Reasoning about Model Complexity with a Multi-scale Approach
Abstract
Model complexity is known to influence uncertainty in predictions through model structure error and insufficiently constrained degrees of freedom. While it has been conjectured that an optimal level of complexity exists to maximise accuracy, it is also recognised that complexity can be difficult to characterise for non-linear models, and fitness for purpose requires more than just accuracy. In a National Socio-Environmental Synthesis Center (SESYNC) pursuit, the authors are exploring the role of deliberation about possible alternative scales in selecting appropriate model complexity. The approach builds on ideas from conceptual modelling, system of systems modelling, model documentation, and model evaluation. Model complexity is conceptualised as an emergent property of a series of decisions that iteratively define the subsystems involved, and the scales at which those subsystems are represented. Possible scale options are made explicit and discussed in terms of the patterns they are expected to be able to reproduce, and corresponding questions and knowledge they can tackle. Emergent behaviour between scales and implications for salience, credibility, and legitimacy are discussed as part of scale selection, providing increased transparency and rigor of argumentation as well as a potential framework for reasoning about integrated system of systems multi-scale and multi-fidelity modelling. Examples from hydrology and integrated assessment and management are discussed.
- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2021
- Bibcode:
- 2021AGUFM.H51C..04G