by Joseph Guillaume
At the iEMSs2016 conference in Toulouse, the session on Decision Making under Deep Uncertainty (see blog report) was accompanied by a more generic one on “Managing Uncertainty”, organised by Joseph Guillaume (Aalto University), Tony Jakeman (Australian National University), Holger Maier (The University of Adelaide), Jiri Nossent (Flanders Hydraulics Research and Vrije Universiteit Brussel) and Evelina Trutnevyte (ETH Zurich).
The session emphasised the diversity of approaches for managing uncertainty. Contributions notably covered sensitivity analysis, scenario analysis, parameter estimation and uncertainty quantification. While not directly tied to Decision Making under Deep Uncertainty, it is important to remember that these techniques form the foundations of our analyses – the means of addressing any uncertainty that is not treated as deep. As argued in a recent publication in Environmental Modelling and Software (Maier at al. 2016), multiple paradigms for modelling the future tend to co-exist, with different parts of an analysis focussed on capturing best available knowledge, quantifying uncertainty, and exploring multiple plausible futures.
Sensitivity analysis contributions included methodological developments generalising existing methods (Variogram Analysis of Response Surfaces), and applications using sensitivity to guide model structure selection in hydrological models, and to guide collection of data in honeybee models, accounting for changes in importance of parameters over time.
Scenario analysis included a variety of examples of different approaches, illustrating the breadth of existing practices as well as ongoing creativity with new practices and new ways of applying them:
• melding an MCDA framework and deep uncertainty concepts into existing airport expansion planning;
• development of an interactive online tool using scenario ensembles for the public to express informed preferences regarding energy transitions;
• use of cross-impact balance analysis to improve consistency of societal aspects in context scenarios;
• development of exploratory scenarios for disaster risk planning based on challenges to resilience and challenges to mitigation;
• use of decision trees in a variety of ways to draw insights from a group of scenarios related to water resource management;
• combination of participatory scenario building with use of cellular automata to explore land use change
Parameter estimation and investigation of model identifiability is critical to understanding to what extent observed data help to understand system behaviour. Contributions included case studies using multiple data sources to reveal limitations of those data sources and of a vegetation model; accounting for uncertainty in data to help better constrain system behaviour; and using alternative data sources to calibrate models in data scarce conditions, in hydrological models.
Uncertainty quantification case studies included: a systematic analysis of effect of assumptions in creating design droughts; effect of using different weather input datasets in agriculture; use of mathematical uncertainty propagation techniques in water distribution networks; multi-method calculation of groundwater recharge, including spatialisation; and effect of climate data scale in net primary production modelling.
Apologies to the authors for any oversimplification in this synthesis. Abstracts and full papers are available in the proceedings on the iemss.org website, with session B5 on pages 537-569 of volume 2.
iEMSs2018 will be in Fort Collins, Colorado.