Level II radar

Level II is what a WSR-88D actually produces: every elevation angle, every moment, unaveraged, at the radar’s native resolution - 250 m in range and about 0.5° in azimuth. Everything else you have seen is a product derived from it, usually flattened to one tilt.

What a volume actually contains

Measured on one real file: 21.7 MB, 25 cuts, 7.4 minutes.

The thing worth knowing, because it decides what a single volume can answer: dual-pol moments and velocity are never in the same sweep at low tilt. The surveillance cut carries reflectivity and the dual-pol moments; the Doppler cut carries reflectivity, velocity and spectrum width. Near 0.5° they alternate, about twenty seconds apart:

cut  1  0.45°  CFP PHI REF RHO ZDR      ← dual-pol, no velocity
cut  2  0.26°  REF SW VEL               ← velocity, no dual-pol
cut  3  0.39°  CFP PHI REF RHO ZDR
cut  4  0.44°  REF SW VEL
 …
cut 13-18      the whole low block again, at t+224 s

SAILS is not a feed. It is extra low-elevation sweeps inside the same volume file - nothing separate to fetch and no separate cadence to subscribe to. What it buys is revisit time at the lowest tilt:

Four samples of the lowest tilt in a 7.4-minute volume, in two pairs. Best case 42 seconds, worst case 182 - not the uniform one to two minutes that is often claimed for SAILS, and on average no faster than the mosaic publishes. What is different is that every sample is unaveraged, and that is the whole point.

The debris badge

Tornado Detected · confirmed is the one thing in WXA/SA that Level II is the only path to.

A tornado debris signature is lofted, non-meteorological material. It shows up as a collapse in the correlation coefficient - rain drops all look alike to a radar, and roof shingles do not.

Confirmed is a badge, not a rank

It renders as an annotation on the tier below - TORNADO DETECTED · confirmed - and never as something above it. Two measured reasons. Most tornadoes never produce confirmation anyone can see: about half of tracked rotations have no low-level dual-pol coverage at all. And confirmation means debris is already airborne, so it carries zero lead by construction. Ranking it top would teach people to discount the tier where the lead time actually lives.

Only the lowest tilt, and that is physics

The signature lives in the bottom 200 to 300 m and nowhere else. Measured at one tornado 12.8 km from its radar, correlation coefficient read 0.21 at 100 m and was already back to ordinary rain values by 1.4 km. Climbing the volume adds nothing, and it actively hurts: the same reading falls again in a hail core aloft, so a detector that scanned every tilt would “confirm” tornadoes off hail.

Forty kilometers, and not one more

Across 116 surveyed strong tornadoes the detection rate is 83% inside 40 km, 11% between 40 and 80, and zero beyond 80.

Intensity does not extend that. The limit is beam width, not height: a 0.95° beam is about 1.8 km across at 108 km, so even a violent tornado’s debris core only half-fills it - and ρHV is a correlation, which partial beam filling pulls straight back toward the rain value.

The gate

A volume is never pulled on a schedule. Only when our own tracker already says there is something to confirm:

storm scan (2 min)
  └─ cells, stitched, hysteresis          → Thunderstorm Risk / Detected
       └─ shear peaks → tracked couplets  → Tornado Risk / Tornado Detected
            └─ an ALERTING couplet ───────┐
                                          │  gate
            ┌─────────────────────────────┘
            ▼
       nearest WSR-88D, radar quality OK, range < 40 km
            ▼
       pull that radar's current volume, lowest-tilt dual-pol sweeps only
            ▼
       ρHV at the rotation, conditioned on reflectivity ≥ 40 dBZ  → + Confirmed

The gate is what makes the badge affordable. Over one outbreak window with 43 tornado reports, 117 tracked rotations came down to 35 volumes from four radars, because rotations share radars and share volumes. Pulling everything that window would have meant roughly 206 GB; the gated version reads about 760 MB.

A realtime chunk stream is already writing the same radar data to local disk as the radar produces it, and a volume control pattern runs its lowest tilts first, which are the only ones this badge looks at. A partial volume is not a compromise here, it is the point: the sweeps we want are readable while the volume is still running.

What it is honestly worth

On that outbreak day, end to end, the badge would have fired on 3 of 71 tornado reports. Precise, and rare.

And a third state matters as much as the other two. Confirmed is debris seen. Not confirmed is looked and found none. Could not look - outside 40 km, no volume published yet, or no low-level echo at the rotation - is neither, and it must never be read as the second.

What Level II does not do yet

Measuring a rotation’s diameter directly, and watching it contract and descend, is the obvious next thing: the resolution is there at 250 m in range and 112 m in azimuth at close range, and it would separate rotation size from rotation strength, which the mosaic cannot.

It waits on velocity dealiasing. Velocity folds at 23.2 m/s on every low-tilt Doppler cut, and a tornadic couplet routinely exceeds that, so the folding is worst exactly where the measurement matters. The debris badge is unaffected - correlation coefficient is not a Doppler quantity - which is why one ships and the other does not.