Extracting Geomorphological Features with Sentinel-2 Data using Machine Learning Algorithm in Google Earth Engine for Himalayan Proglacial Rivers

Keywords: Supervised classification, Google Earth Engine, Proglacial Rivers, Geomorphological Changes
Abstract Type: Paper Abstract

Abstract

The monitoring of hydro-morphological changes in proglacial rivers is crucial to expand the knowledge of unified river management. Himalayan region in South Asia is one of the extremely glaciated, vulnerable mountainous areas on globe and a source of plenty of rivers. The remoteness of proglacial zones and complexity to have access to the majority of sites for field investigation, highlights the viability of remote sensing data. The primary goal of the research is to design a semi-supervised classification model to classify the recent morphological dynamics in Himalayan proglacial river channels on Google Earth Engine Platform with smile random forest (SRF) algorithm using JavaScript API. This investigation is carried out on a number of different multispectral satellite imageries for the duration of 2000 to 2020. Multispectral satellite data acquired by high resolution (10m) Sentinel-2 images and very high-resolution WorldView-2 and QuickBird-2 images (each 2.4m) are processed to extract the ‘channel’, ‘vegetation’ and ‘sediment’ variations over time and space. For the investigation of model performance three important sites; Khola, Ganga and Nubra proglacial rivers are selected from Nepal, India and Kashmir (disputed territory) respectively. The secondary goal of this study is to compare the performance of different satellite sources using accuracy assessments. The average overall accuracy of SRF classifier for Sentinel-2, WorldView-2 and QuickBird-2 was 93.45%, 96.33% and 95.53% respectively. These results indicate the validation of reliability to free accessible high-resolution Sentinel-2 data for similar results to VHR images.