Mapping a Magnetic Superstorm: March 1989 Geoelectric Hazards and Impacts on United States Power Systems

Research Article Open Access

Jeffrey J. Love, Greg M. Lucas, E. Joshua Rigler, Benjamin S. Murphy, Anna Kelbert, Paul A. Bedrosian

First published: 12 May 2022 | https://doi.org/10.1029/2021SW003030

Abstract

A study is made of the relations between geomagnetic and geoelectric field variation, Earth-surface impedance, and operational interference (“anomalies”) experienced on electric-power systems across the contiguous United States during the 13–14 March, 1989, magnetic storm. For this, a 1-min-resolution sequence of geomagnetic field maps is constructed from magnetometer time series acquired at ground-based observatories. Induced geoelectric field maps are calculated by convolving the geomagnetic maps with magnetotelluric impedance tensors. During the storm, anomalies were concentrated where the lithosphere is electrically resistive, and when and where geoelectric field amplitudes were high. This was particularly true in the Mid-Atlantic, Northeast, and the upper Midwest. Few anomalies were experienced in other parts of the Midwest and across much of the West, where the lithosphere is more conductive, and when and where geoelectric field amplitudes were low. Peak 1-min-resolution geoelectric field amplitude ranged from 21.66 V/km in Maine and 19.02 V/km in Virginia to <0.02 V/km in Idaho. Latitude-dependent organization of geoelectric hazards by auroral-zone electrojet currents is detectable, but it is much weaker than geographic organization due to surface impedance. Hazardous geoelectric fields were induced during different storm phases, at different local times, and, by inference, by a variety of ionospheric currents. Compared to geoelectric field amplitudes realized across the United States during March 1989, hazard maps used by utility companies to estimate systems exposure have much less geographic detail and a much smaller maximum-to-minimum range in geoelectric field amplitude. Future research would benefit from denser geomagnetic monitoring, additional magnetotelluric surveying, and access to power-system impact data.

Key Points

• Electric-power system interference was concentrated where surface impedance is high, and when and where geoelectric field amplitudes were high
• High geoelectric hazards and numerous power-system anomalies were realized in the Eastern United States, near many large cities
• Power-system impact data provide important, if partial, validation of retrospectively constructed geoelectric field maps

Plain Language Summary

Electric fields induced in the Earth during magnetic storms can drive uncontrolled currents in electric-power systems, interfering with their operation. Geomagnetically induced currents realized during the magnetic storm of March 1989 caused a blackout in Québec, Canada, and, in the Mid-Atlantic and Northeast United States, they caused operational interference for electric-power companies and damaged a high-voltage transformer. In support of projects for estimating geoelectric hazards and improving power-system resilience, maps are made of March 1989 magnetic-storm geoelectric hazards and corresponding impacts on United States power systems. Results are based on modeling geomagnetic monitoring data, geoelectromagnetic survey data, and a compilation of published reports of power-system interference. During the storm, electric-power system interference was concentrated where the lithosphere is relatively electrically resistive, and when and where the geoelectric field was of high amplitude. This was particularly true in the Mid-Atlantic and Northeast, near many of America’s largest cities, and in the upper Midwest. Retrospective analyses, such as this one for the March 1989 storm, show where utility companies might concentrate their efforts to mitigate the impacts of future magnetic superstorms.

1 Introduction

The magnetic storm of 13–14 March 1989, was, by standard measures, the most intense storm of the space age; maximum negative storm-time disturbance, −Dst = 589 nT (Kyoto), −Dst = 565 nT (Oulu). The March 1989 storm is also notable for the havoc it wreaked on technological systems around the world (e.g., Allen et al., 1989; Cliffswallow et al., 1993). Satellite operations were impaired; some satellites were permanently damaged. Over-the-horizon radio communication, navigational systems, and geophysical surveys were disrupted. As significant as such impacts were, the March 1989 storm is usually recalled for the impact it had on electric-power systems. The Canadian Hydro-Québec power system completely collapsed as a result of interfering quasi-direct “geomagnetically induced currents” (GICs) (Bolduc, 2002; Guillon et al., 2016). In the United States, numerous operational “anomalies” were reported by utility companies, especially in the Mid-Atlantic and Northeast (e.g., Kappenman, 2010), and GICs damaged a high-voltage transformer at a nuclear-power plant in Salem, New Jersey (NJ) (e.g., Subudhi et al., 1994). Some of these effects of March 1989 might have been anticipated from experiences with earlier storms. The storm of August 1972 (Boteler & van Beek, 1999), for example, caused operational interference on North American electric-power systems, and it caused a shutdown of an important telephone cable system operated in the Midwest (Anderson et al., 1974). The storm of March 1940 (e.g., Fleming, 1940; McNish, 1940) caused operational interference on North American power systems (e.g., Davidson, 1940, 1941) and on telegraph and telephone systems (e.g., Germaine, 1940; Ireland, 1940). In light of such historical events, some researchers suggest that our modern, electricity-dependent society is increasingly vulnerable to magnetic-storm hazards (e.g., Kappenman, 2004, 2012; Liu et al., 2010) – a storm as intense as, for example, that of May 1921 (e.g., Hapgood, 2019; Love, Hayakawa, Cliver et al., 2019) or the Carrington event of September 1859 (e.g., Boteler, 2006; Cliver, 2006) might cause widespread and long-lasting interruption of electric-power transmission and carry considerable economic cost (e.g., Barnes & Van Dyke, 1990; Baker et al., 2014; Eastwood et al., 2017).

The deleterious impact that magnetic storms can have on electric-power systems is qualitatively understood. Storm-time geomagnetic field disturbance (or “variation”) induces electric fields in the electrically conducting Earth. At the surface, these geoelectric fields can drive GICs along power transmission lines and through grounded transformers, which can cause half-cycle saturation. This distorts alternating-current waveforms and can trip system relays, precipitating power-system voltage collapse and heating, and, even, damage transformers (e.g., Abda et al., 2020; Kappenman, 2001; Molinski, 2002; Oyedokun and Cilliers, 2018; Samuelsson, 2013). Quantitative modeling of this convoluted chain of causes and effects, encompassing natural hazards and engineered systems, is an important objective of ongoing research and development (e.g., Gaunt, 2016; Pennington et al., 2021; Pilipenko, 2021; Pulkkinen et al., 2017; Thomson et al., 2010). Such work can be seen as timely in light of the fact that storms as intense (or more intense) as that of March 1989 are now known to occur rather frequently – on average, about every four solar cycles (Love, 2021). National and international space-weather organizations identify the quantitative estimation of geoelectric hazards as a high-priority project (e.g., Jonas et al., 2016; Krausmann et al., 2016; National Science and Technology Council, 2019; Schrijver et al., 2015).

In this context, we examine geographic-temporal correlations between geomagnetic field variation, geoelectric field amplitude, Earth-surface impedance, and operational anomalies experienced on electric-power systems across the contiguous United States (CONUS) during the March 1989 magnetic storm. We construct a time-sequence of maps of the surface geomagnetic field variation by spatially interpolating between magnetometer time series acquired at ground-based observatories that operated in March 1989. We construct a corresponding time-sequence of maps of the variation in geoelectric field amplitude by convolving the geomagnetic variation maps with long-period impedance tensors derived from magnetotelluric measurements acquired at numerous survey sites across CONUS in recent years. We compare the geomagnetic variation maps and geoelectric field amplitude maps with each other, with maps of magnetotelluric impedance, and with occurrence times and locations of power-system anomalies reported in publicly available records. Can a particular storm phase or a specific feature in the ionospheric-current system be identified as having been especially effective at generating hazardous geoelectric fields? Were power-system anomalies experienced, as might be expected, when and where local geoelectric field amplitudes were high? How were geoelectric hazards organized? Were they obviously related to the geography of surface impedance? Or were geoelectric field amplitudes just a function of the magnetic latitude of auroral-zone ionospheric currents? Answering these questions should improve our understanding of storm-time geoelectric hazards and usefully inform projects for improving power-system resilience (e.g., Oughton et al., 2019; Piccinelli and Krausmann, 2014; Pirjola et al., 2005).

….

….

12 Discussion and Conclusions

We look back, now, over the sequence of events that were the magnetic storm of March 1989 and the related impacts on United States electric-power systems. After the initial-phase sudden impulse and the subsequent high-latitude substorm that caused the blackout of the Hydro-Québec power-system, with the development of the storm’s main-phase, as measured by a decline in Dst, Figure 5e, substorm ionospheric currents swept down across CONUS. The combination of mid-latitude geomagnetic field variation generated by these overhead currents, Figures 6 and 8, and high surface impedance, Figure 3, due to resistive lithospheric structures, supported the induction of geoelectric fields of relatively high amplitude in the Mid-Atlantic, the Northeast United States, and in parts of the upper Midwest, Figures 7b and 9–11. And this, in turn, caused numerous operational anomalies in power-systems, Figures 7d and 9–11. The high calculated geoelectric field amplitudes in the Mid-Atlantic and Northeast and the anomalies in the same area, notably, represent high hazards for and impacts on electric-power systems serving many of America’s largest cities: Boston, New York, Philadelphia, Baltimore, Washington, DC, Figure 12b – a megalopolis of over 50 million people. The geomagnetic monitoring data available for the March 1989 storm, the magnetotelluric tensors acquired across CONUS, mostly since 2006, and the reported power-system anomaly data allow us to map out the March 1989 sequence of events more than three decades since they transpired – the CONUS power-system experienced stress when and where our calculated storm-induced geoelectric field amplitudes were high.

12.1 Geoelectric Hazard Benchmarks

In contrast to our approach of directly using impedance tensors derived from magnetotelluric measurements, some geoelectric hazard analyses use tensors derived from one-dimensional, depth-dependent models of regional electrical conductivity structure. For example, a one-dimensional conductivity model is developed for the province of Québec (Boteler, 1997; Ferguson & Odwar, 1997). Assuming that this model is representative of extremely resistive geological settings, Pulkkinen et al. (2012) convolve the corresponding surface impedance with 1-min magnetometer time series of the March 1989 storm from 49 different observatories around the world (see also, Ngwira et al., 2013). The maximum peak geoelectric field amplitude that Pulkkinen et al. (2012, their Figure 4a) find is ∼6.00 V/km for an observatory located at ∼52°N geomagnetic latitude. This is lower than what we calculate for 31 of 1253 sites across CONUS, and it is much lower than the 21.66 V/km we calculate for MEE62 in ME, the 19.02 V/km we calculate for VAQ58 in VA, and the 17.33 V/km we calculate for CTI60 in Connecticut, Section 8. We conclude that using the Québec model leads to underestimation of peak geoelectric field amplitudes for the March 1989 storm in some resistive geological settings – and, in particular, the Québec model is not useful for estimating an upper bound on geoelectric field amplitude.

Working under contracts from the Electric Power Research Institute (EPRI), Fernberg (2012) develops one-dimensional, depth-dependent models of “average” electrical conductivity structure for separate physiographic provinces that, together (almost) cover CONUS in a piece-wise fashion (see also, Blum et al., 2015). Some of these physiographic provinces are very large, with lateral length scales >1000 km. One problem is that the regions studied by Fernberg are not clearly defined. For example, the Appalachian Valley and Ridge province appears to be unmodeled, and subsequent studies that use the Fernberg models subsume it with his Piedmont province model (PT-1); see, e.g., NERC (2016, their Figure II-3). Wei et al. (2013), in their study of geoelectric fields induced across CONUS during the March 1989 and October 2003 storms, convolve maps of geomagnetic field variation (their March 1989 maps would resemble ours) with impedances derived from Fernberg’s physiographic models. The peak geoelectric field amplitude that they estimate for the Piedmont province is 1.92 V/km. The NERC (2016) 100-year geoelectric benchmarks are based on a reduced-dimensional latitude-dependent model of storm-time geomagnetic disturbance, also convolved with Fernberg impedances. The NERC benchmark amplitude for VA latitudes in the Piedmont is 2.34 V/km. Both the Wei et al. and the NERC amplitudes are an order of magnitude lower than the peak amplitude of 19.02 V/km we calculate for VAQ58 for the March 1989 storm, and they are two orders of magnitude higher than the peak amplitude of 0.03 V/km we calculate for VAQ55 (both of these sites are within the physiographic region they label as PT-1).

Gannon et al. (2020) develop one-dimensional models of impedances by averaging magnetotelluric tensors over physiographic provinces and, then, extract the part of the average tensor corresponding to a one-dimensional model. They estimate geoelectric amplitudes across all of CONUS using both the (unaveraged) magnetotelluric tensors themselves and their average tensors, convolving each of them with the OTT March 1989 geomagnetic time series – no mapping is done like in our study or in Wei et al. (2013), and no latitude adjustment is made like in NERC (2016). The peak geoelectric field amplitude that Gannon et al. (2020) estimate for the Piedmont province is 42 V/km at VAQ58, more than a factor of two higher than our estimate for this site; lowest peak amplitudes are not emphasized by Gannon et al. (2020). These comparisons are similar to those already made by Lucas et al. (2018), who calculate geoelectric fields and geovoltages on transmission lines for the March 1989 storm using both magnetotelluric tensors and tensors for one-dimensional physiographic models for the Mid-Atlantic United States. Lucas et al. (2018) find that the magnetotelluric tensors produce substantially more geographic granularity in transmission-line geovoltages than the tensors for the physiographic models. In particular, they find that the tensors for the physiographic models can lead to underestimation of geovoltages on some transmission lines by more than an order of magnitude, and overestimation of geovoltages on other transmission lines by, also, more than an order of magnitude.

Accuracy is not a limitation of the one-dimensional average physiographic models of Fernberg (2012), Gannon et al. (2020), and Blum et al. (2015), but geographic resolution is. Limited resolution is why the NERC (2016) benchmarks do not generally give storm-time geoelectric field amplitudes, at any given location, that are close to the amplitudes we find by direct use of the magnetotelluric tensors. Indeed, Gannon et al. (2020) find that impedance across the Piedmont province is geographically too complicated to be accurately described by a single one-dimensional physiographic model. This is the result of lithospheric structures that are varied and complex (e.g., Hoffman, 1989; Rast, 1989). Inversions of the magnetotelluric tensors reveal substantial three-dimensional structure in the Piedmont and neighboring Appalachian Mountains (e.g., Kelbert et al., 2019; Murphy and Egbert, 2017; Ogawa et al., 1996). Other investigators, in comparing impedance tensors derived from one-dimensional physiographic models against magnetotelluric tensors, find that those derived from one-dimensional models generally yield inferior estimates of storm-induced geoelectric fields (e.g., Cordell et al., 2021; Simpson and Bahr, 2021; Torta et al., 2017). To this, it is worth emphasizing that each individual magnetotelluric tensor already embodies a substantial degree of natural averaging. With the diffusion of electromagnetic variation through the volume of the Earth, averaging occurs across electromagnetic diffusive length scales – this principle is fundamental to the formulation of Equation 1. And, furthermore, to calculate a geovoltage on a given transmission line, the projection of the local geoelectric field is integrated over the path of the transmission line (Kelbert & Lucas, 2020; Lucas et al., 2018, 2020) – this is also a type of spatial averaging.

12.2 Additional Survey, Monitoring, Impact Data

In the light of the preceding discussion, we recognize that more complete and, in some places, more detailed magnetotelluric surveying would be extremely useful for geoelectric mapping projects. Most obviously, it would be very useful to perform a long-period survey of Québec, ON, and MB, provinces with broad expanses of electrically resistive, yet geologically complex, metamorphic lithosphere. Ideally, such a survey would have more-or-less uniform site spacing like that of the CONUS long-period magnetotelluric survey, Figure 3. Long-period, national-scale magnetotelluric surveys, as used in our analysis here, serve as useful reconnaissance for identifying regions of high and low surface impedance and geoelectric hazard, but finer-scale hazard mapping, at the level of, for example, individual transmission lines or individual substations, requires denser survey station spacing (Murphy et al., 2021). Therefore, additional surveying could be targeted for areas known to have high surface impedance, Figure 3, high geoelectric hazards, Figure 12a (Lucas et al., 2020, their Figures 8 and 9), and where power-system anomalies are often experienced, Figures 1 and 12b. If the survey tensors are a mixture of long-period and wideband, these can be treated consistently through inverse modeling to construct three-dimensional models of subsurface conductivity structure (e.g., Kelbert et al., 2019). These conductivity models can be used to produce a map of surface impedance (e.g., Kelbert, 2020; Marshalko et al., 2020).

Recalling our discussion of errors in Section 6, the accuracy of geoelectric maps, such as that presented here, would be substantially improved if a denser network of magnetic observatories and magnetometer stations were in operation across North America (e.g., Love, Rigler et al., 2018, their Figure 1). Roughly speaking, we expect that, by simple geometric attenuation, a geomagnetic field generated by a small-scale current system in the ionosphere will have a lateral spatial scale at the Earth’s surface that is longer than the approximate 100-km height of the ionosphere. Indeed, analyses of data from magnetometer networks having close geographic spacing show that 1-min geomagnetic field variation is characterized by scales that generally exceed a few hundred kilometers (e.g., Dimmock et al., 2020; Watermann et al., 2006). Consulting with the SuperMag database (e.g., Gjerloev, 2009), since 1989, numerous variometer stations have been deployed and are operational across Canada (Connors et al., 2009; Engebretson et al., 1995; Mann et al., 2008); notably, this includes stations in Québec (Connors et al., 2016; Russell et al., 2009). The status of variometers in CONUS is very different. Although variometer stations have been deployed (e.g., Chi et al., 2013), over recent years, few have operated and contributed data to SuperMag (Engebretson & Zesta, 2017; Love, Rigler et al., 2018). If opportunities materialize for expanding magnetometer operations across CONUS, as with targeting additional magnetotelluric surveys (discussed above), it would be reasonable to concentrate new stations in areas where impedance is high, geoelectric hazards are high, and where power-system anomalies are often experienced.

The power-system anomaly data we use for our analysis of the March 1989 magnetic storm are descriptions of operational problems, along with timing and location information. They provide important, if partial, validation of our method for mapping geoelectric fields, Section 6 and Lucas et al. (2020). Similarly, descriptive datasets are used with success in analyses of geoelectric fields in other countries (e.g., Eroshenko et al., 2010; Gil et al., 2021; Výbošt’oková and Švanda, 2019). If quantitative power-systems-impact data, like the GIC data shown in Figure 1f for other storms, were available for the March 1989 storm, our analysis could have been more quantitative and more informative for evaluating power-system response to storm-induced geoelectric fields. Although the release of GIC data and other related systems monitoring data is usually viewed as orthogonal to utility company traditions and their commercial sensitivities (e.g., Eastwood et al., 2017; Head, 2015; Schrijver et al., 2015), when such data have been made available to researchers, they have proven to be very useful (e.g., Blake et al., 2016; Dimmock et al., 2019; Divett et al., 2020; Rosenqvist and Hall, 2019; Sokolova et al., 2019; Thomson et al., 2005; Torta et al., 2017). In the United States, utility companies are now required to make storm-time GIC data openly available to the public (Federal Energy Regulatory Commission, 2016). These data will be useful for validating future geoelectric mapping projects. For now, the correlation we see between geoelectric field amplitude and power-system anomalies in Figures 7 and 9–11 provides some degree of confidence in the practicality of the ongoing project of the National Oceanic and Atmospheric Administration and the U.S. Geological Survey for developing a capacity to map geoelectric fields across CONUS in real-time (Balch et al., 2020; Love, Rigler et al., 2018).

12.3 Source Current Systems

It is widely appreciated that the location, intensity, and time dependence of the auroral electrojet are crucial factors in determining the hazardous impact of induced geoelectric fields (e.g., Beggan, 2015; Dimmock et al., 2020; Mac Manus et al., 2017). Some researchers suggest, more specifically, that certain elements of electrojet current systems can be especially effective at inducing geoelectric fields (e.g., Ngwira et al., 2015; Viljanen, 1997). For example, some researchers note geoelectric induction by ionospheric currents associated with sudden impulses (e.g., Pulkkinen, Thomson et al., 2003; Oliveira et al., 2018). Some researchers note geoelectric induction by substorms triggered by particular changes in solar-wind conditions (e.g., Hajra et al., 2020). Others note induction by westward substorm eletrojets in local nighttime (e.g., Viljanen et al., 2006), westward and eastward electrojets in local nighttime (e.g., Kozyreva et al., 2018), and westward and eastward electrojets across a range of local times (e.g., Belakhovsky et al., 2019; Dimmock et al., 2019), and others note induction by geomagnetic pulsations related to localized vortex-shaped auroral-zone current flows (e.g., Apatenkov et al., 2020; Belakhovsky et al., 2019; Chinkin et al., 2021; Hartinger et al., 2020; Heyns et al., 2021; Viljanen, 1997). Still others note geoelectric induction by storm-time currents that are not classically related to substorms (e.g., Huttunen et al., 2002). Considering these results, together, one might conclude that hazardous geoelectric fields and GICs can be induced during different storm phases, at different local times, and as the result of a variety of ionospheric currents (e.g., Kataoka & Ngwira, 2016; Pulkkinen, Thomson et al., 2003; Pulkkinen et al., 2005), although, at different periods of time over the complicated course of an intense magnetic storm, individual current elements rise to be more effective than others at inducing geoelectric fields. Our own conclusions on this subject are necessarily limited by our analysis of one storm, that of March 1989, the time resolution and the quality of the magnetic observatory data (1-min), and the sparsity of the observatory network across North America at the time. Still, we note that power-system anomalies, Figures 7d and 9–11, were experienced during different storm phases and beneath both westward and eastward equivalent electrojets, Figures 6 and 8, across a range of local times (e.g., Boteler & van Beek, 1993; Kappenman, 2010). This also suggests that there is not a single specific storm-time “driver” of hazardous geoelectric fields and GICs.

Acknowledgments

We thank J. M. Carter, R. D. Gold, K. A. Lewis, B. R. Shiro, and E. W. Worthington for reviewing a draft manuscript. We thank J. E. Caldwell for useful conversations concerning magnetic observatory data. This work was supported by the USGS Geomagnetism Program. G. M. Lucas received support from NASA, Grant 80NSSC20K1477.

Read the full open access article here.

HT/Yooper, dnh