Ecological modeling issues

[M. Lang / S. Railsback]

In my presentation to SwarmFest I briefly cited a new paper by Dr.
Volker Grimm; it was one of several interesting papers in the February
issue of Ecological Modelling (vol. 115, pages 110- ), which was
dedicated to individual-based modeling in ecology. Dr. Grimm's paper
points out problems with how IBMs have been built and used, many of them
software-related (I am trying to convert him into a Swarmist). 

He also makes the interesting point that ecologists miss a very
important opportunity when they fail to use IBMs for comparison to
traditional ecological models and theory. (I suspect the same point
applies to economists and other users of agent-based models that are
designed to depict real systems.) The Intro to this paper is posted on
Dr. Grimm's site: http://www.oesa.ufz.de/volker/volker.htm

A second paper in this issue (Lorek, H. and M. Sonnenschein. “Modelling
and simulation software to support individual-based ecological
modelling.” Ecological Modelling 115: 199-216.) discusses 3 kinds of
software: (1) library or framework systems (like Swarm and Dr. Lorek's
ECOSIM package, which is designed specifically for ecological IBMs); (2)
tools for specific kinds of models, which are easy to use but can only
be modified to a limited extent; and (3) simulators for one specific
application. Dr. Lorek was working on a package "WESP" as a type (2)
tool and as a way to make type (3) simulators. Unfortunately, Dr. Grimm
tells me that H. Lorek has left their research institute so the fate of
ECOSIM and WESP are uncertain.

For anyone interested in simulation of real and complex systems and
related software issues, here are excerpts of a note from Dr. Grimm,
after I sent him some comments on his paper (">" denotes my words).
Please note his comments on ecology at SFI, documenting models,
comparing IBMs to classical theory, frameworks for "bottom-up" modeling.

References or interests in any of these issues are welcome! 
If anyone out there is working on protocols for documenting models,
please contact Dr. Grimm. 

Steve Railsback
---------------------------------------
From: 
        Volker Grimm <address@hidden>
    To: 
        address@hidden

Your work is very intersting. I agree that to a large degree it is a
software problem which "keeps the models from being observabel and ...
testable and believable". I had long discussions about this with Helmut
Lorek (author of ECOSIM and WESP-Tool). Helmut did a real great work,
but
still for people like me, who are - as we say in German - "pedestrians"
among the programmers (e.g., I still use Borland Pascal 7.0...), ECOSIM
is
not a tool that I could use. I like the conception of the WESP-Tool (see
paper of Lorek & Sonnenschein in the Special Issue), but it turned out
that
there are very hard technical problems. Unfortunately, Helmut left
science
recently so that I am afraid the WESP project will never be finished.

It's intersting that you use the "Swarm" software - from time to time I
have a look at the Sante Fe homepage and I was all the time surprised
that
nobody in ecology seemed to know or to use "Swarm" (and nobody in Sante
Fe
seemed to be interested too much in ecology and ecological
individual-based
modelling). It is good to know that this gap no longer exists. And, I
would
be very happy to see much more ecology in the Sante Fe Institute.
(ME TOO! SR)

You mention "formulation problems with previous fish IBMs". Do you mean
problem with model structure per se, or problems with the clear
communication of model structure in papers? I am thinking about the
problem
of how to communicate IBMs in a scientifically sound but still practical
way. I was thinking about developing a draft of a standard protocol for
model presentation, presenting this draft in the Forum section of
Ecological Modelling and asking modellers to respond to this proposal
and
then to publish a kind of an "educated consensus". Maybe a good idea,
but
at the moment I am too busy with other projects. What do you think about
ideas like this? Not at all an easy task, but important. As Lorek and
Sonnenschein say: if simulation modellers continue the practice of "the
program is the model and the model is the program", we sooner or later
will
loose scientific credit (?; sorry, my English must seem rather German to
you).

Thank you for your comments on my paper:

>conditions. I think many IBMs are very complex empirical models while
>their authors believe them to be mechanistic models.

Yes, you are right. I did not think about this but I should do. On the
other hand, qualifiers like "empirical" and "mechanistic" are not
absolute
but depend on the level of abstraction one takes. Viewed from the level
of
the logistic equation, most IBMs are indeed "mechanistic", but to a
behavioural ecologist the real "mechanisms" are something completely
different.

>Instead, we have been trying to base important decisions (especially,
>how animals move) on a simple objective like maximizing their
>probability of surviving and reproducing. This approach is conceptually
>appealing because decisions are made with a simple, well-justified, and
>general basis. In one way this simplifies the model. However, we also
>find that in order to let the animals evaluate their probability of
>survival and reproduction, we have to give them a lot of information
>about their environment (including what other animals they compete
>with). So we find ourselves simplifying the fish and having a complex
>model of the environment.

Sounds very good and interesting. I am looking forward to your paper in
Ecological Modelling!

>2. Modellers must be experimenters. Your description of models as a
>black box needing exploration is very apt in many cases. However, there
>is another approach to dealing with this problem besides experimentation
>with the model: making the black box more transparent. The "Swarm"
>software (and, apparently, the ECOSIM software) allows modelers to see
>much of what is going on inside the model. Our fish modeling system has
>an animation window that shows all the fish and their habitat, and lets
>us look at the state variables of both the animals and habitat cells
>whenever we want. These tools are very useful for understanding what
>individuals are doing and why. 

Yes, I totally agree. In our department, we call this "graphical
debugging", that is: you have to SEE what happens, on the very basic
individual level!

>3. Pattern-oriented modelling. I would appreciate very much a copy of
>your 1996 paper on this approach; I am unfamiliar with it, but it may be
>similar to how we propose to begin testing our models.

I will send you two reprints. "Pattern-oriented modelling" is not a
protocol but a general strategy or attitude of ecological modelling. I
proposed the term "pattern-oriented" to show that modelling always has
to
oriented towards something, and in most cases patterns to be observed in
nature should be used as orientors.

>4. IBMs referring to classical theoretical ecology. I agree that there
>is much to be learned by comparing IBM predictions to more traditional
>aggregated ecological models. However, I am still interested in what we
>can learn from a strictly bottom-up approach, since ecological theory
>has not been especially successful in predicting real events. 

Again, I agree. The point I wanted to make is that even the strict
bottom-up approach needs SOME questions at the systems level. I proposed
to
use questions from classical theory, but I hope to see these questions
replaced by more meaningful and more interesting questions in the
future.

>We are starting to think about trying to formalize the bottom-up
>approach by looking at the emerging field of complex adaptive systems
>(CAS) theory. CAS does not yet have much to offer in the way of
>established theory, but it does provide some questions to be asked. 

Great! I love CAS without knowing more about it than you can read in
popular books like those from Waldorp or Gell-Mann. I am, how shall I
say,
"extremely" interested in this approach. I will look at your
presentation
you gave to Swarm users and hope to contact you again regarding this
issue. 


-- 
address@hidden
Lang, Railsback & Assoc.
250 California Ave., Arcata CA 95521
707-822-0453; Fax 822-1868


                  ==================================
   Swarm-Modelling is for discussion of Simulation and Modelling techniques
   esp. using Swarm.  For list administration needs (esp. [un]subscribing),
   please send a message to <address@hidden> with "help" in the
   body of the message.
                  ==================================