Climatology of rapid geomagnetic variations at high latitudes over two solar cycles

We investigate the characteristics of rapid geomagnetic variations at high latitudes based on the occurrence of large time derivatives of the horizontal magnetic field (dH/dt exceeding 1 nT s −1). Analysis of IMAGE magnetometer data from North Europe in 1983–2010, covering more than two solar cycles, confirms and specifies several previous findings. We show that dH/dt activity is high around the midnight and early morning hours, and nearly vanishes at noon and early afternoon. This happens during all seasons, although the midnight maximum is nearly invisible during summer. As indicated by modelled ionospheric equivalent currents, large dH/dt values occur predominantly during westward ionospheric electrojets. Before and around midnight,dH/dt tends to be north-south oriented, whereas in the morning hours, its direction is more west-east directed. dH/dt tends to be more strictly north-south oriented during winter than other seasons. The seasonal occurrence of large dH/dt values is similar to the variation of the maximum amplitude of westward equivalent currents. The yearly fraction of east-west directed large dH/dt vectors at the Kilpisj̈ arvi station (MLAT 65.88) varies from 31 to 47 % without any clear correlation with the general geomagnetic activity nor with the yearly averages of solar wind parameters.


Introduction
Rapid geomagnetic variations have practical significance, since they drive geomagnetically induced currents (GIC) in technological conductor systems (e.g.Watermann, 2007).In most reported cases, the time derivative of the horizontal Correspondence to: A. Viljanen (ari.viljanen@fmi.fi)magnetic field (dH /dt) has a high correlation with GIC.This observation follows from the fact that the primary cause of GIC, the geoelectric field, is related to the geomagnetic field via the surface impedance, which acts as a low-pass filter for dH /dt.Although modelling of GIC for past events is fairly straightforward (e.g.Pirjola, 2002;Viljanen et al., 2004), forecasting GIC based on physical models from the solar wind down to the ground is still highly challenging (e.g.Pulkkinen et al., 2010).Consequently, analysis of past events will probably provide guidelines for improving forecast methods.
Large GIC events occur most frequently at high latitudes in the vicinity of auroral electrojets and other dynamic ionospheric current systems.Viljanen et al. (2001) showed by using the IMAGE magnetometer data from North Europe in 1982-2001 that large dH /dt (exceeding 1 nT s −1 ) primarily occur during night-time events governed by westward ionospheric currents.However, the directional distributions of dH /dt are much more scattered than those of the simultaneous baseline subtracted horizontal variation field vector H . Another study restricted to 833 selected substorms in 1997 and 1999, before a solar maximum, showed a similar behaviour (Viljanen et al., 2006).Turnbull et al. (2009) performed a corresponding investigation for 553 substorms at mid latitudes in 2000-2003 just after a solar maximum.They also found that large dH /dt are generally connected to westward ionospheric currents, but are also affected by smallerscale ionospheric structures.
Using IMAGE magnetometer data of 2002-2003, Pulkkinen et al. (2006) ) studied spatiotemporal scaling properties of horizontal magnetic field fluctuations in terms of structure function analysis.One result relevant to the present investigation was that the spatial symmetry of field fluctuations increases during substorms, indicating the presence of spatially less ordered ionospheric currents.In distribution patterns of dH /dt, this appears as an increasing scatter of directions.In this paper, we extend the timeseries used by Viljanen et al. (2001) to ground magnetic field data of 1983-2010 covering more than two solar cycles.Since large dH /dt events typically occur during substorms, we concentrate on analysing data from the high latitude Kilpisjärvi (KIL) station (MLAT 65.88), supplemented by four other nearby sites with a good data coverage.We also determine ionospheric equivalent currents across a long meridional chain to compare quantitatively main similarities and differences between the characteristics of electrojets and high dH /dt activity.

Results
We start by updating some parts of the analysis of Viljanen et al. (2001) with a more extensive IMAGE magnetometer dataset of 1983-2010 covering completely the solar cycles 22 (1986-1996) and 23 (1997-2008/2009).We will concentrate on results from a selected set of five stations located in the auroral region (Table 1).
We will use the geographic components of the ground magnetic field B = (X,Y,Z), where X points to the north, Y to the east and Z vertically downwards.The vector H = (X,Y ) denotes the horizontal part of B. For the interplanetary magnetic field (IMF), we use the notation IMF B = (B x ,B y ,B z ).We use the magnetic local time (MLT).
The basic dataset consists of all timesteps in 1983-2010 fullfilling the condition that the magnitude of dH /dt exceeds 60 nT min −1 (1 nT s −1 ).We use 20-s values similarly to Viljanen et al. (2001).This is the best resolution available for the whole period under study.The time derivatives are calculated as simple differences between successive values divided by 20 s.The quiet time baseline is subtracted from the horizontal field H , so it represents the field due to ionospheric currents, with some contribution by the induced currents in the earth.Concerning GIC effects, the origin of the field at the earth's surface is not significant, but only its total value matters.For more details about the effect of induction on the ground magnetic field, see Viljanen et al. (2001) and Tanskanen et al. (2001).

Extended analysis of dH /dt
The most pronounced difference between the directional distributions of H and dH /dt takes place at about 02:00-07:00 MLT when a large number of dH /dt vectors tend to point to the (geomagnetic) east-west direction, whereas H points strictly to the south (Fig. 1).It is evident that a westward current must flow in the ionosphere to produce such H .However, the simultaneous distribution patterns of dH /dt cannot be explained by any simple sheet-type current model, but rapidly changing north-south currents and field-aligned currents must play an important role.A comparison to Fig. 9 in Viljanen et al. (2001) based on years 1994-2001 does not indicate any remarkable change with respect to the larger dataset (Viljanen et al., 2001, showed results from SOR, whose latitudinal separation from KIL is about 1.5 deg to the north).
A rough division of dH /dt directions at KIL is shown in Fig. 2. We use two classes: (magnetic) north-south (NS) and east-west (EW).We measure the direction angle α from the geographic north positive clockwise (east).All dH /dt vectors with −55 • ≤ α ≤ +35 • or 125 • ≤ α ≤ 215 • belong to the NS sector, and the rest belong to the EW sector.The selected ranges take into account that the geomagnetic north points about 10 degrees counterclockwise from the geographic one in this region.As is evident from Fig. 1, the horizontal variation field H lies nearly only in the NS sector (85 % of all values).Its time derivative is more likely in the EW sector during the early morning.Of all large dH /dt vectors, 39 % belong to the EW sector.
The results for KIL are quite representative for the whole IMAGE network as we will show next.We considered a chain of 12 stations from Tartu in the south (MLAT 54.5) to Ny Ålesund in the north (MLAT 75.3) in 1994(MLAT 75.3) in -2009 ( (in this case, with 1-min data).Of large dH /dt, 23-52 % belong to the EW sector, but only 11-23 % of H belong to it.As noted by Viljanen et al. (2001), dH /dt is more sensitive to local ground conductivity anomalies than H .This is especially visible at station Masi (MLAT 66.2), where 52 % of dH /dt belong to the EW sector.The smallest occurrence of dH /dt in the EW sector takes place in the subauroral region in the southernmost part of IMAGE.
Directional distributions of dH /dt have prominent seasonal variations (Fig. 3), as already found by Viljanen et al. (2001).Especially in winter, dH /dt is more concentrated to the north-south direction than during other seasons.Autumn and spring are very similar to each other.Summer has the most scattered distribution.The occurrence probability of large dH /dt is 1.46 % (spring), 0.82 % (summer), 1.31 % (autumn), 1.02 % (winter).Tanskanen (2009) reports a corresponding seasonal dependence in the peak amplitude of substorms.On the other hand, the number of substorms has its maximum in winter.
As a new result, we present the diurnal occurrence of large dH /dt during different seasons (Fig. 4).Spring and autumn are again very similar to each other: the midnight and morning maxima are equally large.The difference between winter and summer is clear: it is most likely to have large dH /dt around midnight in winter, whereas the morning time is strongly preferred in summer.This is obviously related to the large scatter of directions in summer (Fig. 3), since pulsations are common in the morning (Fig. 1).
The year-to-year variation of the dH /dt distribution and its diurnal occurrence at KIL are illustrated in Fig. 5).In ten years, the midnight maximum is slightly higher (1983-1985, 1987-1988, 1996, 1998-1999, 2004, 2009).For the rest 18 years, the morning maximum is more pronounced.This behaviour does not seem to depend on the activity level as measured by the yearly number of large dH /dt.
As a side note we remark that year 2009 was exceptionally quiet with only 1736 dH /dt values exceeding 1 nT s −1 at KIL (Fig. 5).The apparently large scatter in the directional distribution in 2009 may be occasional due to the small number of data points.The low activity in 2008-2009 is clearly related to the general decrease of geomagnetic activity during the somewhat diffuse transition from the solar cycle 23 to cycle 24 (cf.Minamoto and Taguchi, 2009).A clear increase of activity is evident in 2010, although it still remains at a very low level compared to the previous years 1983-2008.
The yearly fraction of EW directed large dH /dt vectors at KIL varies between about 31 and 47 % (Fig. 6).It has some correlation with the solar wind magnetic field.Interestingly, correlation with IMF B is high in 1984-1994, after which it clearly decreases.The same behaviour is found for the alpha particle to proton ratio in the solar wind (plot not shown).Density and dynamic pressure are quite unrelated to the scatter of dH /dt.
The solar wind velocity and the dH /dt scatter anticorrelate with each other at a medium level.Years 1994 and 2003 with a very large number of high-speed streams (Table 2) coincide with a small scatter of dH /dt.On the other hand, the smallest scatter occurs in 2007 without any high-speed streams and with quite a typical yearly average velocity.
We also remark that the scatter of dH /dt distributions is smallest close to the minimum of a solar cycle, although not necessarily at the exact minimum year.Especially, the minimum in 1994 occurs two years before the solar cycle minimum, and year 1994 was geomagnetically highly active (see Fig. 5).The previous minimum year of the scatter (1985) was much more quiet.This is even more prominent for 2007.

Auroral electrojets and dH /dt
As already noted, the westward electrojet is clearly related to the majority of the large dH /dt values.For a more quantitative analysis, we determined ionospheric equivalent currents for the period of 1994-2010 when the IMAGE magnetometer stations cover the whole auroral region.We applied the method of spherical elementary current systems (Amm and Viljanen, 1999;Pulkkinen et al., 2003) to calculate the (geographic) eastward current density along the meridian 22.1 • E with a 0.60 deg latitudinal step from 59.0 to 79.4 • N, assuming that the current flow at the altitude of 100 km.We performed the calculation for each day with 1-min data.
As the upper panel of Fig. 7 shows, an eastward electrojet dominates the afternoon hours at about 13:00-21:00 MLT, and a westward current dominates other times having its maximum at about 01:30 MLT.Comparison of this result to the diurnal occurrence of large dH /dt (Fig. 4) shows that there is a general agreement with a high nighttime activity.However, the distinct morning maximum of large dH /dt values around 05:00-09:00 MLT does not have a counterpart with the electrojets.Similarly, there is no high occurrence rate of large dH /dt around the maximum time of the eastward electrojet.
We also determined the daily maximum amplitude of the eastward and westward electrojets, and from these values  1 during the solar cycles 22-23 (1986-2009).Autumn and spring refer to about 90 days centred at equinoxes, and summer and winter are similarly centred at solstices.
Table 2. Yearly number of high-speed streams (HSS) whose velocity exceeded 700 km s −1 and which lasted longer than 6 h (extended from Tanskanen et al., 2005).Year 1993Year 1994Year 1995Year 1996Year 1997Year 1998Year 1999Year 2000  calculated monthly averages (lower panel of Fig. 7).The westward electrojet has a clear seasonal variation with spring and autumn maximum.The eastward electrojet in turn has a single maximum in summer.This behaviour is expected, since a corresponding seasonal variation is known for the global AU and AL indices (Ahn et al., 2000).Since large dH /dt values are mostly related to westward currents, their occurrence has a similar seasonal variation.

Summary and discussion
This paper confirms and specifies several previous findings of the behaviour of large dH /dt (exceeding 1 nT s −1 ) at high latitudes: 1. dH /dt activity is high around the midnight and early morning hours, and nearly vanishes at noon and early afternoon.This behaviour takes place during all  The basic feature of the diurnal behaviour of large dH /dt is evidently well-understood.Before and around midnight, a large-scale westward electrojet is dominating.It produces a southward H and a northward or southward dH /dt depending on whether the amplitude of the electrojet is decreasing or increasing.In the early morning, there is still a strong westward electrojet in the background, but also smaller-scale but rapidly changing structures such as vortices obviously related to geomagnetic pulsations.In such conditions, dH /dt can easily have any orientation.
Although the main characteristics of large dH /dt seem to prevail from one solar cycle to another, we still lack a deeper understanding of their origin.It is obvious that a physical explanation will require consideration of the whole geospace environment from the solar wind through the magnetosphere to the ionosphere.The relative importance of each region as Fig. 5. Upper part: Directional distributions of dH /dt exceeding 1 nT s −1 at KIL.The number of values is given below each yearly plot (due to occasional data gaps, these numbers are not completely comparable).Yearly diagrams are arranged so that the approximate start of each solar cycle is on the left-hand side.Lower part: The normalised diurnal occurrence of large dH /dt values as a function of MLT (00:00-24:00).well as of the related phenomena is unknown.We list and discuss here some specific items: 1. Concerning the solar wind and magnetospheric background, we know that coronal mass ejections (CME) are the major cause of large GIC, or practically equivalently of large dH /dt (Borovsky and Denton, 2006;Huttunen et al., 2008).Huttunen et al. (2008) examined the effectiveness of different structures embedded in a CME (sheath regions, ejecta, and boundary layers) in causing large GIC.When a CME interacts with the magnetosphere, the most intense GIC activity is likely to take place during the passage of the turbulent sheath region.The effectiveness of sheath regions in driving large GIC is possibly due to their capability to trigger substorms and to drive intense directly driven ionospheric activity.Ejecta-associated GIC seem to require an on-going magnetospheric Dst storm, while sheath regions and boundary layers can cause large GIC even when no activity is taking place in terms of Dst.
2. Kataoka and Pulkkinen (2008) showed that GIC amplitudes in the Finnish natural gas pipeline are relatively small during corotating interacting region (CIR) storms in comparison to CME storms.However, Pc3-5 pulsation activity during CIR storms drives long-lasting GIC in the local prenoon sector.Despite of the differences between CME and CIR storms, the maximum hourly value of dH /dt is always an excellent indicator for the maximum hourly amplitude of GIC in the Finnish pipeline for any local time and any storm phase of CME/CIR storms.Although Kataoka and Pulkkinen (2008) did not study auroral latitudes, we may speculate a similar behaviour there.Yagova et al. (2010) consider spectral parameters of high latitude geomagnetic disturbances in the Pc5/Pi3 frequency range (periods of about 250-1000 s).Their Fig. 3 shows a nighttime maximum of the spectral power at CGM latitudes 65-67 partly overlapping the region of the five stations in Table 1.Furthermore, their Fig.7 indicates an enhanced occurrence of high frequency variations in morning hours (06:00-12:00 MLT) at the same time when large dH /dt occur frequently (cf.our Fig.4).Since large dH /dt values are obviously related to high frequencies, the results by Yagova et al. (2010) at least qualitatively support our findings from a different viewpoint.
3. Weigel et al. (2003) studied separately different solar wind parameters to estimate how much influence they have on explaining variations of the northward magnetic field and its time derivative (X and dX/dt) in the region of IMAGE magnetometers.They found that the process that drives X is different from that which drives dX/dt.For X, the solar wind velocity v and IMF B z are the most important drivers when used as model inputs: they explain most of the variance of the X time series.For dX/dt, the form of the coupling is sensitive to the local time, which is evident also from our Fig. 1 showing the large variability of the preferred orientation of dH /dt.At 10:00 MLT, v has the greatest influence and IMF B z has nearly no influence on the prediction efficiency.However, in the postmidnight sector IMF B z explains a substantial portion of the dX/dt variance while v has a much smaller role.The solar wind density was found to have an insignificant effect in agreement with our Fig. 6.
4. During the most active high-speed stream years (1994,2003), a high solar wind velocity v is related to a less scattered dH /dt.We could intuitively expect more scatter during more disturbed periods when ionospheric currents might have a more complicated structure.However, there seems to be no such simple straightforward relationship.Furthermore, the smallest scatter occurs during a quiet year in 2007 with a typical yearly mean of v.
5. Large dH /dt values occur in the nighttime, so they are evidently closely connected to substorm activity.However, there are seasonal differences in the diurnal occurrence of large dH /dt, especially when summer is compared to other seasons.Compared to the substorm occurrence, their number and intensity in summer is smaller than at other times, but on the other hand they last longer (Tanskanen, 2009).An initial study shows that summer substorms start 0.5-1 h later, but this does not explain the lack of dH /dt maximum at midnight.

Fig. 3 .
Fig.3.Seasonal directional distributions of dH /dt exceeding 1 nT s −1 at the five stations in Table1during the solar cycles22-23 (1986- 2009).Autumn and spring refer to about 90 days centred at equinoxes, and summer and winter are similarly centred at solstices.

Fig. 4 .
Fig. 4. Average diurnal occurrence of dH /dt exceeding 1 nT s − at five stations in northern Europe (Table 1) over the solar cycles 22-23 in 1986-2009 during different seasons.

Fig. 7 .
Fig. 7. Characteristics of ionospheric equivalent currents in 1994-2010 across the geographic meridian 22.1 • E in the geographic latitude range of 59.0-79.4• N. Upper panel: Average diurnal variation of the total eastward (red) and westward (blue) currents.Lower panel: Monthly averages of the daily maximum amplitudes.

Table 1 .
IMAGE magnetometer stations used in this study, and their geographic and corrected geomagnetic coordinates (CGM, year 2001).

Table 1
4. The yearly fraction of east-west directed large dH /dt vectors at the Kilpisjärvi station (MLAT 65.88) varies from 31 to 47 % without any clear correlation with the general geomagnetic activity.It has no evident simple connection to the yearly averages of solar wind parameters either.