Electromagnetic ion cyclotron (EMIC) waves may contribute to ring current ion and radiation belt electron losses, and theoretical studies suggest these processes may be most effective during the main phase of geomagnetic storms. However, ground-based signatures of EMIC waves, Pc1–Pc2 geomagnetic pulsations, are observed more frequently during the recovery phase. We investigate the association of EMIC waves with various storm phases in case and statistical studies of 22 geomagnetic storms over 1996–2003, with an associated Dst < −30 nT. High-resolution data from the GOES 8, 9, and 10 geosynchronous satellite magnetometers provide information on EMIC wave activity in the 0–1 Hz band over ±3 days with respect to storm onset, defined as commencement of the negative excursion of Dst. Thirteen of 22 storms showed EMIC waves occurring during the main phase. In case studies of two storms, waves were seen with higher intensity in the main phase in one and the recovery phase in the other. Power spectral densities up to 500 nT2 Hz−1 were similar in prestorm, storm, and early recovery phases. Superposed epoch analysis of the 22 storms shows 78% of wave events during the main phase occurred in the He+ band. After storm onset the main phase contributed only 29% of events overall compared to 71% during recovery phase, up to 3 days. Some differences between storms were found to be dependent on the solar wind driver. Plasma plumes or an inflated plasmasphere may contribute to enhancing EMIC wave activity at geosynchronous orbit.
 The observation of electromagnetic ion cyclotron (EMIC) waves on the ground and by Earth orbiting satellites provides data for diagnostic techniques to study the dynamics of the magnetosphere. These techniques may relate to wave generation and propagation properties or the state of the magnetosphere associated with the waves. EMIC waves are generated as left-hand polarized waves through wave-particle interaction involving 10–100 keV anisotropic ring current protons and sometimes heavy ions [Mauk and McPherron, 1980; Roux et al., 1982; Inhester et al., 1984]. The dispersion relation for EMIC waves propagating parallel to the geomagnetic field in a cold electron-proton plasma shows that the parallel energy of the interacting protons has a minimum at the geomagnetic equator, suggesting that the equatorial region is favored for EMIC wave generation. This is supported by CRRES and other Poynting flux observations which show a source region within ±11° of the equator [Loto'aniu et al., 2005]. EMIC wave energy, typically in packets, propagates from the equatorial region along geomagnetic field lines down to the ionosphere. In the classical paradigm some wave packet energy is reflected back along the field line, through the equatorial region into the opposite hemisphere. This scenario has been questioned following the absence of reflected wave packet energy [Erlandson et al., 1992; Fraser et al., 1996; Mursula, 2007]. However, much of the wave energy reaching the ionosphere is transmitted into the F2 region waveguide [Manchester, 1966; Greifinger and Greifinger, 1968] and the broader ionospheric Alfvén resonator (IAR) [Polyakov and Rapoport, 1981]. In the ionospheric waveguide, following mode conversion, energy propagates horizontally in the right-hand polarized isotropic mode. Energy leaks to the ground through the E region, and consequently EMIC wave energy is seen at a wide range of latitudes and often longitudes [Tepley, 1964; Fraser, 1975]. Therefore it is difficult to locate the magnetospheric field line along which the EMIC wave propagates from ground observations alone unless direction of arrival techniques are utilized [Fraser and Nguyen, 2001]. Wave properties may also be affected by ionospheric transmission [Manchester, 1966; Greifinger and Greifinger, 1968].
 It has been suggested for many years that EMIC waves make an important contribution to ring current ion loss during geomagnetic storms through wave-particle interaction in the magnetosphere. In this context the proton cyclotron instability causes pitch angle diffusion producing a partial refilling of the loss cone and precipitation of protons with energies typically in the range 10–100 keV [Yahnina et al., 2003; Yahnin and Yahnina, 2007; Yahnin et al., 2007]. For example, the instability process, investigated by Criswell , showed that EMIC waves are preferentially generated in regions of overlap of the cold plasma populations in the afternoon sector with energetic ring current ions. Consequently, Criswell  and Kawamura et al.  suggested that EMIC waves might preferentially be seen over a range of L ∼ 2–5. However, this could not explain why waves were seen with greater occurrence outside the plasmapause in the magnetosphere trough region [Anderson et al., 1992; Fraser and Nguyen, 2001]. This dilemma was partially solved through IMAGE-EUV observations of attached radial cold plasma drainage plumes extending out from the plasmapause into the ring current and plasmatrough [Goldstein et al., 2005] and GOES observations of EMIC waves seen within plasma plumes and broad-scale subauroral proton arcs seen by the IMAGE-FUV experiment [Fraser et al., 2005a]. In a following study it was suggested that the radial gradients associated with the “edges” of radial plasma plumes extending out into the outer magnetosphere and convecting sunward may produce the necessary condition for proton instability [Fraser et al., 2006]. This is supported by a recent study by Morley et al.  where a multipoint study of EMIC waves during a storm shows that EMIC waves are likely to be observed on density gradients at the edges of plasmaspheric plumes after storm onset. The drainage plume itself is also highly structured which can support wave guiding and hence extensive growth even within the plume, as suggested by Spasojević et al.  and studied theoretically by Chen et al. .
 It is therefore now well accepted that EMIC waves generated in the equatorial region will precipitate ring current protons, which provides a viable mechanism contributing to ring current decay. The dynamics of the ring current are largely determined by the injection of particles from the plasma sheet during geomagnetic storms and substorms. Consequently it is important to study the characteristics of EMIC waves in relation to storms and substorms.
 Over the years there have been a number of studies made of the association of Pc1–Pc2 waves seen on the ground and in space during and following the main and recovery phases of geomagnetic storms. The association of Pc1–Pc2 EMIC waves with storms has important implications for storm time ring current decay [e.g., Jordanova et al., 2001] and electron loss processes [Meredith et al., 2003] in the radiation belts. Using low-latitude ground-based observations Wentworth  found that Pc1 wave activity maximized in the 3–7 days following the storm main phase. Similar results were found at middle and high latitudes [Plyasova-Bakounina and Matveyeva, 1968; Heacock and Kivinen, 1972]. More recently Engebretson et al. [2008a] confirmed these results using high-latitude Antarctic Pc1–Pc2 data. Using 8 years of Pc1–Pc2 data recorded at low latitudes Bortnik et al.  found that Pc1 waves were 2–3 times more likely to be observed in the 2–4 days following moderate storms and 4–5 times more likely in the 2–7 days following intense storms (Dst < −100 nT). These results were backed up by a 90 day low Earth orbit study of Pc1–Pc2 EMIC waves using data from the ST5 string of 3 spaced satellites, where waves were not seen during the storm main phase [Engebretson et al., 2008b].
 There have been a number of space-based observations of EMIC waves associated with geomagnetic storms. These observations provide the opportunity to study storm phase relationships of EMIC waves without the impediment of transmission through the ionosphere, which may compromise ground observations. Bossen et al. , using geosynchronous ATS-1 data in a statistical study, showed that EMIC wave activity was enhanced within an 8 h period centered on the storm main phase minimum, associated with Dst < −40 nT. In a detailed study Bräysy et al.  used data from the electric field experiment on the low-altitude Freja satellite to study EMIC waves during the great magnetic storm of 2–8 April 1993, where Dst minimized at −169 nT. Prior to the main phase EMIC waves were observed associated with magnetospheric compressions similar to those reported by Anderson et al.  and Anderson and Hamilton . This pattern changed dramatically at the start of the storm main phase. The wave amplitude increased by an order of magnitude and during the main phase the waves were seen at lower latitudes during the late evening. Lower-frequency oxygen band waves were also seen during the main phase. The EMIC wave occurrence rate was found to be enhanced by a factor of five during the main and recovery storm phases, compared with nonstorm quieter times by Erlandson and Ukhorskiy  using DE-1 magnetic field data covering 3.5 < L < 5. However, the main phase statistics in this study are dominated by one large storm which contributed to 21 of 22 events considered. More recently Engebretson et al.  using Cluster data during the October 2003 Halloween storms (Dst = −(300–400) nT) found a somewhat rare EMIC wave propagating obliquely to the ambient field during a positive Dst excursion in the main phase. No further waves were seen until the recovery phase.
 A simple classification of Pc1–Pc2 waves seen on the ground identifies two types of waves: structured and unstructured [Hayashi et al., 1981]. The unstructured waves tend to occur at the higher latitudes, possibly in the plasmatrough, while the structured waves showing typical banded “pearl” fine structure may be confined to the plasmapause or plasmasphere. Observing both types of waves at middle latitudes Kerttula et al.  noted that while both disappeared during the main and early recovery phase, the structured wave occurrence increased by 4–5 times maximizing on day 4 after the onset day. In contrast unstructured waves occurred less frequently and were only weakly affected by the storm evolution.
 The importance of EMIC and Pc1–Pc2 wave observations in space and on the ground at storm times relates to the modeling of ring current ion decay and radiation belt electron losses and associated dynamics. In addition to charge exchange and Coulomb scattering, the accepted two main contributors to ring current decay, ion cyclotron resonance has been shown through modeling to play a significant role [Fok et al., 1996; Jordanova et al., 1997; Khazanov et al., 2006; Kozyra et al., 1997; Jordanova et al., 2007]. In this scenario the cyclotron instability produces pitch angle diffusion leading to partial refilling of the loss cone, and precipitation of keV protons. According to current modeling this is most likely to occur during the main phase of storms when wave amplitudes are enhanced and waves move to lower L shells producing higher wave frequencies [Kozyra et al., 1997; Jordanova et al., 2001; Khazanov et al., 2006]. However, ground-based Pc1–Pc2 observations indicate that waves are generally not seen early in the storm, particularly the main phase. This has variously been attributed to attenuation in the ionosphere or the ionospheric wave guide [Kerttula et al., 2001; Engebretson et al., 2008a; Bortnik et al., 2008], unfavorable ground station location and the masking of Pc1–Pc2 wave activity by intense storm time broadband Pi1 noise. The ionospheric attenuation argument has been discounted by Engebretson et al. [2008b], who found very few waves in the storm main phase above the ionosphere at low Earth orbit. Consequently the question of whether Pc1–Pc2 or EMIC waves are observed during the main phase of geomagnetic storms still needs to be answered.
 In the current study we employ selected high-resolution Pc1–Pc2 magnetometer data from the GOES satellites over a seven year period to study geomagnetic storm effects at geosynchronous orbit in the outer region of the ring current. It is considered that this location in the middle magnetosphere near the equatorial EMIC wave source region [Fraser et al., 1996; Loto'aniu et al., 2005; Morley et al., 2009] will provide more representative data than fast moving low-altitude satellites such as Freja used by Bräysy et al. . Following a description of the data, and analysis criteria, two storm event studies will be given and statistics on the overall data set to place the events in context. Section 6 discusses the results in light of previous work.
2. Data and Analysis
 Observations from the triaxial fluxgate magnetometers on the GOES series of geosynchronous satellites operational over 1996–2003 were used to study the storms reported here. The GOES 8–12 satellites are three-axis stabilized spacecraft and fluxgate magnetometer data are typically analyzed in the spacecraft coordinate frame. In this (Hp, He, Hn) system, Hp is parallel to the Earth's spin axis for zero inclination orbit, which is approximately parallel to the geomagnetic field. The He component is perpendicular to Hp and directed Earthward. The Hn component completes the orthogonal system and is directed eastward. The resultant total wave field is Ht. Although EMIC waves in space show greatest power in the azimuthal magnetic component [Fraser, 1985] this study will utilize the total wave power in event studies and statistics. The location of the GOES satellites during the storms in this study are listed in Table 1. The high-resolution data, sampled at 0.5 s provides information on the low end of the EMIC wave spectrum over 0.1–1 Hz. The response between 0.5–1.0 Hz is reduced by a five-pole Butterworth low-pass antialiasing filter [Fraser et al., 2005a]. This frequency response allows the observation of EMIC waves with amplitude ≥ 1 nT to be seen up to approximately 0.8 Hz. Below 0.5 Hz the noise level is ∼0.1 nT. Fortunately EMIC waves seen on the ground and in space are mostly below 1 Hz during the local afternoon and evening sector [Fraser, 1968; Anderson et al., 1992; Fraser and Nguyen, 2001]. It is important to note that EMIC waves with frequencies > 1 Hz may be seen during storm times [e.g., Bräysy et al., 1998] and will not be observed by GOES. While this may bias the current results it should be noted that there is theory predicting waves seen during the main phase of storms are more likely to appear in the O+−He+ bands [Thorne and Horne, 1997], which are typically below 0.8 Hz at geosynchronous orbit.
Table 1. Location of GOES Satellites Over 1996–2003
Local Time (hours)
Apr 1996 to May 2003
UT − 5
UT − 13
Feb 1996 to Jun 1998
UT − 9
UT − 13.67
Sep 1998 to Oct 2003
UT − 9
 In situ geosynchronous plasma density data were provided by the Magnetospheric Plasma Analyser (MPA) instrument on board the Los Alamos National Laboratories satellites [McComas et al., 1993]. Here various data from the MPAs on three LANL satellites, LANL-97A, 1991-080 and 1994-084 have been used for ion density measurements at energies <100 eV. These ions are predominately protons although O+ and He+ heavy ions are also included.
 The GOES EMIC wave data set used in this study was derived from high-resolution data assembled to study radiation belt electron dropout mechanisms. In the current study from the 30 storms available we have selected 22 so-called “classical” and sometimes modest storms with distinctive, quiet geomagnetic conditions prior to storm onset, a single main phase with a minimum Dst < −30 nT followed by a recovery phase [Gonzalez et al., 1994]. The analysis employs superposed epoch analysis techniques on the total EMIC wave field Ht where the time interval spanned was ±3 days centered on the onset of the main phase identified by the start of a negative excursion of Dst. Variation in the length of the data set for each storm was dictated by data availability (Table 2). The +3 days limit allows study of the early recovery phase but waves seen in the late recovery phase of slow recovery storms may not always be seen. This is not a disadvantage since we are primarily interested in EMIC waves seen in the main and early recovery phases, and an indication of quiet time wave activity can be obtained from the prestorm quiet interval.
Onset is zero crossing time, to the nearest Dst hour, where Dst commences to move consistently negative to at least −30 nT. Main phase duration is difference in time between “Onset (UT)” and “UT Hours of Dst Minimum.” G8, G9, and G10 indicate data availability with respect to onset day. The asterisks indicate the two case study storms shown in Figures 1 and 4.
0900, 17 Apr
0500, 23 Oct
0100, 18 Feb
2100, 10 Mar
+2 to +3
+2 to +3
+2 to +3
0700, 4 Apr
2200, 2 Oct
+2 to +3
+2 to +3
2200, 13 Nov
0800, 17 Apr
0200, 31 Jul
0000, 23 Sep
CME & CIR
0700, 22 Oct
CME & CIR
1000, 13 Dec
2200, 11 Jan
2300, 5 Feb
1200, 12 Feb
0600, 24 May
+2 to +3
+2 to +3
0000, 18 Sep
0400, 17 Jul
0800, 22 Jul
0100, 10 Sep
1400, 16 Sep
0800, 24 Sep
 Dynamic spectral analysis of the GOES 0.5 s Ht data was undertaken in approximately hourly steps of 7000 data points or 58.33 min. An FFT was performed on a window of 200 points (1.6 min) with an overlap of 50 points (0.4 min), providing 140 FFTs over the hour. The FFT frequency resolution was 10 mHz. Input parameters for the statistical study described in section 4 were obtained from 4 h dynamic spectral plots using 120 point FFTs with a 120 point step and no overlap resulting in a frequency resolution of 16.6 mHz. In all cases the initial magnetic field was high pass filtered at 0.01 Hz using a 200 point smoothing function. The FFT intervals were then identified, linearly detrended, tapered with a Hanning window in the time domain, compensated for the Hanning window in the frequency domain and then normalized by dividing the spectrum by the frequency resolution.
 EMIC waves and their properties associated with two storms, selected from the 22 storm data set, 21–26 October 1999 and 9–14 January 2000, will be described in detail. These are then placed in context by considering statistics over the 22 storms
3. Individual Storm Studies
 Detailed studies of two individual storms with minimum Dst of −237 nT and −81 nT, as listed in Table 2 and indicated by an asterisk, have been undertaken.
3.1. DOY 294–298: 21–25 October 1999
 The storm began on 21 October and was identified by a sudden change in Bz (IMF) at ACE from 20 nT to −25 nT commencing at 2337 UT, as seen in Figure 1. Time has been adjusted for the propagation from ACE using the average of V, the solar wind velocity. A check of the GOES 8 magnetogram shows Bz turning negative at 2337 UT. The southward turning preceded a small positive peak in Dst at 0000 UT which then decreased, becoming negative into the main phase of the storm. Over the same interval the solar wind velocity of 450–500 km s−1 gradually increased to 550 km s−1 at 0710 UT on 22 October at the main phase minimum, and increased faster to 700 km s−1 by 0900 UT. A small increase in solar wind dynamic pressure from 3 to 7 nPa occurred before onset. The pressure showed a spike over 0700–0900 UT, maximizing at 38 nPa. Bz abruptly returned to near zero at this time also. Dst reached its minimum at 0700 UT on 22 October with Kp = 8. During the recovery phase Bz was near zero and the solar wind pressure and velocity showed gradual decreases through to the end of the data on 24 October. Preceding this storm a coronal mass ejection (CME) shock arrived at 0220 UT on 21 October, followed by a high-speed stream beginning in the recovery phase.
Figures 1f and 1g show the EMIC waves observed by GOES 8 located at 75°W (geog) and GOES 10 at 135°W (geog). The wave frequency normalized to the local proton cyclotron frequency at the satellite is plotted. This provides a separation of waves into the oxygen (O+), helium (He+) and proton (H+) bands. Here we define the proton (H+) band with frequencies fHe+ < f < fH, the He+ band, fO+ < f < fHe+ and the oxygen band f < fO+. The subscripted frequencies are the cyclotron frequencies of the heavy ions. Prior to storm onset waves were seen in the He+ and H+ bands. GOES 8 and GOES 10 are separated by 4 h in local time and generally see different waves in different bands at different times although they often cluster in broad similar UT time intervals indicating broad temporal coincidence. For example, both satellites see waves in the H+ and He+ bands over 0000–1200 UT on 21 October before storm commencement. Waves are observed in the He+ band during the main phase of the storm by GOES 10 while at GOES 8, the waves ceased at onset. Waves were sparse at both satellites during the recovery phase.
 A comparison between the solar wind dynamic pressure in Figure 1c with GOES 8 and 10 EMIC wave occurrences suggests some wave activity is associated with relatively small magnetospheric compressions. This relationship was found to occur on the dayside over 08–16 MLT at L > 6 by Anderson and Hamilton , for almost half of the EMIC wave events seen by AMPTE-CCE. The sudden increase in solar wind dynamic pressure in Figure 1 at 0225 UT coincided with a magnetic field increase of ∼10 nT at GOES 10 located at 1730 LT on 21 October, and the initiation of EMIC wave activity in the He+ band. The role of compression here is unclear since the waves were seen in an extended dusk plasmasphere where the enhanced cold plasma density may contribute to wave instability (Figure 11).
 Detailed dynamic Ht spectra of the EMIC waves in the 0.1–1 Hz band noted in Figure 1 are shown in Figures 2 and 3. These waves are indicated in Figures 1f and 1g by small triangles showing the wave occurrence with respect to Dst. Figure 2 shows dynamic spectra of the prestorm interval with waves occurring over 0700–2000 UT at GOES 8 and 10. In Figure 2b over 0745–0845 UT GOES 8 sees a narrow band of rising frequency from 0.4–0.6 Hz, in the H+ band. This continues in Figure 2c from 0845–0950 UT with a frequency of 0.63 Hz. Figure 2d shows broader band waves over 0.2–0.6 Hz in the He+ band at 1857–1919 and 1930–2007 UT, some 3 h prior to storm commencement. The narrow band waves are seen postmidnight at 0245–0450 LT on the nightside whereas the lower-frequency broader band waves occur in the afternoon over 1355–1507 LT. This follows the pattern where the lowest-frequency EMIC waves seen on the ground occur in the local afternoon sector [Fraser, 1968]. In contrast, Figure 2a shows GOES 10 in a low magnetic field environment near midnight and waves in both the He+ and H+ bands over 0723–0810 and 0832–0900 UT. These may be broadband events over 0.15–0.4 Hz where the frequency slot is created by the nonpropagation stop band between the helium cyclotron and cutoff frequencies, a result of He+ presence in the background plasma [Fraser et al., 2005b, 2006]. No other waves are seen by GOES 10 prior to storm onset.
Figure 3a shows that GOES 8 observes short bursts of EMIC wave activity in the He+ band over 0.2–0.5 Hz ceasing at storm onset when Bz (IMF) commences to fall at 2337 UT. At this time GOES 8 is in the dusk sector at 1810 LT. In Figure 3b GOES 10 in the early afternoon sector shows a series of twelve He+ band 0.25–0.6 Hz wave bursts over 2332–0015 UT which appear to be a result of modulation by long period Pc5 mixed mode ULF waves [Mursula, 2007; Loto'aniu et al., 2009]. The first of these bands seems to occur prior to storm onset at 2337 UT by about 2 min. There is a delay of 22 min between the waves observed by GOES 8 and GOES 10. This time delay is equivalent to a drift rate of 2.4o min−1 and relates to sunward gradient-curvature drifting protons with energies in the range 44–48 keV assuming pitch angles of 90o and 45o. More intense waves in the He+ band are seen over 0015–0105 UT. The GOES 10 waves show a decreasing frequency trend with the midfrequency decreasing from 0.5 Hz to ∼0.1 Hz over 33 min.
3.2. DOY 10–14: 10–14 January 2000
 As seen in Figure 4 this storm commenced gradually with Dst decreasing and turning negative at 1200 UT on 11 January before showing an increase prior to the main negative Dst excursion at ∼1600 UT. The storm main phase reached a minimum Dst = −81 nT at 2200–2300 UT, and was followed by a gradual recovery over more than four days. The solar wind dynamic pressure increased from 3 nPa to 6 nPa over 1400–1800 UT prior to the storm onset on the previous day, 10 January. A sudden increase to 13 nPa occurred at storm onset at ∼1500 UT. The pressure had returned to preonset levels by the end of the main phase. The solar wind velocity showed a gradual increase from 350 km s−1 late on 11 January to a maximum of 600 km s−1 at 0600 UT on 12 January, in the early recovery phase. This is typical of a storm associated with a corotating interaction region (CIR) commencing with the increase in solar wind speed at ∼2200 UT on 10 January and the high-speed stream interface arriving at 1500 UT on 11 January [e.g., Tsurutani et al., 2006; Denton et al., 2008].
Figures 4f and 4g illustrate the EMIC wave activity occurring before, during and after this storm. EMIC waves occurred before the storm on 10 January in the early hours at GOES 10 in the He+ and H+ bands under quiet geomagnetic conditions. GOES 8 observed waves near f O+ at this time. More waves were seen in the He+ and H+ bands by both satellites following the pressure increase commencing at 1400 UT on 10 January, continuing until Dst went negative at 1200 UT on 11 January. Waves were seen again in the main phase and the early recovery phase, from 1700 UT until about 0100 UT on 12 January. There were only a few waves observed in the two days of the recovery phase, shown at 1900 UT on 12 January and 0900 UT on 13 January at GOES 8 and GOES 10, respectively.
 In Figure 5 we show examples of the preonset spectra from GOES 8 and 10 on 11 January. Figure 5a shows GOES 10 EMIC waves in the He+ band over 0.2–0.3 Hz at 0300–0405 UT and 0425–0435 UT, in the dusk sector 1800–2000 LT, while GOES 8 in Figure 5b sees waves in the H+ band over 0710–0840 UT in the early morning 0200–0400 LT. The prestorm waves seen over 0700–1200 UT on 11 January (identified in Figure 4 but not shown here) were short duration bursts of no longer than 5–10 min in the H+ band, and located on the nightside at GOES 8 (0200–0600 LT) and GOES 10 (0200–0300 LT). These events occurred in association with a solar wind pressure increase and minor peak over 0800–1200 UT on 11 January, seen in Figure 4.
 Following storm onset at ∼1500 UT, EMIC waves were observed from 1635–2015 UT, around noon and into the early afternoon, at GOES 8 in the He+ band (Figure 6). These waves were more intense than those seen prior to storm onset and showed a packet structure modulation associated with the presence of very regular Pc5 long period ULF waves [Loto'aniu et al., 2009]. After 1845 UT the modulated EMIC waves showed signatures in the H+ band accompanied by a slowly decreasing frequency trend. GOES 10 did not see EMIC waves until the main phase minimum and early into the recovery phase at 2345–0050 UT on 11–12 January. These storm time waves, shown in Figure 7a, were in the He+ band and exhibited similar Pc5 modulation to the GOES 8 waves in the afternoon sector. It is interesting to note that GOES 10 was located in the early morning hours 0730–1130 LT and did not see main phase waves until 1445 LT (2345 UT) when it passed through the local time zone where GOES 8 saw main phase waves earlier. Later in the recovery phase GOES 10 observed patchy short duration wave packets in both the He+ and H+ bands (Figure 7b).
4. Statistical Analysis
 The EMIC waves associated with the storms described are just two examples from the 22 storm data set covering 1996–2003 and comprising storms which showed a quiet interval prior to storm onset, an easily identifiable storm onset and a single main phase minimum. This required a significant excursion of Dst to −30 nT or less and an identifiable main phase followed by a gradual recovery phase. It should be remembered that the storms in this study were identified only for those times where high-resolution GOES magnetometer data were available.
 A superposed epoch analysis, with respect to onset times, for the 22 storms shown in Table 2 has been undertaken using EMIC wave properties, magnetospheric indices and upstream solar wind and IMF parameters. Figure 8 shows a plot of the local time and wave frequency response over 3 days before and after the storm onset time as indicated by the negative excursion of Dst at the commencement of the main phase. Each data point represents an EMIC wave event identified from dynamic spectra. Spectra were plotted over a 4 h interval. Distinct waves were identified using the PC mouse cursor and wave activity was binned every hour with long duration events occupying a number of sequential bins. For each event the wave power, frequency, UT, LT and Dst were logged. To determine wave power, a box was constructed centered around the maximum power seen in a 1 h dynamic spectral interval. The box spanned 15 min with a frequency width of 0.1 Hz. Maximum wave power was the maximum inside this box and average power the average inside the box. From Figure 8 it is seen that more EMIC wave events appear in the afternoon 1200–2000 LT interval than during the night and morning hours. This is to be expected from the diurnal variation in EMIC wave occurrence [Anderson et al., 1992; Fraser and Nguyen, 2001] and is probably associated with the presence of an extended plasmapause bulge region, or if convection is present, possibly a radial plasma plume in the post noon and evening magnetosphere plasma trough region, which includes geosynchronous orbit [Fraser et al., 2005a, 2006; Spasojević et al., 2005; Immel et al., 2005]. In contrast to the afternoon-evening sector, waves in the H+ band dominate the midnight to noon hours. The greatest concentration of waves is seen during the main phase and at frequencies in the He+ band.
Table 3 shows a summary of waves seen in the H+ and He+ bands with respect to prestorm, main phase and early recovery (up to 3 days post onset). Of the 22 storms listed in Table 2, 13 showed EMIC wave events in the main phase. For the 448 wave events identified, 243 (54%) occurred within 3 days prior to onset and 205 (46%) in the 3 days following. There were almost equal numbers of events in the He+ and H+ bands with no significant preference in the 3 days before or after onset. Considering after onset waves only, more waves (78%) occurred in the He+ band during the storm main phase while during the recovery phase the H+ band (61%) showed more wave events. The most significant result from Table 3 is the preference for He+ over H+ band waves to be generated during the main phase and the reverse during the recovery phase.
Table 3. Statistics of EMIC Wave Events
 Another important parameter is the power in the EMIC waves and how this varies over the storm phases. Figure 9 shows the EMIC wave power spectral density for events from the 22 storms in Table 2 separated into the He+ and H+ bands with respect to storm onset time. The highest wave power, up to 1000 nT2 Hz−1, is seen in the He+ band from onset to 1.5 days after onset with an apparent relative minimum approximately 0.5 days after onset, followed by high power again over 2.5–3 days. There is also higher power in the He+ band prior to the storm than during the recovery phase beyond 1 day. In contrast, the peak power in the H+ band was generally much lower, at less than 150 nT2 Hz−1, again with higher power prior to storm onset. It is also noted that higher power averaged over 3 days was seen prior to onset compared to after onset. The higher wave power seen in the prestorm phase compared with the recovery phase could be due to the presence of a cold plasmasphere extending out to geosynchronous orbit and contributing to higher growth rates [Borovsky and Denton, 2009]. Other statistics (not shown) indicate that geomagnetic activity prior to onset and indicated by Dst was, on average, typically quieter at EMIC wave event times than during the recovery phase.
 The storm data set in Table 2 has been analyzed with respect to CME and CIR storms. This shows 9 storms were CME associated, 10 CIR associated with 2 of the remaining 3 a mix of CME/CIR and one associated with a slow solar wind stream. The relationship between EMIC wave peak power for CME and CIR storms relative to storm onset time is shown in Figure 10. Here CIR and CME driven storms show similar distributions with power up to 500 nT2 Hz−1. One exception is a CME with power at 1000 nT2 Hz−1. However, CME power is typically between 30–500 nT2 Hz−1 during the main phase while the equivalent CIR power has lower threshold and a greater range of 6–500 nT2 Hz−1. Power distribution is similar for CME and CIR driven storms after one day into recovery dropping from 500 nT2 Hz−1 to 20–30 nT2 Hz−1 at the end of day 3.
 The distributions of a number of parameters, including peak power and local time occurrence, were tested for statistically significant (>95% confidence) differences using a two-sided Kolmogorov-Smirnov test [e.g., Conover, 1999]. The local time occurrence distributions of He+ and H+ band EMIC waves were found to be different in all storm phases. The distributions of normalized wave frequency in the main phase were also found to differ between CME- and CIR-driven storms: EMIC in the main phases of CME-driven storms were observed almost entirely (∼90%) in the He+ band. A further difference can be noted between CME- and CIR-driven storms across this data set: the peak power in the late recovery phase (>1 day from epoch) tends to higher values for CIR-driven storms.
5. Plasma Plumes
 It has been shown that EMIC waves seen by the GOES satellites are often associated with plasma plumes which extend to geosynchronous orbit [Fraser et al., 2005a, 2006; Morley et al., 2009]. In this situation it is of interest to study the effects of plumes on EMIC waves at storm times. Figure 8 shows that the distribution of the waves in local time is reasonably evenly spread over 1200–2400 LT before the storms with less waves over the 0000–1200 LT morning time. Following onset the distribution was more evenly spread over the whole day with a preference for the earlier afternoon hours. However, it should be recalled from section 4 that the statistical hypothesis testing showed no clear differences in LT distributions between different intervals relative to epoch. The association of EMIC wave events with plumes is illustrated in Figure 11 for the 21–25 October 1999 storm. Here enhanced plasma ion concentration data from the MPA experiment on board three LANL geosynchronous satellites (L91, L94 and L97) are plotted with respect to local time. Figure 11 (top) shows where the LANL-MPA instrument measured enhanced low-energy (1–130 eV/q) ion concentrations. The abscissa is universal time, and the ordinate is local time; thus, a geosynchronous satellite traces a diagonal path covering 24 h of local time in one day. The thin black lines mark where data are available from the MPA instruments on the operational LANL spacecraft. Where the low-energy ion concentration, Ni > 10 cm−3, a small symbol is plotted; if Ni > 25 cm−3 a larger symbol is plotted; when Ni > 40 cm−3 the largest symbol is used. Additionally, the three levels of ion concentration have been color-coded such that the lightest gray corresponds to Ni > 10 cm−3 and the darkest gray indicates Ni > 40 cm−3. Each symbol type represents the LANL spacecraft from which the data were obtained (square: 1991-080; triangle: 1994-084; diamond: LANL-97A). The overplotted colored bars indicate where EMIC waves were observed by GOES 8 (blue) and GOES 10 (green). Figure 11 (middle) shows the Kp index across this interval, and Figure 11 (bottom) shows the Sym-H index, a 1 min resolution version of the Dst index. Prior to storm onset a wide dayside distribution of plasma is seen over 0800–2100 LT on 21 October, possibly indicating a full plasmasphere extending out to geosynchronous orbit. About a day later the plasma enhancement has narrowed down to about 4 h over 1100–1500 LT. This is suggestive of a plasma plume forming on the dayside where sunward convection processes erode the plasma on the dusk edge of the plume while counteracting corotation rotates the plume to later local times. The time history of this narrowing and wrapping of a plume is well described by Spasojević et al.  and Goldstein and Sandel . At the main phase of the storm there is still an indication of a plume at geosynchronous orbit. However, no further plasma enhancement is observed at geosynchronous orbit until 24 October, the third day of the recovery phase, when a narrow plume is observed over 1300–1400 LT. It could be assumed that during the main phase and early recovery the plasmasphere is significantly eroded to altitudes below geosynchronous. The lack of cold plasma may explain the absence of EMIC waves during the early recovery phase. During the prestorm interval in Figure 11 EMIC waves are seen within the plume by both GOES 8 and 10 near the outer or eastern edge of an extended plasmasphere. This may be a consequence of convecting keV plasma impinging on the cold plasma density gradient at the edge of the plume, resulting in EMIC wave generation.
 This study of the relationship of EMIC wave occurrence and properties associated with geomagnetic storms is based on in situ GOES satellite results at geosynchronous orbit over 1996–2003. The GOES data set included 22 storms shown in Table 2 and selected as “classical” from a list of 30 storms showing a defined onset time followed by a negative Dst culminating in the main phase minimum and followed by a typical recovery phase. It is important to appreciate that the particular data set became available as a consequence of another study on storms associated with outer radiation belt electron dropouts, and does not necessarily include all storms occurring in these years and the span of the data set for individual storms is not always ±3 days, as noted in Table 2. It has been suggested that electron dropout may show an L shell dependence [Bortnik et al., 2006] which could influence our data set. Whether extended plasma plumes, tongues of cold plasma stretching beyond the plasmapause and often out to geosynchronous orbit extend the region of EMIC influence on electron dropout is unknown. The data set is, however, considered sufficiently representative of storms to enable the current study to be undertaken. The advantage of direct magnetospheric observations compared to ground-based observations is elimination of the complications of the ionosphere. Ground observations have been described in section 1 and are summarized in detail by Engebretson et al. [2008a] and Bortnik et al. . The primary observation from ground data is that Pc1–Pc2 waves, the ground signatures of EMIC waves, are rarely observed, if at all, during the storm main and early recovery phases [Engebretson et al. 2008a]. The reasons for this, as noted in section 1 may be ionosphere related, including direct transmission attenuation through the ionosphere, or attenuation and propagation in the ionospheric F2 region waveguide and ionospheric Alfven resonator (IAR) [Manchester, 1966; Fraser, 1975; Demekhov et al., 2000]. As a consequence ground-based observations cannot easily determine the location of the EMIC magnetospheric source flux tube ionospheric footprint region which may be some distance from the observing location. Also ground stations will view more than one source flux tube. With single point satellite observations individual EMIC source flux tubes are traversed, albeit with a moving frame of observation. Geosynchronous orbit, at the outer edge of the ring current and radiation belt, provides a stable location and is ideal for studying the effects of storms on EMIC waves. The GOES high-resolution magnetometer data with its Nyquist frequency of 1 Hz is somewhat limited in frequency bandwidth for EMIC wave observations. In this situation we typically see unstructured bursty EMIC waves of short duration and may be missing the higher-frequency waves seen at storm times. However, observations show Pc1–Pc2 waves seen on the ground mostly have frequencies <1 Hz during afternoon hours [Fraser, 1968]. Using the Freja satellite at low Earth orbit Bräysy et al.  observed frequencies up to ∼3 Hz, with the higher frequencies occurring at lower latitudes. Interestingly, Bräysy et al.  did not appear to observe main phase or recovery phase waves above 65° CGM latitude, but did observe significant EMIC waves during the main phase of the one geomagnetic storm studied. It is important to note that the Freja observations were two components of the electric field and experience with CRRES in the middle magnetosphere suggests that electric field EMIC wave measurements show a better signal-to-noise ratio than magnetic field measurements. Consequently Freja may have had the advantage of seeing more EMIC waves than the equivalent magnetic field measurements would show.
 Case studies of two storms in October 1999 and January 2000 seen by GOES at geosynchronous orbit and near the equatorial source region of EMIC waves show wave activity during the storm main phase. In both storms the waves occurring in the quieter periods prior to storm onset are narrower banded than those seen during the storm. Higher wave power is observed during the recovery phase of the October 1999 storm and in the main phase of the January 2000 storm. The statistics plotted in Figures 9 and 10 show wave power in the prestorm phase is similar to that seen in the main and recovery phases. For the 21 October storm GOES 8 sees waves at storm onset under quiet field conditions at dusk while GOES 10, earlier in local time, sees broader band waves under more disturbed conditions some ∼22 min later. This can be explained by 44–48 keV protons convecting toward noon. During both storm main phases EMIC waves showed modulation in the 1300–1600 LT afternoon sector associated with long period Pc5 wave activity. The Pc5 waves are similar to the mixed mode waves observed by Greenstadt et al.  and associated with EMIC waves by Fraser et al.  and more recently studied by Loto'aniu et al. .
Figure 9 shows EMIC wave power spectral densities may exceed 500 nT2 Hz−1 during the main and early recovery storm phases. In particular, three storms showed these large EMIC wave powers: 15 November 1998, a CME driven storm; 23 October 1999, CME/CIR driven; and 5 February 2000, CIR driven. The 23 October waves with amplitude 8 and 10 nT on Hn and He components, respectively, occurred at 0312 UT on GOES 10 and are seen in the early recovery phase as the small dots above and below the He+ cyclotron frequency in Figure 1. The 5 February waves (not shown) are seen at the commencement of the main phase following onset as Dst was passing through zero on its way to a minimum of −34 nT, 2.5 h later. The largest amplitude waves of 16 nT were seen in the He+ band over 0.15–0.55 Hz between fO+ and fHe+. Some energy was also seen in the H+ band above the nonpropagation He+ stop band, over 0.6–0.8 Hz. Over the full 22 storm data set plotted in Figure 8, the He+ band showed power spectral densities in the range 0.2–500 nT2 Hz−1 with one at 1000 nT2 Hz−1, over the prestorm, main and recovery phases, within one day either side of storm onset. There was no obvious statistical enhancement of wave power at onset or during the main phase, although the 5 February 2000 storm did show large amplitude waves at onset. The uniformity of wave power over all phases is unexpected and a new result. However, He+ band waves with higher average power were observed during the main phase. In contrast very few H+ band waves were seen with power above 200 nT2 Hz−1 and at all times ranged over 0.1–200 nT2 Hz−1.
 The amplitude of EMIC waves is important in determining electron pitch angle diffusion by cyclotron resonance interaction with EMIC waves as a cause of outer radiation belt electron loss through precipitation [Summers and Thorne, 2003; Loto'aniu et al., 2006]. Using quasi-linear theory and a Gaussian wave power spectrum Summers and Thorne  derived a functional form for pitch angle diffusion where the coefficient Dαα was proportional to the parameter R where R = δB2/Bo2, the ratio of the magnetic field power integrated over the signal bandwidth to the ambient magnetic power. For a large amplitude wave packet occurring during a solar wind compression at onset of the 5 February 2000 storm δB = 3.5 nT (rms) was measured from a 3 min duration wave packet in the 0.17–0.31 Hz band. The rms amplitude was estimated by fitting the observed spectrum to a three-point Gaussian and summing power over the FWHM spectral width, then multiplying by the frequency resolution before taking the square root. This amplitude is larger than the maximum δB = 2.5 nT (rms) observed in a higher-frequency event by Loto'aniu et al.  and suggests that the GOES derived diffusion coefficients may be larger than those from CRRES data at roughly the same radial location.
 An important result is the absence of waves in the H+ band during the main phase and a preference for the He+ band. This may be explained by theoretical considerations associated with simulations of EMIC wave growth during storm conditions. Thorne and Horne  using the HOTRAY code have shown that pronounced intensification of EMIC wave path integrated gain (up to 100 dB) in the He+ band occurs for locations near the edge of the plasmapause, where the waves are guided by the negative density gradient and thus remain essentially field aligned over a growth region of ±10o near the equator. This mechanism may also apply to EMIC waves generated within enhanced density structured plumes at geosynchronous orbit [Chen et al., 2009]. If the injected ion ring current population is anisotropic enough, enhanced convective growth can also occur in the He+ band [Horne and Thorne, 1993, 1997]. This tends to be more effective at higher L (>7) and so may relate to geosynchronous orbit. However, these waves are more likely to be related to substorm injection than storms [Horne and Thorne, 1993].
 Results from the statistical study shows 9 storms were generated by coronal mass ejections (CME), 10 by corotating interaction regions (CIR), and 3 were otherwise defined (Table 2). Figure 10 shows a superposed epoch analysis of the distribution of peak EMIC wave power prestorm and poststorm with respect to CME and CIR driven storms. On separating out these storms no significant differences were found in the EMIC wave occurrences with respect to storm onset, local time or the distribution of peak wave power, over ± 3 days. However, CME generated storms did show a tendency to higher average wave spectral density during the main phase (70–500 nT2 Hz−1) than CIR generated storms (6–500 nT2 Hz−1). This may be associated with the more intense CME driven storms compared to the slower buildup with CIR driven storms, and associated injection and response of the ring current [Borovsky and Denton, 2006]. Looking for statistically significant (>95% confidence) differences using a two-sided Kolmogorov-Smirnov test has suggested that the local time occurrences of EMIC waves were different in the He+ and H+ bands for all storm phases, while waves in the main phase were observed mostly in the He+ band. In CME and CIR driven storms the distributions of normalized wave frequency in the main phase were different, while peak power was higher in the recovery phase for CIR driven storms. Although we have a small sample of storms we consider these results significant at better than the 95% confidence level.
 The relationship of EMIC waves to plumes during storms is somewhat elusive at this time and has not been studied in detail. MacDonald et al.  have suggested that EMIC waves observed in the quiet times in the days prior to storm onset may be related to the presence of high-density plumes providing favorable conditions for wave-particle interaction instability. This appears to be supported by the present study which shows that EMIC wave event occurrence and powers in the days prior to storm onset are comparable to those during the storm main phase (Figures 9 and 10). The case study of the 21–25 October storm supports this scenario with the LANL MPA data showing enhanced plasma densities over 21–22 October prior to storm onset which narrow from a broad structure over 0800–2100 LT to a narrower 1100–1500 LT structure at storm onset (Figure 11). After the main phase minimum plasma densities are low and waves are not seen again until just before midnight on 23 October. These may be associated with a narrow plume seen at the same local time, about 7 h later. In summary, there may be a relationship between storm phase and drainage plume occurrence that determines the occurrence of EMIC waves. It may be that waves are not seen at geosynchronous orbit during the early recovery phase due to inward motion of the plasmapause and waves are not seen again until later in recovery when the plasmasphere reinflates or plumes are formed.
 In recent years it has been shown that electromagnetic ion cyclotron (EMIC) waves play an important role in magnetosphere dynamics, including contributions to ring current ion losses and radiation belt electron losses [Spasojević et al., 2003; Loto'aniu et al., 2006]. Theoretical studies suggest that some of these processes may be most effective during the main phase of geomagnetic storms [Jordanova et al., 1997]. However, ground-based signatures of EMIC waves, Pc1–Pc2 geomagnetic pulsations, are rarely observed during the main and early recovery phases, and more frequently occurred later in the recovery phase [Engebretson et al., 2008a, 2008b]. The current study used geosynchronous orbit in situ data, to investigate the association of EMIC waves with the various storm phases in two case studies and a statistical study of 22 geomagnetic storms over 1996–2003 with Dst of −30 nT or less, from a data set of 30 storms. High-resolution EMIC wave data in the 0–1 Hz band over approximately ±3 days with respect to storm onset, defined as commencement of the negative excursion of Dst, were provided by the GOES 8 and 10 geosynchronous satellite magnetometers. Thirteen of the 22 storms analyzed showed EMIC waves occurring during the main phase. In case studies of two storms waves were seen with higher intensity in the main phase in one and in the recovery phase of the other, with power spectral densities up to 500 nT2 Hz−1.
 Statistics show similar wave power in the prestorm, main and recovery phases. During the main phase EMIC wave power was greatest in the He+ band, confirming a result predicted by theory. A superposed epoch analysis on the 22 storm data set supported these results with 78% of wave events occurring in the He+ band during the storm main phase. Following storm onset the main phase contributed only 29% of events compared to 71% during recovery phase, up to 3 days following onset. Approximately 8% more events occurred in the 3 days before compared with after storm onset. Some differences between storms were found to be dependent on whether the solar wind was driven by a CME or CIR storm. CME storms showed higher power during the main phase while CIR storms showed higher power in the recovery phase. A case study showed that enhanced density plasma drainage plumes in addition to contributing to enhancing EMIC wave activity outside the plasmapause at geosynchronous orbit prior to storm onset, may also influence main phase EMIC occurrence. This is consistent with the study of Morley et al. .
 In the future it will be worthwhile considering the relationship between EMIC wave and plasma drainage plumes with respect to plume evolution at storm and quiet times. A combined study of the GOES magnetometer and corresponding LANL plasma density geosynchronous data sets would be an ideal way to improve our understanding of the generation of EMIC waves under all geomagnetic conditions.
 This research was supported by Australian Research Council Discovery Project grants DP0772504 and DP0663643 and Linkage International grant LX0882515. Infrastructure support has been provided by the University of Newcastle and the Space Weather Prediction Center, NOAA, Boulder, CO. Richard Thorne (UCLA) is thanked for useful discussions.
 Zuyin Pu thanks Mark Engebretson and another reviewer for their assistance in evaluating this paper.