Advancing mesoscale process representation in ocean models through machine learning

go through: PhD. Anna Danville-Sommer

Mesoscale eddies are whirlpool-like motions that play a crucial role in ocean circulation and the global energy budget. On a scale of 10 to 300 kilometers, These dynamic features transmit hydrologic properties (physical and chemical properties of seawater, such as temperature and salinity) and redistribute energy across spatial and temporal scales, affecting large-scale ocean dynamics and biogeochemical processes. Representing their effects in ocean models is critical for accurate long-term climate predictions, as mesoscale eddies influence sunlight reaching deeper sea levels, ecosystem health, and climate feedbacks [Olbers et al., 2012].

However, achieving a balance between resolution and computational efficiency remains challenging. High-resolution ocean models are capable of resolving mesoscale eddies (so-called allowed eddies), but are computationally expensive (CPU time and storage) and are limited by numerical stability requirements, such as high viscosity and dissipation, and Reynolds neglecting eddies Constrained Reynolds stress [Zanna et al., 2017]. In turn, the coarse-resolution models used in climate and Earth system models cannot explicitly represent mesoscale processes. This emphasizes the need for innovative approaches to improve their representation without excessive computational cost.

Machine learning (ML), especially deep learning, offers a promising solution. By leveraging large data sets, machine learning can reconstruct missing information across different spatial and temporal scales. For example, machine learning techniques can now be used to model processes that are not visible to satellites or cannot be parsed by crude models. A notable example is the use of deep neural networks Represents all subgrid atmospheric processes in climate models and successively replaces traditional subgrid parameterizations in global general circulation models [Rasp et al., 2018].

One of the driving mechanisms for the emergence of mesoscale features in the ocean is baroclinic instability (density-driven flow instability), especially in winter [Boccaletti, et al., 2007; Capet et al., 2008; Fox-Kemper and Ferrari, 2008; Mensa et al., 2013; Oiu et al., 2014; Sasaki et al., 2014]. This instability can be captured using the eddy buoyancy flux (you'b'), which quantifies the correlation between 3D velocity and tracer anomalies.

There are several studies dedicated to addressing this issue (Bolton & Zanna, 2019; Zanna & Bolton, 2020; Guillaumin & Zanna, 2021; Bodner et al., 2024). In our work, we first focus on the vertical component of the buoyancy flux w′b′. To address these challenges, we tested a 3D convolutional neural network (3DCNN) approach to reconstruct w′b′ from large-scale ocean variables (temperature, salinity, and velocity). The model is trained on the output of the eNATL60 simulation, a high-resolution regional configuration of NEMO (Common Model of Global Ocean Circulation) covering the North Atlantic with a horizontal resolution of 1/60°x1/60° and 300 vertical levels. This high-resolution dataset clearly resolves mesoscale processes and provides an ideal basis for training. 3DCNN connects coarse-resolution inputs to mesoscale fluxes. Testing consisted of averaging eNATL60 output to simulate various coarse resolutions to create a robust training library.

Figure 1. Vorticity field of eNATL60, zooming in on the Gulf Stream region with mid- to mid-scale activity. The grid on the zoom plot is 1°x1° to show mesoscale features that may be missed in heading resolution models.

By improving the characterization of mesoscale processes (Fig. 1), this work paves the way for enhanced biogeochemical and physical modeling in climate simulations. Once validated in other ocean regions, this method will greatly improve the accuracy of coarse-resolution global ocean and climate models. Watch the scientific literature over the next year to learn about the results of this program as it progresses.

refer to

1. Boccaletti, G., Ferrari, R., and Fox-Kemper, B. Mixed-layer instability and restratification, Journal of Physical Oceanography, 37(9): 2228-50, doi: 10.1175/JPO3101.1, 2007.
2. Bodner, A., Balwada, D., Zanna, L. Data-driven methods for parameterizing submesoscale vertical buoyancy fluxes in the ocean mixed layer, arXiv preprint arXiv:2312.06972, https://arxiv.org/abs /2312.06972, 2023.
3. Bolton, T. and Zanna, L. Application of deep learning to ocean data inference and subgrid parameterization, J. Adv. Model. Earth Systems, 11, 376–399, doi: 2019.
4. Capet, X., Campos, EJ and Paiva, M. Submesoscale activity over the Argentine continental shelf, Geophysical Research Letters, 35, 2-6, doi:10.1029/2008GL034736, 2008.
5. Fox-Kemper, B. and Ferrari, R. Parameterization of mixed-layer vortices. Part 2: Predictions and impacts, Journal of Physical Oceanography, 38, 1166-79, doi: 10.1175/2007JPO3788.1, 2008.
6. Guillaumin, AP and Zanna, L. Stochastic deep learning parameterization of ocean momentum forcing. Journal of Advances in Earth System Modeling,13(9), e2021MS002534, doi: https://doi.org/10.1029/2021MS002534, 2021.
7. Mensa, JA, Z Garraffo, Z., Griffa, A., Ozggokmen, TM, Haza, A., and Veneziani, M. Seasonality of submesoscale dynamics in the Gulf Stream region, Ocean Dynamics, 63, 923-41 , doi: https://doi.org/10.1007/s10236-013-0633-1, 2013.
8. Olbers, D., Willebrand, J., Eden, C. Ocean Dynamics, Springer, Heidelberg, 2012.
9. Rasp, S., Pritchard, MS, and Gentine, P. Deep learning to represent subgrid processes in climate models, PNAS 115 (39), 9684-9689, https://doi.org/10.1073/pnas.1810286115, 2018 .
10. Sasaki, H., Klein,P., Qiu B., andSasa, Yi.. Impact of ocean-scale interactions on atmospheric seasonal regulation of ocean dynamics, Nature Communications, 5, 5636 https://doi.org /10.1038/ncomms6636, 2014.
11. Zanna, L. and Bolton, T.. resource. Letters, 47, e2020GL088376,https://doi.org/10.1029/2020GL0883762020.
12. Zanna, L., P. P. Mana, J. Anstey, T. David, and T. Bolton. Scale-aware deterministic and stochastic parameterization of eddy-average current interactions, Ocean Modeling, 111, 66–80, https://doi.org/10.1016/j.ocemod.2017.01.004, 2017.