A multi-system comparison of forecast flooding extent using a scale-selective approach

Fluvial flood forecasting systems increasingly couple river discharge to a flood map library or a real-time hydrodynamic model to provide forecast flood maps to humanitarian agencies. The forecast flood maps can be linked to potential impacts to inform forecast-based financing schemes. We investigate a new application of scale-selective verification by evaluating three flood forecasting systems. Two simulation library systems, Flood Foresight (30 m) and GloFAS Rapid Flood Mapping (1,000 m) and one hydrodynamically modelled system, the Bangladesh Flood Forecasting and Warning Centre (FFWC) Super Model (300 m), all made predictions of flooding extent at different spatial scales (grid lengths, in brackets) for the Jamuna River flood, Bangladesh, July 2020. The flood maps are validated against synthetic-aperture-radar-derived observations of flooding using a scale-selective approach that can compare directly across different spatial scales. At short forecast lead times, the Super Model outperforms the other systems. Near to the Bangladesh border, the trans-boundary benefits of the two global systems are evident. We find that scale-selective methods can quantify the skill of systems operating at different spatial scales so that the benefits and limitations can be evaluated. Multi-system comparison of flood maps is important for improving impact-based forecasts and ensuring funds and response activities are appropriately targeted.