A comparison of mesosphere and lower thermosphere neutral winds as determined by meteor and medium-frequency radar at 70°N



[1] There has been much discussion as to the veracity of neutral wind measurements made using medium-frequency radar (MFR) employing the spaced-antenna technique. Such systems are able to operate continuously, providing information on mesosphere and lower thermosphere dynamics with typical resolutions of 3 km in altitude and 5 min in time, and thus represent a low-cost monitoring of the atmosphere. It is similarly important to be able to trust the results, and therefore we make a dedicated comparison between the Tromsø MFR (70°N, 19°E) and the newly installed and colocated Nippon/Norway Tromsø meteor radar. The agreement is particularly good between 75 and 85 km.

1. Radars

[2] The Tromsø medium-frequency radar (MFR), located on the Norwegian mainland at 70°N, 19°E, has been described in detail by Hall [2001]. It provides measurements of horizontal neutral winds at 5 min intervals and with a height resolution of 3 km through the upper mesosphere and lower thermosphere (MLT). The transmitter antenna illuminates the atmosphere with a beam of width 17° at half power and at a frequency of 2.78 MHz. Resulting partial reflections from structures in electron density create a pattern on the ground moving according to the advection of the reflecting structures by the neutral wind. The movement of this pattern is detected by 3 equispaced receiver antennas. This method of spaced receivers and the correlation analysis of signals from them has been fully described by Meek [1980] and Briggs [1984], but it should be noted that development of spaced-receiver data analysis is an ongoing process [e.g., Holdsworth and Reid, 1995]. Altitudes of the range gates, earlier somewhat uncertain, have recently been calibrated by Hall and Husøy [2004]. Determinations of winds from this and other similar systems have been used extensively in studies of dynamics from gravity wave to general circulation scales [e.g., Manson et al., 1999, 2004; Hall et al., 2003] and for model development [e.g., Rycroft et al., 1990; Hedin et al., 1996]. The measurements are considered, by many workers, to be good representations of the true wind amplitudes and directions. Nevertheless, some multi-instrument studies have cast doubt on the veracity of the MFR method (mainly speeds, and not directions), a particular problem being disagreement with incoherent scatter measurements over short timescales and especially during geomagnetically disturbed conditions [e.g., Nozawa et al., 2002]. On the other hand, many more studies have suggested that the MFR-determined winds are indeed correct regarding directions, but with speed biases of circa 20% [Manson et al., 1996; Meek et al., 1996]. An aspect of using MF radars, not least in the auroral zone, is that the ionosphere (viz. total reflection at 2.78 MHz) often extends down to or below 90 km such that the precalculated echo altitudes are only virtual because of group delay of the radio wave [Namboothiri et al., 1993]. The aforementioned altitude calibration does not attempt to take the group delay into account, and the effect must be borne in mind throughout this paper.

[3] The Nippon (NIPR)/Norway Tromsø meteor radar (NTMR), is of the type often referred to as a meteor wind radar or recently simply as a meteor radar (MWR), and was installed in late 2003. This system is virtually identical to a corresponding installation on Svalbard (78°N, 16°E), operates at 30.25 MHz, and is colocated with the Tromsø MFR. The description of the Svalbard system applies also to the Tromsø installation and may be found in work by Hall et al. [2002] and Holdsworth et al. [2004]. MWRs illuminate a large region of sky and echoes from meteor trains are detected by receivers arranged as an interferometer. Descriptions of determination of winds may be found in work by Aso et al. [1979], Hocking et al. [2001a], and Tsutsumi et al. [1999], for example. A MWR at high latitude suffers less from strong diurnal variation of meteor rate than at low latitude, and therefore it is possible to obtain a time resolution of 30 min. The resulting 30 min average winds represent, however, a spatial averaging over perhaps 200km at the peak echo occurrence height of 90 km. This averaging, like the group delay problem at MF, is an important factor to be borne in mind when comparing results from the two systems. To further punctuate the difference in spatial averaging between the radars, Figure 1 shows a meteor echo distribution as function of zenith angle for a typical day; virtually all echoes originate from between 70 and 100 km. The time and height resolutions and fields of view of the MFR and NTMR are summarized in Table 1.

Figure 1.

Zenith angle distribution of NTMR echoes (a total of over 10,000) for 4 April 2004, almost all occurring between 70 and 100 km altitude. The effective field of view of NTMR is therefore ±70°. The MFR 17° field of view (i.e., half-power full-width antenna aperture) is shown by the red circle.

Table 1. Salient Parameters of Tromsø MFR and NTMR
  • a

    The effective MFR field of view also depends on aspect sensitivity (which decreases with height, probably because of gravity waves tilting the horizontally stratified scatterers).

  • b

    The effective field of view of NTMR depends on the meteor echo distribution; the time resolution of NTMR is chosen from the meteor count rate.

Height resolution, km31
Time resolution, min530
Field of view, deg17∼140

2. Winds Comparison

[4] In the period from January to April 2004, for each 30 min averaged wind profile (i.e., both zonal and meridional components) obtained by NTMR, a corresponding 30 min averaged profile was formed from up to six 5 min profiles from the MFR. On occasions simultaneous data were not available because of either system failure or atmospheric conditions (e.g., total absorption of the radio wave by the ionosphere). Figure 2 shows the numbers of 30 min averages common to both radars as a function of height.

Figure 2.

Histogram of numbers of 30 min average wind values common to both radars, as a function of height.

[5] Comparisons of data sets akin to those presented here, for which the measurement uncertainty is unknown for both sets, has been addressed by Hocking et al. [2001b]. These authors presented a novel algorithm for deriving a linear relationship between two such data sets, and this is indeed our goal here. Although it would have been possible to have employed this algorithm, we have, instead, applied more traditional techniques and from varying initial assumptions as described below. Let us examine the winds at 79 km, this being well below heights at which we expect group delay to affect the MFR; the number of common wind measurements is near the maximum of over 3000 (out of a possible 5800 half-hour intervals over 4 months) as seen in Figure 2. We have constructed a scatterplot of MFR versus MWR values for each of the zonal and meridional components in Figure 3. In order to perform a least squares fit to this data we must first assume which wind measurements are to represent the independent variable and which are dependent. Taking the MWR values as the independent variable and using a minimum absolute deviation linear fit, a gradient is obtained which is less than that which might have been found had the fit had been performed by eye. A method minimizing the chi-square error statistic results in an almost identical fit although this method is more sensitive to outlying data points. Taking the MFR data as the independent variable results in a gradient somewhat steeper that would result if fitting by eye. Unfortunately, we cannot deem, a priori, which data should be used as the independent variable; each radar has a different sampling volume and the wind measurement is indeed characteristic of that particular volume. The MFR can be expected to exhibit more variability due gravity waves, whereas for the MWR, gravity waves of horizontal wavelength smaller than the radar field of view will “disappear” in the spatial averaging. In order to investigate this problem, we have smoothed the data in time using a boxcar filter in order to remove the effects of gravity waves (and for that matter high-frequency tidal modes). Compensating for differences in fields of view by NTMR data selection or modification of the MFR were not viable options. Figure 4 shows the result of this smoothing: least squares (minimum absolute deviation) fits were performed using both MWR and MFR data as the independent variable, and for each of the zonal and meridional components, as a function of boxcar length. In Figure 4, the boxcar length is given in hours, the length in data points is double this (the spacing between data points being 30 min). In addition we have averaged the gradients obtained by the two approaches. We see, gratifyingly, that with increasing boxcar length, the gradients converge. However, while the convergence for the zonal component is toward a constant value, that for the meridional component the ratio appears to decrease with increasing smoothing. This suggests that while increasing the boxcar length may remove gravity wave contributions, longer-period waves only present in the meridional component are seen by one radar and not the other. It should be recalled that Figure 4 portrays the case for 79 km and we have not attempted to investigate the altitude dependence of this characteristic, nor do we wish to embark on the potentially open-ended experiment of extending the filter length beyond 24 hours.

Figure 3.

Scatterplots of MFR versus MWR winds for 79 km: (top) zonal component and (bottom) meridional component. The blue line shows least absolute deviation linear fit with the MWR values as the independent variable; the yellow line shows linear fit by minimizing the chi-square error statistic; and the red line shows least absolute deviation linear fit with the MFR values as the independent variable. The rationale for these different fits is explained in the text.

Figure 4.

Results of smoothing the 79 km data from each radar prior to performing least squares linear fits: (top) zonal component and (bottom) meridional component. The gradients resulting from using the MWR as the independent variable (red line) and the MFR (blue line) and the average of these two are plotted as functions of the length in hours of the boxcar used for the smoothing.

[6] As we have seen, a number of linear regression algorithms exist. Alternative methods include minimizing the sum of squares of normals from the points to the regression line thus avoiding any a priori assumption as to which measurement was correct (if any), and the method of Hocking et al. [2001b]. As stated earlier however, here we choose to perform the more traditional minimalization of the sum of absolute deviations in one coordinate, taking first one and then the other measurement as being the independent variable, and in this way it is possible to see the consequences of assuming a particular radar/method is the correct one.

[7] With only qualitative criteria, we shall continue the investigation with a 2 hour smoothing; the means of the gradients are still near unity for both wind components and the results will be valid for evaluating MFR wind validity for timescales as short as any tidal mode. Using this philosophy, we extend the comparison to the MFR range gates at {70, 73 … 103} km altitude, again for the two wind components, in Figures 5 and 6. The numbers of points in each panel have already been given in Figure 2. Although data are sparse at 70 and 73 km, the fits still appear believable when qualitatively comparing adjacent heights. On the other hand, at 100 km and above there is little correlation between measurements. The spread in points at all heights is in part, at least, attributable to the differing dynamics scales observed by the two radars and gravity wave amplitude growth with height combined with the differing fields of view of the radars may induce an altitude dependency in the comparison. We may then plot these gradients (MWR as independent variable, MFR as independent variable, average of these, zonal, meridional) as functions of height in Figure 7, and simultaneously, we may now think of these as the ratio between MFR and MWR measured winds.

Figure 5.

Scatterplots of MFR and MWR zonal winds for altitudes between 70 and 103 km inclusive. The data have been smoothed by a 4-point (2 hour) boxcar. The blue lines indicate the minimum absolute deviation linear fits with the MWR data as the independent variable, and the red lines indicate the minimum absolute deviation linear fits with the MFR data as the independent variable. The black lines are the averages of these.

Figure 6.

As in Figure 5, but for the meridional component.

Figure 7.

Fit gradients as functions of height: (top) zonal and (bottom) meridional. Blue crosses show MWR as the independent variable; red crosses show MFR as the independent variable, and the black line shows the average of these. As seen by the axis notation, we now think of these as the ratio of amplitudes of wind components as observed by the two radars.

[8] We have hitherto assessed the differences in both zonal and meridional wind amplitudes measured by each radar, and therefore implicitly the differences in total wind and direction. While knowledge of the total wind speed uncertainties is important for assessing the reliability of tidal and wave amplitudes and the magnitude of the general circulation, directions and their variability are important for the study of tide and wave phases. To assess this aspect we have chosen to abandon the scatterplot approach since spread in azimuth around 360° results in points clustering not only around the (hopefully) unity gradient regression line, but also in low-abscissa/high-ordinate and high-abscissa/low-ordinate regions of the plot (viz. top left and bottom right in the scatterplot formats elsewhere in this paper). Moreover, while a scaling factor is a useful concept for speed, a difference is more appropriate for direction, and therefore we have chosen to determine the difference between wind directions (i.e., azimuth) measured by the two radars. For selected heights, we have constructed histograms for these differences using a 1° bin size: Figure 8. At 103 km there are relatively few data points (cf. Figures 5 and 6) and only below 76 km do any marked differences begin to emerge.

Figure 8.

Histograms of differences in observed wind direction (azimuth) between the two radars (MWR-MFR) using 1° bins and at the same selection of heights as in Figures 5 and 6.

3. Discussion and Conclusions

[9] Visually, from the scatterplots we have presented, the agreement between the MFR and the MWR is generally good, at least below 100 km. At 70 km and above 103 km the data are sparse. While agreement at 70 km and below might be demonstrated to be good had more data been available, increasing spread in the scatterplots above around 90 km suggests that the MFR is performing less well there. This can be due to the group delay causing the MFR echoes to originate from lower height than the MWR winds, and also due to a breakdown of the assumption that drift of the structures seen by the MFR is purely due to the neutral wind. The effect is also clear from Figure 7 in which divergence of the different independent variable profiles (particularly for the zonal component and with MWR wind as the independent variable) becomes dramatic from 90 km and upward. Below 90 km, using the average of the independent variable profiles, however, we can see that the MFR-determined zonal wind lies between 75% and 110% of the MWR values and similarly for the MFR-determined meridional wind. There is an indication of a more systematic (20%) underestimation of the meridional wind by the MFR, which is in good agreement with the findings of Manson et al. [2004] while comparing the Tromsø MFR with a MWR some 200 km distant. For the zonal wind, the difference is modestly altitude-dependent; Manson et al. [2004] noted that the MFR and MWR zonal winds agreed rather well in summer but that the MFR indicated increasingly lower values (20–50% in the region 82–94 km) in winter. Amplitudes in tidal modes vary with height, season, component and location, such that the altitude variation of the MFR-MWR difference we see and the winter-summer difference reported by Manson et al. [2004] may both be consequences of the respective radars' fields of view (Manson et al. [2004] used data from radars with a separation of 200 km, that is, of the same order as the MWR sampling volume diameter). We believe this may be due to spatial variation of gravity wave flux, since orographic forcing is likely to vary over the footprint of the MWR sampling volume. At Tromsø, for example, the surface topology is one of mountains at the radar site, with increasingly open sea to the NW and flatter elevated terrain to the SE. Moreover, gravity wave fluxes are similarly seasonally dependent, and so deposition of momentum (and therefore tidal acceleration for example) can be expected to vary: (1) over the field of view, (2) with season, and (3) with altitude [McIntyre, 1989; McLandress, 1998].

[10] Comparisons between MFR-derived neutral winds and those determined by the nearby EISCAT incoherent scatter radar facility [e.g., Nozawa et al., 2002, and references therein] have revealed substantial differences. Assumptions as to the stationarity and/or homogeneity of the electric field raise a degree of doubt as to the reliability of the incoherent scatter method in which neutral wind velocity is derived from the ion drift velocity with appropriate assumptions concerning the ion neutral collisions and field aligned electric field. Thus the advent of the NTMR MWR has represented the first opportunity to compare the Tromsø MFR results with a colocated system that determined neutral winds by a technique believed to be very robust. We find that below 90 km, the MFR and MWR provide neutral wind determinations correct to within around 20% but above (91–94 km) the speed bias is larger, while directions/phases remain useful for tidal studies. Below 76 km we see differences in the wind directions determined by the respective radars—apparently as much as 50° at 70 km—a phenomenon for which we are currently unable to offer an explanation except that the MWR echo frequency begins to tail off rapidly around these heights. Studies of tides and larger-scale-size phenomena should therefore be performed with care at these lower altitudes.

[11] As discussed above, we have found that MFR winds tend to be around 20% lower than the corresponding MWR measurements, in agreement with similar findings by Manson et al. [2004]. In addition to differences in differing fields of view being an explanation for this discrepancy, contributions can originate from aspects of the MFR experimental setup. Cervera and Reid [1995] also found MFR to yield lower values than MWR, but by somewhat less than 20%, this using a narrow beam MWR; improving the MFR receivers, however, further improved the agreement, indicating that the discrepancy can be attributable, at least in part, to the receiver dynamic range. MFR measured velocity has furthermore been shown to decrease with increasing receiver noise [Holdsworth and Reid, 1995; Zhang et al., 2004] and also with decreasing antenna spacing [Holdsworth, 1999].

4. Conclusions

[12] Several comparisons similar to this have been performed earlier, notably Thayaparan and Hocking [2002], who employed the Hocking et al. [2001b] method, and Cervera and Reid [1995] who used a method similar to the one used here but forcing the regression line to pass through the origin. It is evident that one cannot perform such a comparison and apply the result to all instruments of the same genre. Discrepancies between instruments may arise from differing fields of view, signal-to-noise ratio, and radar characteristics. The extent to which fields of view contribute may well depend on geographic location, surface topology and also season since mean winds affect gravity wave filtering and therefore momentum deposition in the middle atmosphere. Finally, we stress that we have not been so presumptuous as to assume that a given radar yields definitive values by which the other radar should be judged; furthermore we cannot exclude that both methods yield correct values of the neutral wind specific to their respective fields of view.


[13] The first author thanks the Norwegian Research Council for support. The NIPR authors thank the Ministry of Education, Science and Technology for the grant-in-aid money on the Arctic upper atmosphere coupling study in constructing the NTMR meteor radar system. The Canadian authors gratefully acknowledge grants from the University of Saskatchewan, the Institute of Space and Atmospheric Studies, and the national agency NSERC.