The marine Carbon Dioxide Removal (mCDR) industry faces a fundamental challenge: traditional ocean modeling approaches take months to set up, cost hundreds of thousands of dollars to run, and still can’t capture the complex physics that determine whether carbon removal actually works. For mCDR developers, this creates an impossible choice—either wait years for modeling capabilities that might arrive too late, or move forward without the quantitative understanding needed for credible measurement, reporting, and verification (MRV).
But what if that choice is already obsolete?
The Bedford Basin Breakthrough
This week, our postdoctoral researcher Erik Johnson, co-advised by Prof. Ruth Musgrave at Dalhousie University, setup and ran a realistic simulation of Bedford Basin, achieving a 40-meter horizontal resolution with 40 vertical levels, capturing estuarine circulation dynamics. The entire simulation runs in just 46 minutes per simulated day on a single NVIDIA A100 GPU.
Real-time visualization: Sea surface temperature and salinity dynamics in Bedford Basin at 40m resolution
To put this in perspective: traditional CPU-based models like ROMS or MITgcm would require months of setup time and massive computational clusters to achieve similar resolution. Erik’s model captures the same dynamics on hardware that costs less than a used car.
Speed Meets Physics
At 1/2000° resolution with sub-minute temporal steps, this isn't just faster modeling—it's modeling that can finally capture the turbulent mixing, stratification, and buoyancy-driven processes that determine how alkalinity additions actually behave in real coastal environments.
Why Bedford Basin Matters for mCDR
Bedford Basin isn’t just a convenient test case—it’s a perfect representation of the complex coastal environments where mCDR operations need to work. The basin exhibits the challenging physics that make or break carbon removal efficiency: estuarine circulation driven by freshwater inputs, stratification dynamics that control mixing depths, and variable exchange with the open ocean that determines residence times.
These are exactly the processes that simplified models miss and that empirical approaches can’t generalize across locations. Models that capture these dynamics are not only for academic curiosity—it’s the difference between verified carbon removal and expensive guesswork.
What Makes This Operational
GPU-Native Architecture: Built on Oceananigans.jl, achieving 100x speedup over legacy CPU models while maintaining full physics
Flexible Configuration: Days to adapt for new regions vs. months for traditional approaches
Integrated Biogeochemistry: OceanBioME coupling enables direct carbon cycle tracking
Scalable Resolution: From regional planning (500m) to near-field processes (5m) in unified framework
From Research to Operations
Erik’s collaboration between Dalhousie University and atdepth demonstrates how breakthrough academic research becomes operational capability. Prof. Musgrave’s extensive Bedford Basin datasets provide the validation framework that gives these models credibility with carbon markets.
But computational speed alone isn’t enough. atdepth’s infrastructure transforms breakthroughs in mocean modeling into operational services that can spin up quickly for new sites, scale economically across multiple operations, and provide quantitative certainty that meets carbon credit standards.
The Platform Advantage: Our cloud-based Ocean Digital Twin (ODT) infrastructure initializes high-resolution simulations for new locations within days. The single-GPU efficiency enables thousands of concurrent operationsm, continuously integrating monitoring data to provide real-time validation. Comprehensive ocean modeling is now economically viable for continuous use.
Performance Metrics
- Resolution: 40m horizontal, 40 vertical levels
- Speed: 46 minutes per simulated day on single A100 GPU
- Setup: Days vs. months for traditional approaches
- Cost: ~100x reduction in computational requirements
Research Collaboration
The End of the Modeling Bottleneck
Erik’s Bedford Basin work proves that high-resolution, physics-based ocean modeling can run fast enough, cheap enough, and flexibly enough to serve operational needs. The question isn’t whether mCDR operations can afford comprehensive ocean modeling—it’s whether they can afford to operate without it.
GPU-powered ocean models running on cloud infrastructure deliver the speed, accuracy, and economic efficiency that mCDR operations require. For developers, this means access to quantitative ocean intelligence previously available only to research institutions—delivered as an operational service. For carbon markets, it means verification capabilities that scale with industry growth while maintaining scientific rigor.
Most importantly, it means removing the modeling bottleneck that has constrained mCDR development and enabling the rapid, scientifically-grounded scaling our climate timeline demands.
Ready for Commercial Deployment
atdepth is already deploying Digital Twin solutions around the world, at any scale, spinning up comprehensive modeling capabilities in days rather than months and providing our customers with the metrics, answers, and data they need.
The future of operational ocean intelligence is here.
Learn more about how atdepth’s Ocean Digital Twin platform can support your mCDR operations at atdepth.org