Collaborative Research: NSFGEO-NERC: Understanding surface-to-bed meltwater pathways across the Greenland Ice Sheet using machine-learning and physics-based models

该项目由美国国家科学基金会地球科学理事会 (NSF/GEO) 和英国国家环境研究委员会 (NERC) 通过 NSF/GEO-NERC 牵头机构协议共同资助。该协议允许美国/英国的单一联合提案由其调查员拥有最大预算比例的机构提交并进行同行评审。在成功联合确定资助建议后，每个机构都会资助相应比例的预算，以支持各自国家机构的科学家。了解气候变暖导致冰盖表面的冰融化如何影响冰盖的运动非常重要。随着气候变暖，冰盖表面的冰融化形成融水池。为了对格陵兰冰盖的运动产生影响，融水必须到达并润滑冰盖底部。例如，冰盖表面的湖泊可以通过裂缝排水，并在几个小时内到达冰盖底部。为了了解这些裂缝的形成以及格陵兰冰原湖泊排水的原因，我们计划使用深度学习（一种人工智能算法）在卫星图像中查找裂缝和排水湖泊的位置。基于这个新数据集，我们将使用数学模型来了解新裂缝的形成及其对冰盖运动的影响。我们的方法包含令人兴奋的观察结果和数学模型的组合。使用人工智能检测裂缝和湖泊排水的能力为推动对冰盖规模的新认识提供了机会。更广泛的影响：该项目将支持 (1) 美英合作； （二）学生和初级科学家； （3）为北极科学界开发开源人工智能代码； (4) 建立一个关于格陵兰冰盖裂缝和湖泊排水的综合且免费的数据库； (5) 一个名为 COMPACT 的社区主导的指导计划（社区主导的冰冻圈实习生指导计划），该计划将促进美国和英国冰冻圈社区内少数族裔博士生的多导师网络。总的来说，这项研究将帮助我们了解冰盖如何应对气候变化。冰盖表面形成的融水可以渗入连接冰盖表面和冰床的冰臼和裂缝，润滑冰盖底部并影响其动态。格陵兰冰盖上普遍存在地表到床层融水通道。然而，我们目前缺乏了解地表到地层融水路径形成所需的全大陆冰臼、裂缝和排水湖地图。利用深度学习技术自动检测和绘制冰盖表面特征可以极大地增强冰川学界分析高分辨率卫星图像的能力，从而带来新的发现。通过利用深度神经网络，该项目旨在生成全大陆范围的表面特征数据库，这些数据库可用于对冰盖条件进行机械建模，从而创建新的地表到冰床路径及其对冰盖动力学的影响。将特征检测扩大到冰盖规模的能力可以丰富遥感和建模社区。该项目将促进美英合作，涉及初级首席研究员、博士后研究人员和研究生。该项目旨在为更广泛的冰川学界开发开源深度学习代码、遥感算法和冰下水文模型代码。由此产生的格陵兰冰盖地表到冰床路径位置以及冰上湖泊排水日期和位置的数据库将开源。主要研究人员还将合作在美国和英国冰冻圈社区内建立一个社区主导的指导计划，以促进留住代表性不足的博士生和初级教师/科学家。这项研究的社会效益将是更好地了解冰盖过程，并提高预测气候变暖时冰盖变化的能力。了解格陵兰冰盖不断变化的水文仍然是一个重要的话题，因为通过水力压裂融水排出可能在冰盖对不断扩大的消融区域的动态响应中发挥未知但可能重要的作用。该奖项反映了 NSF 的法定使命，并通过使用基金会的智力价值和更广泛的影响审查标准进行评估，被认为值得支持。

This is a project jointly funded by the National Science Foundation’s Directorate for Geosciences (NSF/GEO) and the National Environment Research Council (NERC) of the United Kingdom (UK) via the NSF/GEO-NERC Lead Agency Agreement. This Agreement allows a single joint US/UK proposal to be submitted and peer-reviewed by the Agency whose investigator has the largest proportion of the budget. Upon successful joint determination of an award recommendation, each Agency funds the proportion of the budget that supports scientists at institutions in their respective countries.It is important to understand how melting ice on the surface of ice sheets, caused by a warming climate, affects the movement of the ice sheets. As the climate warms, melting ice at the surface of ice sheets form ponds of meltwater. In order to have an impact on the movement of the Greenland Ice Sheet, the meltwater must reach and lubricate the bottom of the ice sheet. For example, lakes on the surface of the ice sheet can drain through cracks and reach the bottom of the ice sheet within a few hours. To understand the formation of these cracks and the cause of draining lakes on the Greenland Ice Sheet, we plan to use deep learning, an artificial intelligence algorithm, to find the locations of cracks and draining lakes in satellite imagery. Based on this new dataset, we will use mathematical models to understand the formation of new cracks and their impact on the movement of the ice sheet. Our approach contains an exciting mix of observations and mathematical models. The ability to use artificial intelligence to detect cracks and draining lakes offers opportunities to drive new understandings at the ice-sheet scale. Broader Impacts: This project will support (1) a US-UK collaboration; (2) students and junior scientists; (3) the development of open-source artificial intelligence codes for the Arctic sciences community; (4) the production of a comprehensive and freely available database of the Greenland Ice Sheet cracks and draining lakes; and (5) a community-led mentoring program called COMPACT (COmmunity-led Mentoring Program for Advancing Cryosphere Trainees), which will facilitate multi-mentor networks within the US and UK cryospheric communities for minority doctoral students. Overall, this research will help us understand how ice sheets respond to a changing climate.Meltwater that forms on the surface of the ice sheet can seep through moulins and fractures that connect the surface to the bed, lubricating the bottom of the ice sheet and influencing its dynamics. Surface-to-bed meltwater pathways are prevalent across the Greenland Ice Sheet. However, we currently lack the continent-wide maps of moulins, crevasses, and draining lakes needed to understand the formation of surface-to-bed meltwater pathways. Utilizing deep learning techniques for automated detection and mapping of ice sheet surface features can greatly enhance the glaciology community's capacity to analyze high-resolution satellite imagery, leading to new discoveries. By harnessing deep neural networks, this project aims to generate continent-wide databases of surface features that can be used to mechanistically model the ice sheet conditions that create new surface-to-bed pathways and their impact on ice-sheet dynamics. The ability to scale up feature detection to the ice sheet scale can enrich both remote sensing and modeling communities. This project will foster a US-UK collaboration involving junior principal investigators, postdoctoral researchers, and graduate students. The project aims to develop open-source deep learning code, remote-sensing algorithms, and subglacial hydrology model code for the broader glaciological community. The resulting database of Greenland Ice Sheet surface-to-bed pathway locations and supraglacial lake drainage dates and locations will be made open source. The principal investigators will also collaborate to establish a community-led mentoring program within the US and UK cryospheric community to promote the retention of underrepresented doctoral students and junior faculty/scientists. The societal benefit of this research will be a better understanding of ice sheet processes and an improved ability to predict ice sheet change in a warming climate. Understanding the evolving hydrology of the Greenland Ice Sheet remains an important topic given the unknown, but potentially significant, role that meltwater drainage via hydro-fracture may play in the ice-sheet’s dynamic response to an expanding ablation area.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.