This article discusses a coupled water level forecasting system constructed for the River Dee (UK) using parsimonious, physically interpretable, time series models. Tidal forecasts, provided by a simple harmonic model, and observed water levels at the upstream boundary are used to drive a nonlinear hydrological model which forecasts water levels at three gauged sites on the flood plain. The assimilation of observed data and use of the model for real-time forecasting is presented. The results generated indicate that the forecasts of river water can be both timely and accurate except close to the tidal boundary where the the tide is affected by the weir at Chester.