Scholarly record

AN EXPLORATION OF POLARIMETRIC AND COHERENCE INFORMATION OF SINGLE LOOK COMPLEX DUAL-POL SENTINEL-1 SAR DATA FOR FLOOD MAPPING IN RURAL LANDSCAPE

Vu Giang Nguyen, Vu Kim, Bruno Verbist, Ben Somers

Abstract

Hydrological modeling is the conventional approach to simulate, evaluate, forecast, and react to floods. However, the development and validation of these models are more challenging where the monitoring stations are limited, unevenly distributed, and particularly scarce in many developing countries. Alternatively, satellite images allow detecting the multi-temporal flood extent over large horizontal areas with a cost advantage for hydrological calibration and validation. Many approaches have been explored to distinguish floodwater from Synthetic Aperture Radar (SAR) satellite data. However, most of the SAR-based flood studies use only intensity bands to detect the water surface. Studies using polarimetric and coherence for flood detection are few because of their complexity. The general objective of this study is to investigate and explore the potential of intensity, polarimetric, and coherence information of SAR Sentinel-1 satellite data to detect flooded areas and enhance the reliability of flood mapping in rural landscapes. This study tested 15 approaches as the combination of intensity bands with textures, polarimetry, and coherence images derived from dual-polarization Single Look Complex (SLC) Sentinel-1 C-band data. Random Forest (RF) model was used to classify these SAR-based image compositions to detect the flood areas. The model Overall accuracies (OA), precision (Producer accuracies - PA), and sensitivity (User accuracies - UA) were used to evaluate the RF classification and contribution of intake SAR-based image components. Results showed that the modest model accuracies were achieved (UA = 75%, PA = 86.9%, OA = 94.3%) when flood image intensity bands were combined with Alpha, Anisotropy, and the co-event coherence images. It was concluded that the use of polarimetric and coherence information can significantly improve the accuracy of flood maps derived from Sentinel-1 SAR data.

Publication Impact Profile

• Captures
• Mendeley - Readers: 4

PLUMX see details

DOI resolver: doi.org/10.5593/sgem2021v/3.2/s12.07

Publication details

Back to publication list