Improving runoff predictions by updating the subsurface state in the DDD model from observed runoff

A catchment is in a continuous state of recession although this can be difficult to observe due to added moisture from precipitation or snowmelt. For the updating procedure developed in this study, we need to use the runoff and the subsurface storage estimated for a state of pure recession. For an observed runoff value, the runoff caused by pure recession at the next timestep need to be estimated. This is obtained by the recession characteristic Lambda=log(Q(t)/Q(t+1)), which is estimated through an iteration procedure. Then, the subsurface storage for a state of pure recession can be estimated by assuming that the catchment behaves as a linear reservoir, with parameters specific for the moment of update. The rate constant of the linear reservoir is not all a constant but a function of the recession characteristic. The difference between observed runoff and runoff due to pure recession (distributed in time due to an UH estimated using the recession characteristic and the distance distribution describing the distances from points in the hillslopes to the river network) is the moisture needed to be added to the subsurface state of pure recession, giving us an estimate of subsurface state for the observed runoff. If the subsurface state in the model is different, then we need to add or subtract water to the model so that so the modelled subsurface state and the estimated subsurface state due to the observed runoff are equal. The model with the updated subsurface state has improved runoff predictions, and increased precision can be observed for several timesteps ahead. These are very preliminary, but promising results, and will, if successful, be of considerable value for the forecasting services.