Data assimilation of flood maps in a 2D flood model for different performance measures

This study explores the assimilation of uncertain flood maps derived from satellite observations into a 2D hydraulic model to calibrate a spatially varying roughness parameter, which is divided into subdomains. First, a sensitivity analysis is carried out to identify the most influential subdomains for various performance measures. Afterwards, Data Assimilation, specifically the three-dimensional variational method, is used on the reduced subset of parameters to calibrate the roughness field. Different cost function formulations are considered using various performance measures to compare the misfit between observed and simulated flood maps. Using synthetic and real observations, the study evaluates how these measures influence the calibration of model parameters and flood prediction accuracy. The application case is the Garonne River in France, where Synthetic Aperture Radar satellite images were available during flood events. The results indicate that the choice of performance measures in the cost function influences the calibration process. Furthermore, the study highlights that the current precision of derived flood maps is not sufficient for calibrating floodplain roughness.