Fetching live threat data...

Clawd Front ๐ŸŒช๏ธ

AI-powered storm intelligence. Real-time tornado probabilities, SPC outlooks, and ML forecasting.

Clawd Front is an independent AI weather intelligence platform combining NWS/SPC data with custom machine learning models to provide best-in-class tornado probability forecasting.

Platform Capabilities
Built on years of storm data, ML research, and real-time data pipelines.
๐Ÿง 
0.919
ML Model v6.0 AUC
Best-in-class tornado probability model trained on ERA5 + HRRR + SPC storm events.
๐ŸŽฏ
Nadocast
ML Integration
Real-time Nadocast probabilistic tornado forecasts ingested and displayed for 60+ metro areas.
โšก
Live NWS
Warning Overlay
Real-time NWS tornado warnings and watches rendered on the live map with auto-refresh.
๐Ÿ“ก
Day 1โ€“3
SPC Outlook Integration
Official SPC convective outlooks with tornado probability risk level mapping.
๐ŸŒ
Vortex v2
Gridded ML Model
High-resolution gridded tornado probability output from our second-generation Vortex model.
๐Ÿ™๏ธ
60+
Metro Areas Tracked
Probabilistic tornado threat monitored for over 60 major metropolitan areas across the US.
How It Works
From raw data to actionable probability โ€” the Clawd Front pipeline.
๐Ÿ“Š
SPC GeoJSON
Day 1โ€“3 tornado outlook polygons
โ†’
๐ŸŒก๏ธ
IEM Mesoanalysis
Surface obs, STP, CAPE, SRH
โ†’
๐Ÿ–ฅ๏ธ
HRRR / ERA5
Reanalysis + NWP model output
โ†’
๐Ÿงฎ
ML Ensemble
XGBoost v6.0 + Vortex v2 grid
โ†’
๐ŸŽฏ
Probability Output
Metro tornado % + sig tornado %

๐Ÿ—บ๏ธ Live Tornado Map

Warnings: โ€” | Watches: โ€” Refreshing in 30s
โš ๏ธ Official NWS/SPC products displayed for reference. Always consult spc.noaa.gov for life-safety decisions.
๐ŸŒ€Loading SPC Day 1 outlookโ€ฆ
TSTM
MRGL
SLGT
ENH
MDT
HIGH
SPC Day 1 Categorical Outlook โ€” View on spc.noaa.gov โ†—

๐Ÿ“‹ Day 1 Issuance Context

The SPC Day 1 Convective Outlook is issued multiple times daily beginning at 0600 UTC. The tornado probability product highlights areas where tornado activity is possible within the next 24 hours.

Risk Categories:
โ€ข TSTM โ€” General thunder possible
โ€ข MRGL โ€” Marginal risk (5โ€“10%)
โ€ข SLGT โ€” Slight risk (10โ€“15%)
โ€ข ENH โ€” Enhanced risk (15โ€“30%)
โ€ข MDT โ€” Moderate risk (30โ€“45%)
โ€ข HIGH โ€” High risk (45%+)
๐ŸŒ€Loading SPC Day 2 outlookโ€ฆ
SPC Day 2 Convective Outlook โ€” View on spc.noaa.gov โ†—
๐ŸŒ€Loading SPC Day 3 outlookโ€ฆ
SPC Day 3 Convective Outlook โ€” View on spc.noaa.gov โ†—
Clawd Front AI Outlook

Clawd Front AI Outlook โ€” ML-generated tornado probability overlay

Loading Nadocast data...
0.919
AUC-ROC

Clawd Front ML v6.0

Our flagship tornado probability model, trained on ERA5 reanalysis + HRRR NWP output + SPC storm event records spanning 2000โ€“2025. XGBoost ensemble with isotonic regression calibration. Evaluated on held-out storm days, achieving a class-leading AUC of 0.919.

ERA5 + HRRR + SPC Storm Events | XGBoost + Isotonic Calibration | 2000โ€“2025

๐Ÿ“ˆ Top Feature Importances

neighbor_sig_cnt
0.27
synoptic_forcing
0.12
STP (sig tornado param)
0.09
CAPE (surface-based)
0.07
SRH (0-3km storm relative)
0.06
LCL height
0.05
0-6km shear (bulk)
0.04
Model Version History
Version AUC-ROC Key Changes Training Data
v5.0
0.814
Initial XGBoost baseline ERA5 2000โ€“2022
v5.1
0.882
HRRR integration, extended features ERA5+HRRR 2000โ€“2023
v5.2
0.882
Isotonic calibration, feature pruning ERA5+HRRR 2000โ€“2023
v6.0 โ˜…
0.919
neighbor_sig_cnt feature, synoptic forcing, Vortex v2 grid ensemble ERA5+HRRR+SPC 2000โ€“2025
Storm Threat Map
Storm Threat Map

CF ML v6.0 storm threat probability output

Nadocast Comparison
Nadocast Comparison Map

Nadocast metro probability overlay for comparison

๐Ÿ”ฌ Methodology

  1. Data Ingestion: SPC storm event reports, ERA5 reanalysis fields (CAPE, SRH, shear, LCL, STP, LI), and HRRR NWP output are pulled and aligned to a common grid for each forecast day.
  2. Feature Engineering: Spatial clustering of significant tornado reports ("neighbor_sig_cnt") is computed within 150km radius. Synoptic forcing index is derived from 500mb height anomaly and upper divergence fields.
  3. Model Training: XGBoost classifier trained on storm-day samples (positive = any EF1+ tornado within 25mi of grid point). 80/20 temporal split ensures no data leakage from held-out evaluation years.
  4. Calibration: Isotonic regression calibration applied post-training to align model output probabilities with observed climatological base rates.
  5. Vortex v2 Ensemble: Gridded Vortex v2 output is blended with the metro-point model for spatial coverage and uncertainty quantification across the full CONUS grid.
  6. Verification: Ongoing Brier Score and AUC verification against NWS LSR reports. Model is retrained when AUC drops below 0.87 on a rolling 90-day window.
โš ๏ธ UNOFFICIAL AI FORECAST โ€” Not an NWS/SPC product. Do not use for life-safety decisions.
Clawd Front AI Outlook
CF AI Outlook

Machine learning tornado probability forecast

๐Ÿ“… Next 24 Hours

Generating forecast...

๐Ÿ“Š Top Threat Areas

๐Ÿ” Forecast Methodology

This AI forecast is generated client-side by synthesizing Nadocast probabilistic metro forecasts with the Clawd Front ML v6.0 model output. It is not a product of the National Weather Service or Storm Prediction Center. For official forecasts, warnings, and watches, always consult weather.gov and spc.noaa.gov.

Clawd Track โ€” Live Storm Nowcast

Reports (6h): โ€” Refreshing in 60s

Legend

Confirmed Tornado
ASOS Station
NEXRAD: Iowa State IEM | Warnings/Watches: NWS
Tornado Warning Svr Tstorm Warning Tornado Watch Svr Tstorm Watch
LSR: NWS Local Storm Reports
๐ŸŒช๏ธ Confirmed Tornado Reports โ€” Last 6 Hours Loadingโ€ฆ
Loading storm reportsโ€ฆ