LLC4320 + machine-learning studies

Goh, E., Yepremyan, A., Wang, J., & Wilson, B. (2024). MAESSTRO: Masked Autoencoders for Sea Surface Temperature Reconstruction under Occlusion. Ocean Science, 20, 1309–1323. (https://doi.org/10.5194/os-20-1309-2024)
Trains a masked autoencoder / ViT-style model on LLC4320 SST tiles for SST gap filling under cloud-like occlusion. LLC4320 is explicitly used as the training set, with LLC2160 as an unseen lower-resolution test case. ([Copernicus Open Science][1])

Zhao, E., Goh, E., Yepremyan, A., Wang, J., & Wilson, B. (2025). Multi-satellite U-Net for high-resolution sea surface temperature reconstruction. EGUsphere [preprint]. (https://doi.org/10.5194/egusphere-2025-4847)
Uses LLC4320 to generate simulated IR and microwave SST observations and trains the MUSE U-Net model for SST gap filling and multi-sensor reconstruction. ([EGUsphere][2])

Xiao, Q., Balwada, D., Jones, C. S., Herrero-González, M., Smith, K. S., & Abernathey, R. (2023). Reconstruction of surface kinematics from sea surface height using neural networks. Journal of Advances in Modeling Earth Systems, 15, e2023MS003709. (https://doi.org/10.1029/2023MS003709)
Uses LLC4320 SSH and Lagrangian-filtered fields to train neural networks for reconstructing surface vorticity, strain, and divergence from SSH; this is one of the most directly relevant SWOT–ML–LLC4320 papers. ([AGU Publications][3])

Gao, Z., Chapron, B., Ma, C., Fablet, R., Febvre, Q., Zhao, W., & Chen, G. (2024). A deep learning approach to extract balanced motions from sea surface height snapshot. Geophysical Research Letters, 51, e2023GL106623. https://doi.org/10.1029/2023GL106623)
Uses MITgcm LLC4320 SSH to train/evaluate a deep-learning framework for separating balanced and unbalanced motions from SSH snapshots. ([AGU Publications][4])

Bodner, A., Balwada, D., & Zanna, L. (2025). A data-driven approach for parameterizing ocean submesoscale buoyancy fluxes. Journal of Advances in Modeling Earth Systems. (https://doi.org/10.1029/2025MS004991)
Trains a CNN on sampled LLC4320 regions to predict mixed-layer vertical buoyancy fluxes for submesoscale parameterization. ([AGU Publications][5])

Martin, S. A., Manucharyan, G. E., & Klein, P. (2025). Generative data assimilation for surface ocean state estimation from multi-modal satellite observations. Journal of Advances in Modeling Earth Systems. (https://doi.org/10.1029/2025MS005063)
Uses generative modeling for surface-ocean state estimation and discusses/compares against high-resolution model data including LLC4320; this is important for diffusion/generative DA framing. ([AGU Publications][6])

Gallmeier, K., Prochaska, J. X., Cornillon, P., Menemenlis, D., & Kelm, M. (2023). An evaluation of the LLC4320 global-ocean simulation based on the submesoscale structure of modeled sea surface temperature fields. Geoscientific Model Development, 16, 7143–7170. (https://doi.org/10.5194/gmd-16-7143-2023)
Not a forecasting model, but a major ML-based evaluation of LLC4320 SST using an unsupervised probabilistic autoencoder / ULMO framework trained on VIIRS SST cutouts. ([GMD][7])

Lenain, L., et al. (2026). An unprecedented view of ocean currents from geostationary satellites. Nature Geoscience. (https://doi.org/10.1038/s41561-026-01943-0)
Introduces GOFLOW, a U-Net-based method trained on high-resolution global ocean models to infer surface velocity and velocity-gradient fields from geostationary SST imagery; the paper explicitly frames GOFLOW as overcoming sparse altimetric sampling and geostrophic limitations. ([Nature][8])

Wan, Z., Zhu, Y., Peng, S., Xie, J., Li, S., & Song, T. (2025). A TransUNet-based intelligent method for identifying internal solitary waves in the South China Sea. Journal of Marine Science and Engineering, 13(6), 1154. (https://doi.org/10.3390/jmse13061154)
Uses deep-learning semantic segmentation for internal solitary wave detection and lists LLC4320 among core keywords/data context; relevant if the list includes feature-identification tasks rather than only reconstruction/DA. ([MDPI][9])

Emerging / preprint / benchmark-style LLC4320 ML work

VISION authors. (2025). VISION: Prompting ocean vertical velocity reconstruction from incomplete observations. arXiv preprint.
Builds a Kuroshio-Dynamics-48 benchmark from LLC4320, mapping incomplete multichannel surface fields—SSH, SST, SSS, U, V—to 3-D subsurface vertical velocity targets. ([arXiv][10])

Agabin, A., & Prochaska, J. X. (2024). Mitigating masked pixels in a climate-critical ocean dataset. Remote Sensing, 16(13), 2439. (https://doi.org/10.3390/rs16132439)
Uses LLC4320 SST fields in an SST masking/reconstruction context and cites MAESSTRO as a related masked-autoencoder approach. ([MDPI][12])

Useful LLC4320 ML datasets / software references

Jones, C. S., Xiao, Q., Abernathey, R. P., & Smith, K. S. (2022). Ocean surface currents, SSH and SST from LLC4320, before and after Lagrangian filtering [Dataset]. Zenodo. (https://doi.org/10.5281/zenodo.6574307)
Provides LLC4320 velocity, SSH, and SST fields before and after Lagrangian filtering; the dataset description explicitly notes that it is useful for machine learning and dynamical-equation studies. ([Zenodo][13])

Pangeo. MITgcm Global LLC4320 Simulations [Cloud catalog].
Provides analysis-ready LLC4320 grid, SSH, SST, SSS, and surface velocity variables. ([Pangeo Catalog][15])

Important correction to the previous answer

[1]: https://os.copernicus.org/articles/20/1309/2024/ "OS - MAESSTRO: Masked Autoencoders for Sea Surface Temperature Reconstruction under Occlusion"

[2]: https://egusphere.copernicus.org/preprints/2025/egusphere-2025-4847/ "EGUsphere - Multi-satellite U-Net for high-resolution sea surface temperature reconstruction"

[3]: https://agupubs.onlinelibrary.wiley.com/doi/full/10.1029/2023MS003709?utm_source=chatgpt.com "Reconstruction of Surface Kinematics From ... - AGU Journals"

[4]: https://agupubs.onlinelibrary.wiley.com/doi/full/10.1029/2023GL106623?utm_source=chatgpt.com "A Deep Learning Approach to Extract Balanced Motions From ..."

[5]: https://agupubs.onlinelibrary.wiley.com/doi/10.1029/2025MS004991?utm_source=chatgpt.com "A Data‐Driven Approach for Parameterizing Ocean ..."

[6]: https://agupubs.onlinelibrary.wiley.com/doi/full/10.1029/2025MS005063?utm_source=chatgpt.com "Generative Data Assimilation for Surface Ocean State ..."

[7]: https://gmd.copernicus.org/articles/16/7143/2023/?utm_source=chatgpt.com "An evaluation of the LLC4320 global-ocean simulation ... - GMD"

[8]: https://www.nature.com/articles/s41561-026-01943-0 "An unprecedented view of ocean currents from geostationary satellites | Nature Geoscience"

[9]: https://www.mdpi.com/2077-1312/13/6/1154 "A TransUNet-Based Intelligent Method for Identifying Internal Solitary Waves in the South China Sea"

[10]: https://arxiv.org/html/2509.21477v1 "VISION: Prompting Ocean Vertical Velocity Reconstruction from Incomplete Observations"

[11]: https://agu.confex.com/agu/agu24/meetingapp.cgi/Paper/1645744?utm_source=chatgpt.com "Reconstruction of High-Resolution Sea Surface ..."

[12]: https://www.mdpi.com/2072-4292/16/13/2439?utm_source=chatgpt.com "Mitigating Masked Pixels in a Climate-Critical Ocean Dataset"

[13]: https://zenodo.org/records/6574307?utm_source=chatgpt.com "Ocean surface currents, SSH and SST from LLC4320, before ..."