Lyman-alpha Variability During Solar Flares Over Solar Cycle 24 Using GOES-15/EUVS-E

The chromospheric Lyman-alpha line of neutral hydrogen (Ly$\alpha$; 1216\AA) is the strongest emission line in the solar spectrum. Fluctuations in Ly$\alpha$ are known to drive changes in planetary atmospheres, although few instruments have had the ability to capture rapid Ly$\alpha$ enhancements during solar flares. In this paper we describe flare-associated emissions via a statistical study of 477 M- and X-class flares as observed by the EUV Sensor on board the 15th Geostationary Operational Environmental Satellite, which has been monitoring the full-disk solar Ly$\alpha$ irradiance on 10s timescales over the course of Solar Cycle 24. The vast majority (95%) of these flares produced Ly$\alpha$ enhancements of 10% or less above background levels, with a maximum increase of ~30%. The irradiance in Ly$\alpha$ was found to exceed that of the 1-8 \AA\ X-ray irradiance by as much as two orders of magnitude in some cases, although flares that occurred closer to the solar limb were found to exhibit less of a Ly$\alpha$ enhancement. This center-to-limb variation was verified through a joint observation of an X-class flare that appeared near the limb as viewed from Earth, but close to disk center as viewed by the MAVEN spacecraft in orbit around Mars. The frequency distribution of peak Ly$\alpha$ was found to have a power-law slope of $2.8\pm0.27$, interestingly different from that of other observables. We also show that the data provide a clean timeseries for studies of ionospheric responses through a comparison with the Solar Flare Effect as observed by the Kakioka magnetometer.


Introduction
The Lyman-alpha (Lyα; 1216Å) line of neutral hydrogen, resulting from the 2p-1s transition, is the brightest emission line in the solar spectrum. During quiescent solar conditions, the wings of the line are formed in mid-chromosphere (∼6,000 K), while the core is formed higher up at the base of the transition region (∼40,000 K). Due to the high abundance of neutral hydrogen in the solar chromosphere, the Lyα line is optically thick with a broad central reversal. During solar flares, however, nonthermal electrons deposit large excess energies in the chromospheric plasma, generating localized heating and ionization at the flare footpoints. This results in enhanced Lyα emission and the line profile is predicted to go into full emission and lose its central reversal (Allred et al., 2005), although to date there have been very few Lyα flare profiles recorded at high spectral resolution. Further Lyα flare emission must come from the corona, where according to the well-established scenario the hot, dense flare plasma cools down from X-ray temperatures and eventually drains back to the lower atmosphere as coronal rain. However, given that the electron density and the abundance of neutral hydrogen are considerably higher in the chromosphere than the corona, it is assumed that the bulk of the Lyα emission discussed in this paper originates at the flare footpoints (Chamberlin et al., 2018).
Lyα photons drive the molecular dissociation of oxygen (O 2 ) in the Earth's mesosphere, allowing ozone (O 3 ) to form, while the photoionization of nitric oxide leads to the formation of the dayside ionospheric D-layer (∼80-110 km; Lean 1985). Changes in the Sun's output at these wavelengths can therefore have significant implications for the dynamics and composition of the terrestrial environment. Several studies have reported on variations in Lyα due to solar rotation and over the course of solar cycles (Lean & Skumanich, 1983;Lilensten et al., 2008;Woods, 2008). Woods et al. (2000) reported that the mean variability of Lyα due to the 27-day solar rotation across Solar Cycles 18-22 was 9%, dropping to 5% at solar minimum and increasing to 11% at solar maximum. Solar Lyα variability of the course of a solar cycle varied between a factor of 1.5 and 2.1. The importance of Lyα emission from other stars is also being realized as the search for habitable exoplanets intensifies. However, stellar Lyα is almost impossible to detect due to extinction by the interstellar medium, and so indirect reconstruction techniques need to be applied (e.g., Linsky et al., 2013). Furthermore, the integration times required to measure stellar EUV emission can be as much as 10 3 s, which would make the detection of any flare-related variations in Lyα very difficult (Christian et al., 2003). Solar variability in Lyα due to solar flares has not been extensively studied, largely due to instrumental limitations and their duty cycles. Canfield & van Hoosier (1980) were among the first to publish temporal variations in the Lyα line profile during two solar flares using the NRL spectrograph as part of the Apollo Telescope Mount onboard Skylab. Brekke et al. (1996) reported a 6% increase in Lyα irradiance using Upper Atmosphere Research Satellite/Solar-Stellar Irradiance Comparison Experiment (UARS/SOLSTICE), while Woods et al. (2004) reported a 20% increase in the line core and a factor of two increase in the line wings during the famous 28 October 2003 X28 flare from a serendipitous Solar Radiation and Climate Experiment (SORCE) SOLSTICE (McClintock et al., 2005) observation. While both SOLSTICE instruments capture disk-integrated emission, their cadence is not sufficient for capturing rapid temporal variations during flares. The Large Yield Radiometer (LYRA; Hochedez et al. 2006;Dominique et al. 2013) instrument on Project for On-Board Autonomy-2 (PROBA2) also captures full-disk Lyα emission with very high cadence (50 ms). Kretzschmar et al. (2013) reported Lyα signatures in 11 flares (C-and M-class) using LYRA, with only a 0.6% increase in emission detected during a detailed study of an M2 flare. The authors comment that this unusually low contrast may be due to the severe detector degradation suffered by the Lyα channel, but did not give a detailed explanation. Raulin et al. (2013) also reported a <1% increase in Lyα emission during seven solar flares using LYRA, but they also did not detect any appreciable ionospheric disturbances. Kretzschmar et al. (2013) also suggested that the Lyα flare emission varied more gradually than expected for an impulsively heated chromospheric feature. A similar discrepancy was reported by Milligan & Chamberlin (2016) for the Extreme-ultraviolet Variability Experiment (EVE; Woods et al. 2012) Multiple EUV Grazing Spectrograph Photometer (MEGS-P) on the Solar Dynamics Observatory (SDO; Pesnell et al. 2012). In the case of EVE, this was eventually attributed to a Kalman filter used to smooth the data during processing; this was ultimately replaced with a Fourier transform filter (Don Woodraska; Private Communication, 2017). Milligan & Chamberlin (2016) also showed that the Lyα lightcurves from the EUV Sensor (EUVS; Viereck et al. 2007;Evans et al. 2010) on Geostationary Operational Environmental Satellite (GOES)-15 behaved much more impulsively, as expected, consistent with Lyman continuum (LyC) observations from SDO/EVE MEGS-B and with the temporal behaviour of the Lyα emission from the flare presented by Woods et al. (2004).
As well as the space weather implications of changes in Lyα emission, Milligan et al. (2014) showed that some 6-8% of the energy deposited in the chromosphere by nonthermal electrons can be radiated away by the Lyα line alone (10 30 erg; see also Milligan et al. 2012). This single line therefore becomes one of the most important observables for studies of flare energetics. We note that da Costa et al. (2009) reached a similar conclusion (<10% of nonthermal energy) using Lyα images from the Transition Region and Coronal Explorer (TRACE; Handy et al. 1999) (see also Nusinov & Kazachevskaya, 2006). Milligan et al. (2017) also showed that the chromosphere responds dynamically to an impulsive disturbance at its acoustic cutoff frequency, as evidenced by 3-minute oscillations in the Lyα time profile from GOES-15/EUVS-E, which appeared in phase with similar periodicities measured in the LyC from SDO/EVE and the 1600Å and 1700Å lightcurves from the Atmospheric Imaging Assembly (AIA; Lemen et al. 2011) also on SDO. Thus some of a flare's energy could be dispersed via wave damping (kinetic energy), leading to different radiation signatures. This paper presents the first self-consistent, statistical analysis of almost 500 major solar flares observed in Lyα emission. The GOES/EUVS data have clean time profiles, with high signal-to-noise ratios for major events, making quantitative analysis possible. The time series abundantly confirm the recent discovery of an impulsive-phase sig-nature for Lyα, akin to that seen in hard X-rays. This opens a new quantitative channel for modeling of the crucially important flare processes at and near the footpoints of coronal magnetic structures. We primarily present a catalog of Lyα energetics, frequency distributions, and center-to-limb variations for future solar-terrestrial studies, and to highlight the availability and importance of an under-utilized dataset. Section 2 describes the data analysis techniques, including accounting for the effects of geocoronal absorption. The findings are presented in Section 3, including a section on ionospheric and geomagetic consequences. The conclusions are presented in Section 4.

Data Selection and Reduction
The GOES series of spacecraft have been providing a near-uninterrupted measurement of the solar X-ray irradiance since 1975 beginning with GOES-1 via their X-Ray Sensor (XRS) instruments (GOES-8 →; Hanser & Sellers 1996). These data have become the "industry standard" for classifying solar flare magnitudes from the peak of the 1-8Å channel. The launch of GOES-13 in 2006 (and subsequently, GOES-14 and -15) saw the inclusion of the EUVS in order to characterize the solar EUV irradiance in a similar manner. EUVS comprises five channels: A, B, C, D, and E, covering the 50-170Å, 240-340Å, 200-620Å, 200-800Å, and 1180-1250Å wavelength ranges, respectively (Viereck et al., 2007), with the E-channel centered on the Lyα line at 1216Å.
The E-channel data have been converted to radiance by scaling to a Whole Heliosphere Interval quiet-Sun reference spectrum (http://lasp.colorado.edu/lisird/data/ whi ref spectra ;Woods et al. 2009). The conversion to physical units for radiance will therefore not reflect flare-related time variations of the line profile, generating some systematic uncertainties. Over time, the E-channel suffers from degradation, which is compensated for by scaling the daily average values to those from SORCE/SOLSTICE. While Lyα measurements from GOES-13 and -14 have been sporadic, GOES-15 has provided continuous coverage since its launch in 2010. GOES-15 continues to take measurements in the EUV although at the time of writing, only data up to 6 June 2016 have been made publicly available (https://www.ngdc.noaa.gov/stp/satellite/goes/doc/GOES NOP EUV readme.pdf). Only flares for which both EUVS-E (Version 4) and XRS data were available from GOES-15 were included in this study.
The SDO/EVE experiment has a MEGS-B component, which returns whole-Sun spectra at EUV wavelengths with 10 s time resolution and 1Å spectral sampling. The same instrument measures spatially-and spectrally-integrated Lyα emission via its MEGS-P photometer, also at 10 s cadence. Unforeseen detector degradation post-launch meant that MEGS-B was only able to observe the Sun in Lyα for 3 hours per day, plus 5 minutes every hour for most of the SDO mission, and so it has observed significantly fewer flares than GOES-15 (see Table 1). In 2015, the flight software was updated to have MEGS-P (and MEGS-B) respond to increases in the soft X-ray emission detected by its ESP channel, making EVE a dedicated flare instrument. Recently (19 April 2018), to further preserve the signal-to-noise ratio as the instrument deteriorates, the cadence was reduced to 60 s.
The left-hand panel of Figure 1 shows the complete six years of currently available GOES-15/EUVS-E data covering Solar Cycle 24. The presence of the 27-day period due to active region rotation is clearly visible. By detrending these data (using a 120-point/20 minute smoothing function) to remove solar-cycle and rotation timescales, the biannual eclipse seasons -seen as dips in the time profile in the right-hand panel of Figure 1 -become apparent. These result from geocoronal absorption by the Earth's uppermost atmosphere, which has substantial opacity to Lyα line-core photons (Meier & Prinz, 1970;Baliukin et al., 2019). Note that the timing and depth of this dip varies over the course of the year as the EUVS instrument peers through different column depths of the geocorona throughout its orbit. The daily dips typically last for around ±4 hours of local midnight, and range from 0.3-6%, with the greatest decreases occurring at the time of the equinoxes. The geocorona is transparent to X-rays, but there are periods when the GOES satellites are in eclipse, blocking out both X-rays and Lyα.
One of the best-established solar flare catalogs available is that from the GOES/XRS photometers, as characterized by the flux of its 1-8Å passband (https://www.ngdc.noaa .gov/stp/satellite/goes/index.html). This list is compiled by the National Atmospheric and Oceanic Administration (NOAA) and extends back to 1977. Over the six years of available GOES-15/EUVS-E data, the catalog lists the start, peak, and end times of 677 M-class flares and 45 X-class (see Table 1). While the NOAA event list does not always provide heliographic locations for each of these flares, Milligan & Ireland (2018) recently compiled a list of flares observed by multiple instruments during Solar Cycle 24, which included flare locations based on SDO/AIA 94Å (∼10 MK) difference images. These (coronal) locations are determined from the SSW Latest Events list, which is accessible through the Heliophysics Events Knowledgebase (HEK; https://www.lmsal.com/hek/). In this study the locations are needed to derive the center-to-limb variation (CLV) of (excess) Lyα emission (Section 3.3). However, this database is incomplete and is missing several months of information. The number of flares for which location information was available was 573 M-classes and 33 X-classes. Given that the effect of geocoronal absorption on the flare excess emission was assumed to be nonlinear, and therefore difficult to correct for, all events for which the GOES start and/or end time lay within ±2σ of the minima of all geocoronal dips were omitted. This left a sample of 446 M-class flares and 31 X's (477 in total). For comparison, from the launch of SDO until 6 June 2016, EVE MEGS-P observed 94 M-class flares and eight X-classes flare in their entirety (between GOES start and end times), not counting those that were partially observing during its 5-minute observing periods. Figure 2 shows a sample of nine X-class flares in both X-rays (grey curves) and Lyα (black curves) to illustrate the quality of the EUVS data and to show the variety of Lyα responses during flares of comparable X-ray magnitudes. Their GOES classifications and heliographic locations are annotated in the top right corner of each panel. Panel a shows the first X-class flare of Solar Cycle 24 as previously reported by Milligan et al. (2012Milligan et al. ( , 2014; Milligan & Chamberlin (2016); Milligan et al. (2017). Interestingly, the flare in panel b is the largest in the sample (X6.9), which only shows a modest Lyα increase, while the more common X1.8 flare in panel c exhibited the largest Lyα contrast (29% increase) of all the flares studied. Some events displayed multiple Lyα bursts (panels d, f , and i), while others showed no variation due to being either behind (panel e) or close to (panel h) the solar limb. Even back-to-back events of similar magnitudes from the same active region can exhibit substantially different Lyα profiles (panel g). Figure 3 illustrates how background subtracted Lyα profiles were generated for all flares considered. The top left panel shows XRS data over a 24-hour period centered on the peak of the X2.7 flare that occurred on 5 May 2015 (SOL20150515). For each flare, the background of the 1-8Å channel was taken to be the minimum value within this 24hour period (denoted by the horizontal red line). The resulting background subtracted profile, between the start and end times of the GOES event (shown as vertical dotted lines in the two left hand panels), is then shown in the top right hand panel. Note the the largest flares often have peak fluxes several orders of magnitude above the background level, and so are not particularly sensitive to background subtraction (Ryan et al., 2012).

Background Subtraction
The bottom left-hand panel of Figure 3 shows the corresponding EUVS-E Lyα lightcurve over the same 24-hour period. Comparing with the panel above, it is clear that only the largest flares produce enhanced Lyα emission above the background level. Also visible is the daily geocoronal absorption dip beginning around 07:00 UT (vertical dashed line). The presence of these geocoronal dips coupled with the fact that changes in Lyα due to flares are not as substantial as those in X-rays requires a more careful subtraction of the solar background. The entire 24 hour period was fit with an inverted Gaussian to account for the geocoronal dip, plus a constant equal to the modal value over the time interval. Subtracting this background reveals the flare excess emission between the start and end times of the corresponding GOES event (bottom right hand panel). The ratio of the peak Lyα flux relative to the background value then defines the contrast (percentage increase above background; see Section 3.1).

Flare Energetics
Having established the flare excess emission in both Lyα and X-rays between the start and end of each GOES event, the total energy emitted over each wavelength range can be determined by integrating each profile over time. Converting from W m −2 to erg s −1 , using: 1 W m −2 = 2 π (1 AU) 2 ×10 7 erg s −1 =1.406×10 30 erg s −1 . Deviations from 1 AU as the Earth -and therefore, GOES-15 -completes its orbit around the Sun were also accounted for. To get the total integrated energy meant also taking the instrumental cadence time into consideration, which is 10.24 s for EUVS and 2 s for XRS. The two XRS channels (0.5-4Å and 1-8Å) also allow the derivation of the temperature (T ) and emission measure (EM ) of the X-ray emitting plasma to be determined using the procedure outlined in White et al. (2005). Combining these results with the radiative loss function allows the radiative loss rate by the total SXR emitting plasma to be calculated through the simple expression 3n e k B T V , where EM = n 2 e V (see Section 3.2).
The biggest uncertainty in the determination of the total energy radiated by each mechanism is that of the end time of the flare as specified in the GOES event list. NOAA stipulate this time to be that when the X-ray flux equals half that of the peak value (https:// www.ngdc.noaa.gov/stp/solar/solarflares.html). In the largest events, X-rays can remain elevated above the background levels for several hours beyond the cataloged end time, while the impulsive nature of Lyα sees it return to background levels much more rapidly. Accounting for the 'real' flare end time would require fitting the decaying emission with, say, an exponential function and extrapolating in time, and is beyond the scope of this paper (see also Ryan et al. 2016). Therefore, to assess the effect of a longer flare duration on the relative energy radiated, the end time of the X-class flare shown in Figure 3 was arbitrarily extended by 90 minutes. As shown in the left hand panel of Figure 4, the X-ray flux (cyan curve) continues to decay beyond the listed end time (vertical solid line), while the Lyα flux (black curve) has almost returned to pre-flare levels. The cumulative radiated energy by the two processes are shown in the right hand panel as dashed curves, along with the background subtracted lightcurves (solid curves). The ratio of the two energies is plotted as a solid red curve, and does not appear to change appreciably between the listed GOES end time and 90 minutes later. The energies for both X-rays and Lyα reported in Section 3 can therefore be considered lower limits, although the ratio of the two may be broadly considered independent of the total integration time.

Flare Contrast
Solar flares are known to vary by over two orders of magnitude in the X-ray portion of the spectrum relative to the background emission during the most extreme events (Ryan et al., 2012), while the corresponding increase in Lyα irradiance can be marginal due to the much higher background at this wavelength. The left hand panel of Figure 5 shows a histogram of the peak Lyα flux for all flares in this study relative to their preflare background. This shows that 95% of flares exhibit a 10% enhancement in Lyα, with a maximum contrast of ∼30%. These variations are comparable to or greater than changes brought about during active region evolution, albeit on much shorter time scales (Woods et al., 2000). The right hand panel shows how this peak Lyα flux varies with the equivalent peak X-ray flux (essentially, GOES class). For M-class flares (∼10 −5 W m −2 ), peak Lyα flux is on average a factor of 10 more intense that that of the X-rays, and as much as a factor of 100 for some events. For X-class flares, the Lyα flux averages about three times that of X-ray flux, with a maximum of a factor of 10.
Previous reports of Lyα enhancements during flares have often focused on individual or small numbers of events. Woods et al. (2004) reported a 20% increase during an X28 flare observed by SORCE/SOLSTICE, while Milligan et al. (2014) reported an 8% increase during an X2.2 flare observed by SDO/EVE. Kretzschmar et al. (2013) and Raulin et al. (2013) both reported enhancements of <1% using data from PROBA2/LYRA. While the M2 flare presented by Kretzschmar et al. (2013) was not included in this current study, it was observed by GOES-14/EUVS-E which measured a ∼3% increase, suggesting that LYRA may have been underestimating flare-related enhancements to the solar irradiance. This study now presents a comprehensive overview of the variability of solar Lyα irradiance due to major solar flares through a systematic analysis of almost 500 events.

Flare Energetics
Both Lyα and soft X-rays are known drivers of changes in the D-layer of the ionosphere. Therefore knowing the proportional amounts of energy that are injected into the terrestrial atmosphere during flares is important for assessing their relative effects. Milligan et al. (2012) showed that twice as much energy was radiated by Lyα compared to X-rays during a single X-class flare that occurred close to disk center. The left hand panel of Figure 6 shows the total Lyα energy radiated relative to the total X-ray energy between the GOES start and end times of each event. Similar to that found for the peak fluxes presented in Section 3.1, approximately 1-100 times more energy was radiated via the Lyα line than the 1-8Å channel for most M-class flares. X-class flares were only up to 10 times more intense in Lyα than in X-rays. Some events showed very weak Lyα energies ( 10 28 erg) as they might not have exhibited an appreciable enhancement above the background due to their proximity to the solar limb, either because of opacity effects along the line of sight, the footpoints or ribbons being foreshortened or occulted by the solar disk (see Section 3.3).
The right hand panel of Figure 6 shows the same Lyα energies relative to the total thermal energy radiated from the flaring coronal loops as calculated in Section 2.2. In this case, the Lyα energies are comparable to, or around a factor of 10 less, than the total thermal energy. The most reliable measurement to date of a flare's 'true' energy stems from observations of flares in the Total Solar Irradiance (TSI) by Woods et al. (2006) using SORCE/Total Irradiance Monitor (TIM; Kopp & Lawrence 2005). The authors state that the total flare energy is approximately 105 times that of the 1-8Å energy, although this was only measured for four events. A follow up study of 38 events was carried out by Emslie et al. (2012). They claimed that the total thermal energy accounted for about one fifth of the bolometric energy. The findings presented here imply that Lyα is indeed a significant radiator of flare energy as suggested by Milligan et al. (2014).

Center-to-Limb Variation
As Lyα is known to be optically thick (Woods et al., 1995), some degree of centerto-limb variation (CLV) is to be expected, given that Lyα emission from flare ribbons nearer the solar limb will be scattered by the chromospheric column mass along the line of sight. Curdt et al. (2008) showed that this was the case for quiescent Lyα using spatiallyresolved data from SOHO/Solar UV Measurements of Emitted Radiation (SUMER; Wilhelm et al. 1995). Similarly, Woods et al. (2006) showed that a CLV was also applicable to the handful of flares observed in the TSI by SORCE/TIM. By measuring the total energy in the TSI flare excess relative to the GOES X-ray energy (which is optically thin, and therefore not attenuated by the solar atmosphere), for each event, the ratio, R, as a function of heliocentric angle was fit with the quadratic expression: where R C is the TSI/X-ray ratio at disk center, k is the limb variation relative to the center, and µ=cos(θ) (see also Brekke & Kjeldseth-Moe 1994). They found k = 0.11. The same technique was applied here to Lyα flares. The left hand panel of Figure 7 shows E Lyα /E X−ray as a function of heliocentric angle for both M-(black diamonds) and Xclass flares (solid red circles), with fits to the data using Equation 1 overplotted in black and red, respectively. The resulting values of R C and k for the M-class flares were 14.6 and 0.36, respectively, and 4.33 and 0.12 for X-classes, in broad agreement with Woods et al. (2006). The right hand panel of Figure 7 shows the corresponding peak contrast values of Lyα also as a function of heliocentric angles (grey diamonds) with the mean values overplotted as black solid circles with 1σ error bars for each integer value of angle. A similar 'tailing-off' can be seen towards the limb.
To underscore this effect, one of the X-class limb flares (X1.1, 19 October 2014, S13E57) was fortuitously observed in Lyα by both GOES  Figure 8) would have appeared close to disk center (S11W31 at Mars). The corresponding lightcurves from both instruments (at 1 AU, after correcting for the Earth-Mars distance and light travel time) are plotted in the bottom panel of Figure 8. The ∼80% increase in Lyα intensity measured at Mars relative to Earth can be attributed to either opacity effects, or the foreshortening of the Lyα emitting footpoints as seen from Earth. This flare occurred just before GOES-15 began to peer through the geocorona around 07:00 UT on that day.

Frequency Distributions
An important property of solar flares (along with other self-organized criticality systems) is the slope of its frequency distribution. It has been well established for decades that flare distributions can be well represented with a power law of the form: where f is the frequency distribution of a flare parameter, x, α > 0 is the power-law index, and C is a scaling constant. In the case of coronal (e.g. EUV) emission, a value of α > 2 would imply that, extending the distribution to lower energies, would contain enough energy to heat the solar corona (Hudson, 1991). This analysis has since been carried out over a wide range of wavelengths (hard X-rays; Hannah et al. 2011, EUV brightenings;Parnell & Jupp 2000, soft X-rays;Veronig et al. 2002). Traditionally, the approach has been to bin the data (e.g. peak flux) logarithmically and fit the distribution with a power-law function. However, this method is prone to selection effects, based on the bin sizes (e.g. Bai, 1993;Ryan et al., 2016). Using a Maximum Likelihood method instead removes most of these issues. We adopt the double power-law approach of Parnell & Jupp (2000), which not only determines the power-law index, α, but also the break in the double power-law, above which the well-sampled distribution can be reliably determined. This flattening of the distribution at lower fluxes is a well known selection effect in which the weaker events are harder to detect and hence undersampled. The resulting empirical frequency distribution and power-law fit for our Lyα flux is shown in Figure 9. Here an index of α = 2.82±0.27 is found for fluxes above 2.86×10 −4 W m −2 . This index is steeper than those found for the X-ray and other flare observables (α = 1.8 for hard X-ray bursts; Dennis (1985), α = 2.1 for soft X-ray fluxes; Veronig et al. (2002)). Where the empirical frequency distribution matches the power-law line in Figure 9 there is a good fit to the data. At higher fluxes we see that there are significantly fewer events than predicted by power law we have found, hinting at a physical effect limiting the event numbers; there is no bias against detecting more powerful events. A similar roll-over may have been measured in hard X-ray flare observations from Ulysses (Tranquille et al., 2009).

Acoustic Oscillations in MAVEN/EUM data
As mentioned in Section 1, Milligan et al. (2017) detected 3-minute oscillations in full-disk Lyα emission from GOES-15 during the 15 February 2011 X-class flare. This was interpreted as the generation or enhancement of acoustic waves in the chromosphere in response to the impulsive release of energy during the initial stages of the flare. The oscillation was found to be independent of rate of heating due to nonthermal electrons as determined from hard X-ray observations. To date, this is the only reported case of acoustic oscillations in full-disk EUV irradiance data. However, for the X-class flare described in Section 3.3 (19 October 2014), the standard wavelet analysis of Torrence & Compo (1998) was applied to the Lyα lightcurve from MAVEN/EUM. Figure 10 shows that the wavelet power around the onset of the flare is also enhanced at a period of 4.4 minutes. This deviates slightly from the value found by Milligan et al. (2017) but could also be indicative of flare-induced acoustic wave detection at 1.5 AU.

Ionospheric and Geomagnetic Consequences
Ionization resulting from solar UV and X-ray emissions mainly contribute to the Earth's ionosphere, which exhibits abundant variability of various forms. Solar flares in particular provide sudden changes in this ionizing radiation, and Lyα carries a substantial fraction of the flare perturbations. The geomagnetic "Solar Flare Effect" (SFE) in fact appeared in conjunction with the first ever recorded solar flare, SOL1859-09-01 (Carrington, 1859) in the form of compass-needle deflections. Until the present time there has been no reliable and systematic monitoring of Lyα as comprehensive and reliable as are now available from GOES.
To illustrate the possible effect of flare-related changes in Lyα on the ionosphere, the SFE due to enhanced conductivity recorded at the Kakioka magnetometer, Japan, for the major impulsive flare SOL2011-09-08 is shown in Figure 11. This figure shows that the flux increase in Lyα (black curve) is around an order of magnitude greater than that of the soft X-rays (cyan curve), and substantially higher than that of He II 304Å, another impulsive chromospheric emission line which was observed by SDO/EVE (green curve). The inset shows that the timing of the geomagnetic disturbance closely follows that of Lyα (and He II), while the X-ray emission peaks several minutes later, ruling it out as a driving mechanism.
In most cases the D-region ionization, as reflected by the Sudden Phase Anomaly ionospheric effect, matches the GOES soft X-ray flux and does not have a good correlation with the more impulsive Lyα emission (e.g. Raulin et al., 2013). In Figure 11 case we see a much more impulsive SFE in the Kakioka Y-component, approaching the impulsive time scale of Lyα. This hints at enhanced E/F-region current systems in this case, possibly related to the large energy flux in the Lyα line.

Conclusions
This paper presents an overview of over six years of solar flare observations in Lyα emission taken with the E-channel of the EUVS instrument on GOES-15. Prior to this, Lyα flare observations had been severely limited, with sometimes contradictory results, hampering the understanding of how these events contribute to the broader field of space weather. After removing flaring events that were affected by geocoronal absorption, this work shows that Lyα is significantly (1-100×) more energetic than the more commonly studied soft X-rays (1-8Å), which are also a driver of ionospheric disturbances. However, the Lyα contribution is less significant for flares that occur close to the solar limb, due to either opacity effects, or foreshortening or occultation of the flare footpoints as seen from Earth. This was verified through a simultaneous observations of a limb flare that was also observed by MAVEN/EUM when Earth and Mars were ∼90 degrees apart. The frequency distribution of peak Lyα flux revealed a power law slope of α=2.8, implying that the cumulative input from smaller flares pose a greater energetic input into Earth's atmosphere than the summation of larger events.
The high time cadence of the GOES-15/EUVS-E data also allows a more comprehensive comparison of the relative ionospheric effects between Lyα and X-rays. Using magnetometer data from Kakioka it was found that increased conductivity in the ionosphere in reponse to a major solar flare was highly correlated with with the increase in Lyα flux, while the corresponding increase in X-rays lagged the ionospheric response by several minutes, indicating that the X-rays could not have been the driver. Raulin et al. (2013) reported no such ionospheric response (as characterized by a VLF phase change) during seven solar flares observed by PROBA2/LYRA. The flares that they studied were of a lower magnitude that those presented here (C-and low M-class), and were therefore unlikely to have produced any enhancement above the solar Lyα background. Some of these flares also occurred close to the solar limb further diminishing the possibility of them contributing to variations in the solar Lyα irradiance.
While the most recent GOES-15 Lyα data (including those from the period of intense flaring activity in September 2017) have not yet been released at the time of writing, the heliophysics and space weather communities are also anticipating further Lyα observations from the next generation of GOES satellites, the GOES-R series. These four spacecraft will include a dedicated suite of EUV and X-ray instruments (EUV and Xray Irradiance Sensors, EXIS; https://www.goes-r.gov/spacesegment/exis.html), which will provide more advanced coverage of the Sun's output over the next 20 years or more. The Lyα line will be sampled across five pixels, giving a rough estimate of variations in the full-disk line profile.
Lyα also plays a very important role in trying to understand the physical processes that underpin solar flares themselves. The current study confirms that flare Lyα emission has a clear impulsive-phase peak. It had been established that Lyα is a significant radiator of flare energy, but this behavior also means that the data can be used to provide clues as to how the solar chromosphere and transition region respond to flare energy release. Future flare studies could look at the link between Lyα and LyC (Machado et al., 2018), or Lyα and Hα (Canfield et al., 1981), for example. The upcoming launch of Solar Orbiter will also include an Extreme Ultraviolet Imager (EUI; Schühle et al. 2011) than contains a Lyα channel as part of its High Resolution Imager (HRI) suite. EUI will image the Sun in Lyα at <1 s cadence, at 1 resolution at 0.3AU. The currently proposed Japanese Solar-C mission (the follow up to Hinode) is also expected to feature a Lyα spectrograph (Teriaca et al., 2011), as well as the Lyman-alpha Solar Telescope (LST) on the Chinese ASO-S satellite (Li, 2016). The findings presented here will also assist in the interpretation of results from these future observing platforms.