Real-time channel flood forecasting model with multi-segmental roughness coefficients updating dynamically based on ensemble Kalman filter

---

Xingya Xu, Hongwei Fang, Lei Huang, Yuefeng Zhang, Ruixun Lai, Xiaobo Liu

Thursday 2 july 2015

12:42 - 12:45h at Europe 2 (level 0)

Themes: (T) Flood risk management and adaptation, (ST) Early warning systems

Parallel session: Poster pitches: 11K. FloodRisk - Early Warning

---

Reasonable estimation of roughness coefficient is one of the important and effective ways to improve the accuracy of channel flood forecasting model. The value of the roughness coefficient is controlled by the channel physical condition, like bed geology and cross-section geometry, and affected by flow condition, like submerged extent of vegetation and flow velocity. Given the longitude variation of channel physical condition and unsteady state of flood flow, roughness coefficient in the flood forecasting model should have temporal-spatial character. A new method is developed to consider both spatial distribution and time-varying process of roughness coefficient in the channel flood forecasting model by using simultaneous stage observations of multiple gauge stations. The river channel is spatially divided into several river segments which are assigned independent roughness coefficients, taking the locations of the gauge stations as the interior boundaries in the channel. At the time step of real-time stage observations available, the optimal stages at the gauge stations will be estimated based on ensemble Kalman filter, which is a Monte Carlo sequential data assimilation algorithm considering both the errors of observation and model. Taking the estimated optimal stages as the targets, roughness coefficients of each segment can be updated to match the current flow state respectively. The updated roughness coefficients are used in the flood forecasting model for the prediction calculation of the coming flood flow. An application case of real-time forecasting of a real flood event in Yellow River downstream shows that, the accuracy of model predictions can be improved effectively with multi-segmental roughness coefficients updating dynamically.