Identifications and Verifications of System Parameters

Applications of Kalman filter to analysisses of snow melting (I)

Abstract

The precise estimates of snow melt are essential to analyze snow melting. Snow melt is estimated based on the DDF method or by the use of the Runoff Model for separation of runoffs due to snow melt. However, since the hydrological surveys regarding the basin are insufficient to analyze snow melting, the both methods may cause frequent errors in estimating snow melt. Especially, the lack of observed data for the vertical distribution of precipitation and snow depth regarding the basin lead have numerous errors.
In this paper, estimated snow melt being based on the DDF method and observed raifall were used as the inputs for LST-II model.
All depths of the LST-II model were regarded as components of the state vector when the Kalman filter was used. The parameters of the Kalman filter were determined for the analysis of snow melting based on daily hydrological data. The details of the procedures and the obtained results are as follows:
1) Value of DDF was set as 6.0mm/(°C·d).
2) Dummy rainfall was adopted for the purpose of improvement of the system.
3) Dummy rainfall was estimated to be 5.0mm/d for the analysis of snow melting in this basin.
4) Covariances of the system noise and observations were determined based on the simulations of the LST-II model using a Kalman filter.
5) The depth of the second tank in LST-II model were controlled within the upper limit restricted by the physical dimensions of the model, so that the Kalman Gains of the four tanks could be modified by referring to the control of the depth of the second tank.
6) It could be shown that the Kalman predictions for daily runoffs by the use of the identified covariances and the dummy rainfall agree well in many cases with observed daily runoffs, with the exception of those with a abrupt rise.