Delineation and Monitoring of Wetlands Using Time Series Earth Observation Data and Machine Learning Algorithm: A Case Study in Upper Ganga River Stretch

Abstract

Wetlands are unique and valuable ecosystems and are the traditional zones between land and water. They are considered to be one of the most important resources on the earth. The rapid loss of wetlands during the past decades necessitates continuous monitoring of wetlands with respect to its hydrological changes, vegetation, soil etc. The Ramsar Convention of wetland conservation plays a significant role in outlining the internationally important wetlands and the efficient measures for their protection. The present study focused on the delineation and monitoring of wetlands situated in the upper Ganga River stretch, a potential Ramsar site in Uttar Pradesh using Sentinel 2A and Landsat satellite images. Periodical monitoring and change analysis of Ramsar wetland and its environment was studied at decadal scale during the period 1991 to 2019. Machine learning-based classification approach Random Forest classifier was used for extraction of surface water bodies and wetland classification. The overall accuracy and overall kappa coefficient of the classified data is nearly 80% or above in all the years of study, while kappa coefficient ranges between 0.77 and 0.92 for all study years. Further, seasonal turbidity of water was assessed temporally based on the spectral index such as Normalized Differential Turbidity Index using satellite data. The sewage discharge between Anupsahar and Narora towns, industrial activities, non-point source pollution of runoff due to fertilizers and pesticides are some of the important factors that led to increased turbidity levels. Further, intensive agricultural practices at the riverbank side and cultivation in the riverbed areas also act a major cause for wetland loss or degradation in the study area.