Improved Implementation of Scenario-Neutral Climate Impact Assessments for Water Resource Systems

Abstract

The uncertainty surrounding climate change presents a major challenge for the management of water resource systems, which are facing stresses to both supply and demand. To provide insight into how a system might perform under change, scenario-neutral climate impact assessments are being used increasingly to supplement ‘scenario-led’ climate projections. Scenario-neutral assessments stress-test a system against a range of potential climate changes, regardless of their plausibility, so that all major modes of system failure can be identified and characterised. This thesis focuses on overcoming several existing challenges with the implementation of scenario-neutral methods for complex systems. The specific aims of this research are to: (i) improve current methods of generating climate perturbed hydrometeorological time series to consider a wider set of changes, (ii) develop a method to identify the critical changes in climate conditions for inclusion in scenario-neutral climate impact assessments, and (iii) identify and demonstrate the key requirements of a scenario-neutral analysis, such that it will be consistent with the outcomes of a scenario-led analysis. These methods are demonstrated using two case studies: the Lake Como reservoir (Italy) for impact assessments, and Adelaide rainfall data (Australia) for time series generation. For the first aim, this research advances the optimisation formulation that underpins an inverse approach to time series generation. This process uses formal optimisation techniques to identify the parameters of a stochastic weather generator that enable the generation of time series with desired climate attributes (statistics of climate variables). The advancements enable a greater number of climate attributes to be perturbed, while ensuring the realism of the time series. This allows scenario-neutral assessments to be implemented for more complex systems that require stress-testing beyond changes to means and seasonality of climate variables. For the second aim, a method to identify the critical climate attributes for a system is proposed, which uses the partial mutual information algorithm to rank a set of candidate attributes in order of significance. Critical attributes are then selected by considering the value of adding an additional attribute given its relative increase in the description of system performance. This allows the resulting scenario-neutral assessment to ensure that the modes of failure identified will be those to which a system is most vulnerable. Applied to the Lake Como reservoir, results show that an impact assessment using identified critical attributes such as frost days uncovers modes of flood prevention and irrigation supply failure not identified by the commonly used attributes average temperature and rainfall. For the final aim, four key pitfalls in the scenario-neutral approach are identified and their effects are demonstrated using a set of diagnostics that compares implementations of the scenario-neutral approach with the results of a scenarioled analysis. Techniques for avoiding the pitfalls are also presented, building on the preceding advances in attribute identification and time series generation. Results show that when these techniques are applied, it is possible to reconcile scenario-neutral and scenario-led approaches. Collectively, this body of research improves the practical application of a scenario-neutral approach to better deliver on its underpinning principles.