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‐days 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.

中文翻译：

---

观测位置，时间和频率对洪水泛化性能的影响

水动力模型中的洪水淹没预测可以帮助洪水的防范，但是输入和参数的不确定性可能导致错误的洪水淹没估计。但是，基于合成孔径雷达（SAR）的洪水范围信息可用于通过数据同化来约束此类模型预测，从而使其更加准确。由于高分辨率SAR卫星只能对中到大型流域提供部分覆盖，因此方便地评估观测覆盖区，时间和频率的组合，这可以最大程度地改善预报。因此，在此模拟了多种基于时空SAR的洪水范围同化方案，以识别用于改善洪水泛滥预报的最佳观测设计。在2011年澳大利亚克拉伦斯集水区洪灾事件的综合设置中，实施了基于互信息的粒子过滤器，以将基于SAR的洪灾范围与水力模型LISFLOOD-FP结合在一起。使用不确定的流入量来强制开环合奏，并针对不同的首次访问和重新访问场景评估了形态均质的河段中同化洪水范围的影响。结果表明，最佳的时间采集策略在很大程度上取决于到达形态和洪水波到达时间。此外，发现在适当的时间使用单一图像可以使大平坦的洪泛区但潮汐影响较小的河段同化，将8天预报提高约95％，而在狭窄山谷的河段中，需要10张以上的图像才能实现同样的结果。