A Connected Pixels Based Approach for Surface Water Layer Extraction from EOS-04 Medium Resolution ScanSAR (MRS) Data

The extraction of water spread area of inland water bodies and their dynamics is crucial for assessing water security and managing water resources effectively. Satellite imagery, especially from Synthetic Aperture Radar (SAR) sensors, offers an efficient means to map water bodies during monsoon season in India. This study presents a region-growing algorithm for extracting water bodies from C-band SAR data acquired by the EOS-04 satellite. The methodology involves pre-processing steps such as noise removal and radiometric calibration, followed by identifying low and high (seed pixels) confidence pixels that were used for region growing. Dynamic thresholding based on local incidence angle was employed to address variations in backscatter response. The algorithm was tested in pre- and post-monsoon periods, showing promising results in classifying water bodies. Evaluation metrics including Kohen’s Kappa (0.8), user's (85%) and producer's (77%) accuracy indicate overall satisfactory performance of the algorithm with scope for further improvement.