Re: Ecological modeling issues

[Marcus G. Daniels]

>>>>> "GR" == glen e p ropella <address@hidden> writes:

GR> The ad-hoc approach is the only one that has worked in the past;
GR> but, it's clearly not up to the task (as you and Dr. Grimm point
GR> out) because people confuse model encodings with models.  The
GR> bottom-up approach (with good doses of the other two, of course)
GR> is our last best hope.

One group with some concrete ideas about how to prevent confusion
between model encodings and models is Ferdinando Villa's IMA group at
the Institute for Ecological Economics at the University of Maryland.

   http://iee.umces.edu/~villa/IMA

The idea is to articulate domain-specific knowledge and model
component interfaces using a taxonomy of types.  The types are to be
represented using a popular markup language called XML, and (initially)
focus on defining what it means for spaces and timescales to
be compatible.  Elaborating this metadata are more paradigmatic
notions for things like `process' and `agent-based' simulation
components.

By imposing a type system on all the parameters in all model
components, three important things are accomplished:

  1) It becomes possible to reason about about the universe of
     parameter values, and to implement automatic support for this
     (e.g. using logic programming languages).

  2) It becomes much easier to implement interactive model
     builder tools, since it can be evident to the tool what are
     correct and incorrect user actions.

  3) It's a common language layer independent of particular modeling tools.
     Besides the obvious benefits for clarity in scientific description, 
     another intent is that tool developers will implement IMA wrappers so
     that tools can interoperate.  (Once things get off the ground, 
     I certainly expect to implement an IMA wrapper for the Swarm 
     activity library interfaces.)

But back to the discussion --

I'm uncomfortable with Dr. Grimm's claim that "bottom-up approaches
will never lead to theories at the system level".  Yes, traditional
top-down modeling has the potential benefit that it can be a faster
search strategy, i.e. humans can apply logic, extrapolation, and
generalization to intuition and definable abstractions, making it
feasible to safely jump around a (albeit confined) space, and to
suggest how the space might be incrementally modified.

But, generally, I think that using what is known to find something new
will only be a robust generative modeling strategy to the extent that
it helps the modeler not rehash well-mapped conceptual territory.

Yep, I'm just going to go right ahead and make the wild claim that
collecting data is a necessary part of science.  ;-)

If it is possible to describe predictive ecological theories in an
academic journal, it ought to be possible to describe them in a
declarative programming language.  (And, damn it, we ought to try!)

With predicates on bottom-up observations (or structural models of
the observations), it should be possible for pure bottom-up modeling
to identify robust relations across experiments and define theoretical
vocabularies.


                  ==================================
   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.
                  ==================================