Real-time discharge estimation method using 2D dynamic wave model and particle filter

One of the most important components is Manning roughness coefficient in estimation and prediction of river discharge and water level. Nevertheless, Manning roughness coefficient is determined empirically. Moreover, River discharge data is essential for water resource management and hydrological model calibration. In spite of such an importance, river discharge is generally estimated by using observed water stage data and a rating curve. As an alternative to get more precise discharge data, hydraulic model simulation has been considered in some cases. However, hydraulic model simulation with a deterministic model condition also has limitations in consideration of the time variant river characteristics, and thus it is difficult to predict the prospective state of river flow. In fact, natural river flow conditions are continuously changed depending on season, river geomorphology, in-stream vegetation and so on. In particular, the variations of river flow conditions during flood events are extremely changed, and this phenomenon has been reported many times. In this study, we aimed to present the method which can reflect the natural river flow characteristics (time variant Manning coefficient) within a hydraulic model simulation for more exact analysis and prediction of river discharges and test the tracking ability in case that the disturbed discharge estimated from hydrological model utilized as input conditions. In dealing with this problem, we introduce a simple 2D dynamic wave model, which can reflect the geomorphologic effect, and Monte Carlo sequential data assimilation scheme (or Particle Filtering scheme), which is adequate to non-linear system and able to reflect the time variation of state variable and parameters. Based on the Sequential Importance Resampling (SIR) method within the Particle Filtering scheme, the parameters and state values of the dynamic wave model are sequentially updated to consider the observed water stage information every hour. However, we have a limited data for verifying our method, so we use the synthetic experiment. The method was applied to the middle reach of the Kastura River in Kyoto, Japan. The length of the modeled river channel is about 10km, and there are 3 water level stations and 4 weirs within the channel.