On The Need for an Information-Based Approach to Evaluating Model Structural Hypotheses (Invited)
Abstract
This talk will discuss the problem of building computational models to facilitate understanding and enable testable predictions as a process of 'learning'. Note that there are two aspects to learning, which we might call 'Data Assimilation' and 'Hypothesis Testing', both of which are classically approached as problems of 'Estimation Theory'. In the former, we seek to assimilate information from data, conditional on the assumption that the model structural hypothesis is correct, whereas in the latter we seek to improve our perceptual-conceptual-theoretical view(s) of the world, expressed as model structural hypotheses (assumptions and conjectures), by reconciling them with observations of dynamical systems behavior. Arguably, the second problem is more fundamental, more difficult, and more interesting. This talk will argue that while conventional Estimation Theory (rooted in Maximum Likelihood and Maximum Aposteriori Bayes) provides a valuable theoretical foundation, the strategies we use have become woefully inadequate as models have become progressively more complex. We suggest that the only meaningful way forward is through a formal understanding of the different ways by which 'information' is coded into both model structural hypotheses (as both assumptions and conjectures) and into data, so that a more systematic and powerful approach to testing and improving Earth Systems models, based in diagnostic evaluation, can be achieved. Ultimately, this insight is really a call to formalize, unify and creatively build upon the host of intuitive strategies for model structural assessment that already exist.
- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2013
- Bibcode:
- 2013AGUFM.H31I..01G
- Keywords:
-
- 9820 GENERAL OR MISCELLANEOUS Techniques applicable in three or more fields;
- 1916 INFORMATICS Data and information discovery;
- 1910 INFORMATICS Data assimilation;
- integration and fusion;
- 1847 HYDROLOGY Modeling