CoolFace
Modelpublic

aerostratospheric/uogw-weather-forecast

sourceHugging Facecc-by-4.0updated 1d agoView on Hugging Face
0likes
Model Card

UOGW Weather Forecast Suite

Publisher: Aerostratospheric / Midwest Stratospheric Data Systems Library: scikit-learn (.joblib heads) · helper: inference.py Site focus: Casey, Illinois + UOGW surface context Sibling data: `aerostratospheric/uogw` · `casey-openmeteo-hourly`

Multi-horizon research forecast heads (temperature, precip occurrence, gust, MSLP, RH, surface regime).

What this is / is not

IsIs not
Open tabular forecast screening for Casey / UOGWOfficial NWS / NOAA forecast product
Trained on Open-Meteo hourly archive (~60 days) + lag/roll featuresA live operational forecast desk
sklearn joblib artifacts + small loaderA foundation / transformers weather model

Trained 2026-09-18T15:34:31Z. Hub re-touch 2026-09-23 — weights not re-trained that day. Intended retrain cadence: ~12:15 UTC after UOGW package sync.

Heads

HeadHorizonTask
temp_1h / temp_3h / temp_6h / temp_24h1–24 hTemperature regression (°C)
precip_1h / precip_6h1 / 6 hPrecip occurrence classifier
gust_1h1 hWind gust regression
mslp_3h / rh_3h3 hPressure / RH regression
surface_regime—5-cluster surface regime

Feature lists and holdout MAE / R² / accuracy: metrics.json.

How to load

python
import json
from pathlib import Path
import joblib
from huggingface_hub import snapshot_download

root = Path(snapshot_download("aerostratospheric/uogw-weather-forecast"))

# Option A — config-driven
cfg = json.loads((root / "config.json").read_text())
heads = {h: joblib.load(root / "models" / f"{h}.joblib") for h in cfg["heads"]}

# Option B — packaged helper (same directory layout)
# import importlib.util
# spec = importlib.util.spec_from_file_location("inf", root / "inference.py")
# inf = importlib.util.module_from_spec(spec); spec.loader.exec_module(inf)
# heads = inf.load_all()

# Predict with a row matching metrics.json features for that head
# yhat = heads["temp_1h"].predict([feature_row])

Requires scikit-learn, joblib, huggingface_hub.

Data provenance & freshness (2026-09-23 CT)

  • —Training source: Open-Meteo Casey hourly archive + UOGW surface context (see metrics.json → source)
  • —Dataset sync: aerostratospheric/uogw 2026-09-23T18:29:53Z
  • —Ground KILCASEY47 on hub overlay: ~2026-09-23 12:05 CT
  • —Radiation tip gmc-800/latest.json: still 2026-09-16 (stale — not used as a live forecast input claim here)

Details: FRESHNESS.md.

Related artifacts

ArtifactURL
UOGW scientific suitehttps://huggingface.co/aerostratospheric/uogw-scientific-suite
Tornado trendshttps://huggingface.co/aerostratospheric/uogw-tornado-trends
Open-models indexhttps://huggingface.co/aerostratospheric/msds-open-models
UOGW Git tiphttps://github.com/Midwest-Stratospheric/Unified-Open-Global-Weather
Sitehttps://midwestsds.com
Introhttps://calendly.com/aerostratospheric/15min

Citation

Aerostratospheric / Midwest Stratospheric Data Systems (2026).
UOGW Weather Forecast Suite. Hugging Face.
https://huggingface.co/aerostratospheric/uogw-weather-forecast

License: CC BY 4.0. Attribute Open-Meteo and agency sources in derived work.

Awareness — Casey open weather + UOGW (2026-09-25)

This forecast packaging sits beside the live open desk:

  • —Data Hub / xDataHub: https://www.midwestsds.com/msds-data-hub.html
  • —UOGW developer overview: https://midwestsds.com/contact/knowledgebase.php?article=93
  • —Casey Open-Meteo hourly dataset: https://huggingface.co/datasets/aerostratospheric/casey-openmeteo-hourly
  • —Flown archive (awareness): Sep 12 x1Albatross 59,000 ft; Sep 19 X2Griffon maiden burst 90,356 ft — https://www.midwestsds.com/flights.html · framing https://midwestsds.com/contact/knowledgebase.php?article=94

Collaboration / data partnership welcome. No in-kind asks.

Disclosure: Aerostratospheric owned model card.