Sensitivity analysis and predictive uncertainty using inundation observations for parameter estimation in open-channel inverse problem

The necessity of an improvement of flood risk management has been emphasized by the importance of the damages caused by several disastrous floods in recent years. An obligatory way to better understand and forecast these events is to be able to accurately simulate the water transport in the river. Some parameters embedded in open-channel flow equations are difficult to calibrate in practice and hydraulic models are currently constrained by lack of data. The objective of this study is to test the potential of data typical of remote sensing imagery which may be used together with data assimilation methods to ascertain the value of a set of open-channel model parameters. Indeed, for the modeling of ungauged rivers, it may be necessary to estimate open-channel flow and/or morphological characteristics of the river. The innovation of the proposed methodology is to reconstitute information about the geometry of the river from top sight. A sensitivity analysis using the generalized sensitivity analysis method is carried out, together with an estimation of predictive uncertainty using the generalized likelihood uncertainty estimation, a methodology for calibration and uncertainty estimation of distributed models introduced by Beven and Binley (1992). The study aims at building a framework for parameter optimization in the case of ungauged rivers. The proposed method has been implemented on a reach of the Leze River, located in the southwest of France.