Remote sensing as a tool to detect sudden glacier detachment events

冰川剥离事件，即大段冰突然从冰川上剥离，有可能在山区环境中造成灾难性破坏。近年来，最值得注意的是，2021年2月印度北阿坎德邦的一次大规模冰岩崩塌造成200多人死亡(Shugar等人，2020年)，2022年7月意大利Marmolada冰川的一次支队造成11人死亡。此外，在世界各地的山区冰川中也观察到了这种分离，可能比最初认为的更频繁的危险(Kääb等人，2021年)。然而，考虑到它们的罕见发生，人们对导致突然分离的过程知之甚少，如果检测到这种过程，可以用来警告即将发生的事件。由于气候变化加速了山脉冰冻圈的衰落，如何使用观测和建模方法有力地跟踪这些变化是一个挑战。分析导致冰川崩塌的事件的研究表明，存在一些可观察到的前兆，例如扩大的裂缝(Leinss等人，2021年)、来自冰川运动的地震信号的变化(Roux等人，2010年)和波浪状的速度加速(Kääb等人，2018年)。这些前兆中的许多在崩塌前几天甚至几周展出。将这些信号从正常的冰川学过程中分离出来，是将它们纳入可能即将崩溃的预警系统的关键部分。新一代山地冰冻圈遥感技术(Taylor等人，2021年)可能能够探测和预测即使在世界最偏远地区的大型冰川分离。随着剥离事件数据集的扩大，机器学习算法可以被训练成区分指示崩塌的独特表面信号和指示正常冰川行为的独特表面信号(正如在冰盖上应用的那样；赵等人，2022年)。此外，低成本的现场传感器可以在高危地区自由部署，以收集冰川崩塌前地面变化的重要数据。该项目的目的是通过遥感限制冰川崩塌事件的可能前兆。将使用广泛的遥感技术，包括在谷歌地球引擎等云计算环境中利用全球卫星档案进行全球范围的分析。极高分辨率的光学和合成孔径雷达图像也将被用来自动检测预示即将崩塌的关键表面特征。机器学习也可用于在更大的空间尺度上自动检测前兆。还将有机会与区域利益攸关方和政策制定者合作，在高风险地区部署低成本延时相机和地震仪等实地工具。最后的实地选址(S)将根据对坍塌风险的评估、监督小组内部与实地合作伙伴的事先联系以及与成功候选人的讨论而选定。潜在的选址包括阿尔卑斯山、喜马拉雅和秘鲁安第斯山脉。目标：1.根据山脉冰川对未来崩塌的脆弱性制作全球范围的冰川风险地图；2.使用非常高分辨率的遥感和机器学习技术，根据历史事件检测突然崩塌的前兆；3.在脆弱地区部署低成本的实地传感器，以检测可能预示突然崩塌的冰川地面变化。

Glacier detachment events, where a large section of ice suddenly detaches from a glacier, have the potential to cause catastrophic damage in mountainous environments. Most notably in recent years, a large ice-rock avalanche in Uttarakhand, India in February 2021 killed more than 200 people (Shugar et al., 2020) and a detachment from the Marmolada glacier in Italy in July 2022 killed 11 people. Further such detachments have been observed in mountain glaciers worldwide and may be a more frequent hazard than originally thought (Kääb et al., 2021). However, given the rarity of their occurrence, little is known about the processes that lead up to sudden detachment, which if detected, could be used to warn of an imminent event. As climate change accelerates the decline of the mountain cryosphere, a challenge exists in being able to robustly track these changes using observation and modelling methods. Research which analyses the events leading up to glacier collapses have shown that there are a number of observable precursors, such as expanding crevasses (Leinss et al., 2021), changes in the seismic signals from glacier movement (Roux et al., 2010), and surge-like acceleration in velocity (Kääb et al., 2018). Many of these precursors are exhibited days, or even weeks, prior to collapse. Disentangling these signals from normal glaciological processes is a key part of incorporating them within a warning system that imminent collapse is likely.A new generation of remote sensing techniques for the mountain cryosphere (Taylor et al., 2021) may be able to detect, and predict, large glacier detachments even in the world's remotest region. With an expanded dataset of detachment events, machine learning algorithms can be trained to separate the unique surface signals which indicate collapse and those that indicate normal glacier behaviour (as is being applied over ice sheets; Zhao et al., 2022). In addition, low-cost field sensors can be deployed liberally across high risk regions to gather vital data into on-the-ground changes in a glacier prior to collapse.The aim of this project is to constrain the likely precursors of glacier collapse events with remote sensing. A broad range of remote sensing techniques will be used, including harnessing global satellite archives in cloud computing environments such as Google Earth Engine to conduct global-scale analyses. Very high resolution optical and SAR images will also be used to automatically detect key surface-based features which indicate imminent collapse. Machine learning may also be used to automate the detection of precursors on a wider spatial scale. There will also be an opportunity to deploy field-based tools, such as low-cost time-lapse cameras and seismometers, in high risk areas, in partnership with regional stakeholders and policymakers. The final field site(s) will be chosen based on the assessment of risk of collapse, prior connections within the supervisory team to on-the-ground partners, and in discussion with the successful candidate. Potential sites include the Alps, Himalaya, and Peruvian Andes. Objectives:1. Produce a global-scale risk map of mountain glaciers based on their vulnerability towards future collapse;2. Use very high resolution remote sensing and machine learning techniques to detect the precursors to sudden collapse based on historical events;3. Deploy low-cost field-based sensors in vulnerable areas to detect on-the-ground changes in glaciers which may indicate sudden collapse."