[Date Prev][Date Next][Thread Prev][Thread Next]   [Date Index] [Thread Index] [Author Index] 

Re: Nature magazine article- On niches and neutralities

Judith et al.,
I read through this pretty carefully, as it relates directly to some
current discussions. I find no quibbles with this analysis. I think I
can be more specific, however, and this is what I think the scientific
professions can best add - a translation into the current terms of
reference floating around in the particular field. This is extremely
important, and at the heart of Howard's objections, I believe. I am in
fact very sympathetic to Howard's arguments, although I think they are
also a bit too strict for a new budding research area that needs freedom
to be a little general and even metaphorical at times.

But first a direct response to the following:

My impression is that the Niche group perceive something about
complexity, but their mindset is still reductionist/mechanistic.  They are
confusing intricacy with complexity. They instinctively have some feel for
what's really going on, which I tend to doubt in the Neutrality group, but
they aren't aware of what complexity means, so how can they improve their
models?


This is pretty right on as a philosophical matter, although it is a very
general description of the case and I wouldn't imply that it is such a
black and white matter of mechanism vs complexity - they do have to be
related as Howard has been arguing (maybe I was too shy on supporting
this aspect of Howard's points). Also, I would add that ecologists are
already thinking functionally out of necessity and have been for some
time. But there hasn't been a fundamental theory of functions that they
could refer to and say "we're doing that" as the mechanists can by
pointing to physics. You will find quite a lot of Rosennean thinking in
ecology, I've concluded, but it tends to go unjustified on theoretical
grounds, or there are very confused attempts to justify it
mechanistically. Hence it is why I think landscape and ecosystems
ecology are good domains to adavance Rosennean theory. For a scientific
critique of the situation, much more specificity and reference to
current thinking is needed; and perhaps I am now experiencing what
Howard was in his domain, as the discussion turns more toward things I
am familiar with.

For those interested in such details, read on.

The niche is where, in environmental space, an organism does well. This
is hardly theory any more, it is as much a fact (from an environmental
view) as the existence of the Earth. Organisms live within environmental
limits - we know that. As I said earlier, neutral theorists do not
suggest that mangroves will invade the Yukon delta (except perhaps in
evolutionary time) because that would be entirely outside their
environmental adaptations, which is the same as saying outside their
generalized niche. So, the whole issue is not really about niche theory
at all, which is axiomatic. It is about niche realization and
segregation. The supposed challenge to niche theory is mostly
sensationalism.

Niche modelers are mostly  applying statistical models that assume
genetic difference in relation to enviornmental difference along
"resource axes." The contrasting approach is to "grow" an organism from
dynamic process equations. The former is good for generalizing
distribution constraints on a landscape; the later is good for
calibrating growth at a location. The two are needed together to get
predictions over a landscape. Neither one considers complexity or even
non-linear dynamics very well. Some attempts at that are being made
using agent-based models to simulate dispersion through a suitability
(niche) landscape, and adding complex factor by allowing the organism to
modify its own suitability factors as it goes. This puts in the
"impredicative loops" that RR spoke of -- like the 3-body problem. While
such impredicativities cannot be represented in explicit dynamic (time
differential) equations (functions of time), they can be approximated by
iterative loops. The result becomes unpredictable by any other means
than running the iteration - very much like fractal images generated
from the Mandelbrot set. There is no deep mystery in this - they are
simple calculations just done iteratively where outputs modify inputs,
thus approximating the simultaneouls causalities that confound
prediction. I would classify this as computational compexity - but the
key is that it becomes a better approximation of real life as one
reduces the interation interval toward zero and include more and more
inter-related factors. One cannot reduce that interval to zero nor can
one include all the factors, so it is still a generalization, but its
organization is much more in line with functional complexity.

When two or more similar species (tropical forest trees) have
essentially the same environmental limits, as is the case in this study
area that is small and homogenious with respect to presumed niche
factors, their distributions can then be dominated by other dynamics
(how they utilize a niche in real time), which can include biotic
interactions (others who are doing the same thing), disturbance, and
more. Niche optimization theory, taken as an extreme, was that organisms
will segregate their requirements over evolutionary time (like Darwin's
finches) and essentially adapt to micro-niches. This was generally shown
to be true, but only for the case where micro-niches are stable long
enough for species to adapt to them. What if they are changing faster
than evolutionary adaptation? Another theory to cover that case is that
they will segregate resources in ecological time and simply develop
patterns of alternating resource use. These patterns may be random or
may settle into some detectable pattern. This was well studied in terms
of animals in the Serengetti, for example, timing their visits to water
holes to avoid unpleasant contact between those who don't noramlly
socialize. The idea that trees may also randomize their locations within
a common suitability domain should not at all be surprising or
unexpected to realistic niche theorists. That's why I said it was more
advertising than actual science news.

For those who ardently defended deterministic optimization and
equalibrium theories as a "nothing butism" perhaps it is a
disappointment, but that school is already well on the decline. It is
well known in niche theory that adaptation is not infinitely precise and
that there will always be a region in environmental space where the
environmental difference will be either uninmportant genetically or
completely swamped by other factors, such as biotic interactions
(trading locations, synergisms, competition, etc.) utilization dynamics
(how a species actually occupies and utilizes the conditions it is
adapted for), the history of disturbance in the area (hurricanes opening
up patches for colonization, etc.), human influences, and just
randomness itself. Furthermore complex life has a tendency to invent
"solutions" to environmental problems that circumvent previous
limitations (the niche boundaries). This is a perfect case where the
mechanistic view should be taken only as a past constraint on an
otherwise functionally driven complex entity that can invent exceptions
- hence as Howard said, the need to develop both perspectives as
complements of each other.

The fact that these very non-linear and complex interactions are best
generalized as being random at this point only says we don't have models
that can predict anything more than a general pattern of common resource
sharing - sort of a chaotic hodgepodge. Imagine, as an analogy, a cloud
mixing in a chaotic pattern of winds - it is in a region of
environmental space where the cloud can form and exist, but the
constraints are so mixed up that you can't say much about its internal
movements. Cloud structure (shapes) segregate when the environment does.
So you can have cumulonimbus vs lenticular clouds but they would not
generally be found in the same place. Where conditions are suitable for
both, you would get a rather unpredictable result. Now a living
functional entity could establish patterns that are functionally driven,
even when environmental constraints are chaotic. That would be the place
to look for Rosennean complexity at work. Neither neutral nor
environmental/niche models test this, and it is hard to do because there
are, in niche theory, an unspecified number of dimensions that are
important. Any discernable pattern is assumed to be the result of one of
these factors, so it is explained as a missing axes in the niche model.
I am working on a method where we could include functional specs as a
niche axes in addition to environmental constraints. That will extend
niche modeling into the functional domain and allow testing of
functional determination hypotheses.

Otherwise, none of the debate about niche models says very much about
complexity. In fact, the dynamics which neither niche models nor
neutrality models can get at is where most of the complexity resides.
The neutrality model's equal failure in this area (which is all that is
demonstrated when "randomness" is shown to be as good as other
approaches) tells us not to apply niche theory where there isn't a true
functional difference (duh), and that neither approach tells us what
happens then, except that things seem to mix unpredictably. That itself
is misleading, because the random mixing of species is not itself a
random process except when viewed in the aggregate. The colonization and
growth of trees is still constrained by suitable conditions on a
micro-scale, but what we're seeing is the random occurrence of those
conditions. In other words, it may be shifting habitat availabilities
that are involved, in which case the niche theory would indeed hold some
promise over neutrality models, which have no potential for improvement
and were invented to test if other hypotheses are confirmed or not (like
comparing a road accident statistic to random probability - it isn't to
say that random explanations are better, it is to test if the other one
means anything). For example, in the lower columbia river we are trying
to model the availability of suitable river factors (salinity,
temperature, flow, depth) for outmigrating salmon. The location of
suitable conditions change dynamically as the river changes. Maybe there
are some persistent locations, or perhaps it all mixes randomly, but
either way the salmon still track the location of suitable conditions
(just as we know we would, as a general matter). Hence it is the niche
that is changing randomly, not the response to the niche (which can have
some plasticity too, particulary if we consider R-complex functions). So
it is a matter of scale. Neutrality models don't test this. Niche models
are rarely tight enough to test it either, as they presume very general
niche dimensions - i.e., a general averaged niche within which an awful
lot of dynamics and true complexity may be going on. Furthermore, if one
is looking for micro-climate variations and modeling the niche space of
a whole species, this is highly iinappropraite as well. At some scale of
micro-response we have to look at individuals, not species. The bottom
line is that this "news" suggests very little except that the dynamics
are ultimately going to be intractable below a certain level of
environmental constraint - sort of an uncertainty principle in niche
theory (which was already there). Its like quantum mechanics
approximation of quantum phenomenon - just a statistical view of the
mechanical uncertainties and regularities.
A functional view may provide better predictors than separate physical
factors and may be able to say more than a random model. For example,
you may perhaps segregate various strategies in a forest. The old R and
K selection theory (generalists and specialists) was a functional
specification, for example, and it is useful where detailed dynamics
would be too difficult to track. Strategies related to patterns of
disturbance have a lot to do with how distributions unfold in real time.
That's introducing functional modeling and could be seen from a
Rosennean perspective. The other place it applies, of course, is in the
evolution and execution of life strategies. Solving functional problems
could be a much more efficient process/theory than adapting to separate
environmental selective factors.

Enough - I'm probably well beyond the interest of all but perhaps Dan at
this point.

Comments welcome, Dan!!

JJK

Judith Rosen wrote:

Here's my analysis of that "Neutrality versus the Niche" article from Nature
magazine I posted the link for:

According to this article, both sets of ideas are only accurate within
certain narrow parameters. Outside of those parameters they produce such
flawed predictions that it is clear that much more work is needed. But the
astonishing similarity in results between the simulations of the Neutralists
and the Niche-based scientists is hard for all of them to understand. They
have such opposing views of the theoretical basis for modelling ecosystems
and yet both groups came up with models that create simulations that "look
very much like" the real thing. And-- even stranger: The results look like
each other, too. What a head scratcher!

My first reaction as I started reading this article was to mentally scoff at
the whole notion that models based on mechanistic simplicity could truly
mimic complex systems. "Neutralists" look like just more of the same old
reductionist thinkers going by a different name... So, my first impression
of Hubbell and Bell's theoretical work was based on skepticism. However, I
read through the article several times and there was something in there that
tripped a memory. I decided to look a couple things up-- I had many
discussions with my father about how biology can be "counter-intuitive" in
the lessons it teaches us.

I found it... The example I remembered. In "Life, Itself," he wrote:

[Robert Rosen]: "Experience with this [morphogenetic ideas of cell-sorting,
used to explain protein folding] approach has been most interesting. As with
the physico-chemical approach, it proceeds by minimizing something: an
objective function. But it is not the free energy of physical chemistry.
Instead of the thousands of variables and parameters inherent in the latter
approach, it contains very few; less than ten. But the interesting point is
that among these ten or so control parameters, which manipulate the hundreds
of spatial degrees of freedom of the folding polypeptide chain, some must
incorporate "global" information, such as distance from a centroid. In turn,
this information comes from a "generic" folded protein. That is, we must use
the properties of such a "generic" folded protein as a model of protein to
be folded. This in turn may be regarded as creating an impredicativity, the
hallmark of complex systems, and precisely the sort of thing which syntax
alone cannot handle."

"On the other hand, approaching folding from this direction reveals it to be
a synergetic process; one in which very few controls can manipulate a much
larger number of configurational degrees of freedom. Such synergies are
everywhere in biology, as they also are in any inordinately constrained
mechanism. My suggestion is, of course, that in biology they are indicators
of complexity rather than of mechanisms under constraints."

Thus, my father's work gives a completely different explanation for WHY the
Neutralists' theories/models/simulations "seem" to work at all, and also why
they only work under certain artificial constraints (i.e.; for only one
level of the "food web" in an ecosystem of a size between 0.2 and 50 square
km.). It is because the theories their models are based on only address one
aspect of biological phenomena. But I think that one aspect is one which
they actually may have gotten PARTLY RIGHT. In a different way, the same is
true of the Niche-driven theories of ecosystems (which similarly only work
some of the time, also under artificial constraints and specialized
conditions). My impression is that the Niche group perceive something about
complexity, but their mindset is still reductionist/mechanistic.  They are
confusing intricacy with complexity. They instinctively have some feel for
what's really going on, which I tend to doubt in the Neutrality group, but
they aren't aware of what complexity means, so how can they improve their
models?

Both schools of thought are partly right, and both are mostly wrong. What
they are calling "complexity", of course, my father  labels
"complicatedness". Rosennean Complexity illuminates why it is that when you
model complex systems from a mind-set of simplicity (characterized by
traditional reductionist/mechanistic approaches), you are bound to end up
with anomalies, paradoxes, and side-effects. "Streamlining" (a more accurate
word than "simplifying" in this context)the parameters of a model is not
automatically bad-- it just needs to reflect accurately the direction that
living systems do it.

Ultimately, I believe that if these two groups spent some time analyzing
where AND WHY they are each wrong (and compared notes), they would get a lot
farther than they will by arguing over where they are both right and each
proceeding alone. This article illustrates (in my opinion) a common
scientific truth: that experimental results, both positive and negative, can
be instructive, if studied with an open mind. Looking at unusual phenomena
without preconceptions to see where the problem itself leads you was some of
my Dad's best advice.

Judith



--
© 2004 John J. Kineman
all rights reserved