Author(s): Kohji Tanaka; Masayuki Sugiura; Hiroki Tsujikura

Linked Author(s): Kohji Tanaka

Keywords: Conceptual runoff model; Flood forecasting system; Unscented Kalman filter; Particle

Abstract: Flood prediction systems for small- and medium-sized river basins were constructed without using the H-Q curve equation. We applied the unscented Kalman filter (UKF) for feedback, because we previously confirmed its application in this scenario. UKF is capable of being applied to the non-linear model; it is superior to particle filtering, which spends too much time calculating particles and assimilating observed data. However, filtering precision, validity of Gaussian distribution for error distribution, and stability of the filtering have not been inspected. In this paper, the accuracy of UKF is compared to that of particle filtering. The error distribution was assumed via a hierarchical Bayesian method. Then, the filtering effect of a rainfall error, given intentionally, was confirmed via the bootstrap method. Therefore, we show that UKF can be applied to a simple water-level model from the aspects of precision, application range, and stability.

DOI:

Year: 2018