Performance and sensitivity analysis of stormwater models using a Bayesian approach and long-term high resolution data

Stormwater models are important tools in the design and management of urban drainage systems. Understanding the sources of uncertainty in these models and their consequences on the model outputs is essential so that subsequent decisions are based on reliable information. Model calibration and sensitivity analysis of such models are critical to evaluate model performance. The aim of this paper is to present the performance and parameter sensitivity of stormwater models with different levels of complexities, using the formal Bayesian approach. The rather complex MUSIC and simple KAREN models were compared in terms of predicting catchment runoff, while an empirical regression model was compared to a process-based build-up/wash-off model for stormwater pollutant prediction. A large dataset was collected at five catchments of different land-uses in Melbourne, Australia. In general, results suggested that, once calibrated, the rainfall/runoff models performed similarly and were both able to reproduce the measured data. It was found that the effective impervious fraction is the most important parameter in both models while both were insensitive to dry weather related parameters. The tested water quality models poorly represented the observed data, and both resulted in high levels of parameter uncertainty.