Sei sulla pagina 1di 144

Quantitative Population Ecology

A. Sharov, Dept. of Entomology, Virginia Tech, Blacksburg, USA.

http://www.ento.vt.edu/~sharov/PopEcol/#mark7

Lecture Handouts

1. Introduction: Population systems and their components.

1.1. What is population ecology?


1.2. Models as analytical tools
1.3. Population system
1.4. Petri nets (optional)
1.5. Questions and Assignments

2. Estimation of population density and size.

2.1. Censusing a Whole Population


2.2. Simple Random or Systematic Sampling
2.3. How Many Samples?
2.4. Elements of Geostatistics
2.5. Stratified Sampling
2.6. Capture-Recapture and Removal Methods
2.7. Indirect Measures of Population Density
2.8. Questions and Assignments

3. Spatial distribution of organisms.

3.1. Tree Types of Spatial Distribution


3.2. Random Distribution
3.3. Aggregated Spatial Distribution
3.4. Indexes of Aggregation
3.5. Density-Invariant Indexes of Aggregation
3.6. Geostatistical Analysis of Population Distribution
3.7. Fractal Dimension of Population Distribution
3.8. Questions and Assignments

4. Statistical analysis of population dynamics.

4.1. Correlation between population density and various factors


4.2. Correlation between factors
4.3. Example: Colored fox in Labrador (1839-1880)
4.4. Autocorrelation of factors and model validation
4.5. Stochastic models based on regression
4.6. Biological interpretation of stochastic models. Response surfaces

5. Reproducing populations: exponential and logistic growth.

5.1 Exponential model


5.2. Logistic model
5.3. Discrete-time analogs of the exponential and logistic models
5.4. Questions and assignments

6. Life-tables, k-values.

6.1. Age-dependent life-tables


6.2. Stage-dependent life-tables
6.3. Questions and assignments

7. Model of Leslie.

7.1. Model structure


7.2. Model behavior
7.3. Intrinsic rate of population increase
7.4. Stable age distribution
7.5. Modifications of the Leslie model

8. Development of poikilothermous organisms, degree-days.

8.1 Rate of development


8.2 Simple degree-day model
8.3 How to measure temperature?
8.4 Improved degree-day model
8.5 Other non-linear models of development
8.6 Physiological time
8.7 How to combine physiological time with the model of Leslie?
8.8 Questions and Assignments

9. Stability, oscillations and chaos in population dynamics.

9.1. Introduction
9.2. Attractors and Their Types
9.3. Equilibrium: Stable or Unstable?
9.4. Quantitative Measures of Stability
9.5. Limit Cycles and Chaos
9.6. Questions and Assignments

10. Predators, Parasites, and Pathogens.

10.1. Introduction
10.2. Lotka-Volterra Model
10.3. Functional and Numerical Response
10.4. Predator-Prey Model with Functional Response
10.5. Host-Parasitoid Models
10.6. Host-Pathogen Model (Anderson & May)
10.7. Questions and Assignments

11. Competition and Cooperation.

11.1. Intra-specific competition


11.2. Competition between species
11.3. Ecological niche
11.4. Cooperation

12. Dispersal and spatial dynamics.

12.1. Random Walk


12.2. Diffusion Models
12.3. Stratified Dispersal
12.4. Metapopulation Models

13. Population outbreaks.

13.1. Ecological mechanisms of outbreaks


13.2. A model of an outbreak
13.3. Catastrophe theory
13.4. Classification of outbreaks
13.5. Synchronization of outbreaks in space

Labs

1. Orientation in software: Microsoft Excel.


2. How to write a scientific paper
3. Population sampling and spatial distribution.
4. Statistical analysis of population change.
5. Model of Leslie.
6. Development of poikilothermous organisms.
7. Model of Ricker: stability, oscillations, chaos.
8. Parasitism and biological control

Statistical tables

1. t-statistics
2. Chi-square statistics
3. F-statistics, P=0.05
4. F-statistics, P=0.01
5. F-statistics, P=0.001
Lecture 1. Introduction.
Population Systems and Their Components
1.1. What is Population Ecology?

Population ecology relative to other ecological disciplines

• Population ecology is the branch of ecology that studies the structure and
dynamics of populations.
• Physiology studies individual characteristics and individual processes. These are
use as a basis for prediction of processes at the population level.
• Community ecology studies the structure and dynamics of animal and plant
communities. Population ecology provides modeling tools that can be used for
predicting community structure and dynamics.
• Population genetics studies gene frequencies and microevolution in
populations. Selective advantages depend on the success of organisms in their
survival, reproduction and competition. And these processes are studied in
population ecology. Population ecology and population genetics are often
considered together and called "population biology". Evolutionary ecology is
one of the major topics in population biology.
• Systems ecology is a relatively new ecological discipline which studies
interaction of human population with environment. One of the major concepts
are optimization of ecosystem exploitation and sustainable ecosystem
management.
• Landscape ecology is also a relatively new area in ecology. It studies regional
large-scale ecosystems with the aid of computer-based geographic information
systems. Population dynamics can be studied at the landscape level, and this is
the link between landscape- and population ecology.

The term "population" is interpreted differently in various sciences:

• In human demography a population is a set of humans in a given area.


• In genetics a population is a group of interbreeding individuals of the same
species, which is isolated from other groups.
• In population ecology a population is a group of individuals of the same species
inhabiting the same area.

Interbreeding is seldom considered in ecological studies of populations. The exceptions


are studies in population genetics and evolutionary ecology.

Populations can be defined at various spatial scales. Local populations can occupy very
small habitat patches like a puddle. A set of local populations connected by dispersing
individuals is called a metapopulation. Populations can be considered at a scale of
regions, islands, continents or seas. Even the entire species can be viewed as a
population.

Populations differ in their stability. Some of them are stable for thousands of years.
Other populations persist only because of continuous immigration from other areas. On
small islands, populations often get extinct, but then these islands can be re-colonized.
Finally, there are temporary populations that consist of organisms at a particular stage
intheir life cycle. For example, larvae of dragonflies live in the water and form a
hemipopulation (term of Beklemishev).

The major problem in population ecology is to derive population characteristics from


characteristics of individuals and to derive population processes from the processes in
individual organisms:

Main axiom of population ecology: organisms in a population are ecologically


equivalent. Ecological equivalency means:

1. Organisms undergo the same life-cycle


2. Organisms in a particular stage of the life-cycle are involved in the same set of
ecological processes
3. The rates of these processes (or the probabilities of ecological events) are
basically the same if organisms are put into the same environment (however
some individual variation may be allowed).

1.2. Models as analytical tools


Population ecology is the most formalized area in biology.
Model is a tool and should never be considered an ultimate goal in ecological studies.
Model and reality are linked together by two procedures: abstraction and interpretation:
Abstraction means generalization: taking the most important components of real
systems and ignoring less important components. Importance is evaluated by the
relative effect of system components on its dynamics. For example, if we found that
parasitism rate in insect pest is always below 5%, then parasitoids can be excluded from
the model.

Interpretation means that model components (parameters, variables) and model behavior
can be related to components, characteristics, and behavior of real systems. If model
parameters have no interpretation, then they cannot be measured in real systems.

Most field ecologists are not good at abstraction. If they build a model they often try to
incorporate every detail. Most mathematicians are not good at interpretation of their
models. Usually they think of clean models and dirty reality. However, both abstraction
and interpretation are necessary for successful modeling. Thus, close collaboration
between ecologists and mathematicians is very important.

Models are always wrong ... but many of them are useful.
How it may happen that the wrong model can give a correct answer? In the same way as
old maps, which assumed a flat earth and used wrong distance relations, where useful
for travelers in the past.

Modeling strategy:

1. Select optimal level of complexity


2. Never plan model development for more than 1 year
3. Avoid the temptation to incorporate all available information into the model
4. Follow specific objectives, don't try to make a universal model
5. If possible, incorporate already existing models

System properties and model properties


1. Many system properties are not represented in the model.
Example: age structure is not represented in both exponential and logistic
models.
2. Some model properties cannot be found in real systems.
Example: solutions of differential equations are always smooth, while
trajectories of real systems are always noisy.
Example of a wrong question: Does this population have an equilibrium density?
The stable equilibrium is a state to which all trajectories of the system converge
infinitely close with increasing time. The model (e.g. the differential equation) may
have an equilibrium density, but real populations don't have it because:
1. Population density cannot be measured with infinite accuracy.
2. Weather fluctuations always add noise to system's dynamics.
3. Time series are never long enough to talk about limits and convergence.

1.3. Population system


Population system (or a life-system) is a population with its effective environment.

The term "life-system" was introduced by Clark et al. (1967). Later, Berryman (1981)
suggested another term "population system" which is definitely better. A short review of
the life-system methodology was published by Sharov (1991).

Major components of a population system

1. Population itself. Organisms in the population can be subdivided into groups


according to their age, stage, sex, and other characteristics.
2. Resources: food, shelters, nesting places, space, etc.
3. Enemies: predators, parasites, pathogens, etc.
4. Environment: air (water, soil) temperature, composition; variability of these
characteristics in time and space.

Temporal and spatial structure of a population system


Temporal structure Spatial structure

Diurnal cycles Spatial distribution


Seasonal cycles Habitat structure
Long-term cycles Metapopulations

Dynamics of Population Systems


Factor-Effect concept: Environmental factors affect population density. When factors
change then population density changes respectively.
Advantage: Causal explanation of population change
Disadvantage: It is useful for one-step instant effects but becomes confusing if the effect
is simultaneously a factor that causes soemthing else, or if there are time delays in the
effect.

Factor-Process concept: Environmental factors do not affect population density


directly, instead they affect the rate of various ecological processes (mortality,
reproduction, etc.) which finally result in a change of population density. Processes can
change the value of factors; thus, feedback loops become possible.
Advantage: Can handle complex dynamic interactions among components of population
systems including time-delays, negative and positive feedbacks.

In 50-s and 60-s there was a discussion about population regulation between two
schools in population ecology. An agreement could not be reached because these
schools used different concepts of population dynamics. Andrewartha and Birch (1954)
used the factor-effect concepts whereas Nicholson (1957) used the factor-process
concept.
The factor-process concept works not only in population ecology but in any kind of
dynamic systems, e.g. in economic systems. Forrester (1961) formalized the factor-
process concept and applied it to industrial dynamics. Later this formalism became very
popular in ecology and is widely used especially in ecosystem modeling.

The model of Forrester is based on tank-pipe analogy. The system is considered as a set
of tanks connected by pipes with vents which can regulate the "flow" of liquid from one
tank to another. The flow of "liquid" between tanks is considered as "material flow".
However there is also "information flow" that regulates the vents. Vents are equivalent
to processes; and the amount of liquid in a tank is a variable or a factor because it can
affect processes via information flow.

The figure above is the Forrester diagram for an insect population system with 4 stages:
eggs, larvae, pupae, and adults. Transition between these states (development) is
regulated by temperature. Influx of eggs depends on the number of adults. Mortality
occurs in all stages of development. Larval and pupal mortality is density-dependent.

Diagrams of Forrester can be easily transformed into differential equation models. Each
process becomes a term in a differential equation that determines the dynamics of the
variable. For example, the number of larvae in the figure above is affected by 3
processes: egg development ED(T), larval development LD(T), and larval mortality
LM(N). Egg and larval development rates are functions of temperature T; whereas
larval mortality is a function of larval numbers N. The equation for larval dynamics is
the following:

dN/dt = ED(T) - LD(T) - LM(N)

Here the term ED(T) is positive because egg development increases the number of
larvae ("liquid" influx). Terms LD(T) and LM(N) are negative because larval
development (molting into pupae) and mortality reduce the numbers of larvae.

Limitations of Forrester's model:

1. Distinction between information and material flows was not clear because
information cannot be transferred without any matter. For example: egg or seed
production is not only an information flow, there is a flow of matter too.
2. Only one type of processes is considered in which "fluid" moves from one tank
to another. This is good representation of organisms changing their state.
However, it is impossible to apply the model to the processes that involve two or
more participants. For example, when a parasite enters a host then it is
impossible to make a "pipe" that starts from two "tanks": host and parasite, and
ends in the "tank" of "parasitized host".
3. Only one level of processes is considered. In some cases it is important to
consider processes at two spatial levels, e.g., the dynamics of phytophagous
insects can be considered within a host plant and within a population of host
plants.

Other models has been developed to represent the factor-process concept. Petri nets
consider interactions of one two, or more participants. However, there is no generic
model with no limitations.

Factors and processes


The dynamics of a system is always viewed as a sequence of states. State is an
abstraction like an arrow hanging in the air but it helps to understand the dynamics.
Because systems are built from components, the state can be represented as a
combination of states of all its components. For example, the state of the predator-prey
system can be characterized by the density of prey (component #1) and density of
predators (component #2). As a result, the state of the system can be considered as a
point in a 2-dimensional space with coordinates that correspond to system components.
This space is usually called the phase space

Components of the system will be called factors because they affect system dynamics.
The state of the component is the value of the factor. Factors are considered with as
many details as necessary for understanding system's dynamics.

Examples of factors:

• In simple models (e.g., exponential or logistic), there is only one factor


(=component): the population itself. Its value is population density.
• In age-structured populations, each age class is a factor and its value is equal to
the density of individuals in that class.
• Natural enemies and various resources can be considered as additional factors.
Then, the abundance of predators, parasites, or food will be values of these
factors.
• Weather can be viewed as a set of factors: temperature, precipitation, etc. Each
of these factors has a numerical value.
• Factors may have a hierarchical structure. If the population occupies multiple
patches in space, then all factors (e.g., population density, density of natural
enemies and resources, temperature) become patch-specific.

Any detectable change in the population system is considered as an event. Events can be
classified according to the components involved in these events. The process can be
defined as a class of identical events. The rate of a process can be measured by the
number of events that occur in the system per unit time. Also, the specific rate of
processes is often used which is the number of events per time unit per one organism
involved in the event.

Examples of processes
• The birth of an organism is an event. The birth rate is the number of births per
unit time. The specific birth rate (= reproduction rate) is the birth rate per 1
female (or per 1 parent).
• Death of an organism is an event. Mortality on a specific stage from a specific
cause (e.g., parasitism, predation, starvation) is a process. Mortality rate is the
number of deaths per unit time. Specific mortality rate is the number of deaths
per unit time per organism.
• Other processes are: growth, development, consumption of resources, dispersal,
entering diapause, etc.

Interaction of factors and processes

Factors affect the rate of processes as shown below:

On the other hand, processes change the values of factors:

A process may be affected by multiple factors. For example, mortality caused by


predation may depend on the prey density, predator density, number of refuges,
temperature (if it changes the activity of organisms), etc.

The value of a factor may change due to multiple processes. For example, the number of
organisms on a specific stage changes due to development (entering and exiting this
stage), dispersal, and mortality due to predation, parasitism, and infection.

Thus, there is no one-to-one correspondence between factors and processes. Life-tables


show the rate of various mortality processes, but they do not show the effect of factors
on these processes. For example, parasitism may be mostly determined by weather;
viral infection may be determined by host plant chemistry. But life-tables do not show
the effect of either weather or and host plant chemistry.

It is dangerous to judge on the role of factors (e.g., biotic vs. abiotic) from life-tables.
For example, a life-table may show that 90% mortality of an insect is caused by
parasitism. This may lead to an erroneous conclusion that parasites rather than weather
are most important in the change of population density. It may appear that the
synchrony between life cycles of the host and parasite depends primarily on weather.

Life tables show various mortality processes in the population but they do not indicate
the role of factors. To analyze the role of factors, it is necessary to vary these factors
experimentally and examine how they affect various mortality processes. These
experiments may show, for example, that the rate of parasitism depends on the density
of parasites, density of hosts, and temperature. To understand population dynamics, it is
necessary to know both the effect of factors on the rate of various processes, and the
effect of processes on various ecological factors (e.g., on population density). This
information is integrated via modeling.

References
Berryman A. A. (1981) Population systems: a general introduction. New York: Plenum
Press.
Clark L. R., P. W. Geier, R. D. Hughes, and R. F. Morris (1967) The ecology of
insect populations. London: Methuen.
Sharov, A. A. 1992. Life-system approach: a system paradigm in population ecology.
Oikos 63: 485-494. Get a reprint! (PDF)

Petri Nets
A more universal factor-process model was developed by Petri (1962). He introduced a
class of nets which later were named "Petri nets". These nets have two types of
components: positions and transitions (positions = factors, transitions = processes).

Example of a Petri net

The dynamics of a Petri net is a sequence of transition "firing". When a transition is


fired then 2 things happen. First, tokens are taken away from positions which have
arrows going from these positions to the transition considered. If more than 1 arrow
goes from position to transition, then the number of tokens removed from that position
is equal to the number of arrows. Second, new tokens are placed on positions indicated
by arrows that originate from the transition. The number of tokens placed corresponds
again to the number of arrows (in the case of multiple arrows).

For example, in the figure above, when transition t1 is fired, then 1 token is removed
from position p1, 1 token is removed from position p2, and 1 token is added on position
p3. When transition t2 is fired, then 1 token is removed from position p3, and 1 token is
added on position p2. Transition t1 can be interpreted as feeding and growth, and
transition t2 as reproduction.

Petri nets can be used to simulate complex ecological interactions within a population
and among populations. The figure below is an example of a population with sexual
reproduction. More examples can be found in Sharov (1991).

Petri net representing a population with sexual reproduction

The rate of processes is not defined within a Petri net, it should be specified separately.
The rate of chemical reactions is usually defined by mass action law. Mass action law is
used in many population ecology models (exponential growth, Lotka-Volterra eqns.).
However, there are much more cases in ecology than in chemistry when mass action
law does not work. In these cases, the rate of equations is defined in some different way.

How to write differential equations from the Petri net:


1. Write left sides of differential equations for each position i: dxi/dt=

2. For each process (transition) t do the following:


a. Estimate process rate Vt = ktxi xj ... where kt is a constant and xi, xj are the number of
tokens in starting positions i, j,... for this process.
b. Subtract Vt from dxi/dt for each starting position i
c. Add Vt to dxi/dt for each ending position i.

Example:

Rates of processes 1 and 2:


Rates of change in token numbers:

Other examples:

V=k

V = kx

Coefficient k depends on the rate of organisms' movement (=temperature). Mass action


law assumes that particles move incessantly. However, organisms are not like gas
molecules. Predators stop hunting after saturation. A similar saturation effect exists in
macromolecules (enzymes). In these cases, mass action law should be replaced by
Michaelis-Menton equation: V = kxy/(1+ax). The rest of the algorithm remains the
same.

Effect of other factors (temperature, age, sex, etc.) can be expressed as variation of
coefficient k. As a result, k will become a function of several variables.

References
Sharov, A.A. 1991. Self-reproducing systems: structure, niche relations and evolution.
BioSystems, 25: 237-249. Get a reprint! (PDF), Short version (HTML)

Questions and Assignments to Lecture 1


1.1. What is the relationship between population ecology and community ecology,
landscape ecology, systems ecology, population genetics, and population biology
(describe common problems and major differences)?

1.2. Can you define what is a "good" model and what is a "bad" model? Be careful in
your definitions: it may happen that the class of "good" models will be empty!

1.3. What is a population and a population system?

1.4. What ecological characteristics of organisms change during the life cycle? Consider
several examples: a bacteria, insect, mammal, and a plant.

1.5. What are ecological differences between males and females? 1.6. Is it possible to
find similarity in variability?
1.7. Describe elements of the temporal and spatial structures in a population system.

1.8. Why factor-effect models are not appropriate for studying population systems?

1.9. Describe a population system that you wish to study in the future (it may be a
natural or a laboratory population). Specify factors and processes and their interactions.
Draw a diagram with factors in rectangles and processes in ovals, show interactions
with arrows.

1.10. Draw a Petri net for a 3-level trophic interactions: an autotrophic population (algae
or plant) consumes several mineral components from soil (or water), the second
population feeds on that autotrophic population, and the third population feeds on the
second one. All three populations should grow in numbers.

1.11. Write differential equations for the following Petri nets assuming that the rate of
all processes correspond to the mass action law:
Lecture 2. Estimation of population density and size
2.1. Censusing a whole population
This method works only if organisms can be easily observed, their numbers are not too
big, and the area is well bounded and is not too large. Examples: trees in a small
isolated forest; all bird nests are censused in the New-York state (see Pielou 1977).

Migrating populations can be counted using aerial photography. This method is used
when the population has seasonal migration. Examples: sandhill cranes and caribou.

2.2. Simple Random or Systematic Sampling


Two possible objectives for sampling:

1. Estimate average population density,


2. Make a map of population density.

Traditionally, random sampling plans were preferred over systematic sampling plans
because random sampling helped to avoid subjective selection of sample locations.
However, systematic sampling has no elements of subjectivity if sample location is
selected prior to examining the area. For example, there is no subjective decisions if we
sample every tenth potato plant and count Colorado potato beetles.

Moreover, systematic sampling has an advantage over random sampling if the number
of samples is large because of more uniform coverage of the entire sampling area. It is
especially important for making population maps. Random sampling can be used if the
objective is to estimate the mean population density and the number of samples is not
large (<100).

Preferential sampling of specific areas (e.g., high-density areas) was always considered
unacceptable. However, modern geostatistical methods and stratified sampling can take
advantage of preferential sampling. This shows that the methodology of sampling
evolves and old textbooks may give obsolete recipes.

Traditional statistical methods include estimation of the mean population density (M),
standard deviation (S.D.), and standard error (S.E.), which is the standard deviation of
the sample mean.

The equation for standard error is derived assuming that all samples are independent.
This is a very strong assumption which is unrealistic in many situations. Samples
separated by small distance are often positively correlated. Before using standard
statistics it is important to test if samples are correlated. Spatial correlations are
examined using geostatistics. The simplest geostatistical test for spatial autocorrelation
is the omnidirectional correlogram:
where z1 and z2 are organism numbers in two samples separated by lag distance h,
summation is performed over all pairs of samples separated by distance h; Nh is the
number of pairs of samples separated by distance h; Mh and sh are the mean and the
standard deviation of samples separated by distance h (each sample is weighted by the
number of pairs of samples in which it is included).

Correlation decreases with distance between samples as shown below.

The range of correlogram is the lag distance h at which correlation reaches (or becomes
close to) zero. Standard statistics can be applied only if inter-sample distance exceeds
the range of the correlogram.

Confidence interval (c.i.) is the interval where the population mean can be found with
probability of (1 - P), where P is error probability (e.g., P = 0.05). The number of
degrees of freedom d.f. = N - 1 (one d.f. goes for estimation of sample mean).

Precision of sample mean is

A = S.E. / M

There is an empirical rule that precision should be below 0.05 (or 0.1). However, this
rule is not universal The only thing that matters in statistics is testing hypotheses. If
null-hypothesis is rejected then it does not matter whether A was above or below 0.05.
However, in each specific research area, it is useful to find a precision level which is
usually sufficient for rejecting null-hypotheses.
Example:. Insect pest population should be suppressed if its density exceeds the
economic injury level (EIL). A null-hypothesis is tested that the average density M is
equal to EIL. If EIL is within the c.i. for M, then the null-hypothesis cannot be rejected
and no decision can be made. In this case, more samples should be taken. If the EIL is
outside of the c.i., then null-hypothesis is rejected, and population is suppressed if M >
EIL, or not suppressed if M < EIL.

2.3. How Many Samples to Take?


There are two major methods for planning the number of samples:

1. two-step sampling, and


2. sequential sampling:

Two-step sampling
The number of samples, N, required to achieve specific accuracy level can be estimated
from equations for standard error (S.E.) and accuracy, A:

where M is sample mean and S.D. is standard deviation. Here the third equation is
derived from the first two equations.

Standard deviation, S.D., is usually not known before sampling. Thus, the first step is to
take N1 samples and to estimate N using the equation above. Then, at the second step,
take N1 = N - N1 samples.

Taking samples in two steps is possible only if population numbers don't change
between two sampling dates.

Sequential sampling
The main idea of sequential sampling is to take samples until some condition (which is
easy to check) is met.

The first example is the sampling plan targeted at achieving specific accuracy. It is
based on the Taylor's power law:

Coefficients a and b can be estimated using linear regression from several pairs of M
and S.D. estimated in different areas with different average population density.
Combining two previous equations we get:

N=
Mean (M) equals to the total number of recorded individuals (S) in all samples divided
by the number of samples (N). Now, we substitute M by S/N, and solve this equation for
S:

Stop-lines for accuracy levels of A = 0.1; 0.07; and 0.05 are plotted below:

The blue line shows the total number of captured individuals in all samples. Sampling
terminates when this line crosses the stop line for selected accuracy level.

The second example is the sequential sampling plan used for decision-making in pest
management. This method was developed by Waters (1955; Forest Sci. 1:68-79). It is
described in Southwood (1978).

Here the blue line again shows the total number of captured individuals in all samples.
While the blue line is between magenta inclined lines, sampling continues. If the blue
line crosses the upper magenta line, then sampling stops and pesticides are applied
against the pest population. If the blue line crosses the lower magenta line, then
sampling stops and pesticides are not applied.

Deriving the solution of this problem it is too complicated. Thus, we will consider the
final result only.

If the population has a negative binomial distribution (see next lecture), then stop lines
correspond to the linear equation:
where:

2.4. Elements of geostatistics


Geostatistics is a collection of statistical methods which were traditionally used in geo-
sciences. These methods describe spatial autocorrelation among sample data and use it
in various types of spatial models. Geostatistical methods were recently adopted in
ecology (landscape ecology) and appeared to be very useful in this new area.

Geostatistics changes the entire methodology of sampling. Traditional sampling


methods don't work with autocorrelated data and therefore, the main purpose of
sampling plans is to avoid spatial correlations. In geostatistics there is no need in
avoiding autocorrelations and sampling becomes less restrictive. Also, geostatistics
changes the emphasis from estimation of averages to mapping of spatially-distributed
populations.

Spatial autocorrelation can be analyzed using correlograms, covariance functions and


variograms (=semivariograms). For simplicity, here we will use correlograms only.
Covariance functions and variograms are discussed in the next lecture.

In brief, geostatistical analysis usually has the following steps:

1. Estimation of correlogram
2. Estimation of parameters of the correlogram model
3. Estimation of the surface (=map) using point kriging, or
4. Estimation of mean values using block kriging
Detailed description of most geostatistical methods can be found in Isaaks and
Srivastava (1989). Here we will discuss only the most important elements of
geostatistics.

Estimation of Correlogram
Correlogram is a function that shows the correlation among sample points separated by
distance h. Correlation usually decreases with distance until it reaches zero.
Correlogram is estimated using equation:

where z1 and z2 are organism numbers in two samples separated by lag distance h,
summation is performed over all pairs of samples separated by distance h; Nh is the
number of pairs of samples separated by distance h; Mh and sh are the mean and the
standard deviation of samples separated by distance h (each sample is weighted by the
number of pairs of samples in which it is included).

Notes:

1. This is an omnidirectional correlogram; "omnidirectional" means that we don't


care about the direction of lag h.
2. Serious geostatistical analysis often includes estimation of directional
correlograms (see next lecture). It may happen that points are more closely
correlated in some direction (e.g., NE-SW) than in other directions. If
correlogram depends on direction, then the spatial pattern is called anisotropic.
If no anisotropy detected, then it is possible to use the omnidirectional
correlogram.
3. Correlogram equation works only if there are no trends in population density in
the study area. If a trend exists, then a non-ergodic correlogram should be used
instead (see next lecture).

Estimation of parameters of correlogram model


The correlogram can be approximated by some mathematical model. Two models are
used most often:

1. Exponential model:

2. Spherical model:

where c1 is sill, and a is range. These parameters can be found using the non-linear
regression.
Estimation of the surface (=map) using point kriging (ordinary
kriging)
The value z'o at unsampled location 0 is estimated as a weighted average of sample
values z2 at locations i around it:

Weights depend on the degree of correlations among sample points and estimated point.
The sum of weights is equal to 1 (this is specific to ordinary kriging):

Weights are estimated individually for each point in a regular spatial grid using the
system of linear equations:

where is the Lagrange parameter; is the correlation between points i and j which is
estimated from the variogram model using distance, h, between points i and j; 0 is the
estimated point; 1,...,n are sample points.

Using matrix notation this system can be re-written as:

The solution of this matrix equation is:

Now, weights are found, and thus, it is possible to estimate the value z'o. When these
values are estimated for all points in a regular grid, then we get a surface of population
density.

The variance of local estimation is equal to:


Estimation of the mean value using block kriging
The only difference of block kriging from point kriging is that estimated point (0) is
replaced by a block. Consequently, the matrix equation includes "point-to-block"
correlations:

Point-to-block correlation is the average correlation between sampled point i and all
points within the block (in practice, a regular grid of points within the block is used, as
shown in the figure).

The variance of block estimation is equal to:

Where is the average correlation within the block (average correlation between all
pairs of grid nodes within the block).

Advantages of kriging:

• It handles spatial autocorrelation


• It is not sensitive to preferential sampling in specific areas
• It estimates both: local population densities and block averages.
• It can replace stratified sampling if the size of aggregations is larger than the
inter-sample distance.

References
Isaaks, E. H. and R. M. Srivastava. 1989. An Introduction to Applied Geostatistics.
Oxford Univ. Press, New York, Oxford. (a very good introductory textbook)
Deutsch, C. V. and A. G. Journel. 1992. GSLIB. Geostatistical software library and
user's guide. Oxford Univ. Press, Oxford. (software code written in FORTRAN; I have
translated a portion of this library into the C-language).

Geostatistics in ecology (review papers):


Rossi, R. E., D. J. Mulla, A. G. Journel and E. H. Franz. 1992. Geostatistical tools
for modelling and interpreting ecological spatial dependence. Ecol. Monogr. 62: 277-
314.
Liebhold, A. M., R. E. Rossi and W. P. Kemp. 1993. Geostatistics and geographic
information systems in applied insect ecology. Annu. Rev. Entomol. 38: 303-327.
2.4. Stratified sampling
Stratified sampling is used if sampled area (or volume) is heterogeneous (Pielou, p.107).
If patch size is much larger than inter-sample distance, then kriging can be used instead
of stratified sampling.

In stratified sampling program, the area (volume) is subdivided into 2 or more portions
which are sampled separately. Example: pine sawflies prefer to spin their cocoons close
to the tree; thus, the area adjacent to trees (within 1 m radius) can be sampled separately
from the rest of the area.

1. Density of samples in each stratum should be proportional to the variance of


organism counts in the stratum -- in this case the maximum precision is reached.
Taylor's power law can be used to predict variance from the mean. Variance usually
increases with mean, and thus, the stratum with higher organism density should be
sampled more intensively.

2. The mean population density, M, is estimated as a weighted mean of mean densities,


Mi, in each stratum, i, with weights, wi equal to the area covered by stratum i:

3. The standard error, SE, of the mean is equal to

where SEi is the standard error for the mean in stratum i.

2.6. Capture-recapture and Removal Methods


Capture-Recapture Method
Suppose the population is of size N, so that N in the number we wish to estimate.
Suppose, M organisms were captured, marked (or tagged) and released back into the
population. After some time that should be sufficient for organisms to mix, n organisms
were captured, and m of these appeared to be were marked. The proportion of
recaptured organisms is assumed to be the same as the proportion of marked organisms:

Population size can be found as: N = nM/m (the Lincoln Index).

The following conditions should be met:

1. No immigration, emigration, births or deaths between the release and recapture


times.
2. The probabilities of being caught are equal for all individuals (including marked
ones).
3. Marks (or tags) are not lost and are always recognizable.

The first two conditions are often non-realistic, and thus, several modifications of this
method has been developed that loosen these conditions. Perhaps the most popular is
the Jolly-Seber method which requires capturing and marking of animals at regular time
intervals. Animals, marked and released each time, should have different marks so that
it is possible to distinguish between individuals marked on different dates. Algorithm is
given in Southwood (1978).

Jolly-Seber method gives an estimate of population size on each specific date; the first
condition can be violated. However, the second condition is still required. It is also
possible to estimate mortality+emigration rate and birth+immigration rate on each
specific day. These rates are assumed to be constent for all individuals (including
marked individuals).

There are numerous other models for capture-recapture experiments, which are specific
for a particular population. For example, age structure of the population may be
important, or some individuals may have higher probability to be caught than others.

Another problem arises if the population has no boundaries. In this case, a grid of traps
can be established, and only the central portion of the grid is used for analysis (because
traps near the edges may be influenced by migration). The area covered by the grid
should be much larger than the average distance of animal dispersal.

Because the biology of different species is variable, it may be necessary to modify the
capture-recapture model.

Removal method
Removal method is based on intensive trapping of animals in an isolated area.
Migration is prevented by some kind of barriers. It is assumed that there are no births or
natural deaths of organisms. The proportion of animals captured each day is the same.
Therefore, population numbers and the number of captured individuals declines
exponentially:

The model of removal: dN/dt = -aN, where a is the removal rate.

The solutionof this differential equation is N = Noexp(-at), where No are initial


numbers.
Then the number of animals captured per unit time is equal to A(t) = aNoexp(-at)
Parameters a and No can be estimated using the non-linear regression.

Pielou (p. 127) used a different method for estimating parameters which is based on the
analysis of 2 first time intervals only. For example, if captures in the first 2 time
intervals were 29 and 18 animals, then

a = (29-18)/29 = 0.38.

This model can be generalized assuming the recruitment of organisms (e.g., emergence
of adult insects from the soil).

2.7. Indirect methods (relative estimates)


These are many measures which may correlate with population density: trap catches,
visual counts, counts of animal products (frass, nests), proportion of infested hosts (for
parasites, in broad sense). Indirect measures can be related to population density using
regression analysis (linear or non-linear).

Let us remember the basic concepts of regression analysis.

Linear regression:
y' = a + bx

The least square method is most often used to draw the "best" line through a cloud of
points. This method adjusts the values of regression parameters (a and b) so that the
residual sum of squares (=sum of square deviations of points from the line) reaches a
minimum.

The residual sum of squares is:

The Least Square Method means that we find such parameter values a and b that the
value of is minimized. It follows from the calculus that derivatives at the minimum
point are equal to zero:

After substituting of and simplification:


The total sum of squares is . The sum of squares for the
factor effect: , where the covariance is estimated using equation:

The residual sum of squares: .

R-square is .

Important things in regression analysis


R = 0.82, N.S.! However,
significance does not mean
1. Significance
biological significance (e.g., if
R = 0.01)
It is important to check if the
2. Influence diagnistics most influential points are
correct.
Possible solutions:
1. Ignore an outlier
3. Outliers
2. Change regression model
3. leave it as it is
Plot data before any
regression analysis. Use
4. Non-linearity polynomial or non-linear
regression if the relationship is
not linear
Use transformations only if
they are biologically
5. Variable transformation
meaningful. It is better to use
non-linear models.

Polynomial regression

This is a non-linear function, but least square estimation leads to a system of linear
equations. Thus, this regression is analyzed by a linear method. This is NOT a non-
linear regression!

Note: use step-wise regression when you test the significance of non-linear terms in the
polynom. The effect is significant if the increment of R-square is large enough
according to F-statistics:
where, in the numerator, there is a difference in R-squares estimated in two consecutive
steps, and are corresponding degrees of freedom (d.f.= number of regression
coefficients minus 1). The difference - is equal to 1 because one term is added at a
time.

Nonlinear regression
Nonlinear regression is estimated numerically. Residual sum of squares is a function of
model parameters. Thus the minimum can be found as a lowest point on the response
hyper-surface:

Several methods are used to search for a minimum. Examples are:

• simplex - slow but more reliable


• gradient - faster but not robust

Danger: you can end up in a local minimum (see the figure above). To avoid it, try to
start from various initial conditions.

It is desirable that the equation represents some theoretical model of a real system. Then
regression coefficients have biological interpretation.

Example. We return back to indirect population measures. Binomial sampling is a


method when instead of counting organisms in each sample, we count the number of
samples where organisms were present. For example, the density of Colorado potato
beetles can be reconstructed from the proportion of potato plants where at least one
insect was found. This is a faster method than counting all insects. If we assume the
random (poisson) distribution of beetles, then the proportion of uninfested plants is
equal to the zero term po of the poisson distribution:

where M is the mean number of individuals per plant. It is clear that the mean density,
M, can be estimated as the negative logarithm from the proportion of uninfested plants,
po.

An alternative theoretical model can be derived from the assumption that beetles are
aggregated on host plants and that their distribution is negative binomial. The zero term
of the negative binomial distribution is:
where k is the aggregation parameter. To test, which model is better, it is necessary to
use the non-linear regression and then to compare R-square.

Questions and Assignments to Lecture 2


2.1. What is population density?

2.2. The weed is more abundant at the edges of the field. Weed biomass is sampled at
random locations within the field. Can we use equation

to estimate the standard error of mean weed biomass per sq.m.? If not, then what
options do we have?

2.3. What is correlogram?

2.4. Density of the Colorado potato beetle should be estimated with accuracy A = 0.1
(10%). Preliminary sampling (N = 30) indicated: M = 0.5 beetles/plant, SD = 0.9
beetles/plant. How many additional samples you need?

2.5. Solve the ordinary kriging system for two sample points:

z are variable values at sample points; estimate z-value at the estimation point.

2.6. Sampling of the European pine sawfly was performed separately within 1 m radius
around trees and outside of these circles using 0.25 sq.m. squares:

Stratum % area No. of samples Mean SD


Near trees 30 40 32 18
Far from trees 70 20 12 7

Estimate average sawfly density and its standard error in the entire area

2.7. Estimate the size of the population of perch in a pond from capture-recapture data:
284 fish were netted, marked and released; 1392 fish were caught after 2 days, and 86 of
them were found to be marked. (use Lincoln index)
Lecture 3. Spatial distribution of organisms

3.1. Tree Types of Spatial Distribution


Three types of spatial distribution are usually considered:

3.2. Random distribution


Random spatial distribution is simulated using poisson distribution.

Simplest example: 100 people are fishing in the same lake for the same time (e.g. 3 h);
they have equal probability to catch a fish per unit time.
Question: How many fishers catch 0, 1, 2, 3 etc. fish?

No. of fish No. of Proportion Poisson


captured, fishers of fishers distribution
i n(i) p(i) n'(i)=Np'(i)

0 11 0.11 10

1 25 0.25 23

2 21 0.21 27

3 25 0.25 20

4 9 0.09 12

5 7 0.07 5

6 2 0.02 2

7 0 0.00 1

Total N=100 1.00 100

Mean number of fish captured by 1 fisher, M = 2.30, and standard deviation, SD = 1.41.

Poisson distribution is described by equation:


where m is the mean and i!= 1×2×3× ... ×i, 0!=1; 1!=1.

Theorem: In poisson distribution, mean = variance:

Two main methods of parameter estimation

1. Method of moments (m = M)
2. Non-linear regression (iterative approximation)

In the table above we used the method of moments: m = M = 2.3.

Chi-square test is used to test if sample distribution is different from theoretical


distribution. The equation is:

where n(i) is the sample distribution (e.g., the number of fishers that captured i fish),
and n'(i) is theoretical distribution (e.g., expected number of fishers that captured i fish
according to poisson distribution). In our example,

The number of degrees of freedom is equal to the number of distribution classes (7


classes in our example) minus the number of parameters that were used to adjust the
theoretical distribution to the sample distribution. We used 2 parameters: m=2.3 and N =
100. Thus,

df = 7 - 2 = 5

Critical value for chi-square for df = 5 and P = 0.05 is 11.07.


Conclusion: Sample distribution does not differ significantly from the poisson
distribution.

Note. Chi-square test cannot prove that sample distribution is the same as the theoretical
distribution! If there is no significant differences, it may mean two things: sample
distribution is really very close to the theoretical distribution, or there may be just not
enough data to distinguish these distributions. Suggestion: Use multiple hypothesis, i.e.,
compare sample distribution with several theoretical distributions.

Relationship between fishers (our example) and spatial distribution. Imagine, we


sample a population by counting organisms in sample areas (e.g., 1 sq.m.). Then, each
sample area is equivalent to a fisher and the number of organisms found is equivalent to
the number of fish captured. The notion "random distribution" can be defined using the
model of random deposition of individual organisms. We start from empty space and
put the first organism by random selection of its coordinates. The organism may end in
a sample area (=fish captured). Then we add the second organism, and so on.

Thus, the proportion of samples in which i organisms were found will correspond to the
poisson distribution.

Quick test for the type of spatial pattern:


Coefficient of dispersion:
if CD << 1 then regular distribution
if CD ≈ 1 then random distribution
if CD >> 1 then aggregated distribution

Poisson distributions are asymmetric at low mean values, and almost symmetric at
higher mean values:

When mean increases to infinity, poisson distribution coincides with a normal


distribution.

3.3. Aggregated spatial distributions


There is no universal theoretical model for aggregated spatial distributions. However,
there are empirical models, e.g., negative binomial distribution (NBD) which work very
well in the majority of cases. NBD is described by the equation:

where m is mean, k is "coefficient of aggregation" (aggregation increases with


decreasing k). The zero term (the proportion of empty samples) equals to:

Other terms can be estimated by iteration:

Estimating NBD parameters using the method of moments:


m=M

For a random distribution, CD=1 and k is infinite. When k increases to infinity then
NBD distribution coincides with poisson distribution.

3.4. Indexes of aggregation


1. Coefficient of dispersion:

2. Mean crowding (Lloyd 1967) is equal to the mean number of "neighbors" in the
same quad:

No. of individuals No. of neighbors


Sample No. N(N-1)
(N) (N-1)

1 5 4 20

2 3 2 6

3 0 -1 0

4 1 0 0

5 7 6 42

Total 16 - 68

It means that the mean number of "neighbors" is = 4.25.

Note: mean crowding has biological sense only if the size of each quad corresponds to
"interaction distance" among individuals.

Consider probability distribution p(i) = the proportion of samples with i individuals.


Then:
For a random (poisson) distribution, CD=1, and thus, = m.

3. Lloyd (1967) suggested a "patchiness" index: /m, which is 1 for a random


distribution, > 1 for aggregated distributions, and <1 for regular distributions. However
it does not make more sense than CD.

3.5. Density-Invariant Indexes of Aggregation


Simple indexes of aggregation are specific to a particular population sampled at
particular time. They cannot be extrapolated neither in space nor in time, and this is
their major limitation. In order to overcome this limitation, several density-invariant
indexes has been developed.

1. The most frequently used is the "power law" (Taylor 1961):

Coefficient b is considered as species-specific. This equation was shown to work well in


the wide range of species density. Of course, it is hard to expect that b will be constant
in any kind of environment, but for populations in similar environments it is usually
stable.

2. Mean crowding regression (Iwao 1968):

Regression coefficients can be used to distinguish between different patterns of spatial


distribution:

3. Negative binomial k. This is not a good index because usually it is not density-
invariant.

Effect of Quad Size on Aggregation Indexes


Different scales of spatial distribution should be considered. The distribution may be
random at small scales and aggregated at larger scales. Thus, it is important to examine
the distribution using different quad sizes (quad size = spatial resolution).
For example, the distribution is random, if and only if the coefficient of dispersion (CD)
is equal to 1 for all the range of quad sizes. Thus, if CD=1 for a specific quad size, then
we cannot conclude that the distribution is random, we need to test other quad sizes.

Any index of aggregation can be plotted against quad size. However, there are
specialized indexes designed for multiple quad sizes. For example, ro index (Iwao
1972) was defined as:

where subscripts 1,2,...i stand for successively increasing sizes of quads. Ro index is
used to determine characteristic distances in a spatial distribution:

3.6. Geostatistical Analysis of Population Distribution


The main idea of geostatistical methods is to relate the spatial variation among
population densities to the distance lag. We already used one of geostatistical tools: an
omnidirectional correlogram. Here we are going to explore a set of related tools:
covariance functions and variograms. Also, we will discuss the phenomenon of
anisotropy.

Directional correlogram is defined as correlation among population counts at


points separated by space lag h. The difference from the omnidirectional correlogram is
that h is a vector rather than a scalar (that is why it is in bold face). For example, if
h={20,10}, then each pair of compared samples should be separated by 20 m in west-
east direction and by 10 m in south-north direction:

In practice, it is difficult to find enough sample points which are separated by exactly
the same lag vector h. Thus, the set of all possible lag vectors is usually partitioned into
classes:
Vectors that end in the same cell are grouped into one class and correlogram value is
estimated separately for each class. The number of directions may be different (4, 8, 16,
etc.)

Correlogram is estimated using equation:

where indexes -h and +h refer to sample points located at the tail and head of vector h;
z-h and z+h are organism counts in samples separated by lag vector h; summation is
performed over all pairs of samples separated by vector h; Nh is the number of pairs of
samples separated by vector h; M-h and M+h are mean values for samples located at the
tail and head of vector h; s-h and s+h are standard deviations of samples located at the
tail and head of vector h.

Other measures of spatial dependence are covariance function:

and variogram (=semivariogram):

The correlogram, covariance function, and variogram are all related. If the population
mean and variance are constant over the sampling area (there is no trend) then:
where C(0) is the covariance at zero lag = variance = squared standard deviation.

Typical variogram has the following shape:

Interpretation of the nugget effect: It shows the pure random variation in population
density (white noise) or it may be associated with sampling error.

Examples of one-dimensional spatial distributions with different nugget effects and


corresponding variograms:

Anisotropy: different spatial relationships in different directions

A variogram with anisotropy and corresponding spatial pattern:


Log-transformation of data may be necessary before variogram estimation:

Transformation is necessary to make the distribution more symmetrical and to remove


the trend in variance. In a log-normal distribution, variance is proportional to the mean
squared. Thus, in high-density areas, the variance is higher than in low-density areas.
After log-transformation the variance becomes uniform.

Note: If zero values are present, use transformation: log(N+1).

Some times indicator transformation is used:

where c is a threshold value ("a cutoff").

Usually, a series of thresholds is used, and variograms are estimated for all of them. If
one threshold has to be selected, then the best is to take the median threshold m which
corresponds to the 50% cumulative probability distribution.

3.7. Fractal Dimension of Population Distribution


Old definition of a fractal: a figure with self-similarity at all spatial scales.
Fractal is what will appear after infinite number of steps.

Examples of fractals were known to mathematicians for a long time, but the notion was
formulated by Mandelbrot (1977).

New definition of a fractal: Fractal is a geometric figure with fractional dimension

It is not trivial to count the number of dimensions for a geometric figure. Geometric
figure can be defined as an infinite set of points with distances specified for each pair of
points. The question is how to count dimensions of such a figure. Hausdorf suggested to
count the minimum number of equal spheres (circles in the picture below) that cover the
entire figure.

The number of spheres, n, depends on their radius, r, and dimension was defined as:

For example, dimension of a line equals to 1 (see figure above):

"Normal" geometric figures have integer dimensions: 1 for a line, 2 for a square, 3 for a
cube. However, fractals have FRACTIONAL dimensions, as in the example below:
Here we use rather large circles, and thus, the precision is not high. For example, we got
D=2.01 for a square instead of D=2.

Dimension of a square and fractal is estimated as follows:

Square: , Fractal:
Below is the Mandelbrot set which is also a fractal:

Fractal dimension, D, is related to the slope of the variogram plotted in log-log scale, b:

D = 2 - b/2 for a 1-dimensional space


D = 3 - b/2 for a 2-dimensional space
In the figure above, b=1, and thus, D=1.5 for a 1-dimensional space.

3.8. Questions and Assignments to Lecture 3


3.1. How to distinguish among the regular, random and aggregated spatial distributions
of organisms?

3.2. How to estimate distribution parameters using the method of moments?

3.3. Colorado potato beetles were counted on 30 potato plants, and there was no
significant difference between the actual distribution of beetles and both poisson and
negative binomial theoretical distributions. How to decide if their distribution is random
or aggregated? If additional sampling is required, then how to determine the additional
number of samples?

3.4. Taylor's power law for the number of fleas on rats is: ln(s ) = 0.1 + 1.5 ln(m). Total
17 fleas were found on 10 rats. How many rats should be examined to estimate flea
abundance with a 10% accuracy?

3.5. What is the relation between the correlogram, covariance function, and variogram?
3.6. Estimate the fractal dimension of the "carpet of Serpinski":

3.7. What is the area of this "carpet"?


Lecture 4. Statistical Analysis of Population Dynamics
4.1. Correlation between population density and
factors
Statistical models are used to analyze the correlation between the population density and
various factors that change in time and/or in space, e.g., temperature, precipitation,
elevation, etc. These models can be used to predict population numbers in the future or
in unsampled spatial locations.

Data may be of 3 types:

1. Time series - a sequence of measurements at the same point in space


2. Spatial series - a set of measurements made (simultaneously or not) at different
points in space
3. Mixed series - a set of time series obtained from different points in space

Statistical models are usually represented by linear or polynomial equations:

where y is the predicted variable (e.g., population density); Xi are factors; and bi are
parameters which can be found using regression analysis.

Autoregressive model is a regression model in which the value of the dependent


variable is predicted from values of the same variable in previous time steps or in
neighboring locations in space. Most statistical models of population dynamics are
autoregressive. For example, the density of the population in the next year can often be
predicted from its current density and/or its density in 1 or 2 previous years. In spatio-
temporal models, population density in a specific location in the next year can be
predicted from the current density in the same location as well as in neighboring
locations. Autoreggressive models may include external factors, i.e., climatic factors.

Geostatistics (kriging, see Lecture 2) is a better method for spatial modeling than
standard regression. It is possible to use a 3-dimensional kriging (2 space coordinates,
and 1 time coordinate) for spatio-temporal models .

The mechanisms that cause correlations between population density and factors are not
represented in these models. Prediction is based solely on the previous behavior of the
system. If in the past the system exhibited specific behavior in particular situations, then
it is most probable that it behave in a similar way in the future. Statistical modeling is
based on the assumption that system behavior remains the same. Thus, these models
have very limited value for predicting new behaviors of the system. For example, if we
develop a new method for controlling pest populations or a new strategy for
conservation of some species, then stochastic models based on the previous dynamics of
these populations may be misleading.
In some cases stochastic models may help to understand some mechanisms of
population dynamics. For example, if insect population density increases in years with
hot and dry summer, then it is possible that insects of this species have low mortality in
these years, possibly due to reduced resistance of host plants. Of course, these
hypotheses require thorough testing.

4.2. Correlation between factors


One of the problems with multiple regression is that factors may be correlated. For
example, temperature is highly correlated with precipitation. If factors are correlated,
then it is impossible to separate the effect of different factors. In particular, regression
coefficients that indicate the effect of one factor may change when some other factor is
added or removed from the model. Step-wise regression helps to evaluate the
significance of individual terms in the equation.

First, I will remind you the basics of the analysis of variances (ANOVA)
Total sum of squares (SST) is the sum of squared deviations of individual
measurements from the mean. The total sum of squares is a sum of 2 portions:
(1) Regression sum of squares (SSR) which is the contribution of factors into the
variance of the dependent variable, and
(2) Error sum of squares (=residual sum of squares) (SSE) which is the stochastic
component of the variation of the dependent variable.

SSR is the sum of squared deviations of predicted values (predicted using regression)
from the mean value, and SSE is the sum of squared deviations of actual values from
predicted values.

The significance of regression is evaluated using F-statistics:

where
df(SSR)= g - 1 is the number of degrees of freedom for the regression sum of squares
which is equal to the number of coefficients in the equation, g, minus 1;
df(SSE)= N - g is the number of degrees of freedom for the error sum of squares which
is equal to the number of observations, N, minus the number of coefficients, g;
df(SST) = df(SSR) + df(SSE) = N - 1 is the number of degrees of freedom for the total
sum of squares.

The null-hypothesis is that factors has no effect on the dependent variable. If this is true,
then the total sum of squares is approximately equally distributed among all degrees of
freedom. As a result, the fraction of the sum of squares per one degree of freedom is
approximately the same for regression and error terms. Then, the F-statistic is
approximately equal to 1.

Now, the question is, how much should the F-statistic deviate from 1 to reject the null
hypothesis. To answer this question we need to look at the distribution of F assuming
the null hypothesis:
If estimated (empirical) value exceeds the threshold value (which corresponds
to the 95% cumulative probability distribution) then the effect of all factors
combined is significant. (See tables of threshold values for P = 0.05, 0.01,
and 0.001)

Note: In some statistical textbooks you can find a two-tail F-test (5% area is partitioned
into two 2.5% areas at both, right and left tails of the distribution). This is a wrong
method because small F indicates that the regression performs too well (some times
suspiciously well). Null hypothesis is not rejected in this case! If F is very small, then
we may suspect some cheating in data analysis. For example, this may happen if too
many data points were removed as "outliers". However, our objective here is not to test
for cheating (we assume no cheating). Thus we use a 1-tail F-test.

The F-distribution depends on the number of degrees of freedom for the numerator
[df(SSR)] and denominator [df(SSE)].

Standard regression analysis generally cannot detect the significance of individual


factors. The only exception are orthogonal plans in which factors are independent (=not
correlated). In most cases, factors are correlated, and thus, a special method called step-
wise regression should be used to test the significance of individual factors. The step-
wise regression is a comparison of two regression analyses:
(1) the full model and
(2) the reduced model in which one factor is excluded.
The full model has more degrees of freedom, and therefore, it fits data better than the
reduced model. Thus, the regression sum of squares, SSR, is greater for the full model
than for the reduced model. The question is: is this difference significant or not? If it is
not significant, then the factor that was excluded is not important and can be ignored.
The significance is tested with the same F-statistic, but SSR and df(SSR) are replaced
by the difference in SSR and df(SSR) between the full and reduced models:

where SSR and SSR1 are regression sum of squares for the full and reduced models,
respectively; df(SSR) and df(SSR1) are degrees of freedom for the regression sum of
squares in the full and reduced models, respectively; SSE is the error sum of squares for
the full model; and df(SSE) is the number of degrees of freedom for the error sum of
squares.

Because only one factor was removed in the reduced model,

df(SSR) - df(SSR1) = 1.

The F-statistic is related to the t-statistic if the denominator has only one degree of
freedom:

Thus, the t-statistic can be used instead of the F in the step-wise regression.
Example of the step-wise regression:
Full model ; SSR =53.2, SSE =76.3, df(SSR) =2, df(SSE) =53.
Reduced model y = a + b x; SSR =45.7, SSE =83.8, df(SSR1)=1, df(SSE1)=54.
F=(53.2-45.7)53 / 76.3 = 5.21; t = 2.28; P<0.05.
Thus, the quadratic term is significant (non-linearity test).

4.3. Example: Colored fox in Labrador (1839-1880)


The time series comes from Elton (1942).

First step is to log-transform the data. When predicting population density, log-
transformation is always better than no transformation.

The distribution of log-transformed data is more symmetric now. We will experiment


with different factors trying to get the most accurate prediction of log population
density.

1. One factor: previous year population counts

Predictor t-ratio P
Xt-1 0.25 0.801 NS

= 0.2%.
This means that our regression does not work any better than using average log-density.

2. Two factors: population counts in two previous years

Predictor t-ratio P
Xt-1 0.35 0.728 NS
Xt-2 2.73 0.010
= 16.8%.
Effect of year t-1 is not significant, but the effect of year t-2 is significant. Non-
significant effect can be ignored. Thus, we can re-estimate regression using year t-2 as
the only predictor:

Predictor t-ratio P
Xt-2 2.75 0.009

= 16.6%.

3. Plotting the regression

It is always necessary to plot the regression to see if there are any non-linear effects. It
seems that some non-linearity may be present (blue line). Let us test for quadratic
effects of year t-2.

Predictor t-ratio P
Xt-2 1.37 0.179 NS
1.14 0.261 NS

= 19.4%.
OK, there are no quadratic effects of year t-2.

4. Adding year t-3.


Now we can try to add year t-3. May be the model will work better.

Predictor t-ratio P
Xt-2 2.88 0.007
Xt-3 0.41 0.683 NS

= 18.9%.
The model did not get significantly better after adding year t-3 as a predictor. Thus, we
cannot improve the model any further.
5. Plotting the residuals.
Now we can plot the residuals ( ) versus population
counts in year t-1 to test if there is any non-linear effect of year t-1.

This graph indicates no non-linear effects of population counts in year t-1

6. Prediction of population numbers


Now, we can use the model to predict fox counts a year ahead (t) using population
counts in the previous year (t-2) as a predictor:

Predicted population counts follow the same pattern as observed values. However,
predicted values have smaller variation because any regression has a "smoothing"
effect.

In the previous graph, we predicted population counts one year ahead at a time. Let's see
what happen if we try to predict the entire time series from two initial values. In this
case, the error will propagate because we will use predicted population counts as the
base for further predictions.

Predicted population counts exhibit damped oscillations. After a few oscillations, they
approach the equilibrium level of x=5.553.

This model cannot be used to predict population numbers more than 1-3 years ahead.
4.4. Autocorrelation of factors and model validation
The variable is called autocorrelated if its value in specific place and time is correlated
with its values in other places and/or time. Spatial autocorrelation is a particular case of
autocorrelation. Temporal autocorrelation is also a very common phenomenon in
ecology. For example, weather conditions are highly autocorrelated within one year due
to seasonality. A weaker correlation exists between weather variables in consecutive
years. Examples of autocorrelated biotic factors are: periodicity in food supply, in
predator or prey density, etc.

Autocorrelation of factors may create problems with stochastic modeling. In standard


regression analysis, which is designed for fitting the equation to a given set of data, the
autocorrelation of factors does not create any problems . The effect of the factor is
considered significant if it sufficiently improves the fit of the equation to this data set.
This approach works well if we are interested in one particular data set.
For example, when geologists predict the concentration of specific compounds in a
particular area they can use any factors that can improve prediction. And they usually
don't care if the effect of these factors will be different in another area.

However, ecologists are mostly interested in proving that some factor helps to predict
population density in all data sets within a specific class of data sets. It appeared that
models may work well with the data to which they were fit, but show no fit to other data
sets obtained an different time or in a different geographical points.To solve this
problem, the concept of validation was developed.

Model Validation is testing the model on another independent data set.

Example 1. In 60-s an 70-s it was very popular to relate population dynamics to the
cycles of solar activity. Solar activity exhibits 11-yr cycles which seemed to coincide
with the cycles of insect population outbreaks and population dynamics of rodents.
Most analyses were done using from 20 to 40-yr time series. However, 2 independent
cyclic processes with similar periods may coincide very well in short time intervals.
When larger time series became available, it appeared that periods of population
oscillations were usually smaller or greater than the period of the solar cycle. As a
result, the relationship between population density and solar activity may change its
sign in a larger time scale.

Example 2. Our fox model (section 4.3.) was developed by fitting the equation to the
time series. Thus, it is not surprising that it fits these data rather well. The question is,
will this model work if tested on an independent data set which was not used for fitting
the equation. We can separate the data into two portions, one of which is used for model
fitting and the other portion is used for model validation.

In our example, we select first 22 years and used them for estimating the regression:

t-ratio = 1.03; P=0.317; = 5.6%. Effect is non-significant! There is nothing to


validate. We can stop at this point and say that there is not enough data for validation.
However we can make another try and separate the data in a different way. In our case,
population numbers in year t depends on population numbers in tear t-2. Thus, we can
estimate the regression using uneven years only and then test it using even years.

Regression obtained from uneven years:

t-ratio = 13.14; P<0.001; = 66.4%. Effect is highly significant.

Now we test this equation using data for even years. The equation is a predictor of
population numbers; thus estimated values from the equation can be used as the
independent variable and actual population numbers are used as a dependent variable.
Then:
R-square = 0.0001
F = 0.0002
P = 0.96
This means that the equation derived from uneven years did not help to predict
population dynamics in even years.
Conclusion: the model is not valid.

Crossvalidation

The idea of crossvalidation is to exclude one observation at a time when estimating


regression coefficients, and then use these coefficients to predict the excluded data
point. This procedure is repeated for all data points, and then, estimated values can be
compared with real values. However, I don't know any methods for testing statistical
significance with crossvalidation. Standard ANCOVA cannot be applied because it does
not account for the variation in regression parameters when some data are not used in
the analysis.

Crossvalidation method is often considered as a very economical validation method


because it uses the maximum possible observation points for estimating the regression.
But, this is wrong! Crossvalidation is NOT A VALIDATION METHOD. It evaluates
the variability of regression results when some data from the given set are omitted. This
method does not attempt to predict the variability of regression results in other possible
data sets characterized by the same autocorrelation.

Jackknife method (Tukey)

Jackknife method is similar to crossvalidation, but its advantage is that it can be used
for testing the significance of regression. Details of this method can be found in Sokal
and Rohlf (1981. Biometry). Let us consider the same example of colored fox. We need
to test if the effect of population density in year t-2 is significant. Jackknife method is
applied as follows:

Step 1. Each data point is left out from the analysis in turn, and the regression is
estimated each time (same as in crossvalidation). When observation i is excluded we get
the equation: . The slope bi characterizes the effect of population
density in year t-2 on the population density in year t. Slopes appear to be different for
different i, and different from the slope estimated from all data points (b = -0.412). The
variability of bi corresponds to the accuracy of estimated slope.

Year, t xt-2 xt i bi Bi
3 6.562 5.952 1 -0.430 0.286
4 7.318 4.343 2 -0.361 -2.429
5 5.729 6.562 3 -0.435 0.446
6 6.540 7.318 4 -0.530 4.184
7 4.813 5.729 5 -0.407 -0.611
8 5.131 6.540 6 -0.412 -0.429
9 6.605 4.813 7 -0.395 -1.099
10 4.525 5.131 8 -0.427 0.172
11 5.036 6.605 9 -0.409 -0.533
12 5.578 4.525 10 -0.426 0.11
13 6.657 5.036 11 -0.398 -0.962

Step 2. We need to determine if the slope significantly differs from 0 (in our case
significantly < 0). But the variation of bi is much smaller than the accuracy of the slope
because each slope was estimated from a large number of data points. Thus, we will
estimate pseudovalues Bi using the equation:

Bi = Nb - (N - 1)bi

where b = -0.412 is the slope estimated from all data, N is the number of observations,
and bi is the slope estimated from all observations except the i-th observation.

Step 3. The last step is to estimate the mean, SD, SE, and t for pseudovalues:

M = -0.403
t = |SE/M| = 2.43
SD = 1.05
P = 0.02.
SE = 0.166

Conclusion: The slope is significantly < 0.

The jackknife procedure is not a validation method for the same reasons as the
crossvalidation. However, jackknife is less sensitive to the shape of the distribution than
the standard regression. Thus it is more reliable.

Correlogram product method

This is a new method which can be used for validating correlations between population
density and other factors. However, this method cannot be used for autoregressive
models and thus we cannot apply it for the colored fox example.
This method was developed by Clifford et al. (1989, Biometrics 45: 123-134). The null
hypothesis is that variables X and Y are independent but each of them is autocorrelated.
For example, if we consider the relationship between population density and solar
activity, both variables are autocorrelated. The autocorrelation in population density
may result from the effect of solar activity or from other factors, e.g., interaction with
predators. In our null hypothesis we consider that solar activity has no effect on the
population density, but we allow autocorrelations in the population density.

In short, the standard deviation of the correlation between 2 independent autoregressive


processes, X and Y equals the square root of the weighted product of correlograms for
variables X and Y:

where h is the temporal or spatial lag, are correlograms for variables X and Y, and
weights, wh, are equal to the proportion of data pairs separated by lag, h.

Thus, to test for significance of the correlation between processes X and Y we need to
estimate the standard error using the equation above. If the absolute value of empirical
correlation is larger than SD multiplied by 2 (t-value), then the correlation is significant.

Example. The correlation between log area defoliated by gypsy moth in CT, NH, VT,
MA, ME, and NY and the mean number of sunspots in 1964-1994 was r = 0.451:

This correlation is significant (P = 0.011), and it seems that we can use solar activity as
a good predictor of gypsy moth outbreaks. However, both variables are autocorrelated,
and thus, we will use the correlogram product method. Correlograms (=autocorrelation
functions, ACF) for both variables are shown below:
Correlograms are periodic indicating a cyclic behavior of both variables. The cycle is
very similar and the period is 9-10 years (location of the first maximum). The weights
wh are the following: wo = 1/N, and

for h > 0, where N = 31 is the number of observations:

Now we apply the correlogram product method: multiply both correlograms and
weights at each lag, h, and then take the square root from their sum (here we used only
for h $lt; 16). The standard error for correlation is SE= 0.337; t = r/SE = 0.451/0.337 =
1.34; P = 0.188.
Conclusion: Correlation between the area defoliated by the gypsy moth and solar
activity may be coincidental. More data is needed to check the relationship.

Another possible way of the analysis is autoregression combined with external factors.
For example, we can use the model:

where Dt is the area defoliated in year t, Wt is the averrage number of sunspots in year t,
and bi are parameters. The error probability (P) for the effect of sunspots will be very
close to that obtained in tyhe previous analysis. However, coefficients bi do not
necessary have biological meaning. For example, the equation above assums that
current defoliation depends on defoliation in previous years, but there may be no
functional relationship between areas defoliated in different years. Their correlation
may simply result from the effect of some external autocorrelated factor (e.g. solar
activity). Thus, it is necessary to use caution in the interpretation of autoregressive
models.

Note: In many cases, the effect of a factor is significant with one combination of
additional factors but becomes non-significant after adding some more factors. People
often select arbitrarily some combination of factors that will maximize statistical
significance of the factor tested. However, this is just another way of cheating. If
additional factors destroy the relationship between variables that you want to prove,
then there is no relationship. It is necessary to use as many additional factors as possible
including autoregressive terms.

The sequence of statistical analysis

1. Data collection. Select the population characteristic to measure, select factors


that may affect population dynamics, make sure that these factors are
measurable with resources available. If you analyze historical data, try to find all
data that may be useful for the period of time and area you are interested in.
Suggest hypothetical equations that can be useful for analysis.
2. Selection of model equation. Try to incorporate all available data.
3. Regression analysis. Remove factors which effect is not significant. Significance
of individual factors is evaluated using sequential regression.
4. Plot residuals versus each factor, check for non-linear effects. If non-linear
effects are found, modify the model and go to #3.
5. Validate the model using independent data, or use the correlogram-product
method. If the model is valid, then use it. If there is not enough data for
validation: use the model with caution (it might be wrong!); collect additional
data for validation. If additional data does not support the model, then go to #2.

4.5. Stochastic models based on regression


Our objective is to reproduce the pattern of population change rather than to predict the
most probable population counts in the next year. Our model for the fox could not
predict the pattern of population change: predicted density approached a steady state by
damped oscillations, whereas in nature there are quasi-periodic cycles.

This model can reproduce the pattern of population change better if stochastic noise is
added which can draw simulated population counts away from the steady state. Noise
can be simulated by a random variable which distribution is normal with zero mean.
Zero mean is important as a non-biasedness condition. The variance (or SD which is the
square root of the variance) should be set equal to the error variance: [SSE/df(SSE)] of
regression.

Stochastic version of the fox population model is:

Random variable has a normal distribution with zero mean and variance equal to the
error variance in regression: Var( ) = 0.26 (S.D.= 0.51).
Stochastic model generates quasi-periodic cycles similar to that in the real population:

4.6. Biological interpretation of Stochastic Models.


Response surfaces.
The major advantage of stochastic models is that they can be built automatically from a
given data set using existing statistical software. No programming and no high
mathematics is needed. However, the cost of this simplicity is that stochastic models
have no direct biological interpretation. This means that stochastic models usually do
not improve our understanding of processes that determine population dynamics.

But, it is possible to modify stochastic models so that they will have some biological
interpretation. Usually this modification does not reduce the fit of the model. Let us
consider the dynamics of a population with discrete generations (e.g., a monovoltine
insect). Population density, Nt in year t can be estimated from population density in the
previous year t-1:

where s is survival and F is fecundity. Both, survival and fecundity may depend on the
density of the population in year t-1 and on environmental conditions in that year.
Survival may also depend on the density of natural enemies (predators, parasites, and
pathogens) in year t-1, which may depend on the density of the prey (host) population in
year t-2. Thus, survival, s, in year t-1 may be a function of population density in year t-
2. Weather conditions in year t-2 may also affect the survival. It is possible to imagine
even longer feed-back loops. For example, if it takes 3 years for host plants to recover
after severe defoliation caused by insect outbreak, then the quantity and/or quality of
food will cause a 3-yr delayed feedback to the insect population.

These density-dependent processes are often called "regulation". However, this term is
very ambiguous and I prefer to avoid it (see Lecture 9).

The product of survival and fecundity in the previous equation is the net rate of
population increase, R, which is a function of previous population densities and
previous weather conditions:
where wt in year t are weather conditions in year t. For simplicity we will ignore
weather conditions, and density effects with delays longer than 2 years. Then the
equation will become

After log-transformation it will be:

where . is
the rate of population increase in year t which can be positive (if the population grows)
or negative (if the population declines). Now, the rate of population increase can be
represented as a linear function of population densities in the current and previous
years:

This equation is still a statistical model because coefficients bi are estimated using
regression analysis. However, this equation has some biological meaning. For example,
if the effect of population density on the rate of population increase is not significant
then the model becomes equivalent to the exponential model. If population growth
declines with increasing current density, b1 < 0, then some non-delayed density-
dependent processes should be present (e.g., competition, or pathogen infection), and
the model is equivalent to the discrete-time logistic model. If population growth
declines with increasing density in the previous generation, b2 < 0, then some non-
delayed density-dependent processes are present, e.g., parasitism by specialized
parasitoids. Delayed density-dependence usually yields oscillations in population
density.

In a more general case, the function is non-linear. It can be represented


as some surface in a 3-d space:
Turchin and Taylor (1993, Ecology, 73: 289-305) suggested the following equation for
fitting non-linear surfaces:

where bi, h and g are parameters which can be fit using non-linear regression.

Note, that population densities are not log-transformed in the right side of the equation.
However, it can be proved that log-transformation is equivalent to setting parameters h
or g close to zero. Thus, log-transformation can be considered is a specific case of
power-transformation when the power is close to zero.

This is called "response-surface methodology" which means that the shape of the
surface is more important than the equation that fits it. Different species may have
different response surfaces which can be found using this general equation. Response
surface cannot identify mechanisms of population change, but it can indicate some
characteristics of these mechanisms, e.g. immediate density-dependence, delayed
density-dependence, etc.

Example. Let us analyze the dynamics of colored fox. This time we will predict the rate
of population increase rather than population density. First we will use the linear
equation that relates the rate of population increase in year t to log-transformed
population densities in years t and (t-1):

where . After estimating regression


parameters the equation becomes .
Predictor t-ratio P
xt 6.93 0.000
xt-1 2.73 0.010

= 76.8%.

The effect of log population numbers in year t-1 on the rate of population increase in
year t is the same as the effect of log population numbers in year t-2 on the log
population numbers in year t (see our previous example). However, we got a significant
effect of population numbers in year t on the rate of population increase in the same
year which seems contradict to our previous result that population numbers in year t did
not correlate significantly with population numbers in the previous year.

However, there is nothing wrong here. Be got different answers because we asked
different questions. In our previous analysis we wanted to know is it possible to
improve the prediction of population density in year t using the information about
population numbers in the previous year, and we got the negative answer. Then we
asked if the rate of population increase is related to the current population density (test
for density-dependence) and got a positive answer.

To understand the difference, imagine a perfect density-dependence which draws


population density exactly to the equilibrium state. We will also assume that
environmental fluctuations add some noise into population dynamics. In this case,
population density in year t is just a random variable entirely independent from
population density in previous years. But this is the result of a strong density-
dependence.

The advantage of using the rate of population increase is in the possibility of


interpreting data in terms of density-dependence. Now we can say that there is a non-
delayed density-dependence which may be associated with intraspecific competition or
pathogens plus delayed density-dependence which may be associated with specialized
predators or parasitoids. Finally we will try to improve our model using the method of
Turchin and Taylor (1993). Using non-linear regression we get the following equation:

where g = 0.0025 and h = 0.824. = 79.2%. Non-significant terms were remover from
the equation of Turchin and Taylor. Prediction has improved a little. The shape of this
response surface is shown below:
There is a strong non-delayed density-dependence (effect of Nt) and also some delayed
density-dependence (effect of Nt-1) which becomes weaker when density increases.
Lecture 5. Exponential and Logistic Growth
Exponential Model
Exponential model is associated with the name of Thomas Robert Malthus (1766-1834)
who first realized that any species can potentially increase in numbers according to a
geometric series. For example, if a species has non-overlapping populations (e.g.,
annual plants), and each organism produces R offspring, then, population numbers N in
generations t=0,1,2,... is equal to:

When t is large, then this equation can be approximated by an exponential function:

There are 3 possible model outcomes:

1. Population exponentially declines (r < 0)


2. Population exponentially increases (r > 0)
3. Population does not change (r = 0)

Parameter r is called:

• Malthusian parameter
• Intrinsic rate of increase
• Instantaneous rate of natural increase
• Population growth rate

"Instantaneous rate of natural increase" and "Population growth rate" are generic terms
because they do not imply any relationship to population density. It is better to use the
term "Intrinsic rate of increase" for parameter r in the logistic model rather than in the
exponential model because in the logistic model, r equals to the population growth rate
at very low density (no environmental resistance).
Assumptions of Exponential Model:

1. Continuous reproduction (e.g., no seasonality)


2. All organisms are identical (e.g., no age structure)
3. Environment is constant in space and time (e.g., resources are unlimited)

However, exponential model is robust; it gives reasonable precision even if these


conditions do not met. Organisms may differ in their age, survival, and mortality. But
the population consists of a large number of organisms, and thus their birth and death
rates are averaged.

Parameter r in the exponential model can be interpreted as a difference between the


birth (reproduction) rate and the death rate:

where b is the birth rate and m is the death rate. Birth rate is the number of offspring
organisms produced per one existing organism in the population per unit time. Death
rate is the probability of dying per one organism. The rate of population growth (r) is
equal to birth rate (b) minus death rate (m).

Applications of the exponential model

• microbiology (growth of bacteria),


• conservation biology (restoration of disturbed populations),
• insect rearing (prediction of yield),
• plant or insect quarantine (population growth of introduced species),
• fishery (prediction of fish dynamics).

Logistic Model
Logistic Model
Logistic model was developed by Belgian mathematician Pierre Verhulst (1838) who
suggested that the rate of population increase may be limited, i.e., it may depend on
population density:

At low densities (N < < K), the population growth rate is


maximal and equals to ro. Parameter ro can be interpreted
as population growth rate in the absence of intra-specific
competition.

Population growth rate declines with population numbers, N, and reaches 0 when N =
K. Parameter K is the upper limit of population growth and it is called carrying
capacity. It is usually interpreted as the amount of resources expressed in the number of
organisms that can be supported by these resources. If population numbers exceed K,
then population growth rate becomes negative and population numbers decline. The
dynamics of the population is described by the differential equation:

which has the following solution:

Three possible model outcomes

1. Population increases and


reaches a plateau (No < K).
This is the logistic curve.
2. Population decreases and
reaches a plateau (No > K)
3. Population does not change
(No = K or No = 0)

Logistic model has two equilibria: N = 0 and N = K. The first equilibrium is unstable
because any small deviation from this equilibrium will lead to population growth. The
second equilibrium is stable because after small disturbance the population returns to
this equilibrium state.

Logistic model combines two ecological processes: reproduction and competition. Both
processes depend on population numbers (or density). The rate of both processes
corresponds to the mass-action law with coefficients: ro for reproduction and ro/K for
competition.

Interpretation of parameters of the logistic model

Parameter ro is relatively easy to interpret: this is the maximum possible rate of


population growth which is the net effect of reproduction and mortality (excluding
density-dependent mortality). Slowly reproducing organisms (elephants) have low ro
and rapidly reproducing organisms (majority of pest insects) have high ro. The problem
with the logistic model is that parameter ro controls not only population growth rate, but
population decline rate (at N > K) as well. Here biological sense becomes not clear. It is
not obvious that organisms with a low reproduction rate should die at the same slow
rate. If reproduction is slow and mortality is fast, then the logistic model will not work.
Parameter K has biological meaning for populations with a strong interaction among
individuals that controls their reproduction. For example, rodents have social structure
that controls reproduction, birds have territoriality, plants compete for space and light.
However, parameter K has no clear meaning for organisms whose population dynamics
is determined by the balance of reproduction and mortality processes (e.g., most insect
populations). In this case the equilibrium population density does not necessary
correspond to the amount of resources; thus, the term "carrying capacity" becomes
confusing. For example, equilibrium density may depend on mortality caused by natural
enemies.

Discrete-time analogs of the exponential and logistic


models
Exponential model analog:

where t is time measured in generations, and R is net reproduction rate. For


monovoltine organisms (1 generation per year), R is the average number of offsprings
per one parent. For example, in monovoltine insects with a 1:1 sex ratio, R =
Fecundity/2.

The dynamics of this model is similar to the continuous-time exponential model.

Logistic model analog (Ricker):

The dynamics of this model is similar to the continuous-time logistic model if


population growth rate is small (0 < ro < 0.5). However, if the population growth rate is
high, then the model may exhibit more complex dynamics: damping oscillations, cycles,
or chaos (see Lecture 9). An example of damping oscillations is shown below:

Complex dynamics results from a time delay in feed-back mechanisms. There are no
intermediate steps between time t and time t+1. Thus, overcompensation may occur if
the population grows or declines too fast passing the equilibrium point. In the
continuous-time logistic model, there is no delay because the rate of population growth
is updated continuously. Thus, the population density cannot pass the equilibrium point.
Questions and assignments to Lecture 5
1. Population numbers of cockroaches double every month (30 d). What is their
intrinsic rate of increase (per day)?
2. What is the intrinsic rate of increase in a human population if every family has 3
children at parent's age of 30 (there are no singles, no divorces, sex ratio 1:1)?
What would be the numbers of human population after 100 years if initial
numbers are 4 billion?
3. A new lake was created after building a dam. The number of fish censused after
2, 4, 6, 8 and 10 years since that time was 1000, 2000, 3500, 5000 and 6000.
Estimate parameters of the logistic model using non-linear regression. Plot the
data and the model on one graph.
4. Use Excel to simulate population dynamics with the discrete-time logistic model
(Ricker's model) for 60 generations. Use K=100; r = 0.1, 0.5, 1.0, 1.5, 1.9, 2.2;
No = 10.
Lecture 6. Life-tables and k-values
In this lecture you will learn how to collect data for the analysis of population
processes.

Analysis of population processes is as easy as balancing your personal budget! You


need to estimate the increases and losses in population numbers due to different
processes. If you are lucky, then the net change in population numbers will be equal to
the algebraic sum of the effects of all studied processes. If you are less lucky, then some
of the increases or losses in population numbers will be missing. In this case, the non-
attributed mortality is considered as the effect of some unknown factor, we can call it
"winter mortality" or "mystery disease". Additional research can be done later to study
these unknown processes.

Ecological processes are usually specific to organisms' age or stage. Thus, they have to
be recorded relative to the life-cycle stage. This information is usually called a "life-
table". Two types of life-tables are generally used: (1) age-dependent and (2) stage-
dependent.

Age-dependent life-tables
Age-dependent life table shows organisms' mortality (or survival) and reproduction rate
(maternal frequency) as a function of age. In nature, mortality and reproduction rate
may depend on numerous factors: temperature, population density, etc. When building a
life-table, the effect of these factors is averaged. Only age is considered as a factor that
determines mortality and reproduction.

Example. Consider a sheep population which is censused once a year immediately after
breeding season:

Probability of No. of female offspring


Age, years surviving to age x born to a mother of age x
(x)
(lx) (mx)
0 1.000 0.000
1 0.845 0.045
2 0.824 0.391
3 0.795 0.472
4 0.755 0.484
5 0.699 0.546
6 0.626 0.543
7 0.532 0.502
8 0.418 0.468
9 0.289 0.459
10 0.162 0.433
11 0.060 0.421

Only females are considered in this life-table. However, there is no problem to include
male populations into the life table. Then, survival rates should be specified separately
for males and females, and the sex ratio of offspring should be taken into account.

Survivorship curves
Survival probabilities lx are often plotted against age x. These graphs are called
"survivorship curves". They show, at what age death rates are high and low. The
following graphs show two survivorship curves for domestic sheep (data from Caughley
1967) and for lapwings or green plovers (Vanellus vanellus) in Britain (data from
Deevey 1947):

Survivorship curve is exponential (with negative growth) for the lapwing. This means
that survival rate is independent of age. In the log-scale, survivorship curve becomes a
straight line (see above).

Age-specific mortality is estimated using equation:

Sheep mortality generally increases with age; and the slope of survivorship curve
becomes steeper to the end. Humans have a similar shape of survivorship curve.

Characteristics derived from life-tables


Net reproductive rate, R0, is the average number of female offspring born to a sheep
considered at age 0. Consider N new-born sheep. Some of them will die without
producing any offspring (zero offspring), others will produce one or several offspring.
R0 is the average number of female offspring produced in the entire group of N sheep.
In our example, R0 = 1×0 + 0.845×0.045 + 0.824×0.391 + ... = 2.513. It means that an
ewe produces 2.513 ewe lambs in average in a lifetime.

Average generation time, T, is estimated using equation:

In our example, T = (0×1×0 + 1×0.845×0.045 + 2×0.824×0.391 + ...) / 2.513 = 12.825 /


2.513 = 5.1 years. It means that the average age of mothers when they give birth to an
eve lamb is 5.1 years.

Note: If organisms breed continuously, then generation time will be overestimated


using this equation because all births are summed over the period between census dates
which is equal to one age step. Generation time can be adjusted by subtracting half of
age step.

Now it is possible to estimate approximate value of intrinsic rate of increase r using the
following logic. We assume discrete generations with generation time T=5.1 years and
net reproductive rate of R0 =2.513. If population size at zero time was N0, then after T
years, the population will grow to NT = N0×R0. According to an exponential model,

where ln is natural logarithm (logarithm with base e =2.718). We got the equation for r:

Note: This is an approximate estimation of r because we used a simplified assumption


that generations are discrete. Accurate estimation of r will be discussed in the next
chapter.

Time units used for age measurement


If survival and reproduction are continuous processes without any cyclic change, then
any time units may be suitable: days, weeks, months, years. Time units should provide
sufficient resolution. For example, if the life span of an organism is 2 months, then
taking 1 month as a unit will result in only two age intervals which is definitely not
sufficient. In most cases, the number of age intervals is in the range from 10 to 50.

If survival or reproduction are cyclic (e.g., seasonal), then one cycle can be taken as a
time unit. In this case it may happen that the number of age intervals will drop to 2 or 3.
However, it may be not very dangerous if reproduction is limited to a short period
within the year because there will be little age difference between organisms born in the
same year.
If survival and reproduction are cyclic but the entire life span is less or equal to this
cycle, then time units should be smaller than the cycle length. Age-dependent life-tables
can be built for the entire population only if the breeding period is short and therefore
organisms' development is synchronized. Otherwise, separate life-tables should be built
for subpopulations that start their development at different seasons.

Determining survival of organisms till age x (lx)


For domestic animals or for populations reared in the laboratory, it is possible to
observe the fate of a large group of individuals that all started life simultaneously.
Survivors can be counted at regular time intervals and lx values can be easily estimated.
Similar technique can be applied to non-moving organisms (plants, sedimentary
animals). It is possible to mark a large number of individuals and to trace their fate.

This kind of analysis is usually impossible in populations of moving organisms. There


are two principal methods to deduce lx values in this case: by determining the age
distribution in the population, or by determining the ages at death.

If the population is stationary (i.e., population numbers and age distribution do not
change) then the number of new-born organisms x time units ago was the same as now;
and survivors of that group of organisms are of age x. Thus,

where N(x) is the number of organisms of age x. Here we assumed that age can be
accurately measured. In many species the number of "growth rings" in specific organs is
equal to age in years. Examples of such organs are: stems of trees, scales of fishes,
horns in sheep, roots of canine teeth in bears, otholits in fishes. The weight of the eye
lens can be used for age measurement in some animal species. However, in many
populations, measuring age is a difficult problem.

Assumption of population stationarity usually is taken a-priori if no historical data exist.


In cases when age distribution has a periodic component, we can speculate that it
resulted from fluctuations in population numbers. Periodic component can be filtered
out using regression methods.

Consider a large number of carcasses whose age have been determined. We assume that
the probability of detecting a carcass does not depend on the age of animal at death. The
proportion of individuals that were at age x when they died is dx. These individuals
survived to age x but did not survive to age x+1. Thus,

The method of estimation of lx from age of carcasses also requires stability of


population numbers and of age structure.
Determining reproduction rates (mx)
Reproduction rate (=maternal frequency) is equal to the number of female offspring
produced per one mother in age interval from x to x+1. Survival of mothers and
offspring during this time interval should be included into m ; i.e., mx is equal to the
total number of offspring produced in one time interval and survived till the end of this
period divided by the initial number of parent females at the beginning of the time
interval.

Reproduction rates are easy to determine in laboratory reared animals or plants. In


natural populations of mammals, maternal frequencies can be derived from the
proportion of pregnant and/or lactating animals. In birds, maternal frequencies can be
determined from the average number of chick per nest. Indirect measures of maternal
frequencies should be used with caution because they may be biased.

Stage-dependent life-tables
Stage-dependent life tables are built in the cases when:

• The life-cycle is partitioned into distinct stages (e.g., eggs, larvae, pupae and
adults in insects)
• Survival and reproduction depend more on organism stage rather than on
calendar age
• Age distribution at particular time does not matter (e.g., there is only one
generation per year)

Stage-dependent life tables are used mainly for insects and other terrestrial
invertebrates.

Example. Gypsy moth (Lymantria dispar L.) life table in New England (modified from
Campbell 1981)

Mortality Initial no. of No. of Mortality Survival k-value [-


Stage
factor insects deaths (d) (s) ln(s)]
Egg Predation, etc. 450.0 67.5 0.150 0.850 0.1625
Egg Parasites 382.5 67.5 0.176 0.824 0.1942
Dispersion,
Larvae I-III 315.0 157.5 0.500 0.500 0.6932
etc.
Larvae IV-
Predation, etc. 157.5 118.1 0.750 0.250 1.3857
VI
Larvae IV-
Disease 39.4 7.9 0.201 0.799 0.2238
VI
Larvae IV-
Parasites 31.5 7.9 0.251 0.749 0.2887
VI
Desiccation,
Prepupae 23.6 0.7 0.030 0.970 0.0301
etc.
Pupae Predation 22.9 4.6 0.201 0.799 0.2242
Pupae Other 18.3 2.3 0.126 0.874 0.1343
Adults Sex ratio 16.0 5.6 0.350 0.650 0.4308
Adult
10.4
females
TOTAL 439.6 97.69 0.0231 3.7674

Specific features of stage-dependent life tables:

• There is no reference to calendar time. This is very convenient for the analysis
of poikilothermous organisms.
• Gypsy moth development depends on temperature but the life table is relatively
independent from weather.
• Mortality processes can be recorded individually and thus, this kind of life table
has more biological information than age-dependent life tables.

K-values
K-value is just another measure of mortality. The major advantage of k-values as
compared to percentages of died organisms is that k-values are additive: the k-value of a
combination of independent mortality processes is equal to the sum of k-values for
individual processes.

Mortality percentages are not additive. For example, if predators alone can kill 50% of
the population, and diseases alone can kill 50% of the population, then the combined
effect of these process will not result in 50+50 = 100% mortality. Instead, mortality will
be 75%!

Survival is a probability to survive, and thus we can apply the theory of probability. In
this theory, events are considered independent if the probability of the combination of
two events is equal to the product of the probabilities of each individual event. In our
case event is survival. If two mortality processes are present, then organism survives if
it survives from each individual process. For example, an organism survives if it was
simultaneously not infected by disease and not captured by a predator.

Assume that survival from one mortality source is s1 and survival from the second
mortality source is s2. Then survival from both processes, s12, (if they are independent)
is equal to the product of s1 and s2:

This is a "survival multiplication rule". If survival is replaced by 1 minus mortality


[s=(1-d)], then this equation becomes:
For example, if mortality due to predation is 60% and mortality due to diseases is 30%,
then the combination of these two death processes results in mortality of d = 1-(1-
0.6)(1-0.3)=0.72 (=72%).

Varley and Gradwell (1960) suggested to measure mortality in k-value which is the
negative logarithms of survival:

k = -ln(s)

We use natural logarithms (with base e=2.718) instead of logarithms with base 10 used
by Varley and Gradwell. The advantages of using natural logarithms will be shown
below.

It is easy to show that k-values are additive:

The k-values for the entire life cycle (K) can be estimated as the sum of k-values for all
mortality processes:

In the life table of the gypsy moth (see above), the sum of all k-values (K = 3.7674) was
equal to the k-value of total mortality.

This graph shows the relationship between mortality


and the k-value. When mortality is low, then the k-
value is almost equal to mortality. This is the reason
why the k-value can be considered as another
measure of mortality. However, at high mortality, the
k-value grows much faster than mortality. Mortality
cannot exceed 1, while the k-value can be infinitely
large.

The following example shows that the k-value represents mortality better than the
percentage of dead organisms: One insecticide kills 99% of cockroaches and another
insecticide kills 99.9% of cockroaches. The difference in percentages is very small
(<1%). However the second insecticide is considerably better because the number of
survivors is 10 times smaller. This difference is represented much better by k-values
which are 4.60 and 6.91 for the first and second insecticides, respectively.

Key-factor analysis Varley and Gradwell (1960) developed a method for identifying
most important factors "key factors" in population dynamics. If k-values are estimated
for a number of years, then the dynamics of k-values over time can be compared with
the dynamics of the generation K-value. The following graph shows the dynamics of k-
values for the winter moth in Great Britain.

It is seen that the dynamics of winter disappearance (k1) is most resembling the
dynamics of total generation K-value. The conclusion was made that winter
disappearance determines the trend in population numbers (whether the population will
grow or decline), and thus, it can be considered as a "key factor". There were numerous
attempts to improve the method. For example, Podoler and Rogers (1975, J. Anim.
Ecol, 44(1)) suggested regressing k over K.

But, this method was criticized recently because the meaning of a "key" factor was not
explicitly defined (Royama 1996, Ecology). It is not clear what predictions can be made
from the knowledge that factor A is a key-factor. For example, the knowledge of key-
factors does not help us to develop a new strategy of pest control.

The key-factor analysis was often considered as a substitute for modeling. It seems so
easy to compare time series of k-values and to find key-factors without the hard work of
developing models of ecological processes. However, reliable predictions can be
obtained only from models.

This critique does not mean that life-tables have no value. Life-tables are very important
for gathering information about ecological processes which is necessary for building
models. It is the key-factor analysis that has little sense.
K-value = instantaneous mortality rate multiplied by time. A population that
experience constant mortality during a specific stage (e.g., larval stage of insects)
change in numbers according to the exponential model with a negative rate r. We cannot
call r intrinsic rate of natural increase because this term is used for the entire life cycle,
and here we discuss a particular stage in the life cycle. According to the exponential
model:

Population numbers decrease and thus, Nt < N0. Survival is: s = Nt/N0 . Now we can
estimate the k-value:

k = -r t
.

Instantaneous mortality rate, m, is equal to the negative exponential coefficient because


mortality is the only ecological process considered (there is no reproduction):

m = -r,
k=mt

Exponential coefficient r is negative (because population declines), and mortality rate,


m, is positive.

We proved that if mortality rate is constant, then k-value is equal to the instantaneous
mortality rate multiplied by time. This is analogs to physics: distance is equal to speed
multiplied by time. Here, instantaneous mortality rate is like speed, and k-value is like
distance. K-value shows the result of killing organisms with specific rate during a
period of time. If the period of time when mortality occurs is short then the effect of this
mortality on population is not large.

If instantaneous mortality rate changes with time, then the k-value is equal to its integral
over time. In the same way, in physics, distance is the integral of instantaneous speed
over time.

Example. Annual mortality rates of oak trees due to animal-caused bark damage are
0.08 in the first 10 years and 0.02 in the age interval of 10-20 years. We need to
estimate total k-value (k) and total mortality (d) for the first 20 years of oak growth.

k = 0.08 × 10 + 0.02 × 10 = 1.0

d = 1 - exp(-k) = 0.63

Thus, total mortality during 20 years is 63%.

Limitation of the k-value concept. All organisms are assumed to have equal dying
probabilities. In nature, dying probabilities may vary because of spatial heterogeneity
and individual variation (both inherited and non-inherited).
Estimation of k-values in natural populations. Estimation of k-values for individual
death processes is difficult because these processes often go simultaneously. The
problem is to predict what mortality could be expected if there was only one death
process. In order to separate death processes it is important to know the biology of the
species and its interactions with natural enemies. Below you can find several examples
of separation of death processes.

Example #1. Insect parasitoids oviposit on host organisms. Parasitoid larva hatches
from the egg and starts feeding on host tissue. Parasitized host can be alive for a long
period. Finally, it dies and parasitoid emerges from it. Insect predators usually don't
distinguish between parasitized and non-parasitized prey. If an insect was killed by a
predator, then it is usually impossible to detect if this insect was parasitized before.
Thus, mortality due to predation is estimated as the ratio of the number of insects
numbers destroyed by predators to the total number of insects, whereas mortality due to
parasitism is estimated as the ratio of the number of insects killed by parasitoids to the
number of insects that survived predation. In this example, predation masks the effect of
parasitism, and thus, insects killed by predators are ignored in the estimation of the rate
of parasitism. The effect is the same as if predation occurred before parasitism in the life
cycle. Thus, in the gypsy moth life table, predation was always considered before
parasitism. Diseases also mask the effect of parasitism and thus they are considered
before parasitism.

Example #2. It is often possible to distinguish between organisms destroyed by


different kinds of predators. For example, small mammals and birds open sawfly
cocoons in a different way. Suppose, 20% of cocoons were opened by birds, 50% were
opened by mammals, and remaining 30% were alive. The question is what would be the
rate of predation if birds and mammals were acting alone. We assume that sawfly
cocoons have no individual variation in predator attack rate, and that cocoons destroyed
by one predator cannot be attacked by another predator. First, we estimate total k-value
for both predator groups: k12 = -ln(0.3) = 1.204. Second, we subdivide the total k-value
into two portions proportionally to the number of cocoons destroyed by each kind of
predator. Thus, for birds k1 = 1.204×20/(20+50) = 0.344, and for mammals k2 =
1.204×50/(20+50) = 0.860. The third step is to convert k-values into expected mortality
if each predator was alone: for birds d1 = 1- exp(-0.344) = 0.291, and for mammals d2
= 1- exp(-0.860) = 0.577.

Questions and assignments to Lecture 6


1. Build a life table for an aphid population (aphids reproduce parthenogenetically).
Estimate lx, dx, mx, Ro, T, and r. (See a picture of aphids!)
Number of Mean number of
Age, days (x)
survivals offsprings per parent
0 1000 0
1 900 0
2 820 0
3 750 0
4 680 0
5 620 0
6 550 1
7 500 2
8 450 5
9 400 10
10 350 12
11 300 10
12 250 8
13 200 6
14 100 3
15 50 1
16 0 0

2. Partial life-table. The European pine sawfly, Neodiprion sertifer, cocoons were
collected at the beginning of August and dissected. Results of dissection of new (current
year) cocoons are the following:

Healthy sawfly eonymph 144


Eaten by predators 125
Exit hole of parasitoid Drino inconspicua 15
Exit hole of parasitoid Pleolophus basizonus 78
Larvae of parasitoid Exenterus abruptorius 210
Exit hole or larvae of gregarious parasitoid
23
Dahlbominus fuscipennis
Fungus disease 205
Total 800

Life-cycle information:

Excellent images of parasitoids are available from the PHERODIP homepage.

• Parasitoids D.inconspicua, P.basizonus and D.fuscipennis have several


generations per year, whereas E.abruptorius has only 1 generation.
• D.inconspicua (Tachinidae) is an endoparasite and attacks larvae (4-5 instar). It
emerges from the host immediately after host cocooning. It develops very fast
and wins the competition with any other parasitoids.
• E.abruptorius is an ectoparasite, attacks host eonymphs a day prior to
cocooning. Parasitoid larvae emerges inside the cocoon, eats the host and
overwinters as larvae inside host cocoon. If the host was previously parasitized
by D.inconspicua, then E.abruptorius dies.
• P.basizonus and D.fuscipennis attack host cocoons. They are ectoparasites. If
another parasite (E.abruptorius) is already present in the cocoon, it will be eaten
first. D. fuscipennis wins the competition with P.basizonus.

Estimate mortality caused by each natural enemy, convert it into k-value. Check that the
sum of all k-values is equal to the total k-value for sawfly cocoons. Write results in the
table, putting mortality processes in the order of their operation.

Simple example:

Healthy eggs 200


Desiccated eggs 100
Parasitized eggs 200
Total 500

Number of eggs in
Number
which this
Mortality process of killed Mortality Survival k-value
mortality process
eggs
can be detected
1. Desiccation 500 100 0.2 0.8 0.223
2.Parasitism 400 (500-100) 200 0.5 0.5 0.693
Total 500 300 0.6 0.4 0.916

3. Estimate mortality in a predator-exclusion experiment. The fall webworm,


Hyphantria cunea, larvae in colonies were counted at the beginning and at the end of
experiment:
A. Control - without protection
B. Exclusion of large predators: prey were protected by a 1/2 inch cell hardware cloth
C. Exclusion of all predators: prey were protected by 1 mm cell mesh
Estimate: mortality caused by large and small predators, convert it to k-value.

Initial larvae in 10
Larvae alive at the end
colonies
Control 3000 1400
Large predators
3500 2900
excluded
All predators excluded 3200 3100

4. Estimate gypsy moth mortality due to virus (NPV).


Gypsy moth larvae were collected in the forest at 7-day intervals and placed
individually in the cups with diet. Incubation period of viral infection (from infection
till death) is 7 days. Estimate: total mortality caused by virus and the k-value.

Larvae that died in 7 days


Collected larvae
since collection
1-st week 200 23
2-nd week 200 7
3-rd week 200 5
4-th week 200 30
5-th week 200 58
6-th week 200 115

5. Estimate the rate of simultaneous mortality processes


Gypsy moth (Lymantria dispar) pupae are destroyed by small mammals and by
invertebrates. 300 laboratory-reared pupae were placed on tree boles. Three days later,
200 of them were damaged by small mammals and 50 were damaged by invertebrate
predators (Calosoma sycophanta). Each pupa can be eaten just once. Estimate mortality
caused by each predator guild if another predator guild was absent (note: use k-values!).
Lecture 7. Model of Leslie
The model of Leslie is one of the most healivy used models in population ecology. This
is a discrete-time model of an age-structured population which describes development,
mortality, and reproduction of organisms. The model is formulated using linear algebra.
This model is mostly used to answer the following two questions:

1. What is the rate of exponential growth (intrinsic rate of increase)?


2. What is the proportion of each age class in the stable age distribution?

7.1. Model Structure


The model of Leslie (1945) describes 3 kinds of ecological processes:

1. Development (progress through the life cycle)


2. Age-specific mortality
3. Age-specific reproduction

Variables and parameters of the model:

• Nx,t = number of organisms in age x at time t (age is measured in the same units
as time t). Usually, only females are considered and males are ignored because,
as a rule, the number of males does not affect population growth.
• sx = survival of organisms in age interval from x to x+1.
• mx = average number of female offsprings produced by 1 female in age interval
from x to x+1 (mortality of parent and/or offspring organisms is included)

There are two equations:


[1]

[2]

Equation [1] represents development and mortality, whereas equation [2] represents
reproduction. Equation [2] specifies the number of individuals in the first age class and
equation [1] specifies the number of individuals in all other age classes. In the equation
[1], the number of individuals in age x+1 in time t+1 equals to the number of
individuals in the previous age and previous time multiplied by age-specific survival
rate sx. In the equation [2] the number of new-born organisms equals to the number of
mothers (Nx,t) multiplied by the numbers of offspring produced (mx). The number of
offsprings is summed over all ages of mothers.

These two equations can be combined into one matrix equation:


where Nx is the vector of age distribution in the population at time t, and A is the
transition matrix.

When a matrix is multiplied by a vector, we take the 1st row of the matrix, multiply
each number by the corresponding number in the vector-column and then sum all
products. This sum is the value of the 1st element in the result vector. Then we take the
2nd row of the matrix, multiply it by the same vector and the sum becomes the 2nd
element in the result vector. In the same way we can estimate all elements of the result
vector.

The first element of the result vector corresponds to the equation [2], and all other
elements correspond to the equation [1].

How to read matrix models?


Each column specifies the fate of organisms in specific state. The number in the
intersection of column i and row j indicates how many organisms in state j are produced
by one organism in state i. In the Leslie model, organisms' state is defined by age only.
For example, the third column corresponds to age a=2. An organism in age 2 produces
m offsprings of age 0 (first cell in the column), and goes to age class 3 with probability
s (the cell under main diagonal).

Most simple stochastic process is defined by a matrix of transition probabilities between


states:

In this case, the sum of matrix elements in each column equals to 1 because it is
assumed that each system passes through a series of states and neither die nor
reproduce.

Leslie model is more complex because the sum of matrix elements in each column is
not necessary equal to 1. This is a "branching process" because the life trajectory of a
parent branches into life trajectories of its offsprings.
Matrix models are easy to iterate in time. In the next time step we again multiply the
transition matrix by the vector of age distribution:

This equation can be used to simulate as many time steps as necessary.

7.2. Model Behavior


Two major characteristics of the behavior of Leslie model:

• A few damping oscillations are followed by an exponential growth


• Age distribution approaches a stable age distribution

These 2 features can be seen in the graphs that show simulations of sheep population
dynamics:

In this simulation we started from a population of 100 new-born sheep:

The first graph shows exponential population growth (it becomes linear in a log scale)
after several initial years. The second graph shows convergence of age distribution to a
stable age distribution.

Here is an example of the model of Leslie implemented using Excel spreadsheet:


Excel spreadsheet "leslie.xls"

You can play with this model by changing model parameters (the matrix) and initial
age-distribution.

You can simplify the analysis of matrix models using PopTools, which is a free Excel
plugin developed by Greg Hood, CSIRO, Canberra, Australia.

7.3. Estimation of the intrinsic rate of population


increase
Previously we discussed an approximate equation:

The model of Leslie gives an accurate estimation of r.

Method #1
For simplicity we take years as time units. However, the same logic can be applied to
days or weeks.

The number of organisms in age x in year t is equal to the number of new-born


organisms (x=0) x years ago multiplied by their survival (lx) till age x:

[3]

where

Now, equation [2] of the Leslie model can be re-written as:

[4]

After initial damped fluctuations, the Leslie model shows exponential growth and age
distribution stabilizes. Thus, the number of organisms in any age class will grow
exponentially. In particular, the number of new-born organisms increases exponentially:

Then,
[5]

Now we combine equations [4] and [5]:

The term N0,t-x can be taken out of the summation expression and then we get the
equation:

[6]

Equation [6] can be used to estimate r. The sum at the left side can be estimated for
different values of r, and then we can select the r-value that makes this sum equal to 1. It
makes sense to start with the r-value estimated using the approximate method discussed
in the previous chapter. Thus, we start with r=0.181 and get the sum [6] equal to 0.9033.
When r increases, then the value of the sum [6] decreases because r is included as a
negative exponent. Obtained value of the sum appeared to be less than 1, and thus, we
need to try smaller values of r. Let's select r=0.16. Then the sum is equal to 1.0092. The
exact value of r can be found by linear interpolation:

Now we check the solution: when r=0.1617 then the sum [6] is equal to 1.00014 which
is very close to 1.

Note: In Pielou (1978), this example is estimated in a different way which is difficult to
understand. She constructed a different matrix by adjusting reproduction rates and
provided no explanation for this adjustment.

Method #2
Intrinsic rate of population increase can be estimated as the logarithm of the only real
and positive eigenvalue of the transition matrix. The theory of eigenvalues is the central
topic in linear algebra. It is used to reduce multidimensional problems to one-
dimensional problems. I recommend to study this topic for those students who plan to
be quantitative ecologists. Here we will only estimate the eigenvalue using available
software without going into details of the algorithm. The only real and positive
eigenvalue of our matrix is equal to =1.176. Then, r = ln( ) = 0.162 which is very
close to the value estimated by method #1.

Checking the result

Obtained value of r can be checked by estimating the regression of log population


numbers versus time. Initial years should be ignored because age structure has not been
stabilized yet. The slope of this regression should be equal to r. If we take the time
interval from t = 25 to 50, then the regression equation is ln(N) = 4.3557 + 0.1617 t.
Regression slope is exactly equal to r estimated by method #1.

7.4. Estimation of stable age distribution


Equation [5] can be re-written as:

Substituting this equation into [3] we get the relationship between the number of
organisms in age x and in age 0 in a stable age distribution:

Now we can estimate the proportion of organisms, c , in age x:

[7]

Age,
x
lx exp(-rx) lxexp(-rx) cx Simulated cx

0 1.000 1.0000 1.0000 0.2413 0.2413


1 0.845 0.8507 0.7188 0.1734 0.1734
2 0.824 0.7237 0.5963 0.1439 0.1439
3 0.795 0.6156 0.4894 0.1181 0.1181
4 0.755 0.5237 0.3954 0.0954 0.0954
5 0.699 0.4455 0.3114 0.0751 0.0751
6 0.626 0.3790 0.2373 0.0572 0.0572
7 0.532 0.3224 0.1715 0.0414 0.0414
8 0.418 0.2743 0.1147 0.0277 0.0277
9 0.289 0.2333 0.0674 0.0163 0.0163
10 0.162 0.1985 0.0322 0.0078 0.0078
11 0.060 0.1689 0.0101 0.0024 0.0024
Total 4.1445 1.0000 1.0000
Age distribution estimated using equation [7] (column 5) coincided with simulated age
distribution after 50 iterations of the model.

7.5. Modifications of the Leslie model


1. Variable matrix elements. Survival and reproduction rate of organisms may depend
on a variety of factors: temperature, habitat characteristics, natural enemies, food, etc.
To represent these dependencies, the elements of Leslie model can be replaced by
equations that specify survival and reproduction rates as functions of various factors.
Equations can be obtained from experimental data.

2. Distributed delays. Age and time are equivalent in the original Leslie model, and
thus, all organisms develop synchronously with constant rate. However, development
rate of invertebrates and plants is not constant: it depends on temperature and may vary
among individuals. Individual variation of development rates is called "distributed
delay" because there is a distribution of time when organisms reach maturity. Transition
matrix can be modified to incorporate these features.

This matrix has non-zero diagonal elements,


This matrix allows organisms to leap
and thus, some proportion of organisms
over several age intervals in one time
remain in the same age class when time
step. As a result, development goes faster
increases. As a result, development goes
than it would be in the original Leslie
slower than it would be in the original Leslie
model. The rate of development can be
model. The rate of development can be
adjusted by changing the length of age
adjusted by changing relative values of
leaps.
diagonal and sub-diagonal elements

3. Partitioning the life cycle into stages. Many invertebrate species have a complex
life cycle that includes several stages. For example, holometabolous insects usually
have 4 stages: egg, larvae, pupae, and adult. Each of these stages may include several
age intervals: In these models, age is no longer measured in calendar time units (e.g.,
days or years). Instead, it is measured in independent units which can be interpreted as
"physiological age". The concept of physiological age will be discussed in details in the
next chapter. It can be used to define "rate of development" as the average increment of
physiological age per calendar time unit.
Lecture 8. Development of poikilothermous organisms,
degree-days
8.1. Rate of development
Homeothermous organisms are warm blooded (mammals, birds)
Poikilothermous organisms are cold blooded (all invertebrates, plants, fishes,
amphibians, reptiles)

Development of poikilothermous organisms depends on temperature, whereas


development of homeothermous organisms is temperature-independent.

The rate of development can be measured by a reciprocal of the number of time units
(e.g., days) that is required for completion of development. Rates of development can be
estimated for the entire onthogenesis or for a specific stage. For example, if it takes 15
days for an insect to develop from egg hatch till pupation, then the rate of larval
development is v = 1/15 = 0.0667.

Development time, Development rate,


Temperature (°C), t
T v=1/T
5 - -
10 200 0.005
15 100 0.010
20 60 0.017
25 40 0.025
30 30 0.033
35 35 0.028

Here is the graph of development rate versus temperature:

In the temperature range from 10 to 30 (°C), development rate changes almost linearly
with increasing temperature. At very low temperature there is no development, and at
very high temperature development is retarded.

8.2. Simple degree-day model


Degree-day model is based on the assumption that development rate is a linear function
of temperature.

In most cases, real development rate is indeed a linear function in the region of
moderate temperatures (15-25°) (see figure above). Deviating points at temperatures
that are too low or too high can be ignored. For example, in the figure, regression line is
plotted for points from 10 to 30°. At 5° all organisms died and there was no
development. At 35° organisms were overheated and development rate of survivors was
reduced. If we are interested in simulating organism development in moderate
temperatures, then the degree-day model will be the best choice.

Terms:

• tmin is the lower temperature limit; this is the temperature at which development
rate reaches zero.
• ET = t - tmin is effective temperature.
• S are degree-days; this is the effective temperature multiplied by the number of
days required to complete development.

Lower temperature limit and degree-days can be estimated from the regression line of
development rate versus temperature. We assume that regression equation is:
v = a + bt

where a is intercept, and b slope.

a + btmin = 0, by the definition of the lower temperature limit. Thus,

In our case, tmin = 0.0104 / 0.00142 = 7.3°.

Duration of development is T = 1/v. Thus, degree-days can be estimated as:

This equation shows that degree-days do not depend on temperature! This is the
principal feature of the degree-day model. In our case, S = 1 / 0.00142 = 704 degree ×
day. The units of degree-days are degrees (centigrade) multiplied by days. Some times,
it is better to use degree-hours if development is very fast.

Because degree-days are temperature-independent, it is possible to use them to predict


development time in experiments with variable temperature. In this case effective
temperatures are accumulated day by day, and when the sum reaches S, then
development is finished.
Example: tmin = 10 , S = 100.

Average Effective temperature, Accumulated


Day No.
Temperature t t - tmin degree-days

1 15 5 5

2 18 8 13

3 25 15 28

4 23 13 41

5 24 14 55

6 18 8 63

7 17 7 70

8 15 5 75

9 18 8 83

10 15 5 88

11 22 12 100

12 25 15 115

Accumulated degree-days reach the value S = 100 on day 11. Thus, it takes 11 days to
complete the development.

8.3. How to measure temperature?


Yes, use a thermometer! The only problem is when and where to measure it. The
simplest way is to measure minimum and maximum temperatures each day and then
take the average. Majority of weather data bases contain minimum and maximum
temperatures. More accurate estimates can be obtained if temperature was measured
several times a day at regular intervals (e.g. every 3 h). Then, you can average these
measures.

The next problem is where to measure. It is obvious, that temperature should be


measured where studied organisms are located. For example, if we study development
of soil insects [e.g., wireworms (Elateridae)], then temperature should be measured in
the soil at ca. 5-10 cm from the surface. Spider mites live on the lower leaf surface
where temperature is several degrees lower than the ambient temperature. In this case,
temperature should be measured under the leaf. Some times it is possible to build a
regression model that predicts the temperature in a specific niche from ambient
temperature recorded at weather stations.
Can we average temperature prior to analysis?

If development rate is a linear function of temperature, then temperature can be


averaged prior to analysis. However, if this function is non-linear (e.g., in the improved
degree-day model, see below), then temperature averaging may result in substantial
errors.

8.4. Improved degree-day model


The purpose of modification is to use the model for a wide range of temperatures.
Effective temperature (ET) is defined as follows:

This model is non-linear because the graph is not straight (see the figure above). Thus,
temperature cannot be averaged! In particular, you cannot use average daily
temperature. Instead, it is necessary to use actual temperature dynamics. Accumulated
degree-days are equal to the area under the temperature curve restricted to the
temperature interval between tmin and tmax:

Light-blue area equals to accumulated degree-days. Here the average temperature is


below tmin but organisms can accumulate some degree-days because daily maximums
are above tmin.

Another example:
Light-blue area again equals to accumulated degree-days. Daily maximum temperature
exceeds tmax however this excess does not count in the accumulation of degree-days.

Non-linear models of development rate require simulation of diurnal temperature


change. In most cases, only daily minimum and maximum temperature are known.
There are two most frequently used methods for simulation temperature change.
"Rectangular" model assumes that temperature stays at maximum for half of the day
and stays at minimum for the other half:

This method is not very accurate but it is simple and fast.

"Sine-wave" method was developed by Allen (1976; Environ. Entomol. 5: 338-396).


This model generates smooth temperature changes as in the second figure above. In the
paper of Allen you will find a FORTRAN code that can be used for estimation of
degree-days.

8.5. Other non-linear models of development


Other non-linear models of development
Improved degree-day model inherited a linear relationship between development rate
and temperature (between two temperature limits tmin and tmax) from the simple degree-
day model. Now it is time to make the next logical step: not to use linear relationships at
all. Below two alternative models are shown: the logistic model and the normal
distribution:
Non-linear models like the logistic model or the normal distribution are easy to interpret
for experiments with constant temperature. In this case, development rate can be defined
as a reciprocal of development time (v = 1/T). But what is development rate if
temperature changes? We cannot accumulate degree-days any more because the model
is non-linear. In attempt to answer this question a new concept has been developed: a
concept of physiological time.

8.6. Physiological time


What is time? This is the most fundamental question in science. It attracted the best
brains in human history from St. Augustin to Einstein. What is the relationship between
time and dynamics? Do we measure dynamics relative to time, or may be time should
be measured with dynamics? Newtonian physics considered absolute and universal time
which was above nature. Alternative point of view is that each system has its own
individual time measured by the number of events. For example, time in human life can
be measured by the number of decisions made. A person who lives passively, and
avoids making decisions, remains a child psychologically even if he is biologically an
adult. Further discussion on the problem of biological time can be found in the paper of
Sharov (1995).

We can apply the concept of biological time to the development of poikilothermous


organisms. The progress in organism's development can be viewed as a biological clock
that measures physiological time. Physiological time goes fast when it is warm and slow
when it is cold. Then, development rate can be defined as advance of physiological time
per unit of calendar time. For example, if development rate of insect larvae is 0.07 per
day then, taking the entire physiological time at larval stage as 100%, in one day the
insect will increment its physiological time by 7%.

To determine the duration of development we will accumulate development rates rather


than effective temperatures. Development completes on the day when accumulated
development rates reach 1 (=100%), i.e., the following equation is true:

where v(t) is the rate of development as a function of temperature, t; t(x) is temperature


as a function of time, x; and T is the development time. If temperature is defined as a
function of time (see the graph below), then at each time the rate of development is
estimated as a non-linear function of time. Finally the integral of function v(t(x)) (the
area under the curve) from 0 to T should be equal to 1.
The discrete-time version of the previous equation is:

How to measure physiological time?


There is no mechanical watch that measures physiological time. Also there are no hour
and minute hands inside an organism. However, in some cases it is possible to find
indicators of physiological time, e.g., concentration of hormones and metabolites, CO2
emission rate, etc. However, these indicators are usually not very accurate. For example,
the respiration rate is high in gypsy moth eggs for ca. 2 weeks after oviposition; then it
decreases and stay at low level. Thus, it is possible to distinguish between these two
periods of egg development.

The problem of measurement can be evaded by preparing organisms of specific


physiological age for the experiment. This is analogues to quantum mechanics where it
is impossible to measure exact location of an electron without changing its momentum.
But it is possible to prepare electrons with specific characteristics. To prepare organisms
in a specific physiological age, we need to keep them in standard conditions for a
specific time. For example, if the duration of insect pupal stage is 10 days in 25°, then
after 5 days in this temperature, pupae will be in the middle of their pupal stage.

What are the advantages of physiological time concept?


As compared to the degree-day model, the concept of physiological time has the
following advantages:

• It makes no arbitrarily assumptions on the shape of development rate function.


This function can be linear, logistic, normal distribution, etc.
• It admits the effect of other factors besides temperature on the rate of
development. The rate of development may depend on food quality and quantity,
photoperiod, and other factors.
Before the concept of physiological time was developed there were attempts to
incorporate the effect of factors other than temperature into the degree-day model. It
was assumed that additional factors change the amount of degree-days required to
complete development. For example, gypsy moth larvae develop faster on preferred tree
species (oaks) as compared to non-preferred species (e.g., maples). It was assumed that
more degree days should be accumulated by gypsy moth larvae on maple than on oak
trees to pupate. This model does not work if larval migration is considered among trees
because degree-days accumulated on different tree species are not comparable. For
example, if larvae move from an oak tree to a maple tree, they become "younger". If
they were ready to pupate on oak before migration, then after migration to maples they
still have to accumulate additional degree-days to pupate.

8.7. How to combine physiological time with the model


of Leslie?
The model of Leslie can be easily modified to incorporate physiological time. In the
original model age and time were equivalent because each time step organisms were
moving to the next age class. Now age will be measured in units of physiological time!
Each column and row in the matrix corresponds to a specific physiological age.

Progress in physiological age in one time step may depend on temperature (3 arrows at
the left side of this graph).

Also, individual variation in development rate (distributed delays) can be taken into
account (branching arrow at the right side of the graph).

At the start of simulation all organisms can be placed into the first age class. Another
option is to add variability in the starting date of development. For example, if we
simulate insect larval development, it is unrealistic to assume synchronous egg hatch in
one day. It is better to assume distributed egg hatch time. Three kinds of distributions
are used most often: normal, logistic, and Weibul.

Cumulative distributions are described by equations:


Two graphs below show how to convert cumulative probability function of egg hatch
time into the proportion of eggs that hatch each day (logistic and Weibull distributions):

Normal and logistic distributions are both symmetrical and are very similar. But
Weibull distribution is asymmetrical. Actual distribution of egg hatch time is often
asymmetrical, and thus, the Weibul distribution is usually better than the normal and
logistic distributions.

8.8. Questions and Assignments to Lecture 8.


8.1 Insect development rate. Development of pea weevil (Bruchus pisorum) eggs was
studied in the laboratory at constant temperatures (Smith, A.M. 1992. Environ.
Entomol. 21:314-321):
Temperature, C Egg development time, days

10.7 38.0

14.4 19.5

16.2 15.6

18.1 9.6

21.4 9.5

23.7 7.3

24.7 4.5

26.9 4.5
28.6 7.1

1. Plot development rate vs. temperature (use Excel)


2. Use linear regression to estimate lower temperature threshold and degree days
required for egg development.
3. Average temperature in a sequence of days was: 15 20 25 20 15 10 15 10 15 20
15 20 15 10 15 20 25 20 15 20
4. When do you expect eggs to hatch if they were laid on the first day? (Note:
ignore diurnal temperature change).
Lecture 9. Stability, Oscillations and Chaos in
Population Dynamics
9.1. Introduction. Main Problems
Some populations have considerable oscillations in their numbers. There are well-
known cycles in rodent populations. Many insect pests have regular or irregular
outbreaks, e.g., larch budmoth in Switzerland, spruce budworm in Canada, gypsy moth
in Europe and in USA.

Example: Gypsy moth population dynamics in Yugoslavia:

Gypsy moth numbers increased by several orders of magnitude. Pest outbreaks resulted
in forest defoliation in large areas.

Main Problems:

• To evaluate population stability using models


• To examine effects of different factors on population stability

9.2. Attractors and Their Types


When the dynamics of a population model is traced over a large number of generations,
then the model exhibits an asymptotic behavior which is almost independent from
initial conditions (e.g., initial population density). Asymptotic trajectory is called
"attractor" because model trajectories converge to this asymptotic trajectory.

There may be several attractors in a model. In this case each attractor has a domain of
attraction. Model trajectory converges to that attractor in which domain initial
conditions were located.
In this example, there are two attractors: a limit cycle (at the left) and a stable
equilibrium (at the right). Domains of attraction are colored blue, they never overlap.
For different starting places (initial conditions), trajectories converge to different
attractors.

Types of attractors:

1. Stable equilibrium (=steady state)


2. Limit cycle
3. Chaos

Examples:

This is a chaotic attractor (Lorenz attractor)

9.3. Equilibrium: Stable or Unstable?


Equilibrium is a state of a system which does not change.

If the dynamics of a system is described by a differential equation (or a system of


differential equations), then equilibria can be estimated by setting a derivative (all
derivatives) to zero.

Example: Logistic model

To find equilibria we have to solve the equation: dN/dt = 0:


This equation has two roots: N=0 and N=K. An equilibrium may be stable or unstable.
For example, the equilibrium of a pencil standing on its tip is unstable; The equilibrium
of a picture on the wall is (usually) stable.

An equilibrium is considered stable (for simplicity we will consider asymptotic stability


only) if the system always returns to it after small disturbances. If the system moves
away from the equilibrium after small disturbances, then the equilibrium is unstable.

The notion of stability can be applied to other types of attractors (limit cycle, chaos),
however, the general definition is more complex than for equilibria. Stability is
probably the most important notion in science because it refers to what we call "reality".
Everything should be stable to be observable. For example, in quantum mechanics,
energy levels are those that are stable because unstable levels cannot be observed.

Now, let's examine stability of 2 equilibria points in the logistic model.

In this figure, population growth rate, dN/dt, is plotted versus population density, N.
This is often called a phase-plot of population dynamics. If 0 < N < K, then dN/dt > 0
and thus, population grows (the point in the graph moves to the right). If N < 0 or N > K
(of course, N < 0 has no biological sense), then population declines (the point in the
graph moves to the left). The arrows show that the equilibrium N=0 is unstable, whereas
the equilibrium N=K is stable. From the biological point of view, this means that after
small deviation of population numbers from N=0 (e.g., immigration of a small number
of organisms), the population never returns back to this equilibrium. Instead, population
numbers increase until they reach the stable equilibrium N=K. After any deviation from
N=K the population returns back to this stable equilibrium.

The difference between stable and unstable equilibria is in the slope of the line on the
phase plot near the equilibrium point. Stable equilibria are characterized by a negative
slope (negative feedback) whereas unstable equilibria are characterized by a positive
slope (positive feedback).
The second example is the bark beetle model with two stable and two unstable
equilibria. Stable equilibria correspond to endemic and epidemic populations. Endemic
populations are regulated by the amount of susceptible trees in the forest. Epidemic
populations are limited by the total number of trees because mass attack of beetle
females may overcome the resistance of any tree.

Stability of models with several variables


Detection of stability in these models is not that simple as in one-variable models. Let's
consider a predator-prey model with two variables: (1) density of prey and (2) density
of predators. Dynamics of the model is described by the system of 2 differential
equations:

This is the 2-variable model in a general form. Here, H is the density of prey, and P is
the density of predators. The first step is to find equilibrium densities of prey (H*) and
predator (P*). We need to solve a system of equations:

The second step is to linearize the model at the equilibrium point (H = H*, P = P*) by
estimating the Jacobian matrix:

Third, eigenvalues of matrix A should be estimated. The number of eigenvalues is equal


to the number of state variables. In our case there will be 2 eigenvalues. Eigenvalues are
generally complex numbers. If real parts of all eigenvalues are negative, then the
equilibrium is stable. If at least one eigenvalue has a positive real part, then the
equilibrium is unstable.

Eigenvalues are used here to reduce a 2-dimensional problem to a couple of 1-


dimensional problem problems. Eigenvalues have the same meaning as the slope of a
line in phase plots. Negative real parts of eigenvalues indicate a negative feedback. It is
important that ALL eigenvalues have negative real parts. If one eigenvalue has a
positive real part then there is a direction in a 2-dimensional space in which the system
will not tend to return back to the equilibrium point.

There are 2 types of stable equilibria in a two-dimensional space: knot and focus

There are 3 types of unstable equilibria in a two-dimensional space: knot, focus, and
saddle

Stability in discrete-time models


Consider a discrete-time model (a difference equation) with one state variable:

This model is stable if and only if :

where is the slope of a thick line in graphs below:


You can check this yourself using the following Excel spreadsheet:

Excel spreadsheet "ricker.xls"

If the slope is positive but less than 1, then the system approaches the equilibrium
monotonically (left). If the slope is negative and greater than -1, then the system
exhibits oscillations because of the "overcompensation" (center). Overcompensation
means that the system jumps over the equilibrium point because the negative feedback
is too strong. Then it returnes back and again jumps over the equilibrium.

Continuous-time models with 1 variable never exhibit oscillations. In discrete-time


models, oscillations are possible even with 1 variable. What causes oscillations is the
delay between time steps. Overcompensation is a result of large time steps. If time steps
were smaller, then the system would not jump over the equilibrium but will approach to
it gradually.

Now we will analyze stability in the Ricker's model. This model is a discrete-time
analog of the logistic model:

First, we need to find the equilibrium population density N* by solving the equation:

This equation is obtained by substituting Nt+1 and Nt with the equilibrium population
density N* in the initial equation. The roots are: N* = 0 and N* = K.. We are not
interested in the first equilibrium (N* = 0) because there is no population. Let's estimate
the slope df/dN at the second equilibrium point:
Now we can apply the condition of stability:

-1 < 1 - r < 1

0<r<2

Thus, the Ricker's model has a stable equilibrium N* = K if 0 < r < 2.

If a discrete time model has more than one state variable, then the analysis is similar to
that in continuous-time models. The first step is to find equilibria. The second step is to
linearize the model at the equilibrium state, i.e., to estimate the Jacobian matrix. The
third step is to estimate eigenvalues of this matrix. The only difference from continuous
models is the condition of stability. Discrete-time models are stable (asymptotically
stable) if and only if all eigenvalues lie in the circle with the radius = 1 in the complex
plain.

9.4. Quantitative Measures of Stability


In the previous section we have discussed qualitative indicators of stability. According
to these indicators, a model is either stable if its trajectory converges to an equilibrium
state or unstable if it diverges from the equilibrium after small disturbances. However,
real populations never converge to an equilibrium because of the random noise
associated with weather and other stochastic factors. Thus, qualitative stability has a
vague biological meaning. Ecologists are more interested in quantitative indicators of
stability which represent the ability of the population to resist environmental
fluctuations.

Robert May (1973) suggested to measure system stability by the maximum real part of
eigenvalues of the linearized model. It was shown that this value correlates with the
variance of population fluctuations in stochastic models.

Sharov (1991, 1992) suggested measures of m- and v-stability that characterize the
stability of the mean (m) and variance (v) of population density (initially these measures
were called as coefficients of buffering and homeostasis, see Sharov [1985, 1986]).
Later they were re-invented by Ives (1995a, 1995b). They can be used to predict the
effect of environmental changes (e.g., global warming or pest management) on the
mean and variance of population numbers

M-stability (MS) was defined as the ratio of the change in mean log population density,
N, as a response to the change in mean value of some environmental factor, v.
M-stability is the reciprocal of the sensitivity of mean population density to the mean
value of factor v. Log-transformation of population density is important because it
makes population models closer to linear.

For example, if v is temperature which is going to change by 2 degrees due to global


warming, and log population density per ha (log base e) will increase from 1 to 1.5, then
the sensitivity is S=(1.5-1)/2=0.25, and m-stability MS=4. The population with higher
m-stability will change less than the population with low m-stability under the same
changes in average factors.

Strong population regulation increases m-stability because regulating mechanisms will


resist to the changes in population density. Let's assume that regulation is caused by
interspecific competition. Then, if conditions become favorable for the population, then
the organisms will increase their reproduction rate. However, as population density
increases, mortality due to competition increases too and partially compensates
increased reproduction rates. If conditions become less favorable, then density will
decline and mortality due to competition will decrease and partially compensate the
decrease in reproduction rates.

If population dynamics is described by a mathematical model then m-stability can be


estimated from that model. The simplest example is the logistic model. Mean
population density in the logistic model equals to carrying capacity, K. If the factor v
affects K, then . If the factor v affects population growth rate, r,
but does not affect carrying capacity, then mean population density will not respond to
factor change, and thus, m-stability will be infinitely large.

V-stability (VS) was defined as a ratio of the variance of additive random noise to
the variance of log population numbers :

Population that has smaller fluctuations of population numbers than another population
that experience the same intensity of additive environmental noise has a higher v-
stability. To estimate v-stability in the Ricker's model we can use the linearized model
at the equilibrium point:

where N is log population density, and is the white noise with a zero mean. Noise is
not correlated with log population numbers. Thus:
This graph shows that v-stability equals to zero at r=0 and r=2 (these are the boundaries
of quantitative stability). V-stability has a maximum at r = 1.

References

Ives, A.R. 1995. Predicting the response of populations to environmental change.


Ecology 76: 926-941.
Ives, A.R. 1995. Measuring resilience in stochastic systems. Ecol. Monogr. 65: 217-
233.
May, R.M. 1973. Stability in randomly fluctuating versus deterministic environments.
Amer. Natur., 107: 621 650.
Sharov, A.A. 1985. Insect pest population management taking into account natural
mechanisms of population dynamics. Zoologicheskii Zhurnal (Zoological Journal), 64:
1298 1308 (in Russian).
Sharov, A.A. 1986. Population bufferity and homeastasis and their role in population
dynamics. Zhurnal Obshchej Biologii (Journal of General Biology), 47: 183 192 (in
Russian).
Sharov, A.A. 1991. Integrating host, natural enemy, and other processes in population
models of the pine sawfly. In: Y.N.Baranchikov et al. [eds.] Forest Insect Guilds:
Patterns of Interaction with Host Trees. U.S. Dep. Agric. For. Serv. Gen. Tech. Rep.
NE-153. pp. 187-198.
Sharov, A.A. 1992. Life-system approach: a system paradigm in population ecology.
Oikos 63: 485-494. Get a reprint! (PDF)

9.5. Limit Cycles and Chaos


We found that Ricker's model is stable if 0 < r < 2. The question is what happens to
model population when stability is lost. The first suggestion is that population will get
extinct. But this is the wrong answer. Below there are simulations of population
dynamics using Ricker's model with different values of r:

r = 0.5 Monotonous increase in numbers.


K = 200

r = 1.9 Damping oscillations.


K = 200
r = 2.3 Limit cycle with period = 2.
K = 200

r = 2.6 Limit cycle with period = 4.


K = 200

r = 3.0 Chaos.
K = 200

In the two upper figures the model has a stable equilibrium, only the patterns of
approaching the equilibrium are different. In the three lower figures there is no stable
equilibrium. Non-equilibrium dynamics may be of 2 types: a limit cycle when the
trajectory repeats itself, and chaotic when the trajectory does not repeat itself.

The bifurcation plot (below) helps to visualize all types of dynamics generated by the
Ricker's model. This graph is plotted as follows: for each value of parameter r which is
incremented with 0.05 steps, population dynamics was simulated for 200 generations.
First 125 generations were discarded because the population may not have reached the
asymptotic behavior. Population numbers in the rest 75 generations are plotted versus
the value of parameter r:

• r < 2 Stable equilibrium


• r = 2 Bifurcation into a 2-point limit cycle
• r = 2.5 Bifurcation into a 4-point limit cycle
• Then there is a series of cycle duplication: 8 points, 16 points, etc.
• r = 2.692 Chaos ("smear" of points)
• For r > 2.7 there are some regions where dynamics returns to a limit cycle, e.g.,
r=3.15.
Chaotic dynamics looks like stochastic noise, however the model is absolutely
deterministic. Chaotic models are widely used for random number generation in
computers.

One of the questions that are often discussed in ecological literature is "does chaos
really exists in population dynamics?". The major argument in favor of chaos is: when
model parameters are fit to known time series of population dynamics then the
dynamics of the model with these parameters is chaotic. Another kind of arguments is
based on attempts to separate chaotic dynamics from stochastic noise. However
detection of chaos is difficult because of several problems:

• There is no evidence that the model is correct. Usually these models ignore
many ecological processes (natural enemies, etc.). It is necessary to use multiple
models for detecting chaos (Ellner and Turchin 1995, Amer. Natur. 145: 343-
375).
• The confidence interval for parameter values is usually large enough to cover
both chaotic and non-chaotic (limit cycle) model dynamics.
• Time series of population dynamics are usually not long enough to separate
chaotic dynamics from stochastic noise.

At this point, chaotic dynamics was detected consistently only in a few microtine
populations (Ellner and Turchin 1995). I suspect that in these cases chaotic dynamics
was induced by the seasonal cycle in population numbers. Chaos was never detected in
time series with 1 year as a time step. Probably chaos is a rare phenomenon in
population dynamics.

Questions and Assignments to Lecture 9


9.1. Population dynamics is simulated by the differential equation:

Draw schematically the phase portrait (rate of population growth vs. population
density). Find all equilibrium points and characterize their stability.

9.2. Characterize the type of attractor in population dynamics: (a) stable equilibrium; (b)
limit cycle, (c) chaos.
Lecture 10. Predation and parasitism
10.1. Introduction
Predation and parasitism are examples of antagonistic ecological interactions in which
one species takes advantage of another species. Predators (see a picture) use their prey
as a source of food only, whereas parasites (see a picture) use their hosts both as a food
and as a habitat. Predation and parasitism are stage-specific interactions rather than
species-specific. Many species are predators or parasites only on specific stages in their
life cycle.

Importance of the study of predation and parasitism:

• In many species predation and parasitism are dominating among ecological


processes. Dynamics of these populations cannot be predicted and understood
without considering natural enemies.
• Pest species of insects and weeds can be suppressed by introduction of natural
enemies or by inundative release of natural enemies (biological control).
• Natural enemies may cause side effects in pesticide applications. The numbers
of arthropod natural enemies may be reduced due to pesticide treatment which
may result in increasing of pest populations.

10.2 Lotka-Volterra Model


Lotka-Volterra model is the simplest model of predator-prey interactions. The model
was developed independently by Lotka (1925) and Volterra (1926):

It has two variables (P, H) and several parameters:


H = density of prey
P = density of predators
r = intrinsic rate of prey population increase
a = predation rate coefficient
b = reproduction rate of predators per 1 prey eaten
m = predator mortality rate

Measuring parameters of the Lotka-Volterra model


The following set of experiments should be done:

1. Keep prey population without predators and estimate their intrinsic rate of
increase (r).
2. Put one predator in cages with different densities of prey and estimate prey
mortality rate and corresponding k-value in each cage. As we know, k-value
equals to the instantaneous mortality rate multiplied by time. Thus, the predation
rate (a) equals to the k-value divided by the duration of experiment.
Example: lady-beetle killed 60 aphids out of 100 in 2 days. Then,
the k-value = -ln(1-60/100) = 0.92, and a = 0.92/2 = 0.46.
Note: if a -values estimated at different prey densities are not close
enough to each other, then the Lotka-Volterra model will not
work! However, the model can be modified to incorporate the relation of a to
prey density.
3. Estimation of parameters b and m:

Keep constant density of prey (e.g., H = 0, 5, 10, 20, 100 prey/cage), and
estimate the intrinsic rate of predator population increase (rP) at these densities
of prey. Plot the intrinsic rate of predator population increase versus prey
density: The linear regression of this line is:

Note: If points do not fit to a straight line (e.g., the intrinsic rate of predator
population growth may level off), then the Lotka-Volterra model is not adequate
and should be modified. Now , parameters b and m can be taken from this
regression equation.

How to solve differential equations


There are two major approaches: analytical and numerical. Analytical methods are
complicated and require good mathematical skills. Also, many differential equations
have no analytical solution at all. Numerical methods are easy and more universal
(however, there are problems with convergence).

The simplest and least accurate is the Euler's method. Consider a stationary differential
equation:

First we need initial conditions. We will assume that at time to the function value is
x(to).

Now we can estimate x-values at later (or earlier) time using equation:

On this graph we estimate


the slope f(x) of the
function at point t = to and
extrapolate this slope
through the entire time
interval.
The main source of error in the Euler's method is estimation of derivative at the start of
time interval. The direction of actual solution may change drastically during this time
interval and numerically predicted point could be far from the actual solution (see the
figure).

Euler's method can be improved, if the derivative (slope) is estimated at the center of
time interval . However, the derivative at the center depends on the function value at
the center which is unknown. Thus, first we need to estimate the function value at the
middle point using simple Euler's method, and then we can estimate the derivative at the
middle point.

k is the function value in the center of time interval l . Finally, we can estimate
function value at the end of the time interval:

This is called Runge-Kutta


method of second order. The
most popular is the Runge-
Kutta method of the fourth
order. However, for our
purposes it is enough to use
the second order method.

This method is applied to Lotka-Volterra equations in the following Excel spreadsheet:

. Excel spreadsheet "lotka.xls"

First, we estimate prey and predator densities (H' and P', respectively) at the center of
time interval:

The second step is to estimate prey and predator densities (H" and P" at the end of time
step l :
These two graphs were plotted using the same model parameters. The only difference is
in initial density of prey. This model has no asymptotic stability, it does not converge to
an attractor (does not "forget" initial conditions).

This figure shows relative changes in prey predator density for both initial conditions.
Trajectories are closed lines.

The model of Lotka and Volterra is not very realistic. It does not consider any
competition among prey or predators. As a result, prey population may grow infinitely
without any resource limits. Predators have no saturation: their consumption rate is
unlimited. The rate of prey consumption is proportional to prey density. Thus, it is not
surprising that model behavior is unnatural showing no asymptotic stability. However
numerous modifications of this model exist which make it more realistic.

Additional information on the Lotka-Volterra model can be found at other WWW sites:

• Interactive Simulation Server (Lund)


• Model Description and Realization on "STELLA"
• Model Realization on "Models"

References:

Lotka, A. J. 1925. Elements of physical biology. Baltimore: Williams & Wilkins Co.
Volterra, V. 1926. Variazioni e fluttuazioni del numero d'individui in specie animali
conviventi. Mem. R. Accad. Naz. dei Lincei. Ser. VI, vol. 2.

10.3. Functional and Numerical Response


Holling (1959) studied predation of small mammals on pine sawflies, and he found that
predation rates increased with increasing prey population density. This resulted from 2
effects: (1) each predator increased its consumption rate when exposed to a higher prey
density, and (2) predator density increased with increasing prey density. Holling
considered these effects as 2 kinds of responses of predator population to prey density:
(1) the functional response and (2) the numerical response.

Modeling Functional Response


Holling (1959) suggested a model of functional response which remains most popular
among ecologists. This model is often called "disc equation" because Holling used
paper discs to simulate the area examined by predators. Mathematically, this model is
equivalent to the model of enzime kinetics developed in 1913 by Lenor Michaelis and
Maude Menten.

This model illustrates the principal of time budget in behavioral ecology. It assumes that
a predator spends its time on 2 kinds of activities:

1. Searching for prey


2. Prey handling which includes: chasing, killing, eating and digesting.

Consumption rate of a predator is limited in this model because even if prey are so
abundant that no time is needed for search, a predator still needs to spend time on prey
handling.

Total time equals to the sum of time spent on searching and time spent on handling::

Assume that a predator captured Ha prey during time T. Handling time should be
proportional to the number of prey captured:

where Th is time spent on handling of 1 prey.

Capturing prey is assumed to be a random process. A predator examines area a per time
unit (only search time is considered here) and captures all prey that were found there.
Parameter a is often called "area of discovery", however it can be called "search rate" as
well.

After spending time Tsearch for searching, a predator examines the area = a Tsearch, and
captures aHTsearch prey where H is prey density per unit area:

Hence:
Now we can balance the time budget:

The last step is to find the number of attacked prey Ha:

The graph of functional response that corresponds to this equation is shown below:

This function indicates the number of prey killed by 1 predator at various prey densities.
This is a typical shape of functional response of many predator species. At low prey
densities, predators spend most of their time on search, whereas at high prey densities,
predators spend most of their time on prey handling.

Holling (1959) considered 3 major types of functional response:


Type I functional response is found in passive predators like spiders. The number of
flies caught in the net is proportional to fly density. Prey mortality due to predation is
constant (right graph on the previous page).

Type II functional response is most typical and corresponds to the equation above.
Search rate is constant. Plateau represents predator saturation. Prey mortality declines
with prey density. Predators of this type cause maximum mortality at low prey density.
For example, small mammals destroy most of gypsy moth pupae in sparse populations
of gypsy moth. However in high-density defoliating populations, small mammals kill a
negligible proportion of pupae.

Type III functional response occurs in predators which increase their search activity
with increasing prey density. For example, many predators respond to kairomones
(chemicals emitted by prey) and increase their activity. Polyphagous vertebrate
predators (e.g., birds) can switch to the most abundant prey species by learning to
recognize it visually. Mortality first increases with prey increasing density, and then
declines.

If predator density is constant (e.g., birds, small mammals) then they can regulate prey
density only if they have a type III functional response because this is the only type of
functional response for which prey mortality can increase with increasing prey density.
However, regulating effect of predators is limited to the interval of prey density where
mortality increases. If prey density exceeds the upper limit of this interval, then
mortality due to predation starts declining, and predation will cause a positive feed-
back. As a result, the number of prey will get out of control. They will grow in numbers
until some other factors (diseases of food shortage) will stop their reproduction. This
phenomenon is known as "escape from natural enemies" discovered first by Takahashi.

Estimation of Parameters of Functional Response


Experiments should be done as follows: predators are kept in large-size cages
individually. Large-size cages are important because search abilities of predators should
be limited. Different number of prey are released in these cages. Each prey density
should be replicated to get sufficient accuracy. More experiments should be done with
low prey density than with high prey density because the error of mortality estimates
depends on the total number of prey. Experiments are usually set for a fixed time
interval. At the end of experiments, survived prey are counted in each cage.

This is an example of experimental data:

No. of
Total Average no.
prey per No. of
prey of prey killed 1/Ha 1/(HT)
cage replications
killed Ha
H
5 20 50 2.5 0.400 0.1000
10 10 40 4.0 0.250 0.0500
20 7 55 7.9 0.127 0.0250
40 5 45 9 0.111 0.0125
80 3 38 12.6 0.079 0.0062
160 3 35 11.6 0.086 0.0031

Cage area was 10 sq.m., and duration of experiment was T=2 days.
Holling's equation can be transformed to a linear form:

The linear regression has the following coefficients:

y = 3.43 x + 0.0612

Th = 0.0612 T = 0.1224 days = 2.9 hours

a = 1/3.43 = 0.29 cages = 2.9 sq.m.

Another possible method of parameter estimation is non-linear regression. It may give


better results at high prey density than the linear regression method.

Type III functional response can be simulated using the same Holling's equation with
search rate (a) dependent on prey density, e.g.:

Numerical Response
Numerical response means that predators become more abundant as prey density
increases. However, the term "numerical response" is rather confusing because it may
result from 2 different mechanisms:

1. Increased rate of predator reproduction when prey are abundant (numerical


response per se)
2. Attraction of predators to prey aggregations ("aggregational response")

Reproduction rate of predators naturally depends on their predation rate. The more prey
consumed, the more energy the predator can allocate for reproduction. Mortality rate
also reduces with increased prey consumption.

The most simple model of predator's numerical response is based on the assumption that
reproduction rate of predators is proportional to the number of prey consumed. This is
like conversion of prey into new predators. For example, as 10 prey are consumed, a
new predator is born.

Aggregation of predators to prey density is often called "aggregational response". This


term is better than "numerical response" because it is not ambiguous. Aggregational
response was shown to be very important for several predator-prey systems. Predators
selected for biological control of insect pests should have a strong aggregational
response. Otherwise they would not be able to suppress prey populations. Also,
aggregational response increases the stabilility of the spatially-distributed predator-prey
(or host-parasite) system.

References:

Holling, C.S. 1959. The components of predation as revealed by a study of small


mammal predation of the European pine sawfly. Canad. Entomol. 91: 293-320.
Holling, C.S. 1959. Some characteristics of simple types of predation and parasitism.
Canad. Entomol. 91: 385-398.

10.4. Predator-Prey Model with Functional and


Numerical Responses
Now we are ready to build a full model of predator-prey system that includes both the
functional and numerical responses.

We will start with the prey population. Predation rate is simulated using the Holling's
"disc equation" of functional response:

The rate of prey consumption by all predators per unit time equals to

The equation of prey population dynamics is:

Here we assumed that without predators, prey population density increases according to
logistic model.

Predator dynamics is represented by a logistic model with carrying capacity


proportional to the number of prey:

This equation represents the numerical response of predator population to prey density.

The model was built using an Excel spreadsheet:

Excel spreadsheet "predfunc.xls"


You can modify parameters of this model to simulate various patterns of population
dynamics. Differential equations are solved numerically, and it may happen that the
algorithm (2nd order Runge-Kutta method) will not work for some combination of
parameter values. Thus, change parameters with caution. If you suspect that the
algorithm does not work properly, reduce the time step (cell A5) until results become
independent from the time step.

Simulation results are presented below. This model exhibits more various dynamic
regimes than the Lotka-Volterra model.

r = 0.2
H

K = 500
a=
No
0.001 oscillations
Th = 0.5
r = 0.1
P

k = 0.2
r = 0.2
H

K = 500 Damping
oscillations
a = 0.1
converging to
Th = 0.5 a stable
r = 0.1
P
equilibrium
k = 0.2

r = 0.2
H

K = 500
a = 0.3
Limit cycle
Th = 0.5
r = 0.1
P

k = 0.2

This model can be used to simulate biological control. The goal of biological control is
to suppress the density of the pest population using natural enemies. We will assume
that the prey in our model is a dangerous pest, and that the predator was introduced to
suppress its density. Withour predators the density of prey population is equal to the
carrying capacity, K = 500. After a predator with a search rate, a = 0.001, was
introduced, the equilibrium population density, N*, declined to the value of 351.
Beddington et al. (1978, Nature, 273: 573-579) suggested to measure the degree of pest
suppression by the ratio:

For example, if a = 0.001, then q = N*/K = 351/500 = 0.7. Biological control is


successful if the value of q is low (at least, <0.5). If we increase the search rate of the
predator (i.e., we introduced a more effective predartor species) to a = 0.1, then the pest
population is suppressed to the density of 19 (q = 0.038).

It could be expected that more effective predators will cause more suppression of the
prey population density. But this is not true, because more effective natural enemies
also cause larger oscillations in population density. For example, at a = 0.3, the
equailibrium is not stable and populations exhibit periodic cycles (see graph #3 above).
Periodically host density reaches the value of 190.

The transition between the stable equilibrium and the limit cycle occurs approximately
at a = 0.244. The transition from one type of dynamics to another one is often called
"phase transition" (e.g., transitions between liquid and gas phases or between solid and
liquid phases). The phase transition in our model will be less ubrupt if we introduce
noise. With noise, the system will exhibit oscillations even if the equilibrium is stable.
The closer we are to the critical value, a = 0.244, the larger will be these oscillations.
These oscillations result from the interaction of predators with prey.

This example illustrates that pest regulation (or control) by natural enemies is an
ambibuous notion. First, it does not refer to the type of dynamics (stable equilibrium vs.
limit cycle), and second, the excess of "regulation" may cause large oscillations of prey
density.

10.5. Host-Parasitoid Models


Parasitoids are insect species which larvae develop as parasites on other insect species.
Parasitoid larvae usually kill its host (some times the host is paralyzed by ovipositing
parasitoid female) whereas adult parasitoids are free-living insects (see images of
parasitoids). Most of parasitoid species are either wasps or flies.

Parasitoids and their hosts often have synchronized life-cycles, e.g., both have one
generation per year (monovoltinous). Thus, host-parasite models usually use discrete
time steps that correspond to generations (years).

Model of Thompson (1922)


The model assumes that female parasitoids lay their eggs randomly on host individuals
and do not distinguish between healthy and already parasitized hosts. In this case, the
number of parasitoid eggs laid on one host should have a poisson distribution:

where p(i) is the proportion of hosts that get i parasitoid eggs, and M is the mean
number of parasitoid eggs per one host.

Survived hosts are those which get 0 parasitoid eggs. The proportion of survived hosts
is equal to p(0) = exp(-M).

Variables:

• P = Density of parasitoid females


• H = Density of hosts

Parameter

• F = Parasitoid fecundity (no. of eggs laid by 1 female)

PF = Density of eggs laid by all parasitoid females per unit area

= Average no. of eggs per host individual


Then, host survival is

The full model is:

The first equation describes host survival and reproduction. The numbers of survived
hosts are multiplied by Ro which means reproduction.

In the second equation, each parasitized host produce one adult parasitoid in the next
generation. P is the density of females only. Thus, the numbers of parasitoids is
multiplied by the proportion of females = q.

In the model of Thompson, it is assumed that parasites always lay all their eggs. Thus,
realized fecundity equals potential fecundity. This assumption implies unlimited search
abilities of parasitoids. In nature, parasites often do not realize their potential fecundity
just because they can not find enough hosts. Thus, the model of Thompson may
overestimate parasitism rates especially if host density is low.

Model of Nicholson and Bailey (1935)


This model is more realistic than the Thompson's model and is widely used by
ecologists. It assumes that parasitoid female is able to examine area a ("area of
discovery") during its life time. When a host is found, parasitoid lays only one egg in it.
However, the same host can be found again later and then the parasite will lay another
egg in it because we assume that parasites do not distinguish between healthy hosts and
already parasitized hosts.

Because each encounter with the host results in depositing 1 egg, the realized fecundity
equals the product of the area of discovery and host density: F = aH. Substituting this
value of F into the Thompson model we get:
In the Nicholson and Bailey model, the potential fecundity of parasites is not limited.
Parasites lay an egg at every encounter with the host even if the number of encounters is
very large (e.g., if host density is high). Thus, this model may overestimate parasitism
rates at high host density.

Model of Rogers (1972)


The model of Rogers applies the model of Holling, which was originally developed for
predator-prey systems, to host- parasite systems. It assumes two kinds of limitations in
host-parasitoid interactions: limited parasitoid fecundity (as in the model of Thompson)
and limited search rate (as in the model of Nicholson and Bailey).

We will use the Holling's disc equation (see section 10.3) to model the functional
response of parasitoids. The number of hosts attacked by one parasitoid female is equal
to

We can modify this equation by setting T=1 because search rate is considered per life
time of parasitoid female. Life time can be coded as 1 because the time step is equal to 1
generation. The ratio is the maximum fecundity of parasitoid female.
Then:

When parasitoid female attacks a host it lays an egg. Thus, realized fecundity F = Ha.
Substituting this value of F into the Thompson's model we get:

In the model of Rogers, realized fecundity is different from the potential fecundity
whereas in previous models this distinction was not present.

All models of host-parasitoid system are unstable: they generate oscillations with
increasing amplitude.
This is the dynamics of the model of Nicholson and Bailey.

However, in nature host-parasitoid population never show oscillations with infinitely


increasing amplitude. This is not because the models do not capture the mechanisms of
host-parasitoid interactions, but because additional ecological processes (e.g.,
intraspecific competition in hosts or in parasitoids) can partially or completely stabilize
the system. It was also shown that spatial heterogeneity and parasitoid dispersal among
host patches may also stabilize the population system.

References:

Nicholson, A. J., and V. A. Bailey. 1935. The balance of animal populations.


Proceedings of the Zool. Soc. of London. 1: 551-598.
Rogers, D. J. 1972. Random search and insect population models. J. Animal Ecol. 41:
369-383.
Thompson, W. R. 1929. On the relative value of parasites and predators in the
biological control of insect pests. Bull. Entomol. Res. 19: 343-350.

10.6. Host-Pathogen Model (Anderson & May)


Host-pathogen models are similar to predator-prey and host-parasite models. Below is
the model of Anderson and May (1980, 1981) which describes insect diseases. The host
population consist of two portions: susceptibles which are healthy organisms, and
infected individuals. The model describes changes in density of susceptibles (S),
infected individuals (I) and pathogens (P):
This model is capable to generate epidemic cycles. It was used to study evolutionary
strategies of pathogens.

Models of epidemics in mammalian hosts (including humans) consider immune


organisms as a separate category.

Host-pathogen systems may include vectors. For example, malaria is transmitted by


mosquitoes. In these systems, hosts become infected only when they have contact with
the vector. Thus, the number of pathogens is not that important as the numbers of
vectors carrying the patogens. The host-vector-pathogen system can be described as the
change in numbers of 4 kinds of individuals: healthy hosts, infected hosts, uninfected
vectors, and infected vectors. An example of such a model is given in the following
Excel spreadsheet:

Excel spreadsheet "vector.xls"

References:

Andreson, R. M. and R. M. May. 1980. Infection diseases and population cycles of


forest insects. Science 210: 658-661.
Andreson, R. M. and R. M. May. 1981. The population dynamics of microparasites and
their vertebrate hosts. Phil. Trans. of the Royal Soc. of London 210: 658-661.

Questions and Assignments to Lecture 10


10. 1. Estimate parameters of type II functional response from 2 experiments in which
individual predators were kept in cages of the same size size (1 sq.m.) with different
prey density for a period of 2 days.
Number of prey Number of Total number of prey Total number of prey
per cage replications in all cages killed by predators
5 20 100 35
50 5 250 60

10.2. Assume that population dynamics of spruce budworm depends mostly on


generalist predators (birds) which density (P) can be considered constant in time and
which have a type III functional response. The model of spruce budworm dynamics is:

where search rate (a) depends on prey density: a = bN/(N+c)

Parameter values are: r=1.5 per yr; Th = 0.0003yr; P=0.01 birds/sq.m.; b=1500
sq.m./yr; c=30 larvae/sq.m.
Find all equilibrium densities of spruce budworm population. Which equilibria are
stable and which are unstable? At what population density birds fail to control spruce
budworm population?

10.3. Tachinid fly Parasetigena silvestris is a parasitoid of gypsy moth. Both the host
and the parasite have one generation per year. P.silvestris attacks large larvae and
emerges from pre-pupae or pupae. The following results were obtained after three years
of study:

Density of host
Percentage
Year large larvae
of parasitism
per sq.m.
1988 0.45 12
1989 2.15 8
1990 4.10 ?

In 1990, parasitism was not recorded. Estimate expected percentage of parasitism in


1990 according to: 1) the model of Thompson, and 2) the model of Nicholson & Bailey.
Estimate parameters of both models. Additional information: P.silvestris has a sex
ration of 1:1 and pre-adult mortality of 30%.

10.4. Why the distribution of parasitoid eggs laid on hosts may be different from the
Poisson distribution?

10.5. In what host-parasite models (Thompson, Nicholson & Bailey, or Rogers):

• Fecundity is limited
• Search rate is limited
• Host mortality does not depend on host density
• Host mortality does not depend on parasite density?

10.6. In what type of functional response (I, II, or III):

• Predator saturation is considered


• Predator search rate increases with prey density
• The proportion of killed prey is constant
• The proportion of killed prey first increases and then declines with increasing
prey density
• The proportion of killed prey always declines with prey density?

10.7. Host density is 5 individuals per square meter; parasitoid density is 1 female per
square meter; one parasitoid female can parasitize maximum 100 hosts, and its search
rate (area of discovery) is 1 sq.m. per life. What is the proportion of parasitism predicted
by each of three models: Thompson, Nicholson & Bailey, and Rogers?
Lecture 11. Competition and Cooperation
11.1. Intra-specific competition
Intraspecific competition results in a reduction of population growth rate as population
density increases. We already studied several models that consider intraspecific
competition: logistic model and Ricker's model. In these models population growth rate
steadily declines with increasing population density.

However, in nature competition effect may be completely absent until population


density reaches some threshold at which resources become limited.

Forest insect defoliators often have a "scramble" competition. If larvae can find foliage
they still survive, but when all foliage is destroyed then mortality increases very rapidly.
This is because these insects have seasonally-synchronized life cycles. If food is
exhausted before they can pupate, then none of them pupates.

Mechanisms of competition:

1. Simultaneous resource utilization. Examples: forest defoliators, dung insects


2. Direct interaction. Examples: aggression, cannibalism, territoriality

Simultaneous resource utilization is usually associated with threshold-type relationships


between resource abundance and population growth rate; in some cases "scramble"
competition happens.

Direct interaction among organisms makes competition balanced and usually results in a
gradual decline of population growth rate with the decrease in the amount of resources.

11.2. Competition between Species


Competition among ecologically similar species is the major factor that determines the
structure of animal and plant communities. The main question is, can competing species
coexist or not, and what are the major factors that affect coexistence. This topic is a
bridge between population ecology and community ecology.

Major problems:
1. In conservation ecology: to prevent extinction of particular species; predict
potential losses in species composition after introduction of competitors; to
reduce competition effects.
2. In biocontrol: to find an exotic natural enemy which will successfully fit into the
community of existent natural enemies; to find exotic non-pest competitors that
may oust the pest species.

In the logistic model, population density converges to the carrying capacity K, as it is


shown below:

Now, we will introduce the second (competing) species. As a result, the figure becomes
two-dimensional:

In this example, species #1 becomes extinct as a result of its competition with species
#2.

Competitive exclusion principle was first formulated by Grinnell (1904) who wrote:

"Two species of approximately the same food habits are not likely to remain long evenly
balanced in numbers in the same region. One will crowd out the other; the one longest
exposed to local conditions, and hence best fitted, though ever so slightly, will survive,
to the exclusion of any less favored would-be invader."

If competing species are ecologically identical (use the same resource), then inter-
specific competition is equivalent to intra-specific competition. Each organism
competes with all organisms of both populations. As a result, population growth rate of
each population is determined by the sum of numbers of both populations:
Excel spreadsheet "lotkcomp.xls"

In this case, both isoclines are parallel and have a slope of 45° (see figures above). The
species that have a higher carrying capacity (K) always wins. Higher carrying capacity
means that the species can endure more crowding than the other species (e.g., due to
more effective search for resources). Competitive exclusion is called K-selection
because it always go in the direction of increasing K.

If competing species are sufficiently different then intra-specific competition is stronger


than inter-specific competition. Organisms of another species are not considered as
"full" competitors. As a result, the numbers of inter-specific competitors is multiplied
by a weight wi<1:

Theoretically it is possible that weights wi>1. This means that organisms of another
species are stronger competitors than organisms in the same population. I don't know
any example of this sort. But this situation is always discussed in ecological textbooks.
If wi>1 and isoclines intersect, then one species will oust the second one, but what
species will be excluded depends on initial conditions (initial numbers of both
populations):
This system has an unstable equilibrium which separates 2 areas of attraction: (1) where
the first species ousts the second one and (2) where the second species ousts the first
one.

Excel spreadsheet "compet.xls"

Thus, species coexistence is possible if intraspecific competition is stronger than


interspecific competition. This occurs if competing species have different preferences in
resource usage.

When the principle of competitive exclusion became widely known among ecologists, it
seemed to contradict with some well known facts and this contradiction was formulated
as "paradoxes". For example, "plankton paradox" focused on the variability of plankton
organisms which all seemed to use the same resources. All plankton algae use solar
energy and minerals dissolved in the water. There are not so many mineral components
as compared to a large variability in plankton algae species.

There is no final solution for this paradox. However, it became clear that coexistence of
species that use the same resource is a common phenomenon. Mathematical models
described above are correct, but they are oversimplified; thus it is difficult to apply them
to real species. More complicated and more realistic models indicate that species
coexistence is possible. For example, plankton algae have distinct seasonality in their
abundance which is ignored in the simple Lotka-Volterra model. Cyclical dynamic
regime allows species to coexist even if they cannot coexist in stable systems. Another
important factor is spatial heterogeneity which effect is substantial even in such
homogeneous systems as the ocean.

11.3. Ecological Niche


The ecological niche is not a notion of quantitative population ecology despite of
several attempts to define it quantitatively. There were numerous definitions of
ecological niche. Grinnell (1917) defined it as all the sites where organisms of a species
can live (where conditions are suitable for life). Elton (1927) described the niche as the
function performed by the species in the community of which it is a member. The first
definition emphasized the "address" of the species and the second one emphasized its
"profession" (Miller 1967).

Hutchinson (1957) defined a niche as a region (n-dimensional hypervolume) in a multi-


dimensional space of environmental factors that affect the welfare of a species. This
definition is more close to Grinnell's definition. It became popular because the range of
tolerance to ecological factors can be easily measured, whereas species "profession" is
hardly measurable. It is believed that the intensity of competition is proportional to the
degree of niche overlapping. However, this kind of statements should be accepted with
caution because: (1) measurement of niche volume is a subjective procedure, (2) some
important dimensions of the niche may be not known, (3) niches change in the life-
cycle, (4) niches change from one geographical region to another.

More information about niches can be found in Pielou, chapter 13.


11.4. Cooperation
Intra-specific cooperation results in increased reproduction and/or survival of organisms
in groups as compared to isolated organisms. This is called group effect which was first
analyzed Allee (1931) and is often called "Allee effect".

Group effect results in the increase of population growth rate with increasing population
density when density is low. There is more danger of extinction in populations with
group effect because there is a minimum population density below which population
declines until it is gone.

Cooperation between different species is relatively rare in nature and we will not study
its models. Usually it has a form of symbiosis.
Lecture 12. Dispersal and Spatial Dynamics
12.1. Random walk
Early population studies concentrated on local population dynamics. However, spatial
processes are very important in life-systems of most of the species. They may so
significantly modify system behavior that local model would be unable to predict
population changes.

Several ecological problems cannot be addressed without analysis of organism


dispersal. Examples are: spread of invading species, epidemics, etc.

Let's take the problem of pest insect control as an example. The first question is what
area to treat. If this area is too small it will be immediately colonized by immigrants.
Crop rotation is often used to prevent propagation of pests, but the distance between
fields with the same crop in two consecutive years should be separated further than
migration distance. Finally, many insect pests are sampled using traps (pheromone-
baited traps or UV-traps). To determine pest density from trap catches it is important to
know dispersal abilities of the insect.

The main problem: how many organisms disperse beyond a specific distance?

Random walk is simulated here assuming that 50% individuals stay at the same place,
25% move to the left, and 25% move to the right. After several time steps the
distribution of organisms becomes close to the normal distribution:

Excel spreadsheet "diffus.xls"

Normal (=gaussian) distribution corresponds to equation


Random walk can be defined in a 2-dimensional space. If organisms were released at
the center of coordinates (0,0), then their distribution can be described by 2-dimensional
normal distribution:

This is a 2-dimensional normal distribution:

12.2. Diffusion Models


Advantage of diffusion models is that they can be applied to any initial distribution of
organisms. The most simple diffusion model in 1-dimensional space is:

where N is population density, and D is diffusion coefficient. This equation indicates


that the rate of population change is proportional to the curvature of population density.
Examples below show that population increases where curvature is positive and
decreases where it is negative.

Skellam (1951) combined diffusion equation in a 2-dimensional space with exponential


local population growth:
Model assumptions:

• all individuals simultaneously disperse and reproduce;


• there is no variation in dispersal abilities of individuals.

Skellam's model predicts that if a population was released at a single point, then its
spatial distribution will be a 2-dimensional normal distribution:

One of the most interesting features of this model is that it predicts the asymptotic rate
of expansion of population front. The rate of population expansion, V, is defined as the
distance between sites with equal population densities in two successive years:

Skellam's model gives the following equation for the rate of population expansion:

Excel spreadsheet "skellam.xls"

Both parameters, r and D, can be estimated in independent experiments. Intrinsic rate of


population increase can be determined from the life-table. Diffusion coefficient D can
be estimated using mark-recapture experiments. For example, if marked animals are
released within a uniform grid of traps, then diffusion coefficient is estimated as:

where M(t) is mean displacement of organisms recaptured t units time after their release
(Skellam 1973).

Example:
The muskrat (Ondatra zibethica) was introduced to Europe in 1905 near Prague. Since
that time its area expanded, and the front moved with the rate ranging from 0.9 to 25.4
km/yr. Intrinsic rate of population increase was estimated as 0.2-1.1 per year, and
diffusion coefficient ranged from 51 to 230 sq.km/yr. Predicted spread rate (6.4-31.8
km/yr) corresponds well to actual rates of spread.

12.3. Stratified Dispersal


One of the major limitations of diffusion models is the assumption of continuous spread. In
nature many organisms can move or can be transferred over large distances. If spread was
continuous, then islands would never be colonized by any species. Discontinuous dispersal may
result in establishment of isolated colonies far away from the source population.

Passive transportation mechanisms are most important for discontinuous dispersal. They
include wind-borne transfer of small organisms (especially, spores of fungi, small insects,
mites); transportation of organisms on human vehicles and boats. Discontinuous long-distance
dispersal usually occurs in combination with short-distance continuous dispersal. This
combination of long- and short-distance dispersal mechanisms is known as stratified dispersal
(Hengeveld 1989).

Stratified dispersal includes:

• establishment of new colonies far from the moving population front;


• growth of individual colonies;
• colony coalescence that contributes to the advance of population front.

The area near the advancing population front of the pest


species can be subdivided into 3 zones:

• Uninfested zone where pest species is generally


absent
• Transition zone where isolated colonies become
established and grow
• Infested zone where colonies coalesced
A good example of a population with stratified
dispersal is gypsy moth. Gypsy moth egg masses
can be transported hundred miles away on human
vehicles (campers, etc.). New-hatched larvae
become air-borne and can be transferred to near-by
forest stands. Distribution of gypsy moth counts in
pheromone traps in the Appalachian Mts. (Virginia
& West Virginia) in 1995 is shown in the right
figure. Isolated populations are clearly visible. The
US Forest Service Slow-the-Spread project has a
goal to reduce the rate of gypsy moth expansion by
detecting and eradicating isolated colonies located
just beyond the advancing population front.
Metapopulation model of stratified dispersal
Sharov and Liebhold (1998, Ecol. Appl. 8: 1170-1179. [Get a PDF reprint!]) have developed a
metapopulation model of stratified dispersal. This model is based on two functions: colony
establishment rate and colony growth rate. The probability of new colony establishment, b(x),
decreases with distance from the moving population front, x

Population numbers in a colony, N(a), increases exponentially with colony age, a

n(a) = noexp(ra)

where no are initial numbers of individuals in a colony that has just established, and r is the
marginal rate of population increase.

The population front is defined as the farthest point where the average density of individuals
per unit area, N, reaches the carrying capacity, K:

N = K.

The rate of spread, v, can be determined using the traveling wave equation. We assume that the
velocity of population spread, v, is stationary. Then, the density of colonies per unit area m(a,x)
of age a at distance x from the population front is equal to colony establishment rate a time
units ago. At that time, the distance from the population front was x + av. Thus, m(a,x) = b(x +
av). The average numbers of individuals per unit area at distance x from the population front is
equal to:

where n(a) is the number of individuals in a colony of age a. The population front is defined by
the condition N(0) = K. Thus, the traveling wave equation is
This equation can be used for estimating the rate of population spread. To estimate this integral
we need to define explicitly functions b(x) and n(a). We assume a linear function of the rate of
colony establishment:

The population of each colony increases exponentially


n(a) = noexp(ra)

After substituting these functions b(x) and n(a) into the traveling wave equation, we get the
following equation

where V = v/xmax is the relative rate of population spread. This equation can be solved
numerically for V; and then the rate of spread is estinmated as v = V xmax.

This model can be used to predict how barrier zones (where isolated colonies are detected and
eradicated) reduce the rate of population spread. We will assume that the barrier zone is placed
in the transition zone at some particular distance from the population front. Because new
colonies are eradicated in the barrier zone, we can set the colony establishment rate, b(x), equal
to zero within the barrier zone as it is shown in the figure below

If this new function b(x) is used with the traveling wave equation, we get the rate of spread
with the barrier zone. The figure below shows the effect of barrier zone on the rate of
population spread. Relative width of the barrier zone is measured as its proportion from the
width of the transition zone. Relative reduction of population spread is measured as 1 minus the
ratio of population spread rate with the barrier zone to the maximum rate of population spread
(without barrier zone).
This model predicted that barrier zones used in the Slow-the-Spread project should reduce the
rate of gypsy moth spread by 54%. This prediction was close to the 59% reduction in the rate of
gypsy moth spread in Central Appalachian Mountains observed since 1990 (when the strategy
of eradicating isolated colonied has been started).

Cellular automata models of stratified dispersal


Cellular automata is a grid of cells (usually in a 2-dimensional space), in which each cell is
characterized by a particular state. Dynamics of each cell is defined by transition rules which
specify the future state of a cell as a function of its previous state and the state of neighboring
cells. Traditional cellular automata considered close neighborhood cells only. However, in
ecological applications it is convenient to consider more distant neighborhoods within specified
distance from the cell.

The figure above shows 3 basic rules for the dynamics of cellular automata
that simulates stratified dispersal:

1. Stochastic long-distance jumps


2. Continous local dispersal
3. Population growth (population numbers are multiplied by R)

Results of cellular automata simulation for several sequential time steps are
shown at the right figure. It is seen how isolated colonies become established,
grow, and then coalesce. This model was used for prediction of barrier-zone
effect on the rate of population spread and the results were similar to those
obtained with the metapopulation model..
References
Sharov, A. A., and A. M. Liebhold. 1998. Model of slowing the spread of gypsy moth
(Lepidoptera: Lymantriidae) with a barrier zone. Ecol. Appl. 8: 1170-1179. Get a reprint!
(PDF)
Sharov, A. A., and A. M. Liebhold. 1998. Bioeconomics of managing the spread of exotic
pest species with barrier zones. Ecol. Appl. 8: 833-845.
Get a reprint! (PDF)
Sharov, A. A., and A. M. Liebhold. 1998. Quantitative analysis of gypsy moth spread in the
Central Appalachians. Pp: 99-110. In: J. Braumgartner, P. Brandmayer and B.F.J. Manly [eds.],
Population and Community Ecology for Insect Management and Conservation. Balkema,
Rotterdam.
Get a reprint! (PDF)
Sharov, A. A., A. M. Liebhold and E. A. Roberts. 1998. Optimizing the use of barrier zones
to slow the spread of gypsy moth (Lepidoptera: Lymantriidae) in North America. J. Econ.
Entomol. 91: 165-174.
Get a reprint! (PDF)

12.4. Metapopulation models


Metapopulation is a set of local populations connected by migrating individuals.

Local populations usually inhabit isolated patches of resources, and the degree of
isolation may vary depending on the distance among patches:

Metapopulation models consider local populations as individuals. Dynamics of local


populations is either not considered at all, or is considered in a very abbreviated way.
Most of metapopulation models are based on colonization-extinction equilibrium.

One of the first metapopulation models was developed by MacArthur and Wilson
(1967). They considered immigration of organisms (e.g., birds) from a continent to
islands in the ocean. The proportion of islands colonized by a species, p, changes
according to the equation:

Equilibrium proportion of colonized islands can be found by solving the equation dp/dt
= 0:
Assume that extinction rate declines with increasing island diameter S:

and colonization rate declines with increasing distance D from the continent:

Now, the proportion of colonized islands becomes a function of island size and its
distance from the continent:

If there is a group of species with similar biology and similar migration capabilities,
then the proportion of colonized islands is proportional to the number of species that
live on an island. Now the model can be tested using regression:

where

Example: There are 100 bird species on the continent and the number of species on
islands is in the following table:

Island Diameter, Distance from Number of Proportion of ln(p*/(1-


No. km (S) continent, km (D) bird species species (p*) p*))
1 10 30 40 0.40 -0.405
2 3 50 5 0.05 -2.944
3 20 100 20 0.20 -1.386

Using linear regression of ln(p*/(1-p*) against two factors: S and D we get the
following model parameters: α = 0.229; β = 0.0467; and γ = -1.29. These parameters
can be used to predict the number of species on other islands using information about
island size and its distance from the continent.

A similar metapopulation model describes metapopulation dynamics without a


continent. In this case, neighboring islands become a source of colonization. The
proportion of colonized islands changes according to differential equation:
where c = colonization rate per 1 island, and e = extinction rate. Equilibrium proportion
of colonized islands can be found by solving the equation dp/dt = 0:

This model can be expanded by incorporating island size and degree of isolation in the
same way as the previous model.
Lecture 13. Population Outbreaks
Ecological Mechanisms of Outbreaks
Population outbreaks are characterized by rapid change in population density over
several orders of magnitude. Only a small number of species have outbreaks: e.g., some
insect pests, pathogens, and rodents. Population outbreaks often cause serious
ecological and economic problems. Examples of outbreak species: locusts, southern
pine beetle, spruce budworm, gypsy moth

Two kinds of outbreaks:


1. Introduction of a species to a new area
2. Growth of a native population.

The second case is most interesting because it is important to understand why a


population suddenly increases in its density. Usually an outbreak goes through the
following phases:

The building phase in insect pests often goes unnoticed because the effect of pests on
host plants is still very small. Regular monitoring helps to detect population growth
before the pest species devastates host plants over large areas.

Initially ecologists tried to explain outbreaks by direct impact of environmental factors.


However, the magnitude of change in these factors was always much smaller than the
magnitude of change in population density. Attempts to find a "releasing factor" usually
fail.

Thus, there should be an "amlipier" of a small initial disturbance.

Examples of "amplifiers":

1. Inverse density-dependence (positive feedback)

1.1. Escape from natural enemies: gypsy moth.


Mortality caused by generalist predators with a type II or III functional response
decreases with increasing prey density. Thus, the greater is the population
density, the faster it grows. Escape from natural enemies may also result from a
delay in the numerical response of natural enemies (pine sawflies, gypsy moth).
1.2. Group effect: bark beetles, sawyer beetle (Cerambicidae), locusts.
Bark beetles succeed in attacking a healthy tree only when the number of beetles
is large. Adults of the sawyer beetle, Monochamus urussovi, feed on small
branches of Siberian fir. When the density of adults is high, then they cause
considerable damage and the tree looses its resistance to developing larvae of
this species. Locusts change their behavior at high population density, and their
reproduction rate increases.
2. Density-independent processes.
2.1. Plant response to disturbance: spider mites.
Population of spider mites grow very fast at high temperature. They live on plant
leaves where local temperature is lower than the ambient temperature. During
the draught, plant transpiration is reduced, and thus, the temperature of leaves
increases causing rapid reproduction of spider mites.
2.2. Insect physiological response to disturbance: sawflies.
Pine sawflies, Diprion pini, have >50% of their population in a prolonged
diapause lasting from 1 to 5 years. Draught may cause reactivation of a large
proportion of diapausing sawflies. This effect is combined with subsequent
escape from natural enemies.
These amplifiers can be triggered only at specific state of the population system. When
an outbreak is already in progress, additional disturbances have almost no effect. Only
when an outbreak cycle is finished, then the population may again respond to another
disturbance. In some cases, even small disturbances cause an outbreak, and then the
population is permanently in an outbreak cycle.

Outbreaks collapse usually due to one of the following mechanisms:

• Destruction of resources
• Natural enemies
• Unfavorable weather

A model of an Outbreak
The eastern spruce budworm (Choristoneura fumiferana) is a forest pest insect. It
defoliates spruce stands in Canada and Maine. Outbreaks occur in intervals of 30-40
years. Last outbreaks were in 1910, 1940, and 1970. They resulted in defoliation of 10,
25, and 55 million hectares, respectively.

Clark and Holling (1979, Fortschr. Zool. 25: 29-52) developed a simple model of the
spruce budworm population dynamics. It includes (1) logistic population growth, and
(2) type III functional response of polyphagous predators (birds). Biological background
of this model is not very solid because several important factors were ignored (e.g.,
parasitism, diseases). However, the model captures dynamic features of the system and
we will use it as an example of an outbreak models.

Excel spreadsheet "budworm.xls"

Population dynamics is described by the following differential equation:


Technical details: The second term (predation) is derived from the Holling's disc
equation assuming that the search rate of predators is proportional to prey density (a =
qN). Then, alpha = qP, and beta = qTh, where P is the density of predators and Th is
handling time.

We will start with the following parameter values:


r = 1; K = 1000; alpha = 0.5; beta = 0.04

This model describes the start of an outbreak due to escape from natural enemies. The
phase plot of the model is show below. There are 2 stable equilibria: at the lower
equilibrium (N* = 2.5) population numbers are stabilized by predation, and at the higher
equilibrium (N* = 2.5) the population reaches its carrying capacity. The switch point
between these equilibria is N = 10.

If we run the model with a small environmental noise, we will get the following output:

The population density once came very close to the switch point (N = 10) but then it
returned back due to unfavorable conditions at that specific time. If we increase the
amplitude of noise, then the population eventually passes the switch point and the
outbreak starts. The lower graph is the magnification of the upper graph:
Now we will increase parameter r (intrinsic rate of increase) to the value of r = 1.1. The
phase plot changes and the distance between the lower equilibrium and the switch point
becomes smaller:

Now the outbreak starts even with a low noise (as in the first graph) and much earlier:
The lower graph is the magnification of the upper graph.

This model does not include mechanisms that may cause the collapse of outbreak
populations. Thus, the outbreak continues forever in a model population. In nature,
outbreak populations of the spruce budworm cause severe defoliation and destroy the
forest. As a result, the population collapses. Interaction of spruce budworm with host
trees will be considered in the following section.

Catastrophe theory
Catastrophe theory was very fashionable in 70-s and 80-s. Rene Thom was one of its
spiritual leaders. This theory originated from qualitative solution of differential
equations and it has nothing in common with Apocalypse or UFO.

Catastrophe means the loss of stability in a dynamic system. The major method of this
theory is sorting dynamic variables into slow and fast. Then stability features of fast
variables may change slowly due to dynamics of slow variables.

The theory of catastrophes was applied to the spruce budworm (Choristoneura


fumiferana) (Casti 1982, Ecol. Modell., 14: 293-300). We will use the model which was
considered in the previous section and modify it by adding a slow variable: the average
age of trees in the stand.

The performance of spruce budworm populations is better in mature spruce stands than
in young stands. Thus, we will assume that the intrinsic rate of increase (r) and carrying
capacity (K) both increase with the age of host trees:

where A is the average age of trees in a stand. Now the model is represented by the
equation:

The first term is the logistic model, and the second term describes mortality caused by
generalist predators which have a type III functional response. Equilibrium points can
be found by solving this equation with the left part set to zero (dN/dt = 0):
The left graph shows phase plots for various forest ages from A = 35 to A = 85, and the
right graph shows equilibrium points (where the derivative is equal to 0). Only one non-
zero equilibrium exists if N<38 or N>74. If 40<A<74, then there are 2 two stable
equilibria separated by one unstable equilibrium. Equilibrium line folds.

The age of trees continue increasing with time. Age can be considered as a "slow"
variable as compared to population density which is a "fast" variable. Dynamics of the
system can be explained using the graph:

Fast processes are vertical arrows; slow processes are thick arrows. Slow processes go
along stable lines until it ends, then there is a fast "jump" to another stable line.

Direction of slow processes. When the density of spruce budworm is low, then there is
little mortality of trees and the average age of trees increases. Thus, the slow process at
the lower branch of stable budworm density is directed to the right (increasing of stand
age). The upper branch of stable budworm density corresponds to outbreak populations.
Old trees are more susceptible to defoliation and they die first. Thus, the mean age of
defoliated stand decreases, and the slow process at the upper branch goes backwards.
Population dynamics can be described as a limit cycle
that includes 2 periods of slow change and 2 periods
of fast change. Transition to a fast process is a
catastrophe. This model is built in Excel:

Excel spreadsheet "budworm.xls" (open Sheet 2)

We can add stochastic fluctuations due to weather or other factors. As the age of trees
increases, the domain of attraction of the endemic (=low-density) equilibrium becomes
smaller. As a result, the probability of outbreak increases with increasing forest age. If
an outbreak occurs in a young forest stand, then it is possible to suppress the population
and return it back to the area of stability. But if the stand is old, then the endemic
equilibrium has a very narrow domain of attraction, and thus, the probability of an
outbreak is very high. Finally, the lower equilibrium disappears and it is no longer
possible to avoid an outbreak by suppressing budworm population.

The model suggests to reduce forest age by cutting oldest trees. This will move the
system back into the stable area.

Classification of Outbreaks
Classification of insect outbreaks was independently developed by Berryman (1987)
and Isaev and Khlebopros (1984).
Unstable High
Stable High Equilibrium
Equilibrium)
Sustained eruption Pulse eruption

Stable Low Equilibrium

Permanent eruption (no Cyclical eruption


examples)
Unstable Low
Equilibrium

Below is the classification of all types of population dynamics (not only outbreak
species)
• Gradient populations: respond directly to external factors (no density-
dependent amplification). They have high density in favorable conditions and
low density in unfavorable conditions (both in space and time). High-density
populations never spread (cause population increase in surrounding
populations).
• Eruptive populations: the effect of external factors is amplified by inverse
density-dependence (=release effect). Amplifying mechanisms were discussed in
the first section. Outbreaks of eruptive populations are able to spread (traveling
wave). See the spreading outbreak of the southern pine beetle.
• Sustained eruption: environmental fluctuations may cause the transition of the
population from the low equilibrium to the high equilibrium. Examples: bark
beetles, spruce budworm
• Pulse eruption: environmental fluctuations trigger an outbreak which collapses
immediately (e.g., due to parasites). Examples: gypsy moth, pine sawflies, etc.
• Cyclical eruption: Both equilibria are unstable and populations cycles around
them. Examples: Zeiraphera diniana, Cardiospina albitextura.

In bark beetles and some sawyer beetles Cerambicidae, two stable population equilibria
exist because of the positive feedback. Massive beetle attack on a tree overcomes its
resistance, and thus, the greater is population density the more resources (weakened
trees) are available.

Mechanisms of beetle attack may be different. Bark beetles make holes in the bark. If
there are only few beetles, then the holes become filled with resin and beetles die. If
thousands of beetles are making their holes simultaneously, then the tree has not enough
resin for self-defense.

Sawyer beetles (Monochamus urussovi) oviposit into boles of weakened trees. Adults
feed in tree twigs and can weaken a tree if population density is high. As a result, more
oviposition sites become available.

The figure shows the phase-plot of a bark beetle


population. Thick lines are stable equilibria. The high
equilibrium is often called "metastable" because the
bark beetles eventually destroy their habitat.
Synchronization of Outbreaks
Examples of synchronized outbreaks:

• Defoliation caused by the gypsy moth, Lymantria dispar L., in Massachusetts in


1971-86. From Liebhold and Elkinton (1989; Forest Sci. 35: 557-568).
• Outbreak of the mountain pine beetle, Dendroctonus ponderosae Hopk., in
lodgepole pine forests of Glacier National Park. From McGregor et al. (1983;
USDA Report 83-16).

Three main hypotheses:

• Direct weather effect (this is a good explanation for gradient species)


• Migration from outbreak area triggers an outbreak in adjacent areas (e.g., bark
beetles)
• Weather synchronizes population oscillations = Moran's effect (gypsy moth?)

Potrebbero piacerti anche