Keywords
hydrologic modelling, identification, uncertainty, consistency, behavior
Start Date
1-7-2006 12:00 AM
Abstract
Current model identification strategies often have the objective of finding the model or model structure which provides the best performance in reproducing the observed response of a system at hand. Such a strategy typically favours more complex (bottom-up) models with a higher degree of freedom and thus larger flexibility. While this bias can be reduced through punishing models for being more complex, real advancements in our understanding with respect to appropriate system representations are made if we quantify the extent to which our model is consistent with the available data. In particular the idea of an optimal parameter set is very weak in the context of highly uncertain environmental modelling exercises using uncertain data and models. This paper discusses the problem of testing model consistency with the aim of falsifying models that are inconsistent with observations or underlying assumptions (e.g. stationary model parameters). Such a strategy can then be included in a general framework for evaluating performance, uncertainty and consistency for model identification.
Consistency versus Optimality in Environmental Model Identification under Uncertainty
Current model identification strategies often have the objective of finding the model or model structure which provides the best performance in reproducing the observed response of a system at hand. Such a strategy typically favours more complex (bottom-up) models with a higher degree of freedom and thus larger flexibility. While this bias can be reduced through punishing models for being more complex, real advancements in our understanding with respect to appropriate system representations are made if we quantify the extent to which our model is consistent with the available data. In particular the idea of an optimal parameter set is very weak in the context of highly uncertain environmental modelling exercises using uncertain data and models. This paper discusses the problem of testing model consistency with the aim of falsifying models that are inconsistent with observations or underlying assumptions (e.g. stationary model parameters). Such a strategy can then be included in a general framework for evaluating performance, uncertainty and consistency for model identification.
