orangekraken◉radio

RF Forecast — next 6 hours, frozen and graded

Nine shortwave bands, one question each: will a receiver within a couple of hundred kilometres of Great Falls decode a WSPR beacon from at least 3,000 km away during the next six-hour UTC window? The number is written down before the window opens and cannot be changed; the next engine run asks wspr.live what actually happened and keeps the score. Today the number is the climatology — last month's base rate for this time of day — and it stays that way until a model beats it on a graded record.

as of 2026-08-31 · committed snapshot · Great Falls, MT (city-level)

In plain English

Every day the engine freezes, per shortwave band, the probability that a receiver within a couple of hundred kilometres of Great Falls decodes a WSPR beacon from at least 3,000 km away during the next six-hour UTC window. The number is the trailing-30-day base rate of that same time of day on that band, read from wspr.live; the next day the same database grades it 1 or 0 and the Brier score is kept beside the climatology's.
What you'll see: A per-band, six-hour radio forecast for this corner of Montana that is demonstrably better than looking at last month — with every hit and miss on the public record.

Next 6-hour window · 2026-09-01 00:00Z → 06:00Z

frozen 2026-08-31T19:03:34Z · site DN47 Great Falls, MT (city-level)

Per band: P(a receiver in the DN47 neighbourhood decodes a WSPR path ≥ 3,000 km in the window). p = climatology (trailing 30 days, same time of day, wspr.live); the null beside it is the same number — the model IS the null until it earns in.

  • 160 m
    0.50
    frozen

    160 m → ≥3,000 km from Montana: 0.5 (coin — 4 day(s) of history)

  • 80 m
    0.50
    frozen

    80 m → ≥3,000 km from Montana: 0.5 (coin — 4 day(s) of history)

  • 40 m
    0.50
    frozen

    40 m → ≥3,000 km from Montana: 0.5 (coin — 4 day(s) of history)

  • 30 m
    0.50
    frozen

    30 m → ≥3,000 km from Montana: 0.5 (coin — 4 day(s) of history)

  • 20 m
    0.50
    frozen

    20 m → ≥3,000 km from Montana: 0.5 (coin — 4 day(s) of history)

  • 17 m
    0.50
    frozen

    17 m → ≥3,000 km from Montana: 0.5 (coin — 4 day(s) of history)

  • 15 m
    0.50
    frozen

    15 m → ≥3,000 km from Montana: 0.5 (coin — 4 day(s) of history)

  • 12 m
    0.50
    frozen

    12 m → ≥3,000 km from Montana: 0.5 (coin — 4 day(s) of history)

  • 10 m
    0.50
    frozen

    10 m → ≥3,000 km from Montana: 0.5 (coin — 4 day(s) of history)

Scoreboard

graded
0 · void 27 · open 9
Brier (forecast)
—
Brier (climatology)
—
bands beating climatology
0 / 3 needed
earned in
false

The bar: climatology holds p until a model earns in: >= 20 graded windows AND Brier better than climatology across >= 3 bands, on the frozen record, before any non-climatology model may write p; promotion is human-only

prereg/2026-08-28-skywatch-rf-forecast · frozen 2026-08-28

How to read a row

  • bar — the frozen probability; faint while it is a coin (fewer than 10 days of history → 0.5).
  • null — the pre-registered climatology it must beat; identical today by construction.
  • frozen — the window has not closed; graded ✓/✗ — wspr.live said yes/no; void — no receiver near us decoded anything, so nothing was observable (never a miss).
▸Learn more
There is no model yet. The probability shown IS the climatology, so the forecast cannot beat its own null until a model earns in on the pre-registered bar (>= 20 graded windows, Brier better than climatology on >= 3 bands). Until then the tiers are climatology, not skill. A window in which no receiver near us decoded anything is void, not a miss. Very few WSPR receivers report from north-central Montana, so early base rates are low and the coin (0.5) holds while history is short.

What can I hear right now?

No model has earned in. These tiers are the climatology sorted by probability — a watermark, not a rating. The names Excellent / Good / Poor are withheld until the bar is met.

tier 1 p ≥ 0.67

no band in this tier

tier 2 0.33 ≤ p < 0.67
  • 160 m0.50
  • 80 m0.50
  • 40 m0.50
  • 30 m0.50
  • 20 m0.50
  • 17 m0.50
  • 15 m0.50
  • 12 m0.50
  • 10 m0.50
tier 3 p < 0.33

no band in this tier

The rulers to try first: WWV Fort Collins, CO (908 km, 2.5/5/10/15/20 MHz) · CHU Ottawa, ON (2,713 km, 3.33/7.85/14.67 MHz). The forecast is about ≥3,000 km WSPR paths; a beacon closer than that arriving says nothing about the call, and a beacon farther than that is one real path the sky allowed.
▸Learn more
The tiers become ratings only after the pre-registered bar: at least 20 graded windows with a Brier score better than climatology on at least three bands.
Provenance — the exact wspr.live SQL

Source: wspr.live (db1.wspr.live ClickHouse HTTP, FORMAT JSON). Neighbourhood squares: DN47, DN46, DN48, DN37, DN57, DN36, DN38, DN56, DN58. Record: omen-wealth data/skywatch_rf/forecast.jsonl (append-once; a frozen row is never rewritten).

Climatology (the p and the null)
SELECT toDate(time) AS d, band, max(distance) AS kmMax, count() AS n FROM wspr.rx WHERE time >= '2026-08-02 00:00:00' AND time < '2026-09-01 00:00:00' AND toHour(time) >= 0 AND toHour(time) < 6 AND substring(rx_loc,1,4) IN ('DN47', 'DN46', 'DN48', 'DN37', 'DN57', 'DN36', 'DN38', 'DN56', 'DN58') GROUP BY d, band ORDER BY d, band FORMAT JSON
Grade (run once the window has closed by 60 min)
SELECT band, max(distance) AS kmMax, count() AS n FROM wspr.rx WHERE time >= '2026-09-01 00:00:00' AND time < '2026-09-01 06:00:00' AND substring(rx_loc,1,4) IN ('DN47', 'DN46', 'DN48', 'DN37', 'DN57', 'DN36', 'DN38', 'DN56', 'DN58') GROUP BY band ORDER BY band FORMAT JSON

Rule: outcome = 1 if kmMax for band 1 in the window >= 3000 else 0; VOID (unobservable, never a miss) when the query returns no row on any band (no receiver in the neighbourhood decoded anything); graded no earlier than 60 min after the window closes (upload lag)

rf-climatology-v1: trailing-30d base rate of the same 6 h window-of-day x band (wspr.live), shrunk to 0.5 while base has < 10 days; = the null; fixed

▸What do these words mean?
WSPR — Weak Signal Propagation Reporter — thousands of hobby transmitters send a tiny, slow, timestamped signal on fixed frequencies, and every receiver that decodes one reports it to a shared public database. Each report is one proven radio path: who, from where, heard where, how strongly. We read that database, and our own receiver adds its reports to it.
Maidenhead grid — The Maidenhead locator: the world cut into squares named with letters and numbers (DN47 is the square that holds Great Falls). Radio amateurs report locations this way, so every WSPR report already carries a rough position.
Propagation — How a radio wave gets from transmitter to receiver: straight along the ground, bounced off the ionosphere, bent through the lower atmosphere, or scattered. The whole subject of this site.
Skip — A bounced path — the signal rises, reflects off the ionosphere and comes down far away. The ground between the transmitter and the first landing point hears nothing: the 'skip zone'.
MUF (maximum usable frequency) — The highest frequency that will still bounce between two given places at this moment. Above it, the signal punches through the ionosphere into space. It rises by day, falls at night and climbs across the solar cycle.
Ionosphere — A layer of the upper atmosphere (about 60–1,000 km up) that sunlight turns into a weak electrical mirror. Shortwave radio bounces off it, which is why a signal from Montana can land in Japan. Its height and strength change with the Sun, the hour and the season.
Beacon — A transmitter that sends a known signal from a known place on a schedule, so anyone can use it as a ruler. WWV (Colorado), WWVH (Hawaii) and CHU (Ottawa) are time-signal stations we use exactly this way.
SKYWATCH LIVE…