Stronger Southward Magnetic Field and Geoeffectiveness of ICMEs Containing Prominence Materials Measured from 1998 to 2011

Interplanetary coronal mass ejections (ICMEs) could be classified into magnetic clouds (MCs) and non-MCs according to their magnetic field signatures, and into prominence-inside ICMEs (PIs) and non-PIs based on whether they contain colder and higher helium abundance plasmas than the solar wind. It is known that the MCs often lead to magnetic storms. However, whether or not the PIs have significant geoeffectiveness is unclear. This statistical work studies the southward interplanetary magnetic field (IMF) magnitude of the PIs, and the related magnetic storms’ level. The data include the IMF and plasma moments measured by ACE and WIND, and the Dst index from 1998 to 2011. The hypothesis test based on the proportions of two groups is used to analyze 95 ICMEs related to single storms (SSs). The results show that the magnetic storms caused by the PIs mostly distribute at a strong level, while that caused by the non-PIs and by all the 95 ICMEs mostly distribute at a moderate level. The PIs have a significantly higher probability of generating SSs than the non-PIs. Moreover, the MCs containing carbon-cold and helium-enhanced materials (MC&PIs) have the highest fraction of minimum Bz, less than −11 nT. Since the MC&PIs have large-scale magnetic flux rope and prominence material, the stronger southward IMF is probably provided by the prominence. It is in accordance with the observed injection of enhanced twisted flux ropes to prominence. Therefore, the detailed eruption and propagation processes of the three-part coronal mass ejections deserve more concern from a space weather perspective.


Introduction
Coronal mass ejections (CMEs) are observable changes in coronal structure that occur within a few minutes to hours. They were first discovered in 1971 by the coronagraph on board the seventh Orbiting Solar Observatory(OSO-7; Manchester et al. 2017). A CME can be identified by a new brightness enhancement(in white light) moving outward in a coronagraph image (Hundhausen et al. 1984;Yashiro et al. 2004). The typical geometry of CMEs is a three-part structure that contains a bright loop, a dark cavity, and a bright core. The three parts are usually interpreted as the plasma pile-up, the flux rope, and the eruptive prominence respectively (Illing & Hundhausen 1985;Hudson et al. 2006;Webb & Howard 2012). Several statistical studies reported that from 30% to 60% of the observed CMEs have three-part structures (Gopalswamy 2006;McCauley et al. 2015) And this three-part structure is the basic model in studying the initiation of the CMEs (Forbes 2000;Lin & Forbes 2000;Forbes et al. 2006;Manchester et al. 2017;Song et al. 2019).
When the CMEs propagate into the interplanetary space, their counterparts measured in situ are called interplanetary coronal mass ejections(ICMEs). ICMEs could be identified by stronger magnetic field, lower temperature of protons and electrons, enhanced helium abundances relative to protons, bidirectional electrons, and occasional enhancement in minor ions of lower charge (Cane & Richardson 2003;Richardson & Cane 2004). However, these signatures do not necessarily occur simultaneously (Wimmer-Schweingruber et al. 2006;Zurbuchen & Richardson 2006;Richardson & Cane 2010). In previous studies, the ICMEs can be classified by the magnetic field structure. If the ICMEs have enhanced and smoothly rotating magnetic field during a long time, they are called magnetic clouds(MCs; Burlaga et al. 1981).
The materials inside a few ICMEs showing lower proton temperature, higher proton density, higher helium abundance, and lower charge state ions than the solar wind are considered to be the prominences(bright cores) of the CMEs (Hirshberg et al. 1972;Schwenn et al. 1980;Priest 1989;Lepri & Zurbuchen 2010;Yao et al. 2010;Song et al. 2016;Wang et al. 2018a). Feng et al. (2018) presented a new catalog based on the average charge state of carbon ions. If the ICMEs containing carbon ions with mean charge lower than the solar wind mean value minus three times standard deviation and the ionization temperature of carbon ions is lower than 10 6.05 K, they are called carbon-cold ICMEs(CCs). The CCs take about onethird of the studied ICMEs. Therefore, the ICMEs could also be classified by ion composition.
The CMEs propagating along the earthward direction are halo CMEs. Interplanetary counterparts of the halo CMEs are the most important phenomena that drive geomagnetic storms (Gopalswamy et al. 2007;Kilpua et al. 2017b). Geomagnetic storms feature a strong disturbance on the horizontal component of global geomagnetic field lasting for tens of hours to days. The Dst index represents the horizontal magnetic field variation caused by an enhanced ring current during the storm (Sugiura 1964). And its minimum value is used to estimate the storm intensity. A geomagnetic storm could be typically divided into three phases including the sharp increase of the Dst index as the initial phase, the decrease of the Dst as the main phase, and then the recovery phase (Gonzalez et al. 1989(Gonzalez et al. , 1994Yokoyama & Kamide 1997;Maltsev 2004;Gopalswamy 2006;Nikolaeva et al. 2012;Monreal MacMahon & LLop-Romero 2014). Statistical studies reported that the Dst minimum is related to the magnitude of the southward (B z <0) interplanetary magnetic field(IMF; Burton et al. 1975;Akasofu 1981;Yermolaev et al. 2010;Adekoya & Chukwuma 2018).
Since the MCs have a large-scale magnetic flux rope, many researchers studied the geoeffectiveness of the MCs and the non-MCs. Statistical results showed that the MCs are the most geoeffective subset compared to the non-MCs, corotational interaction region(CIR), fast solar wind, and shock (Kilpua et al. 2017a(Kilpua et al. , 2017b. Furthermore, the intensity of the geomagnetic storm caused by the MCs is also stronger than that of the other types mentioned above (Echer et al. 2005;Zhang et al. 2007;Nikolaeva et al. 2011).
However, the geoeffectiveness of the PIs and the non-PIs has not been studied so far. To examine if the PIs are a real physical subset of ICMEs or an artificial one by ion composition criterion, it is necessary to check their effect on the geomagnetic field and their forming mechanism on the solar atmosphere. Since the PIs represent the CMEs with the prominence eruption, it is also interesting to investigate if the details of the eruption process of the three-part CMEs deserves more attention from the space weather perspective.
In this work, we perform a statistical analysis on the southward IMF of the 95 single ICMEs and on the intensity of the storms caused by them measured from 1998 to 2011. The data and the criteria used to identify the storms are described in Section 2. The statistical results of the storm intensity and IMF B z are shown in Section 3. The possible mechanisms are discussed in Section 4. Finally, the conclusion is given in Section 5.

Data
In this work, we use the IMF measurements, solar wind proton (N p ), α particle density (N α ), and the geomagnetic indices. The B z component of the ICMEs in the Geocentric Solar Ecliptic (GSE) coordinate system, and the number density of solar wind protons and α particles are available on CDAWEB (https://cdaweb.gsfc.nasa.gov/index.html). The IMF B z component is level 2 (verified) data measured by the magnetic field experiment (MAG; Smith et al. 1998) on board the ACE spacecraft having a temporal cadence of 1 hr. The N α /N p is measured by the solar wind electron proton alpha monitor (SWEPA; McComas et al. 1998) on board the ACE spacecraft, having a temporal cadence of 64 s. Since there are data gaps caused by measurement failure on ACE data, and the WIND spacecraft is very close to ACE, the WIND data is also used. The 92 s-cadence N α (n/cc) and N p (n/cc) from nonlinear analysis are measured by the solar wind experiment (SWE; Ogilvie et al. 1995) on board the WIND spacecraft. The Dst index is derived from hourly averaged horizontal magnetic field variation at low latitude (Sugiura 1964). It is provided by WDCG (World Data Center for Geomagnetism) in Kyoto (http://wdc.kugi.kyoto-u.ac.jp/dstdir/index.html).

ICME List and Event Selection
First, the studied ICMEs are from the 219 ICMEs in Feng et al. (2018) including 69 CCs and 99 MCs measured from 1998 February to 2011 August. Second, we check the geomagnetic variation caused by these ICMEs according to their Dst index. Considering the traveling time of the ICMEs from ACE/WIND spacecraft to the Earth, there is about a 2 hr delay in the Dst response (Zhang et al. 2007;Yermolaev et al. 2010).
Based on previous studies, we quantify the criteria of different types of geomagnetic disturbances caused by the ICMEs. Since not all the storms have the initial phase, we focus on the main phase and the recovery phase. The start of the main phase is fixed by the maximum Dst value between the ICME shock and the minimum Dst. The minimum Dst called Dst min marks the end of the main phase and the start of the recovery phase. We define the end of the recovery phase as the Dst approaching 10% of the Dst min for Dst min > −60 nT or approaching 30% of the Dst min for Dst min −60 nT. At the end of the recovery phase, we require that Dst>−30 nT. This means that the negative Dst must have an absolute value less than 30 nT, otherwise the Dst should be positive (Yokoyama & Kamide 1997). To guarantee the unique correspondence between the ICMEs and the magnetic storms, we divide the geomagnetic variations caused by ICMEs into four types: single storms(SSs), multiple storms(MSs), storm-like dis-turbances(SDs), and non-storms(NSs), shown in Figure 1.
The geomagnetic disturbances showing Dst min > −30 nT are identified as NSs. The geomagnetic disturbances having the main phase and recovery phase, when there is only one ICME, are classified as SSs. It should be noted that there are some additional constraints to select SSs: (1) The onset of the main phase should be at a relatively quiet level (Dst > −30 nT). Meanwhile, the Dst decrease amplitude in the main phase should be greater than 30 nT. (2) The Dst min should occur between the shock time and 3 hr after the end time of ICMEs.
(3) There must be no other ICMEs during the recovery phase.
The long-lived MSs are caused by two or more continuous ICMEs. They have the main phase and the recovery phase with Dst min −30 nT (Xie et al. 2006). The geomagnetic disturbances that do not meet any of the above criteria for SSs, MSs, and NSs are classified as SDs. As a result, 95 single storms are identified. The 95 single ICMEs out of the 219 ICMEs are related to the SSs. They are selected to carry on the statistical analysis, shown in Figure 2. In this study, we analyze the levels of single storms caused by different types of ICMEs. We use the Dst min to classify the intensity of storms into five levels based on Loewe & Prölss (1997): weak (−50 nT < Dst min −30 nT), moderate(−100 nT < Dst min −50 nT), strong(−200 nT < Dst min −100 nT), severe(−350 nT < Dst min −200 nT), and great(Dst min −350).
Additionally, we divide all the 219 ICMEs into different types according to their magnetic field structure, carbon ionic charge state, and the helium abundance relative to protons. We adopt the criteria from Burlaga et al. (1981) and Feng et al. (2018) to identify MCs. Considering the prominence is colder and the presence of helium enhanced chromospheric material, we focus on the two signatures that could be identified as prominence-inside ICMEs(PIs): the carbon cold and the helium enhanced plasmas. The ICMEs containing carbon cold materials(CCs) are obtained from the list in Feng et al. (2018). According to Zurbuchen et al. (2016) and Wimmer- Schweingruber et al. (2006), the plasmas with N α /N p > 0.08 are possible prominence materials. Therefore, we use this criterion to identify the helium enhanced(HEs) ICMEs. In addition, since the measured N α has large uncertainty, we require the N α /N p >0.08 lasting for 2 hr inside the ICMEs. In all the 219 studied ICMEs from 1998 to 2011, there are 99 MCs, 69 CCs, and 69 HEs shown in Figure 2. In the 95 ICMEs causing single storms, there are 55 MCs, 39 CCs, and 37 HEs. 17/95 ICMEs are simultaneously CCs and HEs, and 12/95 ICMEs are simultaneously MCs, CCs, and HEs, listed in Table 1. In addition, the detailed information of the 95 ICMEs and the related single storms are provided in Table 2 in the Appendix.

Method
We use three statistical methods to quantify the significant differences between different types of ICMEs. The original information that we obtain is the proportion of ICMEs causing SSs to the total number of certain types of ICMEs. However, different proportions could be generated by a random process. Moreover, different proportions do not necessarily mean different probabilities. To reveal the statistical significance of the different proportions from different types of ICMEs, we first calculate the probability of the random process. In statistics, it is assumed that the event with small probability does not occur. If the probability is smaller than 0.05, it could be interpreted that a certain proportion is not controlled by random process at a 95% confidence level. Second, we use one of the hypothesis tests on two groups to check if the higher proportion represents the higher probability. It is the Z-test(the chi-squared test) based on proportions of the independent two groups assuming the probabilities are the same (Shewart & Wilks 2004). In the equation, p 1 and p 2 are the proportions for group 1 and group 2 having certain signatures. The probabilities of the independent two groups are represented by q 1 and q 2 , which are assumed to be the same as q 1 =q 2 . The sample numbers are n 1 and n 2 for the two groups. At the 95% confidence level, if the Z value is larger than 1.645, the null hypothesis could be denied. This means that the different ratios of events with certain signatures to the total number in the two groups have statistical significance. The group having a larger proportion has larger probability. The other original result is the distribution of single storms caused by different subtypes of ICMEs on the Dst min and IMF B z . To examine how significant the difference between the two groups is, we adopt the Kolmogorov-Smirnov test(KS-test) (Massey 1951;Miller 1956;Marsaglia et al. 2003). The key parameter of the test is the p-value between the cumulative fraction function of the two distributions. If the pvalue is smaller than 0.05, the two distributions are statistically different at the 95% confidence level.

Result
In the studied 219 ICMEs, 95 ICMEs caused the SSs and 56 ICMEs caused the NSs, listed in Table 1. These 151 ICMEs could be divided into the CCs and the non-CCs, and HEs and non-HEs according to their carbon ionic charge and the helium abundance. Also, they could be divided into the MCs and the non-MCs according to their magnetic field structure. The numbers of ICMEs in each catalog are shown in Figure 2  We make the Z-test on the proportions of the CCs(39/69) and non-CCs(56/150), and that of HEs (37/69) and non-HEs (58/150), causing single storms, assuming the probabilities of the two groups are the same. The Z values are 2.76 and 2.07, which are larger than the critical value at the 95% confidence level: Z 0.05 =1.645. This means that the null hypothesis should be rejected. Thus, the probability of the CCs or the HEs to generate single storms is significantly higher than the non-CCs or the non-HEs. Next, we use the same hypothesis test on the proportions of the MCs(55/99) and the non-MCs(40/ 120) groups. The Z value is 3.43, which is also larger than 1.645. It is in accordance with the previous studies that the MCs have higher probability of causing geomagnetic storms than the non-MCs. Finally, we make the Z-test on the proportions of NSs occurrence in different types of ICMEs. Obviously, fewer CCs, MCs, and HEs generate non-storms than their opposite types. However, except for the HEs, the other two types do not show statistical significance at the 95% confidence level.
Furthermore, we analyze the intensity of 95 SSs caused by various types of ICMEs. The intensity of SSs is described by Dst min . We classify the SSs into five levels according to their Dst min : weak, moderate, strong, severe, and great. The counts and fraction of each level of single storms caused by the MCs and CCs are shown in Figure 3. Figure 3(a) shows the distribution of all 95 SSs in five levels. It is obvious that the fraction of moderate SSs is the largest in the 95 SSs. In Figures 3(c) and (g), the distributions of non-CCs and   Figure 4. It is obvious that the HE, CC&HE, and MC&HE tend to generate strong storms. More interesting, the CC&MC&HE is most likely to generate a severe storm. To compare the difference of the SSs caused by MCs, CCs, HEs, and CC&MC&HEs, we obtain the cumulative distribution of Dst min of 95 SSs. The result is shown in Figure 5. It is obvious that the CC&MC&HEs have the highest cumulative fraction. The HEs have the second highest fraction followed by the CCs and MCs. The non-MCs, non-CCs, and non-HEs have the least fraction in similar trend. Both the storm level distribution and the Dst min cumulative distribution indicate that the single storms caused by the CC&MC&HEs are obviously stronger than the others.
As far as we know, the MCs are ICMEs with a magnetic flux rope. Thus they tend to have strong negative B z causing magnetic storms. However, the classification of CCs and HEs is not based on the magnetic field structure but on the presence of prominence material. Why the CC&MC&HEs have the strongest geoeffectiveness is a question that deserves further investigation. To reveal the differences of the CCs, HEs, and MCs, we further classify the 95 ICMEs causing SSs into combination subsets, listed in Table 1.
We notice that CC&MC&HEs have the largest propor-tion(0.80) to cause the SSs, and they have the second smallest proportion(0.13) to cause the NSs. If we assume the different proportions are generated by random processes, we find that the two probabilities of CC&MC&HEs causing SSs and NSs are » 0.0028 . Such a small probability means that the CC&MC&HEs tending to cause SSs is not controlled by a random process at the 95% confidence level. It is not a random process for the CC&MC&HEs to generate nonstorms at the 85% confidence level. Next, we use the Z-test to check if the differences between the proportions of the SSs and NSs caused by CC&MC&HEs and non-CC&non-MC&non-HEs are statistically significant. The Z values are 3.72 and 1.97, respectively, which mean that the CC&MC&HEs do have a higher probability of causing SSs, and have a smaller probability of causing weak magnetic disturbance, at 95% and 85% confidence levels, respectively. Above all, we suppose that the CC&MC&HEs are the most geoeffective of the studied ICMEs.

Discussion
The statistical analysis results show that both the CCs and MCs tend to generate stronger storms, and together they have a larger probability to generate storms than each alone. Since the reason for the MCs to generate a strong magnetic storm is the southward magnetic field in its flux rope, we speculate that if the CCs or HEs have stronger southward B z as well. The cumulative fraction of maximum southward IMF larger than a certain value is shown in Figure 6. The ICMEs that are simultaneously CCs and MCs have the largest fraction compared to other subtypes related to the MCs, CCs, and the opposite types, shown in Figure 6(a). It becomes even more interesting when we consider the HE and its subtypes, shown in  Figure 6(b). The cumulative distribution of southward B z of the 12 CC&MC&HEs have the highest value. It is well known that the solar prominence is featured by the low temperature and high helium abundance chromospheric material staying on the loops. According to Yan et al. (2015), the prominence in the active region(AR) could have twisted flux ropes formed by shear flow. The AR prominence could also stay with emerging flux ropes from the photosphere (Wang et al. 2019). Furthermore, jets are observed to transport a large amount of chromospheric material and magnetic field intensity from lower atmosphere to AR prominence (Wang et al. 2018b). Therefore, the eruptive prominence is accompanied by enhanced twisted flux rope. This may explain why the ICMEs that are simultaneously the MCs, CCs, and HEs have the largest fraction containing strong southward magnetic field, which are most probable to generate magnetic storms, especially severe storms. Since the MCs are featured by the large-scale flux rope, and the CCs and HEs represent the cold prominence, respectively, in the CMEs. The ICMEs being CC&MC&HEs are related to the three-part CMEs in the solar  corona. This means that the three-part CMEs have significantly stronger geoeffectiveness.

Conclusion
This work studies 219 ICMEs from 1998 to 2011, in which 95 single ICMEs causing 95 single magnetic storms. These ICMEs could be classified by carbon ionic charge signatures into 39 CCs and 56 non-CCs, or by magnetic flux rope structure into 55 MCs and 40 non-MCs, or by N α /N p into 37 HEs and 58 non-HEs. Out of the 95 ICMEs, 26 are simultaneously CCs and MCs, 17 are CC&HEs, and 12 are CC&MC&HEs. First, we examine the intensity level of all the 95 single storms and that of SSs caused by different subsets of ICMEs. The most probable intensity level of the 95 single storms is moderate. Differently, the intensity distribution of single storms caused by the CCs is concentrating at a strong level. The storms caused by the non-CCs have quite similar intensity distribution to the 95 SSs. It can be noted that more than half(6/11) of the severe and great storms are caused by the CC&MCs, and the CC&MC&HEs have the largest fraction(5/12) of severe and great storms of any other subset of ICMEs. It should be noted that the single storms caused by the CCs, MCs, and CC&MCs concentrate at a strong level, those caused by the CC&MC&HEs concentrate at severe level, and those caused by the non-CCs, non-MCs, non-HEs, and non-CC&non-MC&non-HEs mostly distribute at a moderate level. Besides, we use the KS-test to examine the difference in the cumulative distribution of Dst min of 95 SSs caused by different ICMEs. The result shows that the difference is significant at 95% confidence. The CC&MC&HEs have the highest cumulative fraction, followed by the HEs as the second highest, the CCs as the third, and the MCs as the fourth.
Next, we investigate the magnitude of southward IMF of the ICMEs. The CC&MC&HEs show a significantly larger fraction of B z < −11 nT than any other subset of the ICMEs. Thus, whether or not the ICMEs are the MCs should not be the only variable to consider in evaluating their southward magnetic field and the magnetic storm level. The ion composition should also be taken into account, as the erupted prominence is proved to have enhanced twisted flux ropes. We would like to conclude that the ICMEs related to the three-part CMEs containing both the large-scale flux rope and the cold prominence have the largest fraction of strong southward magnetic field and have the largest probability to cause strong and even higher level magnetic storms.
Our work suggests that both the prominence-inside ICMEs and MC ICMEs should be considered in evaluating the geoeffectiveness and in the space weather forecast. The ICMEs containing prominence material and large-scale flux ropes are related to the three-part CMEs on the solar corona. This means that the three-part CMEs are not only important for the study of the eruption mechanism in the source region, but also important for the study of space weather. Besides, another question that is raised from our work is why the different signatures of prominence material seldom occur in the same ICMEs simultaneously. Is it an effect of the spacecraft trajectory or a result of the prominence propagation in the heliosphere? The Solar Orbiter Mission to be launched in 2020 having the perihelion of 0.28 au will provide the best chance to locate and understand the same three-part CME from its origin to the heliosphere.