Figure b shows detailed time profiles of the northern Scandinavian magnetograms at >65∘ GGlat, which are located near local noon when the X9.3 flare took place. Although the Fig.  period is in the middle of the space weather event that started on 4 September 2017, the magnetic storm did not start until the end of 6 September. Furthermore, the solar wind during this period was stable, at around 450 km s-1 (slightly declining in time), and IMF was weak, with a total field less than 3 nT (BX=-1 nT, BY=1 to 0 nT, BZ=-2 nT during 12:00–13:00 UT in the geocentric solar magnetospheric (GSM) coordinate), as shown in Yamauchi et al. (2018; Fig. 1). Such a stable condition caused the preceding substorm activity before the X9.3 flare to diminish before the flare onset, as seen in the geomagnetic indices (Fig. ). The IMF BY condition indicates that the cusp crochet must be small or invisible.

The subsolar crochet was observed as being short-lived geomagnetic deviations starting nearly simultaneously as the X-ray flux increased at the Earth and lasted about 10–15 min at Sørøya (SOR; 70.5∘ GGlat), Tromsø (TRO; 69.7∘ GGlat), Kiruna (KIR; 67.8∘ GGlat), and Rørvik (RVK; 65.0∘ GGlat.). All equatorward stations show the same type of disturbances (Curto et al., 2018). This is also seen in ASY-D (63 nT at 12:04 UT) and ASY-H (77 nT at 12:05 UT), as shown in Fig. a, with an amplitude change by the flare of about 60 nT. One can even recognise a crochet-like signature in ASY-D when an X2.2 flare occurred at around 09:00 UT.

The new finding is the subsequent geomagnetic disturbances; a large positive ΔH deviation (northward ΔB) also started right after or even during the negative ΔH spike (southwestward ΔB) of the subsolar crochet, with much higher amplitudes, as shown in Fig. b. This positive ΔH continued for hours, with the peak at around 13:00 UT at Bear Island (BJN) at 74.5∘ GGlat (>200 nT), 13:20 UT at 70∘ GGlat (SOR and TRO – 180 nT) and 68∘ GGlat (KIR – 140 nT), and 14:50 UT at 65∘ GGlat (ROR – 70 nT). With a larger amplitude and longer duration than the subsolar crochet, this geomagnetic signature is visible even in AU, as shown in Fig. b, although the baseline is as large as 100 nT due to the previous substorm activity, and it can hide the subsolar crochet if any exists.

The development is also quick. At BJN (74.5∘ GGlat), ΔH reached ΔH>65 nT at 12:05 UT already, i.e. at the peak time of the subsolar crochet, and reached 130 nT at 12:20 UT. Since the long duration already indicates that the generation mechanism is different from that of the subsolar crochet (redistribution of electron), the positive ΔH of this crochet with diminishing amplitude towards lower latitudes should cancel the negative ΔH of the subsolar crochet at high latitudes. In fact, ΔH exceeded the value before the flare at 12:10 UT at 70∘ GGlat, 12:12 UT at 68∘ GGlat, and 12:17 UT at 65∘ GGlat.

These large ΔH, however, are observed in a limited latitudinal range, diminishing towards higher latitudes with 140 nT at 76.5∘ GGlat (Hopen – HOP; 12:50 UT) and not visible at 78.2∘ GGlat (Longyearbyn – LYB). Since the geomagnetic latitude of LYB is only 75.3∘, and IMF is weak with BY=0 nT, the positive ΔH is limited to the geomagnetic closed region outside the cusp or polar cap. The closed geometry is also indicated by the EISCAT Svalbard radar data (Yamauchi et al., 2018). The polar cap (PC) index, which corresponds to the polar cap activity, shows enhanced values in the same period but not as prominently as in AU, as shown in Fig. .

On the other hand, positive ΔB at around 75∘ GGlat was observed in a wide local time range, as shown in Fig. c and d (ΔX is nearly the same as ΔH in both stations). Danmarkshavn (DMH; 19∘ W) and Dikson Island (DIK; 81∘ E) showed ΔX of about 120 nT and 100 nT at around 13:00 UT, respectively, compared to the values before the flare. Together with the zero IMF BY condition, the observed large ΔH cannot be a cusp crochet. Considering its location and pre-existing activity, this crochet is neither a subsolar crochet nor an auroral crochet, although some part of the observed ΔH could be affected by the auroral crochet.

For example, DIK is located near the evening terminator (it is still under sunlight near the horizon), and the geomagnetic activity before 12:00 UT indicates some auroral activity. Therefore, the first peak at around 12:20 UT, which is larger than that of BJN or HOP and with more westward ΔB, can be the auroral crochet rather than the extension of the new crochet. However, the second and third peaks are in the same direction (northward ΔB) as ΔB near local noon, and multiple peaks are not expected for an auroral crochet under a smoothly declining X-ray flux. Therefore, the positive ΔX at DIK at around 13:00 and 13:40 UT can be interpreted as being part of the new high-latitude crochet rather than the auroral crochet, although we cannot dismiss the possibility of the auroral crochet.

With such a large amplitude, the crochet is visible even in AU, although the AE stations are not located at favourable GGlat. During 12:00–14:00 UT, AU has three positive peaks (with provisional values of 240 nT at 12:12 UT, 190 nT at 12:55 UT, and 160 nT at 13:43 UT, while further baseline subtraction might be needed). The timing of these peaks corresponds to the subsolar crochet and the high-latitude one at BJN, but the provisional AU value reflects DIK data (DIK is one of the AE stations), as shown in Figs. c and b. Although the amplitude is larger at BJN than DIK for the second and third peaks, BJN is located far poleward of the AE stations and did not contribute to AU.

From Fennoscandian, Icelandic, and Greenland magnetometer data, we also calculated the ionospheric equivalent currents (including Sq current), using the Spherical Elementary Current System (SECS) technique (Amm, 1998; Amm and Viljanen, 1999). Here, we obtained epsilon = 0.037 in a similar fashion to Wygant et al. (2012). Note that the quiet levels that represent the internal and crustal geomagnetic field (even without Sq) were removed before applying the data to the SECS technique. They were calculated by a least square root approximation (rather than least squares or means), where the square root emphasises the small variations around the quiet level rather than larger disturbances such as substorms. To have a sufficient amount of variations with low activity for the calculation, while avoiding contamination of main field secular variation, the removal of the quiet levels is performed on 10 d of data centred on the day of interest. The uncertainty is estimated to be less than 10 nT (see Edvardsen et al., 2013, for more details).

The results are shown in Figs.  and . Figure  shows the latitudinal distribution of the eastward current density crossing two meridians (Scandinavia – 20∘ E; Greenland – 50∘ W) where the actual magnetometer networks are deployed. In Fig. , one can see a sudden appearance in the ionospheric current in a wide region at around 12:00 UT when the X-ray flux from the X9.3 flare increased at the Earth. The enhancement is westward at lower latitudes (<70∘ GGlat at 20∘ E or 13:00 LT, and <72∘ W at 50∘ W or 09:00 LT), and eastward at higher latitudes in both meridians. They correspond to the subsolar crochet current in the northern hemisphere (Curto et al., 2018) and the new high-latitude crochet mentioned above, respectively. Figure a and b show the 2D vector directions, corresponding to the timing right before the flare (11:50 UT), and at the peak of subsolar crochet (12:04 UT). The low-latitude side composes the anticlockwise current that agrees with the return current direction of the subsolar crochet at high latitudes (Curto et al., 2018; Annadurai et al., 2018). The high-latitude side forms another independent anticlockwise current, with a strong eastward current near BJN (cf. Fig. ), as mentioned above. The resultant shear, which is formed poleward of BJN, corresponds to the upward field-aligned current, i.e. to the afternoon Region 1 field-aligned current (Iijima and Potemra, 1976; Yamauchi and Slapak, 2018).

The eastward current expanded quickly in the longitudinal direction and towards lower latitude as soon as the subsolar crochet diminished. At the same time, the current density gradually increased towards its multiple peaks. Figure c and d show the 2D vector directions at these peaks; the first minor peak of ΔH (at around 12:20 UT) after the end of the subsolar crochet was located at around 75∘ GGlat, and the major peak of ΔH (at around 13:20 UT) was located at around 70∘ GGlat, respectively. By 12:20 UT the area of this westward current that is the most intense at around 72–74∘ GGlat expanded in a wide local time from Greenland to eventually reach Siberia (DIK at 81∘ E as shown in Fig. d), i.e. more than 130∘ in longitude. The entire current lies in the geomagnetically closed region, as mentioned above, and its peak latitude gradually moved equatorward. By the peak time at around 13:20 UT, all regions over 5∘ in latitude and >100∘ in longitude are intensified, with a much higher intensity than the subsolar crochet current.

The eastward current direction (or positive ΔH) is the same as the ionospheric current in the evening auroral oval (we here mean the region between the upward Region 1 field-aligned current and downward Region 2 field-aligned current; Iijima and Potemra, 1976; Akasofu, 1977; Yamauchi and Slapak, 2018), and the observed eastward current continued until the next substorm onset took place at 15:00 UT (see Fig. b). However, no outstanding substorm is visible in AE (Fig. b) or DIK data (Fig. c) before this substorm with continued weak IMF condition. The eastward current patch even started at the Greenland meridian at 09:00 LT and continued towards the afternoon sector although IMF BY=0 nT. Since there was no auroral current signature at BJN before this crochet, this current system is not the auroral crochet current. Rather, the question is how much this new crochet contributes to ΔB in the evening sector, e.g. compared to the auroral crochet. In this sense, we cannot judge for the moment whether the crochet detected in DIK is an evening extension of this crochet or auroral crochet or both effects mixed.

For this X9.3 flare event on 6 September 2017, the duration of this new crochet matches with that of the high X-ray flux. It was at the X-class level until 13:40 UT and at the M-class level until about 16:50 UT, as shown in Fig. . From this coincidence, speculated about the possibility of the enhancement of a pre-existing Sq0 (solar quiet tidal-driven) current without showing detailed data. However, the Sq0 has long been expected to be small at high latitudes (Yamasaki and Maute, 2017). Therefore, we need direct evidence with this Sq0 scenario. For that purpose, ionospheric electron density and ion velocity, both observed by EISCAT VHF radar at Tromsø (224 MHz), are shown in Fig. .

Figure b shows that the electron densities in the 100–200 km altitude range were significantly enhanced by the enhanced X-ray flux, starting at around 12:00 UT (doubling at 100 km altitude and seen up to 200 km altitude). Figure a shows that northward ion convection was also enhanced, and more importantly, that the background Sq0 ion convection (seen as large geomagnetic ΔH>0) starting from around 09:00 UT is already strongly northward (away from the subsolar region), with large values of as much as 150 m s-1 at 71∘ magnetic latitude (Mlat) at 100 km altitude.

Since the increase in the electron density means an increase in the ion-neutral density ratio too, the ionospheric current is expected to flow at lower altitudes where the tidal (Sq0) ion convection is stronger. Such a change can enhance the pre-existing ionospheric Sq0 current significantly, although this scenario does not easily explain why it is found in a limited latitudinal range and wide longitude.

The time development of the electron density enhancement, together with the elevated ion velocity at 100 km, matches the ΔH time profile at BJN that is located under the area observed by the EISCAT VHF radar. Note that ion velocity direction (northward) is the electric field direction at this altitude, where only ions are collisional with neutrals but not electrons, and hence, collision-free electrons drift westward, resulting in an eastward Hall current (that causes ΔH>0 geomagnetic disturbance).

The EISCAT data in Fig. d also revealed a decrease in electron density above 300 km after 12:00 UT. Since photochemistry predicts density increase at all altitudes, this density decrease must be caused by ionospheric dynamics, such as the 3D distribution of ionospheric current. This decrease does not affect the total current because, in addition to the low collision frequency that prevents conduction current, the ion velocity at >300 km did not change very much compared to the ion velocity at <200 km and does not contribute to the ionospheric current.

Although magnetic stations have been extended towards higher latitudes beyond 68∘ GGlat since 1980s, this new crochet has never been reported, at least not to our knowledge. One possible reason is that the phenomenon is limited to a relatively small range in the geographic latitude (68–75∘ GGlat), while the station should be completely outside the geomagnetic cusp (<75∘ Mlat). This criterion excludes many geomagnetic stations over Greenland and Canada from finding the new crochet.

There are many questions to answer on this phenomenon. One obvious question regards the conditions that cause this new high-latitude crochet. When we made a survey using quick plots, we compared the indices with the X-ray flux (red lines in Fig. ) only, but other obvious factors, such as the season and geomagnetic latitude, have not been not sorted out. We thus need to make more solid analyses. For example, we need to include amplitudes (we examined only yes or no responses), examine the closure of the ionospheric current system at high latitudes, analyse global magnetometer network data, and take statistics of them. Such a global perspective would also indicate for which conditions the effect can be detected in AU or AL.

One important note is that the difference between the GGlat and Mlat (i.e. UT dependence) must be considered. Also, we have to note that the current system might be different between different events (the size of the flare may affect the intensity and size, while the season may affect the distribution pattern and profiles of the solar zenith angles of magnetic stations). In addition, if the intensification of the Sq current is important, the new crochet might be enhanced near the equinoxes (rather than summer solstice) through inter-hemispheric coupling (Yamasaki and Maute, 2017), which avoids the saturation of convection-driven charges. Therefore, it might be difficult to obtain consistent results, but at least a common feature can be obtained.

As shown in Fig. , this current system might be related to the enhancement of the Sq0-like background current through the enhancement of both the ion and/or electron density and ion velocity (Pedersen electric field). While the density enhancement is explained by the flare radiation, the direct cause of velocity enhancement is not clear. In addition to this obvious question, we need to know the relative importance of these enhancements on the high-latitude crochet, and we need to understand the relation to the nightside crochet because we also found many cases in the nightside as well. This requires identifying the criterion for how to classify the observed crochet as the auroral crochet or nightside extension of the new crochet, which we have not yet found.

To answer these questions, we need similar radar data for different events. Since the availability of radar data in favourable observation modes and geometry (such as the EISCAT data on the 6 September 2017) is limited, we need other radar data, including future facilities such as EISCAT 3D, for a solid answer. Such work will also probably give some hints as to why the electron density decreased at >300 km in Fig. . Future satellite missions that cover ionospheric E-region, such as Daedalus (Sarris et al., 2020), are highly necessary.

In the preliminary survey, we found many coincidental cases in which a large gradient of ΔH occurred at the same time as a substorm onset or a sudden intensification of a substorm when the X flare took place. Figure  shows one such example, when the X6.2 flare took place at 19:13 UT, and the substorm onset took place immediately after. This substorm is most likely associated with the southward IMF period during 18:29–18:57 UT (not shown here) that is detected at the Sun–Earth first Lagrange point L1 by the Advanced Composition Explorer (ACE), but the triggering mechanism is not necessarily associated with IMF (e.g. Yamauchi et al., 2006). Solar wind density and velocity were stable after the coronal mass ejection (CME) arrival 5 h before, making us consider the crochet as a possible trigger. Larger AU than AL right after the onset is also in agreement with crochet rather than substorm activity because the substorm onset is characterised by large negative ΔH and |AL|≫AU.

Theoretically, the crochet mechanism may trigger a substorm through a sudden intensification of ionospheric density and electric field through a magnetosphere–ionosphere (M–I) coupling (e.g. Kan et al., 1988). Since several different onset mechanisms may cause a substorm, it is quite possible that crochet may also trigger a substorm as one of the minor onset mechanisms (Yamauchi, 2019). The investigation of such a scenario is a future task.

The same question arises with respect to the M–I coupling. The latitude range of the new crochet is inside the geomagnetically closed region (near local noon), while it is close to the dayside Region 1 field-aligned current. This means that the new crochet might influence the field-aligned current system. Such a study requires satellite observations at the right location and the right timing.

If the long-lasting high X-ray flux influences the ionospheric current, some effects might arise from the X2.2 flare at 09:00 UT on 6 September 2017, i.e. on the same day as the X9.3 flare (see Fig. a). One candidate is the density moderation that synchronises with the Pc5 pulsation during the recovery phase of the substorm that started at 09:37 UT, which peaked at 05:58 UT with AL=-666 nT (Fig. ). In fact, ΔH showed large amplitude oscillations with a periodicity of about 30 min during 09:30–11:20 UT (Fig. b). However, the electron density at 150–200 km altitudes showed irregular oscillations with a different periodicity (∼15 min), although no modulation is seen at 100 km in altitude. The periodicity in the ion convection is also about 15 min at 150–200 km altitudes. The only candidate that may match with the 30 min periodicity is irregularity in the ion velocity at 100 km altitude, but the profile does not really match with the ΔH variation. This suggests that the Pc5 pulsation can be modulated by the density variation at 150–200 km altitudes.

The large extra ΔH at 68–75∘ GGlat might, if it happens a during substorm, even be relevant to the space weather hazards because a large sunspot may cause a coincident occurrence of strong solar flare and large substorm powered by the CME. If the crochet mechanism can interact with a substorm, such as reinforcing each other, and if the strong solar flare takes place within 1 h after CME hits the Earth, then we expect extremely strong ionospheric currents and resultant ground-induced currents (GIC) that are hazardous. Thus, crochet study is potentially related to other research fields.