Observing Thinking

Observing Thinking
Observing Thinking

Sunday, June 12, 2011

June 12, 2011 Computer Models



Computer Models


In this column I want to discuss computer models and especially those that model Climate Change but  first let’s understand just what we mean by a model and then we can define a computer model. By a model we mean any representation of a real thing and not the thing itself. We can have physical  models like model ships or airplanes or we can have abstract models of phenomena like population growth, how the brain works, and weather systems. This raises a new question: what is an “abstract” model?

If I were to tell you that the next paragraph is going to be very “abstract” --- what is your first gut reaction?  Anxiety? Pain?  Despair?

Well, those are perfectly normal reactions but only because the term “abstract” has got a bum rap. It has the connotation of something difficult and complex, requiring a lot of thought but in fact, it is just the opposite. An abstraction is a simplification. For example, the graphic shows an abstraction of an evergreen tree. Notice that all of its features, except for its shape and color have been abstracted (extracted) away so that we can concentrate on just those two properties of  all evergreens. In reality, evergreen trees are much more complicated than our abstraction --- to name just a few properties we have abstracted away we could list its smell, height and girth, not to mention details like twigs, bark and needles.



So, of what use are abstractions? Since they simplify an otherwise complex system, they are easier to understand and easier to make  predictions about the behavior of the simplified system. The crucial part of building an abstract model is the decision about which properties of the system to include and which to ignore..For example, in a weather forecast model we know that the temperature, pressure and wind velocity at every point in the forecast volume is essential --- and the more points we have for this data, the better will be our forecast.  But how about the phase of the moon? Should we include that in the model? Probably not, but we would certainly include it in in a model of tidal flow. So it is very important to identify which properties to include and which to exclude in an abstract model. Finally, an abstract model usually uses an abstract language like mathematics to represent the system being modeled.

Now that we know what an abstract model is (I hope) we may define a computer model as a particular kind of abstract model that uses an algorithm to guide a computer toward the solution of a problem.  If the problem we are solving is weather prediction and the algorithm ( a well-defined, step-by-step process --- like a recipe for apple pie) can be translated into a language that the computer can understand and execute then this is called simulation. But rather than an apple pie, we get weather prediction.

This is why a computer Climate model is so difficult to construct. First we have to identify the germane variables and leave out the irrelevant ones. Even now, scientists are not completely sure of the effects of cloud cover on the weather, let alone the phase of the moon. Then we have to develop the mathematical relationships between the salient variables. Then we need to convert the math to a computer language and then we can run or execute the model on an actual computer. Finally we need to test the model: is it valid and how accurate are its predictions?

It’s fairly straighforward to test the predictive power of small systems like automobiles. We can compare the results of actual crash tests with the damage predictions of the model and evaluate on that basis. Once we are convinced the model works, at some point we can eliminate the actual crash tests as running the model is faster and cheaper and allows us to try design changes more quickly and easily. Unfortunately we cannot set  up controlled lab experiments on such a large system as Climate Change. But there is a way.

One way scientists test their predictive (e.g. Climate Change) models is to use past data to predict the present situation.  If you have climate data going back, say 100 years, you can plug it into the model and see how well it predicts the climate now. If the results look good, this increases your confidence in the model’s  prediction for the next 10 to 20 years.  Like Science itself, Climate Change models are evolving and making  better, more accurate predictions. 

The final way we test our models is in the marketplace of ideas. If the overwhelming majority of the expert  scientific community are convinced of the validity of a particular model,  then it is likely our decision makers will also be persuaded. But this is not always the case; sometimes politics trumps logic. But just as we all  hope that Good will eventually triumph over Evil, I continue to hope that reason will rise above politics and that the good work of scientists worldwide will preserve our planet and our species. As the TV pundits say, “Only time will tell.” 


Let’s hope that as we create our future environment we are guided by reason and science rather than those with a personal economic or political agenda.

Search This Blog