人工智能驱动的科学研究(AI for science)正在重塑地球系统科学的范式。然而,面对呈指数级增长的多源异构数据(如遥感影像、历史档案与物理模型输出),冰冻圈科学在知识整合、复杂过程推理及跨模态信息交互方面仍面临巨大挑战。大语言模...人工智能驱动的科学研究(AI for science)正在重塑地球系统科学的范式。然而,面对呈指数级增长的多源异构数据(如遥感影像、历史档案与物理模型输出),冰冻圈科学在知识整合、复杂过程推理及跨模态信息交互方面仍面临巨大挑战。大语言模型(LLMs)凭借卓越的语义理解与推理能力,为突破上述瓶颈提供了新的技术契机。本文系统综述了LLMs在冰冻圈科学中的应用潜力,构建了从数据挖掘到知识传播的智能化框架。分析表明,LLMs不仅能显著提升文献综述与历史观测资料数字化的效率,更在辅助冰冻圈数值模型开发(代码智能体)、复杂耦合过程的归因分析以及风险沟通中展现出独特的认知价值。尽管前景广阔,但通用模型在科学领域的应用仍受限于事实幻觉、物理一致性缺失及数据语义鸿沟。针对这些核心挑战,本文提出了面向冰冻圈的专用大模型(CryoLLMs)研发路径——通过构建多模态对齐的领域知识图谱解决数据异构性问题,引入物理约束机制与检索增强生成以确保推理的可信度,并基于多智能体协同架构实现从单一任务向自动化科研工作流的跨越。本综述旨在厘清LLMs在冰冻圈科学中的技术边界与伦理框架,为构建下一代具备物理感知与逻辑推理能力的冰冻圈智能研究系统奠定理论基础。展开更多
冰川对气候快速响应的同时,冰川流域径流过程也随之发生变化,准确模拟与预测冰川径流变化过程对于区域水资源调控、冰川洪水/泥石流灾害防控至关重要。本文基于贡嘎山东坡海螺沟冰川流域的观测资料及遥感产品数据,利用SWAT与冰川能量-...冰川对气候快速响应的同时,冰川流域径流过程也随之发生变化,准确模拟与预测冰川径流变化过程对于区域水资源调控、冰川洪水/泥石流灾害防控至关重要。本文基于贡嘎山东坡海螺沟冰川流域的观测资料及遥感产品数据,利用SWAT与冰川能量-物质平衡模型相结合的方法对流域冰川物质平衡与冰川径流进行模拟,分析冰川快速消融的驱动因素与径流变化的影响。结果表明,1990—2020年间海螺沟流域冰川整体呈现快速亏损的状态(-0.05 m w.e.·a^(-1))。其中,物质平衡线升高近400 m,物质平衡线以上物质积累量呈减少的状态(-0.22 mw.e.),物质平衡线以下消融区冰川物质平衡加速亏损,受表碛覆盖影响物质平衡由末端呈先增加后减少的趋势。气温升高、固态降水补给减少和反照率降低是冰川快速消融的主要驱动因素。冰川融水与径流量存在较为一致的趋势,其中夏季冰川融水占径流量的36.4%,与SWAT模型的耦合中,相比降雨、融水和基流,冰川融水占比达57.1%,基于SSP2-4.5与SSP5-8.5情景模式,冰川径流量将表现出持续减少的趋势,进而导致对水文过程的调控作用减弱。本研究能够为贡嘎山地区气候变化下的冰川水资源与相关灾害管理提供有效的科学参考依据。展开更多
High-altitude glacier-lake systems in the eastern Pamir Plateau,Tajikistan,are highly sensitive elements of Central Asia’s cryosphere and are vital for sustaining regional water resources.The Yashilkul Lake is locate...High-altitude glacier-lake systems in the eastern Pamir Plateau,Tajikistan,are highly sensitive elements of Central Asia’s cryosphere and are vital for sustaining regional water resources.The Yashilkul Lake is located within a tectonic depression dammed by an ancient rockslide,forming a large alpine lake.This lake is currently impacted by intensified warming,glacier retreat,and poorly quantified hydrological shift.The primary objective of this study is to assess multi-decadal changes in the Yashilkul and Bulunkul lakes and their surrounding cryosphere between 1994 and 2024.The changes were analyzed using multitemporal Landsat imagery and unmanned aerial vehicle (UAV) surveys,complemented by in situ meteorological observations from the Bulunkul meteorological station spanning the period from 1990 to 2024.Glacier and lake boundaries were extracted from Landsat data,primarily by applying the normalized difference water index,supplemented by manual delineation.UAV photogrammetry characterized dam morphology and adjacent ponds,and climate trends were evaluated with the modified Mann-Kendall test.A significant warming trend of 0.096℃/a and pronounced interannual precipitation variability have driven persistent glacier retreat and lake surface area fluctuations.The Yashilkul Lake’s surface area decreased from 36.40 (±1.15) km^(2) in 2010 to 31.94 (±0.54) km^(2) in 2020 and partially rebounded to 33.99 (±0.60) km^(2) in 2024,while the Bulunkul Lake’s surface area remained nearly stable owing to limited glacial influence.Additionally,UAV surveys conducted in 2022 and 2024 revealed main features of the Yashilkul Lake:rockslidedammed origin,perched ponds along the dam body,and an artificial canal regulating its outflow.Nearby glaciers,particularly Glacier No.369,exhibited strong frontal retreat and proglacial lake expansion.The proglacial lake expanded nearly fourfold from 0.08 (±0.01)km^(2) in 2000 to 0.33 (±0.02) km^(2) in 2024,raising concerns about potential glacial lake outburst floods that could impact the Yashilkul Lake and compromise the integrity of its natural dam.The findings highlight accelerating hydrological and cryospheric transformations in the Pamir Plateau,emphasizing the need for sustained monitoring of glacier-lake systems owing to their critical implications for water security,ecological stability,and downstream hazard management.展开更多
文摘人工智能驱动的科学研究(AI for science)正在重塑地球系统科学的范式。然而,面对呈指数级增长的多源异构数据(如遥感影像、历史档案与物理模型输出),冰冻圈科学在知识整合、复杂过程推理及跨模态信息交互方面仍面临巨大挑战。大语言模型(LLMs)凭借卓越的语义理解与推理能力,为突破上述瓶颈提供了新的技术契机。本文系统综述了LLMs在冰冻圈科学中的应用潜力,构建了从数据挖掘到知识传播的智能化框架。分析表明,LLMs不仅能显著提升文献综述与历史观测资料数字化的效率,更在辅助冰冻圈数值模型开发(代码智能体)、复杂耦合过程的归因分析以及风险沟通中展现出独特的认知价值。尽管前景广阔,但通用模型在科学领域的应用仍受限于事实幻觉、物理一致性缺失及数据语义鸿沟。针对这些核心挑战,本文提出了面向冰冻圈的专用大模型(CryoLLMs)研发路径——通过构建多模态对齐的领域知识图谱解决数据异构性问题,引入物理约束机制与检索增强生成以确保推理的可信度,并基于多智能体协同架构实现从单一任务向自动化科研工作流的跨越。本综述旨在厘清LLMs在冰冻圈科学中的技术边界与伦理框架,为构建下一代具备物理感知与逻辑推理能力的冰冻圈智能研究系统奠定理论基础。
文摘冰川对气候快速响应的同时,冰川流域径流过程也随之发生变化,准确模拟与预测冰川径流变化过程对于区域水资源调控、冰川洪水/泥石流灾害防控至关重要。本文基于贡嘎山东坡海螺沟冰川流域的观测资料及遥感产品数据,利用SWAT与冰川能量-物质平衡模型相结合的方法对流域冰川物质平衡与冰川径流进行模拟,分析冰川快速消融的驱动因素与径流变化的影响。结果表明,1990—2020年间海螺沟流域冰川整体呈现快速亏损的状态(-0.05 m w.e.·a^(-1))。其中,物质平衡线升高近400 m,物质平衡线以上物质积累量呈减少的状态(-0.22 mw.e.),物质平衡线以下消融区冰川物质平衡加速亏损,受表碛覆盖影响物质平衡由末端呈先增加后减少的趋势。气温升高、固态降水补给减少和反照率降低是冰川快速消融的主要驱动因素。冰川融水与径流量存在较为一致的趋势,其中夏季冰川融水占径流量的36.4%,与SWAT模型的耦合中,相比降雨、融水和基流,冰川融水占比达57.1%,基于SSP2-4.5与SSP5-8.5情景模式,冰川径流量将表现出持续减少的趋势,进而导致对水文过程的调控作用减弱。本研究能够为贡嘎山地区气候变化下的冰川水资源与相关灾害管理提供有效的科学参考依据。
基金funded by the Key Program of National Natural Science Foundation of China (42230708,42361144887)the Tianshan Talent Project of Xinjiang Uygur Autonomous Region,China (2022TSYCLJ0056)。
文摘High-altitude glacier-lake systems in the eastern Pamir Plateau,Tajikistan,are highly sensitive elements of Central Asia’s cryosphere and are vital for sustaining regional water resources.The Yashilkul Lake is located within a tectonic depression dammed by an ancient rockslide,forming a large alpine lake.This lake is currently impacted by intensified warming,glacier retreat,and poorly quantified hydrological shift.The primary objective of this study is to assess multi-decadal changes in the Yashilkul and Bulunkul lakes and their surrounding cryosphere between 1994 and 2024.The changes were analyzed using multitemporal Landsat imagery and unmanned aerial vehicle (UAV) surveys,complemented by in situ meteorological observations from the Bulunkul meteorological station spanning the period from 1990 to 2024.Glacier and lake boundaries were extracted from Landsat data,primarily by applying the normalized difference water index,supplemented by manual delineation.UAV photogrammetry characterized dam morphology and adjacent ponds,and climate trends were evaluated with the modified Mann-Kendall test.A significant warming trend of 0.096℃/a and pronounced interannual precipitation variability have driven persistent glacier retreat and lake surface area fluctuations.The Yashilkul Lake’s surface area decreased from 36.40 (±1.15) km^(2) in 2010 to 31.94 (±0.54) km^(2) in 2020 and partially rebounded to 33.99 (±0.60) km^(2) in 2024,while the Bulunkul Lake’s surface area remained nearly stable owing to limited glacial influence.Additionally,UAV surveys conducted in 2022 and 2024 revealed main features of the Yashilkul Lake:rockslidedammed origin,perched ponds along the dam body,and an artificial canal regulating its outflow.Nearby glaciers,particularly Glacier No.369,exhibited strong frontal retreat and proglacial lake expansion.The proglacial lake expanded nearly fourfold from 0.08 (±0.01)km^(2) in 2000 to 0.33 (±0.02) km^(2) in 2024,raising concerns about potential glacial lake outburst floods that could impact the Yashilkul Lake and compromise the integrity of its natural dam.The findings highlight accelerating hydrological and cryospheric transformations in the Pamir Plateau,emphasizing the need for sustained monitoring of glacier-lake systems owing to their critical implications for water security,ecological stability,and downstream hazard management.
