Comparison Between Full Polarized L-Band SAR Data and Dual Polarized C-Band SAR Data for Inundated Vegetation Mapping in Eastern North Carolina

Flood disasters cause damage to properties, loss of crops and even death to humans and animals worldwide. Recently, flood events have become intense and more frequent due to heavy rainfall and hurricanes caused by global warming. Real or near-real time accurate floodwater extent maps help emergency management agencies and flood relief programs to direct their resources to the most affected areas. Remote sensing data can be utilized to create floodwater extent maps with less cost compared to other methods. Synthetic Aperture Radar (SAR) is a remote sensing method that is capable of collecting data during night and day, independent of sunlight for illuminations. SAR sensors acquire data using longer wavelengths than optical sensors, which makes SAR signals capable of penetrating clouds, snow, dust, and vegetation canopies. Therefore, SAR data is superior to optical data for floodwater mapping particularly in vegetated areas, despite the difficulty of visually interpreting SAR images. Many Airborne and Satellite Platforms carry various SAR sensors. Usually Satellite SAR image covers a larger area with less cost than Airborne SAR image. Utilizing SAR data from different platforms provide high temporal resolution that allows monitoring flood evolution. However, the accuracy of floodwater extent maps depends on SAR systems characteristics and environmental conditions. In this study, we examined the reliability of floodwater maps derived from Sentinel-1B, C-band SAR data compared to floodwater extent maps derived from full-polarized L-band Uninhabited Aerial Vehicle Synthetic Aperture Radar (UAVSAR) data. This study also evaluates the applicability of using dual polarizations C-band Satellite SAR data to detect floodwater under vegetation in the eastern NC state. We processed and analyzed the Satellite and UAVSAR data during flooding in the North Carolinas coastal plain areas resulted from Hurricane Florence. In response to this event, UAVSAR monitored the areas by collecting daily data over several flight lines with repeat paths, beginning on September 18 through September 23 in the year of 2018. Sentinel-1B sensed the areas on September 19, 2018. For this research, UAVSAR data and Sentinel-1B SAR data on the same date were collected and analyzed. UAVSAR data were processed using polarization decompositions method to identify land cover classes based on their scattering mechanisms. We used UAV high resolution optical imagery to delineate and label land cover classes samples to train random forest classifiers. Also, reference samples from optical data were randomly selected to validate classification results obtained from the two datasets. Overlapping area of UAVSAR data and Sentinel-1B data was analyzed for comparisons. Land cover types at the areas of overlap between the two data were examined to identify which dataset provides higher accuracy for floodwater detection with respect to land cover type. This material is based on work supported by the National Science Foundation under Grant No. 1800768.