Skywarden (Taivaanvahti in Finnish, Himlakollen in Swedish; https://taivaanvahti.fi, last access: 25 August 2026) is a database of observations of celestial phenomena (aurora, solar halos, deep-sky objects, noctilucent clouds, thunderstorms, eclipses, fireballs …) made by citizen scientists. It has been created, developed and maintained by the Ursa Astronomical Association in Finland. It was established in 2011 but it also includes a handful of historical observations which have been digitised recently; the oldest ones are a series of photographs of the aurora taken in Sodankylä in late 1927 – probably the oldest aurora photographs taken in Finland . Skywarden has already proved useful as a database for studies investigating proton aurora and stable aurora red arcs as well as the May 2024 geomagnetic storm .

We used the Skywarden application programming interface API; to search for observations made by citizen scientists during which dunes were identified. We carried out the search between 1 January 2000 and 31 December 2025, and restricted it to observations containing at least one photograph where the dunes were visible. Dune observations were confirmed through three steps:

The observers themselves, in most cases, indicated that their pictures contained structures that may be dunes. Some experienced observers are familiar enough with the dunes to reliably identify them in their material, whereas others may be unsure about it but were encouraged to review their observations aided with freely available material providing guidelines for auroral form identifications – both integrated into Skywarden itself and as open material such as the ARCTICS Aurora Field Guide .

In addition, we also collected all the aurora observations reported in Skywarden during the same time interval (1 January 2000–31 December 2025), to serve as a baseline when interpreting the seasonal and local time distributions of reported dune aurora events. To extract these data from Skywarden, we also used the API; we did not restrict the search to observations containing at least one picture in this case, nor did we visually inspect all the reports. In total, we obtained 16 479 aurora observation reports from 2269 individual contributors. Out of these, we retained the 9927 observation reports having both a start and end time, which took place on 1560 individual days.

For this study, we examined geomagnetic activity using two of the SuperMAG indices, calculated from ground-based magnetic measurements at 1 min time resolution by magnetometers from the SuperMAG network . We used the SuperMAG Electrojet (SME) index as a measure of substorm activity, and the SuperMAG Ring current (SMR) index as a measure of geomagnetic storm activity. We downloaded the SME and SMR indices at 1 min resolution from January 2000 to December 2025.

We retrieved solar wind and interplanetary magnetic field (IMF) measurements at 1 h time resolution between January 2000 and December 2025 from the OMNI database . The OMNI data consist of solar wind parameters and IMF observations made by various spacecraft at the L1 Lagrange point of the Sun–Earth system and propagated to the Earth's bow shock, as well as selected geomagnetic indices . For this study, we considered the following OMNI parameters: solar wind proton number density, speed and dynamic pressure, as well as total IMF magnitude, and its By and Bz components in the Geocentric Solar Magnetospheric (GSM) coordinate system.

We use 10 s resolution data from the International Monitor for Auroral Geomagnetic Effects (IMAGE) network to examine ionospheric equivalent currents associated with dune aurora observations. IMAGE currently consists of 57 ground-based magnetometers located in Fennoscandia and surrounding areas. It enables evaluating the ionospheric equivalent currents with a spatial resolution down to about 100 km, hence resolving the so-called mesoscale current structures. One method to derive equivalent currents, known as the Spherical Elementary Current System method (SECS), was introduced in and has been widely used since then e.g..

Figure  shows an overview of all the dune observations reported to Skywarden between January 2000 and December 2025. In practice, the earliest observation available at the time of this study was made in 2004, and it is the only one prior to January 2012. The time span of more frequent dune observations still covers approximately one solar cycle. In Fig. a–d, the colour of the dots indicates the month when the observation took place.

The dune aurora reports come from three main regions (by decreasing number of observations): Northern Europe, Australia and New Zealand, and North America. These regions have particularly active aurora chaser groups exchanging knowledge and information via social media pages. These groups contribute to Skywarden – but also to other platforms such as Aurorasaurus – with their observations. In the North American sector, there are a few observations from Manitoba, Alberta, and Michigan made during the northern-hemisphere autumn, winter, and spring. In the European sector, while the vast majority of reports come from Finland, there is also a number of them from Sweden, and a couple from Norway, Denmark, and Scotland. Most observations were made during the northern-hemisphere autumn and winter, and a few of them in early spring. In New Zealand and Australia, almost all observations were made in May; these correspond to the May 2024 geomagnetic storm. In all three sectors, dune aurora observations were predominantly made near or below 60° geomagnetic latitude, that is, at latitudes usually considered subauroral.

Figure e–g show the same observations but with dots coloured according to the magnetic local time (MLT), calculated with the AACGM-v2 Python library . When it is not possible to determine accurately enough the time stamps in a given Skywarden observation (this is the case for 28 observations), the MLT is not calculated, and the corresponding observations are not shown in these figures. When multiple dune images were included in a given observation report, or when the observer provided the time interval during which they photographed dune aurora, we retain the earliest time with a dune image to calculate the MLT. While MLT hours range from 17.5 to 1.2 across individual dune observations, one can see that most observations started in the pre-midnight sector.

In Fig. a, we show the MLT distribution of individual dune events. A dune event is defined such that observations that took place on the same night and within the same longitude sector (within 40°) are considered forming part of the same dune event. This results in a total of 61 dune events, for which we determine the earliest and latest MLT associated with dune observations. The histogram shown in Fig. a is built in such a way that, for each 1 h bin, we count how many dune events include observations within that bin. For instance, if an event includes an earliest dune observation at an MLT of 18.2 and a latest one at an MLT of 23.5, it will increment the following histogram bins: 18–19, 19–20, 20–21, 21–22, 22–23, and 23–0. It is clear that the large majority of dune events include observations predominantly between 18:00 and 01:00 MLT (89 % of them before magnetic midnight), with a peak within 21:00–22:00 MLT.

In Fig. b, we show the monthly distribution of individual dune events. This distribution exhibits a main peak in October (15/61 events, 25 %), a secondary one in March (11/61, 18 %), and contains the majority of the remaining events in the other months of the northern-hemisphere winter and early spring (November to April). There are only a couple of events in May, August, and September, and no events in June and July. This absence of events during the Northern-hemisphere summer months is to a great extent explained by the fact that most Skywarden contributors are based in Northern Europe, where nights are bright during that time. It is only recently (particularly after the May 2024 geomagnetic storm) that Skywarden became known to the Southern-hemisphere aurora chaser communities, which is why there are currently only a few dune events reported from those latitudes that could in principle bridge the gap during those months (southern-hemisphere winter).

Citizen science observations are not continuous and bear some bias related to the typical human behaviour, such as staying awake past sunset, sleeping at night, and waking up around or after sunrise. Aurora chasers are also more likely to go out and collect data when a geomagnetic storm is predicted to occur, and may not stay outside for the entire duration of the auroral displays, which can lead to an over-representation of earlier MLTs in the collected observations. As it is known that the occurrence rate of auroras is not evenly distributed throughout the year, we compare the MLT and monthly distributions of dune events with a baseline given by all the available aurora observations from Skywarden. Figure c shows the MLT distribution of the 9927 aurora observation reports having a start and end time retrieved between January 2000 and December 2025, the histogram bins being incremented similarly as for the dune events (see above). While a slight bias can be seen for aurora observations taking place in the earlier half of the night, the overall distribution is nonetheless relatively symmetric and has its peak between 23:00 and 00:00 MLT, i.e. two hours later than that of the dune observations. This suggests that the dunes are predominantly a pre-midnight phenomenon, with an increased likelihood to occur in the few hours after 18:00 MLT. The monthly distribution of all aurora event dates is given in Fig. d. Here, for simplicity, we simply retain individual days with at least one aurora report in Skywarden without considering the longitudes, such that, for instance, the 11 May 2024 is counted only once. A similar figure but considering only the Skywarden reports originating from the European sector is given in Appendix  (Fig. ). We can see that the distributions are largely similar to those given in Fig. , the main difference being the number of dune events in March, of which a noticeable fraction originates from outside Europe.

The histogram shown in Fig. d reveals two peaks around the equinoxes (March and September). Notably, 74 % of dunes events and 95 % of all aurora events in Skywarden were observed in Finland. This highlights the importance of analysing local observation conditions, such as cloudiness and darkness conditions, when interpreting the monthly distribution of observed events. In Appendix , Fig.  shows the monthly average of cloudiness in Finland, obtained from the Finnish Meteorological Institute past weather database. We consider nighttime cloudiness expressed in okta (a scale quantifying the fraction of the sky covered with clouds, ranging from 0: fully clear, to 8: overcast). Nighttime is defined when the solar elevation at ground level at a given meteorological station is at least 10° below the horizon (i.e. the solar zenith angle is greater than 100°). Across Finland, the average cloudiness during the darkest winter months (from November to January) typically ranges between 6 and 7 okta, whereas observing conditions are better in spring (March and April) and early autumn (September). This likely explains to a great extent the peaks in auroral observations reported in Skywarden in March and September (Fig. ), in addition to the Russell–McPherron effect which leads to enhanced geomagnetic activity during equinoxes. Besides, as mentioned earlier, the deep trough in dune and auroral observations between May and July is due to the fact that, during that period of the year, nights are not dark enough in Finland for the aurora to be visible. Given that the vast majority of the reports in Skywarden are from Finland, this effect greatly shapes the monthly distributions shown in Fig. b and d.

Interestingly, September is a month for which only two dune aurora events have been recorded (Fig. b), whereas the largest dune event occurrence peak is in October (15 events, 13 of which are from Europe according to Fig. b). The majority of dune events in October were indeed observed in Finland, which has a monthly average nighttime cloudiness close to 6 okta (in comparison, for September it is under 5 okta). These relatively unfavourable observation conditions suggest that it is likely that recorded events contain only a fraction of all dunes events occurring in October. We also note a low number of dune events in February, not reflected in the all-aurora distribution. If not simply due to too small a number of events for the monthly statistics to be robust, the dominance of October and relative lack of dune events in February and September may suggest that the dunes require conditions that are not only related to geomagnetic activity. A possible factor leading to a lower number of dune events in September compared to October is the fact that the early evening hours, when the dunes are typically observed, are not dark in early September. The fact that equinoxes occur around 20 March and 23 September rather than in the middle of those months implies that evening hours are slightly darker in March than in September, hence making it more favourable for the dunes to be observed in March than in September. Nevertheless, this bias in darkness conditions cannot alone explain the low number of dune events in September, as April – when the Sun sets at later hours than in September – has noticeably more (5, all from Europe) dune events than September. Hence, the September–October difference in dune event number might also, for instance, have something to do with atmospheric dynamics which have their own seasonal patterns.

Citizen scientist observations of the aurora are not continuous and hence they are unlikely to capture the dune events in their entirety. However, the large database provided by Skywarden observations enables us to determine conservative lower-boundary estimates of dune observation durations, when either multiple photographs are appended to the Skywarden record or the observer indicated in the description text within which time range they have dunes in their images. Out of the 308 Skywarden dune observations used in this study, this is the case for 151 of them.

Figure a gives the distribution of dune observation durations as determined from individual Skywarden reports. While approximately a third of them (53/151) were photographed for less than 15 min, the distribution exhibits a gradual decrease with increasing durations, including a non-negligible tail beyond 1 h amounting to about 15 % (23/151) of the observations. Three observations included dune photographs for more than three hours; the first one was reported from Finland on 23 October 2022 (3.2 h), the second one from Tasmania on 11 May 2024 (4.2 h), and the third one from Finland on 12 December 2025 (3.1 h). Multiple factors need to be met for the totality of a dune display to be recorded by a single observer. For instance, the observer needs to be photographing before the dunes appear and until they have vanished, and they need to keep the camera pointed in the direction of the dim dune aurora rather than turn it towards brighter and more dynamical auroral structures. Therefore, we consider the distribution shown in Fig. a a lower-boundary estimate of the true duration of the corresponding dune displays. To get a more comprehensive overview of the duration of dune events, we can combine the information contained in multiple observations of the same event. Figure b shows the distribution of dune event durations based on multiple (when available) reports by observers. Only dune events for which either several reports were made or a single report containing information on the duration of dune presence in images are shown in the figure. This is the case for 47 out of the 61 dune aurora events considered in this study. We can see that only half of these events (23/47) have a duration less than an hour, ∼ 20 % (9/47) of them were photographed for between one and two hours, and about a third of them (15/47) have been recorded for more than two hours. Five events are of particularly long duration: the 7 October 2015 event observed from Finland and Scotland (3.8 h); the 17 December 2023 event observed from Finland (4.1 h); the 13 March 2022 event also observed from Finland (4.8 h); the 20 January 2016 event observed from Finland, Norway, and Scotland (5.2 h); and the 11 May 2024 event observed from New Zealand and Australia (6.9 h). This confirms that dune aurora can cover a wide region and last for multiple hours, as was noted in .

A version of Fig.  for European observations only is provided in Appendix  (Fig. ). No major differences can be noted between the two figures.

To assess whether dune aurora is associated with specific solar wind driving conditions, we compare the distribution of selected solar wind and IMF parameters (solar wind proton number density, speed, dynamic pressure, IMF magnitude, and IMF By and Bz components in the GSM coordinate system) associated with dune events and associated with all the aurora observations from Skywarden. This is shown in Fig. , where histograms corresponding to dune events are given in green whereas those for all aurora observations are in black. The total number of data points is given in the top-right-hand corner of each panel, as are the values of the lower quartile, median, and upper quartile.

For every dune event and Skywarden aurora report, the driving parameters are retrieved from the 1 h OMNI data. We retain the average between the value of each parameter two hours and one hour prior to the start time of the observation (that is, two values), to account for the time delay between the arrival of the measured solar wind and IMF at the Earth's bow shock (as given by OMNI data) and their effects in the magnetotail e.g.. For the dune events (green histograms), we consider the driving conditions during the two hours preceding the earliest dune aurora observation (average value during the two-hour window), such that a given event is associated with a single value of the solar wind and IMF parameters.

Figure a–f suggest that dune aurora observations do not tend to take place when solar wind driving is significantly different from during typical aurora observations. The only possibly meaningful difference is found in the solar wind density, with a median value for dune aurora events of 8.8 cm⁻³, compared to 7.2 cm⁻³ for all-aurora observations. As we will see in Sect. , this difference is however not statistically significant. Regarding the IMF, we note that the total magnitude distributions exhibit a similar situation as the solar wind density, with a slight tendency for dune events to be associated with higher values (median of 9.6 nT) than all-aurora observations (median of 8.6 nT). Differences in IMF By and Bz distributions are marginal. Here too, the differences are not statistically significant. For comparison, we made a similar figure considering only the observations from the European sector (Fig.  in Appendix ), which shows very similar distributions as those presented in Fig. .

Next, we examine the distribution of dune aurora observations with respect to geomagnetic latitude, together with geomagnetic activity as measured with the SME and SMR indices. Figure a shows the geomagnetic latitudes of dune observations as a function of time, with red stars corresponding to northern-hemisphere latitudes and blue stars corresponding to southern-hemisphere latitudes. This way of displaying the data enables one to visualise the latitudinal spread of dune observations during individual events. We can see for instance observations spanning more than 5° in geomagnetic latitude during the 7 October 2015 event (57.2–64.0° N), the 7 October 2018 event (55.4–60.4° N), the 14 January 2022 event (54.0–63.9° N), the 17 December 2023 event (57.6–63.7° N), and the 11 May 2024 event (47.9–54.4° S). This suggests that the dunes may be relatively widespread during some events, although one needs to take into account the uncertainty in their exact location (see discussion in Sect. ). In Fig. b, the same data are displayed as a histogram, peaking at latitudes a few degrees below 60°. Considering the data from both hemispheres, 92 % of the dune observations were made at a geomagnetic latitude below 61°. For comparison, the geomagnetic latitude distribution of all aurora observations retrieved from Skywarden is shown in Fig. c. While it is difficult to assess whether the fact that this distribution peaks slightly polewards from that of the dune observations is significant, we note that it contains a more prominent tail beyond 60°, and the fraction of reports from below 61° geomagnetic latitude is 74 %. Keeping in mind that these distributions are certainly biased by the over-representation of observations from southern Finland compared to other regions of the world, this is consistent with the hypothesis that dunes may be a phenomenon more often observed in the equatorward part of the auroral oval.

Figures d–i present, in a similar format as the above panels, the distribution of the SME and SMR indices during the dune and all-aurora observations. These two indices are measures of substorm and geomagnetic storm activity as they are proxies for auroral electrojet and ring current, respectively. Since SuperMAG indices are provided at 1 min temporal resolution, and given that the aurora observations do not have such a high temporal resolution, we retain a 30 min average of the SME and SMR indices around the start time of each observation (i.e., between 15 min before and 15 min after that time stamp). The main results that can be inferred from these figures are the following:

During a dune event, SME and SMR values can change significantly (see Figs. d, g). This implies that the dunes span across not only a large geographic zone as discussed above, but also during a significant time interval. This is consistent with the findings from Fig.  that dune events can last for up to several hours, a time over which SME and SMR can vary significantly.

A similar figure considering only Skywarden reports from the European sector is given in Appendix  (Fig. ). We can see that, while the distributions overall resemble well those shown in Fig. , the strong-activity (in terms of SME and SMR) tails are absent: there are no dune observations associated with SME>1800 nT nor with SMR<-200 nT. A likely explanation is that the North American and Oceanian dune observations, mainly originating from lower geomagnetic latitudes than the majority of those from the European sector (see Fig. ), require higher geomagnetic activity to bring the auroral oval within the field-of-view of the observers. This is a further indication that the dunes can appear under various levels of geomagnetic activity, from moderately enhanced to fairly extreme.

We evaluate the statistical significance of the differences between the distributions of dune events and the distributions of all aurora observations from Skywarden by carrying out permutation tests. For each considered parameter (MLT, monthly occurrence, solar wind density, solar wind speed, solar wind pressure, IMF magnitude, IMF By and Bz components, and SME and SMR indices), we assess whether the differences in key properties of the dune histograms (green) and all-aurora histograms (black) can be explained by chance, if sampling a small number of reports (61, i.e. the number of dune events) out of all the Skywarden reports of aurora observations.

We perform each permutation test with 20 000 iterations based on the histograms shown in Figs. , , and as follows:

We quantify the difference (“effect size”) between dune and all-aurora distributions according to a selected statistic (circular mean for MLT, total variation (TV) distance for monthly distribution, median for all the other parameters). Given two histograms P and Q, TV is defined as 12∑i|Pi-Qi|; it varies from 0 (fully identical distributions) to 1 (fully disjoint distributions), with intermediate values giving the fraction of total variation.

The results of this analysis are summarised in Table . We can see that the difference in the circular mean of the MLT of dune aurora events compared to all aurora observations is highly significant in the statistical sense, as are the differences in the median values of the SME and SMR indices. Although to a lesser extent, this analysis suggests that the monthly distribution of dune aurora events differs significantly (by 28 % in total variation) from that of all aurora observation reports from Skywarden. However, none of the solar wind and IMF parameters passes the significance test. This indicates that it is unlikely that dune aurora events are associated with a set of driving solar wind and IMF conditions that stand out from those associated with the whole range of auroral observations reported in Skywarden. A similar analysis of statistical significance of differences in distributions for dune aurora compared to all aurora observations, but restricted to observations from the European sector, is presented in Appendix  (Table ). The conclusions are the same.

One should in addition keep in mind that this analysis assesses the statistical significance of the differences between dune aurora events and all-aurora observations as reported in Skywarden, which are themselves affected by biases at various levels. Those results should therefore not be interpreted as strictly depicting the statistical occurrence rates and properties of dune aurora intrinsically, but rather in terms of observations thereof as reported in Skywarden. Inferring the intrinsic statistical properties of dune aurora would require being able to accurately model the observational effort, which is affected by citizen scientist behaviour, weather conditions, social media dynamics, equipment sensitivity and settings, among other parameters. Quantifying this observational effort and normalising the dune parameter distributions accordingly would prove an extremely complex task that lies beyond the scope of this study.

In the study of the first dune event by , it was noted that the dunes occurred in concert with a strong eastward electrojet. Here, we leverage the larger dune event database provided by Skywarden reports to assess whether this is a typical situation for all dune events.

We first identify the dune events for which at least three clear dune observations have been made in the same geographic area, such that we have an indication of a nonzero time period over which the dunes were photographed. We only retain dune observations from the Fennoscandian sector, where ground-based magnetometers from the IMAGE network are close enough to each other to resolve mesoscale patterns (that is, ∼ 100 km horizontal scales) in equivalent currents. When, for a given dune event, the observations span more than 6° in geographic longitude (∼330 km in the east–west direction for geographic latitudes around 60°), the observations are separated into subgroups according to their geographic longitude (in longitude bins of equal width). In three cases, the observations are split into two subgroups, and in one case (spanning 15.9° in geographic longitude) they are split into three subgroups. We obtain 15 sets of observations (corresponding to 10 individual dune events) for which the equivalent current analysis can be carried out.

Table  summarises the results of the equivalent current analysis. The first column gives the date and time interval within which the dunes were photographed during the retained observations. The second column indicates the number of observations, for which the mean geographic coordinates of observers are given in the third column. The fourth and fifth columns indicate whether the dunes were observed in conjunction with the eastward (E) or westward (W) electrojet and with the Harang discontinuity, respectively. This was assessed by visual inspection of the equivalent current figures, some examples of which are given in Figs. –.

In practice, for each set of observations, the SECS analysis method is carried out using all the available IMAGE stations. The results are presented in the form of keograms of the geographic east (ϕ) and south (θ) components of the yielded equivalent current density (Jeq) as a function of time and geographic latitude, for the mean longitude of the dune observations, as exemplified in Fig. a–b for the 12 December 2025 event observations from the Finnish sector. Additionally, equivalent current maps can be displayed (see Fig. c), wherein horizontal current direction and magnitude are indicated with black arrows and the curl of the equivalent currents, which is a proxy for field-aligned current density e.g.,, is shown as the background colour. In the following, we look in more detail at such results for three selected events exemplifying the various configurations with respect to the eastward and westward electrojets.

It appears from Table  that there are three types of situations when it comes to the coincidence of the dune aurora with the auroral electrojets. In 10 cases, the dunes are found to occur within the eastward electrojet, like in . However, in one case, they are taking place within the westward electrojet, and in the remaining four cases they start within the eastward electrojet but remain visible throughout the transition to the westward electrojet. Besides, all events appear to take place in the vicinity of the Harang discontinuity.

Figure  shows an example of a set of dune observations associated with the eastward electrojet and a clear Harang discontinuity. In Fig. a and b, the keograms have been obtained along the meridian whose longitude corresponds to the average of the observers' longitudes. Note that the horizontal line appearing around 68° latitude in Fig. b is an artifact produced by the analysis, which does not affect the region we are interested in here. The mean geographic latitude of observers' locations is indicated with a horizontal dashed line, and the two vertical dashed lines mark the earliest and the latest dune photograph from that set of observations, respectively. Throughout the time interval during which dunes have been photographed from Finland, the observers are close to the equatorward edge of the eastward electrojet. It is important to note that the dunes may have been present before the earliest reported observation and after the latest one, and that it is also possible that the dunes were not necessarily visible at all times during the marked time interval.

Figure c shows the equivalent current map at a time approximately halfway through the set of observations. The large-scale features of the reconstructed equivalent currents clearly exhibit a well-delimited and relatively strong (scale shown in the top-left corner of the panel) eastward electrojet between the region-1 (upward, in red) and region-2 (downward, in blue) field-aligned currents. Near the eastern edge of the resolved area, the reversal in the horizontal current direction with increasing latitude indicates the presence of the Harang discontinuity. The individual observers' locations are shown with white stars. To provide a measure of the uncertainty in the actual location of the dunes, we have added a circle around each star, with a radius of 463 km corresponding to 10° elevation at 100 km altitude (see also this point in the Discussion, Sect. ). We can see that, despite the uncertainty in the exact dune aurora location, for all observations they are unambiguously located within the eastward electrojet. The yellow star indicates the mean location of the observers, which is used for making the keograms in the other two panels of the figure.

In a few cases, the dune observations start within the eastward electrojet but continue after transitioning to the westward electrojet. An example of such events is given in Fig.  with the observations of the 14 January 2022 event within the Swedish sector. The eastward component of the equivalent currents near the observers' mean location (Fig. a) is positive until ∼ 20:40 UT, after which it is negative, corresponding to a westward current. We note that the last reported observation of dunes in the sector (at 21:32 UT, vertical dashed line) seems to match the disappearance of a patch of eastward current at high latitudes (above 65°). This might however be a mere coincidence. Figure c shows the equivalent current map at 20:16 UT. A clear Harang discontinuity signature can be seen to the north-east from the observers' locations.

Finally, an example of dune observations reported in conjunction with the westward electrojet is given in Fig. . Four observers from Finland shared photographs of the event, which took place on 26 February 2023 and had observations only during a relatively brief time interval (22 min). The SECS analysis unambiguously indicates that, at the time of those observations, the Finnish sector was within the westward electrojet (Fig. a and c). However, it cannot be excluded that the dunes were present at an earlier time but were simply not observed.

These three examples reveal that, while dune aurora seems to favour eastward electrojet conditions to appear, it can also be seen within the westward electrojet. It is important to stress that, as the dunes are generally accompanied by brighter auroral displays, the spatio-temporal patterns in the electrojet currents likely reflect the dynamical changes in the active auroral structures rather than those of the dunes themselves.