基于历史观测深度学习海洋预测模型构建

基于全域历史观测数据的深度学习海洋预测模型构建可行性与路径深度研究报告

1. 执行摘要

本研究报告旨在全面、详尽地回应关于“构建一个深度学习模型能够利用已有的尽可能多的人类对海洋的历史观测数据,对海洋进行预测”的可行性与技术路径。经过对全球海洋科学、人工智能算法架构、高性能计算以及最新的“AI for Science”研究成果的深入调研与分析,本报告确立了核心论点:构建这样一个模型不仅在理论上是可行的,而且在当前的技术爆发期,已经从早期的概念验证阶段迈向了具有实际操作能力的工程化阶段。然而,要真正实现利用“尽可能多”的原始观测数据,而非仅仅依赖经过平滑处理的再分析(Reanalysis)数据,仍面临着数据稀疏性、物理一致性以及多尺度时空建模的三重挑战。

当前的海洋预测范式正处于从基于物理方程的数值模式(Numerical Ocean Models)向数据驱动的AI模式(Data-Driven AI Models)转型的关键时期。以“曦和”(XiHe)、“文海”(WenHai)、AI-GOMS为代表的新一代海洋大模型,展示了利用数十PB级历史数据进行训练的巨大潜力,其推理速度较传统数值模式提升了数千倍,且在涡分辨率(Eddy-Resolving)尺度上展现出超越传统模式的预测技巧。然而,这些主流模型大多依赖于数值模式产生的再分析数据作为“真值”进行训练,这在一定程度上限制了模型挖掘原始观测数据中高频、非线性特征的能力。

为了回应利用“人类对海洋的历史观测数据”这一核心需求,本报告深入探讨了从“再分析驱动”向“观测驱动”演进的技术路径。特别关注了神经数据同化(Neural Data Assimilation, 如4DVarNet、ADAF-Ocean)技术的突破,这些技术使得深度学习模型能够直接摄入稀疏的Argo浮标数据、非结构化的卫星轨道数据,并将其映射为连续的海洋物理场。此外,物理信息神经网络(PINNs)与神经算子(Neural Operators)的结合,为解决稀疏数据下的物理约束问题提供了强有力的数学工具。

本报告将分章节详细论述全球海洋观测数据的全景图谱、现有海洋AI大模型的架构剖析、连接稀疏观测与稠密预测的关键技术、物理与AI的融合机制、以及该模型在极端天气、生态安全与水下环境保障等领域的应用前景,最终给出一份迈向“全域观测驱动海洋数字孪生”的战略路线图。

2. 引言:海洋预测的范式危机与AI的机遇

2.1 传统数值模式的物理瓶颈与算力墙

长期以来,人类对海洋状态的预测主要依赖于基于物理学的数值模式(General Circulation Models, GCMs)。这些模式的核心是通过离散化方法求解描述流体运动的Navier-Stokes方程组,以及热力学和盐度守恒方程。尽管数值模式在过去的半个世纪中取得了巨大的成功,成为了现代海洋学的基石,但其发展正日益逼近“算力墙”和“物理参数化瓶颈”。

首先,为了解析海洋中的中尺度涡(Mesoscale Eddies,尺度约10-100公里)甚至亚中尺度过程(Sub-mesoscale,尺度<10公里),数值模式需要极高的网格分辨率。计算复杂度的增长与分辨率的提升呈立方甚至四次方关系,这使得进行全球尺度、高分辨率(如1/12°或更高)、长时效的集合预报需要消耗天文数字般的高性能计算(HPC)资源 。[1][2]

其次,数值模式中存在大量无法被网格直接解析的物理过程(如云微物理、湍流混合、海气界面通量交换),这些过程必须通过经验性的“参数化方案”(Parameterization)来近似。这些参数化方案往往基于有限的观测和简化的假设,引入了难以消除的系统性误差,限制了模式在复杂海况下的预测精度 。[3][4]

2.2 数据驱动范式的崛起:AI for Ocean Science

近年来,随着对地观测技术的飞速发展,人类积累了海量的海洋历史观测数据。这些数据包括来自Argo计划的数百万条温盐深剖面、卫星高度计积累的三十年海面高度数据、以及海量的漂流浮标轨迹数据。这些PB级的数据资产,为深度学习(Deep Learning, DL)的应用提供了肥沃的土壤。

深度学习,特别是基于Transformer、图神经网络(GNN)和神经算子(Neural Operators)的架构,具备极其强大的非线性拟合能力。与遵循显式物理方程的数值模式不同,AI模型通过学习海量历史数据中的时空演变模式,构建起从当前状态到未来状态的映射函数(Surrogate Model)。这种“数据驱动”的方法具有两个显著优势:

2.3 本报告的核心命题

用户提出的“利用已有的尽可能多的人类对海洋的历史观测数据”这一要求,实际上触及了当前AI海洋学最前沿、也是最困难的问题。目前的“海洋大模型”(如Pangu-Weather, XiHe, WenHai)大多是在“再分析数据”(Reanalysis Data)上训练的。再分析数据本质上是数值模式与观测数据的融合产物,虽然时空连续,但也继承了数值模式的偏差。

真正的挑战在于:如何构建一个模型,能够直接消化原始的、稀疏的、噪声大的、非结构化的历史观测数据,并从中提取出物理规律进行预测? 本报告将重点围绕这一核心命题展开论述,从数据基础、算法架构到工程实现,全方位解析构建这样一个“全观测驱动”模型的可行性。

3. 全球海洋历史观测数据的全景图谱:AI模型的“燃料”

要构建利用“尽可能多”数据的模型,首先必须对现有的海洋数据资产进行详尽的盘点。海洋数据的复杂性远超文本或图像数据,其具有多模态、多尺度、高维稀疏的特征。

3.1 现场观测系统(In-situ Observations):海洋的“探针”

现场观测数据是海洋真值的最直接来源,也是训练高保真AI模型的基石。然而,它们的时空分布极不均匀,给模型训练带来了巨大挑战。

3.1.1 Argo全球浮标阵列

Argo计划是全球海洋观测系统中最重要的组成部分之一。它由近4000个在海洋中自由漂流的剖面浮标组成。

3.1.2 漂流浮标(Drifters)与锚系浮标(Moorings)

3.2 卫星遥感系统(Remote Sensing):海洋的“天眼”

卫星遥感提供了全球覆盖的高分辨率表面数据,是目前AI海洋模型主要的输入来源。

3.2.1 卫星高度计(Altimetry)

3.2.2 海表温度(SST)与盐度(SSS)

3.3 再分析数据(Reanalysis Data):目前的“事实标准”

尽管用户希望利用“原始观测”,但目前的AI模型大多训练于再分析数据。

4. 深度学习海洋预测模型的架构演进与SOTA分析

在探讨如何利用原始数据之前,我们需要深入剖析当前基于再分析数据取得成功的SOTA(State-of-the-Art)模型。这些模型证明了深度学习在大规模海洋预测中的可行性,并为未来的架构设计奠定了基础。

4.1 “曦和”(XiHe):纯数据驱动的涡分辨率预测

“曦和”模型是中国科研团队开发的一个里程碑式的全球海洋预测模型,其核心目标是实现高分辨率(1/12°)的涡分辨率预测。

4.2 “文海”(WenHai):物理引导的自回归系统

“文海”模型代表了AI与物理融合的另一条路径,它不仅仅是从图像到图像的映射,而是尝试将物理规律内嵌于网络结构中。

4.3 AI-GOMS:骨干-下游(Backbone-Downstream)范式

AI-GOMS(AI-Driven Global Ocean Modeling System)提出了一种类似于NLP领域(如BERT, GPT)的通用开发范式。

4.4 “琅琊”(LangYa):聚焦温跃层的物理自适应

针对深海预测中的痛点——温跃层(Thermocline),“琅琊”模型展示了专用优化的重要性。

5. 核心挑战与突破:从“稀疏观测”到“全场预测”

尽管上述模型取得了成功,但它们大多依赖再分析数据。为了回应“利用尽可能多的人类历史观测数据”这一需求,我们必须解决稀疏数据直接驱动(Direct Driving by Sparse Data) 的难题。这是因为原始观测(如Argo)在时空上是极度稀疏的,无法直接作为标准CNN或Transformer的输入。

5.1 神经数据同化(Neural Data Assimilation, Neural DA)

这是连接稀疏观测与全场预测的关键桥梁。传统的数据同化(如4D-Var)计算极其昂贵,而AI正在重塑这一过程。

5.1.1 4DVarNet:端到端的神经变分同化

5.1.2 ADAF-Ocean:基于神经过程(Neural Processes)的架构

5.2 物理信息神经网络(PINNs)在稀疏数据恢复中的应用

对于Argo浮标覆盖不到的广阔深海,如何利用物理规律补充数据?

6. 构建全域历史数据海洋大模型的可行性与路径

基于上述分析,构建一个利用全域历史数据的模型是完全可行的,但需要遵循特定的技术路线。

6.1 数据基础的可行性

6.2 推荐的技术路线图

阶段一:基于再分析数据的骨干预训练 (Backbone Pre-training)

阶段二:基于原始观测的神经同化微调 (Raw Data Fine-tuning via Neural DA)

阶段三:物理约束与闭环优化 (Physics-Informed Optimization)

7. 典型应用场景:从科研到实战

利用该模型,我们可以解决许多传统手段难以应对的海洋问题。

7.1 极端气候事件预测:突破ENSO预测障碍

7.2 海洋生态安全:有害藻华(HABs)的精准预警

7.3 水下环境保障:声场与潜器导航

7.4 亚中尺度涡旋的识别与追踪

8. 结论与展望:迈向海洋数字孪生

8.1 结论

构建一个利用已有尽可能多人类历史观测数据的深度学习海洋预测模型,不仅是可能的,而且是海洋科学发展的必然趋势

8.2 未来展望

未来的海洋预测将不再是单一的数值模拟,而是数据与物理深度融合的数字孪生(Digital Twin)

这是一个激动人心的时代,AI正在为古老的海洋科学注入前所未有的活力,让人类离“透明海洋”的梦想更近一步。

表 1:主流海洋AI大模型特性对比

模型名称 开发机构/团队 核心架构 训练数据来源 关键创新点 优势
XiHe (曦和) NUDT / OUC Transformer + SIE Block GLORYS12 Reanalysis SIE模块:提取各向异性空间特征;海陆掩膜:专注海洋区域 推理极快 (0.36s/10天),涡分辨率精度高
WenHai (文海) OUC / Laoshan Lab DNN + Physics Guidance GLORYS12 + ERA5 物理引导:嵌入海气通量块体公式;自回归:长时序预测 物理一致性好,垂直剖面预测准确
AI-GOMS OUC / 9cosmos MAE Backbone Reanalysis Data 骨干-下游范式:预训练通用模型,微调特定任务 灵活性高,易于迁移到生化等下游任务
LangYa (琅琊) OUC Encoder-Decoder ERA5 + GLORYS 自适应热跃层Loss:动态加权垂直梯度 深海温跃层结构预测显著优于通过模型
Pangu-Weather Huawei Cloud 3D Earth-Specific Transformer ERA5 (Atmosphere) 3D Transformer:立体建模;层次化时域:多尺度预测 虽主攻气象,但证明了Transformer处理地球流体的强大能力

[3:6][16:1][4:2][22:1][39]
表 2:传统数据同化 (DA) 与神经数据同化 (Neural DA) 对比

特性 传统数据同化 (如 4D-Var, EnKF) 神经数据同化 (如 4DVarNet, ADAF-Ocean)
核心机制 基于物理方程的迭代优化,最小化代价函数 基于神经网络学习观测与状态之间的映射函数
计算成本 极高,通常占据数值预报系统80%以上的计算时间 训练成本高,但在线推理成本极低,接近实时
数据利用 需要复杂的观测算子,对非高斯分布数据处理困难 可直接处理非线性、非高斯、多模态异构数据
稀疏数据处理 依赖背景场协方差矩阵(B矩阵)进行平滑 利用神经过程 (NP)先验学习进行智能补全
发展阶段 成熟,业务化标准 快速发展中,正逐步集成到业务系统

[26:2][24:1]


🔗 正文参考文献 (References)


🧠 思考过程 (Thoughts)

Step 1: 构建海洋大模型的宏观构想

我正着手评估构建一个“海洋大模型”的科学可行性,这不仅仅是预测单一的海洋指标,而是要尝试将数十年积累的全球海洋历史观测数据——包括卫星遥感、深海浮标以及再分析资料——整合进一个统一的智能架构中。我正在思考如何借鉴气象领域的前沿突破,将先进的深度学习技术迁移到更为复杂的多维海洋流体动力学预测中。

Step 2: 攻克数据稀疏性与物理约束

在初步构思中,我发现一个核心挑战在于如何处理空间上的极端不平衡:表层数据相对丰富,而深海数据则极其稀疏。我正致力于探索物理信息神经网络(PINN)与傅里叶神经算子(FNO)等前沿架构,试图通过将流体力学等物理定律嵌入模型,来弥补数据缺失带来的不确定性,从而提升模型在深海预测中的鲁棒性。

Step 3: 探索前沿技术路径与实践

接下来的行动中,我将重点调研全球范围内顶尖的AI海洋预测实例,并系统梳理各类关键数据集的规模与获取方式。我计划通过深入分析各主流架构在应对海洋异构数据时的表现,评估构建“海洋数字孪生”模型的技术红线与未来演进路径,为您呈现一套完整的技术可行性方案。

Step 4: [40][41][42][43][44][45][46][47][48][49][50][51][52][53][54][55][56][57][58][59][60][61][62][63]

Step 5: 全球海洋基座大模型的崛起

我发现海洋预测正经历一场范式转移,大型基座模型如“ORBIT”已拥有上千亿参数,能够实现长达30天的全球气象与海洋耦合预报。这些模型不再仅仅是黑盒回归,而是通过科学机器学习(SciML)将物理定律与海量数据深度结合。我观察到,当前的趋势是利用这类超大规模架构来捕捉此前难以模拟的超局部、高精度海洋变化,短临预报的准确率已在多个实验中突破90%。

Step 6: 多源观测数据的深度整合与重构

在梳理数据资产时,我注意到Argo全球浮标阵列与卫星遥感的协同已成为核心。通过整合如CORA等原位观测资料,研究人员正利用模拟同化技术填补早期观测的历史空白,甚至能够重构出前Argo时代的海洋三维状态。这种将历史碎片化数据转化为连续时空特征的技术,为构建覆盖数十年周期的深度学习模型提供了坚实的数据支撑。

Step 7: 算子神经网络对动力学模拟的突破

我深入思考了神经算子在解决海洋物理偏微分方程中的表现。傅里叶神经算子(FNO)及其改进版本通过引入时间模式,成功克服了长期预报中的数值稳定性难题。这种方法在捕获多尺度波浪传播方面展现出卓越的物理忠实度,其计算效率相比传统数值模拟提升了数个数量级,这暗示着实时海洋数字孪生的技术瓶颈正在被打破。

Step 8: 聚焦特定区域模型与数字孪生架构

接下来的行动中,我将重点剖析如“羲和(XiHe)”与“风虎”等特定前沿模型的内部细节,探索它们在处理中国近海及全球尺度差异化时的独特表现。同时,我将深入调研欧盟“数字孪生海洋”的技术架构,重点分析其如何将实时传感器流与AI推理引擎无缝融合,以期为您提供更具实操性的系统设计参考。

Step 9: [64][65][66][67][68][69][70][71][72][73][74][75][76][77][78][79][80][81][82][83][84][85][86][87][88][89][90][91][92][93][94][95][96][97][98][99][100][101][102][103][104][105][106][107]

Step 10: “羲和”与“AI-GOMS”:三维海洋建模的跨越

我发现海洋预测正从表层模拟转向全深度的三维重构。例如,“羲和”模型已能整合长达25年的高分辨率再分析资料,利用多达33层的垂直架构捕捉深海状态。与此同时,我注意到“AI-GOMS”等模型采用了先进的掩码自编码器技术,在特征提取阶段就实现了多变量的深度融合,这证明了深度学习模型已具备解析数十年跨度复杂观测数据的能力。

Step 11: 破解数据不对称与时空稀疏性挑战

我正在深入思考一个关键的矛盾点:卫星遥感提供了高频覆盖的海面数据,而深海观测(如Argo浮标)在空间上则极为稀疏。我观察到,当前的尖端研究正尝试引入AI变分同化技术,通过非线性映射将零散的深海点位数据与连续的卫星场进行耦合。这种方法不仅是在填补数据空白,更是在逻辑上重建深海与表层之间的动力学关联,以解决深海预测中的不确定性难题。

Step 12: 探究AI驱动的同化机制与垂直动力学约束

在接下来的思考中,我将重点调研“文海”模型的架构细节,并深入解析AI驱动的数据同化算法(如4DVarNet)如何处理历史观测数据中的噪声与残缺。我计划进一步探索物理信息约束在海洋环流模拟中的具体应用方式,以期揭示模型如何在缺乏直接观测的区域,通过物理规律推理出合理的海洋状态演化过程。

Step 13: [108][109][110][111][112][113][114][115][116][117][118][119][120][121][122][123][124][125][126][127][128][129][130][131][132][133][134][135][136]

Step 14: 辨析再分析资料与原始观测的深度桥梁

我发现当前最前沿的海洋大模型(如“羲和”)主要基于高精度的“再分析资料”进行训练,而非直接处理原始的碎片化观测。这意味着模型实际上是站在了传统同化技术的肩膀上,通过深度学习进一步捕捉复杂的动力学特征。我正在思考,AI究竟是在模仿已有的数值同化结果,还是已经具备了直接从零散的深海浮标数据中提取深层物理规律的能力,这将决定模型对历史长周期预测的上限。

Step 15: 破解深海数据孤岛的掩码重构机制

我注意到一个关键的技术路径:像“AI-GOMS”这样的架构正在引入掩码自编码技术,这为解决深海数据极端稀疏的矛盾提供了新思路。通过在训练中主动“遮蔽”部分已知海域,模型被赋予了更强的空间推断能力。我正在评估这种架构在处理空间分布极不均衡的历史观测资料时,如何确保生成的深海流场不仅在统计上合理,更在物理动力学上具备一致性。

Step 16: 探寻模型落地成本与数字孪生工程实践

在接下来的思考中,我将重点剖析超大规模海洋模型在训练阶段的资源消耗,试图通过具体的算力指标来衡量其在大规模历史数据处理中的可行性。同时,我计划深入调研特定区域的“数字孪生海洋”项目,特别是如何将物理信息约束从纯学术原型转化为实际业务预报中的核心引擎,以期为您呈现一套从理论建模到算力支撑的完整技术路线。

Step 17: [137][138][139][140][141][142][143][144][145][146][147][148][149][150][151][152][153][154][155][156][157][158][159][160][161][162][163][164][165][166][167][168][169][170][171][172][173][174][175][176][177][178][179][180][181]

Step 18: 区分再分析资料与原始观测的建模逻辑

我通过对比分析发现了一个至关重要的技术分野:目前主流的海洋基座模型大多是基于“再分析资料”训练的,这实际上是利用了传统数值模式预处理后的稠密数据。我正在思考,若要真正构建一个能直接吞噬数十年原始历史观测的模型,必须引入能处理极端稀疏性和非结构化数据的AI架构,而不仅是模仿现有的数值模拟结果。

Step 19: 跨越稀疏性障碍的AI同化新范式

在合成最新调研信息时,我注意到一种名为“AI驱动数据同化”的前沿趋势。通过引入如4DVarNet或ADAF-Ocean等架构,研究者正尝试绕过传统的复杂物理同化过程,利用神经算子直接从破碎的卫星轨迹和孤立的深海剖面中学习海洋动力学。这种方式赋予了模型更强的空间补全能力,能够更真实地还原前Argo时代的海洋三维状态。

Step 20: 物理保真度与神经算子的稳定性突破

我深入思考了如何确保长周期预测不发生动力学偏移。最新的进展显示,将物理偏微分方程(PDE)的残差显式嵌入到神经算子的损失函数中,已成为提升模型物理忠实度的核心。我发现这种“物理感知”的深度学习方法,不仅在计算效率上比传统数值模型提升了数个数量级,更在捕获多尺度波动传播方面展现出卓越的稳定性。

Step 21: 深化AI同化细节与工程可行性论证

在接下来的行动中,我将重点剖析AI同化模型处理原始异构观测的具体算法实现,并深入调研如“文海”等模型在物理约束层面的底层逻辑。我计划进一步评估构建国家级海洋数字孪生系统所需的算力与存储规模,以期为您呈现一套从海量历史观测到实时高精度预报的完整技术路线图。

Step 22: [182][183][184][185][186][187][188][189][190][191][192][193][194][195][196][197][198][199]

Step 23: 辨析基座模型与端到端同化路径

通过深度对比,我发现海洋预测正分化为两条核心路径。一类是如“羲和”与“文海”等基座模型,它们主要在经过预处理的高密度再分析资料上进行训练,追求极高的预报精度和稳定性;另一类则是以4DVarNet和ADAF-Ocean为代表的端到端架构,其精髓在于直接吞噬原始、破碎的观测数据。我意识到,这不仅是算法的差异,更是对如何利用历史观测资料的根本理解不同:是从成熟的模拟结果中学习规律,还是直接从自然界的观测碎片中重构物理世界。

Step 24: 物理感知与算子神经网络的性能跃迁

我正在剖析物理信息神经网络(PINN)与神经算子在海洋动态模拟中的深度融合。我观察到,诸如“LangYa”等前沿模型通过将流体动力学残差显式嵌入损失函数,成功解决了长期预报中易出现的物理偏移问题。这种设计使得模型在计算效率上比传统数值模拟提升了三个数量级,且在训练中展现出极强的物理保真度。这一发现证明了,深度学习模型不仅能加速预测,更能作为一种新的物理推断工具,在数据稀疏的海域实现逻辑自洽的状态估计。

Step 25: 评估工程可行性与国家级技术演进

接下来的行动中,我将重点调研中国“透明海洋”计划等国家级数字孪生项目的技术细节,特别是其如何将大规模异构观测流与AI推理引擎结合。我计划进一步挖掘“文海”模型在物理约束层面的底层逻辑,并根据具体的训练资源消耗指标,为您评估构建一个覆盖全深海、长周期观测数据的深度学习预测系统所需的算力壁垒与工程实施路径。

Step 26: [200][201][202][203][204][205][206][207][208][209][210][211][212][213][214][215][216][217]

Step 27: 深度学习对垂直动力学的精准解析

我深入剖析了“文海”与“琅琊”等前沿模型,发现它们在架构设计上已超越了简单的特征提取。通过引入“专家引导”的架构逻辑和自适应温跃层损失函数,这些模型能够精准捕捉海洋内部复杂的垂直梯度变化。我意识到,这标志着深度学习已从表层特征模拟,演进到能够利用历史剖面数据重建深海动力学结构的阶段,有效解决了深海观测点位稀疏导致的建模失真难题。

Step 28: 算力飞跃与海洋数字孪生的工程基准

在评估工程可行性时,我获取了关键的性能指标:当前主流模型在推理速度上比传统数值模拟提升了近1000倍。然而,这种跨越式提升建立在庞大的算力底座之上,例如训练一个高精度的全球模型通常需要数十块高性能A800 GPU持续运行数百小时。这一数据点为我构建“海洋数字孪生”系统提供了明确的硬件参考标准,即利用AI实现实时高精度预报在技术上是完全可行的,但需具备相应的工程支撑。

Step 29: 从模型仿真到“透明海洋”的战略闭环

我已系统梳理了包括中国“透明海洋”计划和欧盟“数字孪生海洋”在内的全球核心项目,并厘清了它们在UN“DITTO”框架下的演进路径。我发现,研究重心正从单纯的数值模拟器(Emulator)转向能够直接处理原始观测流的自主推理引擎。至此,我已完成了对模型架构、数据同化、物理约束及工程实现的全方位论证,准备将这些零散的洞察合成为一份关于构建全球海洋预测大模型的技术可行性报告。

🔗 思考过程参考文献


  1. Crafting the Future: Machine learning for ocean forecasting - Reports ↩︎

  2. Principled Operator Learning in Ocean Dynamics: The Role of Temporal Structure - arXiv ↩︎ ↩︎

  3. XiHe: A Data-Driven Model for Global Ocean Eddy-Resolving Forecasting - arXiv ↩︎ ↩︎ ↩︎ ↩︎ ↩︎ ↩︎ ↩︎

  4. AI-GOMS: Large AI-Driven Global Ocean Modeling System - ResearchGate ↩︎ ↩︎ ↩︎

  5. Argo program - NOAA/AOML ↩︎ ↩︎

  6. Combining neural networks and data assimilation to enhance the spatial impact of Argo floats in the Copernicus Mediterranean biogeochemical model - OS ↩︎

  7. Estimating the AMOC from Argo profiles with machine learning trained on ocean simulations ↩︎

  8. Deep Learning Improvements For Sparse Spatial Field Reconstruction - arXiv ↩︎

  9. Deep learning prediction of two-dimensional ocean dynamics with wavelet-compressed data - PubMed ↩︎ ↩︎

  10. 4DVarNet : Data-driven mapping of Ocean Surface Topography using SWOT altimetry ↩︎

  11. 4DVarNet-SSH: end-to-end learning of variational interpolation schemes for nadir and wide-swath satellite altimetry - GMD ↩︎ ↩︎

  12. Ocean-Intelligent-Forecasting/XiHe-GlobalOceanForecasting - GitHub ↩︎ ↩︎ ↩︎

  13. CORA, Coriolis Ocean Dataset for Reanalysis - SEANOE ↩︎

  14. Forecasting the eddying ocean with a deep neural network - ResearchGate ↩︎ ↩︎

  15. XiHe: A Data-Driven Model for Global Ocean Eddy-Resolving Forecasting - ResearchGate ↩︎

  16. Forecasting the eddying ocean with a deep neural network - PubMed ↩︎ ↩︎

  17. Forecasting the eddying ocean with a deep neural network - PMC - NIH ↩︎

  18. A schematic of WenHai's architecture The surface atmosphere and ocean... - ResearchGate ↩︎

  19. Advanced Long-term Earth System Forecasting - arXiv ↩︎

  20. [PDF] AI-GOMS: Large AI-Driven Global Ocean Modeling System | Semantic Scholar ↩︎

  21. [2308.03152] AI-GOMS: Large AI-Driven Global Ocean Modeling System - arXiv ↩︎

  22. LangYa: Revolutionizing Cross-Spatiotemporal Ocean Forecasting - arXiv ↩︎ ↩︎

  23. LangYa: Revolutionizing Cross-Spatiotemporal Ocean Forecasting - ResearchGate ↩︎

  24. End-to-End Nomadic Ocean Forecasting: Bridging Compression, Assimilation, and Task-Specific Metrics Using Deep Learning - IEEE Xplore ↩︎ ↩︎

  25. Neural variational Data Assimilation with Uncertainty Quantification using SPDE priors - AMS Journals ↩︎

  26. Advancing Ocean State Estimation with efficient and scalable AI - arXiv ↩︎ ↩︎ ↩︎

  27. Advancing Ocean State Estimation with efficient and scalable AI - ChatPaper ↩︎

  28. Physics-informed neural network with pretraining optimization for ocean acoustic field prediction - AIP Publishing ↩︎ ↩︎

  29. PHYSICS-INFORMED NEURAL NETWORKS FOR MODELING OCEAN POLLUTANT - arXiv ↩︎

  30. Applying Physical Constraints in Physics-Informed Neural Networks (PINNs): Methods and Examples | by Noor Fatima | Medium ↩︎

  31. Deep learning for skillful long-lead ENSO forecasts - Frontiers ↩︎

  32. Combined dynamical-deep learning ENSO forecasts - the NOAA Institutional Repository ↩︎

  33. Applying Machine Learning to Harmful Algal Blooms - NASA Earthdata ↩︎

  34. Toward a Predictive Understanding of Cyanobacterial Harmful Algal Blooms through AI Integration of Physical, Chemical, and Biological Data - PubMed Central ↩︎

  35. A Metadata-Enhanced Deep Learning Method for Sea Surface Height and Mesoscale Eddy Prediction - MDPI ↩︎

  36. Full article: Multi-scale eddy identification and analysis based on deep learning method and ocean color data ↩︎

  37. Europe Showcases AI-Powered Digital Twin of the Ocean as a Major Outcome of the European Ocean Pact | Vlaams Instituut voor de Zee ↩︎

  38. AI-Powered Climate and Weather Simulation Platform | NVIDIA Earth-2 ↩︎

  39. Forecasting Capability Verification of the Pangu-Weather and IFS HRES for the 2022 Summer Weather in Jiangsu Province, China in - AMS Journals ↩︎

  40. journals.ametsoc.orgOceanPredict'24 Symposium: Advancing Ocean Prediction Science for Societal Benefit in - AMS Journals ↩︎

  41. sp.copernicus.orgCrafting the Future: Machine learning for ocean forecasting - Reports ↩︎

  42. nationalacademies.orgNew Report Provides Forecast for the Next Decade of Ocean Science ↩︎

  43. olcf.ornl.govFine-Tuning Forecasts: ORBIT Brings Long-Range Weather Prediction Within Reach ↩︎

  44. arxiv.orgA Sentinel-3 foundation model for ocean colour - arXiv ↩︎

  45. aoml.noaa.govArgo program - NOAA/AOML ↩︎

  46. journals.ametsoc.orgReconstructing the Ocean State Using Argo Data and a Data-Driven Method in - AMS Journals ↩︎

  47. ncei.noaa.govGlobal Argo Data Repository | National Centers for Environmental Information (NCEI) ↩︎

  48. seanoe.orgCORA, Coriolis Ocean Dataset for Reanalysis - SEANOE ↩︎

  49. argo.ucsd.eduArgo ↩︎

  50. mdpi.comBeyond the Backbone: A Quantitative Review of Deep-Learning Architectures for Tropical Cyclone Track Forecasting - MDPI ↩︎

  51. pubmed.ncbi.nlm.nih.govFrom monochromatic waves to realistic tides: deep learning for short-term forecasting of coastal ocean - PubMed ↩︎

  52. arxiv.orgPrincipled Operator Learning in Ocean Dynamics: The Role of Temporal Structure - arXiv ↩︎

  53. findaphd.comNeural Network Architectures for Enhanced Ocean Current Forecasting at National Sun Yat-Sen University on FindAPhD.com ↩︎

  54. pubmed.ncbi.nlm.nih.govDeep learning prediction of two-dimensional ocean dynamics with wavelet-compressed data - PubMed ↩︎

  55. pubs.aip.orgPhysics-informed neural network with pretraining optimization for ocean acoustic field prediction - AIP Publishing ↩︎

  56. academic.oup.comPhysics-informed neural networks for offshore tsunami data assimilation - Oxford Academic ↩︎

  57. arxiv.orgPHYSICS-INFORMED NEURAL NETWORKS FOR MODELING OCEAN POLLUTANT - arXiv ↩︎

  58. mdpi.comUnderstanding Physics-Informed Neural Networks: Techniques, Applications, Trends, and Challenges - MDPI ↩︎

  59. frontiersin.orgDeep learning for skillful long-lead ENSO forecasts - Frontiers ↩︎

  60. arxiv.orgTowards Long-Range ENSO Prediction with an Explainable Deep Learning Model - arXiv ↩︎

  61. gmd.copernicus.orgENSO-ASC 1.0.0: ENSO Deep Learning Forecast Model with a Multivariate Air-Sea Coupler ↩︎

  62. repository.library.noaa.govCombined dynamical-deep learning ENSO forecasts - the NOAA Institutional Repository ↩︎

  63. pubmed.ncbi.nlm.nih.govDeep learning for multi-year ENSO forecasts - PubMed ↩︎

  64. arxiv.orgXiHe: A Data-Driven Model for Global Ocean Eddy-Resolving Forecasting - arXiv ↩︎

  65. arxiv.org[2402.02995] XiHe: A Data-Driven Model for Global Ocean Eddy-Resolving Forecasting ↩︎

  66. github.comOcean-Intelligent-Forecasting/XiHe-GlobalOceanForecasting - GitHub ↩︎

  67. mercator-ocean.euMercator launches OceanBench for AI forecasting ↩︎

  68. journals.ametsoc.orgForecasting Capability Verification of the Pangu-Weather and IFS HRES for the 2022 Summer Weather in Jiangsu Province, China in - AMS Journals ↩︎

  69. docs.nvidia.comPangu Weather for Global Weather Forecasting — NVIDIA PhysicsNeMo Framework ↩︎

  70. charts.ecmwf.intPangu-Weather - ECMWF | Charts ↩︎

  71. mdpi.comEvaluation of Near-Taiwan Strait Sea Surface Wind Forecast Based on PanGu Weather Prediction Model - MDPI ↩︎

  72. arxiv.orgPangu-Weather: A 3D High-Resolution System for Fast and Accurate Global Weather Forecast - arXiv ↩︎

  73. researchgate.netFengWu-W2S: A deep learning model for seamless weather-to-subseasonal forecast of global atmosphere | Request PDF - ResearchGate ↩︎

  74. arxiv.orgFengWu-4DVar: Coupling the Data-driven Weather Forecasting Model with 4D Variational Assimilation - arXiv ↩︎

  75. en.people.cnChina's AI model advances effective forecast time of global mid-term weather to 11.25 days ↩︎

  76. vercel.hyper.aiSolving the Problems of Data Silos, Computing Consumption, and ↩︎

  77. arxiv.org[2411.10191] FengWu-W2S: A deep learning model for seamless weather-to-subseasonal forecast of global atmosphere - arXiv ↩︎

  78. researchgate.netAI-GOMS: Large AI-Driven Global Ocean Modeling System - ResearchGate ↩︎

  79. agu.confex.comAI-GOMS: Large AI-Driven Global Ocean Modeling System ↩︎

  80. semanticscholar.org[PDF] AI-GOMS: Large AI-Driven Global Ocean Modeling System | Semantic Scholar ↩︎

  81. arxiv.org[2308.03152] AI-GOMS: Large AI-Driven Global Ocean Modeling System - arXiv ↩︎

  82. tandfonline.comFull article: AI performs on high-resolution three-dimensional ocean temperature forecasting: remote sensing data-driven becomes a new possibility ↩︎

  83. scmp.comDigital artwork competition sees contestants use artificial intelligence to paint the future of humanity ↩︎

  84. thequantuminsider.comChinese Researchers Use Quantum Computer to Fine-Tune Billion-Parameter AI Model ↩︎

  85. freepik.comSurreal d depiction of wukong splitting the ocean with his staff waves frozen midmotion vibrant and dynamic with intricate textures | Premium AI-generated image - Freepik ↩︎

  86. iconscout.comAI Sun Wukong Images - Royalty-Free AI Visuals & Art | IconScout ↩︎

  87. github.comJimmyLv/awesome-nano-banana: Awesome curated collection of images and prompts generated by gemini-2.5-flash-image (aka Nano Banana) state-of-the-art image generation and editing model. Explore AI generated visuals created with Gemini, showcasing Google's advanced image generation capabilities. - GitHub ↩︎

  88. vliz.beEurope Showcases AI-Powered Digital Twin of the Ocean as a Major Outcome of the European Ocean Pact | Vlaams Instituut voor de Zee ↩︎

  89. blue-cloud.orgEDITO (European Digital Twin Ocean) | Blue-Cloud 2026 ↩︎

  90. digitaltwinocean.mercator-ocean.euDigital Twin Ocean ↩︎

  91. research-and-innovation.ec.europa.euEuropean Digital Twin of the Ocean (European DTO) - Research and innovation ↩︎

  92. ihr.iho.intDigital Twins of the Ocean can foster a sustainable blue economy in a protected marine environment - IHR - International Hydrographic Review ↩︎

  93. blogs.nvidia.comKeeping AI on the Planet: NVIDIA Technologies Make Every Day About Earth Day ↩︎

  94. nvidia.comAI-Powered Climate and Weather Simulation Platform | NVIDIA Earth-2 ↩︎

  95. developer.nvidia.comForecasting the Weather Beyond Two Weeks Using NVIDIA Earth-2 ↩︎

  96. developer.nvidia.comTag: Climate / Weather / Ocean Modeling | NVIDIA Technical Blog ↩︎

  97. blogs.nvidia.comAmphitrite Rides AI Wave to Boost Maritime Shipping, Ocean Cleanup With Real-Time Weather Prediction and Simulation | NVIDIA Blog ↩︎

  98. pmc.ncbi.nlm.nih.govForecasting the eddying ocean with a deep neural network - PMC - NIH ↩︎

  99. mdpi.comA Metadata-Enhanced Deep Learning Method for Sea Surface Height and Mesoscale Eddy Prediction - MDPI ↩︎

  100. tandfonline.comFull article: Multi-scale eddy identification and analysis based on deep learning method and ocean color data ↩︎

  101. essd.copernicus.orgA submesoscale eddy identification dataset in the northwest Pacific Ocean derived from GOCI I chlorophyll a data based on deep learning - ESSD Copernicus ↩︎

  102. frontiersin.orgIdentification of mesoscale eddies based on improved YOLOv8 model: a case study in the South China Sea - Frontiers ↩︎

  103. earthdata.nasa.govApplying Machine Learning to Harmful Algal Blooms - NASA Earthdata ↩︎

  104. pmc.ncbi.nlm.nih.govToward a Predictive Understanding of Cyanobacterial Harmful Algal Blooms through AI Integration of Physical, Chemical, and Biological Data - PubMed Central ↩︎

  105. ejosdr.comUsing explainable AI to predict algal bloom forecasting and real-time analysis of deteriorating water quality in aquatic ecosystems ↩︎

  106. mdpi.comBloomSense: Integrating Automated Buoy Systems and AI to Monitor and Predict Harmful Algal Blooms - MDPI ↩︎

  107. gmd.copernicus.orgPrediction of algal blooms via data-driven machine learning models: an evaluation using data from a well-monitored mesotrophic lake - GMD ↩︎

  108. pubmed.ncbi.nlm.nih.govForecasting the eddying ocean with a deep neural network - PubMed ↩︎

  109. pmc.ncbi.nlm.nih.govForecasting the eddying ocean with a deep neural network - PMC - NIH ↩︎

  110. arxiv.orgA data-driven global ocean forecasting model with sub-daily and eddy-resolving resolution - arXiv ↩︎

  111. researchgate.netA schematic of WenHai's architecture The surface atmosphere and ocean... - ResearchGate ↩︎

  112. arxiv.orgFuXi-Ocean: A Global Ocean Forecasting System with Sub-Daily Resolution - arXiv ↩︎

  113. ideas.repec.orgForecasting the eddying ocean with a deep neural network - IDEAS/RePEc ↩︎

  114. researchgate.netWenhai Xu's research works | Dalian Maritime University and other places ↩︎

  115. arxiv.org[2511.18732] OceanForecastBench: A Benchmark Dataset for Data-Driven Global Ocean Forecasting - arXiv ↩︎

  116. ieeexplore.ieee.orgEnd-to-End Nomadic Ocean Forecasting: Bridging Compression, Assimilation, and Task-Specific Metrics Using Deep Learning - IEEE Xplore ↩︎

  117. marine.copernicus.eu4DVarNet-OFDA | CMEMS - Copernicus Marine Service ↩︎

  118. github.comCIA-Oceanix/4dvarnet-turbiditymapping - GitHub ↩︎

  119. cambridge.orgEnsemble-based 4DVarNet uncertainty quantification for the reconstruction of sea surface height dynamics | Environmental Data Science | Cambridge Core ↩︎

  120. arxiv.orgPHYSICS-INFORMED NEURAL NETWORKS FOR MODELING OCEAN POLLUTANT - arXiv ↩︎

  121. arxiv.org[2507.08834] Physical Informed Neural Networks for modeling ocean pollutant - arXiv ↩︎

  122. frontiersin.orgPhysics-informed neural networks to reconstruct surface velocity field from drifter data ↩︎

  123. oceania.inria.clASSESSING PHYSICS INFORMED NEURAL NETWORKS IN OCEAN MODELLING AND CLIMATE CHANGE APPLICATIONS - OcéanIA ↩︎

  124. ieeexplore.ieee.orgPhysics Informed Neural Networks for Modeling Large-Scale Wind Driven Ocean Circulation - IEEE Xplore ↩︎

  125. sp.copernicus.orgCrafting the Future: Machine learning for ocean forecasting - Reports ↩︎

  126. sp.copernicus.orgData assimilation schemes for ocean forecasting: state of the art - Reports ↩︎

  127. egusphere.copernicus.orgConvolutional Neural Networks for Sea Surface Data Assimilation in Operational Ocean Models: Test Case in the Gulf of Mexico - EGUsphere ↩︎

  128. researchgate.netMachine learning in ocean data assimilation: Advances, gaps and the road to operations ↩︎

  129. mdpi.comSpecial Issue : Earth System Modeling, Data Assimilation, Artificial Intelligence, Deep Learning and Ocean Information Engineering II - MDPI ↩︎

  130. github.commercator-ocean/oceanbench: Benchmark evaluating ocean forecasting systems against reference datasets and observations. - GitHub ↩︎

  131. proceedings.neurips.ccOceanBench: The Sea Surface Height Edition - NeurIPS ↩︎

  132. openreview.netOceanBench: A Benchmark for Data-Driven Global Ocean Forecasting systems | OpenReview ↩︎

  133. papers.nips.ccOCEANBENCH: The Sea Surface Height Edition - Supplementary Material A Data Challenges - NIPS papers ↩︎

  134. arxiv.org[2309.15599] OceanBench: The Sea Surface Height Edition - arXiv ↩︎

  135. researchgate.netXiHe: A Data-Driven Model for Global Ocean Eddy-Resolving Forecasting - ResearchGate ↩︎

  136. arxiv.orgXiHe: A Data-Driven Model for Global Ocean Eddy-Resolving Forecasting - arXiv ↩︎

  137. arxiv.orgAdvanced Long-term Earth System Forecasting - arXiv ↩︎

  138. researchgate.netForecasting the eddying ocean with a deep neural network - ResearchGate ↩︎

  139. journals.ametsoc.orgLong-Term Predictions of Loop Current Eddy Evolutions Using OceanNet: A Fourier Neural Operator–Based Data-Driven Ocean Emulator in - AMS Journals ↩︎

  140. arxiv.orgOceanForecastBench: A Benchmark Dataset for Data-Driven Global Ocean Forecasting - arXiv ↩︎

  141. arxiv.orgHow to systematically develop an effective AI-based bias correction model? - arXiv ↩︎

  142. journals.ametsoc.orgNeural Prediction of Lagrangian Drift Trajectories on the Sea Surface in - AMS Journals ↩︎

  143. os.copernicus.orgEstimating the AMOC from Argo profiles with machine learning trained on ocean simulations ↩︎

  144. pmc.ncbi.nlm.nih.govSpatiotemporal upscaling of sparse air-sea pCO2 data via physics-informed transfer learning - PMC - NIH ↩︎

  145. arxiv.orgDeep Learning Improvements For Sparse Spatial Field Reconstruction - arXiv ↩︎

  146. ai.stackexchange.comAre deep learning models suitable for training with sparse data? - AI Stack Exchange ↩︎

  147. os.copernicus.orgCombining neural networks and data assimilation to enhance the spatial impact of Argo floats in the Copernicus Mediterranean biogeochemical model - OS ↩︎

  148. ditto-oceandecade.orgDITTO - Digital Twins of the Ocean ↩︎

  149. ditto-oceandecade.orgAbout - DITTO - Digital Twins of the Ocean ↩︎

  150. piernext.portdebarcelona.catDigital Twins, the Cornerstone of China's Port Strategy - PierNext ↩︎

  151. oceandecade.orgInternational DITTO Summit 2023 - Events - Ocean Decade ↩︎

  152. ditto-summit2023.scimeeting.cnInternational Digital Twins of the Ocean Summit 2023 ↩︎

  153. journals.ametsoc.orgDeep Learning Augmented Data Assimilation: Reconstructing Missing Information with Convolutional Autoencoders in - AMS Journals ↩︎

  154. eartharxiv.orgGenerative Data Assimilation for Surface Ocean State Estimation from Multi-Modal Satellite Observations - EarthArXiv ↩︎

  155. frontiersin.orgFusion of ocean data from multiple sources using deep learning: Utilizing sea temperature as an example - Frontiers ↩︎

  156. repository.library.noaa.govUsing Machine Learning for Data Assimilation, Model Physics, and Post-Processing model outputs - the NOAA Institutional Repository ↩︎

  157. github.comOcean-Intelligent-Forecasting/XiHe-GlobalOceanForecasting - GitHub ↩︎

  158. chatpaper.comAdvancing Ocean State Estimation with efficient and scalable AI - ChatPaper ↩︎

  159. arxiv.orgAdvancing Ocean State Estimation with efficient and scalable AI - arXiv ↩︎

  160. researchgate.netAdvancing Ocean State Estimation with efficient and scalable AI | Request PDF - ResearchGate ↩︎

  161. catalyzex.comYanfei Xiang - CatalyzeX ↩︎

  162. researchgate.netAn overview of AI‐based data assimilation framework (ADAF) and the... - ResearchGate ↩︎

  163. journals.ametsoc.orgNeural variational Data Assimilation with Uncertainty Quantification using SPDE priors - AMS Journals ↩︎

  164. meetingorganizer.copernicus.orgSession ITS1.2/OS4.10 - CO Meeting Organizer ↩︎

  165. researchgate.netA Four‐Dimensional Variational Informed Generative Adversarial Network for Data Assimilation - ResearchGate ↩︎

  166. scitepress.orgMultimodal Unsupervised Spatio-Temporal Interpolation of Satellite Ocean Altimetry Maps - SciTePress ↩︎

  167. arxiv.orgA Benchmark for AI-based Weather Data Assimilation - arXiv ↩︎

  168. arxiv.orgNeuralOM: Neural Ocean Model for Subseasonal-to-Seasonal Simulation - arXiv ↩︎

  169. researchgate.netWenhai Xu's research works | Dalian Maritime University and other places ↩︎

  170. arxiv.orgLangYa: Revolutionizing Cross-Spatiotemporal Ocean Forecasting - arXiv ↩︎

  171. ww3.math.ucla.eduFine-Tune Language Models as Differential Equation Solvers ↩︎

  172. academic.oup.comExcitation-time imaging condition reverse-time migration based on a physics-informed neural network traveltime calculation with wavefield decomposition using an optical flow vector | Journal of Geophysics and Engineering | Oxford Academic ↩︎

  173. researchgate.netPractical options for selecting data-driven or physics-based prognostics algorithms with reviews | Request PDF - ResearchGate ↩︎

  174. arxiv.orgXiHe: A Data-Driven Model for Global Ocean Eddy-Resolving Forecasting - arXiv ↩︎

  175. openreview.netFuXi-Ocean: A Global Ocean Forecasting System with Sub-Daily Resolution - OpenReview ↩︎

  176. arxiv.orgOcean-E2E: Hybrid Physics-Based and Data-Driven Global ... - arXiv ↩︎

  177. openai.comGenerative modeling with sparse transformers - OpenAI ↩︎

  178. frontiersin.orgHuman behavior recognition based on sparse transformer with channel attention mechanism - Frontiers ↩︎

  179. pmc.ncbi.nlm.nih.govHuman behavior recognition based on sparse transformer with channel attention mechanism - PMC - PubMed Central ↩︎

  180. youtube.comAttention in transformers, step-by-step | Deep Learning Chapter 6 - YouTube ↩︎

  181. arxiv.orgThe Sparse Frontier: Sparse Attention Trade-offs in Transformer LLMs - arXiv ↩︎

  182. researchgate.netAdvancing Ocean State Estimation with efficient and scalable AI | Request PDF - ResearchGate ↩︎

  183. chatpaper.comAdvancing Ocean State Estimation with efficient and scalable AI - ChatPaper ↩︎

  184. arxiv.org[2511.06041] Advancing Ocean State Estimation with efficient and scalable AI - arXiv ↩︎

  185. arxiv.orgAdvancing Ocean State Estimation with efficient and scalable AI - arXiv ↩︎

  186. researchgate.netAn overview of AI‐based data assimilation framework (ADAF) and the... - ResearchGate ↩︎

  187. marine.copernicus.eu4DVarNet-OFDA | CMEMS - Copernicus Marine Service ↩︎

  188. researchgate.net(PDF) A Four‐Dimensional Variational Constrained Neural Network‐Based Data Assimilation Method - ResearchGate ↩︎

  189. swotst.aviso.altimetry.fr4DVarNet : Data-driven mapping of Ocean Surface Topography using SWOT altimetry ↩︎

  190. gmd.copernicus.org4DVarNet-SSH: end-to-end learning of variational interpolation schemes for nadir and wide-swath satellite altimetry - GMD ↩︎

  191. gmd.copernicus.org4DVarNet-SSH: end-to-end learning of variational interpolation schemes for nadir and wide-swath satellite altimetry - GMD ↩︎

  192. mdpi.comReview of Physics-Informed Neural Networks: Challenges in Loss Function Design and Geometric Integration - MDPI ↩︎

  193. amazon.sciencePhysics-constrained machine learning for scientific computing - Amazon Science ↩︎

  194. medium.comApplying Physical Constraints in Physics-Informed Neural Networks (PINNs): Methods and Examples | by Noor Fatima | Medium ↩︎

  195. openreview.netFrom Pixels to Views: Learning Angular-Aware and Physics ↩︎

  196. repository.library.noaa.govPhysics constraint Deep Learning based radiative transfer model - the NOAA Institutional Repository ↩︎

  197. pmc.ncbi.nlm.nih.govScaling transformers to high-dimensional sparse data: a Reformer-BERT approach for large-scale classification - PMC - PubMed Central ↩︎

  198. researchgate.netArgo data anomaly detection based on transformer and Fourier transform - ResearchGate ↩︎

  199. arxiv.org[2111.12763] Sparse is Enough in Scaling Transformers - arXiv ↩︎

  200. mdpi.comCan the Digital Economy Improve the Quality of the Marine Environment? Empirical Evidence from Coastal Provinces and Cities in China - MDPI ↩︎

  201. mdpi.comThe Digital: A Catalyst for Accelerating the Quality Improvement and Sustainable Development of China's Marine Industry - MDPI ↩︎

  202. prism.sustainability-directory.comFuture of Underwater Military Technology Race - Prism → Sustainability Directory ↩︎

  203. researchgate.net(PDF) Toward digital twin of the ocean: from digitalization to cloning - ResearchGate ↩︎

  204. andrewerickson.comChina Maritime Report No. 3: China's Distant-Ocean Survey Activities: Implications for U.S. National Security - Andrew Erickson ↩︎

  205. arxiv.orgA data-driven global ocean forecasting model with sub-daily and eddy-resolving resolution - arXiv ↩︎

  206. researchgate.netForecasting the eddying ocean with a deep neural network - ResearchGate ↩︎

  207. mdpi.comForecasting Vertical Profiles of Ocean Currents from Surface Characteristics: A Multivariate Multi-Head Convolutional Neural Network–Long Short-Term Memory Approach - MDPI ↩︎

  208. arxiv.orgLangYa: Revolutionizing Cross-Spatiotemporal Ocean Forecasting - arXiv ↩︎

  209. arxiv.orgNeuralOM: Neural Ocean Model for Subseasonal-to-Seasonal Simulation - arXiv ↩︎

  210. openreview.netFuXi-Ocean: A Global Ocean Forecasting System with Sub-Daily Resolution - OpenReview ↩︎

  211. arxiv.orgLangya: Revolutionizing Cross-Spatiotemporal Ocean Forecastin - arXiv ↩︎

  212. researchgate.netLangYa: Revolutionizing Cross-Spatiotemporal Ocean Forecasting - ResearchGate ↩︎

  213. tandfonline.comData science for oceanography: from small data to big data - Taylor & Francis ↩︎

  214. pmc.ncbi.nlm.nih.govPhysics-informed deep-learning parameterization of ocean vertical mixing improves climate simulations - PMC - NIH ↩︎

  215. fio.org.cnApplication of machine learning in ocean data ↩︎

  216. tribune.com.pkConstruction starts on deep sea research center in China's Qingdao - The Express Tribune ↩︎

  217. eweb.ouc.edu.cnProf. Wei Zhiqiang Talks about Intelligent Ocean in the Age of Big Data ↩︎