From 83081cf53847d01bc1fd10cfa62ac202bc76a8cd Mon Sep 17 00:00:00 2001 From: "pre-commit-ci[bot]" <66853113+pre-commit-ci[bot]@users.noreply.github.com> Date: Mon, 18 May 2026 17:17:03 +0000 Subject: [PATCH 1/2] [pre-commit.ci] pre-commit autoupdate MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit updates: - [github.com/pre-commit/pre-commit-hooks: v4.4.0 → v6.0.0](https://github.com/pre-commit/pre-commit-hooks/compare/v4.4.0...v6.0.0) - https://github.com/psf/black → https://github.com/psf/black-pre-commit-mirror - [github.com/psf/black-pre-commit-mirror: 23.3.0 → 26.5.0](https://github.com/psf/black-pre-commit-mirror/compare/23.3.0...26.5.0) - [github.com/kynan/nbstripout: 0.6.1 → 0.9.1](https://github.com/kynan/nbstripout/compare/0.6.1...0.9.1) - [github.com/nbQA-dev/nbQA: 1.7.0 → 1.9.1](https://github.com/nbQA-dev/nbQA/compare/1.7.0...1.9.1) --- .pre-commit-config.yaml | 10 +++++----- 1 file changed, 5 insertions(+), 5 deletions(-) diff --git a/.pre-commit-config.yaml b/.pre-commit-config.yaml index 586679fc9..b46130748 100644 --- a/.pre-commit-config.yaml +++ b/.pre-commit-config.yaml @@ -1,19 +1,19 @@ repos: - repo: https://github.com/pre-commit/pre-commit-hooks - rev: v4.4.0 + rev: v6.0.0 hooks: - id: end-of-file-fixer - id: trailing-whitespace -- repo: https://github.com/psf/black - rev: 23.3.0 +- repo: https://github.com/psf/black-pre-commit-mirror + rev: 26.5.0 hooks: - id: black - repo: https://github.com/kynan/nbstripout - rev: 0.6.1 + rev: 0.9.1 hooks: - id: nbstripout - repo: https://github.com/nbQA-dev/nbQA - rev: 1.7.0 + rev: 1.9.1 hooks: - id: nbqa-black #- id: nbqa-isort From 427df5007130525ea4011fe7e7f5ae43254c385f Mon Sep 17 00:00:00 2001 From: "pre-commit-ci[bot]" <66853113+pre-commit-ci[bot]@users.noreply.github.com> Date: Mon, 18 May 2026 17:18:11 +0000 Subject: [PATCH 2/2] [pre-commit.ci] auto fixes from pre-commit.com hooks for more information, see https://pre-commit.ci --- content/code/_index.md | 18 ++++++------ content/news/2511Blanke.md | 2 +- content/news/2511Danni.md | 2 +- content/news/2511Samudra.md | 2 +- content/news/2511Samudrace.md | 2 +- content/news/2511Sane.md | 2 +- content/news/2512AGU.md | 38 +++++++++++++------------- content/news/2512Otness.md | 2 +- content/news/2512Zanna.md | 2 +- content/news/2601Berner.md | 1 - content/news/2601Falga.md | 2 +- content/news/2601Wu.md | 2 +- content/news/2602Brettin.md | 2 +- content/news/2602Levine.md | 2 +- content/news/2603Falasca.md | 2 +- content/news/2603Kamm.md | 2 +- content/news/2604Gregory.md | 2 +- content/news/2604Liu.md | 2 +- content/news/2604SScireport.md | 2 +- content/news/2605Perezhogin.md | 2 +- content/news/2605Pudig.md | 2 +- content/news/Newsletters/_index.md | 2 +- content/team/AnurupNaskar.md | 2 +- content/team/DiajengWulandariAtmojo.md | 2 +- content/team/SidArora.md | 4 +-- 25 files changed, 51 insertions(+), 52 deletions(-) diff --git a/content/code/_index.md b/content/code/_index.md index 7c283e5d6..aef0cee67 100644 --- a/content/code/_index.md +++ b/content/code/_index.md @@ -10,8 +10,8 @@ heroBackground: '/images/photo-1542831371-29b0f74f9713.jpeg'

- Our codes are hosted on our Github repository. Weights are hosted on HuggingFace.
- + Our codes are hosted on our Github repository. Weights are hosted on HuggingFace.
+

@@ -55,8 +55,8 @@ heroBackground: '/images/photo-1542831371-29b0f74f9713.jpeg'

Samudra: An AI Global Ocean Emulator for Climate
The first 3D global emulator for Climate, 360x faster than traditional models!
- Code | - Weights | + Code | + Weights | OM4-data

@@ -71,8 +71,8 @@ heroBackground: '/images/photo-1542831371-29b0f74f9713.jpeg'

SamudrACE: the coupled climate emulator
Fast and accurate coupled ocean–atmosphere–sea-ice emulator.
- Code | - Weights + Code | + Weights

@@ -87,7 +87,7 @@ heroBackground: '/images/photo-1542831371-29b0f74f9713.jpeg'

CAMulator: Fast Emulation of the Community Atmosphere Model
250× quicker than CESM CAM.
- Code + Code

@@ -100,8 +100,8 @@ heroBackground: '/images/photo-1542831371-29b0f74f9713.jpeg'

FloeNet: A mass-conserving global sea ice emulator that generalizes across climates
- Code | - Weights + Code | + Weights

diff --git a/content/news/2511Blanke.md b/content/news/2511Blanke.md index 5a8120023..1a4cb0e1a 100644 --- a/content/news/2511Blanke.md +++ b/content/news/2511Blanke.md @@ -9,4 +9,4 @@ images: ['images/news/2511Blanke.png'] link: 'https://doi.org/10.48550/arXiv.2505.18017' --- -Deep generative models, AI systems that can learn to create realistic data, are increasingly used to represent complex physical systems. However, these models often produce results that break basic physical laws, such as conservation of energy or mass. This **LEAP [study](https://doi.org/10.48550/arXiv.2505.18017)**, led by **Matthieu Blanke**, introduces a new method, called **Split Augmented Langevin (SAL)**, that ensures AI-generated outputs obey these fundamental constraints. By enforcing physical constraints in the sampling algorithm of pre-trained diffusion models, the approach makes **AI-based simulations and forecasts more accurate and reliable**. The method shows promising results in climate science applications, paving the way for AI tools that better respect the laws of nature. \ No newline at end of file +Deep generative models, AI systems that can learn to create realistic data, are increasingly used to represent complex physical systems. However, these models often produce results that break basic physical laws, such as conservation of energy or mass. This **LEAP [study](https://doi.org/10.48550/arXiv.2505.18017)**, led by **Matthieu Blanke**, introduces a new method, called **Split Augmented Langevin (SAL)**, that ensures AI-generated outputs obey these fundamental constraints. By enforcing physical constraints in the sampling algorithm of pre-trained diffusion models, the approach makes **AI-based simulations and forecasts more accurate and reliable**. The method shows promising results in climate science applications, paving the way for AI tools that better respect the laws of nature. diff --git a/content/news/2511Danni.md b/content/news/2511Danni.md index 7cb810c73..1237fa4a4 100644 --- a/content/news/2511Danni.md +++ b/content/news/2511Danni.md @@ -9,4 +9,4 @@ images: ['images/news/2511DanniDu.png'] link: 'https://doi.org/10.22541/essoar.176083747.76188196/v2' --- -Ocean models often struggle to represent how water mixes vertically, leading to persistent temperature and circulation biases. In this [preprint](https://doi.org/10.22541/essoar.176083747.76188196/v2), **Danni Du** and colleagues use machine learning (ML) to correct those biases by **learning directly from data assimilation outputs in NOAA’s GFDL SPEAR system**. When integrated into the ocean model, the ML corrections improved temperature and mixing accuracy, outperforming existing correction methods. Combining ML with traditional approaches produced even better results, leading to **more realistic sea surface temperatures and ocean structure**. This approach can be applied to other climate models, offering a powerful new way to make ocean simulations more accurate. \ No newline at end of file +Ocean models often struggle to represent how water mixes vertically, leading to persistent temperature and circulation biases. In this [preprint](https://doi.org/10.22541/essoar.176083747.76188196/v2), **Danni Du** and colleagues use machine learning (ML) to correct those biases by **learning directly from data assimilation outputs in NOAA’s GFDL SPEAR system**. When integrated into the ocean model, the ML corrections improved temperature and mixing accuracy, outperforming existing correction methods. Combining ML with traditional approaches produced even better results, leading to **more realistic sea surface temperatures and ocean structure**. This approach can be applied to other climate models, offering a powerful new way to make ocean simulations more accurate. diff --git a/content/news/2511Samudra.md b/content/news/2511Samudra.md index c7f783726..c97457c97 100644 --- a/content/news/2511Samudra.md +++ b/content/news/2511Samudra.md @@ -21,4 +21,4 @@ M²LInES emulator Samudra was recently featured on 2 platforms: 🎬 **Prof Grace Lindsay Youtube channel** - 5 Minute Papers AI for the Planet: How AI can speed up our study of the ocean {{< youtube ijyF16uy0Hk >}} -
\ No newline at end of file +
diff --git a/content/news/2511Samudrace.md b/content/news/2511Samudrace.md index 311218079..dd2cf047a 100644 --- a/content/news/2511Samudrace.md +++ b/content/news/2511Samudrace.md @@ -11,4 +11,4 @@ link: 'https://medium.com/@lz1955/samudrace-a-fast-accurate-efficient-3d-coupled Our latest **[blogpost](https://medium.com/@lz1955/samudrace-a-fast-accurate-efficient-3d-coupled-climate-ai-emulator-fcef3c60b079) dives into the story behind SamudrACE**, the first 3D AI ocean–atmosphere–sea-ice climate emulator. Developed in collaboration with **M²LInES, AI2, and NOAA GFDL**, SamudrACE marks a major milestone in the use of AI for climate science. -The post explores how the team built a model capable of simulating 1500 years of climate in just one day on a single GPU, making state-of-the-art climate modeling accessible to anyone, without the need for supercomputers or deep expertise in numerical modeling. \ No newline at end of file +The post explores how the team built a model capable of simulating 1500 years of climate in just one day on a single GPU, making state-of-the-art climate modeling accessible to anyone, without the need for supercomputers or deep expertise in numerical modeling. diff --git a/content/news/2511Sane.md b/content/news/2511Sane.md index aa09e2a87..e253d2f67 100644 --- a/content/news/2511Sane.md +++ b/content/news/2511Sane.md @@ -9,4 +9,4 @@ images: ['images/news/2511Sane.png'] link: 'https://doi.org/10.31219/osf.io/uab7v_v2' --- -A new [study](https://doi.org/10.31219/osf.io/uab7v_v2), led by **Aakash Sane**, introduces **a two step method to improve how ocean surface mixing is represented in models**. First, neural networks predict vertical diffusivity while respecting key physical constraints. Then, symbolic regression converts these predictions into simple equations that match the neural network accuracy but are easier to interpret. The resulting formulas reveal how friction velocity, buoyancy flux, Earth’s rotation and boundary layer depth shape mixing and expose a flaw in the standard physics based scheme. This approach provides **a transparent, efficient and physically grounded way to model ocean vertical mixing.** \ No newline at end of file +A new [study](https://doi.org/10.31219/osf.io/uab7v_v2), led by **Aakash Sane**, introduces **a two step method to improve how ocean surface mixing is represented in models**. First, neural networks predict vertical diffusivity while respecting key physical constraints. Then, symbolic regression converts these predictions into simple equations that match the neural network accuracy but are easier to interpret. The resulting formulas reveal how friction velocity, buoyancy flux, Earth’s rotation and boundary layer depth shape mixing and expose a flaw in the standard physics based scheme. This approach provides **a transparent, efficient and physically grounded way to model ocean vertical mixing.** diff --git a/content/news/2512AGU.md b/content/news/2512AGU.md index 6205fabb6..e93539dcc 100644 --- a/content/news/2512AGU.md +++ b/content/news/2512AGU.md @@ -5,78 +5,78 @@ heroHeading: '' heroSubHeading: 'AGU 2025 – M²LInES team members and affiliates Schedule' heroBackground: '' thumbnail: 'images/news/agu25.jpg' -images: +images: link: '' --- ### 📅 Monday, 15 December 2025 -Mitch Bushuk — [Antarctic Sea Ice Trends Across a High-Resolution Coupled Model Hierarchy](https://agu.confex.com/agu/agu25/meetingapp.cgi/Paper/1950675) +Mitch Bushuk — [Antarctic Sea Ice Trends Across a High-Resolution Coupled Model Hierarchy](https://agu.confex.com/agu/agu25/meetingapp.cgi/Paper/1950675) 🖼️ Poster | 14:15–17:45 | Hall EFG (Poster Hall), NOLA CC --- ### 📅 Tuesday, 16 December 2025 -Sara Shamekh — [Precipitation Intensity Sensitivity to Large-Scale Thermodynamic State](https://agu.confex.com/agu/agu25/meetingapp.cgi/Paper/1952465) +Sara Shamekh — [Precipitation Intensity Sensitivity to Large-Scale Thermodynamic State](https://agu.confex.com/agu/agu25/meetingapp.cgi/Paper/1952465) 🖼️ Poster | 08:30–12:00 | Hall EFG (Poster Hall), NOLA CC -Nathanael Zhixin Wong — [Investigating how Different Large-Scale Environmental Conditions impact the Shallow-to-Deep Transition of Convection](https://agu.confex.com/agu/agu25/meetingapp.cgi/Paper/1934844) +Nathanael Zhixin Wong — [Investigating how Different Large-Scale Environmental Conditions impact the Shallow-to-Deep Transition of Convection](https://agu.confex.com/agu/agu25/meetingapp.cgi/Paper/1934844) 🖼️ Poster | 08:30–12:00 | Hall EFG (Poster Hall), NOLA CC -Pavel Perezhogin — [NG23A-05 Generalizable Neural-Network Parameterization of Mesoscale Eddies in Idealized and Global Ocean Models](https://agu.confex.com/agu/agu25/meetingapp.cgi/Paper/1896408) +Pavel Perezhogin — [NG23A-05 Generalizable Neural-Network Parameterization of Mesoscale Eddies in Idealized and Global Ocean Models](https://agu.confex.com/agu/agu25/meetingapp.cgi/Paper/1896408) 🎤 Oral presentation | 14:57–15:07 | Room 298–299, NOLA CC -Renaud Falga — [NG23A-06 Towards a Unified Data-Driven Boundary Layer Parameterization for Ocean and Atmosphere](https://agu.confex.com/agu/agu25/meetingapp.cgi/Paper/1959465) +Renaud Falga — [NG23A-06 Towards a Unified Data-Driven Boundary Layer Parameterization for Ocean and Atmosphere](https://agu.confex.com/agu/agu25/meetingapp.cgi/Paper/1959465) 🎤 Oral presentation | 15:07–15:17 | Room 298–299, NOLA CC -Griffin Mooers — [NG23A-08 First Coupled gSAM - Neural Network Simulations to Improve Representation of Precipitation in Climate Simulations](https://agu.confex.com/agu/agu25/meetingapp.cgi/Paper/1980619) +Griffin Mooers — [NG23A-08 First Coupled gSAM - Neural Network Simulations to Improve Representation of Precipitation in Climate Simulations](https://agu.confex.com/agu/agu25/meetingapp.cgi/Paper/1980619) 🎤 Oral presentation | 15:27–15:37 | Room 298–299, NOLA CC -Adam Subel — [NG24A-03 Probing the Dynamical Response of Ocean Climate Emulators (Invited)](https://agu.confex.com/agu/agu25/meetingapp.cgi/Paper/1859610) +Adam Subel — [NG24A-03 Probing the Dynamical Response of Ocean Climate Emulators (Invited)](https://agu.confex.com/agu/agu25/meetingapp.cgi/Paper/1859610) 🎤 Oral presentation | 16:35–16:45 | Room 298–299, NOLA CC -Shuchang Liu — [NG24A-07 CERA: A Framework for Improved Generalization of Machine Learning Models to Changed Climates](https://agu.confex.com/agu/agu25/meetingapp.cgi/Paper/1940186) +Shuchang Liu — [NG24A-07 CERA: A Framework for Improved Generalization of Machine Learning Models to Changed Climates](https://agu.confex.com/agu/agu25/meetingapp.cgi/Paper/1940186) 🎤 Oral presentation | 17:15–17:25 | Room 298–299, NOLA CC --- ### 📅 Wednesday, 17 December 2025 -Alex Connolly — [Data-driven models of a coefficient in a higher-order closure of atmospheric boundary layer turbulence](https://agu.confex.com/agu/agu25/meetingapp.cgi/Paper/1988373) +Alex Connolly — [Data-driven models of a coefficient in a higher-order closure of atmospheric boundary layer turbulence](https://agu.confex.com/agu/agu25/meetingapp.cgi/Paper/1988373) 🖼️ Poster | 08:30–12:00 | Hall EFG (Poster Hall), NOLA CC -Fabrizio Falasca — [Toward Causally-Constrained, Reduced Stochastic Neural Emulators of the Full Ocean](https://agu.confex.com/agu/agu25/meetingapp.cgi/Paper/1945761) +Fabrizio Falasca — [Toward Causally-Constrained, Reduced Stochastic Neural Emulators of the Full Ocean](https://agu.confex.com/agu/agu25/meetingapp.cgi/Paper/1945761) 🖼️ Poster | 08:30–12:00 | Hall EFG (Poster Hall), NOLA CC -Danni Du — [Reducing Coupled Model Biases with Machine Learning Corrections from Ocean Data Assimilation Increments](https://agu.confex.com/agu/agu25/meetingapp.cgi/Paper/1896240) +Danni Du — [Reducing Coupled Model Biases with Machine Learning Corrections from Ocean Data Assimilation Increments](https://agu.confex.com/agu/agu25/meetingapp.cgi/Paper/1896240) 🖼️ Poster | 08:30–12:00 | Hall EFG (Poster Hall), NOLA CC -Pierre Gentine — [B31B-05 Global model data fusion to unravel land carbon sinks and their changes (Invited)](https://agu.confex.com/agu/agu25/meetingapp.cgi/Paper/1859477) +Pierre Gentine — [B31B-05 Global model data fusion to unravel land carbon sinks and their changes (Invited)](https://agu.confex.com/agu/agu25/meetingapp.cgi/Paper/1859477) 🎤 Oral presentation | 09:10–09:20 | Room 261–262, NOLA CC -Pierre Gentine — [B34A-03 Parsimony versus complexity (Invited)](https://agu.confex.com/agu/agu25/meetingapp.cgi/Paper/1859486) +Pierre Gentine — [B34A-03 Parsimony versus complexity (Invited)](https://agu.confex.com/agu/agu25/meetingapp.cgi/Paper/1859486) 🎤 Oral presentation | 16:35–16:45 | Room 265–266, NOLA CC --- ### 📅 Thursday, 18 December 2025 -Anurup Naskar — [Multivariate Estimation of Vertical Profiles to Better Understand the Shallow-to-Deep Transition of Convection in the Bankhead National Forest](https://agu.confex.com/agu/agu25/meetingapp.cgi/Paper/1954999) +Anurup Naskar — [Multivariate Estimation of Vertical Profiles to Better Understand the Shallow-to-Deep Transition of Convection in the Bankhead National Forest](https://agu.confex.com/agu/agu25/meetingapp.cgi/Paper/1954999) 🖼️ Poster | 08:30–12:00 | Hall EFG (Poster Hall), NOLA CC -Matthieu Blanke — [GC42A-04 Physically-Constrained Deep Generative Modeling](https://agu.confex.com/agu/agu25/meetingapp.cgi/Paper/1992938) +Matthieu Blanke — [GC42A-04 Physically-Constrained Deep Generative Modeling](https://agu.confex.com/agu/agu25/meetingapp.cgi/Paper/1992938) 🎤 Oral presentation | 11:04–11:14 | New Orleans Theater C, NOLA CC -Pierre Gentine — [Emulating climate variability and extremes with a diffusion-based model trained on CESM2 and finetuned on ERA5](https://agu.confex.com/agu/agu25/meetingapp.cgi/Paper/1910646) +Pierre Gentine — [Emulating climate variability and extremes with a diffusion-based model trained on CESM2 and finetuned on ERA5](https://agu.confex.com/agu/agu25/meetingapp.cgi/Paper/1910646) 🖼️ Poster | 14:15–17:45 | Hall EFG (Poster Hall), NOLA CC --- ### 📅 Friday, 19 December 2025 -Fabrizio Falasca — [GC51A-04 Neural models of multiscale systems: conceptual limitations, stochastic parametrizations, and a climate application](https://agu.confex.com/agu/agu25/meetingapp.cgi/Paper/1946408) -🎤 Oral presentation | 09:04–09:14 | New Orleans Theater C, NOLA CC \ No newline at end of file +Fabrizio Falasca — [GC51A-04 Neural models of multiscale systems: conceptual limitations, stochastic parametrizations, and a climate application](https://agu.confex.com/agu/agu25/meetingapp.cgi/Paper/1946408) +🎤 Oral presentation | 09:04–09:14 | New Orleans Theater C, NOLA CC diff --git a/content/news/2512Otness.md b/content/news/2512Otness.md index abea4c5a5..5f7107af2 100644 --- a/content/news/2512Otness.md +++ b/content/news/2512Otness.md @@ -9,4 +9,4 @@ images: ['images/news/2512Otness.png'] link: 'https://doi.org/10.48550/arXiv.2510.22676' --- -In this [article](https://doi.org/10.1088/2632-2153/ae1a36), Karl Otness and co-authors present a new multiscale machine-learning approach designed to **improve predictions in dynamical systems**. The method captures information moving both from fine to coarse scales and from coarse to fine, **boosting model accuracy and stability**, with only **minimal added computational cost** compared to standard architectures. The team evaluates the approach on an idealized fluid-dynamics closure task, where the multiscale networks learn to correct a chaotic model by representing unresolved small-scale processes. The work highlights the **potential of multiscale AI architectures to enhance the reliability of physical system modeling.** \ No newline at end of file +In this [article](https://doi.org/10.1088/2632-2153/ae1a36), Karl Otness and co-authors present a new multiscale machine-learning approach designed to **improve predictions in dynamical systems**. The method captures information moving both from fine to coarse scales and from coarse to fine, **boosting model accuracy and stability**, with only **minimal added computational cost** compared to standard architectures. The team evaluates the approach on an idealized fluid-dynamics closure task, where the multiscale networks learn to correct a chaotic model by representing unresolved small-scale processes. The work highlights the **potential of multiscale AI architectures to enhance the reliability of physical system modeling.** diff --git a/content/news/2512Zanna.md b/content/news/2512Zanna.md index 72c33fb87..3d60727af 100644 --- a/content/news/2512Zanna.md +++ b/content/news/2512Zanna.md @@ -9,4 +9,4 @@ images: ['images/news/2512framework.png'] link: 'https://doi.org/10.48550/arXiv.2510.22676' --- -In this [preprint](https://doi.org/10.48550/arXiv.2510.22676), M²LInES demonstrates the power of **AI driven methods in producing reliable climate simulations.** We introduce a new framework that brings physics- and scale-aware machine learning into climate models. Traditional parameterizations of physical processes often produce significant biases, but AI can now learn these processes directly from data. Our team **implements a suite of data-driven parameterizations in the ocean and sea-ice components of a state-of-the-art model**, ranging from deep learning to interpretable equation-based methods. Our results demonstrate that AI-driven parameterizations can run effectively in operational climate simulations, enabling **hybrid atmosphere–ocean–sea-ice modeling. All tools are open source and available to the community.** \ No newline at end of file +In this [preprint](https://doi.org/10.48550/arXiv.2510.22676), M²LInES demonstrates the power of **AI driven methods in producing reliable climate simulations.** We introduce a new framework that brings physics- and scale-aware machine learning into climate models. Traditional parameterizations of physical processes often produce significant biases, but AI can now learn these processes directly from data. Our team **implements a suite of data-driven parameterizations in the ocean and sea-ice components of a state-of-the-art model**, ranging from deep learning to interpretable equation-based methods. Our results demonstrate that AI-driven parameterizations can run effectively in operational climate simulations, enabling **hybrid atmosphere–ocean–sea-ice modeling. All tools are open source and available to the community.** diff --git a/content/news/2601Berner.md b/content/news/2601Berner.md index 029e5b291..45088538b 100644 --- a/content/news/2601Berner.md +++ b/content/news/2601Berner.md @@ -10,4 +10,3 @@ link: 'https://doi.org/10.22541/essoar.176365939.97870596/v1' --- This [study](https://doi.org/10.22541/essoar.176365939.97870596/v1), led by Judith Berner, examines the fundamental limits of subseasonal-to-seasonal weather predictability and the role of land and ocean initial conditions. Using a climate model in a perfect-model framework, the authors show that beyond four weeks, **land surface initialization - particularly soil moisture and snow - dominates predictability over land**, with ocean conditions playing a secondary role. The results point to significant opportunities for **improving extended-range forecasts through better land initialization and land–atmosphere coupling in prediction systems**. - diff --git a/content/news/2601Falga.md b/content/news/2601Falga.md index 4c3564e35..8eab192cc 100644 --- a/content/news/2601Falga.md +++ b/content/news/2601Falga.md @@ -9,4 +9,4 @@ images: ['images/news/2601Falga.png'] link: 'https://doi.org/10.48550/arXiv.2511.01766' --- -Falga et al. present a [new machine-learning–based parameterization](https://doi.org/10.48550/arXiv.2511.01766) for **turbulent momentum fluxes that works consistently across both oceanic and atmospheric boundary layers.** Trained on large-eddy simulations, the neural network captures key turbulent features missed by traditional schemes and **significantly improves boundary-layer wind predictions in climate models**, reducing errors by a factor of 2–3 under convective conditions. The approach is robust to surface flux biases and generalizes well beyond the training data, **highlighting the promise of unified, data-driven turbulence closures for next-generation climate models.** \ No newline at end of file +Falga et al. present a [new machine-learning–based parameterization](https://doi.org/10.48550/arXiv.2511.01766) for **turbulent momentum fluxes that works consistently across both oceanic and atmospheric boundary layers.** Trained on large-eddy simulations, the neural network captures key turbulent features missed by traditional schemes and **significantly improves boundary-layer wind predictions in climate models**, reducing errors by a factor of 2–3 under convective conditions. The approach is robust to surface flux biases and generalizes well beyond the training data, **highlighting the promise of unified, data-driven turbulence closures for next-generation climate models.** diff --git a/content/news/2601Wu.md b/content/news/2601Wu.md index 5c299b488..81b32d2f1 100644 --- a/content/news/2601Wu.md +++ b/content/news/2601Wu.md @@ -9,4 +9,4 @@ images: ['images/news/2601Wu.png'] link: 'https://doi.org/10.48550/arXiv.2503.03990' --- -Air–sea fluxes — the exchanges of heat, moisture, and gases between the ocean and atmosphere —play a key role in shaping weather and climate. Traditional models often treat these fluxes in a fixed, “one-size-fits-all” way, missing their natural variability. This **[LEAP study](https://doi.org/10.48550/arXiv.2503.03990)**, led by Jiarong Wu, introduces a new probabilistic framework that uses neural networks and observational data to **better capture both the average behavior and the uncertainty of these fluxes**. The results show that accounting for **this variability can influence ocean temperature and mixing**, especially during spring, offering a promising step toward more realistic climate and weather simulations. \ No newline at end of file +Air–sea fluxes — the exchanges of heat, moisture, and gases between the ocean and atmosphere —play a key role in shaping weather and climate. Traditional models often treat these fluxes in a fixed, “one-size-fits-all” way, missing their natural variability. This **[LEAP study](https://doi.org/10.48550/arXiv.2503.03990)**, led by Jiarong Wu, introduces a new probabilistic framework that uses neural networks and observational data to **better capture both the average behavior and the uncertainty of these fluxes**. The results show that accounting for **this variability can influence ocean temperature and mixing**, especially during spring, offering a promising step toward more realistic climate and weather simulations. diff --git a/content/news/2602Brettin.md b/content/news/2602Brettin.md index 1719c5030..cc3a2d55b 100644 --- a/content/news/2602Brettin.md +++ b/content/news/2602Brettin.md @@ -9,4 +9,4 @@ images: ['images/news/2602Brettin.png'] link: 'https://doi.org/10.48550/arXiv.2601.17243' --- -This new [preprint](https://doi.org/10.48550/arXiv.2601.17243) led by **Andrew Brettin** introduces a machine-learning framework to **better quantify uncertainty in extreme climate events.** The proposed ReLU-bias loss quantile neural network (RBLQNN) improves the **accuracy and stability of predicted uncertainty ranges**, especially for nonlinear and non-Gaussian processes. Tested on synthetic data, temperature extremes from over 1,500 NOAA weather stations, and satellite-observed precipitation, the **method outperforms standard approaches** and captures complex dependencies that simpler models miss. This work highlights **RBLQNN as a powerful, flexible tool for assessing climate hazards and extremes.** \ No newline at end of file +This new [preprint](https://doi.org/10.48550/arXiv.2601.17243) led by **Andrew Brettin** introduces a machine-learning framework to **better quantify uncertainty in extreme climate events.** The proposed ReLU-bias loss quantile neural network (RBLQNN) improves the **accuracy and stability of predicted uncertainty ranges**, especially for nonlinear and non-Gaussian processes. Tested on synthetic data, temperature extremes from over 1,500 NOAA weather stations, and satellite-observed precipitation, the **method outperforms standard approaches** and captures complex dependencies that simpler models miss. This work highlights **RBLQNN as a powerful, flexible tool for assessing climate hazards and extremes.** diff --git a/content/news/2602Levine.md b/content/news/2602Levine.md index 3a27d3533..162ed933a 100644 --- a/content/news/2602Levine.md +++ b/content/news/2602Levine.md @@ -9,4 +9,4 @@ images: ['images/news/2602Levine.png'] link: 'https://doi.org/10.5194/wcd-6-1241-2025' --- -A new modeling [study](https://doi.org/10.5194/wcd-6-1241-2025) co-authored by **Xavier Levine** examines how climate extremes in the Arctic may evolve as the region continues to warm faster than the global average. Using the variable-resolution CESM2.2 model, the team compares standard global simulations with high-resolution grids refined over the Arctic to better capture heat waves and heavy precipitation. They find that higher resolution improves the simulation of precipitation extremes, while temperature extremes are better represented in coarser global runs. Projections for the end of the century show stronger and longer-lasting heat extremes, fewer cold extremes, and more intense and frequent heavy precipitation, particularly in regions affected by sea-ice loss and ocean warming. \ No newline at end of file +A new modeling [study](https://doi.org/10.5194/wcd-6-1241-2025) co-authored by **Xavier Levine** examines how climate extremes in the Arctic may evolve as the region continues to warm faster than the global average. Using the variable-resolution CESM2.2 model, the team compares standard global simulations with high-resolution grids refined over the Arctic to better capture heat waves and heavy precipitation. They find that higher resolution improves the simulation of precipitation extremes, while temperature extremes are better represented in coarser global runs. Projections for the end of the century show stronger and longer-lasting heat extremes, fewer cold extremes, and more intense and frequent heavy precipitation, particularly in regions affected by sea-ice loss and ocean warming. diff --git a/content/news/2603Falasca.md b/content/news/2603Falasca.md index ea082cd5f..9cd8b362a 100644 --- a/content/news/2603Falasca.md +++ b/content/news/2603Falasca.md @@ -9,4 +9,4 @@ images: ['images/news/2603Falasca.png'] link: 'https://doi.org/10.22541/essoar.177100611.18240844/v1' --- -In this [work](https://doi.org/10.48550/arXiv.2602.13847), **Fabrizio Falasca** and **Laure Zanna** introduce a **flexible framework that combines response theory and score matching to eliminate spurious, noncausal dependencies in reduced-order neural emulators of turbulent systems.** Using the stochastic **Charney–DeVore model** as a proof of concept for low-frequency atmospheric variability, they demonstrate that **enforcing causal constraints significantly improves emulator responses** to both weak and strong external forcings, even when trained solely on unforced data. The framework is broadly applicable to complex turbulent systems and can be seamlessly integrated into standard neural network architectures, offering a principled **path toward more reliable climate emulators**. \ No newline at end of file +In this [work](https://doi.org/10.48550/arXiv.2602.13847), **Fabrizio Falasca** and **Laure Zanna** introduce a **flexible framework that combines response theory and score matching to eliminate spurious, noncausal dependencies in reduced-order neural emulators of turbulent systems.** Using the stochastic **Charney–DeVore model** as a proof of concept for low-frequency atmospheric variability, they demonstrate that **enforcing causal constraints significantly improves emulator responses** to both weak and strong external forcings, even when trained solely on unforced data. The framework is broadly applicable to complex turbulent systems and can be seamlessly integrated into standard neural network architectures, offering a principled **path toward more reliable climate emulators**. diff --git a/content/news/2603Kamm.md b/content/news/2603Kamm.md index 7c905b0be..962b3b6c9 100644 --- a/content/news/2603Kamm.md +++ b/content/news/2603Kamm.md @@ -9,4 +9,4 @@ images: ['images/news/2603Kamm.png'] link: 'https://doi.org/10.22541/essoar.177100611.18240844/v1' --- -Mesoscale eddies are the ocean’s primary reservoir of kinetic energy, yet most climate models cannot fully resolve them due to computational limits. In this [study](https://doi.org/10.22541/essoar.177100611.18240844/v1) led by **David Kamm**, two **machine-learning–based eddy parameterizations**, Zanna and Bolton (2020) parameterization (ZB20) and Guillaumin and Zanna (2021) parameterization (GZ21), are **implemented in the NEMO ocean model and evaluated against high-resolution simulations.** While GZ21 shows systematic biases linked to grid spacing and does not improve coarse-resolution performance, **ZB20 successfully captures subgrid energy transfers, leading to improved kinetic energy spectra and large-scale circulation.** The results highlight that carefully designed, resolution-aware training data are essential for developing robust and generalizable data-driven eddy parameterizations. \ No newline at end of file +Mesoscale eddies are the ocean’s primary reservoir of kinetic energy, yet most climate models cannot fully resolve them due to computational limits. In this [study](https://doi.org/10.22541/essoar.177100611.18240844/v1) led by **David Kamm**, two **machine-learning–based eddy parameterizations**, Zanna and Bolton (2020) parameterization (ZB20) and Guillaumin and Zanna (2021) parameterization (GZ21), are **implemented in the NEMO ocean model and evaluated against high-resolution simulations.** While GZ21 shows systematic biases linked to grid spacing and does not improve coarse-resolution performance, **ZB20 successfully captures subgrid energy transfers, leading to improved kinetic energy spectra and large-scale circulation.** The results highlight that carefully designed, resolution-aware training data are essential for developing robust and generalizable data-driven eddy parameterizations. diff --git a/content/news/2604Gregory.md b/content/news/2604Gregory.md index 36c6b2564..0e72373dd 100644 --- a/content/news/2604Gregory.md +++ b/content/news/2604Gregory.md @@ -9,4 +9,4 @@ images: ['images/news/2604FloeNet.gif'] link: 'https://doi.org/10.48550/arXiv.2603.12449' --- -**Will Gregory** et al. introduce **[FloeNet](https://doi.org/10.48550/arXiv.2603.12449), a machine-learning emulator trained on the GFDL global sea ice model (SIS2)** to reproduce key sea-ice and snow-on-sea-ice processes while conserving mass. The model emulates 6-hour tendencies related to ice and snow growth, melt, and advection. Trained on reanalysis-forced simulations, FloeNet was tested across different climate states, including pre-industrial and an increasing CO₂ scenario. It accurately **reproduces sea-ice mean state, trends, and interannual variability, outperforming non-conservative approaches.** FloeNet also captures the balance between thermodynamic and dynamic responses to forcing and reproduces coupling-related variables such as ice-surface temperature and ocean salt fluxes. These results suggest strong potential for improving the representation of polar processes in climate emulators. \ No newline at end of file +**Will Gregory** et al. introduce **[FloeNet](https://doi.org/10.48550/arXiv.2603.12449), a machine-learning emulator trained on the GFDL global sea ice model (SIS2)** to reproduce key sea-ice and snow-on-sea-ice processes while conserving mass. The model emulates 6-hour tendencies related to ice and snow growth, melt, and advection. Trained on reanalysis-forced simulations, FloeNet was tested across different climate states, including pre-industrial and an increasing CO₂ scenario. It accurately **reproduces sea-ice mean state, trends, and interannual variability, outperforming non-conservative approaches.** FloeNet also captures the balance between thermodynamic and dynamic responses to forcing and reproduces coupling-related variables such as ice-surface temperature and ocean salt fluxes. These results suggest strong potential for improving the representation of polar processes in climate emulators. diff --git a/content/news/2604Liu.md b/content/news/2604Liu.md index 2123c901f..5335004ae 100644 --- a/content/news/2604Liu.md +++ b/content/news/2604Liu.md @@ -9,4 +9,4 @@ images: ['images/news/2604Liu.png'] link: 'https://doi.org/10.48550/arXiv.2603.17750' --- -Autoregressive neural surrogate models can dramatically accelerate simulations of dynamical systems, but they often suffer from error accumulation over long time horizons. This **[work](https://doi.org/10.48550/arXiv.2603.17750), led by Qi Liu, introduces a unifying framework** that formalizes the trade-off between short-term accuracy and long-term consistency, which most previous approaches handled heuristically. Building on this, the authors propose a **new hyperparameter-free method: Self-refining Neural Surrogate (SNS)**, based on conditional diffusion. SNS can iteratively refine its own predictions or enhance existing models, **delivering stable and accurate simulations even over very long time scales.** \ No newline at end of file +Autoregressive neural surrogate models can dramatically accelerate simulations of dynamical systems, but they often suffer from error accumulation over long time horizons. This **[work](https://doi.org/10.48550/arXiv.2603.17750), led by Qi Liu, introduces a unifying framework** that formalizes the trade-off between short-term accuracy and long-term consistency, which most previous approaches handled heuristically. Building on this, the authors propose a **new hyperparameter-free method: Self-refining Neural Surrogate (SNS)**, based on conditional diffusion. SNS can iteratively refine its own predictions or enhance existing models, **delivering stable and accurate simulations even over very long time scales.** diff --git a/content/news/2604SScireport.md b/content/news/2604SScireport.md index 712e14fc4..230f1f0f9 100644 --- a/content/news/2604SScireport.md +++ b/content/news/2604SScireport.md @@ -11,4 +11,4 @@ link: 'https://www.schmidtsciences.org/2025-report/' The very first [Schmidt Sciences Impact Report](https://www.schmidtsciences.org/2025-report/) has been released, offering a snapshot of the impact made by its grantees and highlighting the importance of supporting foundational research. -Among the featured stories is [M²LInES](https://www.schmidtsciences.org/profile-2025/training-the-next-generation-of-climate-models/), recognized for its work on training the next generation of climate models and advancing AI-driven Earth system science. \ No newline at end of file +Among the featured stories is [M²LInES](https://www.schmidtsciences.org/profile-2025/training-the-next-generation-of-climate-models/), recognized for its work on training the next generation of climate models and advancing AI-driven Earth system science. diff --git a/content/news/2605Perezhogin.md b/content/news/2605Perezhogin.md index de9cc122f..54d348df8 100644 --- a/content/news/2605Perezhogin.md +++ b/content/news/2605Perezhogin.md @@ -9,4 +9,4 @@ images: ['images/news/2605Perezhogin.png'] link: 'https://doi.org/10.48550/arXiv.2604.06398' --- -A new [preprint](https://doi.org/10.48550/arXiv.2604.06398) led by **Pavel Perezhogin** introduces a **more systematic approach to reducing biases in coarse-resolution ocean models**, where key processes like mesoscale eddies are often unresolved. Rather than relying on ad hoc tuning, the study frames parameter adjustment as a calibration problem using **Ensemble Kalman Inversion** (EKI), applied to a neural network–based parameterization. This method significantly **improves model performance** — cutting errors in key ocean features and their variability by about half—while remaining robust to the noisy, chaotic nature of ocean dynamics. The results point to a **practical pathway for enhancing the accuracy of global ocean simulations**. \ No newline at end of file +A new [preprint](https://doi.org/10.48550/arXiv.2604.06398) led by **Pavel Perezhogin** introduces a **more systematic approach to reducing biases in coarse-resolution ocean models**, where key processes like mesoscale eddies are often unresolved. Rather than relying on ad hoc tuning, the study frames parameter adjustment as a calibration problem using **Ensemble Kalman Inversion** (EKI), applied to a neural network–based parameterization. This method significantly **improves model performance** — cutting errors in key ocean features and their variability by about half—while remaining robust to the noisy, chaotic nature of ocean dynamics. The results point to a **practical pathway for enhancing the accuracy of global ocean simulations**. diff --git a/content/news/2605Pudig.md b/content/news/2605Pudig.md index f785475ee..1c2710db8 100644 --- a/content/news/2605Pudig.md +++ b/content/news/2605Pudig.md @@ -9,4 +9,4 @@ images: ['images/news/2605Pudig.png'] link: 'https://doi.org/10.1029/2025MS005497' --- -This [article](https://doi.org/10.1029/2025MS005497) led by Matt Pudig explores how to better represent mesoscale turbulence in ocean models that only partially resolve eddies. The work focuses on “backscatter” parameterizations, which reintroduce energy into the system, and examines whether they can also improve how tracers mix along density surfaces. Using idealized simulations, the authors show that **backscatter alone can substantially enhance the realism of this mixing, closely matching much higher-resolution models and outperforming more conventional approaches.** The findings point to a **promising, unified framework for capturing key ocean processes and improving the fidelity of climate-scale ocean simulations** without requiring much finer resolution. \ No newline at end of file +This [article](https://doi.org/10.1029/2025MS005497) led by Matt Pudig explores how to better represent mesoscale turbulence in ocean models that only partially resolve eddies. The work focuses on “backscatter” parameterizations, which reintroduce energy into the system, and examines whether they can also improve how tracers mix along density surfaces. Using idealized simulations, the authors show that **backscatter alone can substantially enhance the realism of this mixing, closely matching much higher-resolution models and outperforming more conventional approaches.** The findings point to a **promising, unified framework for capturing key ocean processes and improving the fidelity of climate-scale ocean simulations** without requiring much finer resolution. diff --git a/content/news/Newsletters/_index.md b/content/news/Newsletters/_index.md index a890e5d8e..3d2e10a8e 100644 --- a/content/news/Newsletters/_index.md +++ b/content/news/Newsletters/_index.md @@ -9,7 +9,7 @@ tags: Links to our past newsletters are below. - + ### 2026 * 05/01/2026 - [M²LInES newsletter - May 2026](https://mailchi.mp/0ea31f7e9316/m2lines-may2026) diff --git a/content/team/AnurupNaskar.md b/content/team/AnurupNaskar.md index 2c15537d0..d8abf89d7 100644 --- a/content/team/AnurupNaskar.md +++ b/content/team/AnurupNaskar.md @@ -5,7 +5,7 @@ image: "/images/team/AnurupNaskar.png" jobtitle: "Affiliate, Graduate Student" promoted: true weight: 26 -Website: +Website: Position: Climate Informatics tags: [Atmosphere, Machine Learning, Climate Model Development] --- diff --git a/content/team/DiajengWulandariAtmojo.md b/content/team/DiajengWulandariAtmojo.md index 66e6fe951..eccb53c9a 100644 --- a/content/team/DiajengWulandariAtmojo.md +++ b/content/team/DiajengWulandariAtmojo.md @@ -6,7 +6,7 @@ jobtitle: "Affiliate, Graduate Student" promoted: true weight: 26 Website: -Position: +Position: tags: [Sea-Ice, Machine Learning, Climate Model Development] --- diff --git a/content/team/SidArora.md b/content/team/SidArora.md index 7f1e1039a..5a0a7c868 100644 --- a/content/team/SidArora.md +++ b/content/team/SidArora.md @@ -5,8 +5,8 @@ image: "/images/team/SidArora.jpg" jobtitle: "Affiliate, Graduate Student" promoted: true weight: 27 -Website: -Position: +Website: +Position: tags: [Machine Learning, Data Assimilation] ---