Assessment of CI-FLOW and remote sensing-derived flood depths during Hurricane Irene

The Coastal and Inland FLooding Observation and Warning (CI-FLOW) project has developed an innovative coupled modeling system that predicts the combined effects of waves, tides, river flow and storm surge in order to provide routine total water level forecasts for tidally-influenced watersheds. Current evaluations of the modeling system are restricted by the number of land-falling storms in the test basins as well as the number of water-level observations against which model output can be compared. Furthermore, even with available observations from gauges or sensors, validation efforts are only relevant at point locations without considerations of spatial variability over an entire model basin. Such spatial variability is most effectively captured by remotely sensed imagery. The premise of this research is that improved remote sensing measurement of water depth can lead to better validation of the CI-FLOW modeling system, and hence, we can better address issues about spatial and temporal uncertainties, improve the modeling components and their connectivity, and continue to establish the basis for the application of CI-FLOW in other watersheds.

The use of satellite remote sensing is advantageous to monitoring flood extent and depth in sparsely-gauged or ungauged basins and also logically offers common data sources valuable for flood detection during land-falling hurricanes within the CI-FLOW model domain. In August 2011, Hurricane Irene offered the opportunity to investigate CI-FLOW performance in the Tar-Pamlico and Neuse River test basins of coastal North Carolina in real-time. This study utilizes satellite data (i.e. MODIS Rapid Response and SPOT datasets) collected before, during, and after Hurricane Irene's landfall in conjunction with DEM datasets to derive spatial flood extent and height at various times during the storm. Water level observations collected by several federal agencies are used as ground truth measurements to compare with derived flood heights and CI-FLOW forecasts. The ISODATA unsupervised classification algorithm was used to separate the remotely sensed pixels into inundated areas, and spatial analysis determines the flood depth by subtracting underlying DEM data from flood heights. The study developed Python scripts to automate the flood depth derivation procedure and allow scenario analysis by modifying input parameters. Preliminary findings indicate that the derived flood heights are adequately representative of ground-truth observations, which demonstrates potential for improved flood detection in the absence of real-time measurements; however, additional work is needed to distinguish the spatial uncertainty of the derived heights in both magnitude and extent in order to compare them with CI-FLOW forecasts.