Forecast vs reality
How our forecast compares to real WSPR spots on fixed benchmark
paths, updated every 3 h. Same numbers we look at ourselves — including
the ones that make us look bad.
loading…
Per band — last 30 days
Benchmark paths
How to read this — and where it lies to you.
- Mean bias is forecast minus observed SNR in dB: positive means we
are optimistic. RMSE is the typical size of a miss, sign ignored.
- Openness hit-rate asks a yes/no question only: did we predict any
mode at least “likely”, and were there really at least 3 WSPR
spots on that band in the hour? It says nothing about how far off the dB were.
- Survivorship. A WSPR average only exists where somebody was
transmitting and somebody else was listening. Weak or dead bands report
nothing rather than reporting a low number, which flatters the observed side.
- Antennas. The forecast assumes dipoles at both ends at 100 W;
real WSPR stations run whatever they own. Spot SNR is normalized for power,
never for antenna gain, height or local noise.
- Grid fields, not stations. Reality is aggregated over the 2-char
Maidenhead field at each end and over both directions, so it is a regional
average, not this exact circuit.
- 160m is outside VOACAP. The engine is specified for 2–30 MHz;
1.8 MHz numbers are extrapolation and should be read as indicative only.
- A 30-day window means the numbers move slowly and a quiet fortnight can
hide a bad week. Thin bands (small n) are noise, not evidence.