Auroral radio emissions reveal physics of beam-plasma interactions and provide possibilities to remotely sense ionospheric plasma processes.Sato et al. (2012) recently discovered that auroral roar emissions, long known to occur at two and three times the electron gyrofrequency (fce), also occur at 4fce. Using data from wave receivers in the British Antarctic Survey Automatic Geophysical Observatories, we confirm the existence of 4fce-roars and observe for the first time 5fce-roars. A search at higher frequencies did not find higher harmonics. Both 4fce- and 5fce-roars only occur in sunlit conditions near summer solstice. Harmonic roar emission frequencies scale with the strength of the geomagnetic field, and combining data from four observatories suggests that the 4fce-roar source height lies around 245 km, lower than the 275 km estimated for 2fce-roar. These observations show that the auroral roar generation mechanism acts under a broader set of plasma conditions than previously considered.
 Earth's aurora is an energetic phenomenon; in addition to optical emissions, impact ionization and heating, Joule heating, and other energetic effects, it produces radio waves which are a window into the ionospheric plasma physics. These emissions are potentially significant not only as diagnostic tools but also as relatively local and easily-measured examples of plasma radiation processes characterizing electron beam-plasma interactions.
 At least three types of natural auroral radio emissions are detectable at ground level: hiss, a broadband emission below 1 MHz; medium frequency burst, a broadband emission at 1.5–4.5 MHz; and auroral roar, a narrowband emission near cyclotron harmonics (review by LaBelle and Treumann ). Auroral roar was first detected at 2.5–2.8 MHz and associated with the harmonic of the cyclotron frequency (2fce) [Kellogg and Monson, 1979]. A similar emission at 3.7–4.3 MHz was associated with the 3fce harmonic [Weatherwax et al., 1993], and recently the 4fce harmonic was observed [Sato et al., 2012]. Satellite-borne receivers have also detected auroral roar-like signals at 1000–3000 km altitude [James et al., 1974] and much further out [Bale, 1999]. From distributions of 2fce-roar frequencies observed at many stations covering a range of magnetic latitudes,Hughes and LaBelle found that the emission frequency increased approximately linearly with magnetic field, with a best-fit slope consistent with emission at 2fceat an altitude of 275 km, on the bottomside of the F-region. However, bimodal frequency distributions observed at South Pole Station [LaBelle and Weatherwax, 2002] and bimodal altitude distributions inferred from a combination of frequency and direction-of-arrival measurements [Hughes et al., 2001] suggested that 2fce-roar often reached the ground from two sources, one on the bottomside and one on the topside of the ionosphere.
 It was recognized early on that the matching condition between the cyclotron harmonic and the upper hybrid frequency (fuh) was significant for upper hybrid wave generation [Kaufmann, 1980], and Gough and Urban put forth that those upper hybrid waves, converted to electromagnetic modes, could explain the ground-level auroral roar.Yoon et al.  showed that the cyclotron maser mechanism acts at ionospheric altitudes to produce the waves, with peak growth rates where fuh = N fce for N = 1, 2. Shepherd et al. , comparing radio receiver and incoherent scatter radar (ISR) data, demonstrated the tendency for this condition to hold somewhere in the ionosphere when auroral roar occurs, and Hughes and LaBelle , using direction of arrival measurements combined with ray tracing into electron density profiles measured simultaneously with ISR, found stronger evidence that the 2fce-roar originates wherefuh = 2fce. Ground-level observations also showed that auroral roar was left-hand polarized, as expected for mode-converted upper hybrid waves [Shepherd et al., 1997]. Two space-based observations provided strong support to the proposed generation mechanism of auroral roar.James et al. observed auroral roar near 2 and 4 MHz emanating from the topside ionosphere, and by ray-tracing calculations based on electron density profiles determined by topside sounding, they inferred that the source was wherefuh = 2fce. An auroral zone sounding rocket serendipitously penetrated a 2fce-roar source region, observing intense upper hybrid waves wherefuh = 2fce as well as evidence for eigenmode structure qualitatively similar to that predicted by theory [Samara et al., 2004].
 The most recent discovery pertaining to auroral roar was the observation of 4fce-roar emissions using ground-level receivers in Svalbard [Sato et al., 2012]. Unlike 2fce- and 3fce-roar which occur almost exclusively in conditions of darkness [e.g.,Weatherwax et al., 1995; LaBelle and Weatherwax, 2002], the 4fceemissions occurred in sunlit conditions, presumably because high F-region densities are required to achieve the conditionfuh = 4fce, and E-region screening is less of a factor for such high frequency waves. These observations demonstrated that the auroral roar phenomenon occurs under a wider range of conditions than previously thought.
 The Sato et al.  observations of 4fce-roar under unexpected conditions inspired us to re-examine historical data for those conditions, seeking confirmation of their discovery. We accomplished that and more, as described below.
2. Data Presentation
 The British Antarctic Survey Automatic Geophysical Observatories (BAS-AGO's) were operated in the 1990's at five sites including Halley Bay Station, Antarctica [Dudeney et al., 1998]. During 1996–1999, up to three of these observatories included LF/MF/HF receiving systems provided by Dartmouth College: A-80 (80.7 S, 20.4 W, 66.53 magnetic latitude), A-81 (80.7 S, 3.0 E, 68.88 magnetic latitude), and A-84 (80.7 S, 26.3 W, 68.21 magnetic latitude). The stations also included flux gate and search coil magnetometers, riometers, photometers, and a multi-frequency ELF/VLF receiving systems.
 The sensors for the BAS-AGO LF/MF/HF receiving systems were 10 m2loop antennas supported by single ∼4.5-m masts erected ≥100 meters from the observatories. Broadband (16 MHz) preamps at the antennas conditioned the signals for transmission to the receivers in the observatories via coaxial cables. The receivers covered 0.1–16.0 MHz. To achieve 20-kHz frequency resolution over such a broad band while staying within the constraints of limited telemetry in an unmanned observatory, the receiver cycled between five sub-bands, dwelling on each for approximately five minutes and requiring 25 minutes to cover all frequencies. Sub-bands covered 1–4, 4–7, 7–10, 10–13, and 13–16 MHz. For each of these five 3-MHz bandwidth sub-bands, the receivers measured 150-point spectra with approximately 8 second cadence for five minutes, followed by a 20-minute gap. Five selected frequencies below 1 MHz were sampled continuously at 8-second cadence with no gaps. Calibration signals were injected every 5 minutes. With this receiving scheme, >1-MHz wave events lasting less than 25 minutes were captured over only a subset of the total 1–16 MHz frequency range, and very short duration events could only be observed if they produced signals in the particular 3-MHz sub-band being sampled at the time of the event. While this scheme had some disadvantages in terms of continuous coverage, it presented an opportunity to search for new phenomena at higher frequencies and with higher frequency resolution than usually monitored by most other LF/MF/HF receiving systems.
 Previous studies focussed on wintertime data under conditions of darkness, when the receiving systems detected auroral hiss, medium frequency burst, and 2fce- and 3fce-roars [e.g.,Hughes and LaBelle, 1998]. The recent discovery of 4fce-roars bySato et al. inspired a re-examination of the summertime data under daylit conditions, which immediately confirmed the discovery.Figures 1 (top left) and 1 (top right) show two examples of 4fce-roar observed at A-81.Figure 1(top left) is a spectrogram of sub-band 2 (4–7 MHz) recorded Dec. 1, 1998, during seven 5-minute intervals starting at 0027, 0053, 0118, 0145, 0211, 0237, and 0303 UT. The emission occurs in the second interval, 0053–0058 UT, at 4665–4845 kHz. At the resolution of this receiver the emission closely resembles 2fce- and 3fce-roar emissions, except that the frequency is higher. The magnetic field at A-81 was lower than at most auroral observatories; for example, according to the International Geomagnetic Reference Field (IGRF) the magnetic field at 250 km above A-81 is 43753 nT versus 48875 nT for the corresponding field at Svalbard, where theSato et al.  observations took place, so the 4fce-roar at A-81 occurs at a correspondingly lower frequency.Figure 1 (top right) shows another example of a 4fceroar observed at A-81 at 0035–0040 UT and 0101–0106 UT on Dec. 18, 1997, covering a somewhat broader frequency range and lasting long enough to be measured in two consecutive 5-minute intervals separated by a 20-minute gap. Observed 4fce-roars have maximum power spectral densities of order ∼20–40μV/mHz1/2.
 The extended frequency range of the BAS-AGO LF/MF/HF receiving system inspired us to search for higher harmonic auroral roar emissions.Figure 1 (bottom left) shows an example of a 5fce-roar observed at A-81 during 0415–0420 UT on Dec. 24, 1998. The emission occurs in the fourth of six 5-minute intervals shown in the figure, at 5650–5880 kHz. The frequency range of 5fce-roar at A-81 is partly obscured by a band allocated to man-made radio transmissions at ∼5.9–6.2 MHz, around 50 meters wavelength, which appear as horizontal lines on the spectrogram. The natural signals are distinguished by their different frequency-time structure, and this example occurs primarily at frequencies just below the band of man-made transmissions.Figure 1 (bottom right) shows another example of a faint 5fce-roar observed at A-81 during 0104–0109 UT on the following day, Dec. 25, 1998, at 5625–5820 kHz. Possibly, some 5fce-roar is imbedded within the man-made transmissions occurring during 0038–0043 UT; if so, this event may have also lasted longer than 20 minutes.
 BAS-AGO LF/MF/HF data were available from A-81 and A-80 between January, 1996, and January, 1999, and from A-84 between December, 1997, and January, 1999, although various instrumental problems and interference rendered some of these data unusable. A comprehensive search of the 4–7 MHz sub-band yielded 78 4fce-roars and 10 5fce-roars.Figure 2 shows a scatterplot of these 88 events as a function of day of year and time of day (UT); red points represent 4fce-roars, and green points represent 5fce-roars. Exactly opposite to the case for 2fce- and 3fce-roars, but in agreement with the 4fce-roars reported bySato et al. , both 4fce- and 5fce-roars occur in summertime daylit conditions; only a single event occurs before day 300 or after day 50. Presumably this dependence reflects the requirement of relatively high electron densities in the F-region in order to achieve the matching conditionfuh = Nfce for N = 4 or 5. Such high densities occur mainly in sunlight. Figure 2 also shows that, in common with 2fce- and 3fce-roars, both 4fce- and 5fce-roars occur favorably in the pre-midnight and midnight magnetic local time sectors. (Midnight MLT corresponds to 0246 UT at A-80, 0226 UT at A-81, and 0304 UT at A-84.) Presumably this dependence reflects auroral activity, such as substorms, which maximizes at those magnetic local times.
Figure 3 shows histograms of the distributions of frequencies of 4fce- and 5fce-roars observed at the three stations, A-80 (red trace), A-81 (green trace), and A-84 (blue trace). For reference, the figure also shows histograms of the distributions of frequencies of 2fce- and 3fce-roars observed at A-80 during 1998 (red traces). As expected, both 4fce- and 5fce-roars occur at higher frequencies at A-84 than they do at either A-80 or A-81, since according to the IGRF, the magnetic field at 250 km above A-84 is 45690 nT, versus 43020 nT and 43753 nT at A-80 and A-81. Centroids of the distributions of harmonic roar events observed at A-80 are 2338 kHz (2fce-roar), 3604 kHz (3fce-roar), 4813 kHz (4fce-roar), and 5842 kHz (5fce-roar), for a ratio of 2.00:3.08:4.12:5.00; i.e., these centroid frequencies lie within a few percent of being in a 2:3:4:5 ratio with a fundamental frequency of 1170 kHz, which corresponds tofceat an altitude 317 km above A-80.
 The data in Figure 3 together with those in Figure 3 of Sato et al.  provide distributions of the frequencies of 4fce-roars from four stations spanning a wide range of magnetic latitudes and magnetic field strengths, allowing us to employ the method ofHughes and LaBelle  to obtain a better estimate of the source height of the emissions. Assuming that 4fce-roar indeed occurs at four times the electron gyrofrequency, the emission frequency should depend linearly on the magnetic field according to:
where meis the electron mass. Based on the IGRF, one can calculate the magnetic field strength for a hypothetical source altitude above all four stations (A-80, A-81, A-84, and KHO/Svalbard). It is then possible to calculate the mean square deviation of the observed 4fce-roar frequencies from the prediction ofequation (1)at that altitude. Repeating this calculation at all altitudes (100–1000 km) yields a U-shaped curve with a minimum at 245 km.Figure 4 shows a scatterplot of the centroid of the distribution of frequencies of 4fce-roars at each station versus the magnetic field 245 km above the station, together with the prediction ofequation (1) (green line). Hughes and LaBelle  applied this method to 2fce emissions observed at 13 stations and found that the best fit corresponded to 275 km. A significant difference between the source heights of 2fce- and 4fce-roars is not surprising, however, due to the drastically different solar zenith angle conditions of the two emissions. The 4fce-roars occur in daytime when the F-layer is considerably thicker and denser than at nighttime when 2fce-roars occur. Qualitatively, the matching conditionfuh = Nfce could well occur at a lower altitude for N = 4 in a typical daytime electron density profile versus N = 2 in a typical nighttime profile.
 Data from the LF/MF/HF receivers at the BAS AGO sites A-80, A-81, and A-84 confirm the recent discovery of 4fce-roar emissions [Sato et al., 2012] and provide significant first observations of 5fce-roars. However, investigation of all summertime spectra from the next highest sub-band of the BAS-AGO receivers, covering 7–10 MHz, yielded no evidence of higher harmonic roar events (at 6fce, 7fce, or 8fce). At all three BAS-AGO sites, 6fce- and 8fce-roar emissions would partially overlap with a strong bands of fixed-frequency transmissions at 7.0–7.5 MHz (40 m band) and 9.4–9.9 MHz (30 m), although in neither case is the overlap any worse than that between the 50-m band and the 5fce-roars, which were observed. More likely, higher harmonics are simply much rarer and not observed in these limited data sets with only 20% duty cycle for any given frequency of interest. After all, only ten 5fce-roars occur, and these appear significantly rarer than the 4fce-roars for which 78 examples occur.
 The recent observations of these higher harmonic auroral roars show that the auroral roar generation mechanism must be able to act under a much broader set of plasma densities than previously considered. They highlight how ubiquitous and robust this mechanism must be in different plasma environments. Weatherwax et al.  explored the use of 2fce- and 3fce-roar measurements to remotely sense electron densities and scale heights; the existence of higher harmonics suggests a broader application for those methods. The aurora continues to surprise us with different pathways for the auroral energy and the complexity of the auroral wave environment.
 The author thanks Mike Trimpi for design and construction of the BAS-AGO LF/MF/HF receivers. The British Antarctic Survey supported the design, construction and operation of the automatic observatories as well as installation of instruments in them. In particular, the author thanks John Dudeney and Andy Smith for arranging for the LF/MF/HF receiver to be hosted in the observatories, and Dick Kressman who made several trips to Dartmouth and subsequently installed the receivers in the observatories. The work at Dartmouth College was supported by National Science Foundation grants ANT-1141817 and ANT-1043230.
 The Editor thanks Paul J. Kellogg and Yuka Sato for assisting in the evaluation of this paper.