“ LATITUDINAL DEPENDENCE OF GEOMAGNETICALLY INDUCED CURRENTS DURING GEOMAGNETIC STORMS „

Major geomagnetic storms drive rapid intensification and variability of magnetospheric and ionospheric current systems that give rise to large geomagnetically induced currents (GIC). Space weather associated GIC pose a serious threat to the reliability of power−transmission systems and other electrically conducting infrastructure such as oil and gas pipelines. The most severe effects are observed at high lati− tudes due to ionospheric currents associated with the aurora. However, as power transmission grid and pipeline infrastructure continues to grow at middle and low−latitudes, GIC hazards are no longer just concerns of high−latitude regions. We investigate how GIC ampli− tude varies in latitude during six major geomagnetic storms that occurred between 1989 and 2004. Due to limited direct GIC measure− ments, a proxy of the geoelectric field is used, i.e. the GIC index. This is calculated for the selected geomagnetic storms using 25 magnetic observatories relatively uniformly distributed in geomagnetic latitude, 14 magnetic observatories with longitudes varying within a range of 45 degrees as well as for the 7 November 2004 storm using 104 observatories. In addition, we suggest a possible way to follow the lat− itudinal displacement of the auroral oval during geomagnetic storms through the maximum value of GIC index, estimated over 2−hour intervals on a wide number of magnetic observatories. province was submerged in darkness for more than nine hours. This episode has taught a lot about the po− tential vulnerability of the most critical ground−based infrastructures to space weather events and, since then, a lot of scientific efforts have been devoted to gain a better comprehension of this phenomenon. Since the origin of GIC are the electric currents flow− ing overhead, the nearer and stronger the currents, the more dramatic their effects. It follows that currents flowing in the ionosphere at polar latitudes (e.g. auro− ral electrojects) are among the most important sources of GIC. These currents flow about a hundred kilometres away from the Earth’s surface and, during geomagnet− ically disturbed periods, can reach intensities up to 4− 5 times their quiet time values, thus giving rise to wide and rapid geomagnetic field variations going from hun− dreds to thousands nanoTeslas [Smith et al., 2017]. Moreover, under deeply perturbed conditions the posi− tion of the auroral oval can consistently move towards lower latitudes posing a GIC risk also for middle lati− tudes countries. There are other possible GIC sources that cannot be neglected [Pulkkinen et al., 2012; Ngwira et al. 2013; Carter et al., 2015], i.e. the equatorial electro− jet (EEJ) and ring current. Despite both these currents produce, for different reasons, signatures on the geo− magnetic field that are less evident than those produced by the polar current systems, there are many proofs of their effects at middle and low latitudes. Just to men− tion a few examples: the damages to some South African transformers occurred during the October 2003 geo− magnetic storm [Gaunt and Coetzee, 2007], those on transformers of a power line system in Hokkaido [Watari et al., 2009] or on the New Zealand's South Island power network during the moderate storm of 7 Novem− ber 2001 [Marshall et al., 2012]. The mechanisms for the generation of GIC are still unclear, especially concerning middle and low latitudes and diverse hypotheses have been proposed. A dominant idea is that the occurrence of the most intense GIC re− lates to sudden impulses or sudden storm commence− ments sometimes preceding a geomagnetic storm [e.g. Fiori et al., 2014; Carter et al., 2016]. These impulses are produced by the increase of magnetopause currents fol− lowing the compression of the magnetosphere by the plasma expelled from the Sun during solar phenomena as coronal mass ejections or corotating interaction re− gions [Adebesin et al., 2016]. Other authors have found that at middle latitudes large voltages can be also ob− served during the recovery phase of geomagnetic storms and are due to Pc5 pulsations [Hejda and Bochníček, 2005]. Besides the strength of the sources producing geomagnetic field variations, an important role is played also by ground conductivity that can produce very high local intensifications of GIC, thus complicating the un− derstanding of this phenomenon. For instance, the am− plification of electric currents that led to the dramatic blackout occurred in 1989 in Canada was due also to Québec's position on a large low−conductivity rock shield that prevented the current from flowing through the earth [Lotko, 2017], under this circumstance the electric current finds a less resistant path along the power lines. During the last decades, the increasing dependence on technology of society has renewed the interest in the investigation of the mechanisms and phenomenology of GIC that are undoubtedly a source of vulnerability to man−made infrastructures due to Earth−directed space weather events. It has been estimated that in the United States and Canada the vulnerability of critical infras− tructures to space weather impacts was almost tripled in only twenty years [Molinski, 2002]. Also equatorial countries are becoming increasingly susceptible to space weather impacts as a consequence of the effects that in− tensifications of the stormtime EEJ [Ngwira et al., 2013] can have on the rapidly developing transnational power grids [Moldwin and Tsu, 2016]. The characterization of geomagnetic latitudes based on different risk levels associated with space weather impacts is among one of the open issues in this field. Such a clas− sification could help countries in taking appropriate pro− tection actions. Pulkkinen et al. [2012] investigated how the geoelectric field, computed using geomagnetic field data and ground conductivity estimates, changed with geomagnetic latitude during the storms of March 1989 and October 2003. They showed that the geoelectric field magnitudes may go through a steep drop at geomagnetic latitudes of about 40°–60°and suggested that further analyses were required to confirm the existence and lo− cation of a possible “latitude threshold boundary”, below which the risk associated with GIC can be considered negligible. Using an extended set of geomagnetic storms Ngwira et al. [2013] reconstructed the latitudinal profile of the maximum GIC amplitude and found that most in− tense GIC are due to auroral electrojets but that the EEJ, too, is responsible for the enhancement of GIC intensity. Ngwira et al. [2013] used, as a proxy for GIC intensity, the amplitude of the geoelectric field estimated by means of the plane wave method [Pirjola, 1982] and the Québec ground conductivity model. In this paper we reconstruct the latitudinal profile of the maximum GIC amplitude by means of the so−called GIC index [Marshall et al., 2010, 2011] as an alternative to the methods previously used. We first select two sets of magnetic observatories, approximately uniformly TOZZI ET AL.


INTRODUCTION
Large−scale time−varying currents flowing in the ionosphere and magnetosphere are responsible for the origin of electric and magnetic fields at the Earth's sur− face. Specifically, electric currents are induced in the conductive ground when particularly intense and rapid variations occur in the ionospheric and magnetospheric current systems due to perturbations of solar origins. These currents, known as geomagnetically induced cur− rents (GIC), can flow through infrastructure networks such as railroads, power transmission lines, and pipelines with damages ranging from the slow degra− dation to the immediate manifestation of ruptures and malfunctioning [Ngwira and Pulkkinen, 2018]. A well− known example of the possible damages inflicted by GIC is the collapse of the Canadian power grid of the entire Hydro−Québec occurred during the geomagnetic storm of October 1989. On that occasion a blackout followed, affecting more than 6 million people and the Québec province was submerged in darkness for more than nine hours. This episode has taught a lot about the po− tential vulnerability of the most critical ground−based infrastructures to space weather events and, since then, a lot of scientific efforts have been devoted to gain a better comprehension of this phenomenon.
Since the origin of GIC are the electric currents flow− ing overhead, the nearer and stronger the currents, the more dramatic their effects. It follows that currents flowing in the ionosphere at polar latitudes (e.g. auro− ral electrojects) are among the most important sources of GIC. These currents flow about a hundred kilometres away from the Earth's surface and, during geomagnet− ically disturbed periods, can reach intensities up to 4− 5 times their quiet time values, thus giving rise to wide and rapid geomagnetic field variations going from hun− dreds to thousands nanoTeslas [Smith et al., 2017]. Moreover, under deeply perturbed conditions the posi− tion of the auroral oval can consistently move towards lower latitudes posing a GIC risk also for middle lati− tudes countries. There are other possible GIC sources that cannot be neglected [Pulkkinen et al., 2012;Ngwira et al. 2013;Carter et al., 2015], i.e. the equatorial electro− jet (EEJ) and ring current. Despite both these currents produce, for different reasons, signatures on the geo− magnetic field that are less evident than those produced by the polar current systems, there are many proofs of their effects at middle and low latitudes. Just to men− tion a few examples: the damages to some South African transformers occurred during the October 2003 geo− magnetic storm [Gaunt and Coetzee, 2007], those on transformers of a power line system in Hokkaido [Watari et al., 2009] or on the New Zealand's South Island power network during the moderate storm of 7 Novem− ber 2001 [Marshall et al., 2012].
The mechanisms for the generation of GIC are still unclear, especially concerning middle and low latitudes and diverse hypotheses have been proposed. A dominant idea is that the occurrence of the most intense GIC re− lates to sudden impulses or sudden storm commence− ments sometimes preceding a geomagnetic storm [e.g. Fiori et al., 2014;Carter et al., 2016]. These impulses are produced by the increase of magnetopause currents fol− lowing the compression of the magnetosphere by the plasma expelled from the Sun during solar phenomena as coronal mass ejections or corotating interaction re− gions [Adebesin et al., 2016]. Other authors have found that at middle latitudes large voltages can be also ob− served during the recovery phase of geomagnetic storms and are due to Pc5 pulsations [Hejda and Bochníček, 2005]. Besides the strength of the sources producing geomagnetic field variations, an important role is played also by ground conductivity that can produce very high local intensifications of GIC, thus complicating the un− derstanding of this phenomenon. For instance, the am− plification of electric currents that led to the dramatic blackout occurred in 1989 in Canada was due also to Québec's position on a large low−conductivity rock shield that prevented the current from flowing through the earth [Lotko, 2017], under this circumstance the electric current finds a less resistant path along the power lines.
During the last decades, the increasing dependence on technology of society has renewed the interest in the investigation of the mechanisms and phenomenology of GIC that are undoubtedly a source of vulnerability to man−made infrastructures due to Earth−directed space weather events. It has been estimated that in the United States and Canada the vulnerability of critical infras− tructures to space weather impacts was almost tripled in only twenty years [Molinski, 2002]. Also equatorial countries are becoming increasingly susceptible to space weather impacts as a consequence of the effects that in− tensifications of the stormtime EEJ [Ngwira et al., 2013] can have on the rapidly developing transnational power grids [Moldwin and Tsu, 2016].
The characterization of geomagnetic latitudes based on different risk levels associated with space weather impacts is among one of the open issues in this field. Such a clas− sification could help countries in taking appropriate pro− tection actions. Pulkkinen et al. [2012] investigated how the geoelectric field, computed using geomagnetic field data and ground conductivity estimates, changed with geomagnetic latitude during the storms of March 1989 and October 2003. They showed that the geoelectric field magnitudes may go through a steep drop at geomagnetic latitudes of about 40°-60°and suggested that further analyses were required to confirm the existence and lo− cation of a possible "latitude threshold boundary", below which the risk associated with GIC can be considered negligible. Using an extended set of geomagnetic storms Ngwira et al. [2013] reconstructed the latitudinal profile of the maximum GIC amplitude and found that most in− tense GIC are due to auroral electrojets but that the EEJ, too, is responsible for the enhancement of GIC intensity. Ngwira et al. [2013] used, as a proxy for GIC intensity, the amplitude of the geoelectric field estimated by means of the plane wave method [Pirjola, 1982] and the Québec ground conductivity model.
In this paper we reconstruct the latitudinal profile of the maximum GIC amplitude by means of the so−called GIC index [Marshall et al., 2010[Marshall et al., , 2011 as an alternative to the methods previously used. We first select two sets of magnetic observatories, approximately uniformly TOZZI ET AL. spaced in geomagnetic latitude, and study how the maximum intensity of GIC changes with latitude dur− ing six intense geomagnetic storms. One set consists of 25 magnetic observatories with longitudes spanning the entire 0°−360° range. The other set consists of 14 magnetic observatories with longitudes varying within 45 degrees. The use of this small set of observations has the purpose to consider the localized in local time na− ture of many of the current systems developing during geomagnetically disturbed conditions. As shown in de− tail by Tsuji et al. [2012], the local time location of the observatory at the universal time (UT) of a storm has a large impact on the amplitude of the measured pertur− bation. As a further analysis we choose the geomagnetic storm of November 2004 and go more in depth using, this time, data from around a hundred magnetic obser− vatories and study the time variation of GIC index lat− itudinal profile with a sampling of two hours.
The paper is organised as follows. After a thorough description of data selection, GIC index is introduced also through a few examples. Then, GIC index is estimated for both the cases of the reduced and extended sets of mag− netic observatories and results described and discussed. Finally, main conclusions are summarised.

DATA
Our analysis is based on the assumption that GIC in− tensity can be well represented by the GIC index intro− duced by Marshall et al. [2010Marshall et al. [ , 2011 and explained in detail in the next section. The choice to use a proxy of the geoelectric field, and hence of GIC intensity, instead of their real measurements, is a consequence of the global character of the study here proposed. Indeed, real GIC measurements are very difficult to be retrieved on a local scale already, even more on a global scale. More− over, even if these data were available, they would very likely be inhomogeneous and patchy. Also, the tempo− ral coverage would leave a lot to be desired. The situa− tion is quite different for GIC index that, as it will be shown in what follows, can be rather straightforwardly estimated by 1−minute values of the Northward (X) and Eastward (Y) Cartesian components of the geomagnetic field collected at ground magnetic observatories. These, in most cases, follow well−defined standards concerning instrumentation, measuring techniques as well as obser− vatory practice in general. INTERMAGNET observatories follow such strict standards [Love and Chulliat, 2013]. INTERMAGNET constitutes a global network of obser− vatories for monitoring the Earth's magnetic field that adopts "modern standard specifications for measuring and recording equipment", as well as for data process− ing, "in order to facilitate data exchanges and the pro− duction of geomagnetic products in close to real time" (citation from www.intermagnet.org).
First of all, we select the ten most intense geomag− netic storms of the last two decades, going further back in time would have meant a lower amount of available geomagnetic data. These storms are listed in Table 1, or− dered by decreasing intensity based on the minimum value reached by Sym-H index [King and Papitashvili, 2005]. Then we look for a set of magnetic observatories, approximately uniformly spaced in geomagnetic latitude, with data available for some of the storms listed in Table  1. The optimal configuration came out to be a set of 25 observatories that, with a few exceptions, had data avail− able for six out of the ten selected storms. Some details on these observatories are listed in Table 2 and their po− sition, in a geomagnetic coordinates grid, is shown in Figure 1 by green circles. Sym-H index [Iyemori, 1990] corresponding to the selected storms is shown in Figure  2 for the four days characterised by the most intense ge− omagnetic activity, over which the analysis has been per− formed. Two of these storms, i.e. those of October 2003 and November 2004, are actually a double and a triple dip storm, respectively [Farrugia et al., 2006]. As men− tioned above, data from the selected magnetic observa− tories are not available for all the selected storms. One case is that of March 1989 geomagnetic storm, the fur− thermost among those chosen, for which data from BEL, SJG, KOU, PHU, GUA, BNG, HUA, ASC, PAF magnetic observatories (most unfortunately covering equatorial latitudes) are not available. The decision to keep this storm anyway derives from the exceptional low value of −720 nT reached by Sym-H index. Other missing data are those from GUA (July 2000 storm), ASC (March 2001 storm), BNG (November 2003 storm) and HBK (Novem− ber 2004 storm). Looking at Table 2 it emerges that the uniformity of spacing in geomagnetic latitude is broken at the equator, where the density of observatories is higher. This stems from the will to well document what happens at equatorial latitudes and from the need to re− place GUA with BNG for July 2000 storm and ASC with HUA for March 2001 storm. Certainly, we could have used all the observatories available for each geomagnetic storm but we preferred to deal always with the same set of samples to better compare the results. Moreover, con− sidering that this set consists of observatories with sparse longitudes, and magnetic local time (MLT) of recorded GIC index maxima are not taken into account, a differ− ent set of observatories for each storm would have com− plicated interpretation of results consistently. Since the results may be biased from storm to storm in terms of an absolute value of GIC index due to the different geo− magnetic longitudes, we display the same analysis per− formed on the set of 25 magnetic observatory also for a set of 14 observatories. These have geomagnetic longi− tudes between about 7°W and 38°E (grey area of Figure  1), namely on a band of around 45 degrees equivalent to a difference of 3 hours in MLT; details are listed in Table  3. The limited number of observatories selected in this case is a direct consequence of magnetic observatories geographical distribution and data availability. Of course if, on one side, such a loose chain of observatories does not allow a well−resolved investigation of the latitudi− nal dependence of GIC intensity, on the other side, it al− lows better discriminating ionospheric and magneto− spheric currents effects based on the MLT position of the observatory. However, even if it were possible to build dense latitudinal chains, the limit to the accurate obser− vation of the latitudinal variation of GIC intensity max− ima would be the rapid variability of ionospheric current systems. For example, the dayside EEJ has a relatively short (2.4 hour) correlation length in longitude [Alken and Maus, 2007], so the largest signature of the EEJ could be missed by the observatories for any given storm.
As already mentioned in the Introduction, a further part of the investigation relies on a much wider set of 104 magnetic observatories, for the study of the November

Date of SSC Time of SSC (UT) Sym-H min [nT]
13   and Y Cartesian components when directly available, or differently calculated them from the horizontal component and declination. Gaps have been linearly interpolated if not occurring during time intervals characterised by the most intense and rapid variations, otherwise the observa− tory has been excluded for the storm under evaluation. As explained in more detail in the next section, linear inter− polation used for gap filling is not expected to change ex− cessively GIC index estimation as far as to alter results.

ESTIMATION OF GIC INDEX
Measurements of GIC intensity are generally made along electric power grids or along pipelines. These, however, are not easily retrievable and do not cover the entire globe. Another way to study GIC is through the geoelectric field induced at the Earth's surface. This can be done both in the time and in the frequency domains [Pirjola, 2002] by assuming a model for Earth's con− ductivity that, in the best case, is a uniformly con− ducting half−space and, in the worst, considers an anisotropic conductivity. The latter requires the intro− duction of tensors, hence ending in the estimation of reasonably complicated formulas. For this reason, sev− eral proxies of the geoelectric field, and consequently of GIC intensity, have been investigated in the years. Of course, resorting to proxies means neglecting very important factors that determine GIC intensity, as the technical specifications of the power grid and ground conductivity. There are several proxies based on geo− magnetic data (e.g. the time derivative of the geo− magnetic field dB/dt, A and K indices, hourly standard deviation of X and Y components, etc.) that can be used as indicative of GIC intensity. All of them some− how represent the way the geomagnetic field varies in time since, as broadly established, the origin of GIC is to be searched in the time rate of geomagnetic field variations [Thomson et al., 2010]. Recently, while look− ing for spectral information on the relationship be− tween the pipe−to−soil potential (PSP) and variations in the horizontal components of the geomagnetic field,   Marshall et al. [2010Marshall et al. [ , 2011 developed a new index, i.e. the GIC index, characterised by a 1−minute sampling rate. For this reason, proxies providing the poorest in− formation on GIC evolution are certainly A and K in− dices being characterised by a sampling rate of three hours. Differently, hourly standard deviation provides results comparable to those obtained with the GIC in− dex but with a lower time resolution, of one hour in− stead of one minute. The comparison between the per− formance of some of the above cited proxies, in particular of dB/dt, with that of GIC index in the rep− resentation of actual measurements of the PSP has shown that this index is the one that best agrees (based on correlation analysis) with real measurements. Val− ues of the correlation coefficient between PSP and GIC index resulted to be higher than those between PSP and dB/dt of at least 0.3−0.4, generally supersed− ing values of correlation coefficient of 0.8. So far, the performance of this index has been tested only at the geomagnetic latitudes of Australia and New Zealand [e.g. Alekseev et al., 2015], i.e. approximately between 20°S and 50°S, therefore its validity at latitudes out− side this range cannot be taken for granted due to the role of local conductivity on GIC intensity. However, due to the not uniform conductivity of Australia and New Zealand themselves we assume that GIC index works well independently of conductivity also outside the 20°S−50°S geomagnetic latitude range. GIC index can be straightforwardly computed from 1−minute values of the X and Y Cartesian components of the geomagnetic field measured at a given obser− vatory according to the following equation: ( 1 ) with Z(f) a filter function defined as ( 2 ) where f is a variable frequency, f N =8.3~mHz is the Nyquist frequency for data sampled at a rate of one value per minute, FTT{} −1 indicates the inverse Fourier transform and || the absolute value. More details on this index can be found in Marshall et al. [2010Marshall et al. [ , 2011Marshall et al. [ , 2012.
Considering the large amount of available geomag− netic data, this index proves to be a powerful tool to perform investigations concerning GIC over large spa− tial scales or wherever real GIC measurements are not available. These are the reasons why we relied on GIC index.
All estimates of GIC x and GIC y indices shown by Marshall et al. [2011] are performed on a single−day of observations (1440 points); before calculating the FFT, the time series are multiplied by a Hanning function in the time domain with the purpose to limit end ef− fects. Here, however, we deal with time series with a length of four days (5600 data points) and the use of the procedure above inevitably produces a 1−day modula− tion of GIC x and GIC y indices, due to the shape of the Hanning function. This function has the disadvantage to excessively damp GIC indices at the borders of the considered time window. For this reason we used a slightly different procedure to calculate GIC indices. First, we removed from the X and Y components their averages estimated over the entire 4−day interval. Then, we considered windows of 1440 points and, to limit end effects and avoid the 1−day modulation, used a mov− ing window approach. In practice, we followed these steps: a) estimate equations 1 and 2, b) record their maximum value at the centre of the window, c) shift the window of 1 point (i.e. 1 minute) and repeat the pro− cedure. Some tests have been performed to evaluate the sensitivity of GIC indices to the width of the window used for FFT estimation. Comparison of GIC indices ob− tained using different widths of the FFT window (1, 3, 6, 12 and 24 hours) has shown their very low sensitiv− ity to the window's width (results not shown).
Based on the way GIC index is defined, it follows that the linear trends used to fill data gaps, when ex− isting, are not expected to amplify it but, in case, to un− derestimate it. In fact, GIC index takes high values where high rates of change in the geomagnetic field are present; this is not the case of intervals covered by a linear trend. Moreover, we recall that when gaps oc− curred during time intervals characterised by rapid and wide variations of the geomagnetic field (i.e. corre− sponding to highly disturbed geomagnetic conditions), data from that observatory have been discarded. Ulti− mately, the underestimation of GIC index due to gap filling is done exclusively during phases of the storm where the index would not have reached its highest values anyway and, as illustrated in what follows, only GIC index maximum values are considered in the anal− ysis. Figure 3 shows an example of GIC index estimation obtained for the X and Y components measured at three observatories at different geomagnetic latitudes (high, middle and low). Figure 3 anticipates some re− sults concerning the intensity of GIC at equatorial lat− itudes that will be discussed in the next section. Here, we just would draw attention to how amplitudes of GIC y index vary from CBB (high latitude) to HUA (equato− rial latitude). We also observe that, being estimated on the X component, GIC y is expected to be generally larger than GIC x because of the different contributions of magnetospheric and ionospheric currents on the two components. Current systems responsible for the strongest external contributions of the geomagnetic field (e.g. ring current, auroral and equatorial electro− jets) exert their action mainly on the meridional plane, thus affecting X component more than Y component. The analysis shown in Figure 3 for CBB, FRN and GUA is performed on both X and Y components mea− sured at the two sets of 25 and 14 observatories, for the selected geomagnetic storms. The same analysis is also performed on the extended set of 104 observatories only for the November 2004 geomagnetic storm. Obtained results are displayed and discussed in the next section.

RESULTS AND DISCUSSION
To obtain the latitudinal profile representing GIC maximum intensity, for each geomagnetic storm, we es− timate GIC x and GIC y indices from data of the selected observatories, and then record their maximum values over the 4−day time interval shown in Figure 2. This procedure is applied following two different approaches: 1) regardless of considerations on the magnetic local time of the observatory (set of 25 magnetic observato− ries) to indirectly measure, through GIC index, the peak of GIC intensity throughout the geomagnetic storm; 2) focusing on a restricted MLT interval (set of 14 magnetic observatories) to obtain a latitudinal profile of GIC in− dices maximum values less biased from storm to storm.
Results corresponding to the set of 25 magnetic obser− vatories are shown in Figure 4 where the single point represents the maximum GIC index estimated at given magnetic observatory during a given geomagnetic storm and the different colours represent the different storms. In this phase of the study, the double and triple dip storms of October 2003 and November 2004, respec− tively, are treated as a single storm. The following main features are recognisable in Figure 4: from the polar cap GIC indices increase and reach their peak values at au− roral oval latitudes, then they gradually decrease mov− ing toward middle and equatorial latitudes where, in some cases, we observe an increase.
The dependence of GIC indices with latitude depend is determined by the location and dynamics of the sources inducing the geoelectric fields. The values found at polar cap latitudes, smaller than those observed at au− roral oval latitudes, can be explained in terms of the lower intensity of the currents flowing there, that is a consequence of the low conductive polar cap iono− sphere [Mc Granaghan et al., 2016]. Moving equator− ward, GIC indices take the highest value over the auro− ral oval due to the intensification of the auroral electrojects and of the field aligned currents; under heavily geomagnetically disturbed conditions they are both characterised by rapid and wide time variations causing analogous variations in the geomagnetic field. Getting far from the auroral oval, a sharp decrease of GIC indices is observed at geomagnetic latitudes of around 50°. These latitudes are less affected by the po− lar ionospheric currents and other sources, as for in− stance the ring current, could play a dominant role. However, the ring current is very far from the Earth's surface and, consequently, its contribution to geomag− netic field variations is about one order of magnitude 9 GICS LATITUDINAL DEPENDENCE less than that of polar ionospheric currents. Finally, Figure 4 evidences an increase of GIC indices at equatorial latitudes, also reported by other authors [Pulkkinen et al., 2012;Nwgira et al., 2013;Carter et al., 2015Carter et al., , 2016de Villiers et al., 2017;Moldwin et al., 2016]. At these latitudes, the EEJ flows in the E−region of the dayside ionosphere (altitudes of around 100 kilometres). This narrow ribbon of current closely fol− lows the geomagnetic equator and corresponds to an eastward electric current. Looking at GIC x index the in− crease is detectable in BNG during the November 2004 geomagnetic storm, the peak occurring at around 00:00 UT on the 9 of November corresponding to an MLT of around 01:30, which makes it difficult to justify an in− terpretation in terms of an increase of the EEJ that flows on the dayside. Moreover, since EEJ flows on the equa− torial plane, its effect is expected to be visible in the X component of the geomagnetic field and not in the Y component, used to estimate GIC x index. Concerning to an MLT of around 08:00. Due to their MLTs, the first two enhancements are well compatible with an increase of GIC y index deriving from an increase of the EEJ that on turn influenced the X component. Although less clear, also the third case can be a consequence of EEJ enhancement, indeed at 8:00 MLT its intensity is cer− tainly lower than that corresponding to 12:00 but it is anyway not negligible. In fact, the maximum value of GIC y index is smaller than those observed in the other two cases. The reason why this enhancement is not ob− served during March 2001 geomagnetic storm relates to fact that the maximum value of GIC y index is reached at 22:00 MLT at HUA, during times when the EEJ is not present overhead. Concerning July 2000 geomagnetic storm, GIC y index maximum is reached at 17:00 MLT, a time when a weak EEJ is still present. In this case, however, the low intensity of the July 2000 geomag− netic storm could have prevented equatorial electrojet from being affected by intense variations and therefore GIC y index from reaching values higher than those observed under less disturbed conditions. At EEJ latitudes, however, also the counter electro− jet (CEJ) can be observed, particularly in the early morning and evening. When CEJ occurs, the daytime EEJ strength is weakened and reverses direction for a short period. It has also been found that EEJ and its CEJ are characterised by a marked longitudinal variability [Rabiu et al., 2017]. This makes the source, i.e. either the EEJ or CEJ, of the equatorial enhancement of maximum GIC indices not obvious. Rabiu et al. [2017] found that CEJ has the greatest % of occurrence in the African sta− tions and that, among the observatories they consid− ered, HUA is the one characterised by the strongest EEJ and the least occurrence of the CEJ. Therefore, the en− hancement of maximum GIC indices we find at HUA can be quite safely ascribed to the EEJ.
Averaging over the six storms the trends plotted in Figure 4, we obtain the latitudinal profiles of GIC indices shown in Figure 5 (purple dotted lines) with associated error bands estimated, for each observatory, as the stan− dard deviation of the six samples. The green line in Figure 5 represents a smoothed and spline−interpolated version of the average. This Figure shows that a ramp begins at geomagnetic latitudes of about −45° and 45°, meaning a high rate of increase of the maximum values of GIC indices and hence an augmented risk of damages due to the building up of geomagnetically induced cur− rents. The latitude threshold boundary found by our analysis seems to be at about 45° − 50° of geomagnetic latitude. The two trends obtained from GIC x and GIC y in− dices are very similar, nevertheless small differences can be observed. Both trends are characterised by different profiles for the Southern and Northern high latitudes. This is probably due to the better sampling of high Northern latitudes than Southern ones, even if it cannot be excluded that the asymmetries between the two po− lar regions provide an effect. GIC y index profile presents a sort of bump (i.e. the effect of the equatorial electro− jet on the X component), not present in GIC x index. However, the general trends obtained (i.e. peak at high− latitudes, dip in mid−latitudes and a smaller peak at EEJ zone) are consistent with the profiles reported, for in− stance, by Pulkkinen et al. [2012] and Ngwira et al. [2013]. Pulkkinen et al. [2012], considering only the two extreme storms of March 1989 and October 2003, found a sudden drop in all the parameters they investigated as geoelectric field proxies, between 40° and 60° of geo− magnetic latitude and an increase at equatorial geo− magnetic latitudes between −5° and 5°. Interestingly, Pulkkinen et al. [2012] observed also a general ten− dency to have a slightly larger amplitude for all the anal− ysed parameters in the Northern Hemisphere than in the Southern one. Ngwira et al. [2013] considered 12 se− vere/extreme geomagnetic storms and used, as a proxy of the geoelectric field, dB/dt. They found the latitude threshold boundary at about 50° − 55° of geomagnetic latitude and attributed this boundary to the movement of the auroral oval. Ngwira et al. [2013] found also that the ground induced geoelectric field at the latitudes of the EEJ can be one order of magnitude larger than outside the EEJ belt and that this enhancement is the conse− quence of the penetration of high latitudes electric field. Concerning North−South asymmetries, their Figure 2 shows a slight difference between the two hemispheres. The capability to capture features of the latitudinal pro− file of GIC intensity found by other authors strengthens the use of GIC index also outside the geomagnetic lati− tudinal range of Australia and New Zealand. Figure 6 is the same as Figure 4 but obtained con− sidering the set of 14 observatories with longitudes within a band of 45 degrees and MLTs within three hours. The top of Figure 6 reports the value of MLT cor− responding to the UT of Sym-H absolute minimum over the 4−day interval. Concerning the latitudinal profile of the maximum intensity of GIC, main features are similar to those shown in Figures 4 and 5, but Fig− ure 6 contains additional information with respect, for instance, to the current systems responsible for the observed enhancement of GIC indices. At high North− ern latitudes, two peaks are clearly visible in the GIC x index at 55° and 75° of geomagnetic latitude. These lat− itudes are compatible with the position, during dis− turbed conditions, of field−aligned currents [Green et al., 2009] that generate the most visible signature in the Y component. GIC indices enhancement at middle and high latitudes reaches the lowest values when the ob− servatories are in the dayside at the moment of the main phase of the storm. This does not hold for equatorial latitudes where the increase of GIC y index is observed almost independently of MLT. These features can be ex− plained in terms of the asymmetries of ionospheric and magnetospheric current systems that, with the ex− ception of the EEJ, are generally more intense in the nightside. This kind of approach demonstrates the im− portance of considering the MLT position of the obser− vatory in the reconstruction of latitude profiles since information that could be useful for GIC related risk protection is provided. Unfortunately, this kind of study is seriously limited by the availability of observations. So far, we have ignored the fact that two of the storm events here considered are actually a double and triple dip storm. Figure 7a shows separately GIC indices latitudinal behaviours for the double dip storm of 7 November 2004 estimated from the X and Y magnetic measurements of 104 ground stations, regardless of MLT. To split this double storm in two, the time inter− val of four days is divided in two periods of 2880 points each. Despite the first storm is more intense in terms of Sym-H index (see Figure 2) than the second, from Figure 7a it seems that noticeable differences be− tween the two storms are observed only at high lati− tudes, with effects stronger during the second storm due to the intense electric currents flowing in the polar re− gions (as can be deduced by AE index reported in Fig−  ure 7b). Looking more carefully at what obtained for the Northern polar latitudes for the second storm, we could interpret the different shape of the profiles of GIC x and GIC y indices in terms of their different sources. The profile of GIC x index around 70° is clearly sharper than that of GIC y index and the double peak visible in Fig−   ure 6 where MLT is considered, is no more visible. Re− calling that GIC x index is estimated from the Y com− ponent, which is affected mainly by electric currents perpendicular to its direction, we could interpret the re− duced band of latitudes where the highest values of GIC x index are observed in terms of the Pedersen cur− rents that cover a small latitudinal band. They flow along meridional planes between the footprints of the region 1 and region 2 field−aligned currents [Prölls, 2004]. Similarly, recalling that GIC y index is estimated from X component we can deduce that the increased band of latitudes where the highest values of GIC y in− dex are observed is that corresponding to the flow of auroral electrojects. The fact that this distinction is not evident during the first storm (right side of Figure 7a) can be due to the less intense auroral activity during the period between 7 November, 00:00 and 8 November, 23:59 as can be verified in Figure 7b displaying AE in− dex. We also observe that the average smoothed trend obtained considering 25 observatories and six geo− magnetic storms well agrees with the trend recon− structed with the 79 additional magnetic observatories and for a single storm. Certainly, further analyses are needed to univocally interpret the latitudinal profile of GIC index in terms of ionospheric current systems.
As a last analysis, we considered the time interval between 16:00 UT Focussing on the Northern Hemisphere where the GIC indices latitudinal profiles are reconstructed with a higher resolution than in the Southern Hemisphere, Figures 8 and 9 show that the profiles at the high lat− itudes well match with the evolution of the geomag− netically disturbed conditions, represented by Sym-H and AE indices. For instance, in the time interval 18− 20 UT it is possible to recognise a sudden increase in Sym-H index that produces, in the same time interval, an increase of GIC indices maxima well visible at high latitudes. Starting from 22 UT we observe a gradual de− crease of Sym-H index, corresponding to an intensifi− cation of the ring current, and an increase of AE index corresponding to an intensification of the auroral cur− rents. With the start of auroral activity, the latitudinal profile grows at latitudes corresponding to those of the auroral oval, showing a widening of the band of lati− tudes most affected by high values of GIC index to− gether with an equatorward shift of the southernmost boundary of this latitude band. Then, as time passes, the band of disturbed latitudes reduces and moves back to its original position. We suppose that this band of lat− itudes could be representative of the latitudes covered by the auroral electrojets. The black line displayed in Figure 8 and 9 seems to be indicative, in this case, of the equatorward boundary of the auroral oval. This be− haviour leads us to speculate that such a use of GIC in− dex could be viewed as a tool to nearly real−time monitor the displacement of the auroral oval during ge− omagnetically perturbed conditions and hence of the latitude threshold boundary. Certainly, the technical feasibility of this idea needs a thorough and dedicated investigation that is not the purpose of this paper.
Once cleared the technical limitations its achievement would depend, anyway, on the availability of real time 1−minute data from magnetic observatories.
To conclude, it is worth spending a few words on two features of Figures 8 and 9. One concerns equatorial lat− itudes. In Figure 8 a signature that could be ascribed to the EEJ is visible both at the commencement and in the main phase of the storm, while in Figure 9 an en− hancement of the maximum GIC y index is seen only at the storm commencement. This can be explained in terms of the different MLT position of the two observa− tories involved, i.e. BNG and HUA. In Figure 8 the en− hancement is seen at BNG that, at 18 UT, was at 19:30 MLT. This enhancement cannot be ascribed to the EEJ due to the MLTs covered by BNG during the storm (i.e. corresponding to evening and early morning) and also to the fact that it is observed in GIC x index (i.e. to the Y component). Differently, the enhancement visible in Figure 9 in GIC y index, and hence in the X component of the geomagnetic field, correctly relates to the EEJ since HUA observatory was at around 13 − 15 MLT dur− ing the storm commencement (between 18 − 20 UT). During the same interval GIC x index, and hence in the Y component of the geomagnetic field, does not show an appreciable increase. The second feature concerns the unexpected peaks visible in Figure 8, practically at all UTs, around −30° of geomagnetic latitude and that we are not able to explain. These data correspond to the ob− servatory of Learmonth (IAGA code: LRM), they have been double−checked but we found no reason to discard them. At 18 UT, LRM was at 1:30 MLT so the observed peak at the commencement and in the main phase cor− responds to MLTs in the night and morning sectors.

CONCLUSIONS
The purpose of this study is the investigation of the impact of geomagnetically induced currents with lati− tude. Although similar researches have been performed in the past [e.g. Pulkkinen et al., 2012;Ngwira et al., 2013], here we propose an analysis based on a differ− ent proxy of the geoelectric field. Instead of using dB/dt, we estimated the GIC index [Marshall et al., 2010[Marshall et al., , 2011 15 GICS LATITUDINAL DEPENDENCE that has proved to correlate with actual measurement of the pipe−to−soil potential better than dB/dt at geomag− netic latitudes between 20°S−50°S. We reconstructed a latitudinal profile of maximum GIC indices values us− ing two sets of magnetic observatories, with different geographical distribution, for six geomagnetic storms. Then we went more in depth using a larger set of mag− netic observatories (104) and investigated what hap− pened during a single geomagnetic storm, that of 7 November 2004. For such storm, we investigated also the time variations of the latitudinal profile with a sampling rate of 2 hours. Findings can be summarised as follows: 1) The enhancement of the EEJ is recognisable in GIC y index; even though values are low it can anyway pose a risk of GIC damages to countries passing across an MLT from 8:00 to 18:00 during a geomagnetic storm; 2) The latitudinal profile of GIC intensity does not vary sensibly from storm to storm and the latitude threshold boundary can be set at about 45°−50° of geomagnetic latitude; 3) Through GIC index latitudinal profile it could be possible to distinguish the sources producing def− inite features; 4) Through the middle and high latitude portion of GIC index latitudinal profile it could be possible to monitor the displacement of the auroral oval dur− ing geomagnetically disturbed conditions; 5) Results support the use of GIC index at all lati− tudes.
An accurate definition of the way maximum GIC in− tensity varies with latitude plays an important role es− pecially at low and middle latitudes that are generally, and erroneously, considered safe from damages related to space weather events. Concerning middle latitudes, although it is true that the bulk of effects are produced by the magnetopause currents there, it should not be ne− glected that the auroral can expand equatorward con− sistently as, for instance, happened during the 28 − 30 October 2003 geomagnetic storm [Thomson et al., 2010].
According to Marshall et al. [2012], the next phase of this study could be to examine separately events dom− inated by the dynamics of the auroral currents or by that of the magnetopause currents. Indeed, it has been found that the decrease of the maximum value of dB/dt from high to middle latitudes depends on the type of event considered, either geomagnetic storms or sudden im− pulses events. To conclude, we underline that an ad− vantage of using GIC index stands also in the results ob− tained by Marshall et al. [2012] who, using a probabilistic approach on a very large number of world− wide occurrences of known faults, as malfunctioning and ruptures of transformers, quantified the threat to power systems in terms of GIC index and established a scale of the risk level from "low" to "extreme". There− fore, an accurate modelling of the dependence of GIC index maximum value with latitude could be used to as− sess the risk level of each country based on their geo− magnetic latitude.
The results presented in this paper rely on data collected at magnetic observatories. We thank the national institutes that support them and INTERMAGNET for promoting high standards of magnetic observatory practice (www.intermagnet.org). Rapid magnetic variations are those calculated and made available by Ebre Observatory from data collected at magnetic observatories and on geomagnetic indices calculated and made available by ISGI Collaborating Institutes. We thank also the data suppliers and the WDC for Geomagnetism (Edinburgh) and the use of NASA/GSFC's Space Physics Data Facility's OMNIWeb (or CDAWeb or ftp) service, and OMNI data. The authors are grateful to two anonymous referees for their constructive reviews. K. Shiokawa, N. Nishitani, R. Kataoka, Y. Kamide, T. Aso, Y. Watanabe and Y. Tsuneta (2009)