The ocean oscillation is now being asked to keep its own seasonal diary
the ocean oscillation is now being asked to keep its own seasonal diary: a deep model called gl geoformer reads wind and three dimensional temperature across tropical basins and holds skill sixteen months out even when started in spring, a glass box dynamical model hindcasts the 2015 super el nino to nineteen months, and a bayesian cnn issues probabilistic nino 3.4 outlooks two years ahead.
the pacific now files its own enso forecast.
Context
GL-Geoformer (Science Advances) is a deep learning model for global tropical ocean and atmosphere prediction. It uses wind and three-dimensional temperature anomalies across the Indian, Atlantic and Pacific basins. The paper reports skillful ENSO predictions up to 16 months ahead when initiated in spring, and pacemaker experiments on the roles of the Indian Ocean Dipole and Atlantic Nino.
The glass-box model (npj Climate and Atmospheric Science, 2026) uses dynamical system deep learning to build a transparent multivariate ENSO predictor. The authors report skillful predictions up to 19 months ahead, a hindcast of the onset, intensity and decay of the 2015 to 2016 super El Nino more than a year in advance when initialized in boreal spring, and a possible super El Nino development in 2026 to 2027.
The Bayesian convolutional neural network paper (Environmental Research Letters, 2024) gives probabilistic Nino 3.4 predictions with a lead time of up to 9 to 10 months across all seasons, with bias correction.
The note says the Bayesian CNN issues outlooks two years ahead. The abstract and the discussion in the paper report a lead of about 9 to 10 months. The paper mentions two years only when describing a different dynamical model, SINTEX-F, so the two-year figure for the Bayesian model is unsupported by the paper, not refuted by a different one. The 16 and 19 month figures also measure different things: GL-Geoformer's 16 months is skill for forecasts started in spring, and the glass-box model's 19 months is stated for the model overall with the 2015 to 2016 event hindcast as a case. A single hindcast case does not show reliability across events, and the 2026 to 2027 super El Nino mention in the glass-box paper is a model outlook, not an observation.
Related work
- Tropical basin interactions reduce spring predictability barrier of ENSO in a deep learning model (Science Advances) ↗The GL-Geoformer paper with the 16 month spring-start result.
- Skillful and interpretable ENSO prediction via a glass-box dynamical-deep learning model (npj Climate and Atmospheric Science) ↗The source for the 19 month lead and the 2015 to 2016 hindcast.
- A probabilistic forecast for multi-year ENSO using Bayesian convolutional neural network (Environmental Research Letters) ↗The source for the 9 to 10 month probabilistic lead, despite the word multi-year in its title.
Watch next
- Check lead-time skill by season and over independent events for each model, and compare all three on one verification period. For the Bayesian model, look for any experiment beyond 10 months.
Sources
- Tropical basin interactions reduce spring predictability barrier of ENSO in a deep learning model (Science Advances)science.org
- Skillful and interpretable ENSO prediction via a glass-box dynamical-deep learning model (npj Climate and Atmospheric Science)nature.com
- A probabilistic forecast for multi-year ENSO using Bayesian convolutional neural network (Environmental Research Letters)iopscience.iop.org
Provenance
The note above is reproduced unedited from the original post, first published on Threads on 4 October 2026 at 03:02 IST. Sources are the papers and datasets the note draws on.
View the original post ↗Embed this note
More notes
The air is now being asked to keep its own ledger
the air is now being asked to keep its own ledger: ecmwf’s aifs compo becomes the first ai model to forecast atmospheric composition globally every three hours, cleanair simulates 365 days of pm2.5 over china in ten seconds, and a unified framework maps six pollutants at one kilometer across the whole country. the air now files its own composition report.
read the note →The current is now being asked to draw its own map
the current is now being asked to draw its own map: china’s langya 2.0 predicts six ocean phenomena including internal waves and mesoscale eddies, a deep net called wenhai resolves eddies globally with air sea flux formulas built in, and scripps infers surface currents from the way temperature patterns deform in satellite images. the ocean now files its own circulation report.
read the note →The soil is now being asked to report its own carbon
the soil is now being asked to report its own carbon: a nix color sensor paired with generative data augmentation predicts soil organic carbon without a lab, random forest drives 74 percent of soil health mapping studies, and sentinel 2 tracks five year carbon change across france and italy from 922 samples. the dirt now files its own carbon account.
read the note →