Links lost in the flare hour, by flare class and band
Median share of dayside HF links that failed to decode during the hour of the flare peak, versus the same hour on the 14 preceding flare-free days. Interquartile range in brackets. Sample: M1–M4 — 683 flares · M5–M9 — 61 flares · X1–X4 — 46 flares.
| Flare class | 7 MHz | 10 MHz | 14 MHz | 18 MHz | 21 MHz | 28 MHz |
|---|---|---|---|---|---|---|
| M1–M4 moderate flare — several per week at solar maximum · n=683 | −33% [−61…−1%] · 5 dB | −18% [−39…−-2%] · 3 dB | −8% [−28…−-7%] · 2 dB | −6% [−30…−-14%] · 0 dB | −7% [−31…−-17%] · 1 dB | −1% [−32…−-36%] · 1 dB |
| M5–M9 strong M flare — a few per month at solar maximum · n=61 | −50% [−80…−21%] · 9 dB | −40% [−63…−14%] · 7 dB | −26% [−45…−3%] · 5 dB | −23% [−48…−0%] · 3 dB | −20% [−48…−-3%] · 3 dB | −11% [−50…−-23%] · 2 dB |
| X1–X4 X-class flare — a few per year · n=46 | −51% [−77…−28%] · 11 dB | −53% [−75…−28%] · 9 dB | −43% [−67…−22%] · 9 dB | −40% [−62…−14%] · 6 dB | −45% [−66…−16%] · 7 dB | −32% [−61…−-7%] · 5 dB |
Colour is a reading aid only — the number carries the meaning: red ≥40% lost, amber ≥20%. Frequency ordering (7 MHz worst, 28 MHz least) is the D-layer absorption law (∝ 1/f²), which is why we trust the instrument: it reproduces the physics, not noise.
How fast HF comes back
Share of links lost, hour by hour from one hour before the peak (T−1) to six hours after (T+6). 14 MHz, all dayside sectors. Each panel is one flare class; bars are medians.
By region of the sunlit side
Same measurement, split by where the receiving stations were. 14 MHz, flare hour. Regional samples are smaller than the global one — treat rows with n < 10 as indicative.
| Receiver region | M1–M4 | M5–M9 | X1–X4 |
|---|---|---|---|
| All dayside sectors | −8% n=683 | −26% n=61 | −43% n=46 |
| Atlantic & Europe (60°W–30°E) | −5% n=385 | −17% n=37 | −38% n=27 |
| Americas (150°W–90°W) | −9% n=397 | −26% n=38 | −43% n=25 |
| Asia & Pacific (30°E–150°W) | −14% n=383 | −32% n=31 | −47% n=29 |
What this is for
- Oceanic and polar operations. HF is still the primary voice link where VHF and, on some routes, satcom are not available. This table says how much of it a given flare removes and for how long — as a measured median, not a model.
- Space-weather procedures. Turn a NOAA R-scale alert into a concrete expectation: an R3 (X1) event costs roughly half of the 14 MHz links on the sunlit side for one to two hours; an M1 costs a tenth.
- Resilience claims. Independent, reproducible measurement to test vendor claims about HF or HF-alternatives during flares.
Method and limits
- Flares — NOAA GOES-16/18 XRS flare summaries; class from peak X-ray flux. One flare per hour (strongest kept).
- Links — WSPRnet weak-signal beacon spots over ≥1000 km, counted per hour, per band, per 30° longitude sector of the receiver. Only sectors within ±60° of the sub-solar longitude (roughly ±4 h of local noon) enter the flare sample: the flare only affects the sunlit ionosphere.
- Baseline — the median of the same sector/band/hour on the 14 previous days, excluding hours within six hours after any other M-class-or-stronger flare. Rows with fewer than five clean baseline days or fewer than 200 baseline links are dropped.
- “Links lost” — 1 − (links in flare hour ÷ baseline links). Median across flares; the interquartile range shows the spread between flares.
- Absorption in dB — the shift of the baseline signal-to-noise distribution needed to reproduce the observed link count (four SNR bins: ≥−28, ≥−20, ≥−10, ≥0 dB). It is an estimate, good to about ±3 dB, not a receiver measurement.
- Limits — beacon density is high over Europe and North America and thin over the Pacific and Africa: regional rows there have small n and are marked. Hours are UTC bins, so a flare peaking at :50 shares its hour with 50 quiet minutes — real peak-minute losses are larger than the hourly medians shown. X5+ is withheld. Four flares reached the bucket after the archive was extended, but they carry only 11-15 hourly observations per band, and on 28 MHz the measured survival comes out at 1.003 — more links surviving than on a quiet day, which is physically impossible and marks the cell as noise. A row that cannot be wrong in a checkable way is not a measurement, so we do not publish it. Coverage: 9 Feb 2024 → 30 Apr 2026, with April 2024 absent from the WSPR archive entirely.
- How many flares, exactly — 785 / 73 / 49 flares (M1–M4 / M5–M9 / X1–X4), plus 4 in X5+, produced at least one usable measurement. The headline table is stricter: each of its cells needs a resolvable quiet-day baseline, so those rest on 683 / 61 / 46. Cell samples vary a lot — from single digits in the thinnest regional cell to 683 in the densest — so read the n shown next to each row rather than a single global number.
- The hour before the peak is not quiet — and the figures above do not subtract it. At T−1 the same measurement already shows 2.1 dB on 7 MHz for M1–M4 and 3.5 dB for M5–M9, because a flare is already rising an hour before it peaks and weaker flares often sit in the tail of an earlier one. Subtracting that pre-flare hour, the increment attributable to the flare itself is about 2.9 dB for M1–M4 and 5.8 dB for M5–M9 on 7 MHz — meaningfully less than the peak-hour figure. For X-class the correction barely matters: T−1 is 0.9 dB against 10.8 dB at the peak, so 9.9 of those 10.8 dB are the flare. Read the M-class rows as "the state of the band during the flare hour", not as "what the flare added".
- Does more data change the answer? — the series grew from 16 to 26 months in August 2026, raising the X-class sample by half. The published figures barely moved: X-class loss at 14 MHz went from 9.1 to 9.2 dB, M-class at 7 MHz from 5.4 to 5.0 dB. A measurement that survives a 50% larger sample without shifting is one you can plan against.
- What "dB" means here — not a receiver measurement of absorption. It is the shift of the quiet-day signal-to-noise distribution that would reproduce the observed drop in decoded links, read off four SNR thresholds (−28/−20/−10/0 dB). The estimator has a hard floor at 0 and saturates around 38 dB, so in the worst cells (X-class on 7–10 MHz, where half the links are already gone) the published number is a lower bound, not a value.
Get the data
The full table (all classes × bands × sectors × hours) is a public JSON endpoint, no key required. Filters: band, sector, class.
curl "https://api.flysafe.zone/v2/space/hf-flare-impact?band=14§or=all&class=X1-X4" Free to cite with a link. Attribution to carry with the numbers: NOAA NCEI GOES-16/18 XRS flare summaries (US Government public data); Derived hourly aggregates of WSPRnet spots (amateur-radio community data, wsprnet.org). Computed 22 Aug 2026 UTC.
Cite this data:
This page is a measured statistical reference, not an operational advisory. For live space-weather alerts use NOAA SWPC and your ANSP's procedures.