Notice: Wiley Online Library will be unavailable on Saturday 27th February from 09:00-14:00 GMT / 04:00-09:00 EST / 17:00-22:00 SGT for essential maintenance. Apologies for the inconvenience.
 The first results of long-term meteor observations made with Syowa MF radar (69°S, 39°E), Antarctica, are presented. In winter, meteor wind measurements can be conducted throughout the day without being severely affected by group retardation or total reflection, while in summer the observations are confined within several hours around 0000 local time (LT). Daily meteor echo rates (roughly 300–1000) are clearly anticorrelated with local K indices, indicating that the observations mostly represent geomagnetically quiet conditions. Echoes are distributed from around 80 km to 120 km, with the peak altitude at around 100 km. The maximum height is set by the relatively slow sampling frequency employed (5 Hz) and is expected to be extended by using a higher frequency. Meteor and full correlation analysis (FCA) winds in winter months are compared. They agree well at around 90 km, while the FCA winds tend to underestimate the meteor winds at upper altitudes. A notable finding is that the directions of wind velocities can also be different. The FCA winds show less height variations above about 90 km and it appears as though the FCA heights are overestimated. A case study of angles of arrival estimated from MF echoes suggests a possibility that meteor echoes contaminate ionospheric echoes down to as low as 80 km. However, quantitative evaluation of meteor contamination effects, if any, on FCA wind velocities has not been done yet and remains a future topic to be studied. Other sources which cause the differences should also be sought. The remarkably wide height coverage of the present MF radar observations (from around 60 to nearly 120 km) using both FCA and meteor techniques simultaneously will greatly contribute to polar mesosphere and lower thermosphere study.
If you can't find a tool you're looking for, please click the link at the top of the page to "Go to old article view". Alternatively, view our Knowledge Base articles for additional help. Your feedback is important to us, so please let us know if you have comments or ideas for improvement.
 Medium frequency (MF) radars are powerful tools for the study of the mesosphere and lower thermosphere (MLT) region, 60–100 km, and have been widely operated at low to high latitude sites. The Antarctic MLT region is not an exception and has been studied using MF radars for more than 2 decades [e.g., Fraser et al., 1995]. However, the region above MF radar observation, that is, the E region, has not been much probed by radars in Antarctica although it is of great interest to study the polar E region, where complicated physical and chemical processes such as aurora take place with neutral and ionized atmosphere interacting with each other. In contrast, the Arctic E region and F region have been much more intensively studied using incoherent scatter radars such as EISCAT radars [e.g., Brekke, 1997]. Incoherent scatter radars have been virtually the only technique capable of measuring wind motions and temperature in the E region with sufficient height and time resolutions. One disadvantage of the incoherent technique, however, is that it requires a large facility and has a high running cost. Thus there are only a limited number of incoherent radars, even in the Arctic region.
 One advantage of the meteor technique is that it can reach the E region by using a relatively low radio frequency such as MF. Using a 2 MHz system, Olsson-Steel and Elford  showed that meteor echoes can exist up to 140 km altitude. Tsutsumi et al.  showed that meteor wind measurement using an MF system is possible from around 90 km up to nearly 120 km. There is also a good possibility that MF meteor wind observations can, at least partly, compensate for the known problems of MF radar correlation techniques above around 90 km [e.g., Holdsworth, 1999]. A limitation of MF meteor measurement should be noted as well. The observation is basically limited to nighttime because total reflection of the radio wave at around 100 km is a common feature during the daytime, and echoes from meteor trails are quite often masked by strong ionospheric echoes [Olson-Steel and Elford, 1987]. However, continuous operation is possible in the polar regions during the polar night.
 The National Institute of Polar Research (NIPR), Japan, installed an MF radar system at Syowa station (69°S, 39°E), Antarctica in March 1999 [Tsutsumi et al., 2001]. The radar is equipped with four receiving antennas and receivers, and has a full capability of interferometry observations. We have successfully developed meteor wind measurement software based on the work by Tsutsumi et al. , and have been conducting first operational MF meteor observations together with conventional correlation analyses since May 1999.
 In this paper we report the first results of the operational MF radar meteor wind measurements using data collected in 1999 and 2000, and also compare wind velocities estimated by meteor and correlation techniques. In the following sections, we describe the instrumentation and observation techniques in section 2. Time and spatial distribution of meteor echoes are presented in section 3. Meteor and full correlation analysis (FCA) winds are compared in section 4 based on long term simultaneous observations made during the polar winter. In section 5 sources that cause discrepancies between the meteor and FCA winds found in section 4 are discussed. Finally, concluding remarks are presented in section 6.
2. Instrumentation and Observations
 The radar was manufactured by Atmospheric Radar Systems (ATRAD), Australia and was installed at Syowa station during the 40th Japanese Antarctic Research Expedition (JARE40) in 1999. (A detailed technical description is given by Tsutsumi et al. .) Basic parameters of the radar system are tabulated in Table 1. The radar is a monostatic pulse Doppler radar operated at 2.4 MHz. The major advantages of the radar are that it is a four antenna system (with each antenna able to independently transmit and receive radio waves), and that it can perform various interferometric observations. The antenna array consists of four crossed dipole antennas, which are located at the corners and the center of an equilateral triangle with side about 150 m long. An equilateral triangle is the most favorable array shape to avoid biases in measured wind velocities [e.g., Meek, 1990; Holdsworth, 1999]. The minimum antenna spacing is 0.7 λ. Many conventional MF radars are equipped with only three receiving antennas with spacing usually much longer than 1 λ, and do not necessarily avoid ambiguities of echo arrival angles when an interferometric technique is applied.
Table 1. Basic Parameters of Syowa MF Radar
28 μs (half power full width)
60 kHz (99% power full width)
Pulse repetition frequency
four crossed dipoles
four receivers each of which is connected to a single crossed dipole
Duration of each record
Sampling range resolution
 Observational parameters are summarized in Table 2. Two sets of parameters are carefully designed in order that acquired raw data can be processed with as many analysis techniques as possible. The two sets of parameters, alternated every 2 min, have been used since the middle of May 1999. Two radio polarization modes are used, as seen in Table 2. The presence of the Earth's magnetic fields yields differences in the propagation characteristics of two polarizations of radio waves, that is, the ordinary (O) and the extraordinary (X) modes. The X mode suffers more absorption and group retardation in ionized atmosphere than the O mode; thus the O mode is usually used for wind measurements of MF radars. The differences in propagation characteristics have been utilized for the estimation of electron density in the mesosphere [e.g., Manson and Meek, 1984].
Table 2. Observation Parameters
Parameter Set 1
Parameter Set 2
Pulse repetition frequency
O and E modes every eight pulses
Number of recorded data points
1024 (512 for each polarization)
 One of the two sets of parameters uses only the O mode as shown in Table 2. These parameters are chosen considering the following facts. Correlation techniques for wind measurements in the D region require a data length of about 2 min or longer, a sampling frequency of about 2 Hz or higher, and a sampling range of 50–100km [e.g., Holdsworth and Reid, 1995]. On the other hand, meteor echo observations require a higher sampling frequency and extended sampling range to detect short-lived and long-range echoes, respectively [Tsutsumi et al., 1999]. Considering the buffer memory size and the data transfer speed, we employed the equivalent sampling frequency of 5 Hz after 16 coherent integrations, and the sampling range of 50–176 km. Note that this sampling frequency is a compromise to conduct simultaneous meteor and correlation analyses. The maximum observable Doppler frequency shift corresponding to the 5 Hz sampling frequency is 2.5 Hz. This further corresponds to the maximum observable radial wind velocity of λ/2 × 2.5 = 156 m/s. This value is large enough for the present purpose because the largest possible horizontal wind velocity in the mesopause region is about 150 m/s. The observed radial velocity is the projection of the horizontal velocity onto the off-zenith angle at which the meteor echoes appear (typically about 30 degrees). Thus it is about 50% of the horizontal wind velocity. The raw data series acquired with this parameter set is processed using three different techniques: full correlation analysis (FCA) [Briggs, 1984], spatial correlation analysis (SCA), and meteor echo analysis. The first two correlation techniques were coded by ATRAD while the meteor echo analysis software was newly developed by the authors for Syowa MF radar based on the study by Tsutsumi et al. .
 The other set of observation parameters shown in Table 2 involves transmission of both O and X mode radio waves alternatively every 0.1 sec. The equivalent sampling frequency is 5 Hz for both modes, which is the same as the first set of observation parameters. However, the maximum sampling range is reduced to 138 km from 176 km of the first parameter set because of the limited buffer memory size. Five analysis techniques are applied to the acquired data: differential absorption experiment (DAE), differential phase experiment (DPE), FCA, SCA, and meteor echo analysis. Since the original correlation analysis suite by ATRAD was not designed to deal with raw data which consists of both O and X modes, we modified the software so that each polarization data can be extracted to reconstruct a new time series and used by FCA, SCA and the meteor echo analysis to produce both O and X mode winds independently. This modification to the analysis suite enables us to perform wind measurements without interruption by the electron density measurements. The X mode wind velocities are actually byproducts, and not currently used for further analyses because the X mode is more strongly absorbed and retarded than the O mode, as already described. Note, however, that the X mode wind measurements are thought to give reasonable estimates in a low electron density environment, and that in the future they can be used for a better wind estimate combined with the O mode winds than the O mode wind alone.
 The MF meteor echo analysis technique employed in the present study is basically the same as the one developed by Tsutsumi et al. . We detect so-called “underdense” type meteor echoes, and use them for wind estimate. In this type of echo the radio frequency is less than the plasma frequency of the trail and the radio wave penetrates into the trail, and is scattered by each electron. Alternatively, the radio wave is reflected on the surface of the trail when the plasma frequency is higher than the radio frequency. This type is called “overdense” echoes. Doppler frequency shift of an underdense echo gives information on wind velocity. On the other hand, that of an overdense echo is affected by the motion of the trail surface which expands in the radial direction very rapidly, and is thus not used for wind measurement. The underdense type echoes exhibit an exponential decrease in power with time due to ambipolar diffusion, and can be discerned from the overdense type.
 The post-analysis code developed by Tsutsumi et al.  was upgraded for on-line meteor echo processing and was added to the Syowa MF radar data analysis suit. Since the meteor observations are presently being made only for wind measurements, only echoes with a sufficient signal-to-noise ratio (SNR) and duration, more than about 1.5 s, are detected and processed as in the work of Tsutsumi et al. . Meteor echoes are detected by searching the time series obtained when echo powers are averaged over all receivers. In the present study we conduct a post-set steering of the receiving antenna beam into several off-zenith directions. Since almost all the meteor echoes are detected at large off-zenith angles, this technique improves the SNR of meteor echoes significantly and increases the number of detected echoes by roughly 50%.
 Receiver phase calibration is another key issue for quality meteor echo observations with accurate determination of angles of arrival (AOAs) of the echoes. It can be assumed that the MF radio wave during daylight hours is mostly reflected by horizontally stratified ionized layers. Thus the phases of the four receivers were calibrated under the assumption that the long time averaged AOAs of echoes from the upper D region during the daytime should point in the vertical direction.
 All the raw data from the beginning of the observation until late July 1999 have been stored for future developments and improvements of observation techniques. The post-set beam steering capability was added to the online meteor analysis software in late April 2000. Raw data for May, June and July 1999 were reanalyzed with the updated code at a later time.
3. Distribution of Meteor Echoes
3.1. Daily Echo Rates
 Daily underdense meteor echo rates in 1999 and 2000 are summarized in Figure 1 together with daily mean local K indices at Syowa station. Note that the echo rates between late July 1999 and April 2000 cannot be simply compared with those of the other months because the raw data between those months were not processed with the upgraded online software described in the previous section. To facilitate a comparison the rates during the period are multiplied by 1.5 and shown as dotted lines in Figure 1. Bearing this in mind, the daily echo rates are typically somewhere between 300 and 1500.
 One can see an annual variation of echo rates with a broad peak in the first half of the year around day 80. A year to year variation also seems to exist. Echo rates in the latter half of 1999 are generally lower than those of 2000. Recently the annual variations of meteor echo rates were reported using VHF meteor radar systems in both Arctic and Antarctic regions [Singer et al., 2004; Janches et al., 2004]. At a Northern Hemisphere site, ALOMAR (69°), Singer et al.  showed that the rates are maximized in June and minimized in February. At the south pole station (90°), Janches et al.  demonstrated a annual cycle with a peak occurrence in December/January period. In both studies summer time enhancement is the same feature, which is quite different from the present study. As shown in section 3.2 the MF radio wave is subject to the total reflection and absorption, mostly during daytime, and these effects seem to be largely responsible for the difference, hindering the MF radio wave from reaching meteor trails.
 A large day to day variation of echo rates is also evident. The echo rates and the K indices are apparently anticorrelated. The relation is more clearly seen in Figure 2 where (Figure 2a) daily echo rates and (Figure 2b) total number of echoes for a given K index are plotted. This anticorrelation is reasonably explained by the fact that high electron densities in the lower E region under disturbed conditions severely attenuate the MF radio wave and suppress the number of meteor returns from the region. Figure 2b indicates that the most meteor echoes in the present study were detected under very quiet to mild conditions (K = 0 ∼ 3). Thus wind velocities estimated using these echoes (shown in section 4) represent winds under undisturbed conditions.
3.2. Height Distributions
Figure 3a shows an example of height distributions of the O mode underdense echoes in winter (solid line). One month of data during the polar midwinter period, June 1999 (23,634 echoes), is used. The echoes apparently distribute from 80 km to 120 km and sometimes even higher. Errors of range and height caused by group retardation of the radio wave are often discussed and estimated for MF radar spaced antenna techniques [e.g., Namboothiri et al., 1993, 1994; Hall, 1998]. It can also be a problem in the present meteor study, although most data were taken under undisturbed conditions as shown in the previous subsection.
 Here we try to estimate the true height distribution undisturbed by group retardation. The X mode is more affected in an ionized environment than the O mode, and is considerably attenuated and retarded even when the O mode radio wave is still almost free from retardation [e.g., Manson and Meek, 1984]. Thus, when the ranges at which simultaneous O and X mode meteor echoes are detected are similar, the effect of group retardation is reasonably thought to be negligible. We regarded meteor echoes as retardation free when they were detected in both O and X modes simultaneously with a range difference equal to or within 2 km (one range gate), and extracted 714 echoes for June 1999. The number of retardation-free echoes detected is very small compared with the total number of O mode echoes. This small number is partly because of the existence of attenuation and retardation and also because the simultaneous O and X mode operation (parameter set 2 of Table 2) was conducted only half of the total operation time, and the maximum sampling range was only 138 km due to the limited amount of memory of the radar system.
 The dashed line in Figure 3a shows the height distribution of the retardation-free meteor echoes, where heights are estimated from ranges of O mode echoes. Both distributions show a peak height around 100 km with a gradual decrease in number with descending height, and reach a minimum at around 80 km. Above the peak height, however, the retardation-free distribution indicates almost no echoes at 120 km while the other indicates the existence of some echoes even above 120 km. The meteor population near 120 km in the latter distribution is thought to be at least partly caused by group retardation, which exists even in a midwinter month. Electron density enhancement by aurora activity is thought to be one of the sources of the group retardation of meteor echoes.
 It is important to note here that the upper limit seen in the retardation-free distribution plot can be somewhat lower than a real value. The more strongly attenuated X mode echoes usually show a much smaller signal-to-noise ratio (SNR) than corresponding O mode echoes and thus become more difficult to detect than O mode echoes at greater heights, where ambient electron density is higher, resulting in a somewhat reduced estimate of the upper height limit. Therefore the real upper limit under the present experimental setup will be somewhere between the two distributions.
Olsson-Steel and Elford  conducted MF radar meteor observations (1.98 MHz) and obtained a height distribution with a peak and upper limit at ∼104 km and above 140 km, respectively. Their peak altitude is almost the same as in the present study. The 20 km difference in the upper limits arose probably because we chose only echoes showing exponential decay with a duration longer than about 1.5 s, and did not detect short-lived echoes, while Olsson-Steel and Elford  studied the height distribution itself, and included echoes with a much shorter duration.
 Here, as an example of nonmidwinter observation, we show height distributions during April 2000 (fall) in Figure 3b, where distributions for all the O mode echoes (solid line) and again simultaneous O and X mode echoes (dashed line) are plotted. The retardation-free distribution looks quite similar to that seen in Figure 3a. The height distribution of all the O mode echoes has some interesting features. Below the peak height of 102 km, the distribution looks almost identical with the retardation-free distribution, suggesting that the effect of the retardation is not very serious, while above the peak it shows significant enhancement, which is unlikely to be real but seems to be the result of group retardation during daylight hours. Thus, as seems natural, the effect of retardation is stronger at greater heights where electron densities are higher.
 Next the local time dependence of meteor distribution is discussed. Figure 4 shows a winter time example of the local time dependence of height distributions observed in June 1999. Each height profile in Figure 4a exhibits a Gaussian-shaped distribution with a peak occurrence around 100 km. Mean heights and standard deviations seen in Figure 4b are mostly around 100 km and 7 km, respectively, except at hours around local noon (0900 UT), when the mean height is increased by a few kilometers, probably due to weak group retardation. Even during the polar night period the atmosphere at the 100 km over Syowa weakly receives the solar radiation for several hours around noon.
 Hourly echo rates (1 month accumulated) shown in Figure 4c exhibit double peaks at around 0500 and 2000 UT unlike the well-known single peak distribution with a maximum occurrence in the morning [e.g., McKinley, 1961]. This peculiar distribution is thought to be mostly related to absorption of the MF radio waves around local noon and also during occasional aurora events, which often occur from somewhat before local midnight to early local morning over Syowa.
 The local time dependence in April 2000 is presented in Figure 5. During 0600–1200 UT (0900–1500 LT), there are few echoes detected and the corresponding height profiles show a broad peak around 110 km or even higher, clearly indicating the effects of retardation and absorption due to relatively high electron densities during that period. Nevertheless, the profiles in other hours, especially around 1700 UT (2100 LT), do not seem to be much affected, showing a peak height around 102 km, which is the same as that seen in the retardation-free distribution in Figure 3b. Height and local time distributions of meteor echoes in other fall and spring months (March, April, August and September) are generally similar to those in Figure 5. The only major difference is that the echo rate is higher in fall than in spring as seen in Figure 1.
 For readers' information, we also show height distributions during January 2000 in Figure 6 as an example of midsummer observations, when the ionosphere over Syowa is continuously lit. It is notable that many meteor echoes are observed for several hours around 0000 LT (1700–2200 UT) even when the sun is still above the horizon. At those hours the height profiles peak at around 100 km altitude with the mean and standard deviation about 103 km and 8 km, showing mostly similar features with nonsummer months seen in Figures 4 and 5. Outside of the midnight hours the height profiles and the mean heights are strongly affected by retardation, absorption and total reflection. For example, at 1500 and 2300 UT the profiles look elongated at higher altitudes and the corresponding mean heights apparently go up by several kilometers, presumably by group retardation. On the other hand, between 0200 and 1400 UT the profiles are suppressed to lower altitudes and the mean height at local noon (0900 UT) is 100 km. This is due to the high electron density, which is high enough to totally reflect the radio wave (2.4 MHz) and block the radio wave from penetrating into the lower E region.
 From Figures 4, 5 and 6 it can be said that with great care MF radar meteor wind observations can be continued all year round over Syowa, even during summertime, at least around local midnight hours.
3.3. Ambipolar Diffusion Coefficient
 We have estimated ambipolar diffusion coefficients, D, from the exponential decay of underdense echoes [McKinley, 1961; Ceplecha et al., 1998]. Because we do not measure ion and electron temperatures and also because the present meteor observations are conducted mostly under quiet conditions, the two temperatures are simply assumed to be equal in the calculation. To avoid any group retardation effect the retardation-free echoes extracted in the previous subsection are used. Figure 7 shows examples of scatterplots of D against height for (Figure 7a) June 1999 and (Figure 7b) April 2000. Model profiles of D [Thomas et al., 1988], adopting atmospheric temperature and pressure from the CIRA model atmosphere, are also shown as a reference. In both months the values of D exhibit exponential increase with height. This is an expected feature of D, which is inversely proportional to atmospheric density. The distributions are similar to the result at a midlatitude site [Tsutsumi et al., 1999]. At the same time it is also apparent that the distribution is clipped at around 100 m2/s. This is seemingly due to the observation limit set by the relatively slow sampling frequency (5 Hz) and meteor detection algorithm. The duration of meteor echoes in this study should be more than about 1.5 s to be detected as mentioned in section 2. Thus the observable largest values of D for echoes with peak SNRs of 10, 20 and 30 dB are 112, 223 and 335 m2/s, respectively. The SNRs of most of detected echoes are less than 20 dB. It is possible that more echoes can be detected above around 105 km by using a faster sampling frequency.
 On the other hand, the durations of echoes reflected at lower altitudes become longer. For example, the value of D at 90 km is about 3 m2/s and the corresponding echo duration is about 30 times longer than that at 110 km when the same peak SNR is assumed. Since echo duration is inversely proportional to the radio wavelength squared [McKinley, 1961], a typical duration of a VHF meteor echo around 90 km, a few hundred ms [e.g., Nakamura et al., 1991] corresponds to about 1 min for a MF meteor echo. Actually, the durations of observed MF echoes at or below 90 km altitude are sometimes comparable to the length of the recorded time series of 102 s, and it is often difficult to distinguish these echoes from echoes of other origins. This seems to be a major reason why only a small number of meteor echoes are detected below 90 km in spite of the fact that a large population is found there in a VHF meteor study.
 The diffusion of meteor trails in the perpendicular direction to the magnetic fields can be much smaller than that in the parallel direction [e.g., Jones, 1991]. Although the degree of the restriction may be less significant than expected according to a recent work by Dyrud et al. , the effect is still serious above around 105 km when the trails are closely aligned with the magnetic field. However, this effect is thought to be negligible or very minor in the present study, where the zenith angle of peak meteor occurrence is around 20–30 degrees and almost no echoes exist at zenith angles greater than 60 degrees because of the limited maximum sampling range of 176 km. Thus, considering the large dip angle at Syowa, about −64 degrees, virtually no trails are closely aligned with the magnetic field and are subject to the effect. Testing the effect by extending the maximum sampling range is an interesting topic for the future.
 Since the value of D has a strong height dependence as seen above, using D as a proxy of height [e.g., Yukimatu and Tsutsumi, 2002] is another way to avoid the retardation effect. Echoes detected around local noon in summer months can be partly utilized for wind study in this manner. However, since it is beyond the scope of the present work, use of summer time data will not be pursued in the subsequent sections.
4. Meteor Winds and Comparison With FCA Winds
 Before showing and discussing measured wind velocities, we discuss possible effects of electric fields on wind measurements using MF radar meteor echoes. Each meteor body has an impinging velocity which is much faster than the motion of the ambient atmosphere. Just after or during the formation of an ionized meteor trail the distribution of temperature and electron and ion velocities in the trail are very different from those of the ambient atmosphere. Because of their smaller mass the cooling time of electrons is much longer than that of the ions. Nevertheless, it is on the order of 10−3 s and 10−1 s at 80 km and 115 km, respectively [Baggaley and Webb, 1977], and is short compared with durations of echoes in the present study, the minimum of which is about 1.5 s as described in section 2. After thermalization, meteoric ions and electrons follow the motion of the ambient atmosphere. As the plasma density in the ionized meteor trail is higher than that of the surrounding atmosphere, the strong Coulomb force inhibits different bulk motions of meteoric ions and electrons.
 In the polar regions, electric fields may decouple the motions of plasma and the ambient neutral atmosphere. Their effects have been studied both theoretically and experimentally [Kaiser et al., 1969; Reid, 1983; Ogawa et al., 1985; Fujii et al., 1998; Forbes et al., 2001]. Fujii et al.  most comprehensively studied the effect over Tromso using the European incoherent scatter radar (EISCAT). They investigated the effects in two cases: a weak electric field condition (5 < ∣E∣ < 10 mV/m) and a strong field condition (∣E∣ > 25 mV/m). They found that the ion motion below 101 km is governed by neutral motion even under the strong electric field condition while under the weak condition the motion follows neutral gas up to higher altitude of at least 109 km and indicates a very strong neutral coupling even in the F region.
 Since measurements of electric fields over Syowa have not been made, the estimate of electric field effect on the measured meteor winds is not possible. However, MF radar meteor measurements can be conducted only under very quiet to calm ionospheric conditions as discussed in section 2; thus the effect is reasonably regarded as minimal. Nevertheless, the ion motion above 110 km may not be totally governed by the neutral motion, as shown by Fujii et al. . The wind data above that level, that is, at 110–120 km in this study, should be treated carefully.
 Now we show the performance of MF radar meteor wind measurement over Syowa, and also compare its results with FCA winds simultaneously obtained with the same radar system. Data taken in June 1999 and April 2000 are used for a detailed comparison. These months are not chosen simply as examples of midwinter and nonmidwinter months but as examples in which relatively nice agreement (June 1999) and significant disagreement (April 2000) are found between meteor and FCA winds as will be shown shortly.
 Meteor winds are estimated for time-height bins with dimension of 2 hours and 4 km. The height resolution is chosen in order to compare with FCA winds whose height resolution is also 4 km. As the vertical wind velocity is generally much smaller than the horizontal components for atmospheric phenomena with a timescale larger than a few hours, it was reasonably neglected in the wind estimation and only horizontal components were calculated as in the work of Tsutsumi et al. . A horizontal wind vector was fitted by employing a least square method only when the number of underdense meteor echoes in each bin was equal to or more than 5. The bin was shifted by 2 km, as in FCA wind estimation, and/or 1 hour, that is, half the size of a bin, and the calculation was repeated. When the daily echo rate is more than 1000 or so, winds at the peak occurrence height of 100 km can be quasicontinuously estimated during nonsummer months. The FCA winds, which were recorded every 2 min and oversampled every 2 km, were averaged every 2 hours with a time step of 1 hour.
Figures 8 and 9show examples of wind velocities observed on 19–22 June 1999 and 24–27 April 2000, respectively. The daily meteor echo rates in these periods were 1000 or more. The corresponding daily mean local K indices showed relatively low values of 2 or smaller as seen in Figure 1, implying that the effect of group retardation was relatively minor. Meteor winds are estimated at least up to 110 km and sometimes at higher altitudes. In June 1999, complicated wave patterns are seen in both meteor and FCA winds. However, a periodogram analysis (figures not shown) shows that a 12-hour component is dominant below 100 km in both techniques and no clear dominant components are found above that level. The meteor and FCA winds agree well in the overlapping height region of 86–100 km. Agreement is especially good in the lower height region. However, the FCA winds tend to underestimate the meteor winds at greater heights.
 The situation in April 2000 is rather different. A dominant wave pattern with a periodicity of 12 hours, presumably a semidiurnal tide, is very clearly seen in both meteor and FCA winds. A periodogram analysis (figures not shown) also indicates that the activity of 12-hour periodicity is conspicuous in both measurements, especially above 86 km. At around 90 km, there is good agreement. However, at greater heights, there is a significant difference, although the 12-hour periodicity is seen in both. The meteor winds grow in amplitude with increasing height and show a downward phase progression, which is a feature of an upward propagating atmospheric wave. The amplitude of the 12-hour component in the meteor winds is as large as 60 m/s at 100 km altitude. Similar height variations are not seen in the FCA winds, but the wave structure looks almost unchanged at 90–100 km except for the zonal component on day 117, where downward phase propagation is apparent around 1200 UT. Overall agreement is much poorer than in June 1999, and there is a clearer tendency for the phase of the 12-hour component in the meteor winds to lead that in the FCA winds.
 In order to evaluate the differences statistically we calculated correlations between the meteor and FCA winds. Simultaneously observed bi-hourly winds were selected for both June 1999 and April 2000. To minimize the effect of group retardation we used winds only before 0700 UT (1000 LT) and after 1100 UT (1400 LT) for April 2000. For June 1999 all the simultaneously observed wind velocities were applied to the correlation analysis. The results are summarized in Figures 10 and 11 for June 1999 and April 2000, respectively. Scatterplots of meteor versus FCA winds are displayed every 6 km altitude for 86–90 km, 92–96 km and 98–102 km together with histograms of the difference between the meteor and FCA winds (meteor - FCA). Each regression line is fitted so that the sum of the squared distances between the line and each point is minimized. Note that this least square fitting method equally weights the meteor and FCA winds assuming the same estimation error in both techniques. Equal weighting may not be the best way but it is virtually the only way to fit a regression line because it is not easy to estimate wind errors for completely different measurement techniques without bias. The very small meteor echo rates below 90 km, as seen in height profiles in Figure 3, can introduce rather larger estimation errors in meteor winds, which may result in less steep than real slopes in Figures 10 and 11.
 In June 1999 the correlation coefficient for the meridional component below 90 km has a rather high value of 0.68. The FCA winds slightly underestimate the meteor winds, the slope being 0.76. However, no DC offset is seen. It is evident that the correlation becomes less significant and the slope becomes more gradual with increasing height. The height variation of the correlation is visualized in the wind difference plots. The standard deviation of 18 m/s at 86–92 km grows to 25 m/s at 98–102 km. On the other hand, the correlation in the zonal component is somewhat worse. The slope at 86–90 km is only 0.48. The tendency of less significant correlation and less steep slope with increasing height is the same as for the meridional component.
 In April 2000 the height variations of the correlation seen in June 1999 are even more pronounced. The slope and correlation coefficients at 86–90 km are 0.68 and 0.61, respectively, for both zonal and meridional components, comparable to the value at 86–90 km of the meridional component for June 1999. The correlation becomes significantly poorer and the slope less steep at higher altitudes. Apparently there is almost no correlation at 98–102 km altitude in the meridional component. The wind difference plots at 98–102 km show a large standard deviation of about 40 m/s in both the zonal and meridional components. This poor correlation seems largely due to the phase lag of the semidiurnal tide between the two measurement techniques shown in Figure 9.
 The semidiurnal phase structure is more clearly recognized in the monthly mean day plots of both techniques shown in Figure 12, where only simultaneously obtained bi-hourly wind data are used. Note that day time data at 0700–1100 UT (1000–1400 LT) are also plotted for April 2000 in order to facilitate better understanding of the phase structure, although a several km group retardation might have occurred near 100 km during those hours. In June 1999 (Figure 12a) a semidiurnal tide exhibits a similar downward phase propagation in both FCA and meteor winds although there is a tendency for the meteor winds to slightly lead FCA winds by an hour or so in the entire height range of 86–100 km. The lag is most evident above 94 km in the zonal component at around 1500 UT, where the rather steep phase progression in the meteor winds is hardly seen in the FCA winds. The amplitudes of the semidiurnal tides below 90 km are somewhat larger in the meteor winds while the differences become apparent at higher altitudes, especially at 100 km in the zonal component, where the amplitudes are nearly 20 m/s and less than 10 m/s in the meteor and FCA winds, respectively.
 In April 2000 the phase lag between the meteor and FCA winds is more striking as seen in Figure 12b. Below 90 km the meteor and FCA winds mostly agree well with the semidiurnal amplitudes of about 20 m/s. However, the smooth height variation of phases seen in the meteor winds is not very obvious in the FCA winds but much more gradual. The remarkable amplitude growth with height in meteor winds is not seen in the FCA winds. In general the FCA winds exhibit less height variation while the meteor winds show a more realistic picture. In other words the FCA winds look as if they are elongated upward or the heights are overestimated by about 6–8 km at 100 km altitude.
 For comparison, monthly mean days composed of not only simultaneously obtained wind data but all the available bi-hourly data are estimated for both June 1999 and April 2000 (figures not shown). The characteristic differences between the meteor and FCA winds discussed above in Figure 12 (86–100 km) reappear, indicating that it is not the data selection criterion employed to produce Figure 12 that caused the differences.
 In this section we discuss possible causes of the differences between the meteor and FCA winds as seen in the previous section. There have been various comparison studies of MF radar FCA winds with other observation techniques such as meteor radars, Fabry-Perot interferometers and satellite borne instruments [e.g., Cervera and Reid, 1995; Manson et al., 1996; Turek et al., 1995; Burrage et al., 1996; Hocking and Thayaparan, 1997]. These studies indicate that the MF radar FCA technique can give reasonable wind estimates in the mesosphere and lower thermosphere. However, it also has a tendency to give lower estimates than wind values simultaneously measured by other techniques, especially in the upper height region, mostly above 90 km. Cervera and Reid  found through simultaneous MF FCA and VHF meteor wind observations that agreement is good below 90 km, while above 90 km the FCA winds underestimate the meteor winds. It is generally accepted that the FCA technique requires careful parameter setting and also adequate antenna and receiver configurations to yield reliable wind velocities. Several possible causes of the smaller MF FCA winds have been proposed such as saturation of receivers at upper altitudes, inadequate sampling frequency (undersampling), improper receiving antenna spacing (the so called “triangle size effect”), and insufficient signal-to-noise levels [e.g., Meek, 1990; Vincent et al., 1994; Holdsworth, 1999].
 In the present study, receiver saturation is thought to be a minor problem because the receiver gains are carefully tuned so that the receivers are not saturated very often during nighttime. Furthermore, simultaneous meteor wind observations are hard to conduct in the first place under conditions such that the ionospheric return is strong enough to saturate the receivers.
 Other causes of biased FCA winds such as the triangle size effect may be partly responsible. However, readers should note that those effects may underestimate the magnitude of wind velocities but will not introduce phase lags of semidiurnal tides as seen in Figures 12 and 13. The observed FCA winds in those figures appear elongated upward as if the winds at lower heights contaminate the winds at the upper heights. This fact suggests that there is a mechanism which introduces a height determination error. Group retardation is a possible source of this error. However, if this is the case, simultaneously observed meteor data should be retarded more severely because almost all of the meteor echoes are detected in oblique directions, meaning that the radio wave propagates through a longer path in partially ionized media. Since this is not the case, group retardation can be ruled out.
Hocking  discussed the effect of the finite pulse width of the transmitter signal on height determination. The half-power-full-width of a typical MF radar transmitter pulse is 3–4 km (28 μs = 4.2 km in the present study) and usually the total width of the pulse is a few times longer. Hence if strong E-region echoes exist, they can contaminate the spread edge of the receiving pulse at lower altitudes, possibly down to 90 km, as illustrated in Figure 7 of Hocking . This effect, however, causes a height error in the opposite sense, contaminating from the upper region to the lower region, and thus can be discarded in the present study.
 On the other hand the validity of meteor winds in the polar region has been questioned. The effects of strong electric fields on the motion of meteor trails have been studied by several researchers [e.g., Ogawa et al., 1985; Forbes et al., 2001]. However, as discussed in Section 4, the effects are thought to be negligible or very small in this study. Since there is no evidence that the observed meteor winds in this study are biased, it will be plausible to assume that the FCA winds are mostly responsible for the differences between the meteor and FCA winds.
5.1. Angles of Arrival of Echoes
 In the current FCA measurement by Syowa MF radar it is assumed that the transmitted signal is partially reflected by horizontally stratified layers. Thus angles of arrival of the echoes are mostly in the vertical direction. If this is not the case but the signal returns off-vertical, actual heights are somewhat biased downward unless correction is made. Further, there is also a possibility that a number of meteor echoes are received at ranges shorter than 100 km together as in the case of VHF meteor studies [e.g., Holdsworth et al., 2004]. To test these possibilities we estimated angles of arrival (AOAs) of echoes in the sampling range of 70–100 km for 19–22 June 1999. Since we stopped raw data acquisition in July 1999, an AOA estimate for April 2000 is unfortunately impossible. The AOAs were calculated only when the corresponding FCA wind for the range and time of interest had been already successfully estimated. Considering the fact that we do not know the characteristics of possible scatterers (such as life time) which may affect FCA measurements, the AOA calculation was made in the following two ways, varying the length of the time series.
 1. Three phase differences from the four receivers are estimated using the whole 102-s time series. This means that a single AOA is determined for each set of time series. Since only two phase differences are actually necessary to estimate the AOA, a few AOA candidates are obtained from the four receiver outputs. Data with a poor signal-to-noise ratio usually yield AOA candidates which are inconsistent with each other. Only when the calculated AOAs are directed within 3 degrees of each other are they assumed to be reliable, and the vectorial average of them is calculated and used for further analyses.
 2. The 102-s-long time series was first cut into eight segments, each of which is 12.8 s long. The AOA for each segment is calculated in the same manner as in calculation method 1. Hereafter we express the AOAs estimated in calculation methods 1 and 2 as A1 and A8, respectively. Readers should note that we assume single scatterer at a time and range in the following AOA analysis.
 An example of A8 plots for 20 June 1999 is shown in Figure 13. The AOAs are distributed around the zenith in all the ranges, 70–80 km, 80–90 km, and 90–100 km. However, it is clear that echoes are observed in larger off-vertical directions at greater ranges. At the range of 70–80 km, the maximum zenith angle is around 10 degrees and only 7.8% echoes exceed 10 degrees, but at the range of 90–100 km the maximum angle is 20 degrees or more and 27.1% echoes exceed 10 degrees. These features are almost unchanged for A1. The only difference is that the obtained zenith angles for A1 are somewhat smaller.
 In order to investigate the possible existence of time variations of the AOAs, we calculated hourly mean values of them. Since a vectorial average of the AOAs will give only small off-vertical angles as can be expected from Figure 13, we averaged the azimuth and zenith angles separately in a manner such that zenith angles were simply averaged and that azimuthal angles were averaged as arctan (Σ sin(azimuth)/Σ cos(azimuth)). The hourly averaged azimuth and zenith angles of A8 are shown in Figure 14 for two range groups, (Figures 14a and 14b) 80–88 and (Figures 14c and 14d) 90–98 km. The overplotted solid lines are the hourly mean azimuth and zenith angles of meteor echoes averaged in the same manner. Note that the hourly average for meteor echoes is based on all the meteor echoes detected in the range 96–176 km. Thus the solid lines in Figures 14a and 14b are the same as those in Figures 14c and 14d, respectively. For reference, hourly meteor echo rates are shown in Figure 14e.
 The hourly mean zenith angles of meteor echoes show large off-zenith angles of about 30 degrees throughout the period 19–22 June 1999, except during hours when the hourly meteor echo rates are extremely low (10 or below). On the other hand, the hourly mean azimuth angles of meteor echoes rotate anticlockwise once a day. This rotation is a well-known behavior of sporadic (nonshower) meteor echoes and originates from the combination of the revolution and self-rotation of the Earth [e.g., Cervera et al., 2004]. Owing to the fact that more meteor bodies enter the atmosphere from the morning side of the earth, an observer fixed to the Earth sees the preferred direction of AOAs rotate overhead as the earth rotates. Because of the perpendicular condition for a meteor trail to be detected, actual directions of reception are 180-degree opposite.
 Now that we understand the behavior of meteor arrival angles, we return to the AOAs of MF signals received in the range 80–100 km. In the range 80–88 km the hourly mean zenith angles of A8 (Figure 14a) are distributed around 5 to 6 degrees. The 4-day averaged value is 5.7 degrees. On the other hand, the hourly mean azimuth angles show a weak tendency to follow the AOA motion of meteors. In the range 90–98 km the hourly mean zenith angles (Figure 14c) align around 8 degrees, with a 4-day average of 7.6 degrees. The corresponding azimuthal angles now indicate a clearer tendency to rotate together with the AOAs of meteors. AOAs in the range 100–108 km are also estimated (figures not shown) although FCA wind techniques are not often applied at such high altitudes. The 4-day mean zenith angle is an even larger value of 13.8 degrees. As for A1, 4-day averaged zenith angles in the ranges 80–88 km, 90–98 km and 100–108 km are 3.8, 5.6 and 11.3 degrees, respectively, about 20–30% smaller than those for A8. These smaller values might suggest that the life times of the scatterers are closer to 12.8 s than 102 s. Scatterplots of the hourly mean azimuths of A8 and meteor echoes are shown in Figure 15 for the two sampling ranges 80–88 km and 90–98 km. The correlation is larger in greater ranges. These AOA analyses suggest a possibility that the MF radar signals in polar winter are affected by meteor returns, possibly down to around 80 km. However, even if we can assume that meteors occupy some portion of the FCA signal, it does not directly indicate that meteors disturb FCA measurement. It is not known at present if meteor echoes can ever modify both magnitudes and directions of FCA winds, as in the examples shown in section 4. This possibility remains a future issue to be studied while other sources should also be sought.
 Here we check possible height errors caused by the oblique incidence of the received signals. From Figure 14 the height errors are estimated to be only about 94·(1 - cos 7.6°) ≃ 0.8 km and 104·(1 - cos 13.8°) ≃ 3.0 km at the ranges of 94 km and 104 km, respectively. These height errors alone cannot account for the differences in magnitudes and directions between the meteor and FCA winds of June 1999. So far we assumed only single scatterer at a time and range. However, when it is not the case, the obtained AOA is only an effective angle of multiple AOAs and can point more closely toward the zenith than reality. In order to test this possibility techniques which allow the existence of multiple scatterers such as IDI (Imaging Doppler Interferometer) technique should be employed [e.g., Adams et al., 1986]. However, it is beyond the scope of the present study, and should be tried in the future.
5.2. Profiles of Semidiurnal Tides
 Before leaving this section we show monthly mean height profiles of semidiurnal components for months from April to August in 1999 and 2000 in order to demonstrate the wide height coverage of Syowa MF radar observations and also to give readers additional information about the agreement and disagreement between the meteor and FCA winds in months other than June 1999 and April 2000. All the available bihourly wind data are applied to the harmonic analysis. Amplitudes and phases are plotted from 64 km to 116 km in Figure 16, where red and blue lines correspond to the zonal and meridional components, respectively. For almost all months, agreement between the meteor and FCA profiles around 90 km is reasonably good. The phases and amplitudes are smoothly connected from FCA heights to meteor heights. The amplitudes below 90 km are 5–10 m/s in all months shown here. The corresponding phase structures are also very similar below 90 km, indicating a vertical wavelength of about 60 km. In contrast, amplitudes above 90 km show remarkable seasonal variation. Those in midwinter months, June and July, at 100–110 km reduce to 0–5 m/s. Amplitudes in nonmidwinter months are mostly larger than those at the lower levels. The most conspicuous example is April 2000 when both zonal and meridional components reached around 50 m/s at around 104 km. Amplitudes in April in subsequent years have been almost the same (figures not shown).
Murphy et al.  studied semidiurnal tides over Antarctica using Davis, Syowa and Rothera MF radars. According to their findings the semidiurnal component in winter months over Syowa is dominated by a migrating semidiurnal tide, at least below about 90 km. The height region above that level, in particular in the E region, has hardly been investigated in Antarctica. We are now conducting tidal wave analyses fully utilizing this wide height coverage of Syowa MF observations. The results will be presented elsewhere.
6. Concluding Remarks
 In this paper we presented the first results of long term meteor observations made with Syowa MF radar (69°S, 39°E), Antarctica. During the polar night period the MF radio wave can penetrate into the lower thermosphere without suffering much absorption or retardation. Thus the whole local time of 24 hours can be covered in winter while summer time observations are not easy. Nevertheless, a significant number of echoes are observed for several hours around 0000 LT in summer. The daily echo rate remains around 300–1000 throughout the year. Comparison of echo rates with local K indices strongly indicates that most MF meteors are obtained under geomagnetically quiet conditions. Therefore the effect of the electric field on neutral wind measurements is thought to be insignificant. Above about 110 km, however, the effect may not be totally negligible, even under quiet conditions, according to IS radar results.
 Height distributions of underdense echoes were carefully studied utilizing the characteristics of O and X mode polarizations. Retardation-free echoes, which were detected in both polarizations at almost the same range, were distributed from 80 km to 120 km with the peak occurrence height at around 100 km. Since the upper limit is thought to be set mainly by the relatively slow sampling frequency employed in the present work (5 Hz), faster sampling should be tried to detect more short-lived echoes at higher altitudes in the future. Nonetheless, the height range is high enough to extend conventional MF radar observations by nearly 20 km. The effect of group retardation seems relatively minor below 90–100 km. It is still not very significant above that level during nighttime, but can be serious during daylight hours together with the effect of strong absorption and total reflection. Much caution is necessary when using day time MF meteor data.
 Meteor and full correlation analysis (FCA) winds at 90–100 km were compared for the two nonsummer months of June 1999 and April 2000. Agreement around 90 km was rather good while the FCA winds underestimated the meteor winds at higher altitudes, being consistent with earlier findings by other researchers. A remarkable finding in this study is that not only the magnitudes but directions of wind velocities can also be different. That is, the semidiurnal tidal phase in FCA trailed that of meteors at upper altitudes, especially in April 2000, when the FCA winds showed less height variation and it looked as if the FCA height was overestimated above 90 km.
 A case study of angles of arrival (AOAs) estimate was made using raw data taken during 19–22 June 1999. AOAs in the range 80–100 km showed a tendency to rotate once a day following the motion of meteor AOAs. The tendency was clearer at greater heights. Although these results suggest the existence of meteor effects on MF signal at ranges shorter than 100 km, this does not directly indicate that FCA technique is affected by meteor echoes. Quantitative evaluation of meteor echo contamination effects on FCA wind velocities should be tried in the future. Besides, even if we can assume that a height error is caused by the oblique incidence of the received signal, it is still too small to account for the differences between the meteor and FCA winds. In the present AOA analysis we assumed only one scatterer at a time. However, it might not be realistic and an analysis which allows the existence of multiple scatterers is needed to avoid any biases in AOA estimation. Other sources which can cause the differences between the meteor and FCA winds should also be sought.
 In this study we showed that the height range of MF radar measurements could be expanded to 60–120 km by merging preferred height ranges of the meteor and FCA techniques. It is expected that MF radar meteor measurement in polar regions, although mostly restricted to winter months, can greatly contribute to the study of polar E region dynamics which, until now, could only be probed using larger-scale observation facilities such as incoherent scatter radars.
 This research has been supported by a Grant-in Aid for Scientific Research (14740287) from the Japan Society for the Promotion of Science (JSPS). The Japanese Antarctic Research Expeditions (JAREs) have carried out the MF radar operation at Syowa.