betaFeedback

WorksPhysics and Astronomy

Distinguishing between flaring and non-flaring solar active regions with height-dependent polarity inversion line lengths

Abstract

Polarity Inversion Lines (PILs) are regions on the Sun's surface across which the vertical magnetic field changes polarity. This study examines whether the total length of PILs within a height range of 0-3.24 Mm above the photosphere can distinguish between strongly flaring active regions, with at least one flare of GOES (Geostationary Operational Environmental Satellite) class M1 or higher, and non-flaring active regions, where the strongest flare remains weaker than GOES class C1. Using magnetic field extrapolations on a dataset of 40 active regions, the study compares the 48-hr evolution of the total PIL length prior to the first major flare in strongly flaring regions with identical time intervals in non-flaring regions. It is seen that the 48-hr maximum of the total PIL length differs clearly between the two groups. This distinction becomes stronger at heights about 1-2 Mm above the photosphere. It is further seen that at about 1.8 Mm, the maximum value of the total PIL length, for around (i) 78 percent of weakly flaring regions (where the strongest flare is a C-class flare) and (ii) 96 percent of non-flaring active regions, is less than the minimum value of the total PIL length of their strongly flaring counterparts. These results suggest that strongly flaring active regions may be distinguishable from non-flaring or weakly flaring ones using thresholds on PIL lengths in the lower solar atmosphere. Further, this study highlights the importance of studying chromospheric magnetic fields in flare precursor investigations.

Introduction

Solar energetic events, such as flares [16] and coronal mass ejections (CMEs) [24, 9] are explosive phenomena driven by magnetic instabilities in the Sun’s atmosphere. Since these phenomena release vast amounts of energy and energetic particles into the inner heliosphere, they significantly impact electronic, telecommunication and satellite-based infrastructure on Earth and geospace, often commonly referred to as the technosphere. Hence, forecasting these energetic phenomena is of key importance to the space weather research community in order to secure our critical infrastructure in the technosphere. Forecasting space weather often involves studying the morphology of solar magnetic fields in the only place they can be measured routinely currently, namely, the solar photosphere. Such observations are available from ground- and space-based magnetographs. This is achieved by analysing and quantifying key parameters of solar magnetograms, that are known as proxies or predictors, characteristic to regions where the magnetic field is complex and disturbed significantly from its minimum-energy state. One such feature is the polarity inversion (i.e., neutral) lines (PILs) in the photosphere and lower solar atmosphere (LSA), which demarcate magnetic flux patches of opposite polarities. The strongest PILs appear in magnetic flux accumulations known as active regions (ARs). While commonly called ’neutral lines’ to indicate magnetic neutrality in a form of a zero vertical or radial field, strong (i.e., flux-massive) PILs are also characterised by areas of high-shear ([46]; see examples in [44, 8]) in their (strongly non-zero) horizontal field. PILs are also closely linked with solar eruptive phenomena (see [37] and references therein). Numerical models for CME initiation also confirm that ARs with strong PILs are invariably associated with high velocity shear flow ([2, 30]). These plasma flows possibly occur as a response to the Lorentz force acting along PILs in the solar atmosphere ([25]). Further, recent studies have shown that shearing arising out of collision between emerging non-conjugated bipoles in the lower solar atmosphere allows for the formation of compact PILs, leading to shearing and flux cancellation, and consequently solar flares and CMEs ([7]). It has been shown that both the vertical component ([34]) and the horizontal component ([39, 41] (Wang 2006)) of the photospheric magnetic field undergo rapid irreversible changes at the time of flare occurrence. Using MHD simulations, Bian & Jiang [3] verified that especially close to the PILs, the horizontal magnetic field increases significantly over a scale of several minutes after flare occurrence.

Having outlined the key characteristics of PILs, it is clear that detection algorithms must account for both the vertical and horizontal magnetic field variations across PILs, in order to properly identify those associated with strong shear (arising from enhanced horizontal fields) and significant vertical flux gradients. In this connection, there exist a number of predictors, either linked to PILs or proxies of PILs themselves, in the literature that have been studied in the past. For example, from a study of PILs, Falconer et al. (2003) found the strong shear length LSSL_{SS} (computed from vector magnetograms) and the strong gradient length LSGL_{SG} (a proxy for LSSL_{SS} computed from LOS magnetograms) to be good indicators for non-potentiality, and thus suitable candidates for the prediction of CMEs. In this context, non-potentiality refers to the presence of free energy in the ARs arising due to stressing, shearing or twisting of field lines. Similarly, Schrijver [33] introduced a parameter called the RR-value, that is an indicator of the flux around PILs that have a strong horizontal gradient in the vertical magnetic flux. Mason & Hoeksema [26] introduced a parameter strongly correlated to LSGL_{SG} called GWILLGWILL that is computed as a line integral of the gradient in the vertical magnetic field along the primary/longest PIL. Based on the study of photospheric flow fields, Park et al. [27] showed that the line integral of the shear flow speed (along PIL segments associated with (i) a strong horizontal magnetic field and (ii) a strong gradient of the vertical magnetic field) is a good precursor for strong flares.

Traditionally, flare precursors and predictors have been studied on the photosphere. However, recently it has been shown that studying flare precursors in a certain height range within the LSA, called the ’optimal height range’ (OHR), could help in the increase of the lead time before strong flares, especially GOES X-class. For example, Korsós et al. [20] found that it is possible to improve the lead time by 2-8 hr if the WGMWGM morphological parameter (Korsós et al. [17]) is studied between 1-1.8 Mm above the photosphere. From investigations of the emergence component of the helicity flux within 0.36-1.5 Mm above the photosphere, Korsós et al. [18] demonstrated that the lead time may be improved by 2-8 hr. Similarly, Biswal et al. [5] showed that a jump in RR-value could be observed between 48-68 hr before the first X-class flare for emerging ARs within a height range of 0.36-3.24 Mm above the photosphere. Motivated by such results, our main objective here is to investigate whether the PIL length in similar height ranges (above the photosphere), obtained by the extrapolation of the photospheric magnetic field, could help distinguish between flaring and non-flaring ARs. This action could potentially give valuable insights on how magnetic field configuration differs between flaring/eruptive and non-flaring, or weakly flaring, ARs.

Since longer PILs usually indicate more complex magnetic field structures ([26]), we choose to study the time series data for PILs in an AR using two parameters: (i) LtotL_{tot} - denoting the total length of PILs observed at any specific point in time, within a given time series and (ii) Ltot∗L^{*}_{tot} - denoting the maximum value of LtotL_{tot} within the time series. While LtotL_{\mathrm{tot}} is useful to trace the overall evolution pattern of PIL length, Ltot∗L_{\mathrm{tot}}^{*} gives an estimate of the PIL length in the AR’s most complex configuration. In this context, details of the LtotL_{\mathrm{tot}} time series dataset for different AR categories are discussed in § 2. This is followed by (i) a brief discussion of the extrapolation methodology and (ii) the algorithm used to compute LtotL_{\mathrm{tot}} in the extrapolated data in § 3. A few examples of PIL detection and LtotL_{\mathrm{tot}} evolution patterns at different heights are presented in § 4.1. Further results on how an OHR (to separate flaring ARs from non-flaring ARs) may be determined using Ltot∗L_{\mathrm{tot}}^{*} are discussed in § 4.2. § 4.3 and § 4.4 then identify the most optimal height within the OHR (determined in § 4.2) using descriptive and inferential statistical approaches, respectively. The overall implication of the study on flare predictability and forecasting, along with the concluding remarks are presented in § 5.

ACTIVE REGION DATASET

Our dataset see Appendix § 7) consists of 40 ARs belonging to solar cycle 24, out of which 24 ARs produced flare(s) with strength ≥\geq C1.0. These 24 flaring ARs are further classified as (i) X-class flaring ARs or (ii) M-class flaring ARs or (iii) CFARs (C-class flaring active regions), depending on the GOES flare class of the strongest flare produced by the AR (see Table 3 in Appendix § 7). For each of the three flaring groups, we study LtotL_{\mathrm{tot}} for a 48 hr interval immediately before the first flare (belonging to the same class as the strongest flare) within a height range of 0-3.24 Mm (see details in § 3). For example, AR 12673 is considered as an X-class flaring AR as the strongest flare was an X9.3 flare (2017/09/06 11:53 UTC). However, because the first X-class flare for AR 12673 was an X2.2 flare (2017/09/06 08:57 UTC), we study LtotL_{\mathrm{tot}} in the 48 hr interval immediately preceding the (earlier) X2.2 flare and also take a note of the Ltot∗L_{\mathrm{tot}}^{*} value in that interval. Since it is known that ARs that have flared previously are more likely to flare again [45, 28], focusing on the pre-flare evolution prior to the first flare ensures fairness in the analysis. For simplicity (see § 4.2), we consider the first two flaring groups as strongly flaring active regions (denoted as SFARs).

Observed data for SFARs
No.ARSHARP No.Interval TstartT_{start}Interval TendT_{end}Flare TimeFlare Class
01111583772011/02/13 01:002011/02/15 01:002011/02/15 01:44X2.2
02111664012011/03/07 23:002011/03/09 23:002011/03/09 23:13X1.5
03112838332011/09/04 22:002011/09/06 22:002011/09/06 22:12X2.1
041152018342012/07/10 15:002012/07/12 15:002012/07/12 15:37X1.4
051201738942014/03/27 17:002014/03/29 17:002014/03/29 17:35X1.0
061215845362014/09/08 17:002014/09/10 17:002014/09/10 17:21X1.6
071229752982015/03/09 16:002015/03/11 16:002015/03/11 16:11X2.1
081267371152017/09/04 09:002017/09/06 09:002017/09/06 08:57X2.2
091162022272012/11/26 21:002012/11/28 21:002012/11/28 21:20M2.2
101171926352013/04/09 06:002013/04/11 06:002013/04/11 06:55M6.5
111173126932013/04/30 04:002013/05/02 04:002013/05/02 04:58M1.1
121181830562013/08/15 18:002013/08/17 18:002013/08/17 18:16M3.3
131225350112015/01/01 09:002015/01/03 09:002015/01/03 09:40M1.1
141240358852015/08/19 01:002015/08/21 01:002015/08/21 01:56M1.2
151242259832015/09/25 20:002015/09/27 20:002015/09/27 20:54M1.2
161249763272016/02/10 10:002016/02/12 10:002016/02/12 10:36M1.0
Observed data for CFARs
171144014842012/03/19 12:002012/03/21 12:002012/03/21 12:38C2.9
181155319592012/08/27 23:002012/08/29 23:002012/08/29 23:40C1.1
191164023372012/12/30 08:002013/01/01 08:002013/01/01 08:47C1.2
201173026912013/04/27 20:002013/04/29 20:002013/04/29 20:22C5.9
211225550222015/01/09 14:002015/01/11 14:002015/01/11 14:42C1.6
221229352492015/03/02 13:002015/03/04 13:002015/03/04 13:35C1.1
231235355962015/05/21 03:002015/05/21 03:002015/05/23 03:27C1.0
241255265992016/06/07 12:002016/06/09 12:002016/06/09 12:28C1.4

Table 3. Table listing the studied flaring ARs with assisting information such as NOAA number, SHARP number, observation time window (TstartT_{start} denoting the start time and TendT_{end} denoting the end time), the time of flare occurrence and GOES flare class

Our dataset consists of 16 non-flaring active regions (denoted as NFARs henceforth), i.e. ARs not producing any flare with strength ≥\geq C1.0. Unlike flaring active regions (X/M/C-class), NFARs do not have a reference time point at which an analysis may be made to compare them with flaring ARs. To explore all relevant possibilities within the transit time of NFARs through the visible solar disk, we sampled the available data into as many distinct, non-overlapping intervals as possible, following which we have 45 independent and distinct 48-hr time intervals (see Table 4 in Appendix § 7).

Observed data for independently sampled CFARs
No.ARSHARP No.Interval TstartT_{start}Interval TendT_{end}No.ARSHARP No.Interval TstartT_{start}Interval TendT_{end}
111440-114842012/03/192012/03/211311730-326912013/04/302013/05/02
211440-214842012/03/212012/03/231412255-150222015/01/062015/01/08
311440-314842012/03/232012/03/251512255-250222015/01/082015/01/10
411553-119592012/08/242012/08/261612255-350222015/01/102015/01/12
511553-219592012/08/262012/08/281712293-152492015/02/272015/03/01
611553-319592012/08/282012/08/301812293-252492015/03/012015/03/03
711553-419592012/08/302012/09/011912293-352492015/03/032015/03/05
811640-123372012/12/302013/01/012012353-155962015/05/212015/05/23
911640-223372013/01/012013/01/032112353-255962015/05/232015/05/25
1011640-323372013/01/032013/01/052212353-355962015/05/252015/05/27
1111730-126912013/04/262013/04/282312552-165992016/06/082016/08/10
1211730-226912013/04/282013/04/30
Observed Data for NFARs
No.ARSHARP No.Interval TstartT_{start}Interval TendT_{end}No.ARSHARP No.Interval TstartT_{start}Interval TendT_{end}
111143-13352011/01/072011/01/092411957-336482014/01/222014/01/24
211143-23352011/01/092011/01/112511957-436482014/01/242014/01/26
311143-33352011/01/112011/01/132612064-141352014/05/152014/05/17
411416-113892012/02/092012/02/112712064-241352014/05/172014/05/19
511416-213892012/02/112012/02/132812064-341352014/05/192014/05/21
611416-313892012/02/132012/02/152912240-149432014/12/152014/12/17
711673-124892013/02/172013/02/193012240-249432014/12/172014/12/19
811673-224892013/02/192013/02/213112240-349432014/12/192014/12/21
911673-324892013/02/212013/02/233212400-158652015/08/112015/08/13
1011682-125012013/02/252013/02/273312400-258652015/08/132015/08/15
1111682-225012013/02/272013/03/013412400-358652015/08/152015/08/17
1211682-325012013/03/012013/03/033512528-164642016/04/042016/04/06
1311710-125992013/03/302013/04/013612528-264642016/04/062016/04/08
1411710-225992013/04/012013/04/033712663-170502017/06/162017/06/18
1511710-325992013/04/032013/04/053812663-270502017/06/182017/06/20
1611813-130492013/08/082013/08/103912664-170582017/06/222017/06/24
1711813-230492013/08/102013/08/124012664-270582017/06/242017/06/26
1811886-133262013/10/292013/10/314112664-370582017/06/262017/06/28
1911886-233262013/10/312013/11/024212664-470582017/06/282017/06/30
2011922-134902013/12/102013/12/124312719-172992018/08/202018/08/22
2111922-234902013/12/122013/12/144412719-272992018/08/222018/08/24
2211957-136482014/01/182014/01/204512719-372992018/08/242018/08/26
2311957-236482014/01/202014/01/22

Table 4. Table listing the studied independently sampled CFARs and non-flaring ARs (NFARs) with their SHARP numbers and observation time windows; both TstartT_{start} and TendT_{end} of the intervals correspond to 00:00 hr UTC

Since CFARs are often considered weakly flaring ARs in the literature (see for e.g. [42, 22]), we also study them like NFARs by sampling the data independently, ignoring the time of occurrence of the first C-class flare. By doing so we follow the behaviour of Ltot∗L_{\mathrm{tot}}^{*} in CFARs beyond the occurrence of the first C-class flare to explore the possibility of a difference in the Ltot∗L_{\mathrm{tot}}^{*} evolution between CFARs in general, and the pre-flaring PIL lengths in SFARs. Hence we also sampled the available data for 8 CFARs into 23 independent and distinct 48-hr time intervals, irrespective of the time of occurrence of the first C-class flare (see Table 4 in Appendix § 7). For example, if we take the case of AR 11730, the first C-class flare, i.e. the C5.9 flare, occurred on 2013/04/29 (20:22 UTC). The 48-hr interval immediately before this flare is studied within the CFAR data (see Table 3 in Appendix § 7). However, when we independently sample the available magnetogram data for this AR, we obtain 3 distinct and sequential 48-hr time series (see Table 4 in Appendix § 7) between Apr 26 - May 2, 2013 (00:00 UTC).

ANALYSIS METHOD

Extrapolation of input data

Since our objective is to find the magnetic field configuration above the photosphere in ARs and subsequently compute LtotL_{\mathrm{tot}} from the reconstructed field, our first step in this regard is to perform a Potential-Field (PF) extrapolation, based on the Fast Fourier Transform approach [1, 12] on each AR magnetogram.

The input vector magnetogram data is the radial component of the magnetic field (BrB_r), taken from the Helioseismic and Magnetic Imager (HMI), a magnetograph onboard the Solar Dynamics Observatory (SDO) mission Pesnell et al. [29]; Scherrer et al. [32]. The data product we use is either the Spaceweather HMI Active Region Patch (SHARP) data or a crop-out of it, in case the original data contains more than one NOAA-numbered AR Bobra et al. [6]. The PF extrapolation generates a 3D vector field solution from the BrB_r of the AR, as observed in the photosphere. In particular, this analysis is restricted to the LSA and uses 10 2D BzB_z maps produced by PF extrapolation which are equally spaced at steps of 0.36 Mm (∼0.5′′\sim0.5^{\prime\prime}) up to 3.24 Mm, similar to the approach of Biswal et al. [5]. The PF extrapolation is performed at intervals of 1 hr for the 48 hr observation time before the flare, generating a time series of magnetic field maps at each height. During the observation time, all ARs in our dataset were located between 60° EW of the central meridian to avoid severe projection effects Bobra et al. [6]. This ensures that LtotL_{\mathrm{tot}} calculations are based on reliable measures of the vertical magnetic field. While there are more elaborate extrapolation methods such as Non-Linear Force Free Field (NLFFF) method Wiegelmann & Sakurai [43], we chose to use the PF method because it is sufficient enough for a quick and computationally inexpensive calculation of the 3D magnetic field, at the same time preserving the main characteristics of the configuration Korsós et al. [19].

Computation and analysis of PIL parameters

To identify PIL(s), one way is to apply PIL masks of suitable field thresholds and appropriate convolution based image smoothing filters in order to highlight magnetic flux patches of opposite polarities and the PIL(s) separating them Schrijver 2007; Wang et al. 2019; Sun et al. 2021; similar approach in Ji et al. 2023. Mason & Hoeksema [26] followed a different approach where the input magnetogram is first smoothed and subsequently the vector potential field is computed for the identification of zero Gauss contours Alissandrakis [1]; Falconer et al. [10]. When these contours are identified, they must be above a certain horizontal field gradient threshold, so that PILs are considered only from strong shearing regions. The procedure is carried out at different smoothing thresholds and the outputs in each case are superposed to find overlapping regions, or the PILs as we understand them (refer to Figure 2 in Mason & Hoeksema [26]). Another interesting approach for detecting PILs in an AR involves the use of a computer vision based image segmentation algorithm Sadykov & Kosovichev [31]. Given the wide variety of algorithms and techniques employed to detect PILs in the literature Falconer et al. [10]; Falconer et al. (2003); Schrijver [33]; Mason & Hoeksema [26]; Sadykov & Kosovichev [31]; Sun et al. [35]; Wang et al. [41]; Ji et al. (2023), we selected an approach that incorporates physics-based control parameters computed from both the vertical and horizontal components of the magnetic field. Hence our method of detecting PILs is adapted from the FLARECAST project2 Georgoulis et al. [13], which is largely modelled along the detection methodology provided by Mason & Hoeksema [26]. In this approach, the control parameters are (i) BhB_h, i.e. the horizontal magnetic field, and (ii) BzgB_{zg}, i.e. the spatially averaged vertical magnetic field computed within a 10×1010 \times10 pixel box (3.5×3.53.5 \times3.5 Mm). The latter is evaluated separately for the positive and negative masks defined as Bz+=max⁡(Bz,0)B_z^+ = \max(B_z, 0) and Bz−=min⁡(Bz,0)B_z^- = \min(B_z, 0). Two thresholds are applied to BhB_h - (i) Bh1>120B_{h1} > 120 G for an initial detection and (ii) Bh2>100B_{h2} > 100 G for a finer detection. Likewise, BzgB_{zg} uses two thresholds - for Bzg+B_{zg}^+, (i) Bzg1>20B_{zg1} > 20 G and (ii) Bzg2>8B_{zg2} > 8 G; for Bzg−B_{zg}^-, (i) Bzg1<−20B_{zg1} < -20 G and (ii) Bzg2<−8B_{zg2} < -8 G, corresponding to the coarser and finer detections, respectively. The superposition of coarser and finer detections provides a list of all PIL fragments present. Assuming that for a given magnetogram NN PIL fragments are detected, then the parameter LtotL_{\mathrm{tot}} is the sum of all NN PIL fragment lengths LjL_j detected at that instance, namely:

Multiple-height stack plots of $L_{\mathrm{tot}}$ for AR 11158, AR 12158, and AR 11143

Figure 2. Multiple-height stack plots of LtotL_{\mathrm{tot}} for (0-3.24 Mm) for (a) AR 11158 (b) AR 12158 (c) AR 11143 as a function of time; the colorbar in each case represents the magnitude of the LtotL_{\mathrm{tot}} (in Mm); the colorbar’s upper limit is set to 400 Mm for (a), 600 Mm for (b) and 50 Mm for (c); the dashed line in (a) and (b) represents the time of occurrence of the X-class flare

Ltot=∑j=1NLj(1)L_{\mathrm{tot}} = \sum_{j=1}^{N} L_j \tag*{(1)}

Once LtotL_{\mathrm{tot}} is computed at all heights and timestamps for a given AR, we visualise the (i) PIL detection plots and (ii) stack-plots of LtotL_{\mathrm{tot}} evolution at different heights above the photosphere for further analysis (see examples in § 4.1, for results of other ARs visit the Zenodo Project repository3 Biswal [4]). A brief discussion on uncertainties applicable to LtotL_{\mathrm{tot}} is provided in Appendix § 8.

RESULTS

Visualising PIL detection in the extrapolated heights

In this Section, we present examples of PIL detections (see Figure 1) using two SFARs (ARs 11158, 12158) and one NFAR (AR 11143). While Figure 1 shows the PIL snapshots at specific points in time, Figure 2 shows the time series data of LtotL_{\mathrm{tot}} at different altitudes, computed from PF extrapolation, corresponding to the ARs visualised in Figure 1. We observe that the evolution of LtotL_{\mathrm{tot}} with height is different for SFARs and NFARs; and no PILs were found above 2.52 Mm for the AR 11143 at any time (see Figure 2c), e.g. Figure 1o shows that the opposite polarities in AR 11143 have a gap between them. The PIL is not detected here as it does not exceed the control parameter thresholds.

Plots showing detected PILs overlaid in red on the extrapolated magnetic field for SFARs and NFAR

Figure 1. Plots showing detected PILs overlaid in red on the extrapolated magnetic field for SFARs - 11158 (a-f), 12158 (g-l) and NFAR 11143 (m-o) at a height of 0 Mm (col 1), 1.80 Mm (col 2) and 3.24 Mm (col 3); the colorbar thresholds are ± 150 G

Further, the two SFARs also exhibited different LtotL_{\mathrm{tot}} evolution patterns. AR 11158, which is known to have exhibited (i) flux emergence ([36, 38, 5]) and in which, (ii) the regions of opposite polarities moved towards each other during the interval of study, showed an overall increase in LtotL_{\mathrm{tot}} (see Figure 2a & Figure 1a-Figure 1f). Both these factors may have contributed to PIL growth, which can be observed at all heights (for e.g. compare Figure 1b & Figure 1e to check the growth at a height of 1.80 Mm). If we consider Figure 1c, about 48 hr prior to the X2.2 flare, we find that regions of opposite polarities barely touch each other (Ltot∼1L_{\mathrm{tot}} \sim1 Mm). However on 15 February 2011 at 01:00 UTC, less than 1 hr prior to the flare, it can be be seen that these regions have converged and thus account for an increase in LtotL_{\mathrm{tot}} to ∼46\sim46 Mm at a height of 3.24 Mm (see Figure 1f). Contrary to AR 11158, AR 12158, which is known to have shown (i) a decrease in unsigned flux in the interval of study ([5]), and in which, (ii) the regions of opposite polarities moved away from each other during the interval of study, showed an overall decrease in LtotL_{\mathrm{tot}} (see Figure 2b & Figure 1g-Figure 1l).

PF extrapolation smooths the magnetic field as one moves up in height (see Figure 1). This may have some advantages and disadvantages. The advantages are that (i) it makes the data less noisy and that (ii) it makes it easier to trace the movement of the regions having opposite polarities, with time. However, on the downside, smoothing may suppress smaller isolated fragments of PILs see, e.g., how some PIL fragments in Figure 1d are missing in Figure 1f).

Decrease of Ltot∗L^{*}_{\mathrm{tot}} with height

Figure 3a shows the decrease of Ltot∗L^{*}_{\mathrm{tot}} with height above the photosphere for X-class flaring ARs when compared with M-class flaring ARs. Based on their mean (μ\mu) and standard deviation (σ\sigma) values, we find that these two groups cannot be easily distinguished. Hence for further analysis, we consider these two flaring groups as a single category of SFARs. Further Figure 3b, Figure 3c and Figure 3d show the variation of Ltot∗L^{*}_{\mathrm{tot}} with height for the NFARs, CFARs and independently sampled CFARs, respectively, when compared to SFARs. We find that a clear distinction exists between SFARs and NFARs at heights ≥0.36\geq0.36 Mm (see Figure 3b) and between SFARs and CFARs within a height range of 0.36-1.80 Mm (see Figure 3c), at least at 1σ1\sigma significance. These height ranges could serve as an OHR (optimal height range) for distinguishing SFARs from NFARs or CFARs, thereby demonstrating the importance of studying PILs in the PF-extrapolated configuration of ARs above the photosphere.

Profile of $L^{*}_{\mathrm{tot}}$ decrease for various AR categories

Figure 3. Profile of Ltot∗L^{*}_{\mathrm{tot}} decrease for various AR categories; the exact category-wise mean (μ\mu) and standard deviation (σ\sigma) values of Ltot∗L^{*}_{\mathrm{tot}} at each height can be found in Table 1

Further, the distribution of Ltot∗L^{*}_{\mathrm{tot}} at all heights for CFARs in terms of (μ\mu, σ\sigma) is not very different from their counterpart for NFARs, which are sampled independently by definition. This suggests that CFARs may have Ltot∗L^{*}_{\mathrm{tot}} values similar to that of NFARs, in the 48 hr period immediately preceding the first C-class flare. However, once these ARs evolve and progressively become more complex after the first C-class flare, they may have a higher chance of having Ltot∗L^{*}_{\mathrm{tot}} values similar to that of SFARs. This is evident from the partial overlap between the shaded regions in Figure 3d. Despite the partial overlap, Ltot∗L^{*}_{\mathrm{tot}} of SFARs, on average, is considerably greater (by a factor between 1.7 - 3.5, see Table 1 for μ\mu values at each height) than the independently sampled CFARs in general. If we make the same comparison between SFARs and NFARs, Ltot∗L^{*}_{\mathrm{tot}} differs by a factor between 3 - 11, depending upon height.

Mean values - μ(Ltot∗)\mu(L^{*}_{\mathrm{tot}}) (in Mm) as a function of height, listed columnwise
AR Type0.00 Mm0.36 Mm0.72 Mm1.08 Mm1.44 Mm1.80 Mm2.16 Mm2.52 Mm2.88 Mm3.24 Mm
SFAR (X)341.99242.92191.87153.60128.35114.1694.5278.9168.2757.90
SFAR (M)443.77258.32181.28142.76109.6896.5074.8660.1352.2947.04
SFAR392.88250.62186.57148.18119.02105.3384.6969.5260.2852.47
CFAR129.2482.1360.3142.1534.9330.5325.7623.2318.5916.11
CFAR (IS)237.03114.7273.6255.2242.2436.2026.5921.6717.0415.03
NFAR128.3863.0941.7430.5823.6818.1513.139.857.644.6
Standard deviation values - σ(Ltot∗)\sigma(L^{*}_{\mathrm{tot}}) (in Mm) as a function of height, listed columnwise
AR Type0.00 Mm0.36 Mm0.72 Mm1.08 Mm1.44 Mm1.80 Mm2.16 Mm2.52 Mm2.88 Mm3.24 Mm
SFAR (X)168.3898.1782.1369.5060.0353.5148.3141.8036.1329.12
SFAR (M)257.74112.0657.2645.7345.5644.2942.3036.0732.4831.03
SFAR216.78102.0868.6157.1152.3748.3245.0238.9434.1929.61
CFAR107.5330.7819.6618.7219.4218.5918.0518.1218.7417.40
CFAR (IS)226.9890.5063.1947.6539.4936.2027.8224.3321.9620.27
NFAR135.5957.7835.0426.5522.4218.8214.611.610.366.51

Table 1. Category-wise μ\mu and σ\sigma values of Ltot∗L^{*}_{\mathrm{tot}} at each height; SFAR (X) represents X-class flaring ARs; SFAR (M) represents M-class flaring ARs; CFAR (IS) represents independently sampled CFARs

Optimal height for distinguishing SFARs from NFARs and independently sampled CFARs

Heights ≥0.36\ge0.36 Mm may serve as an OHR to distinguish SFARs from NFARs, as mentioned previously in § 4.2. Similarly, a height range of 0.36-1.80 Mm may help distinguish between SFARs from CFARs, which are a subset of the independently sampled CFARs. Since we would also like to investigate how well we could separate between independently sampled CFARs and SFARs even though their Ltot∗L^{*}_{\mathrm{tot}}-decrease profiles show a partial overlap, we made a box-plot analysis of the distribution as a function of altitude (Figure 4). The box-plot shows a clear separation between the IQRs (inter-quartile ranges) of SFARs and independently sampled CFARs at all heights ≥0.36\ge0.36 Mm. The IQRs here correspond to the middle 50% of the data and is denoted by the span of the coloured boxes in Figure 4. This finding once again, demonstrates the importance of studying PILs in the extrapolated topology of ARs above the photosphere. Similarly, we find that the IQRs of SFARs and NFARs are well separated at all heights in our study.

Boxplots of total length as a function of height for SFARs, independently sampled CFARs and NFARs

Figure 4. Boxplots of Ltot∗L^{*}_{\mathrm{tot}} as a function of height for (a) SFARs, (b) independently sampled CFARs and (c) NFARs; the lower and upper boundaries of each box denote the 25th (Q1Q_1) and 75th (Q3Q_3) percentiles; the median is shown by a black line inside each box; the whiskers denote Ltot∗=1.5×IQRL^{*}_{\mathrm{tot}} = 1.5 \times\mathrm{IQR} (inter-quartile range = Q3−Q1Q_3 - Q_1) from Q1Q_1 and Q3Q_3; the coloured dots are outliers

The boxplots in Figure 4 also indicate that the heights 1.80 Mm and 2.16 Mm contain the least number of outliers if we consider all the three groups - SFARs, NFARs and independently sampled CFARs, together. Thus these heights could be most suitable within the OHR for distinguishing SFARs from NFARs and independently sampled CFARs. To verify this, we take the minimum Ltot∗L^{*}_{\mathrm{tot}} value for SFARs at all heights as a threshold (LthL_{\mathrm{th}}) and check how many cases within 45 NFAR samples and 23 independently sampled CFARs samples did not exceed this threshold (see Table 2). Our findings reveal that the 1.80 Mm height is the most optimal height (within the OHR) to separate out SFARs from independently sampled CFARs and NFARs. We find that ∼\sim96% of NFARs and ∼\sim78% of independently sampled CFARs are less than LthL_{\mathrm{th}} ∼\sim46 Mm at a height of 1.80 Mm above the photosphere.

Height (Mm)0.000.360.721.081.441.802.162.522.883.24
LthL_{\mathrm{th}} (Mm)114.2499.0379.6872.0755.6745.9931.6227.3817.126.45
NFAR28/4537/4538/4542/4542/4543/4543/4543/4539/4531/45
(Ltot∗<LthL^{*}_{\mathrm{tot}} < L_{\mathrm{th}})%62.22%82.22%84.44%93.33%93.33%95.56%95.56%95.56%86.67%68.89%
CFAR (IS)11/2312/2316/2317/2317/2318/2316/2316/2315/2311/23
(Ltot∗<LthL^{*}_{\mathrm{tot}} < L_{\mathrm{th}})%47.83%52.17%69.57%73.91%73.91%78.26%69.57%69.57%65.22%47.83%

Table 2. Variation of LthL_{\mathrm{th}} i.e. the minimum Ltot∗L^{*}_{\mathrm{tot}} value (row 2) observed for SFARs with height (row 1) and details of the percentage of NFARs and independently sampled CFARs which satisfied the condition: Ltot∗<LthL^{*}_{\mathrm{tot}} < L_{\mathrm{th}} at each height (rows 3 and 4, respectively). The best distinction is achieved at a height of 1.80 Mm (column in bold).

Optimal height for identifying population clusters from the current data sample

In the boxplots of Figure 4, it is observed that the upper whisker is in general longer than the lower whisker for all three groups of ARs at all heights. This is an indicator of the low-value skewness in the distributions and suggests that our data is not normally distributed. In this context, we present a method to identify and distinguish between different population clusters (for e.g. SFARs and NFARs) using the one-sided Chebyshev’s inequality [21] applying to all kinds of distributions. The inequality states that if XX is a random variable with mean μ\mu and standard deviation σ\sigma, then for λ>0\lambda> 0 the following are true (PP denoted probability):

P(X≤μ+λ)≥λ2λ2+σ2≤P(X≥μ−λ)(2)P(X \le\mu+ \lambda) \ge\frac{\lambda^{2}}{\lambda^{2} + \sigma^{2}} \le P(X \ge\mu- \lambda) \tag*{(2)}

If we analyse Figure 3a, we find that the μ+σ\mu+\sigma line for NFARs is well below the μ−σ\mu-\sigma line for SFARs. It is intuitive to note here that if we multiply σ\sigma by a fixed factor k>1k > 1, then the gap between the two shaded regions can be minimised. This gap can be eliminated if:

k=μ(SFAR)−μ(NFAR)σ(SFAR)+σ(NFAR)(3)k = \frac{\mu(SFAR)-\mu(NFAR)}{\sigma(SFAR)+\sigma(NFAR)} \tag*{(3)}

Using Equation (3), we find k={1.17,1.40,1.41,1.27,1.30,1.20,1.18,1.18,1.33}k = \{1.17, 1.40, 1.41, 1.27, 1.30, 1.20, 1.18, 1.18, 1.33\} for the nine heights between 0.36 Mm and 3.24 Mm, in ascending order. The maximum value of k=1.41k = 1.41 is obtained at 1.08 Mm height. At this height, the μ+1.41σ\mu+ 1.41\sigma level for NFARs and the μ−1.41σ\mu- 1.41\sigma level for SFARs are both ∼\sim68 Mm. Letting λ=1.41σ\lambda= 1.41\sigma in Equation (2) in each case, we have:

P(0≤Ltot∗(NFAR)≤68)≥66%(4)P(0 \leq L^{*}_{tot}(NFAR) \leq68) \geq66\% \tag*{(4)}
P(Ltot∗(SFAR)≥68)≥66%(5)P(L^{*}_{tot}(SFAR) \geq68) \geq66\% \tag*{(5)}

This implies that if our data samples for SFARs and NFARs were to be considered as fairly good representatives for the actual populations of SFARs and NFARs, then at a height of 1.08 Mm above the photosphere, a threshold of Ltot∗=68L^{*}_{tot} = 68 Mm may provide classification-type probabilistic guarantees that at least 66% of the SFAR population must be above this threshold and at least 66% of the NFAR population must be below it.

DISCUSSION AND CONCLUSIONS

Implications on flare predictability and forecasting

Even though it is known that PILs are intimately connected to solar eruptive phenomena and PIL length has been correlated to flare and eruptive flare (i.e., flare-associated CME) likelihoods [10, 15] (Falconer et al. 2003), actual PIL lengths, especially above the photosphere, have been barely discussed. As per Falconer et al. [10], the PIL parameter LSSL_{SS} (called the strong shear length), which is determined by calculating the length of the segments on the main neutral line (i.e. the longest PIL) where the horizontal magnetic field Bh>150B_h > 150 G and the shear angle θ>45∘\theta> 45^{\circ}, is well correlated to CME productivity within a ±48\pm48 hr time window. Falconer et al. (2003) have shown that LSSL_{SS} is well correlated to another PIL parameter LSGL_{SG} (called the strong gradient length) with a confidence level exceeding 95%, which is determined by calculating the length of the segments on the main neutral line where Bh>150B_h > 150 G and the absolute value of the horizontal gradient of the line of sight magnetic field ∇hBlos>50\nabla_h B_{los} > 50 G/Mm. Falconer et al. (2003) have also shown that if LSSL_{SS} exceeds 62 Mm or if LSGL_{SG} exceeds 36 Mm, then CMEs are to be expected (accuracy score ∼\sim88%).

Taking these previous results into consideration, we can say that the PIL parameter of focus in this paper - LtotL_{tot} is slightly different from LSGL_{SG} as it includes all PIL segments in an AR, not just the central PIL. While LSGL_{SG} takes into account the vertical magnetic field in the form of ∇hBlos\nabla_h B_{los}, LtotL_{tot} does so through BzgB_{zg}. Even though LtotL_{tot} is computed using less stricter threshold in BhB_h compared to LSGL_{SG}, this difference may be justified by the fact that our approach computes PILs not just at the photosphere but also at heights above it. As we move up in height, the magnetic field values also become weaker, thereby plausibly requiring weaker thresholds for any meaningful PIL detection. Overall, however, the major similarity between LtotL_{tot} and LSGL_{SG} is that both these parameters use the vertical and horizontal components of the magnetic field as physics-based control parameters for PIL detection. In this context, when detecting PILs using a uniform set of thresholds at each height within a height range of 0-3.24 Mm above the photosphere, LtotL_{tot} may be considered as a useful alternative to LSGL_{SG}. We find that Ltot∗L^{*}_{tot}, i.e. the maximum value of LtotL_{tot} over a 48 hr interval immediately before the first X/M-class flare is in general longer than the respective for ARs that do not host any X/M-class flares.

Using descriptive statistics, we find that a height of 1.8 Mm above the photosphere may be the most suitable height to distinguish SFARs from NFARs and independently sampled CFARs. At this height, ∼\sim78% of independently sampled CFARs and ∼\sim96% of NFARs, Ltot∗L^{*}_{tot} was less than 46 Mm. While these results appear promising at this stage to develop a binary (YES/NO) model to detect SFARs at an optimal height of 1.8 Mm, we acknowledge the caveat of a limited sample of ARs. Working on a larger AR sample will help refine our existing results. This is reserved for a future work on this topic. A large scale study could provide information on (i) the relative natural occurrence frequencies of SFARs (f1f_1), CFARs (f2f_2) and NFARs (f3f_3) and (ii) the probabilities of Ltot∗≥LthL^*_{tot} \ge L_{th} for SFARs (p1p_1), CFARs (p2p_2) and NFARs (p3p_3), where LthL_{th} is a threshold based on some relevant statistical metric (for e.g. minimum value, 5%ile value etc.) pertaining to the Ltot∗L^*_{tot} distribution for SFARs at a given optimal height. Based on this information, the overall problem of forecasting M/X-class flares over a 48 hr window may be translated into a simpler problem of Bayesian inference. In real-time, if we find Ltot∗≥LthL^*_{tot} \ge L_{th} at the optimal height, then the probability that it would produce an M/X-class flare within 48 hr shall be:

p=f1p1f1p1+f2p2+f3p3(6)p = \frac{f_1p_1}{f_1p_1 + f_2p_2 + f_3p_3} \tag*{(6)}

Using the one-sided Chebyshev inequality, we inferred that a height of 1.08 Mm above the photosphere may be the most suitable height to demarcate the population of SFARs from NFARs. At this height, at least 66% of SFARs have Ltot∗≥68L^*_{tot} \ge68 Mm and at least 66% of NFARs have Ltot∗≤68L^*_{tot} \le68 Mm. In this regard, a more rigorous statistical study could be helpful for a more accurate statistical separation of the different populations.

Summary

Some general conclusions of this study are summarised below:

  • While examining the evolution of precursors or predictors such as the WGMW_{GM} morphological parameter [20], magnetic helicity [18] and RR-value [5] in the lower solar atmosphere are useful for maximising the flare prediction lead time, analysing the magnitudes of LtotL_{tot} and Ltot∗L^*_{tot} in the extrapolated topology of ARs within an OHR in the lower solar atmosphere may offer critical complementary information concerning the prediction of strong flares.

  • Specifically, LtotL_{tot} - computed with an uniform set of thresholds (same as those originally intended for the photosphere by Georgoulis et al. [13] in the FLARECAST project) - appears to be a useful parameter to study PILs in the LSA (up to about 3 Mm above the photosphere). The overall approach of analysing LtotL_{tot} as a function of height may be used (i) as a method to distinguish between SFARs and NFARs; (ii) as a technique to determine suitable thresholds of Ltot∗L^*_{tot} as a function of height, which may serve as a basis for demarcating the population clusters of SFARs, NFARs and independently sampled CFARs; and (iii) potentially to estimate the probability of occurrence of M/X-class flares within a 48 hr window, provided a statistical study can establish the natural occurrence frequencies of SFARs, NFARs and independently sampled CFARs as well as the likelihood of each category exhibiting Ltot∗L^*_{tot} values above the thresholds at each height.

  • The Ltot∗L^*_{tot} decrease profiles show a clearer separation between SFARs and NFARs at heights ≥0.36\ge0.36 Mm as compared to the photosphere, thereby highlighting the importance of studying the extrapolated magnetic configuration of ARs. In this regard, it was also seen that for CFARs, the 48 hr period immediately before the first flare resembled NFARs in terms of Ltot∗L^*_{tot}, and was quite different from SFARs between a height range of 0.36-1.80 Mm. However, the Ltot∗L^*_{tot} decrease profile of independently sampled CFARs showed a partial overlap with SFARs at all heights, and exhibited an overall pattern that is intermediate to SFARs and NFARs.

  • We used two different approaches to find the most optimal height (within the OHR) which may bring about the best possible distinction between SFARs and NFARs. While the approach centred around descriptive statistics of our data sample shows that the best height is about 1.8 Mm above the photosphere, the other approach suggests that at about 1 Mm above the photosphere may be the most suitable to demarcate clusters of SFARs and NFARs at a population level.

  • It is promising to note from our dataset of 40 ARs and 92 cases (i.e. 48 hr time series) that ∼\sim78% of independently sampled CFARs and ∼\sim96% of NFARs had Ltot∗<46L^*_{tot} < 46 Mm at 1.8 Mm height.

We plan to extend our calculations, as a continuation of this study, to a much larger statistical study and other methods of extrapolations, which (i) may provide more accurate thresholds in the OHR and (ii) may help develop probabilistic models based on PIL lengths (i.e. Ltot∗L^*_{tot}) to forecast M/X-class flares.

Acknowledgement

This research work is supported by Space Weather Awareness Training Network (SWATNet) and it comes under the aegis of the project: ’Three-dimensional solar flare prediction’ (https://swatnet.eu/). SWATNet has received funding from the European Union’s Horizon 2020 research and innovation programme under the Marie Sklodowska-Curie Grant Agreement No. 955620. SB acknowledges the support received from the University of Sheffield Publication Scholarship Programme. MBK, RE and MG acknowledge support from ISSI-BJ (”Step forward in solar flare and coronal mass ejection (CME) predicting”). RE is grateful to STFC (UK, grant number ST/M000826/1) and PIFI (China, grant number No. 2024PVA0043). MBK is grateful for the Leverhulme Trust Found ECF-2023-271. MBK acknowledges support by UNKP-22-4-II-ELTE-186, ELTE Hungary. RE and MBK also thank for the support received from NKFIH OTKA (Hungary, grant No. K142987). This work was also supported by the NKFIH Excellence Grant TKP2021-NKTA-64. AN and SP acknowledge support by the ERC Synergy Grant ‘Whole Sun’ (GAN: 810218).

Analysed Dataset

This Section contains the full list of all 92 different time series studied for X/M/C-class flaring ARs (see Table 3), independently sampled CFARs (see Table 4) and NFARs (see Table 4) as referenced in § 2.

Error Analysis

Our approach of detecting PILs (see § 3.2) is sensitive to the 2D-configuration of the vertical magnetic field BzB_z, and thus, we assume that any uncertainty in the computed value of LtotL_{\mathrm{tot}} is also a function of the overall configuration of the magnetic field. Since the noise level of the vertical photospheric images is ∼10\sim10 G [23], we performed a Monte Carlo simulation of the error in LtotL_{\mathrm{tot}} (i.e. ΔLtot\Delta L_{\mathrm{tot}}) by (i) adding a normally distributed error (μ=0\mu= 0 G, σ=10\sigma= 10 G) to the radial component of the magnetic field (BrB_r) for AR 11158 (15 February, 2011 at 00:00 UTC), (ii) subsequently extrapolating the data (PF) up to 3.24 Mm and then (iii) computing LtotL_{\mathrm{tot}} at each height. This exercise was repeated for 100 iterations at each altitude and the results are tabulated below:

The Monte Carlo analysis presented here (see Table 5) provides an estimate of the uncertainty in LtotL_{\mathrm{tot}} i.e. ΔLtot\Delta L_{\mathrm{tot}}, resulting from the observational uncertainty in the magnetic field. The propagation of errors through the extrapolation and application of the PIL detection algorithm appears to show no consistent pattern as a function of height. While the analysis is based on a single AR and therefore does not provide a statistical characterisation of the uncertainties across different ARs, the uncertainty in BrB_r measurements suggests that the propagated uncertainties ΔLtot\Delta L_{\mathrm{tot}} are generally expected to remain well below 10%. We therefore adopt 10% as a conservative upper-bound estimate of the uncertainty associated with LtotL_{\mathrm{tot}}. Since Ltot∗L^*_{\mathrm{tot}} is defined as the maximum value of LtotL_{\mathrm{tot}} over a 48-hr period, its uncertainty, ΔLtot∗\Delta L^*_{\mathrm{tot}} can also be assumed to be within 10% of the Ltot∗L^*_{\mathrm{tot}} value.

Monte Carlo estimates of LtotL_{\mathrm{tot}} and ΔLtot\Delta L_{\mathrm{tot}} (100 iterations)
No.Height (Mm)Ltot±ΔLtotL_{\mathrm{tot}} \pm\Delta L_{\mathrm{tot}} (Mm)UncertaintyNo.Height (Mm)Ltot±ΔLtotL_{\mathrm{tot}} \pm\Delta L_{\mathrm{tot}} (Mm)Uncertainty
10.00226.4 ± 9.14.0%61.8074.04 ± 0.260.35%
20.36159.0 ± 4.62.9%72.1666.8 ± 3.24.8%
30.72131.45 ± 0.400.30%82.5262.1 ± 2.33.7%
41.08103.5 ± 2.32.2%92.8855.91 ± 0.0060.011%
51.4479.27 ± 0.640.81%103.2455.41 ± 0.120.22%

Table 5. Results of Monte Carlo simulation of LtotL_{\mathrm{tot}} and ΔLtot\Delta L_{\mathrm{tot}} for AR 11158 on 15 February, 2011 at 00:00 UTC; ΔLtot\Delta L_{\mathrm{tot}} and the uncertainty are presented in figures with two significant digits

References

  1. [1]Alissandrakis, C. E. 1981, A&A, 100, 197
  2. [2]Amari, T., Luciani, J. F., Aly, J. J., Mikic, Z., & Linker, J. 2003, The Astrophysical Journal, 585, 1073, doi: 10.1086/345501DOI
  3. [3]Bian, X., & Jiang, C. 2023, Frontiers in Astronomy and Space Sciences, Volume 10 - 2023, doi: 10.3389/fspas.2023.1097672DOI
  4. [4]Biswal, S. 2026, PILs_3D: Data repository for PIL analysis in 3D, v1.0.0, Zenodo, doi: 10.5281/zenodo.22738623DOI
  5. [5]Biswal, S., Korsós, M. B., Georgoulis, M. K., et al. 2024, The Astrophysical Journal, 974, 259, doi: 10.3847/1538-4357/ad6c33DOI
  6. [6]Bobra, M. G., Sun, X., Hoeksema, J. T., et al. 2014, Solar Physics, 289, 3549, doi: 10.1007/s11207-014-0529-3DOI
  7. [7]Chintzoglou, G., Zhang, J., Cheung, M. C. M., & Kazachenko, M. 2019, The Astrophysical Journal, 871, 67, doi: 10.3847/1538-4357/aaef30DOI
  8. [8]Deng, N., Xu, Y., Yang, G., et al. 2006, The Astrophysical Journal, 644, 1278, doi: 10.1086/503600DOI
  9. [9]Dere, K. P., Brueckner, G. E., Howard, R. A., Michels, D. J., & Delaboudiniere, J. P. 1999, ApJ, 516, 465, doi: 10.1086/307101DOI
  10. [10]Falconer, D. A., Moore, R. L., & Gary, G. A. 2002, The Astrophysical Journal, 569, 1016, doi: 10.1086/339161DOI
  11. [12]Gary, G. A. 1989, ApJS, 69, 323, doi: 10.1086/191316DOI
  12. [13]Georgoulis, Bloomfield, D. Shaun, Piana, Michele, et al. 2021, J. Space Weather Space Clim., 11, 39, doi: 10.1051/swsc/2021023DOI
  13. [15]Kontogiannis, I., Georgoulis, M. K., Guerra, J. A., Park, S.-H., & Bloomfield, D. S. 2019, Solar Physics, 294, 130, doi: 10.1007/s11207-019-1523-6DOI
  14. [16]Kopp, R. A., & Pneuman, G. W. 1976, SoPh, 50, 85, doi: 10.1007/BF00206193DOI
  15. [17]Korsós, M. B., Gyenge, N., Baranyi, T., & Ludmány, A. 2015, Journal of Astrophysics and Astronomy, 36, 111, doi: 10.1007/s12036-015-9329-xDOI
  16. [18]Korsós, M. B., Erdélyi, R., Huang, X., & Morgan, H. 2022, The Astrophysical Journal, 933, 66, doi: 10.3847/1538-4357/ac7469DOI
  17. [19]Korsós, M. B., Jarolim, R., Erdélyi, R., et al. 2024, The Astrophysical Journal, 962, 171, doi: 10.3847/1538-4357/ad18bdDOI
  18. [20]Korsós, M. B., Georgoulis, M. K., Gyenge, N., et al. 2020, The Astrophysical Journal, 896, 119, doi: 10.3847/1538-4357/ab8fa2DOI
  19. [21]Lin, Z., & Bai, Z. 2011, Bounds of Probabilities in Terms of Moments (Berlin, Heidelberg: Springer Berlin Heidelberg), 51–66, doi: 10.1007/978-3-642-05261-3_6DOI
  20. [22]Liu, S., Wang, J., Li, M., et al. 2023, Frontiers in Astronomy and Space Sciences, Volume 10 - 2023, doi: 10.3389/fspas.2023.1082694
  21. [23]Liu, Y., Hoeksema, J. T., Scherrer, P. H., et al. 2012, Solar Physics, 279, 295, doi: 10.1007/s11207-012-9976-xDOI
  22. [24]Low, B. C. 1994, in ESA Special Publication, Vol. 373, Solar Dynamic Phenomena and Solar Wind Consequences, the Third SOHO Workshop, ed. J. J. Hunt, 123
  23. [25]Manchester, W. 2008, in Astronomical Society of the Pacific Conference Series, Vol. 383, Subsurface and Atmospheric Influences on Solar Activity, ed. R. Howe, R. W. Komm, K. S. Balasubramaniam, & G. J. D. Petrie, 91, doi: 10.48550/arXiv.0812.3589
  24. [26]Mason, J., & Hoeksema, T. 2010, The Astrophysical Journal, 723, 634, doi: 10.1088/0004-637X/723/1/634DOI
  25. [27]Park, S.-H., Guerra, J. A., Gallagher, P. T., Georgoulis, M. K., & Bloomfield, D. S. 2018, Solar Physics, 293, 114, doi: 10.1007/s11207-018-1336-zDOI
  26. [28]Park, S.-H., Leka, K. D., Kusano, K., et al. 2020, The Astrophysical Journal, 890, 124, doi: 10.3847/1538-4357/ab65f0DOI
  27. [29]Pesnell, W. D., Thompson, B. J., & Chamberlin, P. C. 2012, Solar Physics, 275, 3, doi: 10.1007/s11207-011-9841-3DOI
  28. [30]Roussev, I. I., Sokolov, I. V., Forbes, T. G., et al. 2004, The Astrophysical Journal, 605, L73, doi: 10.1086/392504DOI
  29. [31]Sadykov, V. M., & Kosovichev, A. G. 2017, The Astrophysical Journal, 849, 148, doi: 10.3847/1538-4357/aa9119DOI
  30. [32]Scherrer, P. H., Schou, J., Bush, R. I., et al. 2012, Solar Physics, 275, 207, doi: 10.1007/s11207-011-9834-2DOI
  31. [33]Schrijver, C. J. 2007, ApJL, 655, L117, doi: 10.1086/511857DOI
  32. [34]Sudol, J. J., & Harvey, J. W. 2005, The Astrophysical Journal, 635, 647, doi: 10.1086/497361DOI
  33. [35]Sun, H., Manchester IV, W., & Chen, Y. 2021, Space Weather, 19, e2021SW002837, doi: https://doi.org/10.1029/2021SW002837
  34. [36]Sun, X., Hoeksema, J. T., Liu, Y., et al. 2012, The Astrophysical Journal, 748, 77, doi: 10.1088/0004-637X/748/2/77DOI
  35. [37]van Driel-Gesztelyi, L., & Green, L. M. 2015, Living Reviews in Solar Physics, 12, 1, doi: 10.1007/lrsp-2015-1
  36. [38]Vemareddy, P., Venkatakrishnan, P., & Karthikreddy, S. 2015, Research in Astronomy and Astrophysics, 15, 1547, doi: 10.1088/1674-4527/15/9/011
  37. [39]Wang, H. 1992, Solar Physics, 140, 85, doi: 10.1007/BF00148431DOI
  38. [41]Wang, J., Liu, S., Ao, X., et al. 2019, The Astrophysical Journal, 884, 175, doi: 10.3847/1538-4357/ab441bDOI
  39. [42]Wang, J., Luo, B., & Liu, S. 2022, Frontiers in Astronomy and Space Sciences, Volume 9 - 2022, doi: 10.3389/fspas.2022.1037863
  40. [43]Wiegelmann, T., & Sakurai, T. 2012, Living Reviews in Solar Physics, 9, 5, doi: 10.12942/lrsp-2012-5
  41. [44]Yang, G., Xu, Y., Cao, W., et al. 2004, The Astrophysical Journal, 617, L151, doi: 10.1086/427210DOI
  42. [45]Zirin, H., & Marquette, W. 1991, SoPh, 131, 149, doi: 10.1007/BF00151751DOI
  43. [46]Zirin, H., & Wang, H. 1993, Nature, 363, 426, doi: 10.1038/363426a0DOI

Paper details

vlnc:W_vq7u2ejs
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Physics and Astronomy

Not recorded for this paper

View statement
Looking for related papers…
Contents