海洋环流与波动重点实验室知识库

海洋流场作为海洋动力系统的核心要素之一，其对海洋动力过程、气候系统以及海洋生态系统均有深远的影响。然而，受海洋风场、地转效应等多种复杂因素的共同影响，海洋流场展现出高度的复杂性和变异性。相较于海洋温度和海洋盐度，海洋流场的观测数据相对匮乏，这也极大地增加了海洋流场重构的难度。海洋流场重构的相关研究成果较少，且主要依赖于动力学方法。然而，动力学模型在参数化复杂的动力学过程以及获取模型的边界条件信息方面存在局限性，因此在重构海洋流场时容易产生较大误差。
鉴于当前基于统计学方法的海洋内部流场重构研究尚显不足，为了验证其在海洋内部流场的重构效能，本文首先采用光梯度增强机器算法（Light Gradient Boosting Machine，LightGBM），构建了海表多源卫星观测与1000米深度处地转海洋学实时观测阵（Array for Real-Time Geostrophic Oceanography，Argo）的水平流之间的回归模型，实现了对南大洋1000米深度流场的重构。通过与Argo流场数据验证，LightGBM模型在描绘南大洋流场的时空特征方面优于多个再分析数据集。从定量上看，LightGBM 模型重构的流场与Argo流场大小的相关系数为0.78，均方根误差（Root Mean Square Error, RMSE）为4.07 cm/s，相较于海洋环流和气候估算模式（Estimating the Circulation and Climate of the Ocean，ECCO；其相关系数为0.51，RMSE为6.83cm/s）、全球海洋数据同化系统（Global Ocean Data Assimilation System，GODAS；其相关系数为0.26，RMSE为6.48cm/s）、全球海洋物理再分析（Global Ocean Physics Reanalyisis，GLORYS12V1；其相关系数为0.65，RMSE为6.0cm/s）和海洋再分析系统 5（Ocean Reanalysis System 5，ORAS5；其相关系数为0.27，RMSE为7.3cm/s），其精度表现更为优越。通过对重构流场多年趋势分析，揭示了自20世纪90年代以来，南大洋1000米深度流场的呈现了统计学上显著的增加，这为南大洋平均环流深层加速提供了新证据。
针对统计学方法在全水深水平流场重构中观测数据匮乏的难题，本文进一步提出了一种结合动力模态分解技术、海表卫星观测和Argo流场数据来重构海洋全水深水平流场的方法。动力模态分解被用于计算海洋流场的垂向动力模态系数，以此作为海洋流场的“代理”，以降低海洋流场的垂向维度，从而克服观测资料不足的限制。对海表观测的敏感性分析和重要性测试结果表明，海表流场在内部流场重构中占主导地位，同时海表高度、海表温度以及海表风场也对内部流场重构有积极贡献。Argo流场数据的验证结果显示，本研究选择的海表观测组合能够捕捉超过90%的正压模态变化以及超过80%的第一斜压模态变化，且重构流场展示了与Argo流场高度一致的空间特性。利用多普勒流速剖面仪（acoustic doppler current profiler，ADCP）实测流场的评估结果表明，重构流场的纬向速度相关系数达到0.78，经向速度相关系数达到0.74，并且在各个深度的精度均优于GLORYS12V1流场。相较于LightGBM模型，基于动力模态分解的重构模型在1000米深度的精度基本相当，这进一步验证了采用垂向动力模态系数作为“代理”的重构策略的有效性。该方法不仅保持了水平流场重构方面的高性能，还有效解决了观测数据不足带来的挑战，实现了海洋内部流场的重构。
最后，本文在德雷克海峡区域进行了全水深流场的重构，并估算了南极绕极流在德雷克海峡的总体积输运。与再分析数据的流场对比结果表明，重构的平均流场展示了与GLORY12SV1流场高度一致的空间特征；通过与重复的船测ADCP项目（简称为SADCP）观测流场断面对比，重构流场展现了与之非常吻合的流场横断面结果；与德雷克海峡阵列观测项目海底速度对比显示，重构的海底速度相关系数达到了0.7。对体积输运垂向分布分析显示，大约40%的体积输运发生在1000米以下的深度，因此准确确定深层海洋的流场状态对于估算南极绕极流在德雷克海峡的体积输运是至关重要。综合多年观测数据，本研究估算出南极绕极流在德雷克海峡的总体积输运量为164±2Sv，为深入理解南极绕极流的动态变化提供了重要依据。
 

As one of the core elements of the ocean dynamical system, the ocean flow field has far-reaching effects on ocean dynamical processes, the climate system, and the marine ecosystem. However, under the joint influence of many complex factors, such as the ocean wind and the geostrophic effect, the ocean flow field shows a high degree of complexity and variability. Compared with ocean temperature and salinity, there is a relative lack of observational data on the ocean flow field, which greatly increases the difficulty of reconstructing the ocean flow field. Presently, research on ocean current reconstruction remains limited, predominantly relying on dynamical approaches. However, dynamical models have limitations in parameterizing multiple complex factors and obtaining information about the boundary conditions, and thus are prone to large errors in reconstructing the ocean current field.
In order to reconstruct the interior flow field of the ocean using statistical methods, this paper initially employs the Light Gradient Boosting Machine (LightGBM) algorithm to construct a regression model between the satellite ocean surface observation and the Array for Real-Time Geostrophic Oceanography (Argo) horizontal flow at a depth of 1000 m, thereby enabling the reconstruction of the horizontal flow field in the Southern Ocean. The LightGBM model, validated by Argo, demonstrates superior performance to multiple reanalyzed datasets in depicting the spatial and temporal characteristics of the flow field in the Southern Ocean. Quantitatively, the correlation coefficient between the flow reconstructed by the LightGBM model and that of Argo is 0.78, and the Root Mean Square Error (RMSE) is 4.07 cm/s. This is better than Estimating the Circulation and Climate of the Ocean (ECCO, with a correlation coefficient of 0.51 and an RMSE of 6.83 cm/s), Global Ocean Data Assimilation System (GODAS, with a correlation coefficient of 0.26 and an RMSE of 6.48 cm/s), Global Ocean Physics Reanalyisis (GLORYS12V1, with a correlation coefficient of 0.65 and an RMSE of 6.0 cm/s), and Ocean Reanalysis System 5 (ORAS5, with a correlation coefficient of 0.27 and an RMSE of 7.3 cm/s). A multi-year trend analysis of the reconstructed flow field reveals a statistically significant increasing trend in the 1000 m depth flow field in the Southern Ocean since the 1990s, providing new evidence in support of this finding.
To address the dearth of observational data in the reconstruction of the full bathymetry horizontal flow field through statistical methods, this paper proposes a method for the reconstruction of the full bathymetry horizontal flow field, which combines the dynamical mode decomposition technique, surface satellite data, and Argo data. The dynamical mode decomposition is employed to calculate the vertical dynamical mode coefficients of the ocean flow field, which serves as a surrogate for the ocean flow field, thereby reducing the vertical dimension of the ocean flow field and overcoming the limitation of insufficient observation data. The sensitivity analysis and significance test of the ocean surface observations indicate that the ocean surface current exerts a dominant influence on the reconstruction of the flow field. Additionally, the ocean surface height, ocean surface temperature, and ocean surface wind also contribute positively to the reconstruction of the current. Validation with Argo demonstrates that the combination of ocean surface observations selected in this study is capable of capturing over 90% of the barotropic mode variations and over 80% of the first baroclinic mode variations. Furthermore, the reconstructed flow field exhibits spatial properties that are highly consistent with those observed in Argo. An evaluation of the measured flow field using acoustic doppler current profiler (ADCP) revealed that the reconstructed flow field exhibited zonal velocity correlation coefficients of 0.78 and meridional velocity correlation coefficients of 0.74. Furthermore, the reconstructed flow field demonstrated superior performance to the GLORYS12V1 at all depths. A comparison of the accuracies of the full-depth reconstruction model based on the dynamical modal decomposition with those of the LightGBM model reveals that they are essentially equivalent. This further substantiates the efficacy of the reconstruction strategy that employs the vertical dynamical mode coefficients as a surrogate. This approach not only preserves the high-performance characteristics of the horizontal flow field reconstruction but also effectively addresses the challenge of insufficient observational data.
A reconstruction of the full bathymetry horizontal flow field is finally carried out for the Drake Passage region, and the total volume transport of the Antarctic Circumpolar Current (ACC) in the Drake Passage is estimated accordingly. The results demonstrate that the reconstructed mean flow field exhibits strikingly consistent spatial characteristics with the GLORY12SV1. Furthermore, the reconstructed flow field displays a remarkable degree of agreement with the Repeat Shipboard ADCP Project cross-section (SADCP). Additionally, the reconstructed bottom velocity correlation coefficient reaches 0.7 when compared with the bottom velocity of the Drake Passage Array Project. The analysis of the vertical distribution of volume transport indicates that approximately 40% of the volume transport occurs at depths below 1000 m. Consequently, an accurate determination of the flow field state in the deep ocean is essential for estimating the volume transport of the Antarctic Circumpolar Current (ACC) in Drake Passage. By integrating data from multiple years of observation, this study estimates the total volume transport of the Antarctic Circumpolar Current (ACC) in the Drake Passage to be 164 ± 2 Sv, which provides a crucial foundation for a more profound comprehension of the dynamics of the ACC.
 

第1章 绪论    1
1.1 选题的背景、意义与依据    1
1.2 国内外研究进展与现状    3
1.2.1 海洋流场监测现状    3
1.2.2 海洋流场重构方法进展    5
1.2.3 南极绕极流总体积输运估算的现状    7
1.3 本文的研究工作    8
1.3.1 利用卫星与Argo数据的南大洋1000米水平流场的重构    9
1.3.2 基于动力模态分解的海洋内部流场重构方法研究    9
1.3.3 南极绕极流在德雷克海峡的平均体积输运估算    9
第2章 利用卫星与Argo数据的南大洋1000米水平流场的重构    11
2.1 数据资料介绍    11
2.1.1 卫星海表资料    11
2.1.2 Argo流场产品    12
2.1.3 再分析海洋流场产品    13
2.2 海洋流场重构方法与过程    13
2.2.1 LightGBM基本理论    13
2.2.2 南大洋1000米流场重构过程    14
2.3 流场重构模型确定与精度评定    16
2.4 重构流场与再分析流场产品对比    19
2.5 南大洋1000米流场检索与分析    26
2.6 本章小结    28
第3章 基于动力模态分解的内部流场重构方法研究    30
3.1 引言    30
3.2 数据资料介绍    30
3.2.1 海表观测资料    31
3.2.2 海洋流场资料    32
3.2.3 气候态海水温度与盐度    33
3.3 全水深水平流场重构方法介绍    34
3.3.1 海洋垂向动力模态分解    34
3.3.2 全水深流场重构模型构建    36
3.4 重构模型精度评定    39
3.4.1 重构的动力模态的系数的表现    39
3.4.2 重构流场与Argo流场对比    43
3.4.3 重构流场与ADCP流场对比    47
3.5 本章小结    51
第4章 南极绕极流在德雷克海峡的总体积输运估算    53
4.1 引言    53
4.2 数据资料与方法    54
4.2.1 数据资料情况    54
4.2.2 方法介绍    56
4.3 结果与分析    56
4.3.1 德雷克海峡平均流空间分布情况    56
4.3.2 重构流场评估    59
4.3.3 ACC平均总体积输运的估计    63
4.4 本章小结    66
第5章 结论与展望    68
5.1 主要结论    68
5.2 下一步研究与展望    69
参考文献    70
致  谢    79
作者简历及攻读学位期间发表的学术论文与其他相关学术成果    81