Ashesh Chattopadhyay

Regional ocean dynamics can be better emulated with AI models Researchers show the success of their technical in a critical region: the Gulf of Mexico.

🌊 Research led by #BaskinEngineering Assistant Professor of Applied Mathematics @ashesh6810.bsky.social shows that regional ocean dynamics in the Gulf of Mexico can be better emulated with #AI models—offering new possibilities for navigation and extreme weather monitoring. Read on: bit.ly/4n0AHvj

Simultaneous Emulation and Downscaling With Physically Consistent Deep Learning‐Based Regional Ocean Emulators An AI-based physically consistent long-term regional emulator has been developed for the Gulf of Mexico region A deterministic and stochastic downscaling model has been developed to super-resolve...

📄 Paper:
Lupin-Jimenez et al. (2025)
"Simultaneous Emulation and Downscaling..."
doi.org/10.1029/2025JH000851

💾 Code & data:
zenodo.org/record/14607130

We’d love to hear from collaborators in ocean ML, emulation, and climate AI 🌊🤝

🚀 Why it’s exciting:
✅ Physically grounded
✅ 10x–1000x faster than ROMS
✅ Enables regional “digital twins”
✅ Sets up for coupled ocean–atmosphere emulation
✅ Works across different reanalysis sources

AI meets ocean science.

📊 Results:

Beats interpolation

Matches or outperforms ROMS in short-term accuracy

Stays stable & realistic over 10 years

Captures mean state and eddy variability

Preserves spectral energy across scales

No exploding gradients here 💥

🧠 What’s different:

We don't just super-resolve existing data.
We downscale from an emulator that predicts ocean dynamics.

Plus: our downscaler learns to correct both model bias and physical mismatch (GLORYS → CNAPS). That’s new.

⚙️ Our framework (FCDS):

An FNO emulator predicts SSH, SSU, SSV, SSKE daily at 8 km

A UNet + PatchGAN-VAE downscales to 4 km & corrects bias

Spectral loss + online fine-tuning ensures physical consistency

Together: speed, structure, and stability.

🌍 Why this matters:

Regional ocean models like the Gulf of Mexico are hard—complex coastlines, eddies, Loop Current, chaotic boundary forcing.

Physics models = accurate but slow.
ML = fast, but unstable after a few weeks. We wanted the best of both.

Simultaneous Emulation and Downscaling With Physically Consistent Deep Learning‐Based Regional Ocean Emulators An AI-based physically consistent long-term regional emulator has been developed for the Gulf of Mexico region A deterministic and stochastic downscaling model has been developed to super-resolve...

🚨 New from our group! A stable AI framework for high-res regional ocean modeling-- joint work with Fujitsu Research and NC State led by @baskinengineering.bsky.social PhD students Lenny and @moeindarman.bsky.social.
Now out in JGR: Machine Learning & Computation 🌊🤖
🔗 doi.org/10.1029/2025JH000851 🧵

Led by @baskinengineering.bsky.social PhD students Niloofar & Lenny with Tianning Wu & Roy He @ncstate.bsky.social

If you're working on GenAI for Earth systems, let’s connect — curious to hear your thoughts!
#GenAI #ClimateAI #OceanML #FNO #DDPM #DataAssimilation

Our method is:
⚡️ One-shot
🌀 Physics-consistent
🌐 Scalable

It captures high-wavenumber, fine-scale structures other ML baselines miss. Spectral diagnostics & vorticity metrics confirm this. (4/5)

🧠 The framework combines:
• FNO (Fourier Neural Operator)
• DDPM (Denoising Diffusion Probabilistic Model)

✅ Reconstructs high-resolution states from 1%–0.1% data
✅ Works on synthetic turbulence, GLORYS reanalysis & real satellite altimetry
✅ No forward solver required (3/5)

Ocean observations are often sparse, noisy, and Lagrangian (they move with the flow).
This makes reconstructing fine-scale ocean dynamics like eddies and fronts very hard — especially for forecasting.
We tackle this using a diffusion model conditioned on a neural operator. (2/5)

Generative Lagrangian data assimilation for ocean dynamics under extreme sparsity Reconstructing ocean dynamics from observational data is fundamentally limited by the sparse, irregular, and Lagrangian nature of spatial sampling, particularly in subsurface and remote regions. This ...

🚨 New preprint alert!
“Generative Lagrangian Data Assimilation for Ocean Dynamics Under Extreme Sparsity” is live!
📄 arxiv.org/abs/2507.06479
🌊 Reconstructs high-res ocean states from just 0.1% data using #GenAI. No forward model needed. (1/5)

A physical analysis of #OceanNet, our high-resolution regional ocean digital twins' predictions for the Loop Current led by Anna Lowe in collaboration with Michael Gray, Tianning Wu, and Ruoying He out in AMS AI for Earth systems. journals.ametsoc.org/view/journal...

🌪️ Can #AI predict freak weather events? AI models handle daily forecasts well—but often miss rare extremes. A team including #BaskinEngineering Asst. Prof. @ashesh6810.bsky.social is exploring how adding physics-based principles could improve AI’s accuracy in extreme cases. bit.ly/3Foh7ta

Check out our new work in @pnas.org exploring AI weather's capabilities to predict OOD gray swans.

Explaining the physics of transfer learning in data-driven turbulence modeling Abstract. Transfer learning (TL), which enables neural networks (NNs) to generalize out-of-distribution via targeted re-training, is becoming a powerful to

A key takeaway is that both a priori and a posteriori performance of ML-based parameterization (stability, accuracy, etc) can be derived from insights embedded in the spectral representation of neural networks. Take a look at some of our older work if interested. academic.oup.com/pnasnexus/ar...

We find an interesting distribution of Gabor filters and low-pass filters before and after fine-tuning and predictable spectral dynamics of the hidden layers both during training and fine-tuning phase.

The key idea lies in analyzing the network in spectral space during training, inference, and fine-tuning. Interestingly, more often than not, generalizing to a new system means generalizing to a new shape of the Fourier spectrum and that is a key indicator of model performance a priori.

Fourier analysis of the physics of transfer learning for data-driven subgrid-scale models of ocean turbulence Transfer learning (TL) is a powerful tool for enhancing the performance of neural networks (NNs) in applications such as weather and climate prediction and turbulence modeling. TL enables models to ge...

We released a new pre-print (arxiv.org/abs/2504.15487) on understanding the physics of out-of-distribution generalization (and lack there-of) for turbulence modeling of ocean dynamics. Led by @moeindarman.bsky.social with @pedramh.bsky.social and Laure Zanna.

Looking forward to learning about recent advances in #AI4Climate at the @apsphysics.bsky.social #GlobalPhysicsSummit meeting. Come check out the back-to-back focus sessions, "AI Applications in Weather and Climate I & II," on Tuesday from 9:00 AM to 1:30 PM!
summit.aps.org/schedule/?c=...

Postdoctoral Scholar - Chattopadhyay Lab University of California, Santa Cruz is hiring. Apply now!

I am hiring for a #postdocposition for scientific ML + climate dynamics. Folks with deep learning, scientific computing skills; preferably some background in climate, please reach out! This is part of an #NSF project in collaboration with Nicole Feldl and Geoff Vallis. recruit.ucsc.edu/JPF01844

We have released the codes and the framework as a part of the pre-print. Do check it out if you are interested or work in this space. The framework is adaptable to other areas of geophysics and generally Earth system modeling beyond just the ocean and atmosphere.

The 8Km emulated ocean is then downscaled to a 4KM reanalysis product with a generative model. The coupled emulator + downscaling framework is long-term stable, demonstrates accurate kinetic energy spectrum, and has the right mean and variability over decadal time scales.

Some the key ideas in the work involves building a framework where instead of costly reanalysis products or forecasts which are downscaled, we built an ocean emulator at 8Km over the Gulf of Mexico which is long-stable, does not drift, and remain physically consistent.

Simultaneous emulation and downscaling with physically-consistent deep learning-based regional ocean emulators Building on top of the success in AI-based atmospheric emulation, we propose an AI-based ocean emulation and downscaling framework focusing on the high-resolution regional ocean over Gulf of Mexico. R...

We have released a new pre-print on AI-based long-term regional ocean modeling and downscaling. arxiv.org/abs/2501.05058. This is work led by my PhD students Lenny and Moein with collaborators Roy He, Michael Gray, and Tianning Wu at NCSU and Subhashis Hazarika and Anthony Wong at Fujitsu Research

5. And finally, some cool results on out-of-distribution generalization (a.k.a extrapolation) capabilities and lack-thereof for SOTA AI weather models on gray swan extremes from @pedramh.bsky.social's group and collaborators. agu.confex.com/agu/agu24/me... 5/5

4. We will also show some cool results on pen-and-paper stability analysis of autoregressive emulators, predicting model behavior in an architecture-agnostic fashion, and some interesting scaling laws on error growth and eigenvalues for deep learning emulators. agu.confex.com/agu/agu24/me.... 4/5

3. If you are around on Wednesday, check out our high-resolution ocean emulator with simultaneous downscaling at 4KM regionally. agu.confex.com/agu/agu24/me... 3/5

2. Haiwen will present joint work with Romit Maulik and Troy Arcomano on LUICE, our cheap long-term stable, climate emulator agu.confex.com/agu/agu24/me.... You can also see our paper here: arxiv.org/abs/2405.16297 2/5