Assessing riverbank erosion in Bangladesh using time series of Sentinel-1 radar imagery in the Google Earth Engine

Abstract

Riverbank erosion occurs along many of the Earth's river systems, affecting riverine populations by destroying agricultural land and housing. In this study, we detected past events of riverbank erosion along the Jamuna River in Bangladesh using time series of Sentinel-1 satellite radar imagery, ground-range-detected (GRD) data with a 12 d revisit cycle, available in the Google Earth Engine (GEE). Eroded land is detected by performing a land cover classification and by detecting land cover changes from vegetated areas before the monsoon to sand or water after the monsoon. Further, settlements are detected as persistent scatterers and classified as eroded if they are located on eroded land. We found that with Sentinel-1 data, erosion locations can be determined already 1 month after the end of the monsoon and hence potentially earlier than using optical satellite images which depend on cloud-free daylight conditions. Further, we developed an interactive GEE-based online tool allowing the user to explore where riverbank erosion has destroyed land and settlements along the Jamuna in five monsoon seasons (2015–2019). The source code of our implementation is publicly available, providing the opportunity to reproduce the results, to adapt the algorithm and to transfer our results to assess riverbank erosion in other geographical settings.