Integrating remote sensing derived indices and machine learning algorithms for precise extraction of small surface water bodies in the lower Thoubal river watershed, India

Abstract

Surface water resources have witnessed various challenges namely shortage of water availability, deterioration in water quality and reduction of water bodies due to climate change and anthropogenic activities. Thus, extracting surface water bodies accurately and examining trend in their areal extent assumes greater significance. This study makes a novel attempt to utilize remote sensing derived indices and machine learning algorithms to extract and examine spatio-temporal changes in surface water bodies in the lower Thoubal river watershed. Spectral water indices were derived using Landsat data for the years 1989, 1997, 2005, 2013 and 2020 and were fed into machine learning algorithms for precise extraction of surface water bodies. The random forest model exhibited superior performance in terms of surface water extraction. The Mann-Kendall test was then used to examine the trend of surface water bodies during 1989–2020. The analysis revealed a significant declining trend in water extent over the past three decades. Future simulations using a Markov-Chain model also projected a decreasing trend in water bodies for the years 2030 and 2040 in both wet and dry months. The morphological pattern of the surface water bodies was assessed using morphological spatial pattern analysis (MSPA). It has also shown decrease in area under water bodies. These changes in the dynamics of surface water bodies may be attributed to human encroachment on wetlands for settlement and farming purposes, damming of streamflow, garbage dumping and climate change. Thus, a comprehensive approach encompassing water conservation, demand management, supply augmentation, legislation and long-term planning is essential for fostering sustainable water resource management. The findings of the study may serve as a valuable record for decision-making process and policy responses. We argue that the methodology applied in the study may help in protecting and conserving neglected small surface water bodies in other geographical regions.