Solar Flares, Forecasting, and What It Means for Our Bands
Solar Flares, Forecasting, and What It Means for Our Bands
A flare forecast describes risk. GOES X-rays and ionospheric observations tell you what is happening to the radio path now.
Every operator knows the scene: a useful daylight HF path weakens abruptly, stations disappear, and the waterfall empties. A solar X-ray flare can do that by increasing D-region ionisation and absorption on the sunlit side of Earth. But a paper that estimates tomorrow’s flare probability is not yet a prediction of your band, path or blackout.
That is why I find the model by Daye Lim, Yong-Jae Moon and Hyun-Jin Jeong interesting. It asks a disciplined statistical question: given the total unsigned magnetic flux of a visible active-region patch, what is the probability of at least one C1.0-or-stronger or M1.0-or-stronger flare in the next day, and what historical range of daily peak soft-X-ray flux has occurred for comparable flux?
What the Lim, Moon and Jeong Model Predicts
The study combines SOHO/MDI and SDO/HMI magnetic patches with GOES 0.1–0.8 nm flare records. It uses one input feature: a μ-corrected estimate of total unsigned radial magnetic flux. The authors fit power-law relationships between that flux and:
- the daily occurrence rate of at least one C1.0+ flare;
- the daily occurrence rate of at least one M1.0+ flare; and
- the historical distribution of the strongest GOES X-ray flare in a day, reported through a mean, standard deviation and upper envelope.
The input units are daily TARP or HARP magnetic patches sampled at 00:00 TAI while their central longitude is within ±60° of the solar central meridian. A TARP or HARP is not automatically one NOAA-numbered sunspot region: the paper notes that a patch can contain zero, one or several NOAA active regions. Keep that identity boundary when interpreting its worked example.
The Dataset and Time Split Matter
The paper uses 40,650 daily patch samples spanning two solar cycles. The split is chronological, which is much more informative than mixing future and past samples randomly:
- Model development: 28,533 samples representing 6,161 distinct patches from 23 April 1996 through 31 March 2013.
- Held-out test: 12,117 samples representing 2,363 distinct patches from 1 April 2013 through 21 May 2021.
- C1.0+ test outcome: 1,152 event-days and 10,965 non-event-days, a climatological event rate near 10%.
- M1.0+ test outcome: 169 event-days and 11,948 non-event-days, a climatological event rate near 1%.
The model also carries measurement and association limits. MDI and HMI fluxes were aligned, line-of-sight flux was converted with a simple μ correction, and flares were assigned to a patch by position after differential-rotation correction. The authors explicitly note that the μ correction can be weak for regions with strong horizontal field.
Why 93% and 99% Are Not Forecast Certainty
The reported 0.93 and 0.99 values are maximum categorical accuracy after converting probabilities into yes/no forecasts at selected thresholds. They are not statements that a predicted flare probability is “93% accurate” or “99% accurate.” With only about 10% C1.0+ event-days and 1% M1.0+ event-days in the test set, raw accuracy is dominated by correct non-event forecasts.
The fuller scorecard is more revealing:
| Held-out result | C1.0+ | M1.0+ | What it says |
|---|---|---|---|
| Climatology event rate | 0.10 | 0.01 | Events are imbalanced, especially for M1.0+. |
| Maximum accuracy | 0.93 | 0.99 | Thresholded fraction correct; heavily influenced by non-events. |
| Maximum TSS | 0.71 | 0.80 | Event discrimination at thresholds near 0.09 and 0.02. |
| Brier skill score | 0.38 | 0.13 | Probabilistic improvement over test-period climatology. |
| Critical success index | 0.44 | 0.17 | Event hits relative to hits, misses and false alarms. |
Threshold choice changes the operational behaviour. A threshold selected to maximise accuracy is not the threshold that maximises TSS, and a contest station may assign a different cost to a missed flare than to a false alarm. The paper also found that some metrics and their best thresholds change with the climatological event rate across the solar cycle.
The predicted magnitude output is broad by design. On the held-out data, about 62% of observed daily maxima fell inside the model’s ±1σ interval and about 97% inside ±2σ; all were below its historical maximum envelope. A wide interval can describe uncertainty honestly, but it is not a precise forecast of the next flare class.
The HARP 8088 and 8096 Example
The paper gives an illustrative calculation for 30 March 2022, after the end of the formal test interval. These identifiers are HARP magnetic-patch numbers, not NOAA active-region numbers.
For HARP 8088, the model gave 71.1% probability of C1.0+ and 12.5% probability of M1.0+ within one day. The patch produced 15 C-class flares and a strongest flare of X1.4. HARP 8096 received similarly high probabilities—73.9% and 13.6%—but produced no flare. Its estimated 26% chance of remaining below C class was not zero.
Both outcomes belong in the interpretation. The 8088 result shows why magnetic flux contains useful rate information; the 8096 result shows why a probability is not a deterministic trigger. The wide historical magnitude range explains possible outcomes, but it cannot tell an operator which outcome will occur on that day.
A Flare Class Is Not a Fixed Band-Impact Table
GOES classes are based on peak 0.1–0.8 nm soft-X-ray irradiance measured near Earth. Increased X-ray and EUV radiation raises ionisation in the sunlit D region, increasing absorption along HF paths that traverse it. Lower HF frequencies are often more vulnerable for the same geometry, but absorption also depends on solar zenith angle, path length through the D region, flare time profile, background ionisation and the required link margin. There is no universal rule that C class affects only 80–40 m, that 20 m is safe, or that every X-class flare makes every HF band unusable for hours.
The NOAA Radio Blackout scale starts at M1, not C class:
| NOAA level | GOES threshold | Published HF scope |
|---|---|---|
| R1 | M1 | Weak or minor HF degradation on the sunlit side; occasional loss of contact. |
| R2 | M5 | Limited sunlit-side blackout, potentially lasting tens of minutes. |
| R3 | X1 | Wide-area sunlit-side blackout, potentially around an hour. |
| R4 | X10 | Blackout across most of the sunlit side for roughly one to two hours. |
| R5 | X20 | Potential complete HF blackout across the sunlit side for a number of hours. |
Those are impact categories, not a substitute for a path measurement. NOAA’s D-Region Absorption Predictions use current GOES X-ray and proton flux to estimate the highest frequency affected at declared absorption levels. That frequency- and location-aware picture is much more useful than a fixed amateur-band table.
Do Not Turn a Flare into a 6 m or 2 m Opening
A flare can be accompanied by a solar radio burst. At VHF, UHF or GNSS frequencies that burst is a potential interference and noise event, not a guaranteed propagation enhancement. It can reduce receiver SNR, overload a front end or disrupt communication and navigation.
A flare may also accompany a coronal mass ejection, but flare class alone does not prove that a CME exists, is Earth-directed or will have the magnetic orientation needed for a geomagnetic storm. If an Earth-directed CME later drives a storm, auroral propagation may become possible on VHF at suitable latitudes and geometry. That is a separate event chain observed through coronagraphs, solar-wind measurements and geomagnetic indices—not an immediate 6 m or 2 m gift from the X-ray flare. Sporadic E likewise needs its own ionospheric evidence.
Forecast Risk, Observe the Event, Measure the Path
- Before operating: use the NOAA three-day forecast and Forecast Discussion for current probabilistic R, S and G conditions; use the Solar Region Summary to identify visible NOAA regions. Treat research-model probabilities as one input, with their feature, horizon and validation period attached.
- When HF changes abruptly: check the real-time GOES X-ray flux, the NOAA R-scale alert and D-RAP. Compare the sunlit footprint and frequency-dependent absorption with your actual path.
- Check the radio evidence: compare several beacons, reverse-beacon reports, an ionosonde or a known reference signal. One disappearing station may be local interference, a station change or ordinary path fading.
- Separate later hazards: use GOES proton flux for polar-cap absorption risk. Use CME observations, arrival models, real-time solar wind and Kp for possible later geomagnetic effects.
- Keep a station record: log UTC, frequency, path, daylight geometry, GOES class and time profile, D-RAP estimate, local noise, signal levels and receiver state. That turns “the bands died” into a comparison that can be repeated.
My conclusion is deliberately practical: flare forecasting can improve planning, but it cannot replace observation. The Lim, Moon and Jeong model offers a transparent probability and magnitude-range method from one magnetic-flux feature. For operating decisions, join that forecast to the current GOES event, D-region absorption, your path geometry and what your receiver is actually hearing.
Primary Technical References
- Lim, Moon and Jeong: A Method to Predict the Likelihood and Magnitudes of Solar Flares
- NOAA Space Weather Scales
- NOAA GOES X-Ray Flux
- NOAA D-Region Absorption Predictions
- NOAA Three-Day Space Weather Forecast
- NOAA Solar Region Summary
Mini-FAQ
- What does the Lim, Moon and Jeong model predict? — From total unsigned magnetic flux, it estimates one-day C1.0+ and M1.0+ flare probabilities plus a historical statistical range for the daily peak X-ray flux.
- Does 99% accuracy mean M-class flares are almost certain to be predicted? — No. M1.0+ event-days were about 1% of the test set, so raw accuracy is dominated by non-events; TSS, Brier skill, false alarms and misses are also needed.
- Were 8088 and 8096 NOAA active-region numbers? — No. In the paper they are HARP magnetic-patch identifiers, and one HARP can contain zero, one or multiple NOAA active regions.
- Which HF bands will an M- or X-class flare close? — No fixed band table can answer that. Use GOES class and time profile, D-RAP frequency and location estimates, sunlit path geometry and live radio observations.
- Can a flare improve 6 m or 2 m propagation? — Not as a general rule. A solar radio burst can add interference; any later auroral opportunity depends on a separate Earth-directed-CME and geomagnetic-storm chain.
- What should I check when HF suddenly fades? — Check GOES X-rays, the NOAA R scale and D-RAP, then compare multiple beacons or reports and confirm that your receiver, antenna and local noise have not changed.