The Purpose of Meaning of Computer Models


Before we plunge into creating and experimenting with models of earth systems, it is extremely important to have a clear idea of why we create models, what our goals are, and how we should view the results of the models. Computer modeling may seem simple in that the machine does all the difficult math for you and it spits out impressive numerical results. But do not be seduced by such thinking. One reason for rejecting this idea has to do with the fact that you must do a great deal of thinking when you construct the model (more on this later). Another reason is that it is not so easy to understand how to interpret the results, or more to the point, it is hard to decide what to conclude from the results of these computer models.

The purpose of a model is not to replicate the real world - it is clearly impossible to put the complexity of the real world into a computer. Instead, the goal is usually to understand something about the behavior of systems and also to understand the affects of changes on the system - its response to changes. We try to get a sense for which parts of the system control its behavior. We create and use these models because the their real-life versions are so complex that we cannot generally understand them without some kind of controlled experimentation. At their best, models capture the essence of real world systems, they exhibit similar, though not identical behaviors. In some cases, a model may provide accurate predictions, but that is generally not the goal.

I think that there are two polarized impulses regarding these computer models. Some people have a tendency to think that the computer does not make mistakes - its results are infallible and should be accepted and believed, case closed. This is one extreme. The other extreme is summed up as "garbage in, garbage out", but more generally, this amounts to a universal rejection of all computer modeling results on the grounds that there is no way that a computer model can adequately represent all of the complexity of the real world. I emphasize that these are extreme views and in reality, most people wander back and forth between these two poles as they grapple with the task of interpreting the results of computer models. In a sense, these two poles also represent the range of response that some people have to science - some maintain that science yields immutable Truths and is infallible, while others summarily reject scientific results because of uncertainty or disagreement among scientists. How we deal with the results of computer models has to consider what the model is intended to do.

For instance, let's imagine a model of plate tectonics in the very early part of Earth's history. We obviously don't know too much about the details of the dynamics at this time, but we might make some assumptions and then investigate the implications of those assumptions by experimenting with the computer model. This kind of modeling has been done (see Appendix B) and one result is that normal subduction should not have occurred until about 2 billion years ago if the assumptions used in the model are more or less correct. Pay attention to the wording in that last sentence. The modeling does not prove anything, but it does suggest something interesting and motivates some field work in ancient terranes to try to understand if in fact the assumptions used in the model are wrong or alternatively, if there was some other kind of subduction in the very early history of the earth. So, this is a kind of model where the real physical system is very poorly known and one strategy for advancing our understanding is to investigate assumptions about how the system operated.

This kind of model is probably the most common in the study of the Earth because we still do not understand all of the processes and connections well enough to express them in the form of an equation. But these models are still very useful in generating questions and exploring the implications of assumptions. The numerical output of many of these models is less important that the general trends and relative changes.

Another kind of model represents the opposite end of the spectrum of models. We may have a model of a very simple system that involves processes that are very well known and in this case, the computer model is capable of generating numerical results that are very good predictions of how the system behaves. One simple example might be radioactive decay of some isotope. If the decay constant has been determined from repeatable laboratory measurements, we can create a simple model that will very accurately predict the history of decay of some initial amount of that isotope.

A still different class of models is represented by one of the global carbon cycle that we will use later in this book. The global carbon cycle is clearly a huge, complex system, and is poorly understood, although a great deal has been learned in just the past few years. Many of the processes involved in this cycle are difficult to express in the form of an equation that captures the behavior of a process occurring all over the globe. But, many of the processes can be described in a way that includes rate constants just like the ones we used in the bathtub model. We also know something about how the real global carbon cycle has behaved in the last few decades and it is possible to adjust the rate constants so that the model can reproduce the "known" behavior of the global carbon cycle. This kind of model thus generates numerical results that are somewhere between the other two types of models - we may not have a great deal of confidence in the actual numbers generated by the model, but we have reason to believe that those numbers may not be too far off. Again, the reason for this modest confidence is that the model is capable of reproducing the observed behavior of the system.

We have explored some complicated issues here, so let me summarize:

1) Models are made for a variety of reasons; the intent of the model must be kept in mind when thinking about the results.

2) Most of the models we will investigate in this book incorporate numerous simplifications and assumptions and we are really investigating the implications of these assumptions, and looking at relative changes and trends in the behavior of the model. We are more interested in the qualitative results rather than the actual quantitative output of the model.

3) If a model reproduces some of the behavior of the real-life system it is patterned after, we have some cause to cautiously accept the results.

4) A few models involve well known, easy-to-measure processes that can be expressed in simple equations; the results of these models represent precise predictions of the system's behavior that have a high probability of being correct.

5) In general, the main goal of modeling is to improve our understanding of complex dynamics and to generate more questions or to help us understand what the key questions are.