Hydraulically based models are used to simulate how sewer networks of urban catchments will respond to precipitation events (Salvadore et al., 2015). These models enable planners to design interventions that might resolve current network issues, and to plan for changes in the urban catchment. However, the models also require the network and catchment to be represented at a high temporal resolution with a fixed spatial representation, resulting in accurate simulations but lengthy simulation times and a lack of flexibility in modeling options. As regulations, such as the UK's newly introduced "Drainage and Wastewater Management Planning," require more computationally expensive applications of sewer network models, such as optimization, real-time control, or scenario analysis to explore the impacts of, for example, changes in climate and land cover (Water, 2019), it is increasingly clear that alternative and more flexible approaches are needed to complement traditional high-fidelity sewer network modeling.

Highlights:-Automatic graph partitioning can flexibly reduce the complexity of sewer networks to enable surrogate modelling -CityWat-SemiDistributed can model these reduced networks without needing parameter derivation from high-fidelity simulations -The combined approach provides computationally cheap simulations and performs accurately even when no high-fidelity model is available