7 th International Conference on Hydroinformatics HIC 2006, Nice, FRANCE 1 INVERSE MODELLING FOR FLOOD PROPAGATION ROLAND K. PRICE UNESCO-IHE, P.O. Box 3015, 2601 DA Delft, The Netherlands W. A. YOHAN S. FERNANDO Central Engineering Consultancy Bureau, 415 Bauddhaloka Mawatha, Colombo 07, Sri Lanka DIMITRI P. SOLOMATINE UNESCO-IHE, P.O. Box 3015, 2601 DA Delft, The Netherlands Inverse modelling is used to extract information contained in discharge time series at upstream and downstream gauging stations of a reach of river using a physically based flood routing equation. This equation is dependent on the kinematic wave speed and attenuation parameters, which are non-linear functions of discharge. An approximate time series for the actual lateral inflow along the reach is determined by applying a low pass filter to the lateral inflows calculated knowing the upstream and downstream discharge time series. We need the filter to remove some of the model errors that are inherent to the calculated lateral inflows. Parametric functions are determined for the kinematic wave speed and attenuation parameter by minimising the mismatch between the observed downstream discharges and those calculated from the basic flood routing equation using the filtered lateral inflow. We apply the approach to the River Wye, UK. INTRODUCTION Consider two gauging stations upstream and downstream of a reach of river. The time series for the downstream discharge contains information about the complex geometry and ungauged lateral inflows along the river as well as the upstream discharge. Can this information on the geometry and inflows be derived from the time series in the absence of cross sections and of rainfall-runoff calculations for the lateral inflows? The overall objective is to generate a model that will predict flows downstream and ultimately form the basis of a flood forecasting system. This paper focuses on the calibration using inverse modelling of a non-linear, physically based model that permits extrapolation for more extreme flood events. PROBLEM DEFINITION The 1D Saint Venant equations are a good basis for modelling flow in a reach of river with significant flood plains and no embankments, but their solution requires detailed information of the river geometry and knowledge of temporal variables such as lateral inflows. Without this information, a solution is difficult to obtain. The equations can