Troubleshooting a complex linked one- and two-dimensional hydraulic model

The Ruffey Creek catchment is a 10km2 catchment in eastern Melbourne. AECOM were commissioned by Melbourne Water and Manningham City Council to develop a TUFLOW model of the catchment, which will be used to assess the flood risk related to their assets. Of note is that the modelled council pipe network contains over 9,000 pipes, with a pit connection to the two-dimensional grid at almost every pipe junction, resulting in a model with over 19,000 one-dimensional elements and 690,000 two-dimensional grid cells. Models of this complexity often require extensive troubleshooting but ultimately give stakeholders a more complete understanding of flood behaviour. Distribution of inflows across all pits in the catchment resulted in widespread shallow flooding, with approximately 236,000 wet cells at the peak of the 100 year Average Recurrence Interval (ARI) flood event. Widespread shallow flooding involves a large number of cells wetting and drying during any given time step. Variations in time step and wet / dry depth were tested. Lowering the time step for the two-dimensional domain from 1s to 0.5s caused the final mass error to increase from -0.3% to 2.7% in the 100 year ARI, 30 minute event. Increasing the cell wet / dry depth from 0.002m to 0.05m caused peak cumulative error to quadruple and final cumulative error to be about twenty times as large. Decreasing the cell wet / dry depth to 0.0002 and reducing the time step from 0.75s reduced peak mass balance error in the two-dimensional domain from -6.1 to -4.1% in the 100 year ARI, 2 hour event. Modifications to the software further reduced it significantly, to around +/-0.1% in many cases. In the one-dimensional domain, input data quality was a significant issue. The method adopted to calculate losses at manholes in TUFLOW for this project was the Engelhund Approach, which takes the difference in orientation between inlet and outlet pipes into account. Often mass balance errors occurred where the nature of the pipe network was not completely captured by the GIS data. This paper addresses how the implementation of the Engelhund Manhole Loss Approach resulted in significant improvements in model stability and mass balance errors caused by the pipe network.