Time series analysis of automated surface water extraction and thermal pattern variation over the Betwa river, India

Sand-dominated rivers undergo a temporal change in surface water extent and in the supply of the water during the year. These rivers are mainly recharged through precipitation, which showcases their vulnerability to climate-change-induced aridification. Accurately detecting the surface water spread of the river is very significant for understanding and preserving the ecosystem dependent on the river. The majority of the passive optical satellite sensors are incongruous for continuous periodic monitoring because of the cloud cover issue. However, the microwave satellites such as Sentinel-1 can produce cloud-free images, and the imageries are available for free to the users. In this study, inundated pixel fluctuation has been analysed for Betwa river, India, over a selected stretch for the time period ranging from September 2016 to July 2020 with the aim of managing and protecting the aquatic life of the river. The river for the study stretch has been divided into five segments. Segment 1 has the highest number of inundated water pixels, and segment 2 has the lowest. Air and water temperature pattern have also been studied, and they follow the same pattern almost for the entire study period. Precipitation mostly happens from the month of July to September for all the years under consideration. The Google Earth Engine cloud processing system has been used to process and analyze the optical and microwave satellite datasets. The multi-Otsu thresholding algorithm has been applied for the automated extraction of the surface water cover. VV (Vertical transmit and Vertical receive) polarisation has shown the best result for water extraction. The validation of the results has been done by comparing it with the MNDWI (Modified Normalised Difference Water Index) generated from the Sentinel-2 multispectral images. The results indicate an accuracy of 90.4% (mean overall accuracy). In addition to the water extraction analysis, the temporal thermal pattern of the river has also been studied based on LANDSAT-8 thermal bands. The gridded GPM (Global Precipitation Measurement) datasets along with ERA 5 datasets have also been used in this study. The multi-Otsu-based local thresholding technique applied on each timestamp of the time series has proven to be highly accurate in extracting the surface water mask for the time period September 2016 to July 2020, highlighting the river water dynamics for each segment of the study stretch. (c) 2021 COSPAR. Published by Elsevier B.V. All rights reserved.