Ocean Networks Canada:How a 1.7-petabyte ocean observatory is using AI to hear what the ocean is telling us
What they're doing
Ocean Networks Canada (ONC) operates observatories on Canada's three coasts and in Antarctica, generating 1.7 petabytes of continuous sensor data that grows daily. With $2.5 million from Canada's Ocean Supercluster through the Pan-Canadian Artificial Intelligence Strategy, ONC is building AI tools to help scientists and decision-makers work with data at that scale: automated data quality tools, AI-driven anomaly detection, acoustic event identification, and video data processing that no human team could manage manually.
ONC is a core partner in Building Bridges, a multi-institution project led by Dalhousie University alongside CIOOS, the St. Lawrence Global Observatory, and the Hakai Institute. The collaboration targets the foundational work most AI projects skip: creating AI-ready ocean data — standardized, quality-assessed, annotated datasets that machine learning models can actually train on. Drew Snauffer, ONC's data analytics and quality manager, has been central to developing the quality-assurance pipeline that makes this possible.
In November 2025, ONC co-hosted the Understanding and Predicting the Ocean Using Artificial Intelligence workshop in Halifax, which convened 137 participants from 65 organizations across Canada and 11 other countries. The gathering surfaced shared bottlenecks — annotation gaps, sensor heterogeneity, the challenge of building models that generalize across different hardware configurations — and accelerated coordination across the CIOOS network.
Why it matters
Canada has the world's longest coastline and ocean health is directly tied to fisheries, climate systems, and northern communities. ONC's bet is that the path to better ocean science runs through better AI infrastructure — and that you can't shortcut the foundational work of data quality and interoperability. The tools they're building with Building Bridges partners aren't siloed inside ONC; they're designed to work across the CIOOS network, which means every investment in the shared infrastructure compounds across institutions.
What's working
- The data scale is an asset: 1.7 petabytes of continuous, multi-coast ocean sensor data is the kind of training resource most AI researchers can't access — making ONC's observatory network a natural home for applied ocean AI
- Federating with CIOOS, Hakai, and St. Lawrence through Building Bridges rather than building in silos means AI tools are designed for interoperability from day one
- Open convening works: the November 2025 workshop surfaced bottlenecks shared by 65 organizations simultaneously, compressing what would otherwise be years of bilateral conversations into a single event
What's harder than it looks
- Ocean sensor data is messy by nature — saltwater, pressure, biofouling, and hardware variability create quality problems that make it harder to build robust AI models than comparable land-based datasets
- Annotation at scale is the hidden bottleneck: training AI to identify species, acoustic events, or anomalies requires labeled examples, and labeling 1.7 petabytes is a human-labor problem before it's an AI problem
- The gap between research-grade AI tools and operational tools for fisheries managers or marine regulators is wide — bridging it requires design decisions that aren't typical in academic AI development
Three takeaways for similar orgs
- 01Data quality before AI deployment: ONC's investment in AI-ready data infrastructure — quality tools, annotation pipelines, standardized formats — is the prerequisite most organizations skip, and the reason most AI projects in data-rich environments fail
- 02Scale creates economic justification: 1.7 petabytes of sensor data makes AI investment straightforwardly warranted in a way it isn't for smaller datasets — the same data volumes that make manual analysis impossible make AI the only practical option
- 03Co-investing in shared data standards across institutions (CIOOS, Hakai, St. Lawrence) compounds every individual AI investment — a transferable model for any sector with distributed data collection