Squall · the open scoreboard

Headline numbers

Reliability diagram

Predicted versus observed, binned by tier.

The dashed diagonal is perfect calibration. Each dot is a bin of alerts; the y-axis is how often that bin actually verified as a tornado. Closer to the diagonal is better. Dot size is proportional to the number of alerts in the bin.

The upper-right bins (predicted 0.7 to 0.9) sit visibly below the diagonal. Every one of those "misses" was a verified severe storm (hail, downburst, damaging wind) that did not produce a tornado. That gap is the radar-only ceiling on tornado-vs-severe discrimination; the next reduction has to come from new sensing (GLM lightning jump, ProbSevere v3, dual-pol), not threshold tuning.

By EF bucket

EF	n	POD	FAR	Mean lead (min)	Brier

Tornado: driver attribution

Which signals actually discriminate. Rotation alone fires on 94% of FALSE alarms.

Loading driver attribution.

Driver	Category	TP rate	FP rate	Lift

Tornado: v2 probability model

L2-regularized logistic on canonical drivers. v1 (score alone) is near-random at AUC 0.52; v2 lands at AUC 0.90.

Loading tornado v2 probability model.

	v1 score	v2 logistic	Gain

Favor tornado

Favor false alarm

Tornado: EF3+ critical tier

Zero unique misses at the casualty tier. Squall and NWS tied on EF3+.

Loading EF tier breakout.

Tier	n	POD	POD when NWS caught	Median lead	Unique misses

Live feed · most recent 10

Tornado: lead-time distribution vs NWS

Same tornadoes, same EF mix, head-to-head against the issuing NWS office.

Loading lead-time distribution.

Bin	Squall (n)	NWS (n)

Tornado: storm-mode lead-time split

Brotzge and Erickson 2009 (NWS 25 min supercell, 7 min QLCS) on Squall's own data. Median 62 min on classified supercells.

Loading storm-mode lead-time split.

Mode	n	p25	p50	p75	Literature reference

Hurricane: rapid-intensification predictor

Squall predicts rapid intensification (30+ kt strengthening in 24 h) hours before NHC.

Verified retrospectively on the 2015-2024 Atlantic + East-Pacific HURDAT2 best-track corpus. Model: logistic regression on trajectory features only (current intensity, prior 12 / 24 h intensity change, latitude, motion, calendar position). Trained on 1990-2014 storms (857 storms), held-out on 2015-2024 (436 storms, 12,830 observation points, 653 confirmed RI events).

Reference: NHC SHIPS-RII operational baseline, DeMaria et al 2021 NHC RI verification report (AUC ~0.78-0.83, POD ~0.45, FAR ~0.70). Squall's gain over the operational baseline is driven by perfect-history retrospective evaluation; real-time performance against ATCF advisories is the next verification milestone. Inner-core lightning (GLM), SST, and shear features are not yet in the fit; we expect those to materially improve real-time POD.

Hurricane: genesis (5-day formation probability)

v1 is a climatology-only logistic. Per-disturbance NHC TWO ingest is the v2 milestone; documented limitation, not yet shipped.

Loading genesis verification.

Hurricane: track forecast error

v2 TVCN consensus ingest. Competitive with NHC OFCL at every horizon.

Loading track verification.

Horizon	n	v1 (CLIPER)	v2 (TVCN)	NHC OFCL	v2 gain	v2 vs OFCL

Hurricane: intensity forecast error

v2 statistical climatology beats published SHIFOR at every horizon. Closing the gap to NHC OFCL.

Loading intensity verification.

Horizon	n	v1 (persistence)	v2 (climatology fit)	SHIFOR	NHC OFCL	v2 gain

Hurricane: per-address storm surge

Powell-Reinhold + Asbury parametric verified against NOAA tide-gauge surge.

Loading surge verification.

Intensity	n	Mean observed (ft)	Mean predicted (ft)	MAE (ft)	Bias (ft)

Hail: lead-time distribution vs NWS

Same hail reports, same lat/lon, head-to-head against the issuing NWS office's SV warning.

Loading hail lead-time distribution.

Bin	Squall (n)	NWS (n)

Hail: precision-recall on the radar volume

Every cell scored, matched against SPC reports. Real FAR, real POD, no cherry-picking.

Loading hail PRC.

Threshold	TP	FP	POD	FAR	Precision

Sensitivity to SPC match window (at threshold 0.50):

Radius / Window	POD	FAR	Precision

Hail: v2 polarimetric probability model

Continuous Z + ZDR + KDP logistic. v2 cuts the Brier reliability gap 32% and lifts precision 4pp at matched recall.

Loading hail v2 polarimetric model.

	v1 gated	v2 logistic	Gain

Learned weights

PR curve (v2)

Hail: per-cell size + severe probability

Witt 1998 SHI proxy + Park 2009 polarimetric, calibrated against SPC severe-report ground truth.

Loading hail verification.

Hail size bucket	n	Mean actual (in)	Mean predicted (in)	MAE (in)	Bias (in)