How to combine the efficiency of depth-averaged shallow flow equations with the accuracy of a free-surface model?