A real-time probabilistic channel flood-forecasting model based on the Bayesian particle filter approach

Abstract

Reliable real-time probabilistic flood forecasting is critical for effective water management and flood protection all over the world. In this study, we develop a real-time probabilistic channel flood-forecasting model by combining a channel hydraulic model with the Bayesian particle filter approach. The new model is tested in Highlights • A real-time flood-forecasting model is proposed by assimilating real-time stage observations into a hydraulic model.

A Modified Particle Filter‐Based Data Assimilation Method for a High‐Precision 2‐D Hydrodynamic Model Considering Spatial‐temporal Variability of Roughness: Simulation of Dam‐Break Flood Inundation

“…The model cannot exactly predict the water stages due to the potential errors of model structures, topography, and grid discretization (D. F. Liang, Lin, & Falconer, ; Prestininzi, ). In addition, Manning's roughness coefficient is a spatial‐temporal parameter associated with riverbed roughness and flow conditions (Y. Kim et al, ; Xu et al, ). The HydroM2D model adopts a constant Manning's roughness coefficient, so it is difficult to exactly predict the observed water stages at each gauge with the model.…”

“…First, water stages of particles at gauges far away inflow boundary were less affected by the perturbation of inflow discharge. Second, the diversity of flow states of particles at a certain assimilation time step was significantly reduced after particle resampling through executing the perturbation of Manning's roughness coefficients (Y. Kim et al, ; Xu et al, ). Third, the high‐frequency data assimilation (e.g., 1‐s assimilation frequency) was adopted in simulating the dam‐break flood because it always drastically varied within a short period (Aureli et al, ).…”

“…The EnKF has been widely applied in hydrodynamic modeling. In the EnKF, it is assumed that follows a Gaussian distribution so that the posterior states are only determined by the first two moments (the mean and the covariance) of the prior density (Xu et al, ). However, the assumption is generally invalid in nonlinear hydrodynamic models (Pasetto et al, ).…”

“…The PF is a sequential data assimilation method utilizing Monte Carlo ensemble filter methods (Moradkhani, Hsu, et al, ). Compared to the widely used EnKF, the PF can relax the assumption of the EnKF that the prior distributions of model states are Gaussian distributions (K. Kim et al, ; J. C. Neal et al, ) and is suitable for nonlinear and non‐Gaussian hydrodynamic models and water environment models (Mattern et al, ; Xu et al, ). For the nonlinear and non‐Gaussian HydroM2D model, its recursive formulation of model states can be expressed as where f is the HydroM2D model; x t +1 and x t are, respectively, the model states (water stages and discharges) at t + 1 and t time steps; θ t + 1 and u t +1 are, respectively, the model parameter (Manning's roughness coefficient) and forcing data (inflow boundary conditions) at t + 1 time step; and v t +1 is the random white noise associated with the uncertainty of the HydroM2D model structure at t + 1 time step with a mean of zero and standard deviation σ m .…”

“…Hydrodynamic models are effective tools to simulate water flows in natural and artificial water systems and have been widely used in real‐time flood forecasting (Xu et al, ), dam‐break flow simulation (Ye & Zhao, ), and sediment transport and water quality prediction (Camacho et al, ; Coraci et al, ). Water stages and discharges of dam‐break flood often experience drastic changes within a short period (Aureli et al, ), so the hydrodynamic model with a high performance is required for accurately simulating dam‐break flood inundation.…”