Sensitivity analysis to improve how uncertainty is understood and managed in complex flood risk models: case studies in the River Rhine, Central Europe, and Queensland, Australia
Image: The River Rhine in Bonn, Germany
A flood risk model combines multiple input variables, along with “internal” structural choices, to generate outputs such as expected damages or extreme event scenario losses. Most, if not all, of the input and structural parameters have uncertainty associated with them.
Which of these many parameters should be the focus in efforts to reduce output uncertainties?
Do uncertainties in complex models compound and become too complicated for analysis, leading to an uncertainty paralysis or undermining the use of modelling?
These questions are typical of the complex models that are applied in various contexts to support decisions about environmental risk management and resilience.
In 2017 we started a PhD collaboration with researcher Dr Georgos Sarailidis, working with Prof Francesca Pianosi at Bristol University and Prof Thorsten Wagener (University of Potsdam), and Dr Kirsty Styles from one of our host companies JBA Risk Management Ltd., to explore the use of formal sensitivity analysis methods in flood catastrophe modelling.
The original PhD research led to further development and case studies, supported with models and data from JBA Risk Management Ltd., over large (~100,000 to ~1,000,000 km2) river basins in Central Europe and Australia. Now further outputs from this work have been published in the journal NHESS.
Computing cost is a significant constraint on the use of sensitivity analysis methods with large, expensive and complex models. The paper shows how a principled, multi-variable sensitivity analysis can be approached pragmatically to explore not only the effect on model outputs of individual input variables, but also the effects of interactions between those inputs. The inputs include both the ranges within which model parameters could plausibly vary (for example in the case of depth-damage ratios) and discrete, structural choices (for example, the spatial aggregation level of an exposure portfolio).
A key insight from the work is that the multi-variable exploration of model uncertainties benefits from a balanced and structured understanding of uncertainties in all input and hyper-parameters. Uncertainty about depth-damage functions seemed to have the greatest influence on the outputs.
However, in the Rhine Basin case study, if the uncertainty about these vulnerability functions were to be reduced by a half, then the dominant source of uncertainty would shift to estimations of the probabilities of flood events in the catastrophe event set. In Queensland, the effects of different choices of aggregation level (from point coordinates up to State scale) was seen to be influential regardless of other uncertainties.
The work points the way to extracting better value from knowledge of the uncertainties in individual model inputs, whilst assessing the influence of those uncertainties in relation to choices made in model construction.
Continuity of research partnerships is important to tackle challenging, inter-disciplinary and multi-sectoral problems like this. We are glad to have been able to support a coordinated effort over 10 years, drawing together research strands from multiple sources (see Funding and Support below).
Citation
Pianosi, F., Sarailidis, G., Styles, K., Oldham, P., Hutchings, S., Lamb, R., and Wagener, T.: Towards global sensitivity analysis of large-scale flood loss models, Nat. Hazards Earth Syst. Sci., 26, 1727–1743, https://doi.org/10.5194/nhess-26-1727-2026, 2026.
Available at: NHESS – Towards global sensitivity analysis of large-scale flood loss models
Funding and support
The research strands in the paper draw together support from an EPSRC Centre for Doctoral Training (Water Informatics: Science and Engineering, WISE, grant number EP/L016214/1), an InnovateUK Knowledge Transfer Partnership (KTP 13266), an Alexander von Humboldt Professorship endowed by the German Federal Ministry of Education and Research, and originated from a NERC consortium for research on environmental risk (NERC grant NE/J017450/1).
