This is a report on the first observations of auroral roar emissions near 4 times the ionospheric electron cyclotron frequency (fce) with a passive receiver installed in Svalbard, Norway. 4fce roar emissions were detected from 5.27 to 5.70 MHz during moderate geomagnetic disturbances in 22 days between May and September 2011 only from noon to evening, while no event occurred during the winter season. An analysis of a typical event shows that the electron density profile measured from EISCAT Svalbard dynasonde satisfies the condition that the upper frequency of the 4fce roar is nearly equal to both upper hybrid resonance frequency (fUH) and 4fce at 238-km altitude. These observations support the idea expanded from the most commonly-accepted generation mechanism of 2fce and 3fce roar: the origin of 4fce roar is upper hybrid waves favorably generated under the condition of fUH ∼ 4fce in the auroral F-region ionosphere.
 The polar ionosphere connected with the magnetosphere via field-aligned currents is a source of various electromagnetic and electrostatic waves. In the MF and HF bands (up to 5 MHz), two types of electromagnetic emissions have been identified by ground-based observations: auroral roar and MF burst. Auroral roar is a relatively narrow band emission (δf/f < 0.1) near 2.5–2.8 MHz and 3.7–4.3 MHz [Kellogg and Monson, 1979; Weatherwax et al., 1993]. Because these frequency bands are considered to be close to the second and third harmonics of the electron cyclotron frequency in emission sources, each type of auroral roar is called 2fce and 3fce roars. The polarization character of 2fce and 3fce roars was identified as left-handed polarized waves by Shepherd et al.  and Sato et al. . Both types are commonly attributed to mode conversion to the L-O mode of upper hybrid waves excited in the bottomside ionosphere by auroral electrons [e.g., LaBelle and Treumann, 2002]. However, the observation of higher frequency components has not been reported.
 In this paper, we report the first observations of auroral radio emissions detected in a frequency range of 5.27–5.70 MHz, which roughly corresponds with ionospheric 4fce. We present one event on 16 May 2011 to show typical frequency spectra of 4fce roar in section 3.1 and synoptic studies of its frequency distribution and occurrence time and season in section 3.2.
 In August 2008, we developed new instrumentation referred to as Auroral Radio Spectrograph (ARS) for observation of MF/HF auroral radio emissions at Kjell Henriksen Observatory (KHO) in Svalbard (latitude 78.15°N, longitude 16.04°E, 75.2 magnetic latitude). ARS consists of magnetic loop antennas whose size is 2.7 m × 6.0 m and two types of receivers: ARS-S and ARS-WF. The former is designed for the continuous measurement of wave spectra with a time resolution of 1 sec, and the latter is designed to obtain waveform data digitalized by an A/D converter with a sampling rate of 10 M samples/sec (Nyquist frequency 5 MHz). The data shown in this paper were obtained with ARS-S. On 1 October 2010, the upper limit of the observation frequency of ARS-S was changed from 5 MHz to 6 MHz to examine the existence of higher frequency components of HF auroral radio emissions. Since then, ARS-S has been monitoring wave spectra in a frequency range of 1–6 MHz except for the period between 5 March and 11 May 2011. For synoptic studies, we use the data set obtained from ARS-S during two periods: 1 October 2010 to 4 March 2011 and 12 May to 30 November 2011.
3. Data Presentation
3.1. Event Study: 16 May 2011
Figures 1a–1b show the frequency-time (f-t) diagrams of radio waves observed by ARS-S in frequency ranges of 1.0–6.0 MHz and 5.2–5.8 MHz during 1400–1520 UT on 16 May 2011. Commonly-used geomagnetic indices represent moderately disturbed conditions during this period: Dst of −5 nT, Kp of 4- and IMAGE electrojet index IE of about 190 nT. Multiple horizontal lines and bands in the f-t diagrams correspond to interference from man-made radio transmissions and local noises from electronic devices and power supply in and around KHO. The emissions appearing in a frequency range of 5.43–5.68 MHz during 1406–1513 UT are distinguished from any interference from these man-made sources, but have the same characteristics as auroral roar reported before: narrowband features (δf/f ≈ 0.05) and slow variation in frequency and intensity. If we assume that these emissions are auroral roar emitted at a frequency near 4fce and use the International Geomagnetic Reference Field (IGRF) model, the above-mentioned frequency range corresponds to the altitude range of a generation region between 187 km and 298 km. Figure 1a shows that any other auroral radio emissions such as 2fce roar, 3fce roar and MF burst did not accompany with the 4fce roar. The 4fce roar attained a maximum intensity of 11 dB higher than background noise level at 1447:20 UT. This value corresponds to only 1.5 × 10−19 W/m2/Hz, a few percent of typical power spectrum density of auroral roar [LaBelle and Treumann, 2002].
 Vertical dashed, but barely-discernible lines appearing every 6 minutes in Figure 1 represent signals transmitted from EISCAT Svalbard dynasonde, located about 1-km away from the ARS antennas. The dynasonde provides the altitude profile of electron densities above ARS every 6 minutes. Because the ARS antenna has a relatively-wide beam angle (tens of degrees), the 4fce roar emissions detected by ARS are not always generated in the regions where dynasonde echoes are reflected. However, we can examine dynasonde data for the tendency of altitude profiles of electron densities in or near source regions of the 4fce roar. Figure 2 shows the comparison between the frequency range of the 4fce roar and altitude profiles of ionospheric characteristic frequencies: nfce (n = 1, 2, 3 and 4), fUH and plasma frequency (fpe). These altitude profiles are determined with fce calculated by the IGRF model and fpe estimated from the dynasonde data measured at 1436 UT on 16 May 2011 with “NeXtYZ,” the latest ionogram inversion [Zabotin et al., 2006]. The gray-shaded region covering 5.47–5.54 MHz corresponds to the frequency range of the 4fce roar detected at 1436 UT, which equals 4fce in altitudes of 245–278 km. The condition fUH = 4fce is met near an altitude of 238 km at 5.56 MHz, which is slightly higher than the upper frequency of the 4fce roar. The explanation of this situation is that the 4fce roar might be generated near the altitude where this matching condition is satisfied.
3.2. Synoptic Studies
Figure 3a is the frequency distribution of 4fce roar emissions observed during the analyzed periods. As in the event on 16 May 2011, their occurrence tended to be associated with moderate geomagnetic activity: Dst of −10 nT, Kp of 2, IMAGE IE of 200 nT (on average). The vertical axis corresponds to the number of 4fce roar events. Whether 4fce roar emissions are detected or not is determined every 20 minutes, and each roar event is divided into 12.5-kHz frequency bins. This figure shows that 4fce roar emissions appear between 5.27 and 5.70 MHz and have a peak near 5.53 MHz. This frequency range is separated from that of 2fce roar (2.525–2.925 MHz) and 3fce roar (3.725–4.275 MHz) [Weatherwax et al., 1995].
Figure 3b shows the appearance time and day of 4fce roar between 12 May 2011 and 30 November 2011 (DOY 132–334). Time periods when 4fce roar appears are represented by gray shaded pixels, the color of which shows power spectrum densities averaged over the frequency range where 4fce roar is detected. The three black bars at DOY 224, 253–254, and 286–289 correspond to the aborting of the data recording. Clear dependence in local time can be found: 4fce roar appeared only from noon to evening (1125–1925 LT; local time (LT) = UT + 1) and most often during 1400–1700 LT. After the first detection of 4fce roar event on 15 May 2011, 4fce roar emissions appeared on 22 days as of 30 November 2011 and most often in May and June. Interestingly, no 4fce roar was detected between 17 September and 30 November 2011. Also between 1 October 2010 and 4 March 2011 (not shown in Figure 3b), the wave spectra of 1–6 MHz had been monitored; however, no 4fce roar was observed during this winter season. These no-4fce roar intervals include the polar night period (between the end of October and the middle of February) when the location of the observing station KHO experiences no sunset or sunrise.
4. Discussion and Conclusions
 Few ground-based radio observations tried to detect radio spectra up to 6 MHz especially during daytime in summer. Kellogg and Monson  monitored radio spectra from 2.6 to 6 MHz with a sweeping receiver on several nights during a campaign conducted in March and April in 1977 and 1978 at Churchill (69° invariant latitude). A receiver used in further observations by Kellogg and Monson  in January and December in 1979, 1981 and 1982 swept up to 10 MHz during part of the time. But the narrowband auroral radio emission they detected was only 2fce roar near 3 MHz. Subsequent measurements with sweeping receivers have been conducted below 5 MHz [e.g., Weatherwax et al., 1993; LaBelle et al., 1997; Sato et al., 2008]. The ARS-S observation, conducted for 24 hours every day, enables us to find a new frequency component of auroral roar near 5.5 MHz.
 The 4fce roar did not accompany any other types of MF/HF auroral radio emissions (2fce roar, 3fce roar and MF burst), which are detected only during local darkness and are most common in the premidnight hours [e.g., Weatherwax et al., 1995; LaBelle et al., 1997; LaBelle and Weatherwax, 2002; LaBelle et al., 2005]. This difference is causally-related to dependence of ionospheric radio wave absorption on frequency: during daylight hours, photoionization enhances electron densities in the ionospheric D- and E-regions to increase radio wave absorption, which is severer for lower frequency waves. Auroral roar emissions, emanating from the F-region ionosphere, should lose some part of their energy in propagating through the lower ionosphere because of the collisional effect. As an example of daylight conditions, here we calculate the ionospheric absorption of radio waves propagating vertically through the D- and E-regions. The absorption per unit length nearly equals to 4.6 × 10−5Nν/(ω2 + ν2) in dB/m, where N, ν and ω are the electron density, the electron-collision frequency, and the wave frequency [Davies, 1969]. Using the electron density profile shown in Figure 2, the absorption is roughly estimated to be 34 dB for 2.7 MHz, 16 dB for 4 MHz, and 8.2 dB for 5.5 MHz. This estimation shows that, even during daylight hours, 4fce roar can reach ground level without severe absorption, while the other types with lower frequencies are screened by the dense sunlit D- and E-regions.
 The appearance time and season of 4fce roar are quite different from those of the other types. Its appearance is influenced by the following three factors: auroral accelerated electrons, cold background electron density, and level of interference from distant radio stations and atmospherics. First, auroral accelerated electrons provide free energy to give rise to plasma instabilities. Although electron distribution functions which involve the generation of 4fce roar emissions remain to be elucidated, auroral accelerated electrons which may be attributed to its generation as in the case of 2fce roar and 3fce roar [Yoon et al., 1998] should exist with sufficient energy flux in a wide LT-region from post-noon to nightside [e.g., Newell et al., 1996, 2010]. Actually, ionograms obtained by the dynasonde show intermittent enhancement of electron densities (foF2 > 5 MHz) above 150 km altitude during 1336–1536 UT on 16 May 2011, which indicates that low energy electrons (less than 1 keV) [e.g., Rees, 1963] precipitate above and around the observing site when 4fce roar is observed. Next, background cold electron density is one of the significant factors in determining the growth rate of causative plasma instabilities. If we employ the hypothesis that 4fce roar originates from linear mode conversion of upper hybrid waves generated under the matching condition of fUH ∼ 4fce, the generation of 4fce roar also requires high-density ionosphere to satisfy this condition; 4fce roar cannot be detected during darkness. Finally, the observability of 4fce roar at ground level is highly affected by interference from distant radio stations and atmospherics, which makes it difficult to distinguish 4fce roar. During the night, they can come from a long distance via multi reflections without severe ionospheric absorption. This situation is realized even during bright summer nights since ionospheric absorption in a sunlit area near the observing site is moderate for radio waves above several MHz. No appearance of 4fce roar during bright summer nights may be attributed to the masking effect of high-level interference from distant radio stations and atmospherics. Considering these three factors together, we can understand the appearance time and season of 4fce roar at ground level.
 Theoretical work [e.g., Kaufmann, 1980; Yoon et al., 1998] showed that electrostatic upper hybrid waves are favorably generated when the upper hybrid frequency matches a cyclotron harmonic in the presence of auroral electron distributions; however, no one predicted the occurrence of 4fce roar. Yoon et al.  showed the growth rate for generation of 2fce and 3fce waves in various modes, calculated for a model auroral electron beam including a loss cone. The highest growth rates correspond to the excitation of upper hybrid waves at the matching conditions fUH ∼ 2fce and fUH ∼ 3fce; however, higher harmonics (4fce and above) are stable. While reassessment should be done as to whether sufficient growth rate of upper hybrid waves at 4fce is attained with more realistic electron distribution functions, the observed properties of 4fce roar, such as the frequency matching condition and occurrence time and season, are explained by the hypothesis that it originates from linear mode conversion of upper hybrid waves generated under the matching condition of fUH ∼ 4fce.
 Finally, we further discuss another possibility that nonlinear wave-wave coupling between two upper hybrid waves with opposite k vectors generated under the matching condition of fUH ∼ 2fce makes electromagnetic auroral roar emissions at 4fce. Sato et al.  proposed this idea for the generation of second harmonic Terrestrial Hectometric Radiation (THR), which is one of auroral radio emissions emanating from the topside ionosphere, and predicted ground-level detection of R-X mode auroral roar at about 5.5 MHz, which is nearly equal to twice the 2fce roar frequency usually observed during darkness at KHO. Actually, this predicted frequency corresponds well to the most common frequency of 4fce roar (5.53 MHz); however, the electron density profiles simultaneously observed by the dynasonde tend to have little correspondence with this hypothesis. For example, Figure 2 shows that the matching condition fUH ∼ 2fce is met near 100 km altitude at a frequency of 2.9 MHz, which is not equal to the half frequency of 4fce roar (2.74–2.77 MHz). Furthermore, it is unlikely that electron distribution functions can retain sufficient free energy at such low altitudes. This observation does not seem to support the last proposed idea; however, it should not be ruled out as a possible generation mechanism. In future studies, polarization measurements are required as well as long-term continuous observation.
 The authors are grateful for the provision of the dynasonde database by EISCAT, which is an international association supported by research organizations in China (CRIRP), Finland (SA), France (CNRS, till end 2006), Germany (DFG), Japan (NIPR and STEL), Norway (NFR), Sweden (VR), and the United Kingdom (NERC). The Dst and Kp indices are provided by World Data Center for Geomagnetism, Kyoto. The authors appreciate UNIS/KHO staff's support for the installation and maintenance of ARS in Svalbard and the institutes who maintain the IMAGE Magnetometer Array.
 The Editor thanks Peter H. Yoon and H. Gordon James for their assistance in evaluating this paper.