Automatic calibration of a river reach in northern Italy by coupling a parallel 2D shallow water model and the PEST tool

This paper presents the application of the parameter estimation tool PEST to automate the roughness calibration of a river reach, in the framework of parallel two-dimensional (2-D) shallow water modelling. The 2-D GPU-accelerated numerical model PARFLOOD is adopted as forward model to accurately describe the flood propagation processes. The procedure is applied to estimate the roughness parameters in a 38 km-long reach of the Parma River (Italy) characterized by the presence of meanders and floodplains. The calibration is carried out considering a real flood event, using both the water level hydrograph recorded at a downstream gauging station and the high-water marks collected along the river. The procedure takes a few computational hours on a cluster equipped with P100 NVIDIATM Tesla GPUs. The calibrated model is finally validated by reproducing another flood event occurred on the same river.