Diagnostic approaches to the identification of environmental models based on data and system signatures
Post Doctoral Fellow, Ecohydrology Research Group
Earth & Environmental Sciences, University of Waterloo
Abstract
Mathematical models play a crucial role in supporting our understanding and predicting the response of environmental systems. The most important task of a modeller is the identification of an appropriate model (i.e. combination of a proper model structure and its corresponding parametrization) in the presence of uncertainties in the data, parameters, and the conceptual structure. From a system perspective, the estimation of model parameters and structure requires exploring and learning from all sources of available data. The “classical approach” to model identification is rooted in statistical Regression theory, Likelihood theory and Bayesian analysis. Although this approach is still widely applied in the environmental modelling, it is limited in its applicability to the complex models now used in Earth System science. Building upon this foundation, and considering models as working hypotheses of the environmental systems, recent approaches to model identification rely on the adoption of a more informative “diagnostic approach” to evaluating existing model hypotheses. Such approaches are also based on reconciling models with data to detect and correct model structural inadequacies. This presentation briefly reviews statistical methods for the identification of environmental models in the presence of uncertainty, and further, elaborates on multiple incisive diagnostics that can be used to scrutinize multiple model representations against observed data. Focusing on hydrological modelling case studies, it will demonstrate how system signatures calculated based on different data sources can be implemented in a powerful and systematic approach for model development in hydrology. It will also show how data-based conceptual approaches to understanding the behaviour of natural systems can be used to constrain the outcome of hydrologic models. This presentation is helpful for researchers who are interested in going beyond statistical analysis of their data, and use system signatures to identify a proper model that can reproduce the expected behavior of environmental systems.
Everyone welcome. Coffee provided.