STATISTICAL STUDY OF CORONAL MASS EJECTIONS WITH AND WITHOUT DISTINCT LOW CORONAL SIGNATURES

Since coronal mass ejections (CMEs) were first identified in the early 1970's in coronagraph observations from OSO-7 (Tousey 1973) and Skylab (Gosling et al. 1974), they have attracted the attention of many scientists. The association between CMEs, flares, and filament eruptions is well-established (e.g., Munro et al. 1979; Webb & Hundhausen 1987). At the present time, it is generally accepted that CMEs, filament eruptions, and flares are all different manifestations of a single physical process that involves a disruption of the coronal magnetic field (e.g., Harrison 1996; Forbes 2000; Lin et al. 2003). As solar observations have been made over an increasingly broad wavelength range, CMEs are now regularly observed in association with a multitude of low coronal signatures (LCSs), such as filament eruptions (e.g., Schmieder et al. 2001; Alexander 2006), flares (e.g., Zhang et al. 2001; Mahrous et al. 2009), post-eruptive arcades (e.g., Sterling et al. 2000; Tripathi et al. 2004), coronal waves (e.g., Thompson et al. 1998; Biesecker et al. 2002), coronal dimmings (e.g., Sterling & Hudson 1997; Bewsher et al. 2008; Reinard & Biesecker 2009), and jets (e.g., Liu et al. 2005; Chen et al. 2009). These phenomena, observed below the field of view of white-light coronagraphs, are usually thought of as low coronal manifestations of the CME.

Before the launch of the STEREO spacecraft, it was generally accepted that a CME must be associated with at least one kind of LCS. As a result, if no LCS was present when a CME was observed by a coronagraph, often the CME was believed to have originated from the backside of the solar disk. That assumption seemed reasonable when only one direction of observation was available. In 2006, the STEREO (Kaiser et al. 2008) spacecraft were launched, and they provide us an opportunity to study CMEs along with their associated LCSs from two different lines of sight simultaneously. Robbrecht et al. (2009b) presented the first study of a front-side, large-scale CME without any obvious counterparts in the low corona. This finding challenges our previous understanding of CME initiation and the relationship between CMEs and LCSs, and may even impact our CME detection capabilities and influence the present theory of CME prediction. How dramatically this finding affects our understanding and CME prediction depends to a large extent on the rate of occurrence of this kind of CME without distinct LCS. In this paper, we present a statistical study by using data from the two viewpoints of STEREO/SECCHI (Howard et al. 2008).

The data used in this paper are listed in Section 2. We discuss how we search the database for front-side CMEs in Section 3. The criteria for LCSs and the results of checking for LCSs when front-side CMEs are observed are shown in Section 4. The differences between the CMEs with and without LCSs are studied in Section 5. We discuss possible initiation mechanisms for CMEs without distinct LCSs in Section 6, and we summarize our results and conclusions as well as considering the implications of this work in Section 7.

STEREO consists of two space-based observatories: one ahead (STEREO A) of Earth in its orbit, the other trailing behind (STEREO B; Kaiser et al. 2008). The two spacecraft slowly drift away from Earth in opposite directions, increasing their separation angle by 22° per year. The data used in this paper come from the EUV imager (EUVI) and the coronagraphs COR1 and COR2 on board STEREO. EUVI images the chromosphere and low corona up to 1.7 R_☉, COR1 images the inner corona from 1.3 to 4 R_☉ and COR2 images the outer corona from 2 to 15 R_☉ (Howard et al. 2008).

We use the CME catalog which is automatically generated by CACTus from COR2 observations (Robbrecht & Berghmans 2004; Robbrecht et al. 2009a). The output of CACTus is a list of events, similar to the classic catalogs, with principle angle, angular width, and velocity estimation for each CME. The URL http://secchi.nrl.navy.mil/cactus/ shows the catalog event list.

In order to study whether CMEs have associated LCSs, we first employ the COR data from the two different points of view of the STEREO spacecraft when they are located in positions that suit for identifying whether a CME is front-side or not. The first criterion for our database is therefore the separation angle between the spacecraft, and thus the time interval, which ranges from 2009 January 1 to August 31. Because the separation angle between the STEREO spacecraft increases with time (from 0° to 137° at the time of writing in 2010 March), not all of the data from the STEREO mission are appropriate for our study. We select STEREO data from 2009 January 1 to August 31, because during this period the spacecraft separation angle changed from 885 to 1125 and the lines of sight from STEREO A and B were approximately perpendicular to one another.

In theory, all of the CMEs observed as front-side CMEs by either A or B can be used to determine whether they are associated with LCSs. However, with only one vantage point it is not easy to definitely determine whether or not the CMEs that appear at the solar limb are front-side. Furthermore, the source regions of CMEs usually occupy volumes of space with an ambiguous border on the solar surface. Therefore, the identification of the source regions of CMEs becomes very complex. To guarantee that the source region of a CME definitely appears in the EUVI fields of view, we require that the CMEs are observed as front-side CMEs by both the STEREO spacecraft. In addition, front-side halo CMEs observed by one spacecraft and appearing as limb CMEs for the other spacecraft (during the period from 2009 January 1 to August 31) are also an ideal sample. These two situations are used as the second criterion for inclusion in database. In this paper, we use the phrase "Earth-sided CMEs" to describe the CMEs which fit either of the two situations.

A CME fits our criteria if it is observed to the east in STEREO COR A data and to the west in STEREO COR B data or as a front-side halo event for either A or B. As illustrated in our schematic (Figure 1), we divide the Sun into four sections (1, 2, 3, 4) along the line between the Sun and STEREO A, and along the line between the Sun and STEREO B. According to this division, the Earth-sided CMEs must come from the region which includes the right part of section 2, section 3, and the left part of section 4 (marked by the two ends of the red arrow in the bottom middle panel in Figure 1). The borders of the right part of section 2 and the left part of section 4, determined by the border of halo CMEs, are relatively flexible.

Zoom In Zoom Out Reset image size

To select the Earth-sided CMEs, we imposed two separated criteria. In one criterion, we selected the Earth-sided CMEs by following the CACTus CME list for STEREO A data (right part of Figure 1) and in the other we followed the CACTus CME list based on STEREO B data (left part of Figure 1). Finally, we merged the results from the two data sets. The entire selection process can be described by the following five detailed steps.

• 1.  
  Step 1. To identify CMEs that occurred during this period, we first consulted the CACTus CME catalog compiled from COR2 observations. According to CACTus, between 2009 January 1 and August 31, 89 CMEs were observed by STEREO A and 115 CMEs by STEREO B.
• 2.  
  Step 2. We visually checked all of these CME detections and removed false detections that are generated due to the detection of planets as CMEs or multiple detections of one CME. We also added several weak CMEs, which were not detected by CACTus using either STEREO A or B data. After correcting for these, we were left with 72 CMEs in STEREO A data and 82 CMEs in STEREO B data. Of these CMEs, 43 were observed by both spacecraft. In order to avoid multiple detections of the same CME, we removed the redundant listings of these CMEs and unambiguously identified 111 individual CMEs. According to this study, CACTus is reliable at a level of (72/89) = 81% in STEREO A and to (82/115) = 71% in STEREO B.
• 3.  
  Step 3. We considered the CMEs that are observed to the east of the Sun in COR2 A data (section 2 or 3 in Figure 1) and to the west of the Sun in COR2 B data (section 3 or 4 in Figure 1). Among the 72 CMEs observed by STEREO A, 39 of them appeared to the east in STEREO A (section 2 or 3 in Figure 1); 35 of the 82 CMEs observed by STEREO B appeared to the west of the Sun (section 3 or 4 in Figure 1).
• 4.  
  Step 4. We cross-correlated the data from STEREO A and B for each CME, to constrain the true location of the CME and determine whether it fits our requirements or not. Of the 39 CMEs observed to the east in STEREO A, 24 of them are also observed to the west (section 3 in Figure 1) or seen as a front-side halo CME (indicated by the brown triangle area, part of section 2 in Figure 1) in STEREO B. Of the 35 CMEs observed to the west in STEREO B, 24 are also observed to the east (section 3 in Figure 1) or as a front-side halo CMEs (indicated by the brown triangle area, part of section 4 in Figure 1) in STEREO A.
• 5.  
  Step 5. After checking, we found that 14 CMEs were selected in both lists. Finally, we confidently identify

  $$\begin{equation} 24+24-14=34 \end{equation} \tag{ 1 }$$

  individual Earth-sided CMEs. (Readers are referred to Figure 1 for a schematic flow diagram explaining this process.)

Figure 2 shows an example of an "Earth-sided CME," which is seen as a partial halo CME coming from the northern part of the solar disk by COR2 B and an east limb CME by COR2 A (i.e., fitting the second situation). Considering the positional relationship of COR2 A and B, we can tell that this CME is a front-side halo CME for STEREO B. The EUVI data show that the source region of this CME is located at the northern solar limb in the EUVI A field of view and the northern part of solar disk in the EUVI B field of view. Therefore, this CME can be used for checking whether it is associated with LCS or not.

Zoom In Zoom Out Reset image size

For each of the 34 Earth-directed CMEs, we studied the corresponding Extreme Ultraviolet Imaging (EUVI; Wuelser et al. 2004) data to check for distinct LCSs by visual inspection. The "distinct LCSs" in this paper refer to the following signatures observed in the 195 Å, 171 Å, or 304 Å EUVI passband.

• 1.  
  Filament eruptions.
• 2.  
  Sigmoid eruptions.
• 3.  
  Flares.
• 4.  
  EIT waves.
• 5.  
  Coronal dimmings.
• 6.  
  Post-eruptive arcades or brightening.

In most cases, we used the plain images in 304 Å to check for filament eruptions, the running difference images of 171 and 195 Å to check for EIT waves, the base difference images to check for coronal dimmings, and the plain images in 171 and 195 Å to check sigmoid eruptions and post-eruptive arcades or brightening. The time cadence of the EUVI images in 171 Å, 195 Å, or 304 Å are 2.5 (or 10) minutes, 10 (or 2.5) minutes, or 10 minutes, respectively.

Figure 3 shows several examples of distinct LCSs associated with CMEs. A filament eruption associated with a CME on 2009 January 14 was observed at the solar limb in 304 Å images by STEREO A (panel (a) of Figure 3). In the field of view of STEREO B, the origin of this CME comes from the north part of the solar disk. A distinct brightening appeared in 195 Å (see the inset images in panel (a) of Figure 3: the top and bottom ones are the images before and during the CME). The panel (b) of Figure 3 shows a sigmoid eruption associated with the CME that occurred on 2009 January 17. The sigmoid eruption is very clear in EUVI A (inset in panel (b) of Figure 3), more so than in the EUVI B field of view, because it is a bit backsided as seen from EUVI B. Panel (c) of Figure 3 shows a violent solar eruption that occurred on 2009 February 13, which has been well studied by Patsourakos & Vourlidas (2009) and Cohen et al. (2009). During the eruption of this CME, a sigmoid eruption, flare, dimming, and EIT wave were observed by both STEREO A and B. Panel (d) of Figure 3 shows a jet-like filament eruption associated with a narrow CME that occurred on 2009 June 23.

Zoom In Zoom Out Reset image size

We found that of the 34 Earth-sided CMEs, 11 had no distinct disk LCSs (see Table 1), which are also called stealth CMEs. A CME is stealth if no distinct LCS (such as coronal dimming, coronal wave, filament eruption, flare, post-eruptive arcade) can be found on the disk. For the other 23 events (see Table 2), we were able to identify an associated LCS. Nearly half of the CMEs without LCSs seem to belong to the blowout type.

We found that about one of the three Earth-sided CMEs between 2009 January 1 and August 31 had no distinct LCSs. We note that this time span occupied a period of exceptionally deep solar minimum, and this surprisingly high percentage of CMEs without distinct LCSs may be atypical. Nonetheless, we pose the question: Are CMEs without distinct LCSs special? Do they perhaps form a different class of CMEs?

To explore whether there is any difference between the CMEs with and without LCSs, we examined the characteristics of the 34 Earth-sided CMEs. Our analysis is based on data from EUVI and white-light data from COR1 and COR2.

We first checked the velocity and angular width distributions for the 34 Earth-sided CMEs according to the CACTus record. Figure 4 displays the distribution of the number of CMEs versus the velocities and angular widths. These data were obtained from the corrected CACTus record shown in columns "v" and "da" of the "limb" column in Tables 1 and 2. The red lines indicate CMEs associated with LCSs and blue lines those without LCSs. We see that the velocities of the CMEs without distinct LCSs range between 100 km s⁻¹ and 300 km s⁻¹, generally slower than the velocities of the CMEs with LCSs, which range between 150 km s⁻¹ and 750 km s⁻¹. The CME angular width distribution shows that the angular widths of the CMEs without distinct LCSs are smaller than 40°, while the angular widths for the CMEs with LCSs range between 10° and 70°. However, some CMEs with LCSs are also very narrow. Generally speaking, there is no clear difference in the CME angular width distributions between the CMEs with and without LCSs.

Zoom In Zoom Out Reset image size

To track the CME in COR2, we employ height–time maps, as Sheeley et al. (1999, 2007) did in their work. To obtain the height–time map, we select a rectangular slit along the radial direction of the CME seen in COR2 running difference images (see the top panels of Figure 5 or 6 for examples). Then we extract the data in the rectangular slit for each image where the CME expands through the COR2 field of view. Then we stack the extracted data at different times (see the middle panels in Figure 5 or 6). The determination of CME velocity and acceleration is of considerable importance for comprehending the CME initiation process and the subsequent dynamic expansion. Usually, the leading edge of the CME is used for making such measurements (being a relatively well-defined feature). However, not all CMEs have a well-defined leading edge. In this study, most of the Earth-sided CMEs are relatively slow (see the velocity values in the "v" column of Tables 1 and 2) and their leading edges are too faint and diffuse to be an ideal tracking feature, as was also the case in Robbrecht et al. (2009b). However, the cores of these CMEs, in most cases, can be clearly identified and can be tracked easily (e.g., Srivastava et al. 1999, 2000; Maričić et al. 2004, 2009). Maričić et al. (2009) carried out a statistical analysis of the relationship between the kinematics of the leading edge and the eruptive prominence core of CMEs. They concluded that in the majority of events (78%) the acceleration phase onset of the leading edge is very closely synchronized (within ±20 min) with the acceleration of the eruptive prominence. Equating the eruptive prominence with the core feature seen in white-light data, we assume that tracking the CME core is a reasonable proxy for determining the dynamic characteristics of the CME.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

We manually identify the core of the CME by visual inspection. The diamonds in the middle panels of Figures 5 and 6 show examples of tracking a CME core position, as well as the uncertainties in the selection of the core position. Due to the fuzzy nature of the extended brightenings, the uncertainty in selecting the highest and lowest extents of the brightening is larger than when selecting the bright core. The solid curve is the result of fitting a fourth-order polynomial,

$$\begin{equation} r(t)=t^4+at^3+bt^2+ct+d, \end{equation} \tag{ 2 }$$

to the data. We investigated other fittings (including exponential fittings) and found that the fourth-order polynomial fit had the smallest chi-square and fit our measurements best.

Only 13 CMEs (marked with asterisks in Table 1) without LCSs and 10 CMEs (marked with asterisks in Table 2) with LCSs are suitable for measurement because they have bright cores that can be tracked in the distance–time plots. We note that according to the corrected CACTus list the 13 CMEs with LCSs that have been chosen here are generally slow CMEs (see the "v" column in Table 2), while the other 10 CMEs with LCSs, whose cores are not easy to track, are generally faster. The asterisks in Tables 1 and 2 indicate the reliability of CME velocities calculated by tracking the CME cores. The events with two asterisks indicate that the CME velocities measured by us are considered to be quite reliable and those with one asterisk are considered to be reliable but less than those with two asterisks. For the 11 events without any asterisk in either A or B (Tables 1 and 2), the derived velocities are not considered to be reliable. These are subjective rankings, based on the clarity of the CME features used to determine CME heights as a function of time. For our study of the CME kinematics, we only use the events with asterisks.

The CME kinematics analysis for the 23 CMEs in this sample is shown in Figure 7. The top, middle, and bottom panels in Figure 7 show distance–time, velocity–distance, and acceleration–distance profiles, respectively. The panels in the left column are profiles for CMEs with distinct LCSs, and the right column panels are for CMEs without any obvious LCSs. The result shows that the velocities of stealth CMEs are below 300 km s⁻¹, more than half of them slower than 200 km s⁻¹. The accelerations of the stealth CMEs are smaller than 10 m s⁻², which smoothly change with the distance that they have traveled away from the Sun and which reach their peaks at around 4 R_☉. The CMEs with distinct LCSs usually show a faster speed (see also the velocity estimated by CACTus in Tables 1 and 2) and a larger acceleration compared with the stealth CMEs. From Figure 7, we can conclude that the stealth CMEs are slow CMEs (v < 300 km s⁻¹) that are gradually accelerated, which is consistent with the result of Robbrecht et al. (2009b). Conversely, CMEs that are faster than 300 km s⁻¹ always showed distinct LCSs. However, there also exist slow CMEs that do have discernible LCSs. So from the kinematics point of view the two CME populations are not clearly separated.

Zoom In Zoom Out Reset image size

Figure 8 shows histograms of the velocity and acceleration distributions for the 24 front-side CMEs for which the properties mentioned above were determined. The upper panels show that the distributions of CMEs with LCSs are all shifted to the right as compared to the distributions of stealth CMEs. In general, the stealth CMEs start off slowly and do not reach speeds higher than 300 km s⁻¹ (as noted earlier). CMEs with LCSs tend to start off faster as can be seen in the middle upper panel. The lower panels do show a difference between the two populations: stealth CMEs are only moderately accelerated or decelerated; the values of (a) are in the range of [−7, 7] km s⁻¹.

Zoom In Zoom Out Reset image size

The distribution of maximum velocity for the CMEs without any distinct LCSs is similar to that of those with LCSs (see panel (a) of Figure 8). A difference occurs only below 200 km s⁻¹, where only CMEs without LCSs are observed. However, the sample size is not large enough to establish whether or not this difference is significant. The distribution of CME speeds at first appearance in COR2 data (see panel (b) of Figure 8) shows that CMEs without LCSs tend to be distributed between 50 km s⁻¹ and 100 km s⁻¹. There is no clear difference in acceleration between the CMEs with and without LCSs (see panels (d–f) of Figure 8).

Most CMEs in our data set did not follow a purely radial direction in the COR1 field of view and are deflected by the surrounding streamers toward the equator, as described by, e.g., Delannée et al. (2000). This prevents us from applying to most events the method that we used for COR2 data (described above), to obtain the dynamic characteristics of CMEs during their propagation through the COR1 field of view. Instead, we measured the time interval during the passage of the CME through the COR1 field of view from about 1.3 R_☉ to 2 R_☉ by measuring the time difference between the moment the CME (seen as a rising or expanding coronal structure) first appeared in the COR1 (see "Time 1" in Tables 3 and 4) and COR2 (see "Time 2" in Tables 3 and 4) fields of view, respectively. Using this time interval, we estimated the average velocities of the 34 CMEs as they traveled from 1.3 R_☉ to 2 R_☉, through the COR1 field of view. The time interval uncertainty is about 19 minutes (5 minutes for COR1 time cadence and 14 minutes for COR2 time cadence), which we used as the time uncertainty to calculate average velocities. The results are listed in the "v" column in Tables 3 and 4. The velocity uncertainty can be described as

$$\begin{eqnarray} V_{\rm uncertainty} &=& [0.7 \;R_\odot /({\rm time\;interval}-19)]\nonumber\\ &&-[0.7\; R_\odot /(\rm time\;interval+19)], \end{eqnarray} \tag{ 3 }$$

which is very sensitive to the time interval (time interval = Time2−Time1 in minutes).

Table 3 lists the observation spacecraft, date, "Time 1," "Time 2," average velocity, and the specific LCS associated with the corresponding CME. Table 4 shows similar information for the CMEs without LCSs. The extra columns "HL" and "HH" represent the estimated low and high positions of the coronal structures associated with some of the stealth CMEs (for more detail, see Section 6).

Figure 9 shows the CME average velocities derived in COR1 field of view, blue for CMEs without LCSs, and red for those with LCSs. The results indicate that for the CMEs without LCSs, the velocities are below 20 km s⁻¹ and more than half slower than 10 km s⁻¹. For the CME with distinct LCS, their average velocities range between 5 and 140 km s⁻¹ and more than half of these are larger than 20 km s⁻¹. Generally speaking, the average velocities for the CMEs with LCSs are larger than those without LCSs.

Zoom In Zoom Out Reset image size

This work has shown that a significant proportion of CMEs (at least during 2009 January 1–August 31) are initiated without any distinct disk signature. While the obvious signatures, such as flares, filament eruptions, and coronal dimmings, are not seen in the events, detailed inspection of EUVI data from STEREO indicates that some faint changes are likely associated with the CMEs without LCSs.

EUVI limb observations show that some coronal structures, likely to be part of flux ropes, high in the corona were changing before or during their associated CME processes. As shown in Figure 10, we found certain coronal structures (indicated by the white circle) high in the corona in 8 of the 11 CMEs without LCSs. Figure 10 shows composite images from EUVI and COR1 data; the EUVI images were contrast enhanced using a radial filter technique developed by S. Cranmer (2010, private communication). We roughly estimated the lower and upper heights of these structures observed in EUVI data and the results are shown in "HL" and "HH" columns in Table 4, respectively. While the results still suffer from projection effects, it is clear that the structures are located high in the corona, and the lowest positions of the coronal structures are usually more than 0.2 R_☉ above the solar surface. The upper positions of the coronal structures may even reach 1 R_☉ above the solar surface in the COR1 field of view (panel (e) of Figure 10). The EUVI on-disk observations for these events do not show similar coronal structures. These CMEs without LCSs may be initiated by the disturbance of flux ropes which are suspended high in the corona (see also Robbrecht et al. 2009b; Reinard & Biesecker 2009).

Zoom In Zoom Out Reset image size

For the remaining three events, no distinct disturbance high in the corona could be identified in EUVI observations, but many tiny eruptions in the low corona are observed. Figure 11 is an example of this situation. The top left panel is a composite image from STEREO A (COR2, COR1, and EUVI), which shows the entire coronal structure from the solar disk to the high corona that existed before the CME occurred. The composite image in the bottom left panel is composed of running difference images from COR2, COR1, and EUVI. The morphology of the CME and steamer expansion at the CME's flank (COR1 field of view) are clearly identified. The panels in the middle and right columns are EUVI running difference images at 195 Å from A (right column) and B (middle column). No clear disturbance high in the corona was observed before the associated CME occurred on 2009 January 7 (readers are also referred to the animations on http://stereo-ssc.nascom.nasa.gov/browse/2009/01/07/). Many tiny eruptions in the low corona indicated by magenta arrows in Figure 11 were observed before and during the CME initiation. Additional eruptions to those indicated by arrows can be identified in the animation associated with Figure 11 (available in the online journal). However, the potential relationship between the tiny eruptions observed in EUVI as tiny brightenings and dimmings and larger-scale eruptions that develop as CMEs is not clear yet. Further detailed study of such small-scale eruptive activity is required to clarify the extent and nature of the potential relationship to CMEs.

Zoom In Zoom Out Reset image size

In this paper, we present a statistical study of CMEs with and without distinct LCSs between 2009 January 1 and August 31. According to the SECCHI CME list compiled by CACTus, a total of 111 CMEs were observed and nearly all of them are slow CMEs. We exploited the ∼90° separation of the STEREO spacecraft and identified 34 front-side CMEs. Due to the upper limit of the separation angle of STEREO A and B used in our study, the sample in our work is not large, but nonetheless the results make a significant contribution toward understanding the initiation of CMEs. In particular, our work has implications for automatic CME detection methods and other manual approaches that do not employ coronagraph data.

• 1.  
  Approximately 1/3 of the Earth-sided CMEs studied from 2009 January 1 to August 31 are stealth. However, we note that this statistical result is obtained during a period of deep solar minimum, and the percentage of CMEs without LCSs may be different if we consider all of the CMEs over an entire solar cycle.
• 2.  
  The study of the CME kinematics in the COR2 field of view shows that the stealth CMEs are slow CMEs (v < 300 km s⁻¹) that are gradually accelerated. Conversely, CMEs that are faster than 300 km s⁻¹ always showed distinct LCSs. However, there also exist slow CMEs that do have discernible LCSs. So from a kinematics point of view, the two CME populations are not clearly separated. The average velocities of the 34 CMEs when passing through the COR1 (1.3 R_☉–2 R_☉) field of view show that the expansion velocities of the CMEs without LCSs are slower than those with LCSs.
• 3.  
  Although there may be no distinct disk signatures for the stealth CMEs, in most of the cases some coronal variation could be observed off-limb. We found that more than half of the stealth CMEs in this paper showed some faint change of the coronal structures (likely parts of flux ropes) when they could be observed over the solar limb. These features are usually very difficult to identify in the on-disk observations. In some of the other events a multitude of tiny eruptions can be observed at the same time as the CME process.
• 4.  
  Our results confirm that a significant proportion of front-side CMEs are not always associated with distinct LCSs. Therefore, detection of LCSs is not sufficient for a confident detection of all CMEs at the present time. Space weather detection systems based on flare detection, coronal dimmings, filament eruptions, and so on (independent of coronagraph data) may fail to detect a significant proportion of CMEs.

We sincerely thank the anonymous referee for very helpful and constructive comments that improved this paper. The STEREO/SECCHI data are produced by an international consortium: NRL, LMSAL, NASA, GSFC (USA); RAL (UK); MPS (Germany); CSL (Belgium); and IOTA, IAS (France). We thank Huw Morgan for his work with the COR1 data. We also thanks Jason Byrne, Alisdair Davey, Meredith Wills-Davey, Edward DeLuca, Petrus Martens, Kathy Reeves, and Yingna Su for useful discussions. S.M. thanks Henry Winter, Paolo Grigis, and Alexander Panasyuk for their help with software. This work was supported by NASA grants SP02H1701R and NNM07AB07C. S.M. and J.L. were also supported by Program 973 grant 2006CB806303, by NSFC grants 10873030 and 40636031, and by CAS grant KJCX2-YW-T04 to YNAO. G.D.R.A. gratefully acknowledges NASA grant NNX09AB11G. Part of S.M.'s work was supported by SAO-610089-4210-40610089HH0022. Part of J.L.'s work was performed when he visited CfA, supported with the Smithsonian Institution Restricted Endowment Funds.