CLOUD-FREE FLOOD INUNDATION FREQUENCY MAPPING OF MAJULI ISLAND, ASSAM USING GOOGLE EARTH ENGINE AND SENTINEL-1 SAR TIME-SERIES (2017-2025)

Abstract

Majuli Island, located within the braided channel of the Brahmaputra in upper Assam, India. The island is widely documented as the world's largest inhabited river island and one of the most flood-prone landmasses in South Asia. Repeated monsoon flooding has stripped the island of more than half its mapped area since the late nineteenth century, yet no spatially explicit or decadal-scale record of where flooding actually recurs has been produced. Cloud-free flood inundation frequency mapping for Majuli Island exercise using dual Sentinel-1 SAR Ground Range Detected (GRD) workflows executed within Google Earth Engine. One pipeline, built on VH cross-polarisation descending-orbit imagery to extracted annual monsoon-season flood extents across 2017 to 2025 through bi-temporal backscatter change detection. A second pipeline, using VV co-polarisation from 2017 to 2025, stacked nine annual binary flood masks into a pixel-level Flood Frequency Index. The results show that roughly 43.51% of the island was flooded at least once during the observation period. Persistent inundation was recorded in almost every season, with concentrations along the southwestern channel margin and in low-lying wetland basins. The annual flooded area ranged from 70 sq. km to 294 sq. km, with 2019 recording the largest extent. Validation against Sentinel-2 MNDWI reference imagery yielded a mean overall accuracy of 63.44% and a mean Kappa coefficient of 0.27. The mean spatial overlap with JRC Global Surface Water seasonal zones reached 88.57%. Over 61% of the island's agricultural land fell within the occasional-to-persistent flood frequency categories. The flood frequency dataset produced here is intended as a direct operational input to embankment prioritisation, early warning planning, and agricultural risk assessment in Majuli.