基于多种机器学习及其堆叠式集成方法的月尺度北极海冰预测研究北大核心CSCD

北极海冰范围退缩已对区域和全球气候变化、北极通航性乃至地缘政治格局产生了深远影响,开展北极海冰预测具有重要意义。本文使用美国冰雪数据中心(NSIDC)发布的海冰密集度、海冰冰龄遥感资料以及ERA5再分析资料,结合前人研究和海气耦合模式结果选择预测参量方案,开展了月尺度的海冰冰情预测。比较了支持向量机(SVR)、深度森林(DF)、LightGBM (LGB)、XGBoost(XGB)和CatBoost (CAT)等5种机器学习算法和以树模型LGB、XGB和CAT作为基模型,以贝叶斯回归、岭回归、套索回归和深度森林作为元模型的4种堆叠式集成学习模型,以及深度神经网络(DNN)、卷积神经网络(CNN)、时空卷积网络(ConvLSTM)3种深度学习模型在2000年测试集上对海冰范围和密集度空间分布的预测效果。结果表明:在海冰密集度预测中,ConvLSTM表现最优,套索堆叠集成学习模型预测效果次之。集成学习模型相较于三种单一树模型在预测效果上有约1%~4%的提升。在海冰范围预测中,堆叠式集成学习模型的预测效果最好。本研究为开展机器学习海冰预测奠定了重要基础。     

基于FY-3 MWRI数据的北极海冰密集度反演研究

以ASI算法（ARTIST sea ice algorithm）为基础, 得到基于风云3C气象卫星（FY-3C）微波辐射计（MWRI）数据的纯水与纯冰系点值, 利用插值方法确定基于FY-3 MWRI数据的ASI海冰密集度计算公式, 采用大津法（Otsu算法）得到基于MWRI数据的天气滤波器阈值。以2016年1月数据为例, 对北极海冰密集度进行反演, 并与美国国家冰雪数据中心（NSIDC）以及德国不莱梅大学提供的海冰密集度产品进行对比验证。结果表明： 基于MWRI数据得到的1月平均海冰面积以及平均密集度均介于二者之间, 其中平均密集度与不莱梅产品更接近, 仅相差1.310%。与风云卫星空间分辨率为250 m的中分辨率光谱成像仪（MERSI）数据得到的结果进行对比, 发现二者的海冰外缘线基本一致, MERSI数据得到的海冰密集度以及海冰面积比MWRI数据得到的结果分别高出5.029%、 9.318%。因此, 应用该方法可有效推进MWRI数据反演北极海冰密集度, 进而监测北极海冰分布和变化。     

基于多源共享数据的北极航道海域冰情分析

随着全球气候变暖,北冰洋季节性大规模通航已经可以预见。为开辟北极航道服务,保障航道航行安全,采集航线所经海域及沿岸国家和地区基础地理信息,建立基础数据库是基础。以东北航道所经海域海冰覆盖为例,获取多国海洋信息服务网络的海冰共享数据,按照空间信息系统要求,相互校验,提高数据时效和空间覆盖频率,处理成适用于保障北极航道安全航行的信息,讨论了东北航道所经海区的海冰对航道开通时间的影响,实现航道水域海冰信息的通航期(7月1日到11月30日)逐日发布,非通航期(12月1日到次年6月30日)半月发布一次,为北极航道大规模通航积累经验和基础信息。     

长时间序列北极海冰密集度遥感数据的比较评估

基于空间分辨率和精度更高的BLM海冰密集度数据集, 评估了两种时间跨度较长、 应用广泛的25 km分辨率海冰密集度遥感数据集——NSIDC数据集和SICCI数据集。两种数据集与BLM数据集的海冰面积变化趋势相同, 但均低于基于BLM数据集得到的海冰面积, 其中基于SICCI数据集得到的海冰面积更接近BLM数据集。相比于NSIDC数据集, SICCI数据集的年、 月平均和日海冰面积偏差分别低81.88%、 80.90%、 81.44%, 且其海冰密集度平均偏差为-3.28%, 低于NSIDC数据的4.36%, 因此在进行北极地区整体海冰面积及海冰密集度分析时应选用SICCI数据集。按纬度、 海冰密集度值分情况对两种数据进行比较, 发现NSIDC数据集对开阔水域和浮冰区的区分效果较差, 其在低纬度和低密集度区域的平均偏差分别为10.11%和13.13%, 而SICCI数据集的平均偏差达到0.05%和0.44%, 是研究低纬度和中低海冰密集度区域的首选数据。与之相对, NSIDC数据集对中高纬度高海冰密集区域, 特别是近北极点区域的反映能力优于SICCI数据集, 平均偏差为1.08%, 均方根偏差为7.76%, 因此进行中高纬度高海冰密集度区域分析时首选NSIDC数据集。对两类数据集在北极东北航道上的分段评估结果发现, 低纬度海冰边缘地带或中低海冰密集度区域占比较高的航段区, SICCI数据集更接近BLM数据集, 这些航段应使用SICCI数据集进行分析; 而在中高纬度高海冰密集度区域占比较高的航段区, NSIDC数据集更加贴合, 应为首选数据集。     

1997-2005年北极东西伯利亚海海冰变化特征研究

近年来北冰洋气候系统发生了显著变化,海冰面积减小和厚度变薄是最主要的特征.利用美国国家冰中心的1997—2005年东西伯利亚海海冰密集度数据,研究了该海域海冰覆盖率的多年变化特征.结果表明:1997—2001年海冰冰情偏重,2002—2005年海冰冰情明显减轻,尤其是2005年,海冰面积覆盖率创密集度为近几年最低值.海冰融冰期长度和海冰面积覆盖率最小值与海冰面积指数的相关系数在置信水平为0.95时分别达到-0.90236和0.9098,海表负积温与海冰面积指数的相关系数达到-0.7424.近年来东西伯利亚海海面风场的偏南风有加强趋势,有利于海冰的北向运移和暖空气北上,造成海冰面积持续减小.河流径流量在9a中整体呈现增长趋势,与海冰具有很好的负相关性.     

关于矿床学创新问题的探讨

2003年1月4日至2月15日期间,在5种不同情况下对南极海冰进行了调查研究。包括:(1)基于走航观测的威德尔海至普利茨湾之间海冰分布研究;(2)基于航空拍摄的普利茨湾海冰分布研究;(3)纳拉海峡固定冰和上浮雪厚度钻孔测量以及冰心钻取;(4)中山站附近融化冰的分布研究以及(5)中山站附近海冰早期冻结过程观测研究。结果表明,威德尔海至普利茨湾之间走航观测得到的海冰全部密集度为14.4%,大部分冰(99.7%~99.8%)属于一年冰,观测到冰的厚度在15~150 cm。沿观测航线上海冰最大密集度(80%)出现在威德尔海,从59°56 S到69°22 S以及从040°41 W到076°23 E的区域分布着广阔的水域。这一结果验证了Silvia的海冰漂移理论。普利茨湾沿岸海冰受制于沿岸地形、拉斯曼丘陵以及搁浅冰山的影响,其密集度呈现较大的空间变化。钻孔测量显示,纳拉海峡固定冰平均厚度为169.5 cm。风吹雪的重分布以及日照强度差异是导致纳拉海峡固定冰厚度差异的主要因素。观测表明,中山站附近海冰早期冻结遵循Lange的海冰早期冻结过程“饼状循环”最初的两个阶段。     

基于卫星气候资料的1989-2015年南北极海冰面积变化分析

利用被动微波卫星海冰密集度气候资料,分析了1989-2015年南北极海冰面积和密集度的长期变化趋势。结果表明：研究期内,北极年平均海冰面积减少,南极海冰面积增加,变化趋势分别为-0.569×10⁶ km²·（10a）^-1和0.327×10⁶ km²·（10a）^-1,均通过了0.01水平的显著性检验,两极海冰面积变化趋势表现出明显的"非对称性"。两极总海冰面积出现了下降,变化趋势为-0.242×10⁶ km²·（10a）^-1。年海冰密集度在北极地区普遍减少,而在南极地区的变化趋势存在显著的空间差异,威德尔海、罗斯海北部海冰密集度增加,趋势超过了10%·（10a）^-1,别林斯高晋海、阿蒙森海的海冰密集度出现下降。北极各月海冰面积的变化趋势存在明显的季节差异,7-10月海冰面积减少明显,其中9月减少最显著,趋势为-0.955×10⁶ km²·（10a）^-1。南北极海冰冻结和融化的时间不完全对应,北极融化与冻结时间基本平衡,南极海冰冻结时间明显长于融化时间。南极年内海冰面积的变化幅度大于北极,呈现显著的季节性特征。北极极小海冰面积的变化趋势最显著,达到了-0.636×10⁶ km²·（10a）^-1。南极极大海冰面积出现的时间后移明显,趋势为0.733候·（10a）^-1;极小海冰面积出现的时间非常稳定,没有明显的变化趋势。     

累积海冰密集度及其在认识北极海冰快速变化的作用

为定量分析北冰洋海冰密集度年际差异,提出并采用累积海冰密集度(ASIC)概念。利用SSMR/SSMI的分辨率为25 km的海冰密集度数据,分别研究了1979—2011年北极海冰在融冰期(4~9月)和结冰期(10月至翌年3月)的变化过程以及2个冰期内ASIC的区域差异。研究发现,在1979—1989年、1989—1999年和1999—2009年期间,融冰期海冰发生明显变化的范围都远远大于结冰期海冰发生明显变化的范围。1998—2010年,融冰期内发生加速融化的海区并没有都出现结冰期冰量减小的现象。在此期间融冰期ASIC减小,结冰期ASIC也减小的海域仅集中在楚克奇海、新地岛北部海域以及格陵兰岛东西海岸。融冰期ASIC减小,而结冰期ASIC无明显变化的海域包括波弗特海、东西伯利亚海、拉普捷夫海和喀拉海。这些区域与局地陆地径流侵入的海域重合。研究发现,在这些区域,融冰期ASIC减少是陆地径流增大加速海冰融化引起的。在结冰期,陆地径流加速海水结冰的作用消除融冰期海水吸收大量太阳辐射能后发生推迟结冰的现象,使得ASIC无明显变化。融冰期ASIC减小,而结冰期ASIC增大的区域只有白令海。研究结果证明累积海冰密集度能够去除海冰高频变化而只表现低频变化,能够描述海冰的年际变化特征。同时由于海冰变化与海洋中其他物理参数存在显著关系,变T的ASIC可以更加方便地描述次表层叶绿素最大值层深度的变化。     

白令海冰间湖的数值模拟及影响模拟准确度的关键因素

冬季在北白令海陆架区域频繁地出现潜热冰间湖,对当地的生态系统和北极盐跃层贡献很大.将CICE海冰模式应用到该区域,采用高分辨率(6.37 km)网格,模拟2002年11月至2003年4月的海冰变化过程,模拟的海冰总面积和海冰密集度与AMSR-E/Aqua卫星遥感结果吻合很好,其中两者日平均海冰总面积在模拟期间的相关系数达到0.97.模拟结果表明,东北风将海冰向南输运在东西走向的海岸南部形成冰间湖,反映了潜热冰间湖形成和演化的动力过程.对卫星观测数据,将海冰密集度<75%作为冰间湖的判据;而对数值模拟结果,确定海冰密集度<70%为冰间湖的判据.据此讨论白令海4个区域的冰间湖形成过程,与卫星数据进行比较,大部分冰间湖得到很好的模拟.深入讨论了影响冰间湖模拟准确度的主要因素,认为选用恰当的阈值、提高气象强迫场的空间和时间分辨率有助于提高模拟效果.对部分海域的冰间湖模拟效果不佳,需要发展冰海耦合模式才能最终解决.     

格陵兰海海冰外缘线变化特征分析

格陵兰海作为北冰洋的边缘海之一,容纳了北极输出的海冰,其海冰外缘线的变化既受北极海冰输出量的影响,也受局地海冰融化和冻结过程的影响。利用2003年1月到2011年6月AMSR-E卫星亮温数据反演的海冰密集度产品,对格陵兰海海冰外缘线的变化特征进行了分析。结果表明,格陵兰海海冰外缘线不仅存在一年的变化周期,还存在比较显著的半年变化周期,与海冰在春秋两季向岸收缩有关。格陵兰海冬季的海冰外缘线极大值呈逐年下降的趋势,体现了北极增暖导致的冬季海冰范围减小;而夏季海冰外缘线离岸距离的极小值呈上升趋势,表明夏季来自北冰洋的海冰输出量增大。2003—2004年是格陵兰海夏季海冰融化最严重的2年。2007年北冰洋夏季海冰覆盖范围达到历史最小;而格陵兰海夏季的最小海冰范围最大,表明2007年北冰洋海冰的输出量大于其他年份。此外,夏季格陵兰岛冰雪融化形成的地表径流对海冰外缘线有一定的影响。对海冰外缘线影响最大的不是格陵兰海的局地风场,而是弗拉姆海峡(Fram Strait)区域的经向风,它直接驱动了北冰洋海冰向格陵兰海的输运,进而对格陵兰海海冰外缘线的分布产生滞后的影响。     

基于Landsat-8影像的北极地区入海冰川流速监测

Landsat-8 OLI因其空间分辨率较高、重复周期适中、高辐射分辨率、高图像获取率（图像质量）的特点,在北极地区大范围冰川流速监测研究中有较大优势。利用2017/2018年格陵兰岛、斯瓦尔巴群岛、北地群岛、法兰士约瑟夫地群岛、德文岛5处北极区域的Landsat-8全色波段数据,采用特征追踪方法提取入海冰川消融期流速。结合MEaSUREs冰川流速数据,分析了198条北极地区入海冰川流速的空间分布特征及其影响因素,同时探究了格陵兰岛Kangerlussuaq冰川流速随时间变化特征。结果表明：与北极其他区域相比,格陵兰岛前缘流速在5~10 m·d^-1及10~20 m·d^-1的入海冰川在数量上最多,最大流速达到了31.62 m·d^-1。而格陵兰岛内部的冰川流速存在差异,北海岸入海冰川平均流速最慢（1.99 m·d^-1）,东海岸平均流速（6.13 m·d^-1）大于西海岸（4.14 m·d^-1）。这种流速空间分布差异可能由冰川规模、冰床地势、海流作用、冰盖消融情况等多种因素共同导致。2018年3—10月期间,Kangerlussuaq冰川前缘流速为21.02~22.87 m·d^-1,整体流速为10.02~11.39 m·d^-1。冰川流速在6—7月和9—10月出现峰值,在8—9月出现低谷,主要缘于冰川融水导致的运动加速和冰川物质平衡变化。     

A Study of Monitoring,Simulation and Climate Impact of Greenland Ice Sheet

Greenland Ice Sheet is one of the two largest ice sheets on the planet. Under the background of climate warming, the melting of the Greenland ice sheet and its contribution to sea level rise has become an international hot issue. The whole melting of the Greenland ice sheet can cause the global sea level to rise by about 7.3 meters. However, the dynamic mechanism that affects the mass balance of ice sheet is still unclear and is the greatest uncertainty source for predicting the rise in sea level in the future. The National Key Research and Development Program of China “A Study of the Monitoring, Simulation and Climate Impact of Greenland Ice Sheet” conducts monitoring and simulation studies on the key processes of instability of the “ice sheet-outlet glacier-sea ice” system, and establishes a satellite-airborne-ground integrated observation system, supporting the numerical simulation and impact research of the ice sheet and its surrounding sea ice, laying the foundation for long-term monitoring and international cooperation in Greenland. This program will work to reduce the uncertainty of sea level change projections by improving the ice sheet dynamic model forced by the ice core records, reveal the driving mechanism of sea ice changes around the ice sheet, focusing on the Northwest Passage, evaluate and forecast the navigation window period. The results of the project will deepen the understanding of the changes and impacts of the Arctic cryosphere, serve the safe navigation and operation of the Northwest Passage, and provide scientific support for the comprehensive risk prevention of coastal zones in China.     

Numerical reconstructions of the Eurasian Ice Sheet and climate during the Late Weichselian

Geological investigations undertaken through the Quaternary Environments of the Eurasian North programme established ice-sheet limits for the Eurasian Arctic at the Last Glacial Maximum (LGM), sedimentary records of palaeo-ice streams and uplift information relating to ice-sheet configuration and the pattern of deglaciation. Ice-sheet numerical modelling was used to reconstruct a history of the Eurasian Ice Sheet compatible with these geological datasets. The result was a quantitative assessment of the time-dependent behaviour of the ice sheet, its mass balance and climate, and predictions of glaciological products including sediments, icebergs and meltwater. At the LGM, ice cover was continuous from Scandinavia to the Arctic Ocean margin of the Barents Sea to the north, and the Kara Sea to the east. In the west, along the continental margin between the Norwegian Channel and Svalbard, the ice sheet was characterised by fast flowing ice streams occupying bathymetric troughs, which fed large volumes of sediment to the continental margin that were deposited as a series of trough mouth fans. Ice streams may also have been present in bathymetric troughs to the north between Svalbard and Franz Josef Land. Further east, however, the ice sheet was thinner. Across the Kara Sea, the ice thickness was predicted to be less than 300 m, while on Severnaya Zemlya the ice cover may have been thinner at the LGM than at present. It is likely that the Taymyr Peninsula was mainly free of ice at the LGM. In the south, the ice margin was located close to the shoreline of the Russian mainland. The climate associated with this ice sheet is maritime to the west and, in stark contrast, desert-like in the east. Atmospheric General Circulation Modelling has revealed that such a contrast is possible under relatively warm north Atlantic conditions because a circulation system develops across the Kara Sea, isolating it from the moisture-laden westerlies, which are diverted to the south. Ice-sheet decay began through enhanced iceberg calving in the deepest regions of the Barents Sea, which caused a significant ice embayment within the Bear Island Trough. By about 12,000 years ago, further iceberg calving reduced ice extent to the northern archipelagos and their surrounding shallow seas. Ice decay was complete by about 10,000 years ago.     

Late Weichselian Glaciation of the Russian High Arctic

A numerical ice-sheet model was used to reconstruct the Late Weichselian glaciation of the Eurasian High Arctic, between Franz Josef Land and Severnaya Zemlya. An ice sheet was developed over the entire Eurasian High Arctic so that ice flow from the central Barents and Kara seas toward the northern Russian Arctic could be accounted for. An inverse approach to modeling was utilized, where ice-sheet results were forced to be compatible with geological information indicating ice-free conditions over the Taymyr Peninsula during the Late Weichselian. The model indicates complete glaciation of the Barents and Kara seas and predicts a “maximum-sized” ice sheet for the Late Weichselian Russian High Arctic. In this scenario, full-glacial conditions are characterized by a 1500-m-thick ice mass over the Barents Sea, from which ice flowed to the north and west within several bathymetric troughs as large ice streams. In contrast to this reconstruction, a “minimum” model of glaciation involves restricted glaciation in the Kara Sea, where the ice thickness is only 300 m in the south and which is free of ice in the north across Severnaya Zemlya. Our maximum reconstruction is compatible with geological information that indicates complete glaciation of the Barents Sea. However, geological data from Severnaya Zemlya suggest our minimum model is more relevant further east. This, in turn, implies a strong paleoclimatic gradient to colder and drier conditions eastward across the Eurasian Arctic during the Late Weichselian.     

DEFINITING AND ITS GEOLOGIC MEANING OF SOUTH-NORTH TREND FAULTED STRUCTURE BELT IN QIANGTANG BASIN, NORTH PART OF TIBET

There were more expounding to north—west (west) trend fault and north\|east trend fault within Qiangtang Basin, North Part of Tibet, in the past literature. With increasing of geophysical exploration data, nearly east\|west trend structure began to be taken note to. Since the year of 1995, by a synthetic study to geophysical and geological data, that south\|north trend faulted structures are well developed. These structures should be paid much more attention to, because they have important theoretical meaning and practical significance.1 Spreading of south\|north faulted structure belt According to different geological and geophysical data, the six larger scale nearly south\|north faulted structure belt could be distinguished within the scope of east longitude 84°～96° and near Qiangtang Basin. The actual location of the six belts are nearly located in the west of the six meridian of east longitude 85°,87°,89°,91°,93°,95° or located near these meridian. The six south\|north faulted structure belts spread in the same interval with near 2° longitude interval. The more clear and much more significance of south\|north trend faulted structure belts are the two S—N trend faulted structure belts of east longitude 87° and 89°. There are S—N trend faulted structure belts in the west of east longitude 83°,81°, or near the longitudes. The structure belts spreading features,manifestation,geological function and its importance, and inter texture and structure are not exactly so same. The structure belts all different degree caused different region of geological structure or gravity field and magnetic field. There is different scale near S—N trend faulted structure belt between the belts.     

Mid-Pleistocene ice drainage pattern in the Norwegian Channel imaged by 3D seismic

Several investigations have shown that a huge Late Weichselian ice stream flowed along the Norwegian Channel, and deposited thick debris flow deposits at the North Sea Fan. The development of the channel is probably mainly a result of several cycles of ice stream activity during the Quaternary. A merged 3D seismic image shows a lineated relief pattern interpreted as a uniquely well-preserved footprint of a moving ice sheet. This deep Quaternary horizon corresponds to a slightly irregular reflector on top of a parallel-layered seismic sequence. Seismic tie to the Troll core 8903 south of the study area shows that the sequence comprises Early Middle Pleistocene marine sediments. The pattern of lineations, the seismic stratigraphy, as well as the chronostratigraphic investigations of the Troll core, strongly indicate that the image reflects the initial phase of an extensive Middle Pleistocene glaciation (inferred age ca 0.5 Ma), prior to the development of a massive Norwegian Channel Ice Stream. The northwesterly oriented pattern, seen in the southeastern part of the study area, demonstrates that the ice flowed into the channel mainly from the coastal zone north of Bergen. Farther west various sets of ‘fan-shaped lineations’ partly cross each other, showing that the 3D-image represents a certain short time window. In the northern study area the ice flow was dominantly northwards, and directed towards the area below the present shallow Måløy Plateau. There are no indications that the glaciers north of Sognefjorden affected the marine ice sheet.     

Reconstructing the last Irish Ice Sheet 2: a geomorphologically-driven model of ice sheet growth,retreat and dynamics

The ice sheet that once covered Ireland has a long history of investigation. Much prior work focussed on localised evidence-based reconstructions and ice-marginal dynamics and chronologies, with less attention paid to an ice sheet wide view of the first order properties of the ice sheet: centres of mass, ice divide structure, ice flow geometry and behaviour and changes thereof. In this paper we focus on the latter aspect and use our new, countrywide glacial geomorphological mapping of the Irish landscape (>39 000 landforms), and our analysis of the palaeo-glaciological significance of observed landform assemblages (article Part 1), to build an ice sheet reconstruction yielding these fundamental ice sheet properties. We present a seven stage model of ice sheet evolution, from initiation to demise, in the form of palaeo-geographic maps. An early incursion of ice from Scotland likely coalesced with local ice caps and spread in a south-westerly direction 200 km across Ireland. A semi-independent Irish Ice Sheet was then established during ice sheet growth, with a branching ice divide structure whose main axis migrated up to 140 km from the west coast towards the east. Ice stream systems converging on Donegal Bay in the west and funnelling through the North Channel and Irish Sea Basin in the east emerge as major flow components of the maximum stages of glaciation. Ice cover is reconstructed as extending to the continental shelf break. The Irish Ice Sheet became autonomous (i.e. separate from the British Ice Sheet) during deglaciation and fragmented into multiple ice masses, each decaying towards the west. Final sites of demise were likely over the mountains of Donegal, Leitrim and Connemara. Patterns of growth and decay of the ice sheet are shown to be radically different: asynchronous and asymmetric in both spatial and temporal domains. We implicate collapse of the ice stream system in the North Channel – Irish Sea Basin in driving such asymmetry, since rapid collapse would sever the ties between the British and Irish Ice Sheets and drive flow configuration changes in response. Enhanced calving and flow acceleration in response to rising relative sea level is speculated to have undermined the integrity of the ice stream system, precipitating its collapse and driving the reconstructed pattern of ice sheet evolution.     

西藏地区重磁场特征及藏东构造格局新认识

根据新编制的1∶100万西藏地区重磁图发现,在西藏东部雁石坪—昌都地区存在一组北西向构造,它与西藏中西部发育的近东西向构造呈相交而不是相接关系,这与地质上通常认为的西藏地区主体呈北西—东西—北西西向的平卧"S"型构造格局观点相异,从而对西藏地区东部的地质构造格局有了新的认识。文中清晰地展示了这组北西向构造及近东西向构造的重磁场面貌,并推测认为这组北西向构造可能是从祁连山—东昆仑山北西向构造中沿东昆仑断裂分裂出来的,而先前认为的北西向构造与近东西向构造相接的观点反映的应是侏罗纪之后受印度地块向北推挤,上覆浅层构造特征,这个新发现对认识西藏地区地质构造格局具有重要意义。