SamudrACE: Fast and Accurate Coupled Climate Modeling with 3D Ocean and Atmosphere Emulators
This paper presents a new AI-based climate model called SamudrACE, which can simulate climate conditions much faster than traditional models. It combines different components that represent the atmosphere and ocean to provide accurate climate predictions.
Content & Liability Disclaimer
This article and its accompanying video are automated summaries derived from the original research paper by Unknown authors. The original research was conducted solely by the paper's authors; PDFdigest did not conduct any of the research and makes no claims of ownership over the underlying scientific work.
The video narration is generated by artificial intelligence and references the paper's authors for attribution. The video is not narrated by any of the paper's authors. This content may contain inaccuracies, omissions, or misinterpretations of the original research. First-person language (e.g., "we found", "our results") reflects the original authors' voice, not PDFdigest's. Always read the original paper for accurate, verified information before making any decisions based on this content.
This content is provided "as is" without any warranties, express or implied. Simulated systems OÜ, its officers, directors, employees, and agents shall not be liable for any direct, indirect, incidental, special, consequential, or punitive damages arising from your use of, reliance on, or access to this content, including but not limited to errors, omissions, or misinterpretations of the original research. This disclaimer applies to the fullest extent permitted by applicable law.
- 1 Future versions aim to mitigate this over-smoothing by incorporating stochastic training with CRPS or spectral energy score losses, enabling longer autoregressive trajectories without sacrificing spectral fidelity.
- 2 Since then, atmosphere model emulators have continued to mature and support Atmosphere Model Intercomparison Project (AMIP) compatible simulations .
- 3 Wang et al., 2024) and stable simulation of present-day and CO 2 -modulated equilibrium climate, their simplified ocean representations are insufficient to support accurate coupled atmosphere-ocean variability such as ENSO.
- 4 In this paper, we present SamudrACE, which is constructed by coupling the ACE2 3D atmosphere emulator to the Samudra 3D ocean emulator which has been extended to predict sea-ice concentration and thickness.
Introduction
The advent and success of machine learning (ML)-based weather prediction has led to similarly data-driven global atmosphere emulators trained on output of numerical models, such as the atmosphere-only version of the Ai2 Climate Emulator (ACE) . Coupled atmosphere and ocean emulation is needed to learn and generate realistic climate trends (e.g., through the time-evolving spatial patterns of ocean heat uptake and sea-surface temperature rise).
While these approaches enable accurate seasonal forecasts (C.
Fortunately, three-dimensional ML ocean emulators have been recently developed for data-driven ocean forecasting on timescales up to 1-2 years (Chen et al., 2023; El Aouni et al., 2024; Xiong et al., 2023; X.
While SamudrACE reproduces the general distribution of daily precipitation (Figure S15 ), it slightly underestimates extremes due to the exclusive use of MSE loss, a known limitation of the ACE2 component .
Research Question
Future versions aim to mitigate this over-smoothing by incorporating stochastic training with CRPS or spectral energy score losses, enabling longer autoregressive trajectories without sacrificing spectral fidelity.
Methodology
This training duration is supported by sensitivity analysis (Text S3; Figures S17 and S18 ) showing that longer records are essential for capturing internal variability like ENSO, even as time-mean error metrics stabilize with shorter datasets. Sensitivity analysis (Figure S18 ) confirms that the full training volume is necessary for Niño 3.4 amplitude to converge toward the reference.
Study Design
Spectral analysis (Figure 4b ; see Text S2 for methods) reveals that while SamudrACE reproduces the 3-year peak, it generally exhibits excessive power in the 2-4 year band and a deficit at lower frequencies (> 4 years).
Teleconnections are also well-represented: the regression of precipitation onto the Niño 3.4 index matches the reference (Figure 4c ).
How PDFdigest Helps You Understand Research
Instant Paper Analysis
Get structured summaries and key findings from dense PDFs in seconds.
Visual Explanations
Turn complex methods, figures, and results into clearer visual breakdowns.
AI-Powered Q&A
Ask focused questions and get answers grounded in the paper.
Results & Findings
Since then, atmosphere model emulators have continued to mature and support Atmosphere Model Intercomparison Project (AMIP) compatible simulations . This paper will demonstrate early progress toward the natural next step in this progression, a global climate model (GCM) emulator, which consists of modular coupled atmosphere, sea ice, land, and ocean emulators, capable of running the Coupled Model Intercomparison Program (CMIP) DECK simulation suite .
- Since then, atmosphere model emulators have continued to mature and support Atmosphere Model Intercomparison Project (AMIP) compatible simulations .
- This paper will demonstrate early progress toward the natural next step in this progression, a global climate model (GCM) emulator, which consists of modular coupled atmosphere.
- It is also needed to generate the variability in physical phenomena that emerge through the realistic interaction and coupled evolution of atmospheric surface forcing and upper.
- Several recent papers have incorporated simplified forms of ocean coupling into ML atmospheric emulators, e.g. by use of a physically-based slab ocean model expressed in PyTorch.
- Wang et al., 2024) and stable simulation of present-day and CO 2 -modulated equilibrium climate, their simplified ocean representations are insufficient to support accurate coupled atmosphere-ocean.
Several recent papers have incorporated simplified forms of ocean coupling into ML atmospheric emulators, e.g. by use of a physically-based slab ocean model expressed in PyTorch , or by prognostically emulating sea surface temperature (SST) , or with the addition.
Since then, atmosphere model emulators have continued to mature and support Atmosphere Model Intercomparison Project (AMIP) compatible simulations .
Practical Applications
This could later be extended to incorporate other components of the Earth system (e.g., biogeochemical processes). This combination of low bias and stability indicates that SamudrACE may be amenable to additional physical constraints such as energy budget corrections.
The exact cause is unknown, but we hypothesize that one possible way to overcome this issue is by increasing the amount of training data, thereby allowing for greater sampling of low-frequency dynamics (Text S3; Figure S18 ).
Alternative coupling approaches in component emulator latent space may also be able to circumvent these issues, at the cost of lowered physical interpretability and obfuscation of inconsistencies in uncoupled pretraining.
As a natural next step, future work could explore the incorporation of a sea-ice emulator able to prognose sea ice concentration at the original 6-hourly temporal resolution used in forcing uncoupled ACE2, and for handling atmosphere-ocean flux and momentum exchange.
The SamudrACE coupler
The coupler manages the interaction between ACE2 and Samudra, using different time steps for atmosphere and ocean simulations to ensure realistic variability and accurate coupling.
Pretraining ACE2
ACE2 is pretrained using CM4 output data, incorporating additional diagnostic variables to enhance its predictive capabilities for atmospheric conditions.
Frequently Asked Questions
Coupled atmosphere and ocean emulation is needed to learn and generate realistic climate trends (e.g., through the time-evolving spatial patterns of ocean heat uptake and sea-surface temperature rise). Future versions aim to mitigate this over-smoothing by incorporating stochastic training with CRPS or.
This training duration is supported by sensitivity analysis (Text S3; Figures S17 and S18 ) showing that longer records are essential for capturing internal variability like ENSO, even as time-mean error metrics stabilize with shorter datasets. Sensitivity analysis (Figure S18 ) confirms.
Since then, atmosphere model emulators have continued to mature and support Atmosphere Model Intercomparison Project (AMIP) compatible simulations . Wang et al., 2024) and stable simulation of present-day and CO 2 -modulated equilibrium climate, their simplified ocean representations are insufficient to support.
SamudrACE’s overall time mean state is accurate in terms of the RMSE of the generated time mean (Figure S2 ) and globally averaged annual mean series (Figures S3 and S4 ). This combination of low bias and stability indicates that SamudrACE may.
Several recent papers have incorporated simplified forms of ocean coupling into ML atmospheric emulators, e.g. by use of a physically-based slab ocean model expressed in PyTorch , or by prognostically emulating sea surface temperature (SST) , or with the addition of near-surface.
This paper presents a new AI-based climate model called SamudrACE, which can simulate climate conditions much faster than traditional models. It combines different components that represent the atmosphere and ocean to provide accurate climate predictions.