WeatherNext 3 team files

run_wn3.py runs hourly on GitHub Actions (.github/workflows/wn3_forecast.yml). It reads the newest complete 00/06/12/18 UTC run of Google DeepMind’s WeatherNext 3 from the free statistics bucket (gs://weathernext3_statistics_spatial), keeps:

• hourly station series to 48 h for every station in assets/data/asos_stations.json: station-calibrated 2-m temperature and dew point (mean, 10th and 90th percentile), 10-m wind speed (mean, p10, p90) and direction, hourly precipitation (mean, p90, IMERG-calibrated mean), sea-level pressure and cloud cover (total, low, middle, high);
• 6-hourly maps to 240 h on a 0.1° window over the US (20–54°N, 135–60°W): 2-m temperature, dew point and temperature spread (p90 − p10), 10-m and 100-m wind, sea-level pressure, 6-h precipitation (model and IMERG-calibrated) and cloud cover;

and publishes them to wn3/ in the Hugging Face dataset yanxingjianken/nh-forecast-6hourly.

Why the files are encrypted. Live WeatherNext forecasts are covered by the GDM Real-Time Weather Forecasting Experimental Data Terms of Use, which allow sharing only with clearly identified people, for educational purposes, through controlled distribution. Every data file is AES-256-GCM encrypted with a key derived (PBKDF2-HMAC-SHA256, 600 000 iterations) from the team passcode; the WxChallenge page asks for the passcode and decrypts in the browser. The passcode lives only in the GitHub secret WN3_PASSCODE and with the team. The Terms, the LEGALLY_BINDING_TERMS_OF_USE.txt notice and the copyright notice travel with the data.

Secrets: GCP_ADC_JSON (application-default credentials of the allowlisted Google account, from gcloud auth application-default login), WN3_PASSCODE, HF_TOKEN.

Pressure levels are not used: they exist only in the requester-pays 64-member bucket, where the downloader pays Google’s network charges.

Local checks (nothing uploaded; delete the output afterwards):

python run_wn3.py --list
WN3_PASSCODE=test python run_wn3.py --dry-run --out /tmp/wn3_test --ts-hours 3 --map-steps 6,12 --stations KBOS,KTPA