Reconstruction of pretelescopic and early telescopic solar activity cycles from auroral records Authors: Kristof Petrovay, Laura G Magyar, Hisashi Hayakawa Published: 2026 ABSTRACT AbstractThe historical record of low-latitude aurorae is essentially a poorly sampled record of the largest space weather events (SWEs). Its use for the identification of individual solar cycles is hindered by the low event rate and by the fact that the solar cycle profile of the occurrence of SWEs does not closely follow the variation of sunspot numbers. Based on recent studies of the solar cycle dependence of the occurrence rates of large SWEs, here we construct Monte-Carlo simulations of a large number of activity cycles to identify the optimal procedure to infer the characteristics of underlying solar cycles from the sparse record. We find that a reliable reconstruction of the cycle phase (>90% of reconstructed minima corresponding to actual minima within ±2 years) is possible whenever the long-term mean event rate (annual mean number of space weather events resulting in low-latitude auroral sightings) reaches or exceeds a value around 3. This condition is found to be satisfied during most of the the Early Modern Active Period (EMAP), a century-long period of normal solar activity between the Spörer and Maunder Minima. For the numbering of solar cycles in the EMAP we introduce the “telescopic era”, where Tn denotes the nth cycle from the first telescopically observed cycle, T0, ongoing in 1610. Using our optimal procedure we reconstruct a series of 8 solar activity cycles from T–5 to T2 (1560–1640). Earlier cycles starting from 1540 can be reconstructed with a somewhat lower degree of reliability. Comparing our results with radionuclide-based reconstructions and sunspot observations we find a good overall correspondence, with the exception of the last cycle before the Maunder Minimum.Keywords: Solar cycle, Cosmogenic radionuclides, Aurorae, Space climateSubject terms: Astronomy and planetary science, Physics, Planetary science, Space physics Introduction Spectacular auroral displays at moderate geomagnetic latitudes have drawn human attention long before solar flare patrols or geomagnetic measurements1. This resulted in a written record of space weather activity reaching back several centuries2,3 or even millennia4–7. Beyond providing evidence of individual large space weather events (SWEs), the auroral record also offers the possibility to study long-term variations in the Sun’s cyclic activity. This possibility was already recognized in the late nineteenth century8. In the following century, much effort was invested in extending the auroral record with new items and interpreting it in terms of space climate variations9,10. In addition to secular variations, the possibility of reconstructing individual Schwabe cycles in the pretelescopic era has also been explored11,12. The first and best known such reconstruction based on the auroral record was due to Schove13. As this reconstruction was based on the unjustifiable assumption of long-term phase coherence and it resulted in predictions that later proved to be wrong, it is now considered of historical interest only14. Although evidence for an 11-year cyclicity in auroral data has occasionally been noted15,16, more than three decades have elapsed since the last attempt17 to pin down the phase and amplitude of individual historical auroral cycles. This can be mainly attributed to the success of space climate reconstructions based on cosmogenic radionuclides14. The availability of increasingly detailed and extensive series of 14C and 10Be abundances from annually stratified samples (tree rings and ice cores, respectively) resulted in data sets that, albeit noisy and affected by geographical and climatic factors, have been considered arguably homogeneous and objective14,18,19. (Recent studies, however, noted some caveats20.) In contrast, the auroral record consists of sparse, heterogeneous sets of subjective aurora descriptions by early observers, influenced not only by weather conditions and a varying geomagnetic dipole but also by political and cultural factors21,22. Furthermore, even disregarding these factors, the incidence rate of large geomagnetic storms and the associated aurorae does not closely follow the sunspot number curve. Geomagnetic activity normally peaks 1–2 years after the sunspot number, and this phase difference shows a large scatter from one cycle to the next23. These difficulties led space climate reconstructions to focus on cosmogenic radionuclide data. Radionuclide-based work has by now culminated in a reconstruction of individual solar cycles covering the last three millennia24–26. It is to be noted, however, that this 14C based reconstruction comes with different quality flags assigned to each cycle, the suggestion being that for quality flags ≤ 3 (on a scale of 5) cycle parameters or even cycle identification are uncertain: about half the listed cycles fall in this category. This is further borne out by a comparison of different radionuclide-based reconstructions, as discussed later in this paper. On the other hand, recent work has elucidated the most salient features of the geomagnetic storm cycle underlying the auroral cycle and its relation to the sunspot cycle27,28. Furthermore, a careful critical compilation of the low-latitude auroral record was assembled by Křivský2,3. A large number of further auroral sightings have been unearthed in the past three decades, mainly from East Asia4,16,29. In view of these promising developments, the problem of reconstructing solar cycles from the auroral record alone may be worth a second look. In the present work, we make a new attempt at reconstructing individual auroral cycles in the pretelescopic and early telescopic period based on the existing auroral record. Our reconstruction method is guided by recent findings28 on the cycle profile of geomagnetic storm incidence and tested in Monte-Carlo (MC) simulations of a large number of solar cycles following this empirical distribution. Results Simulation of auroral cycles The cycle variation of the largest geomagnetic storms has been established in a recent analysis by Owens et al.28 (O21 in what follows). The occurrence probability of the storms was found to be bimodal, split into an active phase and a quiet phase, represented by a simple step function. The active phase lasts from solar cycle phase 0.18 to 0.79, while the early and late parts of the cycle, outside these limits, represent the quiet phase. The occurrence rate of SWEs is, to a first approximation (“Phase model”), constant in each phase, and a factor F higher in the active phase compared to the quiet phase, so that the cycle-averaged mean occurrence rate is M events/year. In order to test the possibility of reconstructing solar cycles from auroral records with these characteristics, for a large grid of F and M values, we simulated a series of 1000 cycles following the above statistical variations, with cycle lengths randomly varying around 11 years with a standard deviation of 1.15 years, as in the directly observed sunspot record. In line with usual practice in work with auroral records, the resulting series of simulated events was converted to a time series of number of events per calendar year. This time series was then analyzed by a variety of algorithms to identify the years of cycle minima. A resulting minimum year within ± 2 years of the actual minimum is considered a successful identification and the percentage of such successful identifications (hit rate) is used as a measure of the performance of the method. The procedure found to perform best essentially consists in smoothing the annual data by a 12221 slanted boxcar and combining local minima closer than 5.5 years, Further details are given in the Methods section. We note that identifying cycle maxima and other notable epochs was found to be possible with a lower success rate than minima. This somewhat counter-intuitive conclusion is related to the fact that the quiet phase is significantly shorter than the active phase; hence, large storms are scattered randomly over a longer period of time than the quiet phase between them. Figure 1 presents the hit rate as a function of F and M. We set a 90% hit rate (thick contour) as the lower limit for an acceptable reconstruction. As we may assume that extant records likely correspond to the most conspicuous aurorae, i.e. the largest SWEs, the mean event rate M in a period of normal solar activity may be translated to a lower threshold in event size. The statistics in O21 shows that over a period of 150 years, for the top 0.1% of events (99.9% threshold) the mean occurrence rate was M=0.37, whereas for a 99% threshold it was an order of magnitude higher. On the other hand, F was also found to depend on the threshold: for the threshold of 99.9% O21 quote F=9, while from inspection of their figures, for a threshold of 99% F is reduced to ∼ 3.5. Fitting these two points in the F–M plane with a power law results in the dashed blue curve in Fig. 1. Comparing this curve with the contour line drawn at 90% we infer that for a reasonably good reconstruction of solar cycles (> 90% of minima within ± 2 years of the actual date) a mean event rate of M≳ 3 is sufficent. Map of the success rate of the identification of cycle minima against maximum/minimum contrast factor F and mean annual event rate M, based on monte-carlo simulations. The blue dashed line is representative of the M–F relation for the strongest SWE events at times of normal solar activity. Xs mark possible loci representative of the EMAP within the 2σ error interval for the empirically derived contrast. Application to the auroral record Our starting point for the annual auroral observation record is the 1996 extended version of the Křivský auroral catalogue2,3. The variants of this catalogue have been used to evaluate the long-term solar activity in a number of previous studies10,30,31. This compilation does not simply lump together data from its source catalogues. Each catalogue item underwent a critical review, and a significant fraction of reports were discarded as not referring to aurorae. This critical review was performed by the same authors uniformly over the whole data set, which lends a certain degree of homogeneity to this compilation. Such critical reviews have proven important from some case studies, especially to eliminate possible contamination32–35. Beyond the sources used by Křivský, numerous new historical sources relating to aurorae have been collected by various authors. Of these, we collected the records originating before 17004,29,36,37 and merged them with the Křivský catalogue. The only exception where pre-1700 records were not used were the large amount of aurora-like sightings from Korea16,38. It is currently believed39,40 that a large fraction of these do not refer to bona fide polar aurorae but to equatorial airglow originating in the Western Pacific Anomaly of the geomagnetic field, affecting this area in the 15th to 18th centuries, so their inclusion would introduce a strong heterogeneity in the record. In line with earlier work on auroral data, for the present study we use annually accumulated records, i.e. the number of days with auroral sightings for each calendrical year. This circumvents issues related to seasonal variations and allows direct comparison to the annually sampled cosmogenic radionuclide record. Nevertheless, we take into account that large SWEs often last longer than a calendar day. Therefore, auroral sightings on two consecutive days are considered by us as a single event. In this sense, the annual count of events given in our final annual list differs from the number of days with auroral observation. The long-term variation of the record is displayed in Fig. 2. As is well known, strong inhomogeneities are present, the most important of which is the order-of-magnitude increase in the number of auroral sightings from the early 18th century, due to the start of scientific interest in aurorae. In the pre-1700 period, dips related to the known grand minima of solar activity are superimposed on a long-term increase due to the increasing number of source documents, related, among others, to the spread of printing. Comparing the long-term mean event rates plotted in the figure with the M≳ 3 lower limit determined in our simulations, we see that the only extended period where the event rate reached the required levels was in a period of normal solar activity sandwiched between the Spörer and Maunder Minima. We will refer to this period as the Early Modern Active Period (EMAP). Based on the conventional dates of grand minima14, the EMAP covers the period 1530–1640. From Fig. 2 we see that the period where auroral cycle minimum epochs can be expected to be reliably determined is 1560–1640, covering a large part of the EMAP. Annual record of SWEs resulting in auroral sightings 1000–1700, smoothed with a 25-yr low-pass Butterworth filter with 2 degrees of freedom. The conventional dates14 of grand minima of solar activity are marked with light blue bands: Oort Minimum (OM), Wolf Minimum (WM), Spörer Minimum (SM), Maunder Minimum (MM). Light yellow shading marks the Early Modern Active Period (EMAP), i.e., the interval between the Spörer and Maunder minima. A dotted horizontal line marks the formal lower threshold for reliable cycle reconstruction; the part of EMAP satisfying this condition is marked with a darker yellow shade. All observations in our catalogue originate from geographic latitudes below . At the middle of the EMAP this translated to geomagnetic latitude (MLAT) less than . The bulk of the historical sources originate from Central Europe: the MLAT of Prague, 52– at this time, may be taken as representative. As only conspicuous aurorae, significantly above the horizon were likely to be recorded in this early period, the equatorward boundary of the auroral oval may be estimated to lie at latitudes not higher than during the observed SWEs. This, in turn, translates to geomagnetic storm amplitudes nT41,42, meaning moderate or larger storms. This is consistent with the fact that there are only a very few years during the EMAP with no auroral observations: indeed, the annual maximum of practically never drops below 50 nT at times of normal solar activity27. The mean event rate over the reliable part of the EMAP (1560–1640) is M=4.2. The value of F may also be determined a posteriori from the reconstructed cycles. This determination, detailed in the Methods section, yields 3.3± 0.2 (2σ error range) for the underlying contrast parameter. The resulting parameter combination (M, F)=(4.2, 3.3) is compatible to the pair (3.7, 3.5) corresponding to a 99% threshold in O21, discussed in the previous subsection. After these preliminary considerations, we employ our optimized cycle-finding algorithm to the 1500–1650 section of the annual time series. The result is shown in Table 1 and in Figs. 3 and 4. Due to the difficulty of identifying cycles during the Maunder Minimum it is not possible to unambiguously extend the conventional numbering of solar cycles backwards to the period concerned. Instead, in the present paper we use the year of the first telescopic sunspot observations, 1610 as a reference point, denoting the solar cycle ongoing at this time with T0, while other cycles in the period studied will be denoted by Tn, where the integer n is the number of cycles counted from T0. Table 1List of reconstructed auroral cycles. The mean event rates per year M may serve as indicators of cycle amplitude, while the empirical contrast parameter can serve as a “quality flag”. Cycle numbering reflects the “telescopic era” introduced in this work, where cycle T0 marks the cycle ongoing in 1610. Reconstructions for the first two cycles, set in italics, are less reliable.Cycle no.StartMCycle lengthT-715401.7100.6T-615503.091.3T-515592.775.4T-415665.2113.0T-315775.5103.8T-215873.0102.1T-115975.2122.2T016093.8101.4T116193.981.3T216274.1131.5T31640 List of reconstructed auroral cycles. The mean event rates per year M may serve as indicators of cycle amplitude, while the empirical contrast parameter can serve as a “quality flag”. Cycle numbering reflects the “telescopic era” introduced in this work, where cycle T0 marks the cycle ongoing in 1610. Reconstructions for the first two cycles, set in italics, are less reliable. Reconstructed auroral cycles in the Early Modern Active Period. Cycle numbers in our “telescopic era” numbering are shown on top. The bars represent mean number of days with aurorae per year in each of the unequal cycle quartiles. Solid vertical lines mark the reconstructed minimum epochs; gray shading indicates the ± 2-year error range. Coloured vertical lines show minimum years from other reconstructions for comparison. Blue dots: GSN43; green dashes: C24; green dash-dots: C44; red dots: Be45; red dashes: Be, this work. (Some of the comparison lines were slightly horizontally shifted for better visibility.). SWE probabilities in the Early Modern Active Period based on our reconstruction results. Cycle numbers in our “telescopic era” numbering are shown on top. The green bars indicate expected annual event rates in cycle quartiles (without rounding to calendar years), calculated from the cycle means given in Table 1, assuming the “Phase” model in O21 with the underlying contrast parameter F=3.3 deduced for the EMAP. Shading of the color bars reflects the reliability of the reconstruction as hinted by the values in the table. Very light: , light: ; medium: ; dark: . Together with cycle start (phase 0) and mid-cycle (phase 0.5), the phase values 0.18 and 0.79 separating the quiet and active phases split each cycle into 4 unequal quartiles, Q1 to Q4. In our annually discretized data we assign the year of each phase division as the first year of the new quartile. The annual mean event rate in each of these quartiles is displayed in Fig. 3 in the form of bars. Summing up the number of events per quartile for cycles T-5 to T2 yields a percentage distribution of events between the quartiles: 9%, 39%, 38%, 14% for quartiles 1 to 4, respectively. Our MC simulations with the same M, F parameter pair result in: 7%, 41%, 39%, 13%. (The asymmetry of the distribution originates from the unequal quartile lengths combined with the discretization into calendar years.) The close agreement supports the correctness of the underlying model. For the EMAP period 1540–1640 we identify 10 cycles, the last 8 of which fall in the period 1560–1640, identified above as suitable for reliable cycle determination. For these 8 cycles, the mean cycle length is 10.1 years. and the scatter of cycle lengths is considerably higher than in the recent sunspot record. These discrepancies may not be statistically significant. Nevertheless, as from the MC simulations of the previous subsection the hit rate of our reconstruction is only expected to be ∼ 90%, the possibility arises that our reconstruction may include a “fake” cycle minimum. The most obvious candidate would be the unusually short cycle T-5. However, this cycle has a high contrast factor, and lumping it together with any of its two neighbours would result in a cycle with even more anomalous length. Furthermore, this cycle is also found in most other reconstructions, as discussed in the next section. Hence, we do not have any obvious candidate for a misidentified cycle even if we consider the whole EMAP period. Prior to cycle T-7 our reconstruction is not expected to be trusted due to the low event rate, and it indeed clearly misses a minimum around 1531, found by most other reconstructions. Comparison with other reconstructions Our reconstruction is compared with other solar cycle reconstructions for this period in Table 2. The aurora-based cycle reconstructions of Link15 and Schröder17were based on simple visual inspection of the frequency curve of selected auroral samples. Radionuclide-based reconstructions24,45,46 involve a heavy modelling component, including the modulation of cosmic ray flux by solar activity variations, and carbon-cycle viz. Be-transport models, respectively. The exception is Fogtmann-Schulz et al.44 where the listed epochs were determined directly from the frequency filtered isotopic abundances. The approach has the advantage of not being model-dependent, and, while cycle amplitudes may not be well reproduced, the reconstructed epochs are in good agreement with other reconstructions. We therefore also applied a similar approach to the Be isotopic fluxes determined from the NGRIP ice core47, as detailed in the Methods section, yielding a further set of reconstructed minimum epochs. Finally, the only reconstruction of the group sunspot number (GSN) overlapping with the EMAP is also shown. Table 2Comparison of reconstructed minimum dates of solar cycles. The ‘?’ signs were adopted from the original sources of the data. Slanted typesetting indicates numbers at odds with the other determinations.Cycle no.Auroral minimaC based minimaBe based minimaGSNThis workL7815Sch9217U2124FS1944M2146MB19This workSv1616T-101512151015111509T-91523152515221517T-81533153215311530T-71540154215391543T-6155015511552/531552154915481550?T-515591559155815581559T-4156615661566156515671566?T-3157715761577/781574157515771573T-2158715861589?158415871586T-1159716011600/01?1595160015971597T0160916101610/111609160616091608T1161916191619/2016201621161916201620T216271633163216331627?16301631T3164016451636?163716411640 Comparison of reconstructed minimum dates of solar cycles. The ‘?’ signs were adopted from the original sources of the data. Slanted typesetting indicates numbers at odds with the other determinations. When comparing minimum epochs determined by different recontructions, in each case we identify the 5-year period (true minimum ± 2years) that comprises the highest fraction of all recontructions. The reconstructions falling outside this interval are then marked as outliers (years set in slanted script in the table). With only one exception (cycle T2), none of our aurora-based reconstructions in this work are found to be outliers, supporting the validity of the method. Nevertheless, the significantly diverging results concerning the minima of cycles T-5 and T2 indicate that only the continuous series of 6 cycles from T-4 to T1 can be considered well established. Discussion We have established a method to reconstruct solar activity cycles from the auroral record when the mean annual event rate is at least 3. Applying this method to the available data a series of 10 cycles have been reconstructed in the EMAP. We note that the sharp criterion M>3 for the reliability of the reconstruction is not to be taken very rigorously. First, the scaling relation plotted in Fig. 1 on which it is based is itself approximate; the value of M will depend on the smoothing applied; and the 90% hit rate required is also an arbitrary choice. Hence, cycle reconstruction may be attempted also for periods for somewhat lower values of M. Indeed, as Table2 shows, our reconstruction for the period 1540-1560 is also in reasonably good agreement with the radioisotopic reconstructions. It may be worth placing some individual observations in the context of this reconstruction. The EMAP was a time when auroral reports became increasingly detailed and objective as aurorae slowly came to be considered a phenomenon of nature rather than divine portents. The Italian polymath Squarcialupi’s description of the aurora of 10 September 1580, observed from Transylvania, has been called the first scientific description of aurora borealis48. This intense storm occurred in the 2nd quartile of the strong cycle T-3; this quartile was characterized by the 2nd strongest level of auroral activity in the whole EMAP. Another recently discussed pretelescopic observation in the EMAP is Kepler’s sunspot of 160712. The deprojection of Kepler’s drawing allowed to infer that the observed sunspot had low heliographic latitude, implying that it marked the late phase of cycle T-1. This is consistent with our reconstruction of a minimum date of 1609 for cycle T0, in contrast to some other reconstructions (cf. Table 2). Methods The data used in this research and the codes for MC simulations and data analysis are publicly available on GitHub. (https://github.com/kpetrovay/Auroral-cycles). They are accompanied by readme files and the codes are well commented. In this section we only summarize the most salient technical aspects. Monte-Carlo approach to optimal cycle identification method The main algorithmic steps in the simulation are as follows. 1. Generate a sequence of 1000 cycles whose lengths are Gaussian random variables with mean 11 year and σ =1.15 year. Separate active and quiet phases in each cycle based on cycle phase. 2. In each time step generate number of events as a Poisson random variable whose expected value is computed from M and F. 3. Re-discretize time by calendar years. 4. Analyze the resulting time series of the annual count of events, evaluating the result in terms of the success rate, i.e., the fraction of reconstruction of cycle epochs (minimum, mid-cycle, quiet/active transitions) within ± 2 years, and select the procedure resulting in the optimal reconstruction. The tested procedures differed in the following aspects. (a) Smoothing [frequency filtering with an n=2 Butterworth filter, varying cutoffs; boxcars with different profiles, 121, 12221, 13631, 1222221 etc.] (b) Minimal allowed distance between identified epochs; epochs closer than this limit were replaced with a weighted average. (c) Final regularization (epochs separating short-long cycle pairs may be shifted to reduce difference). This resulted in the selection of the following procedure: (1) 12221 smoothing (2) identify minima and combine minima separated by <5.5 years (3) Regularization: minima between short–long cycle pairs are shifted by 1 year to improve balance; short/long defined as deviating min. 3 years from mean. (This step has very little effect on the final results in practice.) Empirical determination of the contrast factor F An empirical value for the contrast factor may be introduced as , where is the number of events in the ith quartile. Due to the limited length of the data set (8 reliable cycles) this empirical value can only be determined with a rather large statistical error that may be determined from the MC simulations. From the historical auroral data set for the reliable part of the EMAP this yields . It needs to be taken into account, however, that due to the annual discretization, the empirical value falls systematically short of the actual contrast F of the underlying model. This effect was also studied in our MC simulated cycles: Fig. 5 shows as a function of F and M. Together with M, for the EMAP this allows to read off the underlying true contrast value from the plot as . Left: Map of the empirically determined contrast factor against F and M based on the Monte-Carlo simulations. The contour line for is drawn in thick, while the contours for the limits are drawn in thin. Xs mark potential loci representative of the EMAP within the statistical errors. We finally note that our underlying cycle model, corresponding to the “Phase” model in O21, might be further improved taking into account the observation that odd and even numbered cycles tend to have opposite sign in , i.e. in cycles of one parity storms peak early, while in cycles of the other parity storms tend to peak late (“EarlyLate” model in O21). We performed various tests to check the presence of such odd-even alternation in the auroral data during the EMAP but, with the poorly sampled and short data set, no indication of such behaviour was found. Reconstruction of minimum and maximum epochs from Be data A proper reconstruction process of solar activity variations from cosmogenic radionuclide data would involve a comprehensive treatment to derive the modulation potential of cosmic ray propagation, taking into account the atmospheric production, residence, and transport of the nuclide involved, as well as the effects of the time-varying geomagnetic field21. While the shorter atmospheric residence time and higher threshold cosmic ray energy of Be make this less essential in the case of this nuclide, in the absence of a proper analysis of the above effects we judge it safer to limit ourselves to the reconstruction of cycle phases, without focusing on cycle amplitudes. Data processing: We use the raw Be fluxes FBe10 as determined by Berggren et al.47 in the NGRIP ice core, covering the time period 1389–1995. Schwabe cycles are not immediately obvious in this series due to short timescale noise and trends on a longer time scale. To remove these effects, filtering is applied to the data, retaining only components varying between time scales and . We experimented with different methods of filtering, including smoothing with a sliding window (both with sharp cutoff or a tapered Gaussian profile49) spectral filtering with a sharp cutoff, and a Butterworth filter. For each filtering method a variety of values for and were used. The optimal filtering method and parameters, selected following a calibration process (see below), were identified as a Butterworth filter of order 2 with yr and yr. This is in agreement with the original analysis47. The estimated atmospheric residence time of Be is about one year. In view of this, we assume that solar activity maxima [minima] precede minima [maxima] of FBe10 by 1 year. Local minima [maxima] in the series of filtered annual flux values are identifed as imprints of independent solar cycle maxima [minima] in the record whenever their separation from neighbouring minima [maxima] exceed yr. Otherwise, they are considered as imprints of a single (double peaked) solar maximum [minimum] at an epoch one year before the weighted average of the indiviual Be minima [maxima]. Calibration: We calibrate our method to the series of directly observed solar cycles since the end of the Maunder minimum, characterized by the group sunspot number (GSN). (https://www.sidc.be/silso/groupnumberv3). For compatibility with the FBe10 data, annual mean GSN values were used. The annual GSN series is subjected to the same filtering method, described above, as the FBe10 values. Comparing our series of epochs of extrema in the two series, an agreement is considered to hold whenever an FBe10 minimum [maximum] occurs within a 5-year interval centered on a date following the GSN maximum [minimum] by one year. This one-year time shift is introduced to allow for the atmospheric residence time of Be, as mentioned above. Note that some further phase shift exists between the sunspot number and the (inverted) galactic cosmic ray intensity50. The mean time lag was found to be 8 Bartels rotations (0.6 years), with a ± 100% scatter. This scatter has a bimodal distribution determined by the polarity of the heliospheric magnetic field (HMF). As this delay is an order of magnitude shorter than our 5-year time window and information on HMF polarity is lacking for the period of study in this paper, this further phase shift is not taken into explicit consideration. We determine the percentage of such agreements as a percentage of the total number of minima [maxima] in GSN and in FBe10 for minima and maxima separately, and consider the agreement best when the average of these percentage values is highest. This procedure results in the optimized filtering method and parameters quoted above, where, on average, the epochs agree in 84 %, i.e. an extremum epoch estimated as the epoch of the extremum of FBe10 plus one year agrees with the filtered GSN within yrs in 84 % of the cases (and to yr in 72 %). Epochs of minima and maxima in the calibration period are listed in a table available on the GitHub site of this project. Author contributions K.P.: research concept, auroral data analysis; L.M.: Be-10 data analysis; H.H.: contribution to discussion of early sunspot observations and space weather events. Funding This research was supported by the NKFIH grants TKP2021-NKTA-64 and 2026-4.1.1-MISSZIÓK-2026-00005, and by the European Union’s Horizon 2020 research and innovation programme under grant agreement no. 955620. HH received financial support from JSPS Grants-in-Aids JP25K17436 and JP25H00635, and the Research Grants in the Humanities of the Mitsubishi Foundation (202520034). Data availability The data used in this research and the codes for MC simulations and data analysis are publicly available on GitHub at https://github.com/kpetrovay/Auroral-cycles Declarations Competing interests The authors declare no competing interests. Footnotes Publisher's note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. References Associated Data Data Availability Statement The data used in this research and the codes for MC simulations and data analysis are publicly available on GitHub at https://github.com/kpetrovay/Auroral-cycles