The hydraulic roughness of the main channel of most lowland rivers is dominated by bed forms. River bed forms act as roughness to the flow, thereby significantly influencing the water levels, which are essential for flood forecasting. We compared a time-lag model and a physically based pickup and deposition model to predict dynamic bed form evolution during a flood wave in the flume and the field. The results showed that the explicit computation of bed form and associated roughness predictions perform equally well as a calibrated model for the flume case, but slightly less for the field case. We were able to explain a large part of the roughness of the main channel that is normally calibrated. Using a physically-based roughness prediction improves the accuracy of the modelled water levels for operational flood forecasting.