Ocean observatories, exemplified by the NSF Ocean Observatories Initiative (OOI), aim to transform oceanography from an expeditionary to an observation-based science. To do so, new cyberinfrastructure environments are helping scientists from disparate fields jointly conduct experiments, manage large collections of instruments, and explore extensive archives of observed and simulated data. However, such environments often focus on systems, networking, and databases and ignore the critical importance of rich 3D interactive visualization, asset management, and collaboration needed to effectively communicate across interdisciplinary science teams.

This dissertation presents the design, implementation, and evaluation of an interactive ocean data exploration system designed to satisfy the unmet needs of the multidisciplinary ocean observatory community. After surveying existing literature and performing a multi-month contextual design study that included input from scientists at multiple institutions, I propose a set of guidelines for the system's user interface and design. Motivated by these guidelines and informed by close collaboration with multidisciplinary ocean scientists, I then present the Collaborative Ocean Visualization Environment (COVE), a new data exploration system that combines the ease of use of geobrowsers, such as Google Earth, with the data exploration and visualization capabilities of sophisticated science systems.

To validate COVE'S design, I evaluated its capabilities in three ways. (1) User studies showed that it works efficiently for expert and novice data explorers as well as visualization producers and consumers. (2) Multiple real-world science deployments, both on land and at sea, saw it replace existing systems for observatory design, provide faster and more engaging planning and data analysis for science teams, and enhance mission preparation and navigation for the ALVIN research submarine. (3) An analysis of COVE over local, server and cloud-based resources indicated that its flexible work partitioning architecture is essential for real-world observatory data analysis and visualization tasks.