An overview of auroral evolution in magnetic local time (MLT) outlined by and recently revisited by describes the evening sector as being most populated by auroral arcs, while the midnight and morning sectors are characterized by structurally most complex aurora: the midnight sector aurora driven by geomagnetic activity and the morning sector aurora often appearing irregular and patchy. On the dayside, the auroral emission equatorward of the polar cap boundary is predominantly green, while poleward of the polar cap boundary, on the newly opened field lines, the auroral emission is primarily red .

In an earlier study of auroral occurrence, showed that the occurrence of all auroral observations peaks around magnetic midnight. They analysed visually classified auroral types in image data from eight auroral all-sky cameras (ASCs) in 1973–1997. Six of the eight cameras were located in the Fennoscandian Lapland under the main auroral oval, while one camera was installed on Svalbard and another one in the sub-auroral zone in Finland. Their visual classification of the auroral displays divided auroral structures into arcs, rayed aurora, patches, and other auroral forms. The resulting MLT distribution showed a slight dusk dominance for auroral arcs while both patches and other auroral forms were most frequently observed in the dawn sector. A similar MLT distribution of arcs, patches, and omegas reported by was based on 220 000 automatically classified images from the Canadian camera station in Gillam in 1993–1998. The smallest class was omega bands consisting of 600 images, while the other two auroral classes included thousands of images. The highest auroral arc occurrence was seen at around 21:00 MLT, the number of omega bands peaked at about 02:30 MLT, and patchy aurora occurred most frequently at around 06:00 MLT.

As shown by the above mentioned studies, the “known” auroral classes were small compared to the class of “other auroral forms”. To deal with the “unknown” auroral forms, an automatically calculated structural index called auroral arciness was introduced by . Using this index they were able to include all auroral forms in a long-term analysis instead of only the handful of commonly accepted morphological classes (i.e. an automated version of the approach). They concluded that auroral arcs are always part of auroral displays in both hourly and annual averages. They found a positive correlation with auroral arc occurrence and solar wind driving, although the occurrence of more complex structures varied much more than that of arcs. The arc occurrence maximized in the dusk sector at about 17:00–20:00 MLT, and then steadily decreased towards midnight and dawn, while the portion of more complex aurora increased in MLT after about 21:00 MLT. This temporal evolution is in a good agreement with previously mentioned studies of the MLT evolution in auroral displays. The arciness index was later used in a substorm study to examine the morphological evolution of the aurora during different phases of substorms . Auroral arcs were also found in all substorm phases as a basic element of the aurora, but a clear morphological transition was observed to take place at the substorm onset. Growth phases were found to be more arc-dominated while the auroral display changed into more complex auroral structures within minutes after the magnetic substorm onset.

The morphology of the dayside aurora has mainly been studied using the auroral camera data from Svalbard in arctic Norway. For instance, classified image data from Ny-Ålesund station at 79∘ N in 2004–2009 into four morphological categories. The class of auroral arcs occupied one third the whole dataset. The arc occurrence increased to 60 %–70 % towards 17:00 MLT in the dusk sector and the fewest arcs were observed around midday.

Measurements of auroral peak emission (or lower border) heights are sparse in the literature and primarily consists of case studies of different auroral arc-like structures, as recently summarized by . The nightside and evening arcs are found at 110–150 km heights based on either emission or electron density measurements e.g., while taller rayed arcs, high-latitude arcs, or morning sector arcs have been observed to occur higher, at about 130–170 km e.g..

To facilitate large-scale emission height studies, an automatic method was developed for auroral images by . This method uses simultaneous auroral all-sky images from neighbouring camera stations with overlapping fields-of-view (FoVs). The auroral emissions are projected onto both horizontal and field-aligned planes, and the height value is determined by comparing results between the different projections as well as with correlating the projections from the two stations. Applied to the Fennoscandian MIRACLE (Magnetometers–Ionospheric Radars–All-sky Cameras Large Experiment) network camera data, the automatically calculated peak emission heights were used to study pulsating aurora (PsA) in contrast to other types of aurora preceding the pulsating structures. This study consisted of about 400 PsA events showing that the median peak emission height decreased by about 8 km at the beginning of PsA. The same conclusion was achieved for both green (557.7 nm) and blue (427.8 nm) auroral emissions. This result provided evidence for PsA being associated with higher-energy electron precipitation compared to the types of aurora preceding PsA, which was concluded to be substorm-related auroral displays. Because PsA is a very common type of aurora in the morning sector, lasts for several hours, and covers a large range of latitudes and local times, it is expected to be a dominant feature in the morning sector aurora causing lowering of the peak emission heights.

In this study, we analyse the magnetic local time (MLT) distribution of auroral arciness and peak emission heights calculated for both Lapland and Svalbard image data. These datasets include more than a decade of systematic imaging from both latitude regions. By employing the same methodology on similar data, we can directly compare the dayside and nightside aurora, as well as investigate the dawn–dusk symmetry. We examine the effects of solar wind driving and level of magnetic activity on auroral morphology and emission heights, and we further study the contribution of PsA to the auroral behaviour in the morning sector. Our motivation is to understand the climatology of the auroral emission height and structure evolution. Furthermore, as the auroral peak emission height carries information about the energy of the precipitating electrons, the MLT distribution of emission heights in this large dataset will contribute to the knowledge of electron acceleration mechanisms as well.

The core dataset of this work consists of the Magnetometers–Ionospheric Radars–All-sky Cameras Large Experiment (MIRACLE) network all-sky camera (ASC) images . The network is operated and maintained by the Finnish Meteorological Institute (FMI) in collaboration with Sodankylä Geophysical Observatory (SGO) and INAF-IAPS Institute for Space Astrophysics and Planetology in Italy. We use images from the five Lapland stations (SOD, MUO, ABK, KIL, and KEV) during the winter seasons in 1996–2006 and the two Svalbard stations (LYR and NAL) during the winter seasons in 1999–2012 (for geographic locations and fields-of-view see Fig. ). All seven cameras were identical in design and largely run on the same imaging mode. These ASCs were equipped with narrow bandpass filters for auroral emissions at 630.0, 557.7, and 427.8 nm, so-called red, green, and blue, respectively. While red and blue images were taken once a minute, green images were captured once every 20 s whenever the Sun was more than 10∘ below the horizon. This imaging rate resulted in about one million green images per station per winter season, which spans from September until April in Lapland and from November until March in Svalbard. All images were automatically pruned , and our analysis only consists of pruned images, namely those containing aurora.

This study further utilizes previously published data products of auroral peak emission heights and auroral arciness . They both require overlapping fields-of-view (FoVs) of two neighbouring ASCs and are thus called “paired data”. Any height value from any camera pair in Lapland is considered as a Lapland auroral height in this study without specifying which ASC pair was used, because the FoVs overlap so much that any station pair works well. The full all-sky FoV is reduced to exclude elevation angles below 10∘. Auroral peak emission heights are automatically determined by projecting the emission distributions to both a horizontal plane and many field-aligned planes . When the height values from the two projections differ by less than 20 km, the heights from the field-aligned projection are used for further analysis. While the maximum height difference between the two projections is implemented to exclude outliers, such as unfavourable location of the aurora low in the horizon, our statistical results are not sensitive to the choice of that value, since a height difference of 10 km gives the same results. Fifty percent of height values used in this study differ by a maximum of 7 km. For the Lapland ASC data in 1996–2006, the number of reliable peak emission height values was about 397 000, while the Svalbard ASC data in 1999–2012 gave about 83 000 reliable height values. One height value corresponds to the dominant auroral feature within the common FoV of one image pair. While this means ignoring the peak emission height variations along auroral structures e.g., this dataset provides a sufficient overview for large statistical studies, which is the aim here.

Auroral arciness is an index describing how arc-like an auroral display in an image is. This structural indexing is based on clustering of the brightest pixels in ASC images and fitting polynomials into the bright pixel clusters . The arciness then describes the spread of the bright pixels around the polynomial fits. All arciness data we use are also paired data, so essentially data (image pairs) which also gave peak emission height values. Since the pruning method , like any automatic classifier, is not perfect, there would always be some images of illuminated clouds, for instance, which have been wrongly interpreted as aurora. Although there is strictly speaking no paired data requirement for arciness calculation, this selection guarantees that all analysed data actually contain aurora which is seen from two stations at the same time. The paired dataset adds up to about a million images analysed for arciness and height over both Lapland and Svalbard. The arciness was described in detail and statistically analysed for the Lapland data by . For Lapland arciness, we use the arciness data from the central station KIL in this study, which consists of a total of about 357 000 values. The same procedure is repeated for the Svalbard image data, which provides the high-latitude stations and the dayside contribution to the auroral morphology. The number of Svalbard arciness data is based on the number of images from the Longyearbyen station with the total of about 183 000 arciness values.

To determine the solar wind driving conditions, we use the OMNIWeb data for solar wind speed and interplanetary magnetic field (IMF). These data are propagated to the magnetopause and have a temporal resolution of 1 min. We divide the solar wind driving into high and low conditions, which are defined as solar wind speed less than 400 kms-1 or positive IMF BZ for low solar wind driving and solar wind speed larger than 500 kms-1 or negative IMF BZ for high solar wind driving.

We further use the lower envelope curve of the auroral electrojet index AL, to assess high and low geomagnetic activity conditions. The global AL index data are downloaded from the World Data Centre in Kyoto in 1 min temporal resolution. We define high geomagnetic activity as AL index values less than -300 nT and low geomagnetic activity as AL index values larger than -100 nT.

This study on the MLT dependence of auroral peak emission heights and auroral structures is, to our knowledge, the first large-scale study of its kind, employing about half a million images from two different latitude regions over about a decade. Although the overall findings about the median peak emission height behaviour are not surprising, they provide a solid foundation for the often assumed emission heights of 110–120 km for the nightside green aurora and 130–140 km (or higher) for the dayside green aurora. While the measurements of the peak emission heights tend to be limited to case studies e.g., the studies of the particle precipitation energies e.g. tell a similar story, since 2–5 keV electron precipitation corresponds to a stopping height of about 110–120 km .

The MLT distribution of peak emission heights behaves very similarly for green (Fig. ) and blue emissions (Fig. ). As pointed out by : “The fact that the green and the blue emission heights undergo the same behavior suggests that the green emission, at least in case of pulsating aurora, comes primarily from an energy interchange process between atomic oxygen and molecular nitrogen. This would make the peak emission heights of the green auroral emission also carry information on the precipitation energy.” This is demonstrated to be a valid process for all auroral structures in a companion study by , which analyses the statistical difference between the peak height of green and blue emissions.

Based on the long-term arciness evolution reported by , the solar wind driving conditions can bias the distribution of auroral structures. This, however, is considered to be an insignificant bias in this study, since we only examine the relative portions of Arcs and Other more complex aurora, both of which are present in any solar wind conditions, in any MLT sector and in any substorm phase . The outstanding auroral years of 2003 with very high average solar wind speed and magnetic activity, and 2008–2009 with extremely low solar wind driving and magnetic activity would cause the largest offsets in any auroral datasets. All these extreme years are included in our Svalbard dataset, while the lowest activity years are beyond our Lapland dataset. Thus, the Lapland dataset of arciness may be biased towards more complex aurora as compared to the Svalbard arciness. In the nightside peak emission heights, this may mean a lower average peak emission heights for Lapland compared to the Svalbard data. In these large amounts of data spanning over a decade, however, the solar wind speed distributions are the same for both time periods (1996–2006 for Lapland and 1999–2012 for Svalbard) within a few kms-1.

On the nightside in particular, the fast solar wind is associated with lower peak emission heights. This can be interpreted as solar wind high-speed streams driving higher substorm occurrence rate, more intense substorms, and higher energy dissipation rate into the ionosphere e.g.. Our findings of lower peak emission heights suggesting more energetic particle precipitation during fast solar wind is in agreement with the increased substorm activity during the same conditions. The IMF polarity has its largest effects in the dawn, noon, and dusk sectors, leaving the nightside unaffected. In the dawn and noon sectors, the northward IMF corresponds to lower peak emission heights compared to the southward IMF, which we interpret as contraction of the auroral oval leading to a more energetic precipitation source region moving into the ASC FoV. This is particularly clear at noon where the cusp region of soft precipitation is narrow in latitude, just poleward of the more energetic boundary layer particle sources, and sensitive to the IMF polarity . The nightside auroral oval is wider and the Lapland ASC FoVs are well within the band of discrete aurora. Nightside is thus less sensitive to the expansion and contraction of the auroral oval due to IMF changes with respect to our observation sites. In the dawn sector, where the diffuse aurora extends farthest equatorward during low solar wind driving , the contraction of the auroral oval may cause the typical emission heights to decrease. The PsA events occur mainly during the southward IMF, but as about one fourth of the events last into the periods of the northward IMF, these events can also contribute to the lowering heights during the northward IMF. The dayside auroral emission is weaker than the nightside auroral emission, and the imaging season in the high-latitude polar night is also much shorter than the auroral season over Lapland. Thus, even for a decade-long time series of auroral image data, the number of pruned images (containing aurora) is limited when the data are binned according to other parameters, such as the solar wind speed and the IMF.

Structural changes in the mainland data are most likely due to substorm activity in the midnight sector. This observation is expected and in agreement with the general conclusion of the long-term morphological investigation by , who concluded that the occurrence of active auroral forms follows the solar and geomagnetic activity better than the occurrence of auroral arcs does. Our results confirm that this is also the case in the statistical MLT distribution. Furthermore, this finding is in a very good agreement with the arciness evolution over the substorm onset , where the arc-dominated growth phase changes into a complex aurora-dominated expansion phase. Occurrence of all auroral structures clearly peaks around the magnetic midnight during low solar wind driving, while during high solar wind driving, the arc occurrence distribution widens towards the dusk and the distribution of more complex aurora extends further towards the dawn sector, again in agreement with . This enhanced structural complexity in the morning sector has been known since and is further summarized by . The much larger variation over Svalbard likely relates to the expansion of the auroral oval during active times. While the study shows that one third of all dayside auroral structures are arcs, our statistics suggest that although one third is a representative number for high solar wind driving, arc occurrence reaches 50 %–70 % during low solar wind driving (Fig. ). In agreement with , our arciness dataset suggests that the most arc-dominated MLT sector is dusk at 17:00–20:00 MLT, although our results refer to arcs being more frequent than other auroral forms in Lapland rather than in Svalbard. In the Lapland image data, the portion of arcs exceeds that of other auroral structures when the solar wind driving or geomagnetic activity increases. This can be interpreted as a more frequent growth phase appearance in the FoVs of the Lapland ASCs in the pre-midnight sector. The fewest arcs have previously been found in the MLT noon sector at 10:30–13:30 MLT , while in our data, the minimum arc occurrence appears particularly deep at 10:00–11:00 MLT. This time sector has also been called a “dayside gap” due to a low probability of accelerated particle precipitation leading to an ultimate minimum in the occurrence of all aurora.

The presence of slow solar wind increases the auroral peak emission heights over Lapland strongest in the morning sector at 02:00–04:00 MLT (Fig. 5, left panel). This is in agreement with the correlation between the occurrence of PsA and the solar wind speed , as the increasing solar wind speed would drive higher PsA occurrence, and thus lead to higher precipitation energies and lower peak emission heights. In this dataset, the PsA has the highest occurrence rate at 02:00–04:00 MLT.

The height drop due to PsA in the morning sector seems to be the main reason for the dawn–dusk asymmetry in median heights. In fact, excluding the PsA events from the height distribution (red curve in Fig. ) removes the lowering of the height evolution towards the late morning hours at about 02:00–07:00 MLT where PsA most frequently takes place . What is left is a rather mild increase (2–3 km) in the median heights from midnight until about 03:00 MLT, after which a decreasing trend is seen from about 120 km to about 115 km until 07:00 MLT. This suggests that other morning sector aurora does not have a major role in affecting the auroral peak emission height, which is in agreement with earlier observations of emission heights being lower inside the PsA patches than outside these patches e.g.. The studies by and show that the most notable change in the morning sector precipitation takes place around 06:00 MLT, after which the PsA-related peak height of ionospheric electron density decreases from about 110–120 km down to heights below 100 km as a result of soft precipitation (<10 keV) decay while more energetic electron precipitation still persists . As all auroral height data are analysed together in our study, we merely see a gentle decrease of heights towards the late morning hours, which nearly disappears when PsA events are removed. Although auroral pulsations have been shown to extend to magnetic noon , only eight PsA events were identified in the range of the Svalbard data analysed in this study. Their removal had no effect on the statistical height distribution for the Svalbard data. This is in agreement with , who recently reported a much lower occurrence rate of PsA at high latitudes (Svalbard) compared to Lapland. They further concluded that most high-latitude PsA events are related to less energetic electron precipitation and thus, may not have a strong effect on the auroral peak emission height, even in the case of high occurrence rate.

The increase of emission height by 3–7 km at 03:00–08:00 MLT when PsA events were excluded is in agreement with the 8 km decrease of height at the start of the PsA . However, the height difference found in this study is a measure with respect to the background morning sector diffuse aurora rather than the substorm-type discrete aurora where PsA often starts from. While in the previous study the start of the PsA event was associated with an average peak emission height decrease from about 117 to about 109 km, the all-inclusive median height in the morning sector shows a more gentle but similar order of magnitude decrease in height from about 117 km at 04:00 MLT to about 110 km at 07:00 MLT, even though these data are not organized according to PsA event onsets. Again, this emphasizes the role of PsA in the morning sector precipitation.

This study combines a decade of auroral all-sky image data from Svalbard and the Fennoscandian Lapland to examine the magnetic local time (MLT) distribution of peak emission height and auroral morphology. The median peak emission height for the nightside green aurora is about 120 km while for the dayside green aurora, it is 130–140 km. Fast solar wind and enhanced geomagnetic activity drives lower peak emission heights on the nightside as compared to slow solar wind and magnetically calm conditions. The IMF polarity, on the other hand, has no appreciable effect on the nightside auroral emission heights, but changes the emission heights in the dawn, noon, and dusk sectors, all of which are interpreted as the contracting auroral oval moving the most poleward and softest particle precipitation regions poleward and outside the fields-of-view (FoVs) of the cameras.

The structural analysis is divided into auroral Arcs and “Other” more complex aurora. This shows that the occurrence of more complex auroral forms follow the solar wind driving and geomagnetic activity more closely than the occurrence of auroral arcs, in agreement with earlier studies. The magnetic midnight sector is particularly prone to complex auroral structures due to substorm activity, but also the morning sector is dominated by more complex aurora. Magnetically quiet conditions or low solar wind driving leads to up to 70 % of the aurora being arcs, especially in the evening sector.

Excluding periods of known pulsating aurora (PsA) from the green emission peak height dataset makes the MLT distribution of heights dawn–dusk symmetric by increasing the morning sector auroral heights by 3–7 km. This finding emphasizes the role of PsA in the morning sector energy deposition.