Which Models Are Good (Enough), and When?

Journal Article (Chapter)

With the possible exception of some basic theories such as quantum mechanics and relativity, scientific models are by nature incorrect - at best they correspond to a natural system in limited ways. This is especially true of models used in Earth-surface-process research, whether conceptual, analytical, or numerical; they all involve simplifications, abstractions, and parametrizations, either intentionally or out of necessity. A model can be wrong in an absolute sense - for example, if the processes represented do not correspond at all to those in the natural system of interest. However, in most cases, rather than evaluating whether a model is right or wrong, it is more appropriate to ask whether the model is useful. The answer often depends on the context and the purpose at hand. For example, a highly simplified model developed as a theoretical tool might be useful in developing explanatory insights, and yet not be useful for simulating the behavior of a natural system with some required level of quantitative accuracy. The types of predictions that are most appropriate for testing models differ between those intended for theoretical and simulation purposes.Evaluating models often also involves a subjective element, with many researchers having more faith in models that represent physical conservation laws more explicitly, which can correspond to resolving finer spatial and temporal scales. However, when sediment transport and/or biology are involved in landscape evolution or pattern formation, basing a model explicitly on the conservation of momentum is generally impractical or impossible. All Earth-surface-process models rely on parametrization of processes at unresolved scales, and we use such parametrization whenever we treat variables that emerge at the macroscopic level (e.g., pressure and density). Embracing this approach when addressing landscape change on relatively large scales, and developing empirically based parametrizations at scales much larger than those of laboratory experiments, could facilitate more effective modeling, in terms of quantitative accuracy as well as explanatory clarity. © 2013 Elsevier Inc. All rights reserved.

Full Text

Duke Authors

Cited Authors

  • Murray, AB

Published Date

  • January 1, 2013

Volume / Issue

  • 2 /

Start / End Page

  • 50 - 58

Digital Object Identifier (DOI)

  • 10.1016/B978-0-12-374739-6.00027-0

Citation Source

  • Scopus