This chapter presented results of an investigation of the solar sources and IP characteristics of 229 GMS events (Dst ≤ −50 nT) during SC 23. A particular aspect of these results is that they can be considered as an estimation of solar sources of GMS within on sunspot cycle (an average of 11 years).
While earlier similar studies focused on the sources of intense storms (Zhang et al., 2007), the present study has extended the analysis to moderate storms. The following are the most important results of the analysis:
• Most of the intense GMS (83%) during the period of study were caused by halo CMEs. For moderate storms, only 54% were associated with halo CMEs, while the remaining 46% seemed to originate from CIRs or geoeffective non-halo CMEs.
• The association of intense storms with ICMEs at 1 AU was very high (86%), compared to moderate storms of which only about 44% were associated with ICMEs.
• Using the information for X-ray flares associated to halo CME probable sources of storms, this investigation found that about 69% of intense storms were caused by CMEs originating close to the disk centre [within ±450 of the CMD], and only 51% of
moderate storms had solar sources close to the disk centre.
• There was no significant difference between intense and moderate storms as far as the class of X-ray flare associated with the driving halo CME. Generally, both intense and moderate storms were mostly associated with C and M-class flares.
• A comparison of full halo CMEs and partial halo CMEs, showed that full halo CMEs were generally associated with intense storms with an average Dst = −128 nT, while partial halo CMEs were linked to moderate storms with an average Dst = −92 nT. On the other hand, geoeffective parameters of full halo CME-driven storms (Bz, SW
speed, CME speed) had higher average values than those of partial halo CME-driven storms. In addition, full halo CME-driven storms were generally associated with class M flares, compared to partial halo CME-driven storms which were mainly associated with C-class flares.
• Full halo CME-driven storms were associated with CMEs mostly originating mostly close to the disk centre, while only 44% of partial halo CME-driven storms were asso- ciated with CMEs originating within the CMD.
• This investigation found that multiple halo CME-driven storms represent 26% of all CME-driven storms, the majority (69%) being intense storms. In addition, 92% of multiple CME-driven storms were associated with ICMEs at 1 AU.
• Finally, GMS and associated probable solar sources and IP properties demonstrated a triple peak in GMS, CME and ICME activities. The peaks are observed in each phase, namely in the rising phase (1998), maximum phase (2000) and declining phase (2005). 50% of non-halo CME (or CIR) driven storms were concentrated in the descending phase of SC 23.
The low geoeffectiveness of partial halo CMEs compared to full halo CMEs can be associated to the fact that the majority of partial halo CMEs originate far away from the solar disk center. For both CIRs and halo CMEs, if they originate far away from the the Sun center, the IP structures linked to them will have less geoeffective properties since only the outer flanks are expected to encounter Earth (Gopalswamy, 2008).
In this chapter, a quantitative analysis was conducted to explore probable solar sources and associated IP properties of GMS in SC 23. The aim of this kind of study should not only be limited to the identification and characterisation of the sources of GMS, but should also extend to the development of GMS prediction models for the purpose of minimising space weather effects. Some of the geoeffective parameters described in this chapter were used in NN-based models for predicting the occurrence of GMS. The next two chapters describe in detail the GMS prediction models that were developed.
Predicting the geoeffectiveness of halo
CMEs
Estimating the goeffectiveness of solar transient phenomena is of practical importance for the modelling and prediction of space weather. The ability to predict space weather needs an accurate prediction of GMS, phenomena which represent typical features of space weather. However, space weather prediction is still relatively inaccurate given the fact that the un- derlying physics of the main drivers, e.g. coronal mass ejections (CMEs) and associated X-ray flares is not yet sufficiently understood (Schwenn et al., 2005). This chapter describes the NN-based models developed for predicting the probability occurrence of GMS from halo CMEs and associated geoeffective parameters.
5.1
On the predictability of GMS with neural networks
Currently models for predicting GMS include statistical, empirical and physics based meth- ods. However, despite previous attempted theoretical models to forecast the magnetic storm occurrence (Dryer, 1998; Dryer et al., 2004), physics based models are still difficult to achieve due to the complex and non-linear chaotic system of solar-terrestrial interaction (Fox and Murdin, 2001; Schwenn et al., 2005). Space weather forecasters often prefer empirical ap- proaches based on observable data (Kim et al., 2010). Various functional relationships have been proposed for magnetic storm predictions. Empirical models for predicting GMS using CME-associated parameters observed at the Sun have been developed, including a recent model by Kim et al. (2010). Other authors prefer statistical methods, e.g. Srivastava (2005) used a combination of solar and IP properties of geoeffective CMEs in a logistic regression model to predict the occurrence of intense GMS.
Empirical methods also include NN methods which are input-output models and have been proven to be efficient in capturing the linear as well as the non-linear processes (Kamide et al.,
1998). As indicated in Section 3.1 of Chapter 3, NN techniques have been described by var- ious authors to be suitable for predicting non linear systems. If a network is well-designed and trained, it can improve a theoretical model by performing generalisation rather than only curve fitting. By changing the NN input values, it is possible to investigate the func- tional relationship between the input and the output and thereby be able to derive what the network has learned (Lundstedt, 1997). In a NN-based model developed by Valach et al. (2009), geoeffective solar events such as solar X-ray flares (XRAs) and solar radio bursts (RSPs) were used to predict the subsequent GMS. In order to improve the forecast of GMS, Dryer et al. (2004) suggested that models should include both solar and near-Earth geoef- fective conditions.
For this study, a combination of solar and IP properties which are characteristic of geoef- fective halo CMEs were used in a NN model to predict the probability occurrence of GMS following halo CME events.