Diagnostic Evaluation of Environmental Models in Data Rich and Data Poor Situations


Prof. dr. ir. dipl-ing. Thorsten Wagener

Chair of Water and Environmental Engineering, Department of Civil Engineering, University of Bristol, Bristol, UK


Predictions of environmental models are a fundamental for solving many scientific and operational problems. What factors control the transport behaviour of a river reach? How sensitive is headwater catchment hydrology to climate and land use change? How uncertain is the outcome of different climate stabilization efforts for different climate change scenarios? These are just three example questions we might ask. While these questions seem very different from each other, I will show how a consistent diagnostic evaluation framework and common methods can be applied to address all of them.


Data availability and limits in our understanding reduce the credibility of environmental model predictions, especially at data poor locations and for projections of change impacts. In this talk, I will discuss and demonstrate a framework for diagnostic evaluation and uncertainty analysis in environmental modelling through a combination of global sensitivity analysis and Bayesian statistics within a signature-based framework. Signatures are indicators that provide insight into the dynamic functional behaviour of environmental systems. These signatures can be derived from historical observations, but also transferred to data poor or ungauged locations through regionalization or from expert elicitation. Using a trading-space-for-time strategy further enables the extrapolation in time, not just in space. Through three different case studies, I will answer the three questions posed in the beginning of the abstract.