大数据:概念、技术及应用研究综述

随着物联网、云计算、移动互联网的迅猛发展,大数据(Big Data)吸引了越来越多的关注,正成为信息社会的重要财富,同时也给数据的处理与管理带来了巨大挑战.首先从大数据概念入手,阐述了大数据的来源、主要挑战、关键技术、大数据处理工具和应用实例等,并对比了大数据与云计算、物联网、移动互联网等技术之间关系,然后剖析了大数据核心技术、大数据企业解决方案,讨论了目前大数据应用实例,最后归纳总结了大数据发展趋势.旨在为了解大数据当前发展状况、关键技术以及科学地进行大数据分析与处理提供参考.     

图像多标签学习的研究概述

随着图像大数据的爆发,特别是用户贡献数据的飞速增长,图像样本的语义内容越来越丰富,标签信息也随之越来越复杂.因此图像多标签学习的研究是近年来学术圈和产业界的研究热点之一,涌现了大量表现优异的方法和技术.基于此,本文将对近年来图像多标签学习上的研究成果进行总结.首先,对多标签学习进行简单介绍,并详述其主流方法的分类;随后,针对目前大数据时代的数据特性,总结了多标签学习面临的新的技术难点及其对应的解决方案;最后,在应用层面上介绍了多标签学习在医学、计算机科学等领域的应用实例.     

深度细粒度图像识别研究综述

细粒度图像分类是计算机视觉中一项基础且重要的工作,其目的在于区分难以辨别的对象类别（例如不同子类的鸟类、花或动物）.不同于传统的图像分类任务可以雇佣大量普通人标注,细粒度数据集通常需要专家级知识进行标注.除了视觉分类中常见的姿态、光照和视角变化因素之外,细粒度数据集具有更大的类间相似性和类内差异性,因此要求模型能够捕捉到细微的类间差异信息和类内公有信息.除此之外,不同类别的样本存在不同程度的获取难度,因此细粒度数据集通常在数据分布中表现出长尾的特性.综上所述,细粒度数据分布具有小型、非均匀和不易察觉的类间差异等特点,对强大的深度学习算法也提出了巨大的挑战.本文首先介绍了细粒度图像分类任务的特点与挑战,随后以局部特征与全局特征两个主要视角整理了目前的主流工作,并讨论了它们的优缺点.最后在常用数据集上比较了相关工作的性能表现,并进行了总结与展望.     

星载合成孔径雷达海洋遥感与大数据

星载合成孔径雷达以其全天候、全天时、不受云雨影响的工作特性在空间对海观测中起到了重要作用，又以其高空间分辨率、多极化、多成像模式的特点展示了其在海洋动力要素反演和海洋多尺度动力过程研究中独特的魅力.起步于20世纪70年代末的星载合成孔径雷达技术，迎来了发展的"黄金时期"，大数据和机器学习又赋予了星载合成孔径雷达海洋遥感更强大的生命力.本文首先阐述了星载合成孔径雷达大数据的5"V"特性，进而以高分辨率海面风场反演、海洋内波中尺度动力过程观测两类典型案例，阐述了大数据、机器学习等现代信息科学技术与卫星海洋遥感结合，实现海洋环境参数高精度反演和海洋动力过程科学深层次认知的研究.最后，展望了星载合成孔径雷达海洋遥感与大数据的发展前景.     

陆地生态系统固碳速率立体监测方法:进展与挑战

全球CO₂浓度增加造成的全球变暖已成为人类亟需解决的问题,陆地生态系统在过去几十年一直扮演着重要的碳汇角色,吸收了30％左右的人类活动排放CO₂。本文调研分析了陆地生态系统固碳速率空间估算方法,包括样地调查、通量监测、模型模拟、遥感估算等,梳理了各种估算方法的研究现状与进展。样地调查、通量观测等方法可以提供点尺度的固碳速率直接测量信息,但存在观测样本有限、空间代表性不足等问题。模型模拟方法可以从机理的角度描述陆地碳、水、能量循环,模拟预测陆地生态系统固碳速率的状态和变化。然而,在模型建立过程中,抽象和简化会引入结构与假设的不确定性,以及模型驱动数据引入的不确定性等问题是碳循环模型模拟方法面临的重大挑战。卫星遥感具有全球覆盖、分辨率精细、时间序列观测等优点,结合机器学习方法,为地球大数据驱动的全球碳源汇估算提供了新的研究范式。但是,当前各种固碳速率的监测方法还没有满足高度时空异质性的陆地生态系统固碳量监测需求,未来需要整合地面观测、模型模拟和卫星遥感等多种技术手段,提供区域和全球尺度的陆地生态系统碳汇精确估算方法体系和科学数据产品。     

科学大数据在全球气候变化研究中的应用

本文从气候系统观测资料、数值模拟资料、经济社会资料和土地利用资料等方面论述了全球气候变化的科学大数据的构成。为推动气候变化大数据的应用,需要进一步发展集成融合、存储共享、数字模拟和数据挖掘等新的科学大数据处理技术方法。文章还分析了科学大数据在全球气候服务框架和未来地球计划中的应用价值,并对其未来发展进行了展望。     

数值预报AI气象大模型国际发展动态研究

数值预报是研究地球系统的重要工具,有助于加深科学家对大气、海洋、气候和环境等复杂系统之间相互作用和变化过程的理解,在防灾减灾、气候变化和环境治理等方面发挥着不可或缺的作用。随着模式复杂度和分辨率的提高,传统数值模式在气候变化研究和气候预测方面取得了迅速的进展,但也面临一些挑战,需要得到数据同化、集合耦合、高性能计算和不确定性分析等多方面的支持。而近年来,“AI+气象”的交叉研究在气象领域引起了广泛关注。基于多种深度学习架构的人工智能大模型,依托强大的计算资源和海量的数据进行训练,能够以新的科学范式进行高效数值预报。气象大模型不断涌现,一些科技公司如华为、英伟达、DeepMind、谷歌、微软等,以及国内外高校如清华大学、复旦大学、密歇根大学、莱斯大学等发布了多个涵盖临近预报、短时预报、中期预报和延伸期预报等不同领域的气象大模型。这标志着人工智能与气象领域的交叉融合已经达到新的高度。尽管气象大模型在现阶段取得了较大突破,但其发展仍然面临弱可解释性、泛化能力不足、极端事件预报强度偏低、智能预报结果过平滑、深度学习框架能力需要拓展等诸多挑战。     

新书架

《气候变化与青藏高原大气水分循环》该书从青藏高原气候变化趋于暖湿化的视角出发,综合论述了气候变化对青藏高原大气水分循环机制产生的重要影响;提出了青藏高原特殊的大气水分循环结构及其概念模型;分析了影响青藏高原大气水分循环变化的驱动和调制因素;剖析了青藏高原水汽输送的变化特征及其对气候变化产生的响应;归纳总结出在气候变暖背景下,青藏高原冰川、湖泊、冻土、湿地对气候变化的响应及其对该地区水资源与生态系统的影响;提出了进行青藏高原多圈层综合观测的设计思路和实施方案,为系统地认识和理解多圈层过程总体效应提供了科学数据。另外,本书还给出了气候变暖背景下青藏高原区域气候和水资源未来趋势预估。在上述综合分析基础上,提出了一系列具有战略性意义的青藏高原气候变化应对决策建议。本书可为青藏高原科学考察和研究提供理论依据,可供大气科学工作者及相关院校师生参考。     

气候变化影响中国粮食安全研究的新思路北大核心CSCD

鉴于气候变化影响粮食安全问题的特殊性和复杂性,本文试图从自然科学和社会科学的交叉研究入手,提出一种新的研究的思路和方法,即:运用计量经济学模型对气候变化数据进行统计分析,使用计量经济学方法来评估气候这一外部驱动因素引发的社会经济系统变化与观测到的气候变化引发的社会经济系统变化之间的关系;在厘清“气候变化影响量”对粮食产量的影响的基础上,预估我国未来30年特别是经济社会发展两个关键节点2035年和2050年的粮食生产的气候变化风险,文章给出了一种新的研究视角,构建了研究内容和研究方法,力争实现定性研究与定量研究相结合,以科学预测为政策指导提供有力支撑。     

基于用户点击数据的细粒度图像识别方法概述

近年来,细粒度图像识别逐渐成为计算机视觉领域的研究热点.由于不同类别图像间的视觉差异小、语义鸿沟问题严重,传统的基于视觉特征的细粒度图像识别性能往往不尽人意.针对这些挑战,目前许多学者都在研究基于用户点击数据的图像识别.本文围绕点击数据在图像识别中数据预处理、特征提取和模型构建3大模块中的应用,总结了已有的基于点击数据的识别算法及最新的研究进展.     

中国未来地球计划的协同推广机制建设初探

本文在回顾科学成果传播方式与模式的基础上,以科学成果集成、网络传播和公众科学素养提升为出发点分析了当前科学传播面临的挑战与时代需求。基于对未来地球计划科学组织模式的解读,分析了未来地球计划成果传播模式的特点,并从科学研究、社会认知和科学决策层面探讨了科学成果推广和普及机制。有针对性地提出了充分引入利益相关者参与、推进创新科学成果推广机制与平台建设等科学传播相关政策建议。     

大规模图中低复杂度分布式算法浅析

近年来大规模图分析问题在网络大数据领域发挥着重要作用.经典的图分析问题包括求图的直径、半径、围长、聚类系数、紧密中心度和介数中心度等.集中式算法求解这些图计算问题一般都需要问题规模的平方甚至立方以上复杂度,显然不适用于大规模图.本文旨在从分布式算法角度介绍对这些基本图计算问题具有最坏性能保证的低复杂度（线性时间）算法.此外,本文还将介绍如何通过通信复杂性理论证明分布式图计算问题的下界.     

人工智能技术气候预测应用简介

近年来,随着人工智能技术在多个领域大数据分析中的应用,许多研究工作者尝试将地学研究与人工智能跨学科结合,取得了很多新的进展,推动了地球科学的发展。其中气候预测与人类生活以及防灾减灾等息息相关,准确的气候预测至关重要。本文简要总结了人工智能技术在气候预测应用方面的研究进展,包括资料同化、模式参数化、求解偏微分方程、构建统计预测模型、改进数值模式产品释用等领域。这些研究证明了利用人工智能提高气候预测技巧的可能性和适用性,可以极大地节省计算成本和时间。然而人工智能应用也存在诸多挑战,例如数据集的构建、模型的适用性和物理可解释性等问题,对这些难点问题的研究和攻克,可以让人工智能在大数据时代中更好地补充传统地球科学方法,产生更多有益的效应,极大地改进气候预测水平。     

Operational climate prediction in the era of big data in China: Reviews and prospects

Big data has emerged as the next technological revolution in IT industry after cloud computing and the Internet of Things. With the development of climate observing systems, particularly satellite meteorological observation and high-resolution climate models, and the rapid growth in the volume of climate data, climate prediction is now entering the era of big data. The application of big data will provide new ideas and methods for the continuous development of climate prediction. The rapid integration, cloud storage, cloud computing, and full-sample analysis of massive climate data makes it possible to understand climate states and their evolution more objectively, thus predicting the future climate more accurately. This paper describes the application status of big data in operational climate prediction in China; it analyzes the key big data technologies, discusses the future development of climate prediction operations from the perspective of big data, speculates on the prospects for applying climatic big data in cloud computing and data assimilation, and puts forward the notion of big data-based super-ensemble climate prediction methods and computerbased deep learning climate prediction methods.     

Operational Climate Prediction in the Era of Big Data in China:Reviews and Prospects

Big data has emerged as the next technological revolution in IT industry after cloud computing and the Internet of Things.With the development of climate observing systems,particularly satellite meteorological observation and high-resolution climate models,and the rapid growth in the volume of climate data,climate prediction is now entering the era of big data.The application of big data will provide new ideas and methods for the continuous development of climate prediction.The rapid integration,cloud storage,cloud computing,and full-sample analysis of massive climate data makes it possible to understand climate states and their evolution more objectively,thus predicting the future climate more accurately.This paper describes the application status of big data in operational climate prediction in China;it analyzes the key big data technologies,discusses the future development of climate prediction operations from the perspective of big data,speculates on the prospects for applying climatic big data in cloud computing and data assimilation,and puts forward the notion of big data-based super-ensemble climate prediction methods and computerbased deep learning climate prediction methods.     

Integrating knowledge to assess coastal vulnerability to sea-level rise: The development of the DIVA tool

This paper describes the development of the DIVA tool, a user-friendly tool for assessing coastal vulnerability from subnational to global levels. The development involved the two major challenges of integrating knowledge in the form of data, scenarios and models from various natural, social and engineering science disciplines and making this integrated knowledge accessible to a broad community of end-users. These challenges were addressed by (i) creating and applying the DIVA method, an iterative, modular method for developing integrating models amongst distributed partners and (ii) making the data, scenarios and integrated model, equipped with a powerful graphical user interface, directly and freely available to end-users.     

Citizens, consumers and sustainability: (Re)Framing environmental practice in an age of climate change

Recent moves by national and local policy makers have sought to encourage individuals to engage in a wide range of pro-environmental practices to address both discrete environmental problems and major, global challenges such as climate change. The major framing device for these developments is the notion of ‘citizen–consumers’, which positions individual ecological responsibilities alongside consumer choice logics in a Neo-liberal socio-economic framework. In the environmental social sciences, there have been recent moves to interpret the citizen–consumer through adopting a social practices approach, which advances the notion that in understanding environmental commitments, a deeper appreciation of underlying norms, values, identity politics and consumption is required to uncover the complex processes that lead to environmental practices in specific contexts. This paper argues that whilst these approaches have considerable utility in tracing the normalisation of established and discrete environmental practices in particular contexts, the issue of climate change represents an independent and over-arching discursive conflict between new and embedded practices that challenges the ability of citizen–consumers to act as agents for change. Accordingly, the data presented in this paper suggest that climate change can be seen as an unsettling and dynamic issue that generates discursive conflict in its own right around fundamental issues of knowledge, responsibility, scale and place. The paper therefore argues that a new and more critical perspective is required within environmental social science to understand (conflicting) discourses of sustainable living between the ‘passive’ normalisation of conventional environmental practice and the ‘contested’ ambiguities of climate change.     

Towards practitioner-initiated interactive knowledge development for sustainable development: A cross-case analysis of three coastal projects

Although the central challenges of sustainable development are well-known, sustainability science has been slow in contributing to effective and feasible solutions for sustainable development. Turning knowledge into action for sustainable development therefore remains a major challenge for sustainability science. Interactive knowledge development is considered a prerequisite for sustainability-oriented action. Most studies approach interactive knowledge development from a researcher's perspective. This paper focuses on practitioners that initiate interactive knowledge development for sustainability-oriented actions. A cross-case analysis is presented for interactive knowledge development in coastal projects. Three cases are analysed through the framework of project arrangements and knowledge arrangements. The projects are located in the Wadden Sea, San Francisco Bay and the Ems estuary and address issues of flood control, nature restoration and liveability. The cross-case analysis revealed 11 causal mechanisms that help explain how project decision-making impacts on interactive knowledge development, how a process of interactive knowledge development functions and what its outcomes are. The mechanisms clarify the key underlying processes of interactive knowledge development in coastal projects. The mechanisms show that interactive knowledge development may result in sustainability-oriented solutions that are feasible for implementation. As such, this paper contributes to a practice-oriented understanding of turning knowledge into action for sustainable coastal development.     

Linking global and local scales: designing dynamic assessment and management processes

This paper identifies challenges inherent in addressing multi-scale environmental problems, and outlines tentative guidelines for addressing such challenges and linking science and policy across scales. The study and practice of environmental assessment and management increasingly recognize the importance of scale and cross-scale dynamics in understanding and addressing global environmental change. These ongoing efforts, however, lack a systematic way of thinking about and addressing the challenges involved in integrating science and policy across multiple scales, for example, in the design of policy-relevant, scientific assessments of problems such as climate change. These challenges include matching scales of biogeophysical systems with scales of management systems, avoiding scale discordance (matching the scale of the assessment with the scale of management), and accounting for cross-scale dynamics. In this paper we propose tentative guidelines for meeting such challenges for both assessors and decision-makers: (1) utilize boundary organizations — institutions which serve to mediate between scientists and decision-makers, and between these actors at different scales; (2) utilize scale-dependent comparative advantages — coordinating the allocation of resources, technical expertise, and decision-making authority to best capitalize on scale-specific capabilities; and (3) employ adaptive assessment and management strategies — constructing long-term, iterative, experiment-based processes of integrated assessment and management.