The value of sensitivity analysis for the evaluation and use of water resource models under deep uncertainty

Computational modelling provides a vital tool to evaluate risks and benefits of different investment or management options on a virtual system before they are implemented on real water resource systems. In England and Wales, models are used to inform a range of decisions across different spatial and temporal scales – from company-level operational decisions during individual drought events to strategic infrastructure investment decisions at the national scale. Model outputs though are conditional on a range of uncertain assumptions and input data, due to our incomplete or imperfect knowledge of the drivers and the properties of the system being modelled. When models are used for long-term planning, the uncertainty about the current properties and drivers of the system is compounded with deep (i.e. poorly characterised) uncertainty about how these will evolve in the future.

In this talk we will present results from the USARIS (Uncertainty quantification and Sensitivity Analysis for Resilient Infrastructure Systems) project [ST/Y003713/1], which aims at setting the foundations for integrating Uncertainty Quantification and Sensitivity Analysis (UQ&SA) functionalities in the UK DAFNI (Data and Analytics Facility for National Infrastructure) platform (https://www.dafni.ac.uk/). We will discuss the value of global Sensitivity Analysis to systematically analyse the impact of varying uncertain factors and decision levers on model predictions and hence improve both the model evaluation and its use for decision-making under deep uncertainty - and demonstrate it by application to Pywr-WREW, the Python-based national-scale water resources model for England and Wales.

We will focus on a complex, multi-reservoir system in the Northumbrian region, and analyse the relative influence of the model’s decision levers (changes to operational preferences and management decisions) and uncertain inputs and properties (future climate, demand and environmental flow requirements) on a range of performance metrics. At the model evaluation stage, the global SA helps us to sense-check the model (i.e. making sure that the “right” input controls the “right” output) and to ensure that model predictions are sufficiently controlled by decision levers relative to the impact of other uncertain factors (otherwise the model would not be suitable for decision-making). At the options appraisal stage, the same methodology can be used (under the assumption “system=model”) to determine the key drivers of the future system performance (e.g. supply-side vs demand-side) and begin to identify “robust” decisions that work sufficiently well across a range of uncertain futures. Finally, we will discuss the scalability of our proposed approach to more complex/larger-scale systems, and blockers and enablers for uptake by practitioners in water companies and environment agencies.