Quantifying flood model accuracy under varying surface complexities

Floods in urban areas which feature interactions between piped and surface networks are hydraulically complex. Further, obtaining in situ calibration data, although necessary for robust simulations, can be very challenging. The aim of this research is to evaluate the performance of a commonly used deterministic 1D-2D flood model, calibrated using low resolution data, against a higher resolution dataset containing flows, depths and velocity fields; which are replicated from an experimental scale model water facility. Calibration of the numerical model was conducted using a lower resolution dataset, which consisted of a simple rectangular profile. The model was then evaluated against a dataset that was higher in spatial resolution and more complex in geometry (a street profile containing parking spaces). The findings show that when the model increased in scenario complexity model performance was reduced, though most of the simulation error was < 10% (NRMSE). Similarly, there was more error in the validated model that was higher in spatial resolution than lower. This was due to calibration not being stringent enough when conducted in a lower spatial resolution. However, overall the work shows the potential for the use of low-resolution datasets for model calibration.

Automated summary

The aim of this research is to evaluate the performance of a commonly used deterministic 1D-2D flood model, calibrated using low resolution data, against a higher resolution dataset containing flows, depths and velocity fields. The findings show that model performance was reduced as the scenario complexity increased, but most of the simulation error was < 10% (NRMSE). Additionally, the validated model with higher spatial resolution had more error compared to the lower resolution model due to less stringent calibration at lower spatial resolution. However, overall the study demonstrates the potential of using low-resolution datasets for model calibration.