diff --git a/content/news/2406Dheeshjith.md b/content/news/2406Dheeshjith.md new file mode 100644 index 00000000..6069f7e8 --- /dev/null +++ b/content/news/2406Dheeshjith.md @@ -0,0 +1,14 @@ +--- +date: 2024-06-02T09:29:16+10:00 +title: "Transfer Learning for Emulating Ocean Climate Variability" +heroHeading: '' +heroSubHeading: 'Transfer Learning for Emulating Ocean Climate Variability across CO₂ forcing' +heroBackground: '' +thumbnail: 'images/publications/dsga_2024.png' +images: ['images/publications/dsga_2024.png'] +link: 'https://doi.org/10.48550/arXiv.2405.18585' +--- + +We are adding the link to the preprint which is now accepted in ICML 2024 workshop and will be presented as a talk and a poster this week in Vienna: [Link to the paper](https://doi.org/10.48550/arXiv.2405.18585). + +This study, led jointly by **Surya Dheeshjith** and **Adam Subel**, showcases Machine Learning Ocean Emulators accurately predicting global ocean surface conditions over 5-8 years under different CO₂ forcing scenarios. While the models struggle to generalize, fine-tuning with small amounts of additional data from warner climates can significantly improve model performance. The study also shows the robustness of the emulators to noise in the atmospheric forcing. **Shubham Gupta**, **Alistair Adcroft**, **Carlos Fernandez-Granda**, **Julius Busecke**, and **Laure Zanna** all contributed to the work. diff --git a/content/news/2406Falasca.md b/content/news/2406Falasca.md new file mode 100644 index 00000000..4b2a6cd9 --- /dev/null +++ b/content/news/2406Falasca.md @@ -0,0 +1,12 @@ +--- +date: 2024-06-01T09:29:16+10:00 +title: "Data-driven dimensionality reduction and causal inference for spatiotemporal climate fields" +heroHeading: '' +heroSubHeading: 'Data-driven dimensionality reduction and causal inference for spatiotemporal climate fields' +heroBackground: '' +thumbnail: 'images/news/2406Falasca.jpeg' +images: ['images/news/2406Falasca.jpeg'] +link: 'https://doi.org/10.1103/PhysRevE.109.044202' +--- + +One of the main challenges in statistical inference is to infer causal relations rather than simple correlations. In this [paper](https://doi.org/10.1103/PhysRevE.109.044202), **Fabrizio Falasca**, **Pavel Perezhogin**, and **Laure Zanna**, propose a framework, based on linear response theory, to investigate and diagnose causal mechanisms in the climate system directly from data. The method is physically-based and scales to high-dimensional systems. Possible future applications include the response of the climate system to external perturbations and evaluate climate models outputs. diff --git a/content/news/2407Gentine.md b/content/news/2407Gentine.md new file mode 100644 index 00000000..369373f2 --- /dev/null +++ b/content/news/2407Gentine.md @@ -0,0 +1,12 @@ +--- +date: 2024-07-02T09:29:16+10:00 +title: "Sampling Hybrid Climate Simulation at Scale" +heroHeading: '' +heroSubHeading: 'Sampling Hybrid Climate Simulation at Scale to Reliably Improve Machine Learning Parameterization' +heroBackground: '' +thumbnail: 'images/news/2407Gentine.png' +images: ['images/news/2407Gentine.png'] +link: 'https://doi.org/10.22541/essoar.172072688.86581349/v1' +--- + +Machine-learning (ML) models could enhance climate simulations by accurately representing small-scale processes like turbulence and convection. However, it's unclear if better standalone (offline) performance leads to better integrated (online) performance in climate models. In this preprint, researchers, including **Pierre Gentine**, conducted extensive experiments with 2,970 hybrid simulations, finding that reducing offline error generally improves online accuracy, but some decisions can destabilize the model. Key improvements include incorporating memory, training on diverse climates, converting moisture input to relative humidity, and avoiding certain error metrics. This study answers crucial questions about ML design for parameterizations in climate modeling, paving the way for more reliable and efficient simulations. diff --git a/content/news/2407Reichl.md b/content/news/2407Reichl.md new file mode 100644 index 00000000..c27ff52e --- /dev/null +++ b/content/news/2407Reichl.md @@ -0,0 +1,12 @@ +--- +date: 2024-07-01T09:29:16+10:00 +title: "Improved Equatorial Upper Ocean Vertical Mixing" +heroHeading: '' +heroSubHeading: 'Improved Equatorial Upper Ocean Vertical Mixing in the NOAA/GFDL OM4 Model' +heroBackground: '' +thumbnail: 'images/news/2407Reichl.png' +images: ['images/news/2407Reichl.png'] +link: 'https://doi.org/10.22541/essoar.170785794.47537760/v1' +--- + +Deficiencies in how upper ocean mixing is modeled cause biases in climate simulations, affecting ocean heat uptake and ENSO predictions. To address this, researchers evaluated different mixing models, To address this, researchers evaluated different mixing models, in this [study](https://doi.org/10.22541/essoar.170785794.47537760/v1) led by **Brandon Reichl**, using detailed simulations and applied improvements to NOAA/GFDL's OM4 ocean model. Enhancements to the model's mixing were informed by observational data and led to more accurate diurnal mixing, ocean currents, and upper ocean temperature profiles. The improved model better represents tropical ocean dynamics, leading to more accurate climate predictions. **Alistair Adcroft** also contributed to the research. diff --git a/content/news/Newsletters/_index.md b/content/news/Newsletters/_index.md index 82a09be5..e4ce53b6 100644 --- a/content/news/Newsletters/_index.md +++ b/content/news/Newsletters/_index.md @@ -10,6 +10,8 @@ tags: Links to our past newsletters are below. ### 2024 +* 09/03/2024 - [M²LInES newsletter - September 2024](https://mailchi.mp/bee785c9dfef/m2lines-sept2024) + * 08/05/2024 - [M²LInES newsletter - August 2024](https://mailchi.mp/673f0ae414a1/m2lines-august2024) * 07/22/2024 - [M²LInES newsletter - July 2024](https://mailchi.mp/f28e00f28543/m2lines-july2024) diff --git a/content/publications/_index.md b/content/publications/_index.md index edcfa5d0..4ec5b4b1 100644 --- a/content/publications/_index.md +++ b/content/publications/_index.md @@ -18,7 +18,7 @@ If you are interested in understanding how M²LInES is using machine learning to

- + DOI icon William Gregory, Ronald MacEachern, So Takao, Isobel R. Lawrence, Carmen Nab, Marc Peter Deisenroth, Michel Tsamados
Scalable interpolation of satellite altimetry data with probabilistic machine learning
Nature Communications 2024 DOI: 10.1038/s41467-024-51900-x @@ -30,13 +30,37 @@ If you are interested in understanding how M²LInES is using machine learning to

- + DOI icon Veronika Eyring, ... Laure Zanna
- SPushing the frontiers in climate modelling and analysis with machine learning
+ Pushing the frontiers in climate modelling and analysis with machine learning
Nature Climate Change 2024 DOI: 10.1038/s41558-024-02095-y

+
+
+ +
+

+ + Jerry Lin, Sungduk Yu, Liran Peng, Tom Beucler, Eliot Wong-Toi, Zeyuan Hu, Pierre Gentine, Margarita Geleta, Michael S Pritchard
+ Sampling Hybrid Climate Simulation at Scale to Reliably Improve Machine Learning Parameterization
+ Preprint submitted to JAMES DOI:10.22541/essoar.172072688.86581349/v1 +

+
+ +
+
+ +
+

+ DOI icon + Brandon G Reichl, Andrew T. Wittenberg, Stephen M Griffies, Alistair Adcroft
+ Improved Equatorial Upper Ocean Vertical Mixing in the NOAA/GFDL OM4 Model
+ Preprint submitted to ESS DOI:10.22541/essoar.170785794.47537760/v1 +

+
+
diff --git a/static/images/news/2406Falasca.jpeg b/static/images/news/2406Falasca.jpeg new file mode 100644 index 00000000..f470e36a Binary files /dev/null and b/static/images/news/2406Falasca.jpeg differ diff --git a/static/images/news/2407Gentine.png b/static/images/news/2407Gentine.png new file mode 100644 index 00000000..2c8e70a8 Binary files /dev/null and b/static/images/news/2407Gentine.png differ diff --git a/static/images/news/2407Reichl.png b/static/images/news/2407Reichl.png new file mode 100644 index 00000000..723ec0a2 Binary files /dev/null and b/static/images/news/2407Reichl.png differ