The sun is now being asked to file its own power forecast
the sun is now being asked to file its own power forecast: a latent diffusion model turns all sky images into probabilistic intra hour irradiance trajectories, a physics guided video generator predicts photovoltaic ramps from sky motion, and google’s weathernext 3 runs global hourly forecasts at five kilometers with energy variables for grid operators.
the panel now files its own output ledger.
Context
FarSky (arXiv, August 2026) is a generative intra-hour solar forecaster. A multi-task autoencoder learns a shared latent space for sky image reconstruction and irradiance estimation, and a latent diffusion model generates future latent states, so sampling gives probabilistic forecasts. It was built on a multi-year all-sky imager dataset from Plataforma Solar de Almeria in Spain. The authors report forecast skill up to 11 percentage points better than baselines and ramp detection F1 above 60%.
The physics-guided video work (arXiv, May 2026) pairs a sky video predictor, PhyDiffNet, with a ramp-aware PV output model, RaPVFormer. It forecasts ramp events up to 16 minutes ahead at 1 minute resolution and reports a 10% gain in Critical Success Index for ramp detection.
Google DeepMind's WeatherNext 3 (blog, September 3, 2026) adds real-time satellite data, hourly refreshes, higher resolution and clean energy variables. The post gives temperature and moisture at 5 km, other surface variables at 10 km and atmospheric variables such as wind at 25 km, and says it is about five times sharper than WeatherNext 2 on a 25 km grid. It names hub-height wind speeds and sun radiation levels as energy products for grid operators.
The note says WeatherNext 3 'runs global hourly forecasts at five kilometers', but Google's post gives 5 km only for temperature and moisture, 10 km for other surface variables and 25 km for atmospheric variables like wind, so the 5 km figure is narrower than the note's wording. The two sky-image papers are different studies on different sites and metrics: FarSky reports skill points and F1, while the PV video paper reports CSI and a 16 minute window. The note's 'all sky images into probabilistic trajectories' fits FarSky, and 'video generator predicts ramps from sky motion' fits the second paper. Both are preprints, and the WeatherNext figures come from Google's own blog.
Related work
- FarSky: Task-Aware Latent-Space Coupling for Generative Intra-Hour Solar Forecasting (arXiv) ↗Source for the latent diffusion all-sky forecaster and its skill and F1 numbers.
- High-Fidelity Full-Sky Video Prediction for Photovoltaic Ramp Event Forecasting (arXiv) ↗Source for PhyDiffNet, RaPVFormer and the 16 minute ramp lead.
- WeatherNext 3: Our most advanced global weather AI model (Google DeepMind blog) ↗Vendor source for the hourly, 5 km and energy variable claims.
- SkyGPT: Probabilistic Short-term Solar Forecasting Using Synthetic Sky Videos from Physics-constrained VideoGPT (arXiv) ↗An earlier probabilistic sky video approach to the same task.
Watch next
- Look for independent verification of WeatherNext 3 energy variables against measured solar and wind output. Test FarSky and the PV video model on the same site and metric.
Sources
Provenance
The note above is reproduced unedited from the original post, first published on Threads on 4 October 2026 at 01:47 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 →