Modelling the time-dependent nature of the rainfall-runoff relationship using on-line identification

This paper focusses on the problem of real-time calibration of the time-variant rainfall-runoff relationship. Two different approaches to adaptive modelling of dynamic systems, i.e. firstly the introduction of a "forgetting' factor and secondly the use of an additive Q matrix into the recursive algorithms for the linear parmameter estimation problem, are compared. It is found that, from an operational standpoint, the use of an additive Q matrix leads to better forecasting performance. The choice of the additional design parameters however is not trivial and in this paper it is based on a simulation study using several storm events for subcatchments of the river Meuse, Belgium. -Authors