Diagnosing the location of uncertainty in urban drainage models with hydrologic and hydraulic signatures: a real case study with a complex internal overflow structure

Utilities are starting to apply digital twins of urban drainage systems in their daily operation, where observations from sensors are applied together with simulation models to provide information on e.g. documentation of service levels or combined sewer overflows or to analyze when to make predictive maintenance (Pedersen et al., 2021a). The uncertainty in the simulation models needs to be addressed and explicitly highlighted, as non-expert users often expect high confidence of the simulation model in digital twin for all processes and situations. Such accuracy and precision is however rarely the case in reality, where model estimates and in-sewer observation can deviate significantly. Typically, a model is calibrated heuristically based on observation from a limited period and considered validated, but caution needs to be taken as there are many other uncertainties that can affect you model results. By applying signatures (Gupta et al., 2008) to models and observations we can identify uncertainties in integrated urban drainage models (Pedersen et al., submitted). Uncertainties can be classified into four locations: context, input (external forcings or system attributes), model structure (conceptual as physical attributes or processes, mathematical as the spatial and temporal variability, and equation or computational uncertainty) and parameter uncertainty (Gupta et al., 2012; Walker et al., 2003). In this work, we show how hydrologic and hydraulic signatures can be used to diagnose uncertainties in integrated urban drainage models with a real-world example of a complex sewer overflow structure.