Lead Investigator

Christine Borgman

Overview

Our research program on data management has two overarching goals: (1) research on data practices to understand how data are created, used, and reused throughout the information lifecycle and (2) employ the results of our data practices research for research and development of systems to capture and manage scientific data in ways that will facilitate immediate use by the data creators and later reuse by the creators and others, and which will reflect fair policies for access. Multidisciplinary collaboration, which is among the great promises of e-Science, depends heavily on the ability to share data within and between fields.  Research in terrestrial ecology and marine biology are much different than research in computer science and engineering, for example, and thus these collaborators vary widely in data practices.  We need to understand the nature of “data” in these respective areas if we are to understand how to capture and manage them in ways that are deemed useful and fair by the user communities. We also need to identify requirements for architecture and policy within each area, and to determine where architecture meets policy.

Figure 1: CENS data variation organized by collection method and use. As shown in figure 1, data from CENS dynamic field deployments can be grouped into four types.  Sensors are used to collect data on the scientific application, on the performance of the sensors themselves, or – for robotic sensor technology – proprioceptive data about the world to use in navigation.  The fourth category is hand-collected data for the scientific application, such as water samples.  Each of the four data types has multiple variables; these are examples from a longer list.  Some data serve only one purpose, but most serve multiple purposes as illustrated by the intersecting sets.

Our research questions on data practices address the initial stages of the information life cycle in which data are captured, and subsequent stages in which the data are cleaned, analyzed, published, curated, and made accessible. The questions can be categorized as follows:

• Data characteristics: What data are being generated? To whom are these data? To whom are these data useful? 
• Data sharing: When will scientists share data? With whom will they share data? What are the criteria for sharing? Who can authorize sharing? How do data sharing criteria vary between scientific and educational applications?
• Data policy: What are fair policies for providing access to these data? What controls, embargoes, usage constraints, or other limitations are needed to assure fairness of access and use? What data publication models are appropriate?
• Data architecture: What data tools are needed at the time of research design? What tools are needed for data collection and acquisition? What tools are needed for data analysis? What tools are needed for publishing data? What data models do the scientists who generate the data need? What data models do others need to use the data?

Figure 2: Life cycle of CENS data. A first step in developing digital library tools and services to support the data life cycle is to identify the stages in that cycle.  We have identified eight stages that appear to be common to the CENS deployments studied and to the resulting data, as shown in Figure 2.  The order of the steps is not absolute, as some stages are iterative.

Our vision for an interoperable data framework is directed not only to the description of sensor data, but to the entire data production life cycle. We anticipate the development of a "fabric" capable of connecting data resources at different stages of their life cycle: from their inception to their storage and dissemination (figure 2). The most promising approach is to participate in the development of the Open Archives Initiative - Object Reuse and Exchange (OAI-ORE, http://www.openarchives.org/ore/) architecture, which is led by the Digital Library Research Group at the Los Alamos National Laboratory. This infrastructure envisions full connection and interaction between heterogeneous data resources via their abstraction to a shared data layer. CENS, holding diverse data such as sensor datasets, deployment descriptions, images and publications, proves an ideal testbed for the development of such a framework for interoperability, as shown in figure 2 (adapted from Warner et al. Pathways: Augmenting interoperability across scholarly repositories. Accepted for IJDL Special Issue on Digital Libraries and eScience).

Figure 3: OAI-ORE framework for interoperability. A layer for interoperability does not aim to standardize the internal formats used by heterogeneous data repositories; rather it interfaces with diverse formats via a shared data model. This model allows digital objects to be represented as "surrogates" and moved across the layer via specific service protocols (obtain, harvest, put). In the figure, five possible data repositories have been represented: 1) OAIster - a metadata aggregator and provider, 2) ENS repository - a possible external OAI-compliant repository of ENS literature, 3) Biblio DB - CENS's very own OAI-compliant publication repository, 4) CENS DC- the Deployment Center, 5) SensorBase.

For heterogeneous digital objects to be capable of cross-linking, referencing and including each other, they are required to be stored in a format and accessible via a protocol that is universally recognized. Thus, the first step of our workplan is to ensure that all CENS data resources, namely the sensor datasets, the deployment description and the publications, comply with this requirement. Concerning datasets, we are working in collaboration with the Sensorbase project, CENS's repository of sensor data, to introduce the possibility of exporting and importing data via a shared data model. Given our interview study on data practices we have developed a set of scenarios for a CENS data repository, which we can use to perform a human factors analysis to evaluate the functionality of Sensorbase with respect to how the application researchers need for Sensorbase to function. Similarly, for deployment information, we are closely working with the nascent CENS Deployment Center, to ensure that project, sensor and deployment descriptions are expressed using standard formats.

As for the publications, an interoperable repository of CENS literature does not currently exist. We intend to collaborate with existing archive projects hosted by the California Digital Library, in particular the UC-wide eScholarship Repository (http://repositories.cdlib.org/escholarship/) to make CENS and ENS-related publications become both usable within the ORE framework and more visible. The publication repository should be operational by the end of March 2007. Once set up, all metadata, fulltext and related datasets in the archive will be openly harvestable via an ad-hoc protocol Open Archives Initiative - Protocol for Metadata Harvesting (OAI-PMH, http://www.openarchives.org/pmh/). Ultimately, the publications themselves will act as containers of data, or more specifically, digital "surrogates" of the datasets. The datasets as well will be annotated in such a way that they will refer back to specific deployments and sensor instrumentation, making the entire data life cycle more intelligible and trustworthy. In the long run, we are exploring the use of author and researcher data (manually input when submitting deployment or publication information) to automatically populate a CENS people directory.

Figure 4: Interlinking databases for describing CENS data. In order to capture necessary information for the interpretation of sensor data we need to describe the context surrounding the data: the sensing equipment, the deployment, the dataset, the people responsible, and the publication. Each of these pieces plays a part of the story of the data. As indicated in Figure 4 we currently have or are working towards fully functional deployment, dataset, and publication description, while still lacking in adequate description of sensor and personal information.

Accomplishments

• Analysis of interview data, resulting in publications and talks
• Auditing data formats and standards for possible application
• Development of the CENS Deployment Center proposal
• Analysis of the current CENS bibliographic database functionality
• Proposed a new bibliographic database solution
• Developed behavioral models of data practices (figures 1 & 2)
• Developed functional requirements for the organization, storage and maintenance of sensor data
• Developed initial visualization for CENS social networks based on bibliographic data
• Human factors analysis of SensorBase as a data repository (in progress)

Future Plans

• Develop CENS Directory and Sensor Registry pieces (figure 4)
• Port over bibliographic database contents into the eScholarship Repository
• Continue auditing emerging standards and their appropriateness for CENS
• Application of interview study conclusions to CENS initiatives
• Continue human factors analysis of CENS data interfaces
• Develop visualizations for understanding bibliographic data and CENS social networks based on publications

People

Faculty:

• Christine Borgman, Information Studies, UCLA

Staff:

• Jillian Wallis, CENS, UCLA

Graduate Students:

• Alberto Pepe, UCLA
• Matthew Mayernik, UCLA

External Research Partnerships

• Microsoft – Secured a gift for the hire of another graduate student researcher, and in talks with regards to a more formal collaboration.
• Open Geospatial Consortium – (in process) We joined the OGC Technical Committee and plan to develop a proposal for an interoperability experiment between CENS, the WATERS network and other institutions such as the EPA for the institution of a data standard with regards to sensing of water quality measures.
• Open Archives Initiative –  (planned) Enter CENS as a testbed for the implementation of OAI harvestable metadata and data