Deep learning techniques for observing the impact of the global warming from satellite images of water-bodies

Global warming is a threat to modern human civilization. There are different reasons for speed up the global average temperature. The consequences are catastrophic for human existence. Seafloor rise, drought, flood, wildfire, dry riverbed are some of the consequences. This paper analyzes the changes in boundaries of different water bodies such as fresh-water lakes and glacial lakes. Over time, the area covered by a water body has been varied due to human interventions or natural causes. Here, variants of Detectron2 instance segmentation architectures have been employed to detect a water-body and compute the changes in its area from the time-lapsed images captured over 32 years, that is, 1984 to 2016. The models are validated using water-bodies images taken by the Sentinel-2 Satellite and compared based on the average precision (AP), 99.95 and 94.51 at $$ and $$ metrics, respectively. In addition, an ensemble approach has also been introduced for the efficient identification of shrinkage or expansion of water bodies.