: a1 > b1 indicates that low trophic level organisms take only a short time to consume food relative to the time taken to attack/find it, essentially in a

near constant state of consumption; a2 < b2 as higher trophic level organisms eat far less often than it takes for them to attack their prey, perhaps once a week; and finally d 1 > d2 as lower trophic level organism generally have shorter lifespans than those in higher trophic levels. To further augment the simple food chain model proposed by Lotka and Volterra, we incorporated an additional component to the food chain designed by Hastings and Powell. Instead of simply Y consuming X and Z consuming Y, we closed the loop to allow X to consume Z. As such, another functional response was added to both dx/dt and dz/dt. Such an adjustment to the model was not made arbitrarily; we sought to model an even more complex food web that could occur in nature. For example, such a food web could occur if X is a parasite/low trophic level organism synergy. The parasite would have no effect on species X (organism), or on species Y when it consumes X (organism/parasite), but can actively infect Z once it consumes Y. This infection by X (parasite) on Z constitutes the consumption of X (parasite) by Z thereby killing Z. To close the loop, species X (organism) could be one that feeds on the dead Z species (killed by the parasite) thereby consuming the parasite and producing more organism/parasite synergies (increasing X population size) to be consumed by Y. Thus, the new equations become:

( ( ) ( )

( ) ( ) ( )

( )

Results: Reproducing the results found in Hastings & Powell was quite straight forward. We implemented the differential equations in Matlab using a stiff ode solver which utilizes differential step sizes to account for the varying oscillatory time scales. In plotting we used the following initial conditions based on qualitative observation of the plots in Hastings & Powell (1991): x = 0.72, y = 0.16, z = 9.90. Figure 1 illustrates the complex oscillatory dynamics of the three species, feed forward food chain; the many peaks and troughs in the
Figure 1: Reproduction of complex oscillatory dynamics in Hastings & Powell (1991) of simple, three species, feed forward food chain.

populations of species X and Y occur as they oscillate with respect to one another and ride on top of much slower oscillations dependent on species Z. Figure 2 shows the 3 dimensional phase plot produced by the model. The teacup shape of the plot is indicative of the 2 oscillatory time scales for X vs. Y and Y vs. Z. One can also get a glimpse at the chaotic dynamics in the system when observing the handle of the teacup phase plot: the values there are quite close to one another and yet the oscillation that results from each of them diverges considerably.
Figure 2: Three dimensional phase plot.

To add to the model we went about calculating the fixed point by setting each of the differential equations equal to 0:

[(

Automatically, the fixed points (x,y,z)=(0,0,0) can be found, however, this is not useful or interesting in real life situations if all the species were not present in the first place/extinct! However, looking at the equations, by dividing x, y, z in the first, second, and third equation respectively, the term that is left in each equation must be 0: ( )

Because a1, b1, a2, b2, d1, d2 are known, x, y, and z can be solved for: x* =0.819245, y* = 0.1250, and z* = 9.8082 through simple algebra. When the initial conditions are set to the fixed points, the plots of x, y, and z stay around the initial conditions with no overt oscillations. However, when the initial conditions are perturbed slightly from the fixed point, the system returns to its original oscillatory dynamics. Figure 3 depicts the system at its steady state values with small initial oscillations that die out quickly and lead to stable values for all 3 populations.

Similar to Hastings & Powell we altered the initial conditions from the steady state ever so slightly, from 0.819245 to 0.82, and observed oscillations. Once again, this is a clear demonstration of a chaotic system, whereby slight changes in initial conditions produce massive changes in the solution. To complement the plots in Figure 3, we conducted a linear stability analysis by determining the Jacobian matrix of the partials of each equation with respect to each variable and plugging in the fixed point to determine the stability of the system at this point. (( J=
( ) (
Figure 3: Hasting & Powel (1991) with fixed point as initial conditions.

)
( ) )

@(0.81945, 0.1250,9.8082) The eigenvalues are then calculated at this point and one negative and two imaginary eigenvalues are found: 1 = -0.6112; 2 = 0.0387 + 0.748i; 3 = 0.0387 0.748i. This shows that the system is stable at this point as there is a one negative real part and oscillates ever so slightly as there are minute imaginary eigenvalues. As long as initial conditions dont lead to any species going extinct, it will cause oscillations and will produce a negative real part

eigenvalue with two imaginary values. Most other initial conditions will cause close to no oscillations and will actually cause one species to die out before an oscillation can occur. As stated in the modelling section above, we also wanted to add to the model by imagining a slightly more complex food chain that was cyclical in nature rather than merely feed forward. We implemented this by adding an additional functional response to species Z and X as explained above. The results of our modeling efforts are depicted in Figure 4. Though the equations for the cyclical system become more complex, the actual solution to the system is quite a bit simpler relative to the
Figure 4: Simple oscillatory dynamics of cyclical food chain model.

original model. The system oscillates at a single time scale, one population peaking after the next as each is dependent on the population below it and above it which can be seen in Figure 5, the phase plot observed from X vs Z, X vs Y, and Y vs Z. This result is qualitatively similar to that of the genetic represilliator system. For the additional functional response f3(z) added to dx/dt and dz/dt we used values of a3 = b3 = 1.0. We chose these values as we assumed that the attack rate and consumption rate of the parasite X on Z were proportional to one another (a = h). That is, the moment that Z consumes Y, X is able to infect
Figure 5: Cyclical food chain model phase plots.

(attack) Z which allows for these simple oscillatory dynamics. Upon changing a3 to 2.0 and b3 to 4.0 (a = h = 2) we observed similar dynamics which we took as evidence for a relatively accurate attack/consumption rate assumption. Similar to finding fixed points for the Hastings and Powell model, fixed points for our extended model can be found to by setting all differential equations to be 0:

[(

The problem with this scenario is that there is no specific fixed point. Fixed points change due to a change in x, y, or z as x, y, and z, rely on each other. Furthermore, qualitatively, there is location along the population dynamics plots where dx/dt=dy/dt=dz/dt =0. By adding a third functional response and making the food chain cyclical, one should expect that the population of each species will oscillate one after another. Conclusion: Through the reproduction and manipulation of Hastings and Powells three species food chain model it is clear that even relatively simple interspecies interactions within an ecosystem give rise to complex population dynamics. Also, the saturating functional response in these simulations has biological significance and has been used by a variety of other food chain modeling studies. However, although the conclusions drawn from our and Hastings & Powells

simulations are indicative of chaotic dynamics in natural food webs, such idealized and simplistic models hardly scratch the surface of the true complexity inherent in ecological systems. Natural food webs consist of hundreds of species interacting in ways that go far beyond simple predator-prey dynamics as well as those in the functional response used in this papers simulations. A more accurate model must not only consist of interactions between vast numbers of species but also take into account other ecological phenomena including agedependent interactions, omnivory, immigration/emigration etc. Incorporation of such phenomena clearly causes a simple food forward trophic level scheme to be obsolete. A more recent paper written by Polis et al. (1996) discusses a few of these biological phenomena, including the addition of parasitic species such as with our model, and their effects on an ecological food web. Instead of modeling food web dynamics with simple trophic level schemata for which population sizes and feed forward dynamics constitute a basis, Polis et al. suggest a more comprehensive theoretical model. Their model takes into account a variety of different phenomena including those discussed above by considering energy and nutrient transfer across the trophic spectrum. This trophic spectrum consists of multiple species that interact across idealized trophic levels as well as species responsible for decay, migration etc. At the end of their paper Polis et al. call upon experimentalists to design experiments to record parameters from existing ecological systems and the computationally oriented to make attempts at models of such a system. Perhaps when these efforts are carried out we will be able to observe true complexity and chaos in a model system that we have caught a glimpse of through the simple simulations conducted in this paper and in Hastings & Powell (1991).

Citation:
Hastings, A., & Powell, T. (June 01, 1991). Chaos in a Three-Species Food Chain.Ecology, 72, 3, 896-903.

Polis, G. A., & Strong, D. R. (January 01, 1996). Food web complexity and community dynamics. American

Naturalist, 147, 5, 813-846. <http://www.jstor.org.offcampus.lib.washington.edu/stable/2463091?seq=1>

Appendix: File 1:
%% CODE FOR AMATH 422 PROJECT %% CHAOS IN A 3 SPECIES FOOD CHAIN % Replication of Results in Hastings & Powell (1991) utilizing feedforward % food chain of three species. %clear all; clc;

% Parameters a1 = 5.0; b1 = 3.0; a2 = 0.1; b2 = 2.0; d1 = 0.4; d2 = 0.01; % fixed points: plug values in below to observe stable dynamics xfix = 0.819245; yfix = 0.125; zfix = 9.8082; % solves with initial conditions [T,Y] = ode15s(@fcsolver,[0 10000], [ 0.72, 0.16, 9.9]); t = T; x = Y(:,1); y = Y(:,2); z = Y(:,3); figure(1) subplot(3,1,1) plot(t(end/2:end/1.5),x(end/2:end/1.5)) xlabel('time') ylabel('x') title('Chaotic Dynamics in Hastings & Powell Model')

subplot(3,1,2) plot(t(end/2:end/1.5),y(end/2:end/1.5)) xlabel('time') ylabel('y') subplot(3,1,3) plot(t(end/2:end/1.5),z(end/2:end/1.5)) xlabel('time') ylabel('z') figure(2) plot3(x(end/2:end),y(end/2:end),z(end/2:end)) xlabel('x') ylabel('y') zlabel('z')

File 2:
function [dy] = fcsolver( t,y ) % Set up differential equations for ode solver %y(1) = x %y(2) = y %y(3) = z % Parameters a1 = 5.0; b1 = 3.0; a2 = .1; b2 = 2.0; d1 = 0.4; d2 = 0.01; dy = zeros(3,1); dy(1) = y(1) * (1 - y(1)) - (a1*y(1)/(1+b1*y(1))) * y(2); % dx/dt dy(2) = (a1*y(1)/(1+b1*y(1))) * y(2) - (a2*y(2)/(1+b2*y(2))) * y(3) - d1 * y(2); % dy/dt dy(3) = (a2*y(2)/(1+b2*y(2))) * y(3) - d2 * y(3); % dz/dt end

File 3:
%% Calculates Eigen Values from Jacobian for LSA AMATH 422 % HASTINGs & POWELL MODEL % fixed point and jacobian matrix calculated by hand xfix = 0.819245; yfix = 0.125; zfix = 9.8082;

jacobian = [(-a1*yfix/(1+b1*xfix) + xfix*(a1*b1*yfix/(1+b1*xfix)^2 - 2) + 1), -a1*xfix/(1+b1*xfix), 0;... a1*yfix/(1+b1*xfix)^2, (a1*xfix/(1+b1*xfix) a2*zfix/(1+b2*yfix)^2 - d1), -a2*yfix/(1+b2*yfix);... 0, a2*zfix/(1+b2*yfix)^2, (a2*yfix/(1+b2*yfix) -d2)]; eigs1 = eig(jacobian)

File 4:
%% CODE FOR AMATH 422 PROJECT %% CHAOS IN A 3 SPECIES FOOD CHAIN, CYLICAL MODEL % The model was modified to make the food chain cyclical instead of merely % feed forward in nature. This was done by adding a f(z;a,b)*x term to the % dx/dt equation. An example of such a food chain in a biological ecosystem % is one where x represents the combination of a host species on a low trophic % level as well as a parasitic species that only physically effects species z. % As x is consumed the population of the host organism is reduced and the % parasite is inactive until y is consumed by z whereby the parasite x now % infects/feeds on z thus producing more parasite/host pairs of x. %clear all; clc;

% Parameters a1 = 5.0; b1 = 3.0; a2 = 0.1; b2 = 2.0; a3 = 1.0; b3 = 1.0; d1 = 0.4; d2 = 0.01; [T,Y] = ode15s(@fcsolver2,[0 10000], [ 0.721, 0.16, 9.9]); t = T; x = Y(:,1); y = Y(:,2); z = Y(:,3); figure(1) subplot(3,1,1) plot(t(end/2:end/1.9),x(end/2:end/1.9)) xlabel('time') ylabel('x') title('Chaotic Dynamics in Hastings & Powell Model') subplot(3,1,2) plot(t(end/2:end/1.9),y(end/2:end/1.9)) xlabel('time') ylabel('y')

subplot(3,1,3) plot(t(end/2:end/1.9),z(end/2:end/1.9)) xlabel('time') ylabel('z') figure(2) subplot(3,1,1) plot(x(end/2:end/1.9),y(end/2:end/1.9)) xlabel('x') ylabel('y') subplot(3,1,2) plot(x(end/2:end/1.9),z(end/2:end/1.9)) xlabel('x') ylabel('z') subplot(3,1,3) plot(y(end/2:end/1.9),z(end/2:end/1.9)) xlabel('y') ylabel('z')

File 5:
function [dy] = fcsolver2( t,y ) % Set up differential equations for ode solver %y(1) = x %y(2) = y %y(3) = z % Parameters a1 = 5.0; b1 = 3.0; a2 = 0.1; b2 = 2.0; a3 = 1.0; b3 = 1.0; d1 = 0.4; d2 = 0.01; dy = zeros(3,1); dy(1) = y(1) * (1 - y(1)) - (a1*y(1)/(1+b1*y(1))) * y(2) + (a3*y(3)/(1+b3*y(3))) * y(1); % dx/dt dy(2) = (a1*y(1)/(1+b1*y(1))) * y(2) - (a2*y(2)/(1+b2*y(2))) * y(3) - d1 * y(2); % dy/dt dy(3) = (a2*y(2)/(1+b2*y(2))) * y(3) - d2 * y(3); % dz/dt end

You might also like