Re: [asa] Level of certainty in science

Don Winterstein wrote:

> On the face of it one of the more disturbing things about trust in climate
> scientists' predictions is their record or lack thereof on forecasts. We
> know how to do hindcasts, and after fiddling endlessly with them they can look
> impressive. But forecasts--if accurate--are where the payoff will be. How
> good are climate scientists at forecasting anything? To my knowledge they have
> no record.
>

I agree that there are limitations on any model used. In general,
when you want to model things like stock markets (physicists
actually can make money for banks!), climate, galaxy formation,
protein folding, and so forth, you _must_ make approximations.

I have not done climate models, however, I have worked with
modeling protein folding, molecular reactions, RNA folding
which are also very complex systems. I have some acquaintance
with modeling fluids and the emission of particles from a
volcano.

So lets consider something quite off the wall here.
For example, quantum electrodynamics certainly plays some role
in the Van Allen belt and therefore, perhaps it could have some
influence on the weather. Extrapolating, it could have influence
on climate if, for example, the intensity of cosmic rays from
the sun were to increase for a very extended period of time (many
years). However, if you insist on including precise calculations
at the level of quantum electrodynamics in your code, you will be
waiting a very long time for _next weeks_ weather report (maybe
1000 years with a state of the art computer).

I guess that would be one way to guarantee perpetual employment.

So the first thing you do, is you look for ways to simplify
the model, either by ignoring things you think are not all
that significant that they will adversely affect the calculation,
or you build empirical models to compensate for things you think
you cannot ignore but also are not at liberty to compute exactly.

Because climate involves the long time frame, you would aim at
large scale effects so you can compress the time window. Something
like quantum electrodynamics, if it plays any significant role at all,
could probably be approximated with a bulk empirical equation,
because whatever influence it has, it is not the individual
electrons that you are concerned with, but the bulk behavior
of the plasma.

So what you do is you construct your best model with all its
simplifying assumptions and whatever empirical equations you
need to include. You apply some kind of heuristics to make the
problem tractable, and you test it. The best indication you
have that the model is correct is that it can reproduce data
that is known to be true -- and only then to the extent that
it is known to be true. Hence, climate models do have some
limitations here, because we have only one limited example to
test the model on.

When you consider the earth with its land masses and oceans
it is a very difficult system to model well. I would _not_ expect
it to predict very well in the quantitative sense, but I would
be included to trust some of its qualitative predictions. These
are based on some basic physical principles that don't depend
strongly on the details of the model. If you douse the atmosphere
with CO2, then eventually the physical laws overwhelm the system.
As you work closer to getting a real system, you may be able to
get more accurate prediction of when and how much, but of course
there will be some undeniable uncertainty.

In as much as time and resource are availible, you should use
the best that you have, and you have to develop models that
make good sense, but I have little doubt that the people
seriously working on this problem are doing exactly that.

 
>
> If so, by taking the forecast capabilities of their models seriously we'd be
> sticking our necks out for people who have essentially no record of either
> success or failure. That would make me nervous if I were betting a lot on
> their being correct.
>
>

In weather forecasting, you have more history to test with, and
you have a lot of supporting data from satellites and past
experience from predictions (good and bad) to weight the data.
That is certainly a lot of help, and it does seem that the weather
reports are much better than they were 20 years ago. Climate
models would have far less information because we still don't
have so much data to work with to fit the models.

I grant that there are limitations, but I don't take seriously
people who make a big deal out of the chaos and the vast complexity
of the system as an excuse to ignore the predictions. Take
them as limited, fine. Take them as qualitative, fine again.
Take them as a rough estimate only, fine, fine. But shouting
"GIGO" is not contributing anything useful to science or to
improving the model, unless you really have a substantially
better model to propose.

by Grace we proceed,
Wayne

To unsubscribe, send a message to majordomo@calvin.edu with
"unsubscribe asa" (no quotes) as the body of the message.
Received on Thu Feb 8 09:14:50 2007