资源环境科技发展态势分析平台

Flood inundation forecasts from hydrodynamic models can help with flood preparedness, but uncertainty in the inputs and parameters can lead to erroneous flood inundation estimates. However, Synthetic Aperture Radar (SAR)‐based flood extent information can be used to constrain such model forecasts through data assimilation thus making them more accurate. Since high‐resolution SAR satellites can only provide partial coverage for medium to large catchments, it is expedient to evaluate the combination of observation footprint, timing, and frequency which can lead to maximum forecast improvements. Consequently, multiple spatiotemporal SAR‐based flood extent assimilation scenarios have been simulated here to identify the optimum observation design for improved flood inundation forecasts. A mutual information‐based particle filter was implemented in a synthetic setup for the 2011 flood event in the Clarence Catchment, Australia, to combine SAR‐based flood extents with the hydraulic model LISFLOOD‐FP. The open loop ensemble was forced using uncertain inflows and the impact of assimilating flood extents in morphologically homogenous river reaches was evaluated for different first visit and revisit scenarios. Results revealed that the optimum temporal acquisition strategy strongly depends on reach morphology and flood wave arrival timing. Further, it was found that a single image at the right time could improve the 8‐day forecast by ∼95% when assimilated at reaches with large flat floodplains but limited tidal influence, while in reaches with narrow valleys over 10 images were needed to achieve the same outcome. Experiments such as the one presented here can therefore inform targeted observation strategies to ensure cost effective flood monitoring and maximize the forecast accuracy resulting from flood extent assimilation.