The river in the sky is now being asked to file its own flood watch
the river in the sky is now being asked to file its own flood watch: a regional ai model downscales vapor transport and precipitation to two kilometers over the western us and holds skill against ar recon dropsondes, a physically guided classifier flags atmospheric river flood days globally at an auc of 0.94, and a noaa experimental model ships five day water vapor outlooks to forecasters.
the moisture now files its own arrival forecast.
Context
West-WRF AI 2-km (arXiv, September 2026) is a stretched-grid AI model with 2 km resolution over the western United States and part of the Northeast Pacific and about 31 km elsewhere. It is fine-tuned on a 2 km regional reanalysis from CW3E and evaluated over winters 2020 to 2023 against gridded precipitation, rain gauges and AR Reconnaissance dropsondes. The abstract reports lower dropsonde errors and better categorical skill at the most extreme vapor transport threshold.
The physically guided classifier (Scientific Reports, 2026) predicts flood occurrence during atmospheric river conditions worldwide. Its best model reaches ROC AUC 0.94 against 0.81 for logistic regression, and the authors find about one third of large midlatitude floods occur under AR conditions.
NOAA's AR-AFS is an experimental Atmospheric River Analysis and Forecast System built on the Hurricane Analysis and Forecast System code. NOAA's Weather Program Office says it gives daily updates up to five days ahead for water vapor, its movement and 24 hour rain or snow, and that forecasters can see it on the DESI map.
These are three different things. The 2 km model is a weather forecast checked against dropsondes at extreme transport, not a general claim of skill. The 0.94 AUC is for classifying whether a flood occurs on an AR day, against a logistic baseline, and AUC does not show how well single events are timed. AR-AFS is a physics-based numerical system, not an AI model: NOAA describes it as built on HAFS code and coupled to an ocean model, so the note's 'noaa experimental model' is accurate but it should not be read as one of the AI systems. NOAA says early results do well on high precipitation events, with no scores given in the page.
Related work
- West-WRF AI 2-km: High-Resolution Prediction of Integrated Vapor Transport and Precipitation (arXiv) ↗Source for the 2 km western US model and dropsonde verification.
- Assessing global flood risk from atmospheric rivers through physically guided machine learning (Scientific Reports) ↗Source for the 0.94 AUC and the one third of large floods.
- Experimental atmospheric river forecast model goes online (NOAA Weather Program Office) ↗Source for AR-AFS and its five day outlooks.
- A regional high resolution AI weather model for the prediction of atmospheric rivers (npj Climate and Atmospheric Science) ↗The earlier 6 km version of the stretched grid AI model.
Watch next
- Read the West-WRF AI results by lead time and compare them to AR-AFS on the same events. Check whether the AUC of 0.94 holds in regions with few flood records.
Sources
- West-WRF AI 2-km: High-Resolution Prediction of Integrated Vapor Transport and Precipitation (arXiv)arxiv.org
- Assessing global flood risk from atmospheric rivers through physically guided machine learning (Scientific Reports)nature.com
- Experimental atmospheric river forecast model goes online (NOAA Weather Program Office)wpo.noaa.gov
Provenance
The note above is reproduced unedited from the original post, first published on Threads on 4 October 2026 at 02:02 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 →