Mapping urban floods in 2023: a study using sentinel-1 SAR and social media data validation

Abstract

This study focuses on mapping urban floods in 2023 in the cities of Delhi, Chennai, Nagpur, and Greece, using Sentinel-1 Synthetic Aperture Radar (SAR) imagery and social media data for validation. The research applies rule-based classification, change detection, and supervised classification methods to identify flooded areas. By plotting histograms for pre-flood and post-flood images, a threshold of -30 dB is chosen to determine flooded pixels. Permanent water is distinguished using the Global Surface Water (GSW) dataset, with pixels having a seasonality band greater than 5 considered as permanent water. Noise reduction is accomplished through isolated pixel masking, eliminating single pixels in areas with fewer than eight connected neighbors. Historical and current flood maps from the Indian Space Research Organization (ISRO) provide the foundation for training data in supervised classification. Validation of the methods is achieved using crowdsourced data from social media platforms such as Twitter and Facebook, as well as news channels and OpenCity. The study concludes by plotting confusion matrices for each method against ground truth data, assessing the accuracy of the proposed flood mapping approaches.

Keywords: urban floods, SAR imagery, rule-based classifier, change detection, supervised classification, social media validation, confusion matrix.