Flood mapping in small-size water rivers: Analysis of spectral indexes using super-resolved Sentinel-2 images

Abstract

Floods are one of the most common and catastrophic natural events worldwide, making studies on the magnitude, severity and frequency of past events essential for risk management. On this wise, remote sensing techniques have been widely used in flooding diagnoses, where Sentinel-2 images are one of the main resources employed in surface water mapping. These studies have developed single band, spectral indexes and machine learning-based methods, which have typically been applied to large water bodies. However, one of the issues in identifying water surfaces remains their size. When water surfaces have sizes close to the spatial resolution of satellite images, they are difficult to detect and map. To improve remotely sensed images' spatial resolution, an algorithm for super-resolving imagery has been developed, giving good results, especially in areas covered by agricultural land with large uniform surfaces. Although this method has proved effective on Sentinel-2 images, it has not been tested for enhancing flood mapping. Thus, flood mapping is still considered an open research topic, as no suitable method has been found for all datasets and all conditions. Consequently, the present study has developed a methodology for flood delineation in small-sized water bodies. The method leverages the advantages of Sentinel-2 MSI data, image preprocessing techniques, thresholding algorithms, spectral indexes and an unsupervised classification method. Thus, this framework includes a) the generation of super-resolved Sentinel-2 images, b) the application of seven spectral indexes to highlight flood surfaces and evaluation of their effectiveness, c) the application of fifteen methods for flood extent mapping, including fourteen thresholding algorithms and one unsupervised classification method and, d) the evaluation and comparison of the performance of flood mapping methods. The technique was applied in the Carrión River, located in the Duero basin, province of Palencia, Spain. This river is classified as a narrow water body, which presents recurrent flood events due to its morphometric characteristics, fluvial dynamics, and land uses. The results obtained show optimal performances when highlighting flood zones by combining AWE spectral indices with methods such as those of Huang and Wang, Li and Tam, Otsu, and momentum-preserving thresholding algorithms and EM cluster classification.