Samudra: An AI Global Ocean Emulator for Climate

Geophysical Research Letters, Год журнала: 2025, Номер 52(10)

Опубликована: Май 24, 2025

Abstract AI emulators for forecasting have emerged as powerful tools that can outperform conventional numerical predictions. The next frontier is to build long climate simulations with skill across a range of spatiotemporal scales, particularly important goal the ocean. Our work builds skillful global emulator ocean component state‐of‐the‐art model. We emulate key variables, sea surface height, horizontal velocities, temperature, and salinity, their full depth. use modified ConvNeXt UNet architecture trained on multi‐depth levels data. show emulator— Samudra —which exhibits no drift relative truth, reproduce depth structure variables interannual variability. stable centuries 150 times faster than original struggles capture correct magnitude forcing trends simultaneously remain stable, requiring further work.

Язык: Английский

---

Advancing ocean monitoring and knowledge for societal benefit: the urgency to expand Argo to OneArgo by 2030

Frontiers in Marine Science, Год журнала: 2025, Номер 12

Опубликована: Июнь 2, 2025

The ocean plays an essential role in regulating Earth’s climate, influencing weather conditions, providing sustenance for large populations, moderating anthropogenic climate change, encompassing massive biodiversity, and sustaining the global economy. Human activities are changing oceans, stressing health, threatening critical services provides to society, with significant consequences human well-being safety, economic prosperity. Effective sustainable monitoring of physical, biogeochemical state ecosystem structure ocean, enable adaptation, carbon management marine resource is urgently needed. Argo program, a cornerstone Global Ocean Observing System (GOOS), has revolutionized observation by real-time, freely accessible temperature salinity data upper 2,000m (Core Argo) using cost-effective simple robotics. For past 25 years, have underpinned many forecasting services, playing fundamental safeguarding goods lives. enabled clearer assessments warming, sea level change underlying driving processes, as well scientific breakthroughs while supporting public awareness education. Building on Argo’s success, OneArgo aims greatly expand capabilities 2030, expanding full-ocean depth, collecting parameters, observing rapidly polar regions. Providing synergistic subsurface extension several key space-based Earth Observation missions GOOS components, will new long-term predictions which deep component. Driving forward revolution our understanding ecosystems poorly-measured be instrumental assess fluxes, acidification deoxygenation. Emerging applications include views mixing, bathymetry sediment transport, resilience assessment. Implementing requires about $100 million annually, increase compared present funding. strategic investment provide decision-makers, both government industry, knowledge needed navigate future environmental challenges, safeguard wellbeing generations come.

Язык: Английский

---

Crafting the Future: Machine learning for ocean forecasting

State of the Planet, Год журнала: 2025, Номер 5-opsr, С. 1 - 9

Опубликована: Июнь 2, 2025

Abstract. Artificial intelligence and machine learning are accelerating research in Earth system science, with huge potential for impact challenges ocean prediction. Such algorithms being deployed on different aspects of the forecasting workflow aim improving its speed skill. They include pattern classification anomaly detection; regression diagnostics; state prediction from nowcasting to synoptic, sub-seasonal, seasonal forecasting. This brief review emphasizes scientific methods that have capacity embed domain knowledge; ensure interpretability through causal explanation, be robust reliable; involve effectively high-dimensional statistical methods, supporting multi-scale multi-physics simulations aimed at parameterization; drive intelligent automation, as well decision support. An overview recent numerical developments is discussed, highlighting importance fully data-driven models future expansion capabilities.

Язык: Английский

---

A Stable Implementation of a Data‐Driven Scale‐Aware Mesoscale Parameterization

Journal of Advances in Modeling Earth Systems, Год журнала: 2024, Номер 16(10)

Опубликована: Окт. 1, 2024

Abstract Ocean mesoscale eddies are often poorly represented in climate models, and therefore, their effects on the large scale circulation must be parameterized. Traditional parameterizations, which represent bulk effect of unresolved eddies, can improved with new subgrid models learned directly from data. Zanna Bolton (2020), https://doi.org/10.1029/2020gl088376 (ZB20) applied an equation‐discovery algorithm to reveal interpretable expression parameterizing momentum fluxes by through components velocity‐gradient tensor. In this work, we implement ZB20 parameterization into primitive‐equation GFDL MOM6 ocean model test it two idealized configurations significantly different dynamical regimes topography. The original was found generate excessive numerical noise near grid scale. We propose filtering approaches avoid issues additionally enhance strength large‐scale energy backscatter. filtered parameterizations led climatological mean state distributions, compared current state‐of‐the‐art backscatter parameterizations. scale‐aware and, consequently, used a single value non‐dimensional scaling coefficient for range resolutions. successful application parameterize offers promising opportunity reduce long‐standing biases global simulations future studies.

Язык: Английский

---

A Machine Learning Approach for Deriving Atmospheric Temperatures and Typhoon Warm Cores From FY‐3E MWTS‐3 Observations

Journal of Geophysical Research Machine Learning and Computation, Год журнала: 2024, Номер 1(3)

Опубликована: Сен. 1, 2024

Abstract Machine learning has gained an increasing popularity in the fields of satellite retrieval and numerical weather modeling. In this study, machine‐learning (ML) neural‐network (NN) models are utilized to retrieve atmospheric temperatures from observations brightness temperature Microwave Temperature Sounder‐3 (MWTS‐3) onboard China's first dawn‐dusk polar‐orbiting Fengyun (FY)‐3E, with ERA5 reanalysis serving as training data sets. The root mean square errors ML‐retrieved at all pressure levels smaller than those obtained by a previously used traditional linear regression method compared over global oceans well radiosonde land. Less 1‐week period is usually sufficient for ML NN model converge less 50–100 iterations. shortest time 3‐days right before testing period. While horizontal patterns temporal evolutions warm cores Typhoon Malakas (2022) Haikui (2023) upper troposphere favorably methods periods. vertical structures extend further down middle lower while confined troposphere. A comparison among results additional sets across different seasons confirms above conclusion.

Язык: Английский

---

Cross-attractor transforms: Improving forecasts by learning optimal maps between dynamical systems and imperfect models

Authorea (Authorea), Год журнала: 2024, Номер unknown

Опубликована: Июнь 6, 2024

Biased, incomplete models are often used for forecasting states of complex dynamical systems by mapping an estimate a “true’ initial state into model phase space, making forecast, and then back to the space. While advances have been made reduce errors associated with initialization forecasts, we lack general framework discovering optimal mappings between reference system Here, propose using data-driven approach infer these maps. Our consistently reduces in Lorenz-96 imperfect constructed produce significant compared configuration. Optimal pre- post-processing transforms leverage “shocks’ “drifts’ make more skillful forecasts system. The implemented machine learning architecture neural networks custom analog-adjoint layer makes generalizable numerous applications.

Язык: Английский

---

Parameterization of Langmuir circulation under geostrophic effects using the data-driven approach

Progress In Oceanography, Год журнала: 2024, Номер unknown, С. 103403 - 103403

Опубликована: Дек. 1, 2024

Язык: Английский

---

The K-profile Parameterization augmented by Deep Neural Networks (KPP_DNN) in the General Ocean Turbulence Model (GOTM)

Authorea (Authorea), Год журнала: 2024, Номер unknown

Опубликована: Май 20, 2024

This study utilizes Deep Neural Networks (DNN) to improve the K-Profile Parameterization (KPP) for vertical mixing effects in ocean’s surface boundary layer turbulence. The DNNs were trained using 11-year turbulence-resolving solutions, obtained by running a large eddy simulation model Ocean Station Papa, predict turbulence velocity scale coefficient and unresolved shear KPP. DNN-augmented KPP schemes (KPP_DNN) have been implemented General Turbulence Model (GOTM). implementation is stable long-term integration as efficient existing variants of schemes. Three different KPP_DNN schemes, varying input output variables, developed trained. performance models compared with that those popular deterministic first-order second-moment closure turbulent parameterizations. Solution comparisons show simulated mixed cooler deeper, aligning closely observations when wave are included In framework, changes shear, which used calculate depth, larger impact on than do magnitude diffusivity. KPP_DNN, depend not only forcing, but also depth buoyancy forcing.

Язык: Английский

---

Machine learning for online sea ice bias correction within global ice-ocean simulations

arXiv (Cornell University), Год журнала: 2023, Номер unknown

Опубликована: Янв. 1, 2023

In this study we perform online sea ice bias correction within a GFDL global ice-ocean model. For this, use convolutional neural network (CNN) which was developed in previous (Gregory et al., 2023) for the purpose of predicting concentration (SIC) data assimilation (DA) increments. An initial implementation CNN shows systematic improvements SIC biases relative to free-running model, however large summertime errors remain. We show that these residual can be significantly improved with augmentation approach, sequential and DA corrections are applied new simulation over training period. This then provides set refine weights network. propose machine-learned scheme could utilized generating conditions, also real-time seasonal-to-subseasonal forecasts.

Язык: Английский

---