Calibrating 2D Flood Models in the Era of High Performance Computing

Hydrodynamic models which solve the 2D Shallow Water Equations, and commonly used for flood modelling, are considered to be mechanistic simulators. However, even they are based on physics, they still incorporate grey-box parameters which should be calibrated. One of the major limitations until now, except the lack of data, was the computational cost. In our era, parallel coding and the boost of High Performance Computing facilities made feasible the calibration of the required parameters. In this work, we discuss this potential using the UniCal simulator at the Mandra (Greece) 2017 flood event. Taking into account that usually the flood datasets are not sufficiently informative since they are not homogenized in the field, we suggest that the principle of parsimony is the most suitable strategy. First, with the reduction of the dimensional space after the screening of the parameters through a global sensitivity analysis method, such as the Morris method, in order to reduce the impact of equifinality. Second, with the use of a simple optimization method, such as the grid-search calibration, instead of the more sophisticated evolutionary algorithms which are time consumable but do not guarantee that they can find an optimum solution much more better than the brute-force methods.