Repeat-pass L-band UAVSAR images for flood extents mapping during Hurricane Florence

Abstract:

Extreme precipitation events are intensifying due to a warming climate. However, it is challenging to delineate flood-affected areas through cloudy optical images, and publicly available C-band Synthetic Aperture Radar (SAR) in the densely forested Carolinas. In the fall of 2018, Hurricane Florence produced heavy rainfall and subsequent riverine flooding, causing substantial damage in the coastal plain. NASA/JPL carried out a series of flights using the Uninhabited Aerial Vehicle Synthetic Aperture Radar (UAVSAR) sensor in the aftermath of Hurricane Florence, collecting numerous L-band SAR tracks over the record-setting river stages. With better vegetation penetration capability than C-band SAR, higher spatial resolution (5 meters), and full-polarized repeat-pass flight tracks at daily intervals, UAVSAR images provide an excellent opportunity for tracking flooded areas. Especially for flooded forests, radar return signals have been enhanced due to the “double bounce effect”.

Here, we constructed a flood detection framework. Specifically, we first derived decomposed variables using polarimetric decomposition methods. By applying a change detection algorithm on decomposed variables from two flights, we delineated the approximate open flooded areas. Finally, based on the references collected from the detected open flooded areas and visual interpretation of the high-resolution Google Earth images as well as the UAVSAR decomposed variables , we developed and verified a Random-Forest (RF) classification model. Validation showed that, compared with commercial Planet optical satellite imaging, the trained RF model effectively delineated flooding areas of daily repeat-pass UAVSAR scenes on different flight tracks. Flooding maps derived here captured the rapidly changing flooding areas. Specifically, we revealed that the floods receded faster near the upper reaches of the Neuse River, Cape Fear River, and Lumber River, while along the flat terrain close to the lower reaches of the Cape Fear River, the floods flowed downstream, resulting in the flooded extents expanding more than 1,000 meters within four days (i.e., Sep. 18th - 22nd, 2018). Flood maps produced here have the great potential for assisting hydrodynamic modeling to provide support for planning flood-resilient communities.