A state-of-the-art multispectral imager has been used to study the flickering component of an auroral event at high spatial (40 m) and temporal (32 fps) resolution. Scale sizes for the flickering patches were found to be regularly smaller than 1 km. The typical flickering frequency observed was in the range 6–8 Hz, although flickering patches at both lower and higher frequencies were identified. The flickering structure was correlated with the coincident non-flickering aurora, showing that although there is a temporal relationship between the two, there is no spatial correlation. These results support the theory that flickering structure is caused by interfering dispersive Alfvén waves.
 When viewed in the magnetic zenith, flickering aurora comprises small spots or patches of periodic luminosity oscillations, usually within a discrete auroral arc. It is often associated with auroral breakup events. Flickering aurora is usually observed with frequencies between 2 Hz and 20 Hz. McHarg et al.  have reported observations of flickering with frequencies as high as 180 Hz. These two main frequency ranges correspond broadly to the ion cyclotron frequencies of O+ and H+ at 1 RE (≈10 Hz and ≈100 Hz respectively). In most known cases only a small proportion of the total aurora intensity is flickering, typically 10–20 percent [Kunitake and Oguti, 1984; Sakanoi and Fukunishi, 2004; Grydeland et al., 2008]. Previous reports of flickering aurora have almost exclusively concentrated on patches with scale sizes larger than 1 km. To the authors' knowledge there has been only one other study devoted to small-scale flickering, reported by Holmes et al. [2005a, 2005b]. They found scale sizes ranging from 215 m to 1 km.
 There is not yet a well-defined theory for the mechanism leading to flickering aurora. There have been several observations of oscillating field-aligned electron bursts (“FABs”) by rocket-based instruments. These have been linked to simultaneous ground-based measurements of flickering aurora [McFadden et al., 1987; Lund et al., 1995]. Temerin et al.  suggested flickering aurora and FABs are produced by electromagnetic ion cyclotron (EMIC) waves which propagate from the magnetosphere to the ionosphere. Particle and wave measurements made by instruments on the FAST satellite were used by McFadden et al.  to show resonant interaction between propagating H+ EMIC waves and both energetic inverted-V electrons and cold secondary electrons, supporting the model of Temerin et al. . Lund et al.  interpreted electromagnetic oscillations detected by a sounding rocket as electromagnetic oxygen cyclotron waves, again supporting the model of Temerin et al. . Sakanoi et al.  modelled interference between multiple EMIC or dispersive Alfvén waves in a resonance cone, successfully reproducing observed spatial and temporal characteristics of flickering aurora. Arnoldy et al.  proposed an alternative generation mechanism in which an on-off inverted-V potential oscillates at ion cyclotron frequencies.
 On 22 October 2006 strong flickering was observed in aurora over Tromsø, Norway, with the multispectral imager ASK (Auroral Structure and Kinetics). Initial results of the scale sizes and temporal and spatial variations of the flickering are interpreted in relation to current theories for the generation of flickering aurora.
 ASK (N. Ivchenko et al., Auroral Structure and Kinetics—A new optical instrument for auroral studies) consists of three highly sensitive Andor iXon EMCCD (Electron Multiplying CCD) cameras, each fitted with a different narrow-passband spectral filter. These are co-aligned with an identical 3.1° × 3.1° field of view centered on the magnetic zenith. This field of view corresponds to roughly 5 km × 5 km at 100 km height. Each EMCCD detector contains 512 × 512 pixels. These are binned into 256 × 256 equal-sized “super-pixels” during acquisition, primarily for technical reasons. This provides a spatial resolution of approximately 20 m at 100 km height, allowing for detailed studies of auroral features on the sub-km scale. At the time of the event studied here the cameras were run at 32 frames per second, allowing for analysis of flickering aurora up to a Nyquist frequency of 16 Hz. In October 2006 ASK was located at the EISCAT site in Ramfjordmoen, near Tromsø, northern Norway (geographic latitude 69.58° N, longitude 19.22° E).
 Data from only one of the three ASK cameras have been used for the analysis presented here. The camera was fitted with a spectral filter centered at 673.0 nm with a full-width half-maximum of 14.0 nm. This filter was designed for studies of N2 1PG auroral emission bands.
 The data used in this study were recorded in a 33 s period between 18:16:20 UT and 18:16:53 UT (≈21 MLT) on 22 October 2006. The initial 4 s of data exhibit a bright, flickering auroral arc moving into the field of view. The arc has a width of about 1°. Dynamic flickering aurora then fills the field of view.
 In the first stage of the analysis each image was binned into 128 × 128 (16,384) equal-sized super-pixels. Part of this binning was performed by the camera hardware during acquisition, as mentioned in the instrumentation section. Each super-pixel corresponds to 4 × 4 on-chip detector pixels. The additional post-acquisition binning was carried out in order to reduce noise and computational time required during the analysis. After binning the images have a spatial resolution of approximately 40 m at 100 km height.
 A third order polynomial was fitted through the first 32 frames (1 second) of data from each super-pixel, and each fit subtracted from its corresponding data. This procedure removes intensity fluctuations caused by auroral structures moving in and out of the pixel, leaving only the flickering. A Hann window was applied over these residuals, before a fast Fourier transform (FFT) was performed on the sequence. The Hann window was used to reduce edge effects when performing the FFT. This process results in a frequency spectrum with a resolution of 1 Hz. The flickering magnitude was obtained by taking the absolute value of the complex FFT result. This was done for each super-pixel, so the spatial structure within the flickering can be identified. A 32-frame sliding window was applied in time steps of 1 frame (1/32 s), repeating the analysis at each time step, to facilitate the study of temporal changes in the flickering structure.
 In addition to the frequency analysis, mean images were made in order to compare auroral structure with flickering structure. The same 32-frame sliding window was used. For each set of 32 frames the binned camera images were averaged pixel-by-pixel, in order to produce a valid image for comparison with the FFT results.
 The most notable feature of the analysis results is the ubiquitous small scale structure in the intensity of the flickering. Selected results are shown in Figures 1 and 2. The eight small plots of Figures 1 (right) and 2 (right) show the flickering power spectral density across the whole image in a frequency range of 2 Hz, centered on the frequency indicated beneath the plots. Figure 1 (left) shows the mean auroral image corresponding to the 32 frames used in the frequency analysis. Figure 1 (left) also highlights the peak flickering frequency for each of the super-pixels in the image.
 When the initial auroral arc moves through the field of view the strongest region of flickering at each frequency lies roughly within the region occupied by the main part of the arc, although the flickering region often does not cover the whole arc, and varies in frequency. In Figure 1, at 18:16:30 UT, 6 s after the main arc has passed, there are small-scale flickering patches with different frequencies which can overlap but are not necessarily co-located. This is typical of the flickering throughout the remainder of the studied event. There are often patches which fill only a small fraction of the instrument field of view. A thorough survey of patch sizes will be undertaken. However, we are able to report that most have widths smaller than 1 km, and there are many examples where patches as narrow as 4 super-pixels (160 km) occur. At most times there are patches and structures which appear and remain in the same location for a few seconds before fading or being replaced by other structures. Some narrow patches move through the field of view. Figure 2 shows a time towards the end of the period studied. It can be seen that there is still small-scale flickering structure, despite the aurora having weakened considerably. An animation of the results in the same format as Figures 1 and 2 is available online in the auxiliary material. This continuous view of the results leads the eye to the occurrence of elongated narrow structures which are often coaligned. This information is crucial for theoretical studies. The ratio between width and length of the structures could be related to the difference in propagation angle between two or more interfering dispersive Alfvén waves. Modelling work done by Gustavsson et al.  has shown that it is possible to produce many different structures by varying the parameters of multiple interfering EMIC waves.
 The dominant frequency of the flickering is between 6–8 Hz for a substantial portion of the 33 s interval. However, it is clear from Figures 1 (right) and 2 (right) and the peak frequency shown in Figures 1 (left) and 2 (left) that other frequencies dominate in small structures at other times. For example in Figure 2 the dominant frequencies of a distinct feature across the image are in the 11–12 Hz range. Usually structures partially overlap across a frequency range of a few Hz.
 In order to study the spatial and temporal nature of flickering aurora, and its relationship to the steady auroral intensity, the mean images have been compared with the power spectral density of the flickering. Figure 3 (bottom) addresses the temporal relationship between flickering aurora and the background steady aurora. The solid grey line shows the mean flickering power spectral density across all frequencies, obtained by averaging over all super-pixels. It is likely that the large peak at 18:16:24 UT is caused by the transient intensity increase resulting from the auroral arc moving into the field of view. The black line shows the mean brightness of the 32-frame mean images. This is equivalent to a 1 s running mean auroral brightness. Figure 3 (bottom) shows that when there is increased auroral activity there is also increased flickering activity. As there is aurora filling the field of view for all but the initial part of the sequence, this high correlation cannot be caused by changes in auroral coverage. Therefore there is a temporal correlation between flickering strength and auroral brightness.
 To address the spatial relationship between flickering aurora and the background steady aurora the mean images have been correlated spatially with the power spectral density of the flickering. Figure 3 (top) shows the time-varying correlation coefficient for three 2-Hz-wide frequency ranges and also for all frequencies (thick black line). The correlation coefficient is obtained for each mean image individually. Each super-pixel is treated as a separate data point, and the image brightness is correlated with the flickering power at the frequencies within the specific range. It can be seen that the correlation is generally low. There are times where the correlation coefficient is considerably positive (or negative) at specific frequencies, but this is simply the result of the flickering matching with the brighter (or less bright) auroral features at those times. This correlation has been tested for each frequency, including those not plotted in Figure 3. The mean correlation coefficient over the whole 33 s period for all frequencies (thick black line) is 0.094, which is not statistically significant. This is also biased by the relatively high correlation coinciding with the period in which the arc is moving into the field of view (first 4 seconds). The mean correlation coefficient for the last 28 seconds of the period (where aurora is filling the field of view) is 0.064. Therefore on small scales the flickering is not necessarily stronger where there is brighter aurora, i.e. there is no spatial correlation. The combined result of the correlations shown in Figure 3 is that flickering is linked temporally to auroral activity, but not spatially on small-scales.
 This work has shown that the intensity of flickering in aurora has small-scale structure on sub-km scales. To the authors' knowledge there has been only one other analysis of flickering patches on such small scales, reported by Holmes et al. [2005a, 2005b]. Our observations agree closely with these reports. This work has produced evidence to suggest that the source of flickering and the source of the general aurora are linked, as flickering strength and auroral activity correlate temporally. However, on small scales the two are not linked spatially. These results are consistent with the theories of Temerin et al.  and Sakanoi et al. , that small-scale structure in flickering is a result of interference between two or more dispersive Alfvén waves, modulating the auroral intensity. According to this theory the spatial structure seen within the flickering of the aurora is an interference pattern, which is not necessarily affected by or linked to mechanisms producing fine-scale structures in non-flickering aurora, or the non-flickering component of flickering aurora. In order to explain the lack of spatial correlation using the oscillating inverted-V theory of Arnoldy et al.  it would be necessary to assume that the auroral brightness is strongly dependent on small-scale non-flickering structures which are not related to the oscillating inverted-V potential. This seems unlikely. Also it is difficult to explain the small-scale structure seen in the flickering power spectral density using this theory, since inverted-V auroral structures are usually of greater horizontal widths than the sub-km sizes described here.
 Further work with ASK data will investigate the phase of flickering patches, and the pattern of phase across the structures. The multi-spectral capability of ASK will allow us to investigate differences in flickering seen in different emissions. As the three cameras are fitted with different spectral filters they observe emissions coming from different altitudes. Therefore it will be possible to study the energy distribution of the precipitating electrons responsible for the flickering. Peticolas and Lummerzheim  used a time-dependent auroral electron transport model to simulate field-aligned bursts with an on-off electron intensity distribution. We intend to carry out similar work using a combined electron-transport and ion-chemistry model, for comparison with ASK results.
 The authors wish to thank Geoff Daniell, Dirk Lummerzheim, and Mike Lockwood for useful discussions during this work. ASK has been funded by PPARC of the United Kingdom. D. Whiter is supported by an STFC studentship grant.