Using SAR imagery to extract flash flood sediment deposition area in the northern Loess Plateau

Abstract

The lack of hydrology data brings challenges to the accurate simulation of the inundation range of flash floods using hydrodynamic models. In addition, the water bodies of flash floods often disappear quickly, which makes it difficult for remote sensing satellites to extract flash flood inundated areas without standing water. There is currently no research using flood sediment deposition areas to characterize flash floods in areas without water accumulation. To address this research gap, we developed a method based on Google Earth Engine (GEE) that uses Sentinel-1 SAR imagery to automatically extract flash flood sediment deposition areas. This method, termed Power Operation of Different Polarization Combination (PODPC), differs from other feature extraction methods in that it utilizes not only the original Vertical-Vertical (VV) and Vertical-Horizontal (VH) polarizations and their simple combinations but also employs power operations and threshold techniques to extrac t sediment deposition areas. We explored the advanced features of PODPC in terms of extraction accuracy, threshold segmentation effect, and sediment deposition area distribution. The results demonstrate that the histograms generated by PODPC exhibit clearer bimodal characteristics, facilitating better threshold selection. VV3, (VV + VH)6, and (VV × VH)3 produced fewer false deposition areas, achieving an overall accuracy of 92 % ∼ 93 %. Compared to commonly used methods such as VV, VH, and VV + VH, PODPC increased the overall accuracy by an average of 23 %. (VV-VH)n and (VV/VH)n are are considered unsuitable for the extraction of sediment deposition areas. The extracted sediment deposition area can serve as indicator of flash flood severity and as validation and input data for flash flood models, enhancing the scientific assessment of flash flood disasters.