© R. Vainio et al., Published by EDP Sciences 2013

1. Introduction

Solar energetic particle (SEP) events are one of the most important elements of space weather (Vainio et al. 2009), being the defining component of solar radiation storms,¹ contributing to radio blackouts in polar regions and being related to many of the fastest coronal mass ejections (CMEs) driving major geomagnetic storms. In addition to CMEs, SEP events are also related to solar flares (e.g., Reames 1999). The occurrence rate of large, space-weather relevant SEP events is about 100 per solar cycle, with a broad distribution in time that extends well into the declining phase of the activity cycle as traced by the sunspot index.

Data Services and Analysis Tools for Solar Energetic Particle Events and Related Electromagnetic Emissions² (SEPServer) is a three-year collaborative project funded under the seventh framework programme (FP7-SPACE) of the European Union. The project started in December 2010. It will construct an Internet server that will provide access to a large number of SEP datasets from different instruments on-board half-a-dozen missions, to electromagnetic (EM) observations related to the events indentified from the SEP data, and to state-of-the art analysis tools that can be used to infer the solar SEP emission time profiles and interplanetary transport conditions prevailing during the SEP events. All datasets will be accompanied by comprehensive metadata as well as assessments of data quality. To assess their validity, the developed data-analysis methods will be compared against each other and solar EM observations. Furthermore, the project will provide analysis results on the events in the form of comprehensive SEP event catalogues that list key properties of the events. The server will be released to the community near the end of the project, in the fall of 2013.

This paper presents the SEPServer project and the results from the first year: the first SEP event catalogue. A systematic scan of the space-weather relevant³ proton intensities (55–80 MeV, or ~68 MeV) observed by SOHO/ERNE was performed over the years 1996–2010. A total of 115 proton events were identified and analysed using data from SOHO/ERNE, SOHO/EPHIN and ACE/EPAM. We will present and discuss here the results of the analysis of these events. We concentrate on the onset analysis utilizing two different methods, velocity dispersion analysis (VDA) and time-shifting analysis (TSA), to get estimates of the particle release times close to the Sun. Furthermore, we analyse the EM associations of 44 of the events, i.e., those that occur during the day time in Europe and have the best coverage in terms of data available at SEPServer. The analysis is not meant to be comprehensive but spawn a number of further efforts by the community, utilizing the data and preliminary analysis results presented in this paper. Likewise, while we attempt to compile a comprehensive list of events for the years 1996–2010, the resulting catalogue is not complete due to several data gaps and since the methodology applied misses some of the events that occur in the aftermath of another. Nevertheless, the community is encouraged to perform further studies making use of the SEPServer event catalogue for which full access will be provided through the project website (http://server.sepserver.eu).

The structure of the paper is the following. We present the SEPServer and the data it provides in detail in Section 2, describe the first SEP event catalogue produced by the project in Section 3, present and discuss the statistical analysis performed on the events in Section 4, and give our conclusions and outlook in Section 5.

2. The SEPServer project

The SEPServer project aims at facilitating access to a number of complementary SEP datasets, allowing in-depth analyses of SEP events to be performed. The emphasis of the project is on the space-weather relevant SEP events, i.e., events with proton spectra that extend to high enough energies to penetrate moderate shielding provided by spacecraft and constitute a space-weather hazard. The SEP datasets included in the project span over solar cycles 21–24 and the observations cover radial distances 0.3–5.4 AU. The SEP datasets distributed through SEPServer are listed in Table 1.

In addition to SEP data, crucial ingredients in the analysis of SEP events are the solar EM observations that are related to the events. Besides from context observations of, e.g., solar-wind plasma and magnetic field and CMEs related to the SEP event, important knowledge of the acceleration processes close to the Sun can be obtained from solar X-ray and gamma-ray observations as well as from solar radio observations that are generated by the accelerated particles, mostly electrons, that interact with the solar atmosphere. This occurs in the chromosphere and low corona for hard X-rays (HXR) and gamma-rays, and between the low corona and interplanetary space for radio-emitting electrons. While the CME observations are relatively easily accessible through a number of sources (e.g., SOHO LASCO CME Catalog⁴, see Gopalswamy et al. 2009, and the automated CACTUS CME-detection facility⁵, see Robbrecht et al. 2009), energetic photon and radio observations relevant to SEP events have not so far been collected to a single access point. Table 2 presents the EM observations distributed through SEPServer. At present, these datasets provide a comprehensive coverage of the identified high-energy proton events of the 23rd solar cycle. In future, they will be extended back to cover key events at earlier times, in particular the Nançay radioheliograph (available as 1-D imaging data for 1980–1995) and AIP/OSRA (to be recovered from microfilm archives) radio observations from the Helios era. These OSRA data have not been available in digital format prior to SEPServer. In addition to SEP and EM observations, SEPServer will provide local magnetic field data from each spacecraft of Table 1 carrying a magnetometer and plasma data from near-Earth solar wind (ACE and Wind).

In addition to a large number of well-documented datasets, SEPServer will provide the scientific community with state-of-the-art analysis tools based on numerical simulations of particle transport in the interplanetary medium. These tools allow one to deconvolve the effects of interplanetary transport from the observations in order to assess the source function of particle injection close to the Sun. This, in turn, can be directly compared with EM observations to perform accurate comparisons of the temporal and spectral relation between the interacting and escaping particles. The inversion tools (Agueda et al. 2008, 2009, 2012) are presently available for electron observations of ACE/EPAM, Ulysses/HI-SCALE and Wind/3DP. Work to implement the inversion methods for proton observations of the same instruments as well as on relativistic proton observations by the world network of neutron monitors is in progress. Note that certain criteria for the quality of observations (in terms of angular coverage and resolution and magnitude of the event) and quietness of the interplanetary medium have to be fulfilled in order for the simulation-based methods to be applicable (see Agueda et al. 2009). Thus, not all events distributed through SEPServer can be analysed using these methods.

The project will also perform scientific analysis of the SEP events using various kinds of data-analysis methods to investigate the relationship of the highest-energy proton and electron events and the corresponding EM solar observations. The methods employed range from the direct comparison of SEP and EM observations, applying TSA methods to crudely eliminate interplanetary particle transport effects, to the application of the simulation-based inversion methods. Direct data-analysis (DDA) methods⁶ will be compared to the advanced inversion methods and to EM observations to assess the range of validity of the direct methods. For first results demonstrating the concept and capabilities of the approach, see Malandraki et al. (2012), who analysed the SEP event of 13 July 2005 in detail using the DDA and inversion methods, which will be distributed through SEPServer.

3. SEPServer proton event catalogue for the 23rd solar cycle

3.1. Event selection

In order to establish a catalogue of high-energy SEP events, which are particularly relevant for investigation of in-orbit space-weather effects, we visually scanned through SOHO/ERNE (Torsti et al. 1995) data for the time period from May 1996 till the end of 2010 extending over a full solar cycle. Intensity enhancements of protons in the energy range 54.8–80.3 MeV (average energy 67.7 MeV) were searched for.⁷ The energy channel was purposely chosen well above the typically-used >10 MeV proton channel (e.g., Laurenza et al. 2009) available from GOES satellites. This was considered appropriate for the goals of the SEPServer project from the point of view of space-weather relevance, as discussed in the Introduction. Using this relatively high-energy range for event selection also allows more reliable identification of individual events in cases when events follow each other in rapid succession. In such cases small events at low energies tend to be masked by previous bigger events, while at high energies they can be better distinguished due to more rapid fall of intensities.

The selection criterion for the events in the SEPServer catalogue was that the 1 min average intensity in the 54.8–80.3 MeV ERNE proton channel was enhanced by a factor of ~3 above the quiet-time background of the appropriate phase of the solar cycle. During the observation period the quiet-time background varied from ~5 × 10⁻⁴ cm⁻² s⁻¹ sr⁻¹ MeV⁻¹ in 1996–1997 to ~3 × 10⁻⁴ cm⁻² s⁻¹ sr⁻¹ MeV⁻¹ in 2001–2003 and up to ~7 × 10⁻⁴ cm⁻² s⁻¹ sr⁻¹ MeV⁻¹ in 2009–2010. The total number of observed SEP events was 115 (Table 3). It should be noted that there were long breaks in the observations of the instrument from 25 June 1998 to 9 October 1998 and from 21 December 1998 to 8 February 1999 and several shorter data gaps as indicated in Table 3. The ERNE High Energy Detector has a very large geometric factor (~20–40 cm² sr) often causing saturation during very large SEP events. Therefore, events following very large ones before the recovery of the instrument may have been missed. Such cases are, among others, the events of 17 and 20 January 2005 (Cane et al. 2010) following the event of 15 January 2005.

3.2. Solar release time determination

We carried out VDA for all 115 events using data with 1 min time resolution. The analysis was based on 20 proton energy channels between 1.58 MeV and 131 MeV (Table 4). SOHO/ERNE covers this energy range by using two different sensors. The Low-Energy Detector (LED) operates in the range 1.58 MeV to 13.8 MeV and the High-Energy Detector (HED) from 12.7 MeV to 131 MeV. Both sensors provide 10 energy channels. The geometric factor of LED is only about 1 cm²sr and thus the statistics in many events are relatively low at the highest-energy channels of LED. The geometric factor of HED is 20–40 cm² sr, depending on energy, making it possible to observe many events up to 100 MeV and above.

VDA of an SEP event is based on determining the onset times of the event at a number of energies and presenting these onset times as a function of the inverse velocity of the particles at respective energies. The velocity dispersion equation at 1 AU can be written as$$t_{\mathrm{onset}}\left(E\right)=t_0+8.33\hspace{0.5em}\frac{\min }{\mathrm{AU}} \mathrm{s} {\beta }^{-1}\left(E\right),$$(1)where t_onset(E) is the observed onset time in minutes at proton kinetic energy E, t₀ is the release time (min) from the acceleration site, s is the apparent path length (AU) travelled by the particles and is the inverse speed of the particles. Thus, by linear fitting of the onset times as a function of the corresponding inverse speed, an estimate for both the release time and the apparent path length of the particles can be obtained. Onset-time determination was done by using the so-called Poisson-CUSUM method (Huttunen-Heikinmaa et al. 2005). This is analogous to a statistical quality control scheme deciding whether or not a process is in control, and if not, giving the exact moment of time when the failure happened. In this case, the failure is an SEP event causing intensities to rise above the pre-determined background. The updated algorithm used in this work allows changing background, which is often necessary to take into account, e.g., SEP events preceding the one under investigation, when the background has not yet reached a constant quiet-time value. Other criteria used for event onset determination were as described in Huttunen-Heikinmaa et al. (2005).

The SEP events investigated and the results of VDA are given in Table 3. Column 1 gives the identification number of the event, column 2 the date and column 3 the onset time (of 54.8–80.3 MeV protons) of the event. The maximum intensity, I_max, in column 4 indicates the size of the event as observed in the energy range 54.8 to 80.3 MeV by ERNE. Events with maximum intensity ≤1.0 × 10⁻³ cm⁻² s⁻¹ sr⁻¹ MeV⁻¹ in this energy range are very weak and generally might not be observed at higher energies by ERNE. Columns 5–7 give the properties of the associated soft-X-ray flare as deduced by Cane et al. (2010), who considered the >25 MeV proton events in 1997–2006. However, we have not included their data on flares that are separated more than 4 h from the start of the event.⁸ Note also that several events marked as “no flare” are actually events, where the parent solar activity occurred behind the limb Cane et al. (2010). The results of the VDA are listed in Columns 8 and 9, i.e., the deduced release time, t₀, and the apparent path length, s, travelled by particles with their standard errors, respectively. Column 10 gives the coefficient of determination (i.e. the squared linear correlation coefficient of the fit) related to the reliability of the results and describing the portion of the total variance in the onset time that can be explained as a linear relationship between the inverse velocity and the onset time. In column 11 the channel numbers of the data points discarded from the VDA are listed. In most of the events one or more data points were discarded from the VDA. The main reasons for discarding points were either high background from a previous event so that no additional enhancement was observed at low energies or the fact that the event did not cause noticeable intensity enhancement at the highest energies. Among the 115 analysed events there were four for which no reasonable velocity dispersion relation could be found. Also, for some events, while producing a fit of acceptable quality, VDA gives a value of the path length that can hardly be regarded as physical (either below 1 AU or much greater than it). As one purpose of this paper is to evaluate the performance of VDA, we have kept those events in the sample.

For electrons, we determined the onset times of the 115 events as observed by SOHO/EPHIN (Müeller-Mellin et al. 1995) and ACE/EPAM (Gold et al. 1998) in three energy channels: 0.18–0.31 MeV (EPAM), 0.3–0.7-MeV (EPHIN) and 0.7–3-MeV (EPHIN). The algorithm determines the average intensity I and the standard deviation σ inside the specified time window and compares the data just ahead of this window with a threshold I + nσ, where n can be chosen by the user (3σ and 4σ are used for EPAM and EPHIN, respectively). The onset is defined as the time stamp of the first point above the threshold. As all these electron channels have mean speeds close to the speed of light (0.73c, 0.86c and 0.98c, respectively), the data do not really allow for VDA to be performed. We determined the onset times of the electron channels for each event and then performed a back-shifting of the onset times to get the release times of the electrons at the Sun:$$t_{\mathrm{rel}}\left(E\right)=t_{\mathrm{onset}}\left(E\right)-8.33 \frac{\min }{\mathrm{AU}} \mathrm{L} {\beta }^{-1}\left(E\right).$$(2)

Rather than using the apparent path length obtained for protons, which is often severely affected by interplanetary scattering (Lintunen & Vainio 2004; Sáiz et al. 2005), we used the nominal length of the Parker spiral L computed for the solar-wind speed u_sw observed during the event, i.e.,$$L\left(u_{\mathrm{sw}}\right)=z\left(r_{\mathrm{s}/\mathrm{c}}\right)-z\left(R_{\odot }\right)$$(3) $$z\left(r\right)=\frac{a}{2}\left[\ln \left(\frac{r}{a}+\sqrt{1+\frac{r^2}{a^2}}\right)+\frac{r}{a}\sqrt{1+\frac{r^2}{a^2}}\right],$$(4)where z(r) is the distance along an Archimedean spiral from the centre of the Sun, r_s/c is the radial distance of the spacecraft from the Sun (here approximated as the position of ACE), R_⊙ is the solar radius, a = u_sw/Ω_⊙ and 2πΩ_⊙⁻¹ = 24.47 d is the equatorial period of solar rotation. For the solar-wind speed, we used data from ACE/SWEPAM⁹ (McComas et al. 1998), when available, substituted by data from Wind/SWE¹⁰ (Ogilvie et al. 1995) when SWEPAM data was unavailable. The release time thus obtained represents an estimate of the latest possible release of electrons from the Sun that is consistent with the determined onset time at 1 AU. For comparison, we also performed the same type of TSA for 55–80-MeV protons observed by ERNE.

The results of the TSA are presented in Table 5. The first column gives the event ID (as in Table 3). The second column gives the nominal length of the Parker-spiral field line, columns 3–5 give the onset times for ERNE 55–80 MeV protons (from column 3 of Table 3), columns 6–7 give those for EPAM (for five events, the 0.18–0.31 MeV electron channel was replaced by the 0.10–0.18 MeV channel) and columns 9–12 for EPHIN. Column 8 gives a qualitative description of the electron anisotropy as observed by EPAM.¹¹ This may give an indication of how much scattering is experienced by electrons during interplanetary transport, but the beaming of the angular distribution may also be reduced by local field fluctuations at time scales shorter than the time resolution of the particle flux data. One could nevertheless regard the events with beam-like anisotropy as those with the most reliable release time. Note that we have added 500 s, i.e., the light travel time per AU, to the release time of each channel to allow an easier comparison with EM observations at 1 AU.

3.3. Electromagnetic observations

SEPServer will also provide a wealth of EM observations on SEP events. In Table 6, we give the results of a preliminary analysis of SXR and radio observations for those 44 events that occur during the European day-time, i.e., for the events that SEPServer will provide the most extensive EM data coverage including results also from the ground-based observatories. In this paper, we concentrate on deka- and hectometric emission, obtained from Wind/WAVES, but further data from ground-based observatories at metric wavelengths is provided at our website in the form of summary plots of electromagnetic emissions.

The three first columns of the table give the event ID, the date and the solar release time (SRT) of protons, as deduced by the ERNE VDA. The fourth column shows the time of the impulsive flare phase. It is defined as the time between the start and the maximum of the SXR burst observed in the 0.1–0.8 nm channel of GOES. The start was computed as the time when the flux exceeded the pre-event background plus three times the noise level. The maximum is the time of the measured maximum of the burst, irrespective of considerations of the noise.

The characteristic times of type III emission were evaluated near 1 MHz. While the early emission is usually a well-defined type III burst, later phases are often mixed with less well-identified spectral features, such as spectrally complex type III bursts, or with type II emission. We inspected the plots of the dynamic spectra and then decided up to which time the emission was type III. The duration of the type III burst (Column 5) is the time interval during which the emission is at 3-sigma level above the background. For future reference, the number of bursts is evaluated from the flux density time profiles and checked by visual inspection of the dynamic spectra. Sometimes plateaus are observed, which look like an unresolved type III burst. These are generally counted as a burst, but the number is therefore to some extent subjective.

4. Results and discussion

4.1. Analysis of VDA results

The analysed 115 SEP events provide some statistics for the derived apparent path lengths. The apparent path length distribution seems to be double-peaked (Fig. 1). The first maximum is at 1.4 AU, close to the nominal Parker spiral length. The second peak is at longer apparent path lengths at 2.2 AU. The statistics, however, are not very high with 14 events in the first and 11 events in the second maximum 0.2-AU wide bin. Thus, as evidenced by the 1σ statistical error limits in the histogram, the statistical significance of the second maximum is not very large. There are also several events for which either very long (>5 AU, 8 cases) or very short (<1 AU, 8 cases) apparent path lengths were found. With the exception of three cases, the linear VDA fit to the data in these events was poor. In two of the exceptional cases (08.11.2000 and 15.01.2005) only seven data points were accepted for the fit.

Fig. 1. Apparent path length distribution of 107 SEP events. Four events with apparent path lengths above 6 AU are not shown in the figure. The numbers on the abscissa denote the lower limit of the path length bin.

The apparent path length and the spiral length of the field line are compared in Figure 2. There is almost no correlation between the two quantities. The VDA path length is typically longer than the length of the spiral field line, as expected, and it seems that other factors, such as pre-event background variability and/or non-simultaneous release at different energies, contribute to the apparent path length much more than the actual length of the field line, although the trend line in the figure has a reasonable slope, i.e., somewhat exceeding unity. However, the statistical significance for the fit is very low so this could be entirely fortuitous.

Fig. 2. The apparent path length obtained from VDA against the length of the Archimedean spiral field line. The error bars are obtained from the VDA.

Next, we investigated the relation between the proton release times obtained from TSA and VDA methods. Figure 3 gives the difference of the two release times as a function of the difference of the apparent VDA path length and the calculated spiral length, s-L. The slope of the trend line, 19.4 min AU⁻¹, is close to v⁻¹ = 23.1 min AU⁻¹ for 67.7 MeV protons, which indicates that the basic assumptions of the VDA analysis are not unreasonable. If one subtracts the trend line from the release time difference, the data points show a standard deviation of 38 min. Thus, there is quite significant scatter of the points around the trend line, which can be due to several contributing factors. First, the scatter reflects the inaccuracies in the determination of the onset time for some events as, for example, previous events can yield a high background masking the start of the event at some energy channels. Second, some of the scatter is probably also due to energy-dependent release times at the Sun.¹²

Fig. 3. The difference between the TSA (67.7 MeV, using the spiral length) and VDA (1.6–130 MeV) release times for protons as a function of the difference of VDA path length and the spiral length. Events with release time difference greater than 3 h have been dropped from the analysis.

If one compares the VDA release time with the 67.7-MeV proton release time as obtained using TSA, but replacing the spiral field-line length L in ^{Eq. (2)} with s obtained from the VDA, one gets an estimate of the scatter that is due to the variation of the onset time in a particular channel with respect to the trend line obtained in VDA. Figure 4 presents the distribution of the difference of the release times of 67.7-MeV protons obtained this way. The mean value and the standard deviation of the difference are 10 min and 17 min, respectively. Note that the value of the mean is close to the 13 min intercept obtained from the trendline in Figure 3. Thus, there is evidence that the release of 67.7-MeV protons occurs somewhat later than the mean release time of particles included in the VDA.

Fig. 4. The distribution of 67.7 MeV proton release time differences as obtained from TSA (using the VDA path length) and VDA, respectively. The value of the difference equals the difference between the actual 67.7 MeV onset time and the value obtained from the VDA fit at 67.7 MeV. Events with differences more than an hour have been dropped. The mean value and the standard deviation of the difference are 10 min and 17 min, respectively. The error bars denote the statistical error, only. The numbers on the abscissa denote the lower limit of the time difference bin.

One way of comparing the apparent path length to the spiral length of the field line is to analyse the statistics of their ratio. This is because we can write$$\mathrm{d}s= \nu \mathrm{d}t=\frac{\nu \mathrm{d}l}{\nu \mu }={\mu }^{-1}\mathrm{d}l,$$(5)where dl is the length increment of the field line connecting the source and the observer and µ is pitch-angle cosine of the particles, so the obtained value of the apparent path length could be interpreted as$$s=\overline{{\mu }^{-1}}l,$$(6)where l is the actual length of the field line connecting the source and the observer and the overbar denotes an average value obtained for the first-arriving particles during their propagation from the source near the Sun to 1 AU. If one takes l = L, the ratio L/s gives an estimate of the effective value of µ as . Figure 5 presents the distribution of µ_eff = L/s obtained from all the 111 events with a VDA result. Excluding the unphysical values (the integral bin), the mean value and standard deviation of µ_eff are 0.54 and 0.21, respectively, which seem reasonable in light of simulation studies (Lintunen & Vainio 2004; Sáiz et al. 2005).

Fig. 5. The distribution of the ratio of the length of the Archimedean spiral field line to the apparent path length obtained from VDA. The quantity represents an estimate of the effective value of µ of the first-arriving particles. The total number of events is 111. The error bars denote the statistical error, only. The numbers on the abscissa denote the lower limit of the µ bin and the last bin is integral.

4.2. Analysis of solar release times and flare onset

Next, we studied the estimated release times (from TSA) of relativistic electrons. The difference of the release times in two energy channels is plotted in Figure 6. Although the scatter is significant, electrons at higher energies have earlier TSA release times than electrons at lower enegies. There is an average difference of 4.6 min between the estimated release time of 0.7–3 MeV and 0.3–0.7 MeV electrons, in this case so that the higher energy electrons are released later. This could partly result from underestimation of the path length travelled by first-observed electrons, but since the delay of the lower energies is quite large, 9.0 min on average for events included in Figure 6, this explanation would require the true path length to exceed the spiral field-line lengths by more than 5 AU, which is unlikely. Moreover, instrumental effects cannot be ruled out as the two channels are measured by two different instruments.

Fig. 6. The distribution of the difference of the apparent release times of SOHO/EPHIN 0.3–0.7 MeV and ACE/EPAM 0.18–0.31 MeV electrons. Twelve events with an onset-time difference of more than an hour were omitted. The error bars denote the statistical error, only. The numbers on the abscissa denote the lower limit of the time difference bin.

Finally, we compared the particle release times with the start times of the related SXR flares as given by Cane et al. (2010). The delay of near-relativistic (NR; 0.18–0.3 MeV) electron emission with respect to X-ray emission is plotted as a function of the flare longitude ϕ_fl in Figure 7a. The shortest delay is obtained in the region of well-connected flares, around 60–80° solar longitude. Figure 7b presents the same points as a function of longitudinal distance of the solar flare from the footpoint of the Parker spiral leading to Earth, as obtained from the solar-wind speed observed during the SEP event onset:$$\Delta \varphi =\left|{\varphi }_{\mathrm{fl}}-\frac{{\mathrm{\Omega }}_{\odot }{r}_{\mathrm{S}/\mathrm{C}}}{u_{\mathrm{sw}}}\right|.$$(7)

Fig. 7. The apparent SRT of NR electrons, as observed by ACE/EPAM, relative to the flare onset as a function of (a) the longitude of the parent flare; and (b) the longitudinal distance of the flare from the footpoint of the Archimedean spiral field.

Both pictures show a clear delay of the onset of events with poor magnetic connection to the flare. The linear fit to the data points in Figure 7b gives a slope of 0.34 min/degree, which would correspond to an azimuthal speed at the solar surface of 590 km s⁻¹, which is of the same order of magnitude with typical lateral speeds of CME-related disturbances in the corona (Warmuth & Mann 2011; Cheng et al. 2012). This possibly indicates that the evolution of the CME related to the event has a role in the release of the accelerated particles on field lines connected to the observer. Note, however, that the quality of our SRT determination, as evidenced by the large scatter of the individual points in Figures 7a and 7b, is not enough to reach a definitive conclusion on this issue. We note also that the onset-time determination for widely separated events might be compromised by lower early intensities in these events.

Figure 8 gives the release times of protons as obtained from VDA as a function of apparent path length. This figure indicates that nearly all events with 1 < s ≲ 3 AU have reasonable proton release times that are independent of the path length. However, at path length values outside this range there is a tendency of the release time to shift towards earlier values for larger values of the path length. This would be expected if the true path length in these events actually had a value inside the expected range. Thus, probably the apparent path length values outside 1 < s ≲ 3 AU are artefacts of the analysis and the related release times should not be trusted. A similar result was obtained by Lintunen & Vainio (2004) analysing simulated data.

Fig. 8. The apparent SRT of protons (SOHO/ERNE VDA) with respect to the associated X-ray flare onset as a function of the apparent path length.

4.3. Associations with EM emissions

Next, we will turn to the timing of proton release with respect to the escape to space of energetic electrons at keV energies as traced by DH type III bursts. For a preliminary discussion we use the 44 events that occurred during the day time in Europe, listed in Table 6. We will compare the SRT of protons inferred from ERNE with the time of DH type III bursts observed by Wind/WAVES and the time of the impulsive flare phase inferred from GOES SXR observations. In Table 6 we only consider SXR and radio emission near the SRT inferred from the ERNE VDA. This explains why some of the identifications differ from those in the Cane et al. (2010) event list, which we used in Table 3.

The duration of the impulsive phase of the SXR emission (Column 4) is the interval between the time where the GOES time profile (0.1–0.8 nm wavalength) exceeded the background by 3σ and the time of maximum. Slight differences may exist with the start times listed in Table 3. Decametric-kilometric type III bursts were detected in all events, but sometimes they were embedded in type III storms, and an exact delimitation of the bursts related to the parent activity of the SEP event with respect to the storm bursts was difficult. Column 5 of Table 6 gives the time interval where the emission at 1 MHz was above the background + 3σ level. Event E75 (2002 July 16) occurred during a type III storm, but had no conspicuous DH type III burst near the SRT window. The brightest type III burst between the solar release time inferred from velocity dispersion analysis and the start of the event at ERNE occurred near 9 UT east of central meridian, as observed by the Nançay Radioheliograph. Because of its location it is not an obvious tracer of the release of the protons detected by ERNE. Event E89 was also classified as uncertain in Table 6, because no clear type III emission was observed at the high-frequency limit of the WAVES spectrograph. But prominent type III and type II emission was seen at lower frequencies during the release time interval of the protons inferred from the velocity dispersion analysis.

In the other 42 events we found the following timing of the DH type III bursts with respect to the SRT of 68 MeV protons:

• In nine events (21%) the SRT interval inferred from VDA preceded the DH type III emission (E3, E8, E17, E25, E47, E54, E78, E82, E106) by between a few minutes (E17) and more than 4 h (E106).

• In nine other (21%) events (E15, E24, E29, E37, E44, E51, E68, E71, E100) the SRT interval lagged behind the end of the DH type III emission by between a few minutes (E100) and nearly 3 h (E29).

• In 24 events (57%) the SRT interval overlapped with an interval of intense DH type III burst emission.

The majority of the events (to which event E89 has to be added) hence show a time coincidence between the inferred window of initial proton release and the DH type III bursts, which signal electron beams travelling through the high corona into interplanetary space.

The differences found in other events indicate either a failure of the VDA or physical delays between the releases of deka-MeV protons and keV electrons (cf. Malandraki et al. 2012, for the 2005 July 13 event). Comparing the path lengths derived from VDA and the type III timing in the three event categories derived from the type III timing, we find the following:

• Extremely long path lengths inferred from VDA are found in events E3 (s = 3.94 AU), E82 (s = 3.84 AU) and E106 (s = 16.34 AU). In these three events the SRT preceded the DH type III groups by particularly long durations, ranging from 34 min to nearly 5 h (E106). In the other events where the SRT was inferred to occur before the type III bursts the delays varied between 8 and 24 minutes.

• Events with unphysically short path lengths (s < 1 AU; E24, E29, E44) are in the group where the SRT inferred from VDA lags behind the end of the DH type III bursts by between several tens of minutes and several hours. The events with the longest lags, about 3 h (E29) and 2.2 h (E44), had both unphysically short path lengths.

• Events where the solar release time inferred from VDA overlaps with the DH type III bursts have apparent path lengths between 1.05 and 3.46 AU.

This corroborates our conclusion that extreme path lengths inferred from VDA, above 3-4 AU as well as below 1 AU, are due to a failure of the method. The initial solar release of deka-MeV protons appears to be closely related in time to the DH type III bursts.

Next, we compare the timing of the radio and X-ray emission. Out of a total of 44 SEP events listed, nine had no clear SXR burst, probably because they occurred behind the solar limb. One event (E106) had no clear DH type III burst at the time of the inferred solar release of protons and the associated SXR burst. So 34 events allow for a clear association of SXR emission with DH type III bursts. The relative timing of the DH type III bursts with respect to the SXR burst is the following:

• In 6 events (18%) DH type III bursts occur only during the impulsive phase, i.e., during the rise phase of the SXR profile (GOES, 0.1–0.8 nm). These events are labelled with an “asterisk” in Column 7 of Table 6.

• In 28 events (82%) the DH type III emission starts during the impulsive flare phase and extends beyond the peak of the SXR emission.

• In no case does the DH type III emission start after the SXR peak.

Thus, it seems that the electrons producing the type III burst have direct access to open field lines during the energy release of the flare. Note that these electrons are not necessarily the same as those observed in situ, so the result is not in contradiction with the delayed release of NR electrons of poorly connected events. However, an alternative interpretation would be that the delay is not real but due to the compromised timing of particle release in nominally poorly connected events, as discussed above.

5 Conclusions and outlook

We have performed a scan of 55–80 MeV proton fluxes observed by SOHO/ERNE in 1996–2010. A total of 115 SEP events were identified from this time period. We performed a VDA on all the proton events, using ERNE observations in 20 energy channels at 1.6–130 MeV, to get estimates of the SRT and path length in the interplanetary medium of the first-arriving protons. The results were compared to a TSA method utilizing the Archimedean spiral field-line length as the path length of first-arriving protons. In addition to protons, TSA was performed also to relativistic electrons observed by ACE/EPAM and SOHO/EPHIN.

A preliminary statistical analysis was performed on the VDA and TSA results, comparing them to each other and to the timing of the GOES X-ray flares related to the particle events, as identified by Cane et al. (2010). VDA was seen to produce reasonable results, in the sense that the two parameters, s and t₀ (relative to the flare onset), appeared independent of each other, provided that the apparent path length obtained from the analysis was in the range 1 < s ≲ 3 AU. Outside this range the parameters appeared to be anticorrelated. TSA compared to VDA produced the expected result that TSA release time (for 67.7 MeV protons) appeared later in most cases. This was true also if instead of the spiral length of the field line, the VDA path length was used to obtain the TSA release time.

Interpreting the higher-than-nominal values of the apparent path length of protons obtained from VDA as a signature of interplanetary scattering, we detemined the distribution for first-observed particles. The events showed a distribution that had a maximum in the 0.5–0.6 bin, with a mean value calculated over the sample as µ_eff = 0.54. This indicates that the scattering mean free path in the interplanetary medium is typically well below 1 AU, since even the first-arriving protons propagate with path lengths about 1.9 times their nominal values.

TSA for 0.18–0.31 MeV electrons was performed to study the release time of electrons with respect to the onset of the SXR flare. The time delay of the electron release with respect to the X-rays was studied as a function of flare longitude. While there was significant scatter among the events, the trend of the delay was consistent with the longitudinal expansion of the electron emission region at the speed of about 600 km s⁻¹. This seems consistent with the typical lateral expansion speed of the erupting structures in the corona, but the result may also be fortuitous, as the onset-time determination for widely separated events might be compromised. In fact, our result indicating the coincidence of initial proton release and DH Type III radio emission for events that conform to the expected range of VDA path lengths suggests that this might be the case.

The results of our preliminary analysis can be freely utilized in future studies. In addition to the tabulated information presented in this paper, our website (http://server.sepserver.eu) will offer access to summary plots of particle fluxes and EM observations related to the events. In addition to the catalogue of 1-AU observations made during the 23rd solar cycle, we will prepare catalogues of energetic particle events observed by the Helios and Ulysses missions and upgrade the 1-AU observations with particle events of the 24th solar cycle, including events observed by the twin-spacecraft STEREO mission (Kaiser et al. 2008).

Acknowledgments

The research leading to these results has received funding from the European Union’s Seventh Framework Programme (FP7/2007-2013) under Grant Agreement No. 262773 (SEPServer).

References

All Tables

All Figures

	Fig. 1. Apparent path length distribution of 107 SEP events. Four events with apparent path lengths above 6 AU are not shown in the figure. The numbers on the abscissa denote the lower limit of the path length bin.
In the text

	Fig. 2. The apparent path length obtained from VDA against the length of the Archimedean spiral field line. The error bars are obtained from the VDA.
In the text

	Fig. 3. The difference between the TSA (67.7 MeV, using the spiral length) and VDA (1.6–130 MeV) release times for protons as a function of the difference of VDA path length and the spiral length. Events with release time difference greater than 3 h have been dropped from the analysis.
In the text

Fig. 4. The distribution of 67.7 MeV proton release time differences as obtained from TSA (using the VDA path length) and VDA, respectively. The value of the difference equals the difference between the actual 67.7 MeV onset time and the value obtained from the VDA fit at 67.7 MeV. Events with differences more than an hour have been dropped. The mean value and the standard deviation of the difference are 10 min and 17 min, respectively. The error bars denote the statistical error, only. The numbers on the abscissa denote the lower limit of the time difference bin.

In the text

Fig. 5. The distribution of the ratio of the length of the Archimedean spiral field line to the apparent path length obtained from VDA. The quantity represents an estimate of the effective value of µ of the first-arriving particles. The total number of events is 111. The error bars denote the statistical error, only. The numbers on the abscissa denote the lower limit of the µ bin and the last bin is integral.

In the text

	Fig. 6. The distribution of the difference of the apparent release times of SOHO/EPHIN 0.3–0.7 MeV and ACE/EPAM 0.18–0.31 MeV electrons. Twelve events with an onset-time difference of more than an hour were omitted. The error bars denote the statistical error, only. The numbers on the abscissa denote the lower limit of the time difference bin.
In the text

	Fig. 7. The apparent SRT of NR electrons, as observed by ACE/EPAM, relative to the flare onset as a function of (a) the longitude of the parent flare; and (b) the longitudinal distance of the flare from the footpoint of the Archimedean spiral field.
In the text

	Fig. 8. The apparent SRT of protons (SOHO/ERNE VDA) with respect to the associated X-ray flare onset as a function of the apparent path length.
In the text