Half a Century of X-Ray Solar Flares Reveals Unpredictable Cycle-to-Cycle Behavior

Solar flares are among the most consequential events in the Solar System, capable of disrupting radio communications, navigation systems, satellite operations, and power grids on Earth. Forecasting them has long depended on a deceptively simple premise: that a sunspot region’s appearance today tells you something reliable about its flaring behavior, and that this relationship holds steady from one solar cycle to the next. A new statistical analysis published in the journal Solar Physics challenges that assumption in a striking way. Examining nearly five decades of X-ray flare records spanning solar cycles 21 through 25, from 1976 to 2025, the study finds that neither the intensity distribution of flares nor their rates across different sunspot classes can be considered consistent between cycles, raising uncomfortable questions for operational space weather forecasting.

The study, conducted by Owen Giersch, analyzed X-ray flares of M class and higher—the moderate to extreme events that matter most for space weather operations. The research took two complementary approaches. First, flares were sorted into logarithmically spaced intensity categories and compared across sunspot cycles. Second, flare rates were tabulated according to the McIntosh classification of the sunspot group that produced each event, allowing a direct test of whether a given type of sunspot region flares at the same rate regardless of which cycle it belongs to. Both tests delivered results that depart from the comforting assumption of cycle-to-cycle consistency.

The McIntosh classification scheme, which underpins the second part of the analysis, has a long pedigree. Sunspot classification began with Alfred Cortie’s scheme of 1901, which is no longer used. Observers in Zurich developed an alternative system in the 1930s, using nine categories to describe the evolution of sunspot groups. In the 1960s, Patrick McIntosh modified this Zurich system, removing two classes and adding parameters describing the penumbra of the largest sunspot in a region and the compactness of spots within the region. The resulting Modified Zurich Classification—now universally known as the McIntosh system—combines the Zurich class, the penumbral character, and the compactness into 60 permitted classes. Because sunspot classification serves as a proxy for the magnetic complexity of an active region, it remains the backbone of most flare forecasting practice.

The data underpinning the analysis came from two archival sources. For flares prior to 1996, records were obtained from the National Centers for Environmental Information, and from 1996 onwards from the Space Weather Prediction Center, both drawing ultimately on the GOES satellite series that has monitored solar X-ray output continuously since cycle 21. Sunspot region reports came from United States Air Force Solar Observing Optical Network sites, whose daily observations include quality flags ranging from 1, indicating very poor seeing conditions, to 5, indicating excellent conditions. Because pre-1996 reports were listed from up to five individual sites without being reduced to a single classification, the author developed a method to consolidate multiple daily reports: retaining the highest-quality report, then adopting the most frequently occurring class, with random selection only in the rare case of irreconcilable ties.

The data cleaning itself illustrates the messy reality of half a century of observational records. Numerous invalid classifications appeared in the archive—classes with missing or implausible parameters—and a set of explicit rules was applied to correct them, drawing on the physical logic of the McIntosh scheme. After all corrections, only 193 regions could not be assigned a valid class, and these were omitted from the analysis. For the flare-rate comparison by sunspot class, only flares with an assigned region were included, while the raw cycle-to-cycle flare counts used all reported events. This careful curation matters, because the study’s conclusions hinge on comparisons between data streams that were assembled by different teams, with different instruments and practices, across 50 years.

The statistical machinery was deliberately conventional, designed to be transparent rather than exotic. To compare flare intensity distributions between two cycles, a chi-squared test was employed, with categories combined wherever either distribution contained fewer than five flares, and p-values computed so that distributions could be deemed similar at the 95 percent confidence level. To compare flare rates by McIntosh class, both chi-squared tests and linear regression were used. Under the regression approach, if two cycles behaved identically, plotting one cycle’s flare rates against the other’s should yield a line with a gradient of one, an intercept of zero, and a correlation coefficient close to one. Departures from that ideal indicate genuine differences in how sunspot classes translate into flares.

The results on intensity are nuanced but troubling. At the 95 percent confidence level, the flare intensity distribution of cycle 21 cannot be considered similar to those of cycles 22 through 25—a discrepancy the author attributes in part to the fact that cycle 21 was the first cycle during which the GOES satellites operated, raising the possibility of calibration or measurement issues. The comparisons between cycles 22 and 25 and between cycles 23 and 25 also failed the similarity test. A broad pattern emerged: the greater the separation between two cycles in time, the more dissimilar their flare intensity distributions tend to be. The striking exception is the pairing of cycles 22 and 24, which produced the highest p-value of any comparison, suggesting those two cycles flared in remarkably similar proportions despite differing in overall amplitude.

The second finding is more fundamental. When flare rates were compared by McIntosh class across cycles, chi-squared tests found essentially no pairs of cycles that could be considered similar. The sole nominal exception—a comparison of cycle 21 as the observed distribution against cycle 22 as the expected distribution—flipped to dissimilar when the roles were reversed, a signature of a statistical false positive. The regression analysis reinforced the picture: while the intercepts of seven out of ten cycle comparisons were close to zero, as expected for the least active sunspot classes, only the comparisons involving cycles 25 and 22 produced gradients close to one. In plain terms, the rate at which a given kind of sunspot group produces M-class or larger flares varies significantly from one solar cycle to another.

What explains this variability? The study is candid that the answer remains unclear. One possibility is observer bias: sunspot classification is a human judgment, and many researchers have documented problems with sunspot group analysis, particularly with sunspot areas, which form one input to the McIntosh parameters. If area measurement techniques—and therefore the resulting values—have drifted over decades, other derived parameters such as extent and compactness may have drifted too, contaminating any cycle-to-cycle comparison. The alternative is more provocative: that the difference reflects a genuine physical change in how the Sun organizes magnetic energy from cycle to cycle. Intriguingly, recent work by other researchers found that as the McIntosh Zurich class increases from A to F, the total magnetic energy of active regions rises, but the free energy—the portion available to power eruptions—does not change significantly, hinting that classification alone may not capture what makes a region flare-ready.

Despite the ambiguities, the analysis closes with something immediately useful for forecasters. Assuming flares follow a Poisson process, the probability of at least one M-class or larger flare occurring in a given region over a specified interval can be computed from its flare rate, and with multiple regions on the Sun—common near solar maximum—the combined probability follows from multiplying the individual exponential terms. The full dataset generated for the study has been released in a public GitHub repository, and the author emphasizes that further analysis of the USAF sunspot classifications is needed to determine whether the cycle-to-cycle variation is intrinsic to the Sun or an artifact of evolving observational practice. Either answer carries weight: one rewrites the physics of flare statistics, the other a cautionary tale about the archives on which space weather science is built.

Subject of Research: Statistical analysis of X-ray solar flare intensity distributions and flare rates by McIntosh sunspot class across solar cycles 21 to 25

Article Title: Analysis of X-ray Solar Flare Rates from 1976 to 2025

Article References: Analysis of X-ray Solar Flare Rates from 1976 to 2025. (n.d.). https://doi.org/10.1007/s11207-026-02743-w

Image Credits: AI Generated

DOI: 10.1007/s11207-026-02743-w

Keywords: solar flares, X-ray flares, sunspots, McIntosh classification, solar cycle, space weather, GOES satellites, flare forecasting, solar physics, statistical analysis, sunspot classification, active regions