A GPU-accelerated simulation framework for running large ensembles of simplified ocean models for real-world domains.

Operational ocean forecasting models are computationally expensive and are therefore often run as a single deterministic simulation at given intervals. This means that they often lack information about forecast uncertainties, which are significant given that there are relative few observations of the ocean compared to its sheer size. Information on uncertainty is, however, vital for various applications, including search-and-rescue operations at sea.

What is GPU Ocean?

GPU Ocean is a simulation framework designed to explore the short-term uncertainty in forecasted ocean currents. It achieves this by running ensembles of simplified ocean models nested within operational ocean forecasts. These simplified models efficiently solve barotropic dynamics described by the shallow-water equations using GPUs.

Key Features:

• Efficient simulation: GPU Ocean utilizes GPU acceleration and state-of-the-art finite-volume methods to solve shallow-water equations efficiently.
• Model nesting: Easily import initialization data, terrain information, forcing, and boundary conditions from NetCDF files containing operational 3D ocean forecasts.
• Drift trajectory prediction: Conduct online drift trajectory simulations for drifting objects within the ocean models, directly assessing uncertainty in drift trajectories based on the uncertainty in the ocean currents.
• Data assimilation: GPU Ocean includes tailored data-assimilation methods for sparse in-situ observations.
• Python with CUDA performance: Rapid prototyping and easy orchestration, pre-, and post-processing using Python while getting the computational performance of CUDA.
• Parallel processing: MPI support for running even larger ensembles on multiple GPUs.

GPU Ocean aims to be a powerful tool to complement ocean current forecasts through estimating and accounting for forecast uncertainties. By combining local observations and advanced data assimilation methods, users can make more informed decisions based on the latest oceanic data.

Installation

See here.

Academic publications using GPU Ocean

• F. Beiser, H. Holm, M. Sætra, N. Kristensen, K. Christensen (2024) Combining Barotropic and Baroclinic Simplified Models for Drift Trajectory Predictions, Journal of Operational Oceanography, 17(3), 187-206. DOI: 10.1080/1755876X.2024.2364975
• F. Beiser, H. Holm, J. Eidsvik (2024) Comparison of Ensemble-Based Data Assimilation Methods for Sparse Oceanographic Data, Quarterly Journal of the Royal Meteorological Society, 150(759), 1068–1095. DOI: 10.1002/qj.4637 [Preprint: arXiv:2302.07197].
• H. Holm, F. Beiser (2023) Reducing Numerical Artifacts by Sacrificing Well-Balance for Rotating Shallow-Water Flow. In: Franck, E., Fuhrmann, J., Michel-Dansac, V., Navoret, L. (eds) Finite Volumes for Complex Applications X — Volume 2, Hyperbolic and Related Problems. FVCA 2023. Springer Proceedings in Mathematics & Statistics, vol 433. Springer, Cham. DOI: 10.1007/978-3-031-40860-1_19
• A. Brodtkorb, H. Holm (2021) Coastal Ocean Forecasting on the GPU using a Two-Dimensional Finite Volume Scheme. Tellus A: Dynamic Meteorology and Oceanography, 73:1, 1-22, DOI: 10.1080/16000870.2021.1876341 [Preprint: arXiv:1912.02457]
• H. Holm, A. Brodtkorb, M. Sætra (2020) Data Assimilation for Ocean Drift Trajectories Using Massive Ensembles and GPUs. In: Klöfkorn, R., Keilegavlen, E., Radu, F.A., Fuhrmann, J. (eds) Finite Volumes for Complex Applications IX - Methods, Theoretical Aspects, Examples. FVCA 2020. Springer Proceedings in Mathematics & Statistics, vol 323. Springer, Cham. DOI: 10.1007/978-3-030-43651-3_68
• H. Holm, M. Sætra, P. van Leeuwen (2020) Massively parallel implicit-equal weights particle filter for ocean drift trajectory forecasting. Journal of Computational Physics: X, volume 6, 100053. DOI: 10.1016/j.jcpx.2020.100053 [Preprint: arXiv:1910.01031]
• H. Holm, A. Brodtkorb, K. Christensen, G. Broström, M. Sætra (2020) Evaluation of selected finite-difference and finite-volume approaches to rotational shallow-water flow. Communications in Computational Physics, volume 27, pp. 1234-1274. DOI: 10.4208/cicp.OA-2019-0033
• H. Holm, A. Brodtkorb, M. Sætra (2020) GPU Computing with Python: Performance, Energy Efficiency and Usability. Computation, volume 8, number 1:4 (Special issue on Energy-Efficient Computing on Parallel Architectures). DOI: 10.3390/computation8010004. [Preprint: arXiv:1912.02607]
• H. Holm, A. Brodtkorb, M. Sætra (2020) Performance and energy efficiency of CUDA and OpenCL for GPU computing using Python. Advances in Parallel Computing, volume 36, pp. 593-604. DOI: 10.3233/APC200089

Preprints

• F. Beiser, H. Holm, K. Lye, J. Eidsvik (2024) Multi-level Data Assimilation for Simplified Ocean Models, preprint in Nonlinear Processes in Geophysics npg-2023-27

Development and funding

GPU Ocean is developed through a collaboration between the Norwegian Meteorological Institute and the Applied Computational Science research group at SINTEF Digital. We are greatful for the support from the Norwegian Research Council under grant numbers 250935 (GPU Ocean) and 310515 (Havvarsel).