The role of model complexity in assessing water supply system resilience

Assessing water supply system resilience is becoming increasingly important as system pressures, including climate change and population, increase. The resilience of water supply systems in the UK to droughts of differing intensity and duration is now assessed through the simulation of a range of extreme droughts using water resource models. This process can be time consuming, data intensive and in some cases, may not be proportionate to the system response. The drought response surface has become a useful tool in UK water resources planning to quickly visualise modelled outputs, and therefore resilience, to droughts of differing intensity and duration. Work undertaken as part of the NERC Historic Droughts project has sought to compare results within the drought response surface framework to different levels of modelling complexity. Models of increasing complexity were developed that are rapid, accommodate varying data quality and provenance, and can be applied worldwide. Droughts were characterised for a range of reservoir systems based on duration and system response in the historic record. Flexible linear and polynomial regression models were trained on historic storage data and increasingly complex inputs – firstly rainfall, then site-translated historic reconstructed streamflow, and finally site-specific historic reconstructed streamflow – to hindcast reservoir storage to 1891. System understanding was important in drought duration selection, which was dependent on the length and distribution of reservoir storage in the observed record. The models permit the input of fuzzy data points, accommodating anecdotal evidence related to drought intensity and duration. Extreme value analysis of storage allowed for water company levels of service to be considered, therefore providing a description of the system response of an event in terms of the customer experience. Drought response surfaces produced under increasingly complex inputs were compared with those produced through water resource modelling to demonstrate the implications of different levels of modelling complexity. While simple, statistical modelling approaches that do not incorporate physical parameters should not replace deterministic system modelling, they allow for identification of periods and systems of interest that can focus further, more detailed analysis using more complex models and may have particular applications in data and model poor regions