Author(s): P. Gopi; R. Manjula; K. Chandrasekar

Linked Author(s):

Keywords: Flood Detection; Synthetic Aperture Radar; Digital Elevation Model; Vertical Transmitter and Horizontal Receiver; Vertical Transmitter and Vertical Receiver; Convolutional Neural Network

Abstract: Due to the unpredictable nature of flood occurrences, prompt and accurate flood detection was critical for disaster management. Recent advancements in deep learning offered immense potential for flood detection, yet the scarcity of high-quality flood datasets posed a challenge. The Godavari River Basin was selected for flood modeling due to its diverse communities often affected by floods, making it essential to study the socio-economic impacts and develop strategies to protect these populations. The 2019 flood incident in the Godavari River Basin was categorized into training, testing, and application sets. Assessment of various convolutional neural network models using these datasets had demonstrated significantly higher efficiency. The study was examined the impact of VH polarization, VV polarization, and auxiliary DEM involvement on flood detection. VH polarization emerged as more favorable for flood detection, while the addition of DEM showed limited influence within the Godavari River Basin. Leveraging the strongly labeled datasets, a convolutional neural network was applied to real-time flood detection for the 2019 flood event. The results from these experiments aligned with earlier findings from the robust datasets, affirming the efficacy of the approach. The higher accuracy of F1 Score 0.91 and 0.92 were achieved particularly on 06th August 2019 and 18th August 2019 post flood events respectively.

DOI: https://doi.org/10.3850/iahr-hic2483430201-282

Year: 2024