A Tutorial For Understanding Ecological Modeling Papbrs For THB
A Tutorial For Understanding Ecological Modeling Papbrs For THB
A Tutorial For Understanding Ecological Modeling Papbrs For THB
...
Understanding ~
..
Ecological Modeling
PapBrs for thB
M
ODELING PAPERS PLAY A CRUCIAL ROLE in further proach entomology from a different perspective
ing the science of entomology. In complex and often use a different set of tools to pursue our
ecological systems, modeling can be used to science. Many of my colleagues shudder at the
predict outcomes, clarify questions, facilitate com- thought of anything mathematical, and, in their
puter experiments, and manipulate key variables circles, it is rare that such tools are ever used. This
that would be impossible to do experimentally be- primer is for them.
cause of cost, logistics, or ethics. Yet, to many ento-
mologists, models remain the domain of Statistical Versus Process Models
mathematicians, statisticians, and others from more Models can be classified in several different ways.
quantitative disciplines. As a result, modeling pa- Like any taxonomist, we would like a nice, neat
pers sometimes are ignored by the biologists for dichotomous key, but modeling can be an amal-
whom these papers are written. This is unfortu- gam of many different elements. Although there is
nate because a particular model can provide in- overlap between model types, with one blending
sights that might help those working on an smoothly into the next, there are some general rules
empirical problem. Conversely, modelers can al- and some dichotomies. We will work through these.
ways benefit from those working closely with the But keep in mind that these many models might all
biology of the systems that they are trying to model be combined to produce a more complex model. A
by fairly assessing the results and assumptions the list of definitions of all the model types is given in
modelers have used to describe the systems. It is the glossary.
important that both modelers and nonmodelers The first distinction that needs to be made is
speak to one another. Modelers, however, seem to between statistical models and process or descrip-
have a language of their own: analytic models? simu- tive models, which are the kind of mathematical
lation models? individually based models? stochas- models I describe here. Statistical models are used
tic models? Readers unfamiliar with modeling might to give a probabilistic interpretation of the data.
wonder how these terms relate to the modeling Familiar statistical models include techniques such
paper they are trying to understand. as simple linear regression that fits a line through a
Becoming familiar and comfortable with any series of data points. Statistical models do not try
subject is often a matter of getting to know the to describe underlying processes through an un-
terminology and basic tools associated with the derstanding of biological mechanisms but rather
discipline. If you do modeling on a regular basis, attempt to find a set of parameters that can be used
there is little need to read on. This article is not for to predict relationships that are described by the
you. Some of the more arrogant modelers might data. The science of statistics is one of the most
even feel a wave of condescension, believing that valuable tools we have for understanding data. It
such things ought to already be a part of the tool can help us find hidden patterns, predict future
kit of every entomologist. Nevertheless, we all ap- trends, find differences between one set of biologi-
Analytic Model: A model for which a spe- programming in which objects can be defined
cific mathematical form for the model can be with specificattributes and functions that modify
written in an equation or set of equations. The those attributes. This is a very useful program-
ahility to write an equation for a model allows ming environment for individual-based mod-
the use of many techniques developed to exam- els. For example, a popular object-oriented
ine the prediction and behavior of these mod- programming language is C++ (pronounced "C
els. (Compare with Simulation Model) plus plus").
Cellular Automata Model: A spatial model Ordinary Differential Equations: Used in
where both time and space are discrete. continuous time analytic mathematical models.
Computer Model: See Simulation Model. Partial Differential Equations: Used in ana-
Continuous Versus Discrete: Continuity re- lytic mathematical models in which both time
r=2.3 r=2.9
0
0
0 8
~ 0
''""
zQ)8 zQ) §
.~ §
(/)
N N '" N
C
'00 0
g§ c 0
0 ~
ct3'" '';::
::J ct3
a. S 8
00 a.8
a.8 o -
<0
a.
8
0
"'
8
0
, ,
" 0 10 20 30 40 so 0 10 20 30 40 50
t t
Fig. 1. The behavior of a model using the deterministic logistic equation Nt+1 = Nt + rNt K K Nt
to predict the size of an insect population through time. The behavior of the model differs depending on the
value of the growth parameter r (see text for details). When r = 2.3, the population has predictable oscillations.
When r = 2.9, the behavior of the model is very erratic and enters into the realm known to mathematicians as
"chaos."
Table 1. A sample of modeling papers published in Environmental Entomology between 1989 and 1998 classified by
whether they are analytic or simulation, and deterministic or stochastic.
Nonspatial models are in black; spatial models in blue (see Spatial Models section).
t
begin time t+ 1
5: proportion of field planted in susceptible cotton
T: proportion planted in transgenic cotton
transgenic cotton (Peck et al. 1999), the model was
constructed in just over 1 month, but testing and
refining took more than 1 year, thousands of runs,
~ (c) and a substantial investment of time to convince us
end time t (f)
that the model was doing what we hoped it would
adults migrate genotype--plant- do. Most of the hard thinking started after the
t (e) (d)
struct lull you into thinking that they are a quick-
and-dirty method of exploring complex questions.
When done correctly, they can be among the most
time-consuming methods of modeling. There has
genotype-- density dependent been a lively debate on simulation versus analytic
pl(lnt-tj'pC spCClflc modeling, and the role of population dynamics
survival modeling (Berryman 1991, 1997; Onstad 1991;
adult emergence Logan 1994; Hess 1996a). These articles in the
American Entomologist describe in detail some of
the issues and problems with these two types of
models.
Fig. 2. This spatial model (Peck et al. 1999), which explores the regional development of Often one will see several of the model types
resistance in Heliothis virescens (F.) to a Bacillus thuringiensis Berliner (Bt) d-endotoxin in described in this article combined into a larger more
transgenic colton, has several compartments. In each day and within every field, the complex model. For example, a compartment
activities of the insects are represented in the model by using compartments to handle
model will consist of several analytic, and/or simu-
the separate aspects of their life history. Starting in the upper left-hand corner (a), the
insects randomly mate and lay eggs, which are then divided among the plantings of lation models combined to produce a single model.
susceptible and transgenic colton in the field. These eggs hatch and the larvae move Typically, these are explored through simulation.
between the two crop types (b); the size of the arrows indicate that most movement is The analytic models are run in "compartments,"
from the transgenic crop to the susceptible plants because larvae are more likely to drop and the information from these runs is passed on
once they have tasted the Bt-containing leaves. These events are followed by (c)
to other compartments, which in turn model dif-
selection through mortality (based on their allele type and what kind of plant they are on),
(d) density dependent survival (if the population is large), (e) plant-type (transgenic or ferent aspects of the system. The flow of such a
susceptible) specific adult emergence, and (f) migration to neighboring fields, after which model is illustrated in Fig. 2. Each of the compart-
a new day begins and the same processes are repeated. Each of these compartments is ments is controlled by a mathematical or simula-
a separate model that captures a particular life history stage of the insect and then
tion model that passes information to the next
passes the information it has generated to the next compartment. By sharing information
among the compartments, the entire life history of the insect can be modeled. compartment after modeling the processes repre-
sented inside the individual compartment.
based modeling, but, often higher level languages The Representation of Space in Models
like C, C++, FORTRAN, PASCAL, or BASIC are Increasingly, it has been recognized that the spa-
used. In particular, object-oriented programming, tial dimension in which organisms move plays a
which is available in some newer computer lan- major role in ecological processes (Levin 1989,
guages such as C++, provides an excellent envi- Karieva 1990). Living things move. Individuals,
ronment for rule-based programming. populations, and even entire ecosystems change
Object-oriented programming languages allow one their location at spatial scales ranging from milli-
to define objects with certain attributes and func- meters to thousands of kilometers. This annoying
tions that are associated with these objects. For fact has plagued ecological studies since the time
example, I developed a program in C++ that de- that insects first descended upon our ancestors'
fined insects as one class of objects. Associated plantings of domesticated crops. If living things
with an insect object was its sex, genetic makeup, would just stay put, they would be so much easier
location on a plant, and its fecundity. Rules asso- to study, understand, and quantify. But they do
ciated with the in,ect-object included how the in- not-and we just have to deal with it. To under-
sect should respond when it met another insect, stand what effect space has on the dynamics of a
or move when it was living on a dying plant. The population in the field can be challenging. Models,
dp
•
•• ••
~