The Disappearing Solar Filament of 2013 September 29 and Its Large Associated Proton Event: Implications for Particle Acceleration at the Sun

The unsigned magnetic reconnection (or ribbon) flux, defined to be the amount of flux undergoing reconnection during a solar flare, is a relatively new flare diagnostic parameter (Qiu & Yurchyshyn 2005; Qiu et al. 2007). Kazachenko et al. (2017) recently determined the ribbon flux based on observations by AIA at 1600 Å and the Helioseismic and Magnetic Imager (HMI; Scherrer et al. 2012) on SDO for 3137 ≥C1 class flares from 2010 April to 2016 April (http://solarmuri.ssl.berkeley.edu/~kazachenko/RibbonDB/). The ribbon flux determination procedure is illustrated in Figure 3 for the 2013 September 29 DSF event. The left panel in the figure shows the HMI photospheric magnetogram for Br with the contours of the cumulative AIA 304 Å flare ribbons at the end time of the two-ribbon flare associated with the DSF overplotted. Integration of unsigned Br inside these contours yields the ribbon flux. The right panel shows the corresponding temporal and spatial evolution of the 304 Å flare ribbons with each pixel color-coded by the time of its initial brightenings. The observations at 304 Å yield an unsigned reconnection flux of 6.4 × 10²¹ Mx (Figure 4). A comparable figure could not be made for the 1600 Å images used in Kahler et al. (2017) and in this study because the emission at that wavelength was too weak to determine the reconnection flux.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

The ribbon flux has been found to be correlated with such parameters as CME speed (Qiu & Yurchyshyn 2005; Deng & Welsch 2017; Pal et al. 2018), the magnetic flux of interplanetary CMEs (Qiu et al. 2007; Gopalswamy et al. 2017b), and the flare peak SXR flux (Kazachenko et al. 2017; Tschernitz et al. 2018). Kahler et al. (2017) were the first to apply the reconnection flux to the question of proton acceleration at the Sun. Quoting from Kahler et al. (2017), "The photospheric magnetic flux swept out by flare ribbons is thought to be directly related to the amount of magnetic reconnection in the corona and is therefore a key diagnostic tool for understanding the physical processes in flares and CMEs." The underlying assumption is that the reconnection flux is a more reliable indicator of the amount of energy release in flares than parameters based on, e.g., SXR or radio emissions. The reconnection link to the flare-resident acceleration process might be direct as in electric field acceleration (e.g., Litvinenko 1996, 2006) or indirect via a stochastic process (e.g., Petrosian 2012), where reconnection creates turbulence in magnetic loops.

Using an early version of the Kazachenko et al. (2017) ribbon flux database that only considered ≥C8 events, Kahler et al. (2017) identified 15 reconnection events for flares located from W20 to W45 that were associated with SPEs on a list of >25 MeV proton events (I.G. Richardson 2017, personal communication) that updated a previous compilation from Richardson et al. (2014). The lower longitude limit was chosen to consider only magnetically well-connected SPEs in order to reduce proton propagation effects, and the upper limit was mandated by the difficulty of measuring unsigned flux far from disk center. Kahler et al. also identified 111 non-SPE reconnection events under the assumption that the absence of an entry in the Richardson list for an event in the Kazachenko et al. database implied a negligible peak ∼25 MeV intensity (i.e., background intensity of ≤10⁻⁴ protons (cm² s sr MeV)⁻¹). That assumption turned out to be incorrect; approximately half (55) of the 111 events had actual background values larger than this, ranging up to 10⁻¹ protons (cm² s sr MeV)⁻¹. Of the 56 remaining events, 35 were M- or X-class SXR flares. Figure 5, patterned after a similar figure in Kahler et al. (2017), is a scatter plot of peak ∼25 MeV intensity for the 15 SPEs in Kahler's sample along with the 56 non-SPEs. In addition, the figure includes a point (open red square upper limit) for the 2013 September 29 DSF event that was not considered by Kahler et al. because the peak SXR flux (C1.2) was <C8.

Zoom In Zoom Out Reset image size

Comparison of AIA images at 304 Å (Figure 3) and 1600 Å for the DSF event reveals that the upper chromosphere was significantly more affected by the DSF than the lower chromosphere. While a ribbon flux of 6.4 × 10²¹ Mx was obtained from AIA images at 304 Å (Figures 3 and 4) and HMI observations, only an upper limit reconnection flux of 10²¹ Mx could be determined using the 1600 Å observations of the DSF. This upper limit value, plotted in Figure 5, indicates that, in addition to solar events with large reconnection fluxes that do not produce protons, there are events with negligible reconnection flux (at 1600 Å) that give rise to large (S2) SPEs. To date, no comprehensive database has been constructed for the 304 Å based reconnection flux from which a scatter plot similar to that in Figure 5 could be constructed. We determined the ribbon flux based on 304 Å images for the two events in Figure 5 that had ∼25 MeV peak intensities comparable to that of the 2013 September 29 event (2011 August 4 (coordinates in Figure 5 (0.92, 0.04)) and 2014 April 18 (0.98, −0.4)). In each case the log of the unsigned flux increased, from 0.92 to 1.38 for the 2011 event and from 0.98 to 1.57 for the 2014 event versus a corresponding increase from <0.0 to 0.81 for the 2013 event. Thus, for a plot based on 304 Å observations, the data point for the 2013 September 29 event would remain an outlier, to the left of the adjusted points for the 2011 and 2014 SPEs and an order of magnitude (or more) above points having the same adjusted 304 Å reconnection flux.

Examination of the 56 events with only upper limit ∼25 MeV peak proton intensities of 10⁻⁴ protons (cm² s sr MeV)⁻¹ reveals that they lacked associated decametric-hectometric (DH) type II bursts (0/55 cases; data gap for one event) based on observations from the Waves radio instrument (Bougeret et al. 1995) on the Wind spacecraft (Acuña et al. 1995).¹⁰ DH slow-drift type II bursts, the radio manifestation of coronal shock waves, are highly correlated with large SPEs (Gopalswamy et al. 2002; Cliver et al. 2004). Only 3 of the 56 non-SPEs in Figure 5 (5%) had metric type II association.¹¹ Of the 56 non-SEP events, 36 lacked CMEs in the Large Angle and Spectrometric Coronagraph (LASCO; Brueckner et al. 1995) CME catalog (https://cdaw.gsfc.nasa.gov/CME_list/; Yashiro et al. 2004; Gopalswamy et al. 2009) on the Solar and Heliospheric Observatory (SOHO; Domingo et al. 1995). The median CME speed and width for the 20 events with CMEs was ∼395 km s⁻¹ (range from 171 to 762 km s⁻¹) and 43° (range from 6° to 360°), respectively, versus ∼660 km s⁻¹ (366–2175 km s⁻¹) and 360° (99–360°) for the 15 SPE-associated events in the original Kahler et al. (2017) study. Of these 15 events, 10 (67%) had DH type II association, and 4 of the remaining 5 had associated metric type IIs.

In contrast to the 56 non-SPE events in Figure 5, the DSF on 2013 September 29 was associated with a >1000 km s⁻¹ halo CME (1179 km s⁻¹) (Figure 6 (top)) and a DH type II burst (Figure 6 (bottom)). The open red square upper limit for this event shown in Figure 5 corresponds to a peak ∼25 MeV intensity of 0.5 protons (cm² s sr MeV)⁻¹ (Richardson et al. 2014). Including the DSF event, the combined type II association rate for the SPE events in Figure 5 is 94% (15/16). The flare, CME, shock, and SEP parameters for the 56 non-SPE and 16 SPE events are given in Table 1.

Zoom In Zoom Out Reset image size

Figure 7 gives a scatter plot taken from Grechnev et al. (2015) showing a correlation (delineated by the dashed oval) between >100 MeV proton fluence and 35 GHz radio fluence for >100 MeV SEP events from 1990 to 2014. Flare microwave emission and hard X-ray emission (e.g., Arnoldy et al. 1968; Kosugi et al. 1988) are signatures of nonthermal electron acceleration in the flare impulsive phase. The correlation in Figure 7 could provide evidence for the concomitant acceleration of protons to high energies in a flare-resident process. Grechnev et al. (2013, 2015) used the 35 GHz frequency emission because this frequency is in the optically thin branch of the gyrosynchrotron spectrum.

Zoom In Zoom Out Reset image size

The five square data points in Figure 7 were not encompassed in the narrow dashed oval of Grechnev et al. (2015) because those authors attributed them to acceleration of SEPs by shocks rather than to an unspecified flare-resident process as was assumed for the events in the oval. The rationale for doing so was the weak 35 GHz flare emission of these events relative to their "abundant" proton fluences. For reasons given below based on Cliver (2016), we believe that the black square outliers should not be excluded from the correlation analysis.

In their analysis, Grechnev et al. only considered events that had 35 GHz bursts with peak flux densities ≥1000 solar flux units (sfu; 1 sfu = 10⁻²² W m⁻² Hz⁻¹) or SPEs with peak >100 MeV intensities >10 pfu. The 2013 September 29 fell well short of meeting either of these criteria and thus was not considered in their study. We used the 35 GHz radio data from Nobeyama (Tanaka & Kakinuma 1957; Nakajima et al. 1985; http://solar.nro.nao.ac.jp/norp/html/event/) to determine the x-axis coordinate of the September 29 data point in the >100 MeV proton fluence versus 35 GHz fluence scatter plot. The radio time profiles at the highest frequencies (those >9.4 GHz) are compromised because the event began close to the sunrise time of 21:50 UT at Nobeyama. To remove ground interference effects, we subtracted the flux time traces for the six Nobeyama observational frequencies (ranging from 1 to 35 GHz) for the two previous observing days (September 27/28 and September 28/29) from those for September 29/30, with the resulting differential fluxes based on the September 28/29 comparison shown in Figure 8. Bad weather at Nobeyama on 2013 October 1 and 2 precluded use of the data for those days. Fluctuations at 3.75 GHz are due to radio interference from geosynchronous satellites near the autumnal equinox. In Figure 8(f), the general decrease in the 35 GHz flux following sunrise is attributed to increasing cloudiness on September 29 based on photographs taken at 0:00, 3:00, and 6:00 UT every day. There is no clear burst in the 35 GHz differential flux profile during the ∼22:00–22:25 UT interval of impulsive emission (indicated by the dashed lines) observed at 17 GHz (Figure 8(e)). The standard deviation of the differential flux at 35 GHz during the event at 17 GHz is ∼10 sfu. For a 1σ upper limit flux at 35 GHz of 10 sfu over the 22:00–22:25 UT time interval, we obtain an upper limit 35 GHz fluence of ∼1.5 × 10⁴ sfu s. To obtain the >100 MeV proton fluence for the 2013 September event (∼5 × 10³ pfu s), we integrated the background-subtracted GOES >100 MeV proton intensity time profile (shown in Figure 2) from 00 UT on September 30 to 08 UT on October 1 (http://www.ngdc.noaa.gov/stp/satellite/goes/dataaccess.html).

Zoom In Zoom Out Reset image size

Cliver (2016) noted that the observation of large/fast CMEs and DH type II shocks for the "main-sequence" events in the oval in Figure 7 argued for a shock-related acceleration process for all major SEP events. In fact, the main-sequence CMEs were more energetic than those for the outliers. Thus, we redid the Grechnev et al. analysis here to obtain Figure 9. To minimize SEP propagation effects, we only considered SPEs and non-SPEs in Table 1 of Grechnev et al. (2015) that originated between W21 and W90, using the longitude correction for >100 MeV proton fluences given by those authors.¹² The filled gray circles at the bottom of the figure represent large 35 GHz bursts that lacked associated SPEs; these are plotted arbitrarily at a >100 MeV fluence of 10 pfu s. We obtained a power-law fit (y = 0.0002×^1.4; R² = 0.51) to the black diamonds and squares (main-sequence and abundant proton events, respectively, in Figure 7) in Figure 9 and drew an oval, with the fit line as the major axis, to envelop these data points. We excluded the data point for the 2000 November 8 event from the curve fit because it, like the 2013 September 29 event, appears to be an outlier from the expanded main sequence. In Figure 9, both of these data points are plotted as light-blue squares.

Zoom In Zoom Out Reset image size

As was seen in Figure 4, there are examples of large flares in Figure 9—in this case manifested by 35 GHz bursts (with peak fluxes ≥1000 sfu and fluences ranging from 10⁴ to 10⁶ sfu s with a median of 4 × 10⁴ sfu s)—that originated in well-connected (W21–W90) flares that were not followed by detectable near-Earth (>100 MeV) proton events (filled gray circles placed at a Φ_{100 MeV} value of 10 pfu s). Of these 17 bursts, only 1 of the 11 events that occurred from 1994 to 2016 (for which Wind Waves data are available; a data gap from 2014 October 28 to 2014 November 5 contained one other event) had an associated DH type II burst. In all, 7 of the 17 cases (including the event with an associated DH II) had an associated metric type II burst for an overall type II association rate of 41% (7/17). Conversely, 18 of the 24 (75%) well-connected SPEs in Table 1 of Grechnev et al. (2015) with Wind Waves coverage had associated DH type II bursts, and the overall type II association rate (DH and/or m) for SPEs in Figure 9 was 96% (26/27). In regard to these statistics, we note that metric type IIs in themselves are not a reliable indicator of a proton event at Earth. Cliver et al. (2004) found that only ∼25% of metric type II bursts from western hemisphere flares from 1995 to 2001—that were not accompanied by DH type IIs—were associated with >20 MeV proton events. When a metric II had a DH counterpart, the association rate rose to 90%.

The 2000 November 8 SPE, the other outlying event in Figure 9, was one of the largest events of the space age. To the best of our knowledge, the peak prompt (nonshock) >10 MeV proton intensity of ∼1.2 × 10⁴ pfu is the highest yet recorded. Despite its large size at >10 MeV and detection in the GOES > 700 MeV proton channel, the 2000 November SPE failed to produce a ground-level event (Mewaldt et al. 2012; Thakur et al. 2016). Remarkably, there is evidence that the 2000 November SPE, like that of 2013 September, had its origins in a DSF. The 2000 November event arose in a complex of three active regions near the west limb (∼W75) of the Sun that had sunspot areas ranging from 50 to 120 millionths of a solar hemisphere. Nitta et al. (2003a) compared Transition Region and Coronal Explorer (TRACE; Handy et al. 1999) images and SOHO LASCO data for this event. In reference to Figure 10, reproduced from their paper, they wrote, "In TRACE 171 Å images, multiply oriented loops (seen in the larger box (in the left panel)) were seen to rise at 10–20 km s⁻¹ before the data gap of 22:11–22:35 UT The loops were probably linked to the dark filament seen in absorption against the solar disk (in the smaller box). During 22:49–22:54 UT, the lower part of the loops as marked by the dotted line in the second image (right panel) was seen to move at ∼80 km s⁻¹. The rising motion (was accompanied by the disappearance of the dark filament and) left a faint two-ribbon flare, as circled in white (not visible in the GOES light curve). This was at an area outside two minor active regions (ARs 9218 and 9222), to the north of the M7.7 flare (circled in black), which originated in AR 9213. The CME was first seen at a height of 3.9 R_⊙ at 23:06 UT so the associated SEP event appears to have originated in the non-active region eruption rather than the M7.7 flare (which reached maximum at ∼23:25 UT), although the entire complex of minor active regions was probably involved at some level. During the eruption, the nonthermal signature was very weak, as indicated by the 17 GHz time profile." Given the lack of high-cadence inner coronagraph observations in 2000 November (LASCO C1 (1.1–3 R_⊙) images are not available after 1998 June), however, the height–time plot of the leading edge of the CME in the 2000 November 8 event (Figure 11) does not allow us to say definitively that the CME arose in the faint two-ribbon flare rather than the M7.7 SXR flare (see Zhang et al. 2001).

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

That said, there is indirect evidence that suggests a DSF source for this large SPE. Figure 6 in Thakur et al. (2016), reproduced here as Figure 12, indicates a possible type II onset at ∼22:58 UT with a starting frequency of ∼40 MHz (see also Figure 5(b) in Cane et al. 2002). More clearly, a DH type II was observed by Wind Waves with a start time of 23:20 UT and a starting frequency of ∼4 MHz. Both of these frequencies lie below the average starting frequencies of the type II bursts associated with the high-energy SPEs (GLEs) detected by neutron monitors (107 MHz) and "regular" (i.e., non-GLE and non-DSF) SPEs (81 MHz) and are consistent with those for DSF-associated SEP events (22 MHz) (Gopalswamy et al. 2017a). Additional support for a DSF association for the 2000 November 8 SPE is provided by the low ratio of time-integrated iron (Fe) and oxygen (O) intensities for this event in the 30–40 MeV nucleon^–1 range (0.041 ± 0.046), the second lowest of that for a sample of 44 large gradual SEP events analyzed by Tylka et al. (2005). In the current picture of SEP acceleration (e.g., Tylka et al. 2005; Reames 2013), low Fe/O ratios are taken to be a signature of quasi-parallel shock acceleration. The 1998 April 20 event, which had the lowest values of this parameter (0.019 ± 0.003), arose in a decaying active region and appears to have been near the DSF end of the spectrum of eruptive events (Nitta et al. 2003a).

Zoom In Zoom Out Reset image size

As is the case in Figure 5, and similar to the 2000 November event, the 2013 September 29 event occupies parameter space to the left of the broad (oval) main sequence in Figure 9. Gopalswamy et al. (2015) report a starting type II frequency of 13 MHz for this event. In the "hierarchal" model of Gopalswamy et al. (2017a), low starting frequencies are linked to low initial CME acceleration rates and soft SPE spectra. Conversely, type II bursts with high average starting frequencies (∼100 MHz) tend to be associated with CMEs with high initial acceleration rates and SEP events with harder spectra. We are not aware of published Fe/O ratios for the 2013 September 29 event and, more generally, for the SPEs of cycle 24.

Figure 13, adapted from Cliver et al. (2012), is a scatter plot of log (prompt peak > 10 MeV proton intensity) versus log (1–8 Å peak SXR flux) for a sample of well-connected SPEs (W20–W85) from 1997 to 2005. In the plot, the data point for the 2000 November 08 SPE is highlighted along with that for 2002 May 22 (identified as a DSF-associated SPE by Gopalswamy et al. 2015) and two additional DSF-associated events from Gopalswamy et al. that fell outside of the 1997–2005 time range (2011 November 26 and 2013 September 29). This plot also includes a data point for the last GLE (on 2006 December 13; X3.4, 680 pfu) of cycle 24. We examined other events on the upper side of the scatter in proton intensities for similarities in solar sources to those of the four highlighted events. Four such events were identified. Of these four events, one (2000 April 04) has been associated with a DSF in the literature (Huttunen et al. 2002; see also Gopalswamy et al. 2015). For the other three events, there is plausible evidence that—like the 2000 November 08 event—the SPE arose in a relatively weak field source on the Sun (for details on the origins of these eruptions, see the notes to Table 2 below). We will refer to these three events, along with the 2000 November 8 event, as quasi-DSF-associated SPEs. The 1998 September 30 event originated in a magnetically simple (magnetic class α, single polarity) decaying active region with spot area of 100 μsh. While the 2001 November 22 event was associated with an M9.9 SXR event and a large active region (550 μsh), the flare was accompanied by the disappearance of one or more adjacent filaments (or segments of longer filaments) as recorded in images from Kanzelhoehe Observatory. The active region associated with the 2004 July 25 SPE was even larger (1340 μsh; M1.0 SXR flare), but the eruption appeared to involve a transequatorial loop system overlying relatively weak magnetic fields (Figure 14). Thus, for both the 2001 November 22 and 2004 July 25 events, as for the 2000 November 08 SPE, the presence of a nearby large flare and/or a major active region may be misleading regarding the nature of the eruption. Ideally, the solar circumstances should be examined for all the events in Figure 13, using imaging data as was done for the 2000 November 8 event by Nitta et al. (2003a).

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Figure 15 is a plot of all the events in Figure 13 for which a determination of the type II bursts associated with the high-energy Fe/O ratio was available; here we use the 25–80 MeV Fe/O values of Cane et al. (2006) and Mewaldt et al. (2012). We estimated the Fe/O ratios for the two cycle 24 DSF-associated SPEs (2011 November 26 and 2013 September 29) identified by Gopalswamy et al. (2015) using data from the Solar Isotope Spectrometer (SIS; Stone et al. 1998a) on the Advanced Composition Explorer (ACE; Stone et al. 1998b) (http://www.srl.caltech.edu/ACE/ASC/level2/new/intro.html). Both Tylka et al. (2005) and Cane et al. (2006) used an Fe/O value of 2.0 to mark the separation of SEP events into two groups (c.f., Cliver 2009). In the Cane et al. (Tylka et al.) formulations, Fe/O ratios <2.0 indicate shock (quasi-parallel shock) acceleration, and those ≥2.0 signify flare (quasi-perpendicular shock) acceleration of SEPs. Figure 15 contains only large (gradual) SPEs; no large impulsive SEP events from the lists of Reames & Ng (2004) are included. The largest impulsive SPEs on the Reames & Ng list have peak intensities of ∼1–3 pfu (Cliver & Ling 2007; Cliver 2009). The data points for the SPEs with low (<2.0) and high (≥2.0) Fe/O ratios in Figure 15 are color-coded as light blue and black, respectively. In Figure 15 the DSF- and quasi-DSF-associated SPEs have been grouped into a light-blue oval. Analogous to Figure 7, the black oval encompasses the main sequence.

Zoom In Zoom Out Reset image size

In Table 2, we list the date, SXR peak time, SXR class and fluence, and solar source location of each of the eight identified events in Figure 15 along with CME speed, type II starting frequency, DH type II association, 1 MHz radio fluence, prompt peak of >10 MeV proton intensity, >10 MeV "proton yield" (defined as the ratio of peak proton intensity above a given energy to the 1 MHz fluence; Cliver & Ling 2009), and Fe/O ratio at ∼30 MeV. In Table 2, the entries for the four quasi-DSF events are given in italics. The Fe/O ratios for seven of the eight listed events are <2.0. The SPEs on the main sequence had a more even distribution of low (15 events) and high (16) Fe/O ratios. The 40 MHz median type II starting frequency of the eight events lies below those for "regular" SPEs (81 MHz) and GLEs (107 MHz) determined by Gopalswamy et al. (2017a). The median CME speed of the seven events for which these data exist was 1333 km s⁻¹ versus a median speed of 1891 (Gopalswamy et al. 2012) for the nine GLEs plotted in Figure 15.

The results in Figure 15 and Table 2 are based on a relatively small sample of events. They will need to be substantiated using data from cycle 24.

The last two flare emissions we will consider that have been compared with SEPs are flare SXR fluence and ∼1 MHz radio fluence. In their Empirical model for Solar Proton Event Real Time Alert (ESPERTA) to provide short-term warning of ≥S1 and ≥S2 SPEs, Laurenza and colleagues (Laurenza et al. 2009, 2018; Alberti 2017) used these two parameters, along with flare location on the Sun, as inputs. In ESPERTA, the SXR emission is taken to be a measure of overall flare energy and the ∼1 MHz radio emission, formed at a height of ≳10 R_⊙, is used as a gauge of particle escape, while flare location takes SEP propagation into account. Note that the ∼1 MHz emission (primarily type III) indicates escaping electrons, with the assumption that such electrons are accompanied by protons (e.g., Cane et al. 2002).

Cliver & Ling (2009) used the proton yield parameter to distinguish between classic impulsive and gradual SEP events. Here we use it to characterize the DSF- and quasi-DSF-associated SPEs. The median >10 MeV proton yield of 1.35 × 10⁻³ pfu/(sfu min) of the events listed in Table 2 is well above (∼90 times) the median value of 1.54 × 10⁻⁵ pfu/(sfu min) for a sample of 20 independent, large, well-connected events considered by Cliver & Ling (2009). For context, the median proton yield of the classic impulsive SPEs considered by Cliver & Ling was 3.37 × 10⁻⁸ pfu/(sfu min), much smaller than either of these two values. The gradual SPEs with high proton yields based on 1 MHz radio emission are analogous to the proton-abundant events of Grechnev et al. (2015), although only the 2000 November 08 SPE in Table 2 is among the black square outliers in Figure 7. The other four outliers fell on the main sequence in Figure 15.

Figure 16, adapted from Laurenza et al. (2018), is a scatter plot of 1 MHz radio fluence, measured by the Wind Waves experiment, versus GOES 1–8 Å SXR fluence.¹³ The dashed contour line is a probability threshold that is chosen to maximize the number of correct predictions of ≥S1 SPEs while minimizing the number of false alarms. In this plot for all ≥M2 SXR flares from 1995 to 2014 that had flare longitudes from W20 to W120, events with data points located above the dashed-line probability threshold are well associated (∼65%; 42/64) with large (≥10 pfu at >10 MeV) SEP events (indicated by diamonds above the contour, stars below, with the peak >10 MeV intensity color-coded), and those below the contour are less likely to have SEP association (∼6% (17/281); all non-SEP-associated events are indicated by small open black circles). Thus, Figure 16 can be used to provide short-term alerts of impending SEP events. In the ESPERTA forecast tool, alerts are not issued for SPEs associated with <M2 SXR flares in order to reduce the false-alarm rate. Thus, five of the eight events in Table 2 are counted as "missed" events, i.e., SPEs meeting the NOAA forecast threshold of >10 pfu at >10 MeV that were not predicted. The data points for these events based on the parameters in Table 2 are plotted as open red squares in the figure. In each case these points fall outside the contour. Three of the four SPEs from Table 2 that fall within the dashed-line probability contour—marked by black circles—were quasi-DSF-associated events: 1998 September 30 (M2.7), 2000 November 8 (M7.0), and 2001 November 22 (M9.5). As appears to have been the case for the 2000 November 8 event, however, these formally correct predictions may be due to a shock-driving eruption from a weak magnetic field region that occurs in conjunction with a peripheral large flare.

Zoom In Zoom Out Reset image size

In Figure 17 we present a schematic, patterned after Figure 15, that characterizes our view of the general relationship between SPE and flare size parameters in large, well-connected prompt SEP events. This view is based on comparisons of SEP parameters versus gamma-ray line emission (Cliver et al. 1989), SXR emission (Cliver et al. 2012; Laurenza et al. 2018), 1 MHz radio emission (Cliver & Ling 2009; Laurenza et al. 2018), 35 GHz radio fluence (Grechnev et al. 2015), and reconnection flux (Kahler et al. 2017), covering a range of SEP energies from >10 to >100 MeV. There are three main zones or regions in the diagram:

• (1)  
  DSFs with associated SPEs. This region indicates that large flares are not a necessary condition for a large SEP event. This is the least populated zone in the schematic because DSFs generally may be thought of as "soft" low-energy two-ribbon flares (Kiepenheuer 1964) and thus are less likely to have high-speed CMEs that can drive shocks. That said, the assessment of Klein & Dalla (2017) that "whatever the interpretation of the filament-associated SEP events, there are at best very few SEP events associated with a CME and no alternative signature of particle acceleration in the corona" misses the point about the importance of anomalies (e.g., Sturrock 2017). Also, the characterization by Belov (2017; quoted in Section 1) that such SPEs either originate in sources behind the Sun's western limb, for which the flare source is occulted, or are small proton events occurring on a high SXR background that masks the flare source does not hold for the 2013 September 29 event (Figures 1–3) and the other events in Table 2. We note that the separation of the ("weak flare/strong SPE") region from the main sequence (i.e., zone (1) from zone (2)) in Figure 17 is for illustrative purposes only—to highlight the importance of DSF-associated SPEs for our understanding of SEP acceleration. It is not meant to imply a difference in the primary mechanism of SEP acceleration (CME-driven shocks) operating in this zone and on the main sequence, although the details of this process (see Section 3.2.2) may have a quasi-systematic variation. Nor do we expect a sharp demarcation of SEP properties between zones 1 and 2, but rather a gradient as observed for the Fe/O ratio in Figure 15.
• (2)  
  The main sequence, large flares with SPEs. The general correlation of SEP and flare parameters on the main sequence is attributed to the big flare syndrome (Kahler 1982). Big flares tend to have more of everything, so such high-scatter correlations are inescapable in comparisons of flare and SEP parameters. The main sequence is a key reason for the resilience of the view that a flare-resident acceleration process plays an important role in proton acceleration at the Sun, particularly at higher energies (e.g., Dierckxsens et al. 2015; Grechnev et al. 2015). In Dierckxsens et al. (2015) and Trottet et al. (2015), as well as Cliver et al. (2012), the authors considered a list of SEP events and their associated flares, not taking into account comparable flares (in the parameter considered) that lacked such SEP association. Figures 4 (adapted from Kahler et al. 2017) and 9 (after Grechnev et al. 2015)—which both take non-SEP flares into account—show how ignoring flares without associated SEPs overstates the SEP versus flare correlation.
• (3)  
  Flares without associated SPEs (only background SEP intensities). This region indicates that large flares are not a sufficient condition for a large SEP event. For the reconnection flux (Figure 4), this is the most populated zone of the diagram, while there are relatively few gamma-ray line flares that lack associated SPEs (see Figure 2 in Cliver et al. 1989). This difference is a reflection of the big flare syndrome—larger, more energetic flares such as gamma-ray flares are more likely to have associated fast CMEs, shocks, and therefore SEPs. In comparison, only half (36/72) of the reconnection flux events in Table 1 had associated CMEs. This zone of events consists of two subclasses: (1) flares without CMEs and (2) flares with relatively slow (median speed ~400 kms^-1) CMEs. Both conditions argue against shock acceleration of SEPs (see, e.g., Figure 2 in Cliver & D'Huys 2018).

Zoom In Zoom Out Reset image size

As zones 3 and 1 of Figure 17 show, respectively, big flares can lack SEP association and some weak flares can be associated with large SPEs. Taking both of these regions into account weakens SEP versus flare correlations (e.g., Figures 4 and 9). In addition, zone 1 underscores the prevailing view that fast CMEs and the shocks they drive are the principal requirements for large SEP events.

We are not saying that protons are not accelerated in impulsive flares (or during the impulsive phase of fully developed flares). They clearly are, including to high energies (see, e.g., Forrest & Chupp 1983; Forrest et al. 1985; Chupp et al. 1987; Ackermann et al. 2012). Also, it has long been known that type III radio bursts, the characteristic emission of the flare impulsive phase (Wild et al. 1963), are associated with electron-rich (Lin 1970) and ³He-rich events (Reames et al. 1985) at 1 au—so particles can escape from flares. Such SPEs will populate parameter space below zones 1 and 2 and above zone 3 in Figure 17 (as in, e.g., Figure 11 in Cane et al. 2010). What is lacking at this point, however, is compelling evidence that flares, either during their classic impulsive phase or in the extended phase of type III emission first identified by Cane et al. (1981), are able to make a significant, let alone dominant, contribution to large (gradual) proton events. The principal problem is the proton deficiency of flare-accelerated particles. The electron-to-proton ratios (for both (0.5 MeV electrons/>10 MeV protons) and (0.5 MeV electrons/>100 MeV protons)) of the largest impulsive SEP events are approximately two orders of magnitude larger than those of the largest gradual SPEs (Cliver & Ling 2007; Cliver 2016). It is not possible to produce a large SPE by scaling up a classic impulsive SPE without giving rise to electron events much larger than observed to date. For the extended or delayed 1 MHz emission, termed type III-l by Cane et al. (2002), Cliver & Ling (2009) showed that the median >30 MeV proton yield for a sample of large proton events was ∼280 times larger than that for classic impulsive events and concluded that the greater proton yield of the gradual events was due to the fact that such events are strongly associated with DH type II shocks, in contrast to large impulsive SPEs.

Cane et al. (2010) criticized the Cliver & Ling (2009) analysis, writing, "(1) ... Cliver & Ling (2009) failed to take account of the timing of the type III emission relative to the flare emissions. (2) Their comparison of radio intensities and proton intensities takes no account of the fact that they compare inferred total electron intensities relatively close to the Sun with observations at 1 au, where only a part of the particle beam would be intercepted." Regarding item 1, Cane et al. are correct that Cliver & Ling combined the impulsive and delayed type III emission for gradual events, but excising the impulsive phase type IIIs to isolate the delayed emission would only make the proton yield larger for gradual SPEs, further distinguishing them from the low-yield impulsive SPEs. Regarding item 2, what Cliver & Ling (2009) did in their Figure 6 (a scatter plot of proton intensity vs. 1 MHz fluence) is no different conceptually than what Cane et al. (2010) did in their Figure 11 (a scatter plot of proton intensity vs. SXR intensity). In fact, the assumption both studies followed—that flare electromagnetic emissions are (or could be) representative of SPE size parameters at Earth—underlies all such flare–SPE comparisons. One cannot rule out the possibility that the late-phase type III bursts associated with large (≥10 pfu at >10 MeV) SPEs are a signature of a nonshock acceleration process that produces SPEs with substantially different composition and higher proton yields than those associated with impulsive phase type IIIs, but this remains to be proven.

Gopalswamy et al. (2015, 2016, 2017a) have recently reported that, on average, the CMEs associated with DSFs have lower initial acceleration rates (leading to greater heights of shock formation and lower type II starting frequencies) than the eruptions associated with GLEs. This linkage of shock and SPE properties counters the argument of Grechnev et al. (2015) that the soft SEP spectra of "non-flare-related-SPEs" weighs against the capability of shocks to accelerate protons to high energies on the main sequence. More directly, the consensus of several recent analyses of late-phase solar gamma-ray events that are associated with main-sequence flares (median SXR class of X1.1; Table 1 in Share et al. 2018) is that CME-driven shocks are capable of accelerating protons to the highest energies (≳300 MeV) that can be inferred from gamma-ray observations (e.g., Pesce-Rollins et al. 2015; Ackermann et al. 2017; Plotnikov et al. 2017; Gopalswamy et al. 2018; Kahler et al. 2018; Jin et al. 2018; Omodei et al. 2018; Winter et al. 2018; c.f., Grechnev et al. 2018; Hudson 2018; Klein et al. 2018).

The hierarchal view of SEP acceleration of Gopalswamy et al. (2016, 2017a), viz., the correlation of initial CME speeds (i.e., early acceleration rates) and SPE spectra, had a precursor in the Nitta et al. (2003a) study that suggested differences in SPE spectra/composition/charge state between eruptive flares exhibiting explosive and gradual (during an extended pre-eruption phase) outward motions. The hierarchal picture can be seen as being complementary to the Tylka et al. (2005) shock formulation if quasi-perpendicular shocks are most effective low in the corona where the CME expands laterally, and quasi-parallel shocks dominate above the nominal ∼2 R_⊙ maximum height of closed loops. The analytical model of Tylka & Lee (2006) incorporated shock geometry and seed particle populations to provide an explanation of the event-to-event variation of elemental composition at high energies first noted by Breneman & Stone (1985). In this model, SPEs with high Fe/O ratios are attributed to quasi-perpendicular shock acceleration of flare suprathermals (c.f., Mewaldt et al. 2007, 2012) and events with low Fe/O ratios to quasi-parallel shock acceleration of coronal or solar wind suprathermals. If the predominantly low Fe/O ratios of DSF- and quasi-DSF-associated SPEs in Table 2 indicate quasi-parallel shock acceleration, then the near-even mixture of the number of SPEs with high and low Fe/O ratios (17 and 15, respectively) in the black oval in Table 15 implies a greater role for quasi-perpendicular shock acceleration on the main sequence.

A key test of understanding of SEP acceleration at the Sun is the ability to issue reliable alerts of impending SEP events based on observations of eruptive flares. Consideration of the DSF-associated SPE of 2013 September 29 in the proton alert study of Laurenza et al. (2018) indicates that attempts to understand large SPEs in terms of a flare-resident acceleration process, using flare diagnostics (Figure 16), are off track. The flare parameters for the 2013 September 29 DSF are an order of magnitude too small to provide an alert without an unacceptably high false-alarm rate. While flare-based alert methods such as ESPERTA have a demonstrated utility for timely warnings, significant advances in SEP forecasting require a focus on fast CMEs and shocks. Recent studies by St. Cyr et al. (2017) and Richardson et al. (2018) have placed more emphasis on these phenomena.