Jump to Main Content
A modified two-level estimator for real time forecasting
- SINGH, KRASHNENDRA, MAJUMDAR, MANDAKINEE
- Hydrological sciences journal 1993 v.38 no.5 pp. 417-430
- algorithms, hydrology, models, probability distribution, rain, uncertainty
- A modified two-level estimator is proposed and implemented for real time flood forecasting for a hydrological system plagued by a high order of uncertainties because of ungauged tributaries and scanty nonrepresentative rainfall information. A finite set of time varying models running in parallel constitutes a coarse estimator for the proposed two-level estimator. The results of this coarse estimator are combined on the basis of a posterior probability distribution; the weighted model thus obtained is fine tuned by a Restricted Exponential Forgetting (REF) algorithm based on a fine estimator. In addition to the usual link between the coarse estimator and the fine estimator, a backward fine-to-coarse estimator link is also suggested. This not only helps in the crucial task of compilation of the finite model set of the coarse estimator but also materially affects the performance of the other components of the estimator. The estimator is shown to yield improved results.