CMWRXVI . 1 DATA ASSIMILATION TO IMPROVE FORECAST QUALITY OF RIVER BASIN MODELS ANNE KATRINE V. FALK, MICHAEL B. BUTTS, HENRIK MADSEN, JOHAN N. HARTNACK 1 1 DHI – Water & Environment, Agern Allé 5, DK-2970 Hørsholm, Denmark ABSTRACT This paper evaluates alternative data assimilation formulations based on the Ensemble Kalman Filter for flood forecasting. A general framework is developed for a combined river and catchment model to perform uncertainty propagation and data assimilation for hydrological forecasting and simulation. This Ensemble Kalman Filter framework allows inclusion of uncertainty in the observations, parameters and boundary conditions. Data assimilation improves forecast accuracy by using observations up to the time of forecast thereby providing the best initial conditions for a forecast. In this study two implementations of data assimilation using measured river flows are investigated. In the first only states in the hydraulic (river) part of the model are updated assuming that uncertainty arises directly from uncertainty in the catchment inflows from the rainfall-runoff model. In the second, updating is carried out on states in both the hydraulic (river) and hydrological (catchment) parts of the model, assuming uncertainty arises from the catchment rainfall. The results show that for volumetric errors, data assimilation in both the river and catchment states provides more accurate forecasts over longer lead times than updating on the river channel alone. However, in the case of phase errors the performance of the two methods are comparable. 1. INTRODUCTION Ideally, real-time flood management decisions must be based on an understanding of the uncertainties and associated risks. It is therefore central for effective flood forecasting tools to provide reliable estimates of the forecast uncertainty [Butts et al., 2004]. Only by quantifying the inherent uncertainties involved in flood forecasting, can effective real-time flood management and warning, be carried out [Cadman et al., 2006]. Estimating forecast uncertainty requires the estimation of the uncertainties associated with the hydrological model inputs (e.g. observations or forecasts of precipitation), model structure, parameterisation and calibration, and methodologies that predict how the uncertainties from different sources propagate through the hydrological and hydraulic system. During a flood both lives and property are at risk and therefore any means to reduce forecast uncertainty are highly desirable. WMO has identified updating or data assimilation as an essential requirement for accurate flood forecasting [WMO, 1992]. Within the EU 5 th framework project FLOODRELIEF, an ensemble-based approach has been developed to estimate forecasting uncertainties and reduce uncertainty using data assimilation. This general stochastic framework for flood forecast modelling is based on the Ensemble Kalman Filter. The Ensemble Kalman filter was introduced by [Evensen, 1994] and provides a natural framework for determining how the different sources of uncertainty