Comparative Assessment of Artificial Neural Networks and RORB Model in Simulating Catchment Runoff: A Case Study of East Melbourne

Australia's surging climate vulnerability status has amplified the likelihood for prevalence of more recurrent and austere intense rainfall events, triggering an incessant residual flooding risk within the urban landscape. The complex topography and inherent non-linearity of influential hydrological processes of urban catchments limits the prediction accuracy of conventional hydrological models, leading to over/(under)estimation of expected flood flow. Recently, Artificial Neural Networks (ANNs) have shown commendable progress in offsetting most limitations allied with conventional models - especially in simulating highly complex, non-linear relationships without understanding underlying physical processes. Thus, this research proposes an ANN-based enhanced accuracy flow estimation model for accurate simulation of expected flood flow in urban catchments. Gardiners Creek catchment, located in east Melbourne was selected as the study area, where the model is calibrated with historical storm event data and performance is assessed against Victorian Water Industry standard hydrological model, RORB results. The study results depict that the ANN model outperforms the Runoff Routing Burroughs (RORB) model, in terms of superior performance, excellent predictive skill and lower minimal error, highlighting that the ANN model has good understanding and is capable of accurately modelling the current catchment response state to storm events, and can be considered for implementation in improving the accuracy of current flood flow estimation practices, and also for tackling emerging challenges in flood investigation process due to climate change impacts.