The extraction of a glacial lake is the process of monitoring, mapping, and detecting water bodies from glaciated mountains using satellite images(Mitkari, Arora, & Tiwari, 2017). This process of glacial lakes monitoring is a crucial task that plays an important role in reducing natural disasters, global warming, and various human threats on a regional and global scale (Verpoorter, Kutser, & Tranvik, 2012). In the last three decades, glacial lakes are rapidly growing worldwide due to climate change and glacier retreats (Glacial Lakes Have Grown Rapidly Worldwide- Satellite Images _ Earth, n.d.). According to Shugar et al. (Shugar et al., 2020), the glacier lakes volume increased by around 48% to 156.5km3 between 1990 and 2018 globally. So, the increasing glacial lakes are not only the most vital climate indicators and water resources, but they also act as a cause of many glacial hazards such as Glacial Lake Outburst Floods GLOFs (Yao, Liu, Han, Sun, & Zhao, 2018).
Similarly, at a regional scale, Hindukush Karakorum and Himalayan (HKH) mountains are mainly sensitive to climate change therefore, the number of glaciers and glacial lakes is constantly changing with time and expanding or born new glacial lakes rapidly (Ashraf, Naz, & Iqbal, 2017). However, in the late 1990s, the big debate "Karakorum Anomaly" arises among scientific societies and researchers that many glaciers of the central Karakoram region are stable or advances which is peculiar behavior than rest of world glaciers(Hewitt, 2005). Later, (Bazai, Cui, Carling, & Wang, 2020)(Ashraf et al., 2017) found that glacier and glacial lakes in various river basins indicated different patterns (may stable or may not) depending on geographic location in the HKH region. Like in the Hunza river basin of Karakorum range, there were five GLOF events that occurred during 2007 and 2008, which harshly affected the nearby societies and modeled a threat for the future(Ashraf, Naz, & Roohi, 2012). Thus, the monitoring of glacial lakes on a regional and local scale is therefore an enormous function. Because it plays a significant role in natural hazard prevention, global warming, and varied human therapies (Verpoorter et al., 2012). Other than that, glacier lake mapping will mitigate risks to GLOFs that are very risky to downstream including infrastructure and vegetation. (Yasmeen, 2013).
All the above-mentioned (Mitkari et al., 2017)(Yasmeen, 2013) studies indicate that the significance of mapping glacial lakes from high altitude mountain is more important and need of time. Accordingly, the research (Elsahabi, Negm, & Ali, 2016) was performed to extract glacial lakes from steeped mountains using multiple forms of data and techniques. Such as the study(Chen, Zhang, Tian, & Li, 2017) performed to extract glacial lakes annually in the entire Tibet Plateau region and were used Landsat-8 images and non-local active contour techniques. The authors(Chen et al., 2017) noticed that the method introduced is theoretically applicable to large-scale initiatives related to the mapping of water bodies. On the other side, Landsat 8 has a cumulative research area of 35.2 observations. Out of overall measurements, 21.2 measurements are decent and reliable regardless of cloud cover. Likewise,(Wangchuk & Bolch, 2020) proposed mapping glacial lakes and frozen glacial surfaces. Their method used comparatively multi-source data such as SAR and optical. They found that different factors challenge the mapping of water bodies in the alpine region. These factors include the shadow of mountains, cloud cover in optical data, and small glacial lakes. Thus, most of the studies used visual remote sensing data (Landsat, sentinel-2A/B, Advanced Spaceborne Thermal Emission and Reflection Radiometer (ASTER), and Moderate Resolution Imaging Spectroradiometer (MODIS), etc.) to monitor the spatial and temporal changes of glacial lakes (Zhang, Chen, Tian, Liang, & Yang, 2019a). This optical data is only adequate during the absence of clouds. However, clouds are mainly appearing in high-altitude mountains, especially in the Karakoram Mountains of northern areas. In this case, the use of optical data is more challenging as some information related to an area of interest might be missing just because of cloud cover. Hence, to address optical data, we need such a type of Satellite data that penetrates even through clouds or any kind of weather condition in mountainous areas. At the same time, that is "Microwave sentinel-1 SAR GRD" data. Moreover, to avoid cloud cover in optical data, we have to take more additional processing steps. This type of processing may be time-consuming processing (Wangchuk & Bolch, 2020). Therefore, our study uses or contributes Sentinal-1 GRD data in an alternative way as it is independent of all-weather conditions. Apart from that, our study also helps to reduce the negative impact on livelihood by giving future precautionary measurements of GLOFs events.
Additionally, machine learning algorithms mainly offer an effective and efficient classification of remote sensing data. It can handle high-dimensional datasets and map classes with complex characteristics (Maxwell, Warner, & Fang, 2018). Therefore, our study comparatively analyzes some specific machine learning algorithms that can enhance the performance analysis of GRD images in more reliable way. The main objectives of this study are: 1) aimed to explore the importance and application of GRD data for extraction of glacial lakes. 2) to modify a conventional method for effective and automated analysis of SAR imageries. 3) to evaluate the efficiency and accuracy of machine learning algorithms for the extraction of water bodies in mountainous areas.
Mainly, the north part of Pakistan is composed of mountainous regions. The high mountain chain of Hindukush, Karakoram, and Himalayas (HKH) is home to several glaciers and glacial lakes in Gilgit-Baltistan. Accordingly, the GB Hunza basin is located in the extreme northern part of the Upper Indus Basin (UIB) at the junction of four high mountains. These mountains include Pamir, Tianshan, Karakoram, and Hindukush near the Pakistan-China border. The Hunza basin lies between latitude and longitude of 350 N to 360 N and 750 E to 760 E, respectively. The total area of the basin is 13,567.23 km2, as calculated in ArcMap.
Additionally, the Hunza basin glaciated area lies between 2280m to 7850m (Second Summer School on Integrated Water Resources Management 27–31, 2018). The Batura Glacier is an area of interest selected from the Gojal region of the Hunza basin. It is the fifth-longest glacier outside the polar regions as 57km its length, and spread over 285sqkm.