Multi-instrument derivation of 90 km temperatures over Svalbard (78°N 16°E)



[1] For 90 km above Svalbard at 78°N, 16°E, neutral air temperatures derived from potassium lidar and OH spectrometer observations are combined and used to calibrate the corresponding estimates derived from meteor echo fading times. While lidar and spectrometer results depend heavily on observing conditions, a meteor radar can easily provide daily measurements; the three instruments thus complement each other to yield daily temperature estimates, hence providing a temporal coverage hitherto unprecedented at this location.

1. Introduction

[2] The continuous monitoring of mesospheric temperatures has long been a notoriously difficult task. Radars depend on gradients or even discontinuities in refractive index to obtain echoes, and in the mesosphere neither water vapor nor electrons necessarily abound to provide these, at least, not for the scattering processes which give temperature information. Thus while radars are able to operate continually, the near absence of scalar tracers makes it difficult to measure the neutral atmosphere. When sufficient ionization is present, for example during auroral precipitation or in sunlight, the results obtained can only be said to be, a priori, representative of those particular geophysical conditions. As we shall see, the problem is alleviated by using a meteor radar in which irregularities in the meteor trail provide the necessary refractive index gradients.

[3] Optical observations often rely on dark conditions, not to mention clear skies. If one assumes hydroxyl (OH) molecules are at the same temperature as their environment, measuring their emissions yields the average temperature of the hydroxyl layer. At high latitude, no summer observations are possible, and during early and late winter, measurements are representative of nighttime tidal phases and thus day averages are unobtainable. Resonance lidars, such as the potassium lidar described subsequently, are able to operate in daylight conditions. They require personnel, however, and automatic year-round operation is impracticable. In fact, any kind of observation requiring personnel can be problematic, especially in winter, at high latitude due to weather conditions (both Arctic and Antarctic) and polar bears (Arctic).

[4] Another method is that of in situ sounding. Such measurements are of course snapshots of the atmosphere and thus a high repetition rate must be used to eliminate, at least, tidal biases. A continuous observational program would thus be prohibitively expensive in terms of both personnel and nonreusable ordnance. Furthermore, if passive payloads are used (e.g., to reduce costs), a reference temperature often has to be used in the derivation of a temperature profile, thus requiring optical instrumentation anyway.

[5] In this pilot study we demonstrate how it is possible to utilize the best characteristics of a meteor radar, an OH spectrometer and a potassium temperature lidar (K lidar) in a complementary way in order to obtain day average temperatures from 90 km above Svalbard, at 78°N, 16°E. In future developments of the method we shall describe, it will be possible to extend the altitude range and to increase the time resolution. In addition to the study of temperature variation itself, the results may be combined with those from incoherent scatter radar, such as the nearby EISCAT Svalbard Radar (ESR) in order to investigate positive ion composition and negative ion concentration [e.g., Hall, 1989].

2. Instruments

[6] The Nippon/Norway Svalbard Meteor Radar (NSMR) has been described by Hall et al. [2002]. In summary, the sky is illuminated at 31 MHz and echoes from meteor trails detected by a 5-antenna interferometer array. The maximum echo occurrence occurs at 90 km, with typically 800 echoes per day at the peak itself, the half-width of the occurrence peak being around 7 km and the total number of echoes around 6000 per day. The range resolution is 1 km, and the height resolution is therefore of the same order, depending on the spatial distribution of echoes within the field of view. By grace of the high latitude of this instrument, echoes are received at all times of the day (in contrast to lower-latitude systems which exhibit strong diurnal variation), and thus day averages do not suffer from a tidal bias. Similarly, the azimuthal distribution is reasonably homogeneous within a field of view of 50° zenith angle. The fading times of underdense echoes are employed for estimation of the ambient neutral air temperatures using two methods, described respectively by McKinley [1961] (employing a pressure model) and, more recently, Hocking [1999] (using a temperature gradient model). The system has been operating since March 2001, and improvements to antenna tuning, interferometer calibration and range resolution were made in the autumn of that year. A system failure caused an interruption of some weeks in the early summer of 2002, but otherwise measurements are available for almost all days since the start of operations to present. Rather than attempt to determine temperatures over a height regime, we have limited our scope to only the height of maximum echo occurrence, 90 km. Within 5–10 km above the height of maximum echo occurrence, the echo fading times may often become affected by electrodynamics [Dyrud et al., 2001], whereas below, in the upper mesosphere, fading may be augmented by turbulence [Hall, 2002]. Contrary to the impression one might obtain from the literature, derivation of temperatures from meteor echo fading times is non trivial, and seldom results in believable values: indeed, examination of Hocking [1999] reveals that a form of calibration is necessary to bring the meteor-echo derived temperatures “into line” with concurring independent measurements. More recently, Hocking et al. [2004] and Singer et al. [2004] have improved the temperature gradient model used in the Hocking [1999] approach. This improvement has proven so satisfactory that further scaling is no longer necessary. While temperature estimations from meteor radar observations at lower latitudes have been able to employ multitudes of independent measurements, not least the empirical model of Lübken and von Zahn [1991] for 69°N, this has not been the case for Svalbard (78°N) until recently. Hitherto, we have resisted the temptation to use independent temperature measurements from 69°N for fear of introducing an a priori assumption that the temperature structure at 78°N is similar. Moreover, few temperature data are available for Svalbard, most being winter measurements by OH spectrometer, and summer data in particular are sparse; indeed, the data sparsity over the high arctic is the very motivation for this study. Our philosophy here, therefore is not to strive toward best possible temperature determinations using NSMR alone, but rather to accept external calibration will be necessary and then to take advantage of simultaneous optical measurements. In a way, NSMR is being used a tool to assist interpolation between irregularly spaced optical measurements.

[7] An OH spectrometer is situated some 5 km to the northwest of the NSMR site. The instrument is of the Ebert-Fastie type and is described by Sigernes et al. [2003, and references therein]. The temperatures derived are representative of an 8 km thick layer centered at 87 km [Baker and Stair, 1988]. As stated earlier, the success of obtaining day average temperatures from such an instrument depends on clear skies, and darkness (including in the absence of moonlight). In addition, temperatures can usually only be derived absence of aurora (whose spectrum contaminates that of hydroxyl). While it is possible to exclude moonlight using software, and thin cloud can be tolerated, data are visually checked and aurora-contaminated and low-signal-to-noise spectra rejected.

[8] The K lidar as described by von Zahn and Höffner [1996] was located 15 km to the northwest of the radar. Almost all observations have been performed under daylight conditions which require a special daylight filter [Fricke-Begemann et al., 2002]. The K lidar was in operation between June 2001 and August 2003, and most observations were performed throughout the summer seasons 2001 and 2003. The K lidar is able to measure temperatures over a large altitude range and long periods of time; however measurements are limited by the necessary clear sky conditions. Furthermore, the possible altitude range and accuracy of the temperatures depend on the seasonal potassium density variations. By focusing on average temperatures at 90 km altitude with 1 km vertical resolution the K lidar has observed altogether temperatures from the end of February to the beginning of October (although not each year). The typical statistical error is in the order of a few Kelvin and the inherent uncertainty in the measurement system is estimated to be less than 5 K at the most difficult polar summer conditions. The analysis method including a discussion of errors is given by, again, von Zahn and Höffner [1996] and also Lautenbach and Höffner [2004].

[9] Thus, while there are spatial separations between the instruments, the volumes viewed by both spectrometer and lidar are within that viewed by the radar. Nevertheless, the volume sizes are not the same, and so an important assumption is that the temperatures determined by each method are intercomparable.

3. Method

[10] The philosophy of the temperature calibration is simple yet, as we shall see, effective, and is not unlike the approach of Hocking [1999]. Any more complicated method would almost certainly involve implicit assumptions about the relationship between diffusivity of the ions in the meteor trail in air and be contrary to the theory underlying most similar derivations. For each day average independent measurement (i.e., both OH and lidar), we identify the corresponding temperature obtained from the meteor radar on the corresponding date. To do this we invert the expressions Chilson et al. [1996] give for ambipolar diffusion coefficient in terms of temperature and combine with atmospheric pressures obtained from the constituent number densities provided by the MSISE-90 model [Hedin, 1991] and assume a zero field reduced mobility of 2.4 × 10−4m2V−1s−1 for ions in the meteor trail (the impact of the model pressures on the final result will be discussed later). Since, as already stated, the OH measurements are representative of 87 km altitude, we have examined both CIRA-86 [Rees et al., 1990] and MSISE-90 models and determined corrections to extrapolate the values to 90 km. These are specified in Table 1, a value being simply subtracted from the measurements according to month. The time span for OH measurements is November 2001 to February 2003, and that for K lidar measurements is June 2001 to August 2003.

Table 1. Reduction in OH Temperature in Order to Extrapolate From an Assumed Layer Centered at 87 km to the Altitude of the Meteor Trail Echo Preferred Height, 90 km
MonthReduction, K

[11] The original temperature estimates before calibration are shown in Figure 1 to facilitate readers comparing with other meteor radar results. Scatterplots of meteor radar temperature measurements versus independent results were then made and linear regressions performed, as shown in Figure 2. For the OH measurements there is a one-to-one correspondence in dates; however, for historical reasons, the K lidar day averages are from 1200 to 1200 UT. For these points, therefore, we have used the average meteor radar temperature for the two days spanning the lidar averaging period. Since there appears to be a considerable spread in OH measurements with respect to K lidar measurements, we have performed the regression with both the combined optical data sets and with the K lidar data set only. The results are given in Table 2. These may be contrasted to Hocking [1999, equation (11)], namely, Ttrue = 0.774 Tmeteor − 42.8, where Ttrue is the estimate of the true temperature and Tmeteor is the completely uncalibrated temperature derived from the theory. We thus employed the following calibrations of the meteor radar determined temperatures:

equation image

again where Ttrue is the estimate of the true temperature and Tmeteor is the completely uncalibrated temperature. The above regression lines are shown in Figure 2. The reasons for the necessity of calibrating the meteor radar derived results are suggested by Chilson et al. [1996]; however, since it seems that some degree of “magic” scaling is invariably necessary, and also that we have required the use of a model atmospheric pressure, we shall refrain from hypothesizing as to the possible causes. Furthermore, we explicitly assume the spectrometer and lidar derived temperatures to be correct.

Figure 1.

Neutral air temperatures at 90 km altitude over Svalbard (78°N) deduced from measurements of meteor echo fading times according to the method described by Chilson et al. [1996]. Month annotation indicates the 1st of the respective month. The values until the end of September 2001 were of lower quality during the installation period, the radar operating in a low height resolution single-pulse mode and with questionable interferometer calibration, and are shown as points. Only the data points depicted by pluses are used in this study.

Figure 2.

Scatterplots of independent day-average temperature measurements by potassium lidar (pluses) and OH spectrometer (diamonds) versus corresponding values from Figure 1. Each point represents a specific date as opposed to a day of the year. A least squares linear fit is shown by the solid line. Cases corresponding to the slope and intercept ±1-sigma uncertainties are indicated by dashed lines. (top) Linear regression of meteor radar temperatures on combined spectrometer and lidar data. (bottom) Regression on lidar data only.

Table 2. Results of Linear Regression of the Meteor Radar Determinations of Temperature on Those From the Optical (OH Spectrometer and K Lidar) Instruments As Shown in Figure 2
K lidar measurements only−144 K ± 5 K0.47 ± 0.03
Combined measurements−138 K ± 5 K0.52 ± 0.03

[12] Figure 3 shows the results of calibrating Tmeteor using the combined optical data. Owing to the aforementioned changes in altitude resolution and other initial problems with NSMR, 2001 data prior to September may be suspect. In addition, Figure 3 includes the temperatures predicted by CIRA-86 [Rees et al., 1990] and MSISE-90. Also indicated are the original temperature determinations from the optical instruments. Uncertainties in the corrected temperatures are indicated by error bars, as explained subsequently. While it would seem logical to improve statistics by including as much independent data as possible, the summer minima observed by the K lidar are slightly better reproduced in the calibrated meteor radar results when only the former is used in the regression, despite the fact that it might be expected that the OH temperatures would only affect the winter results. However, the differences would be virtually undetectable in the format of Figure 3 and are therefore not shown here.

Figure 3.

Corrected temperatures (black pluses) for 2001, 2002, and 2003 using the linear calibration deduced in Figure 2 and using both spectrometer and lidar results. The green error bars indicate the anticipated uncertainty, i.e., ±17 K (see section 5). Month annotation indicates the 1st of the respective month.

4. Discussion

[13] The method of derivation of neutral atmosphere temperature from meteor radar echo fading times combined with independent measurements by other instruments certainly seems to yield plausible results. There are a number of assumptions, however, which we shall list here.

[14] 1. The radar has a considerably larger field of view than either the OH spectrometer or the K lidar, so we must assume that the temperature field is homogeneous over all fields of view. On the other hand, long (1 day) integration times should largely reduce this effect.

[15] 2. The winter contribution to the calibration comes largely from the OH measurements; it is assumed that the function-of-month OH layer temperature adjustment (i.e., Table 1) for extrapolating from 87 to 90 km to the radar scattering volume 89–91 km is suitable.

[16] 3. Day averages from each instrument are truly representative and not substantially biased by tides/waves (discussed in the next section).

[17] 4. The calibration obtained is constant with time of year, and that using a linear (as opposed to higher order) regression is justifiable. We have tried to illustrate this by including the optical instruments' results in Figure 3, in which we see that the calibration might be yielding overestimates during late summer and autumn but good agreement in winter.

[18] 5. Electrodynamics and turbulence do not contaminate the meteor echo fading times (a reasonable assumption at 90 km).

[19] While CIRA foresees a well-behaved sinusoidal variation in temperature over the year, our results show considerable temporal structure. In fact, the very departure of our results from the model illustrates the potential problems of temperature calculation which relies on a pressure model as input. The pronounced seasonal variation with summer minima around 110 K just after the solstice dominates the picture; however there is considerable variability in the winter half of the year such that maxima are not necessarily around the winter solstice. We can see that MSISE-90 does indeed exhibit asymmetry about the solstices, with a shift of the winter maximum toward November, and this is more in line with our observations. While the fall in temperature in spring roughly follows CIRA, the transition from summer to winter states is much more abrupt. This is a little surprising since Hall et al. [2003] report that the dynamics' summer state persists longer at 78°N than at 69°N, and in both cases, the wind reversal is in late September. By this time, in fact during the course of August, the temperature has already attained its average winter value of around 200 K.

[20] A particularly interesting feature is the approximately 1-month periodicity in the autumn. This strong periodic variation is also a feature of the Antarctic mesosphere [Tsutsumi et al., 2001], and considering the large differences in surface topology between arctic and Antarctic regions, this is unlikely to be attributable to orographic forcing alone (longitudinal variation has recently been studied by Shepherd et al. [2004a], but only for equinox conditions). Moreover, Singer et al. [2004] present results for 69°N in which both a rapid autumn transition and high late autumn variability are featured, while Shepherd et al. [2004b] focus on the spring transition. Heating of the stratosphere by solar rotation modulated UV [e.g., Luo et al., 2001] may be responsible for planetary waves of approximately one month period, and these may be filtered by the stratosphere/lower mesosphere temperature and background wind structures such that they are only manifested in the 90 km temperature field during certain seasons. Liu and Roble [2002] and very recently Cho et al. [2004] discuss the response of the mesosphere and lower thermosphere to stratospheric warming; the mechanisms here are also relevant for explaining our temperature observations.

5. Errors

[21] While sources of errors have been touched upon above, we shall address them in more depth. Users of temperature information will require varying degrees of precision depending on the application, so it is important to document exactly what the meteor radar system is offering. Although meteor echoes are received all day, there is a degree of diurnal variation, although as mentioned earlier, not as acute as at lower latitudes, and combined with varying temporal coverage by the optical instruments (e.g., due to cloud and daylight), comparison of results from the instruments may be more justifiable on some days than others. Indeed, the K lidar temperatures often exhibit large variances during the course of a day. Despite the day averaging, we might expect particularly poor agreement between observations due to the larger spatial averaging of the OH and meteor methods and the fact that a K lidar “day” is from 1200 to 1200 instead of 0000 to 2400 UT. We have hoped to alleviate this problem by using a large number of days of measurements.

[22] Inclusion of radar data with questionable height calibration and 3 km range resolution, such as those prior to October 2001 resulted in a considerably poorer general agreement between the calibrated radar results and the optical observations and unrealistically low summer temperatures. It was concluded that mixing altitude resolutions in such a study was to be avoided; insufficient data are available to establish whether or not acceptable results could be obtained using only 3-km resolution radar data.

[23] The basic temperature calculation (i.e., Figure 1) employs atmospheric pressure provided by the MSISE-90 model, which is our rationale for also showing the CIRA model in Figure 3. In order to determine the influence this choice of model has on the calibrated values, we have performed the analysis using a fixed pressure (in fact the year average from MSISE-90). The resulting calibrated temperatures were very similar to those for when the daily model values were used, although the fit to the optical observations was visually poorer, quantified by the mean absolute deviation being about 25% larger. We conclude that while using a model time-varying pressure model is desirable, its choice does not appreciably affect the final calibrated values. In turn, we conclude that the displacement of the winter maximum to well before the solstice is not some manifestation of a similar feature of MSISE-90.

[24] Again, it should be stressed that we have not put in a large effort to derive realistic temperatures using the meteor echo decay times alone; we have rather accepted that the method we use is somewhat inadequate (illustrated by the scaling Ttrue = 1.9 Tmeteor − 263 as contrasted with Hocking's [1999]Ttrue = 0.774 Tmeteor − 42.8). Rather, we have obtained a parameter Tmeteor linearly related to the temperature, and have obtained the latter by combining observations from colocated instruments.

[25] As a check, we have performed a linear least squares fit of the calibrated meteor radar derived temperatures to the independent measurements (both K lidar and OH spectrometer), and confirmed that the slope and intercept are unity and zero respectively (Figure 4). The mean absolute deviation was found to be 17 K and this is deemed to be a good representation of the uncertainty one would expect in any temperature predicted by the radar. The spread of points either side of the unity slope line in Figure 4 indicates the variability in the measurements and not necessarily errors. It is the combination of this variability together with the intermittent nature of the optical observations that is primarily responsible for discrepancies between optical and meteor results on given days. In winter, tidal amplitudes have been found to be of the order of a few K according to earlier OH observations; gravity wave amplitudes, on the other hand, may easily attain 20 K (e.g., as indicated by inspection of K lidar measurements before day averaging) and are thus of the same order as the mean absolute deviation illustrated by Figure 4. For some studies, such as those related to noctilucent clouds, and temperature sensitive reaction rates, such uncertainties are unacceptably large, but for others such as dynamics, the calibrated temperatures may be more valuable.

Figure 4.

Calibrated meteor radar derived temperatures as a function of the corresponding lidar and spectrometer measurements. As expected the regression line has unity slope and zero intercept, as shown by the solid line. The mean absolute deviation is indicated by the dotted lines above and below.

6. Summary

[26] It is a sad fact that the theory for determination of ambient temperature from meteor echo fading times has been, to a degree, inadequate: the values resulting from a calculation based on the theory alone must invariably be scaled in some way, in practice. The method described by Hocking [1999] involves such a scaling; however, we do not know whether this is suitable for 78°N. Recently, Singer et al. [2004] and Hocking et al. [2004] have very successfully derived temperatures using meteor radar in combination with temperature gradient models and obviated the need for further scaling. To avoid the need for such a temperature gradient model for 78°N, we have therefore obtained independent temperature values from two separate instruments measuring within the scattering volume of the meteor radar to form a calibration which is thus peculiar for that geographic location and radar configuration. The resulting temperature variation only agrees reasonably with model predictions during midwinter and spring and even then apparent wave activity generates discrepancies of up to 30 K, Furthermore, we find a degree of asymmetry, the spring cooling at 90 km is relatively gradual compared to the more abrupt recovery to the winter state through August. The winter half year average temperature is similar to the model prediction, and there is an indication that the summer minimum is a little cooler than the model prediction. Planetary wave activity is less in summer than in winter, and there appears to be a tendency for strong activity in the autumn when the meteor radar indicates considerably higher temperatures than the models. This overall spring-autumn temperature difference is apparently not evident at lower latitudes [e.g., Shepherd et al., 2004a] and may therefore be a high arctic feature. Similarly while a transient peak is evident in early March 2002 at 78°N, there is otherwise little evidence for the late March–April increase reported for midlatitudes [Manson et al., 2002].

[27] This pilot study has attempted to establish a satisfactory methodology for determining mesospheric temperatures on a regular basis by combining results from three instruments. Only two years of results have been presented here, yet the NSMR operations are open-ended; with time we expect to build statistically reliable temperature estimates which will in turn contribute to improved models and provide input to interpretation of results from nearby incoherent scatter and MST radars.


[28] The first author thanks NFR for support. The second author thanks the Ministry of Education, Science and Technology for the grant-in-aid money on the Arctic upper atmosphere coupling study in constructing the NSMR system.