DMUU: Statistical Disclosure Limitation for Geospatial Image Data

为了可用于科学和政策目的，在社会科学和自然科学中无处不在的遥感数据，如卫星图像，必须给予其他形式的数据长期提供的同样的保密保护。该项目将制定地理空间图像数据统计披露限制(SDL)的初步公式和技术，使用纵向分析来框定问题，并作为指向试验台数据集的指针。特别是，这项研究将探讨现有的SDL风险-效用公式、现有的记录链接技术以及现有的计算方法和系统，特别是地理信息系统(地理信息系统)，能够在多大程度上产生直接影响。在这一过程中，项目组将开发新的抽象概念和技术，以便向多个利益攸关方提供记录联系、披露风险和数据利用。将创建可扩展的数据结构和算法，以处理复杂的高维图像数据。应对气候变化的合理决策的公式、实施和评估在很大程度上依赖于对地理空间图像数据的纵向分析，在决策前(问题是否变得更糟？)以及决策后阶段(政策是否奏效？)然而，许多这样的数据是在没有征得土地所有者等数据主体的同意甚至不知情的情况下收集的。为了使获取图像数据的科学益处超过隐私侵犯，必须保持机密性。这项研究将提供能够有效使用纵向图像的技术，同时保护隐私，并适应图像与传统-数字或分类-数据之间的深刻差异。该项目将对与地理空间图像数据相关的隐私问题有更深入的了解，并作为未来研究的基础。这一发展奖作为2003财政年度人类和社会动力优先领域不确定性决策特别竞赛的一部分得到支持。

In order to be usable for scientific and policy purposes, remotely sensed data such as satellite images, which have become ubiquitous in the social and natural sciences, must be accorded the same confidentiality protection long provided to other forms of data. The project will produce initial formulations and techniques for statistical disclosure limitation (SDL) for geospatial image data, using longitudinal analyses to frame questions and as pointers to testbed data sets. In particular, the research will address the extent to which existing risk-utility formulations for SDL, existing techniques for record linkage and existing computational methods and systems, especially geographical information systems (GIS), can be adapted to yield immediate impact. In the process, the project team will develop new abstractions and techniques for record linkage, disclosure risk and data utility to multiple stakeholders. Scalable data structures and algorithms will be created that can handle complex, high-dimensional image data.Formulation, implementation and evaluation of sound decisions to deal with climate change depend significantly on longitudinal analyses of geospatial image data, at both the pre-decision (Is the problem getting worse?) and post-decision stages (Is the policy working?). Much such data is, however, collected without the consent or even the knowledge of data subjects such as landowners. In order that the scientific benefits of capturing image data outweigh privacy intrusions, confidentiality must be maintained. The research will provide techniques that enable effective use of longitudinal image while preserving privacy and accommodating profound differences between images and traditional---numerical or categorical---data. The project will yield a deeper understanding of privacy issues associated with geospatial image data, and serve as the basis for future research. This developmental award was supported as part of the Fiscal Year 2003 Human and Social Dynamics priority area special competition on Decision Making Under Uncertainty (DMUU).