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

Abstract

The particle filter‐based data assimilation method is an effective tool to adjust model states based on observations. In this study, we proposed a modified particle filter‐based data assimilation method with a local weighting procedure (MPFDA‐LW) for a high‐precision two‐dimensional hydrodynamic model (HydroM2D) in dam‐break flood simulation. Moreover, a particle filter‐based data assimilation method with a global weighting procedure (PFDA‐GW) for the HydroM2D model was also investigated. The MPFDA‐LW and the PFDA‐GW for the HydroM2D model, respectively, adopted spatially nonuniform and uniform Manning's roughness coefficients. The MPFDA‐LW considering spatial‐temporal variability of Manning's roughness coefficient could significantly improve the performances of the HydroM2D model in simulating water stages at all gauges simultaneously, whereas the PFDA‐GW considering temporal variability of Manning's roughness coefficient could only slightly improve the performances of the HydroM2D model in simulating water stages at a few gauges. The MPFDA‐LW is more suitable for improving the performance of 2‐D hydrodynamic models in flood inundation simulation than the PFDA‐GW.

Comments on “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” by Cao et al.

“…In addition, the surface roughness at the 10 selected locations is no different from other places, because the ground of the physical model was made of cement and the surface had been manually smoothed. The smooth surface of the physical model can be seen from the photo of the physical model shown in Figure 2(a) in Cao et al (). As a consequence, the roughness can be considered the same everywhere.…”

“…Cao et al () tested the HydroM2D hydrodynamic model characterized by a constant Manning coefficient based on a benchmark test case (the physical model of the Toce River), and they found discrepancies between the simulated and observed water levels at several observation locations. To reduce the discrepancies, Cao et al () allowed the Manning coefficient to vary at several places; specifically, Cao et al () adjusted the Manning coefficient at 10 selected computational grid cells. They chose these cells simply because water levels were measured at these locations (the MPFDA‐LW method calculated particle weights based on simulated and measured water levels, and then the expectations of the Manning coefficients at these specific grid cells were estimated).…”

“…Another important reason is that the accuracy of the observation positions was inconsistent with the accuracy of numerical models (Caleffi et al, ; Prestininzi, ). The water level measurements were carried out at specific locations, whereas each grid cell covered a certain area, for example, the grid cell in Cao et al () is 0.1 m 0.1 m. Discrepancies between simulated and measured values become more perceptible at locations where the topography of the physical model changes significantly (Paquier & Mignot, ).…”

“…The Manning coefficient temporally varies mainly because of the variation of flow depth. However, the Manning coefficient weakly varies with flow depth, and a change in water depth usually increases or decreases the Manning coefficient by less than 20% to 30%, rather than an order of magnitude change (see Figure 8 in Cao et al, ). A simple evaluation of the impact of varying flow depth on the Manning coefficient can be carried out based on the Keulegan equation: …”

“…The is a parameter varying with different assumptions, and equation performs quite well when (James & Jordanova, ). For the 2‐D hydrodynamic model used in Cao et al (), the ratio of the area of 10 computational grid cells to the overall model area is 1/3519, and thus, changing resistance coefficients at the 10 locations by Cao et al () has little influence on the overall resistance coefficient. In other words, a flooding process is barely affected by changing the Manning coefficient at those locations.…”