Changes in mesospheric dynamics at 78°N, 16°E and 70°N, 19°E: 2001–2012



[1] Mesospheric wind data from meteor wind radars situated on Svalbard (78°N, 16°E) and the Norwegian mainland (70°N, 19°E) are examined for evidence of systematic change during the interval 2001–2012. For both locations, we find changes that suggest a strengthening of the summer westward jet, a weakening of the local winter eastward flow and, yet at the same time, weak evidence for any significant corresponding trend in the winter poleward flow. There is also a suggestion of an increase in the altitude of the summer polar jet, but more data will be required to confirm this. The main finding of a strengthening zonal flow is consistent with earlier studies and also with a contemporary scenario of progressive strengthening of the Brewer-Dobson circulation. We show that inclusion of sudden stratospheric warmings strongly influences trends. There is no obvious causality between the changes detected over the 2001–2012 time interval and the solar cycle parameterized by total solar irradiance.

1 Introduction

[2] While increasing greenhouse gas (GHG) concentrations are usually thought of as responsible for climatic warming in the troposphere, following their transport to the middle atmosphere, they act as refrigerants and hence cause cooling and shrinking of the stratosphere and mesosphere [e.g., Roble and Dickinson, 1989; Rishbeth and Roble, 1992; Rishbeth and Clilverd, 1999]. Not unexpectedly, we can anticipate consequences for dynamics at various altitudes and spatial and temporal scales [e.g., McLandress and Shepherd, 2009; Hall et al., 2007; Laštovička et al., 2008, 2010; Keuer et al., 2007; Jacobi et al., 2003]. Using radars, we are able to monitor winds in the region ~70–100 km (depending on radar system and atmospheric conditions): for the meteor radars in Adventdalen, Svalbard (Nippon/Norway Svalbard Meteor Radar (NSMR)) at 78°N, 16°E [Hall et al., 2002] and at Ramfjordmoen, on the Norwegian mainland (Nippon/Norway Tromsø Meteor Radar (NTMR)) at 70°N, 19°E, the resolutions are 30 min and 1 km. It is the NSMR system alone that can currently provide time series of mesospheric winds extending to one solar cycle in the Norwegian high-latitude sector. In addition, NSMR can deliver neutral temperatures [e.g., Dyrland et al., 2010], although we shall not include this possibility here. Otherwise at high latitude, corresponding decadal changes in the wind field have been investigated by Portnyagin et al. [1993], in particular, from Heiss Island (80°N, 58°E) but for the period 1968 to 1977. A recent study by Jacobi et al. [2012] reports the findings at middle latitudes but also represents an up-to-date source of references to work in this field hitherto. On the whole, however, virtually all recent studies of long-term changes in mesospheric dynamics have been for latitudes somewhat lower than those results presented here. It is therefore impossible to make direct comparisons with other observations during the same 10 year period, but later, in a dedicated section, we compare similar results from earlier years and/or other latitudes.

[3] For the purpose of this study, we shall smooth the original data from the radar depending on what features we wish to emphasize. Our philosophy here will be to identify trends detectable over approximately one solar cycle and report them. We shall also focus on the uncertainties in the observed trends and whether or not they can be considered significant, following methods described by Taylor [1982], Tiao et al. [1990], Weatherhead et al. [1998, 2002], and Working and Hotelling [1929]. Studies of, for example, the ionosphere in order to detect falling pressure levels that could be attributable to a net shrinking of the underlying neutral atmosphere [e.g., Ulich and Turunen, 1997] have attempted to identify the dependence of the observable on solar flux (parameterized by, for example, sunspot number or UV flux). Apart from a qualitative comparison with total solar irradiance (TSI), we shall refrain from this variant here, leaving the causality—anthropogenic or solar—a debate to further research, although it should be mentioned that the observations can be interpreted as compatible with a scenario of increasing CO2 concentration [Keuer et al., 2007], a process of anthropogenic origin also leading to a shrinking of the middle atmosphere. Some of the 2001–2011 (i.e., approximately solar cycle) changes described herewith exceed those predicted to be attributable to solar flux changes between solar minimum and maximum [Gray et al., 2010; Tsutsui et al., 2009], suggesting that additional forcing, possibly anthropogenic [e.g., Jacobi et al., 2003], must be at work. The particular years involved, the geographical locations, and altitudes of these earlier studies dictate that care must be taken in making direct comparisons with our findings. While keeping in mind that only one decade of observation, and furthermore from only two geographical locations, is somewhat short and limited to draw any real conclusions as to solar influence, we are still reporting the 2001–2012 (for 78°N) and 2004–2012 (for 70°N) changes actually observed, time series which are continually evolving and which can and will be valuable for climate study.

[4] When evaluating the results that follow in the context of the overall dynamics of the polar region, it must never be forgotten that the observations are from two, and only two, geographic locations and therefore a priori representatives only of 78°N, 16°E and 70°N, 19°E. Driven originally by solar forcing, the biggest influences on mesospheric winds at high latitudes are the polar vortex in winter and, in summer, upwelling and the associated westward jet. The notions of “zonal” and “meridional” winds are rather arbitrary in the vicinity of the polar vortex: in this region, zonal symmetry is largely destroyed for a number of months each year due to meandering and shifting circumpolar wind patterns and periodic breakup of the vortex itself. The mesosphere to stratosphere coupling tends to be where the winter vortex is the most intense, and in this context, we shall illustrate the consequences of including sudden stratospheric warming (SSW) [e.g., Charney and Drazin, 1961] signatures in the data. The geometric shift of the entire vortex and the associated wind patterns in its vicinity as well as changes in the strength and shape of the vortex show up superficially as fluctuations in meridional and zonal winds, especially when observed from a single location. Any true trend in the mesospheric wind patterns (whether zonal or meridional) measured over a single location can only be concluded when this aspect has been addressed in a coherent way, i.e., by assimilating observations from a number of locations. In order to further address the dynamics, establishing any trend in the Brewer-Dobson circulation requires observations in which seasonal differences in wind velocities have been measured at relatively uniform distances from similar polar vortex patterns or must be derived from local observations at multiple locations. Model simulations of the Brewer-Dobson circulation predict a change of only ~1–2 m/s over 150 years near the poles even at altitudes of 10 hPa. [McLandress and Shepherd, 2009]. To demonstrate similarity to model predictions, experimental observations of winds measured over a decade must be extremely accurate, even assuming a stationary vortex. Thus, given the limitations of the observations reported in this study, care must be taken not to “over-interpret” the 2001–2012 changes we shall present but rather consider them to be a contribution to a future understanding of climatic change in middle atmosphere dynamics.

2 Results

[5] As described by Hall et al. [2002], both NSMR and NTMR echo receptions are performed by five antennas and respective receivers operating as an interferometer; this yields phase differences at a 2 km resolution (a 4 bit complementary code used at NSMR and NTMR) whence velocities at which meteor trails are transported by the background neutral wind are determined. Every 30 min, the average horizontal wind is determined over the field of view and, using 2 km altitude bins, by a least-squares fit using all available echoes. The 2 km altitude bins overlap by 1 km resulting in an apparent 1 km resolution. Thus, the horizontal wind data we shall work with have an initial altitude resolution of 1 km and time resolution of 30 min. Due to the sporadic nature of meteor arrivals, not all altitude bins are populated during each 30 min time slice; however, subsequent averaging/smoothing of the data generally yields contiguous profiles at least 10 km either side of the peak height for echo occurrence, viz. ~90 km for both NSMR and NTMR. A major consideration during acquisition and subsequent implementation of the two radars was that they should be virtually identical such that the results can be directly comparable. Any instrumental errors (differences in which could, for example, arise from numbers of meteors detected at the two sites) are swamped by the intraday and (in the analysis presented here) monthly variances. Before working with these data, for each altitude, magnitude-based data rejection is applied to remove unrealistic values. In Figure 1, we show altitude profiles of means and standard deviation (σ) for both zonal and meridional components from NSMR for the period 2002–2012 inclusive. We have chosen to reject magnitudes exceeding 3σ although we have repeated the analyses described forthwith for different criteria. For 90 km, for example, 3σ implies an upper limit of ~80 ms−1 and well outside the dynamic range observed by, for example, the Ramfjordmoen MF radar (69°N, 19°E) [Hall, 2001] co-located with NTMR, and at low latitude [e.g., Eckermann et al., 1997], although Meek [2012] did report isolated instances of stronger winds. In order to illustrate the overall nature of the winds' data series, we show basic data set at various stages of preconditioning. To avoid overwhelming the reader, we have selected data from 80 km altitude. The original 30 min observations for the NSMR zonal component are shown in Figure 2 (first panel). Thereafter, the same data are shown following rejection of all values outside the 3σ limit as described above. Finally, in Figure 2 (third and fourth panels), the data (zonal and meridional components, respectively) are shown smoothed with a 2 day (96 point) running mean purely for illustrative purposes to eliminate the diurnal and semidiurnal tides and thus better emphasize the seasonal variation. We have selected this altitude because the features are clear to see and because 80 km can be considered part of the mesosphere during both summer and winter. At lower altitudes, increasingly fewer meteor trails are available for analysis; earlier, we moved into the lower thermosphere and also features like the seasonal variation become less pronounced. Short horizontal lines are caused by data gaps. Positive values indicate eastward winds (westerlies), and negative values indicate westward winds (easterlies). The most striking feature overall is the seasonal variation seen in both components, the amplitude of which exceeds all other variability for the zonal wind but is more comparable to the shorter period variability for the meridional component; the minima—really maxima in westward flow—are present in summer and show the upper part of the summer mesospheric jet. The shorter period variability encompasses planetary waves at various periods, the activity being somewhat suppressed during the summer months where the flow is westward (easterly winds) and stronger during autumn and winter. Significant changes were made to the radar (NSMR) in autumn 2001, so in the analyses that follow, we exclude data prior to October 2001. In Figure 3, we have smoothed the winds using a 15 day boxcar running mean revealing the month-to-month variation and a clear trend (over the 10 year time span) for summer data, but not for winter. We have defined summer months to be June and July and winter months to be November and December and have emphasized these in the figure. Using a simple least-squares fit, linear regressions were performed separately for summer and winter daily data. At 78°N in winter, there is a trend of −0.7 ± 0.06 ms−1 decade−1, i.e., virtually unchanging eastward flow over the 11 year period; in summer, the westward flow intensified at a rate of 6.4 ± 0.06 ms-1 decade−1 over the period. The 1σ uncertainties in the fits were only 0.06 ms-1 decade−1 for both winter and summer. At 70°N, in winter, the trend is −1.8 ± 0.13 ms-1 decade−1; in summer, −5.3 ± 0.07 ms-1 decade−1, i.e., decreasing winter eastward flow, and increasing summer westward flow. Tiao et al. [1990] suggested that for a trend to be significantly nonzero at a 95% confidence level, the absolute value of the trend should exceed 2σ, and this condition is satisfied here for both winter and summer at both latitudes. For comparison, we have also indicated the results of using the minimum absolute deviation method by the dashed lines in the figure: at 78°N, the winter and summer trends are −0.0 ± 0.1 and −7.0 ± 0.1 ms-1 decade−1, respectively. Thus, for summer, the 95% confidence level would be achieved even if the uncertainty was taken to be the difference between minimum absolute deviation and least-squares methods [Branham, 1982], whereas for winter, there is no significant nonzero trend. It should be pointed out, however, that this requirement assumes a Gaussian distribution of possible slopes given a (large) ensemble of formulations of the stochastic component of the time series. The actual nature of the processes giving rise to “noise” in such time series is the theme of current research [viz. Hall et al., 2011]: while true noise adds to the uncertainty of a regression analysis, memory (either short term as in an autoregressive AR(1) process, e.g., [Weatherhead et al. [1998]] or long term, e.g., [Lennartz and Bunde, 2009]) diminishes the confidence in the value of the determined slope since each data point has a degree of dependence on its predecessor. We see from the figure that the qualitatively larger uncertainty in the winter slope resulted from the greater planetary wave activity during that season, whereas during summer, the wave activity is less and the peaks in westward flow are better defined. Furthermore, substantial excursions can be seen in several Januaries—these are responses to sudden stratospheric warmings we have intentionally excluded by selecting November and December as the winter months.

Figure 1.

(top) Profiles of wind means and (bottom) their standard deviations, above Adventdalen, from 2002 to 2012 inclusive. Red, zonal component; blue, meridional component.

Figure 2.

Winds above Adventdalen, 78°N, 80 km altitude, with tidal fluctuations smoothed out to highlight seasonal variation. Top to bottom: (first panel) zonal component at 30 min resolution, positive eastward; (second panel) zonal component after rejection of values over 3σ (see text); (third panel) zonal component after rejection of values over 3σ followed by 2 day running-mean smoothing; (fourth panel) corresponding meridional component, positive northward.

Figure 3.

(top) Svalbard, 78°N. (bottom) Ramfjordmoen, 70°N. Month-to-month variation of zonal winds at 80 km altitude. Highlighting for positive values indicates data from November and December; for negative values, June and July. Straight lines indicate linear regressions using least-squares fits on these winter and summer subsets, showing little change in the eastward winter flow and an increase in summer westward flow. The dashed lines indicate the results of using a minimum absolute deviation method instead of least-squares.

[6] The preceding figures were intended as illustrations of the nature of the wind data, the overall seasonal variation, and the trends we can identify over 11 year (78°N) and 9 year (70°N) observation periods using daily wind values. We can now extend this strategy to more heights but, at the same time, restrict ourselves to the upper mesosphere regime 75–100 km. In order to better take care of data gaps, we have determined means for the summer and winter months (defined earlier) for each year of observation and for each altitude (1 km resolution). Somewhat anomalous winds in summer 2012 at 78°N may be due to radar problems, and therefore, we exclude this final summer hereafter. Linear regressions are then performed as before (viz. using the least-squares method), and 1σ and 2σ uncertainties are calculated. The results for 78°N are shown in the four panels of Figure 4: summer data (left column), winter data (right column), results for the zonal component (top row), and the meridional (bottom row). In each case, the linear regression line slope is depicted by the solid line. For zonal components, negative values indicate positive trends in westward flow, whereas positive values indicate positive trends in eastward flow. Similarly, for the meridional components, positive trends in poleward flow are given by positive values and positive trends in equatorward flow are given by negative values. Moreover, the 1σ and 2σ uncertainties are shown by the shaded areas. Where the absolute value of a trend's 2σ (viz. 95% confidence) limit exceeds zero, the trend is highlighted to indicate that the determination is significantly nonzero. The only really significant trends are in the zonal component in the approximate height regime 77–92 and 97–98 km in summer and the meridional component at 90–95 km also in summer. Other instances of changes that are significantly nonzero are at individual heights or, at best layers, only 2 or 3 km thick. On the other hand, it can be argued that even though a trend is not significant (e.g., at the 95% level), a whole height regime of perhaps 8 km of similar trends suggests that “there is something there”. This is perhaps the case for both components in winter. For the meridional component in summer, the slope from the regression is relatively small at all heights and significant at only a few. Note that the 80 km trends indicated in Figure 3 differ from those subsequently indicated in Figure 4 because, in the latter, we use monthly means (in all, 20 summer values and 22 winter values) which we consider more robust than simply fitting trend lines through smoothed (originally 30 min resolution) data. Figure 4 is based on true monthly means, whereas in Figure 3, the seasonal values are influenced by adjacent months due to the boxcar smoothing (e.g., in Figure 3, “winter” includes not only November and December data, but also October and January to a certain degree, whereas in Figure 4, it is strictly only November and December).

Figure 4.

Svalbard, 78°N. Trends as altitude profiles: (top left) summer, zonal; (bottom left) summer, meridional; (top right) winter, zonal; and (bottom right) winter, meridional. Solid line shows derived trend (in ms−1 decade−1) surrounded by shading, indicating the 1σ and 2σ uncertainties. Horizontal highlighting indicates where the detected trend is significantly nonzero with 95% confidence.

[7] In Figures 3 and 4, we have rejected winds of magnitude exceeding 3σ (σ being a function of altitude) as described earlier. From Figure 1, this corresponds to, for example, ~60 ms−1 at 80 km altitude. To check the consequences of this data rejection, we have repeated the analysis underlying the results shown in Figure 4 with different degrees of data rejection. The results are shown in Figure 5, for upper limits for wind speeds of 1–5σ (red: summer winds; blue: winter winds). We see that for all rejections, the decadal changes in summer zonal wind are all increases westward and that for 3–5σ rejection are all similar. Furthermore, as seen by the vertical bars, the uncertainties in the derived trends are similar and largely independent of the data rejection for limits greater than 2σ. In winter, there are no corresponding changes that are significant, irrespective of data rejection.

Figure 5.

Svalbard, 78°N. Derived trends together with uncertainties at 80 km (red, summer; blue, winter) as a function of data rejection. The independent variable is the number of standard deviations (viz. found from Figure 1) within which data are assumed to be reliable. Here the same monthly averages are used as are portrayed in Figure 4.

[8] In Figure 6 we investigate the consequences of defining winter months as December and January. This figure corresponds to Figure 4 (top right), viz. zonal component in winter at 78°N. It can be seen that the inclusion of SSWs drastically affects the trends we arrive at. Here we have simply assumed that SSWs tend to occur in the latter half of winter; to truly exclude SSWs from the trend analysis would be somewhat more demanding as minor warmings can occur in early winter. For the meridional component (not shown here), there is little difference except that standard deviations are greater. It is a matter of discussion as to whether SSWs should be included in trend analyses, particularly for longer datasets: should the “background” situation be treated separately, or are SSWs to be considered as part of the dynamics, or perhaps there is a climatic change in SSWs (e.g., both frequency and intensity) that should be considered alone?

Figure 6.

Svalbard, 78°N. Result of defining December and January as winter months as opposed to November and December. This figure corresponds to Figure 4 (top right). The difference is attributable to the occurrence of sudden stratospheric warmings in the latter half of winter.

[9] Addressing 70°N now, the equivalent of Figure 4 is shown in Figure 7. For the zonal component in summer, the situation is similar to that at 78°N except that the height regime exhibiting increasing westward flow is more restricted (80–87 km for significant values) and the decadal change is somewhat less. For the summer meridional component, however, the situation is reversed in that there is evidence for weak intensification of poleward flow between 80 and 88 km. In winter, there are no significantly nonzero changes, and again, variability is correspondingly larger than in summer, just as at higher latitude.

Figure 7.

Ramfjordmoen, 70°N. Trends as altitude profiles: (top left) summer, zonal; (bottom left) summer, meridional; (top right) winter, zonal; and (bottom right) winter, meridional. Solid line shows derived trend (in ms−1 decade−1) surrounded by shading, indicating the 1σ and 2σ uncertainties. Horizontal highlighting indicates where the detected trend is significantly nonzero with 95% confidence.

[10] For the last 11 years, a picture then emerges of fairly stable zonal wind climatology in the mesosphere and lower thermosphere, except in the approximate height regime 78–92 km. In that regime in summer (see Figures 4 and 7 (top left panels)), we detect significantly nonzero trends wherein the westward flow has increased at a rate of up to ~10 ± 3 ms−1 decade−1 (the slopes of the regressions are negative, but the wind was negative, i.e., westward in the first place). In winter, above 90 km, the eastward flow has decreased at a rate of up to ~10 ± 5 ms−1 decade−1 at both 78°N and 70°N (see Figures 4 and 7 (top right panels)) (the slopes of the regressions are negative, and the wind was positive, i.e., eastward in the first place). Associated with the increasing westward flow in summer, we detect little or no significant change in the meridional climatology, suggesting (only suggesting since these measurements are from two locations only), perhaps surprisingly, an intensifying westward circulation that is not associated with a corresponding increase in equatorward flow. On the other hand, associated with the decreasing eastward flow in winter above 90 km at 78°N, there is a degree of evidence (due to lack of significant trends and similarities between adjacent altitudes) for an unchanging poleward component. Also, it is interesting to note that reduction of eastward flow in winter is found when January data are included (this applies to both 70°N and 78°N although the former is not shown here), suggesting that increasingly with time, SSWs may cause deposition of horizontal momentum in the mesosphere, decelerating the zonal flow.

[11] Following Brasseur and Solomon [2005], in general, the polar mesosphere is characterized by westward flow in summer associated with a quasi-geostrophic equatorward component giving rise to an upwelling over the pole and hence the cold summer mesopause. Although this westward flow has increased during the last decade at 78°N, 16°E, it is remarkable that the equatorward component has remained almost constant on average; in fact, we observe a weak decrease in equatorward flow between 2004 and 2012 at 70°N. In winter, the mesospheric winds are characterized by poleward flow below 90 km, again associated with the eastward circulation [again, Brasseur and Solomon, 2005]. However, Hall et al. [2003] noted that the winter poleward flow was less ubiquitous in winter than in models [viz. Hedin et al., 1996] suggested and that, at least for 78°N 16°E, there was equatorward flow during short period midwinter in the upper mesosphere. The prime focus of this paper is changes in the wind field observed at 78°N and 70°N in the Scandinavian sector rather than the climatology, and the reader is encouraged to consult Hall et al. [2003]. A scenario of an unchanging midwinter equatorward flow yet combined with a diminishing eastward circulation around ~97 km is a possible description of the winter cases in Figure 4. Indeed, the change in equatorial flow can be expected to be much smaller in magnitude than that in the associated zonal flow and could be undetectable within the uncertainty of our observations hitherto. Thus, considering the lack of statistical significance and, not least, full geographic coverage of the vortex region, we abstain from exploring this feature in more depth: a longer data set may help resolve the problem.

[12] Despite the caveats, the conclusion from the trend analysis, at least of the zonal wind, is that the summer mesospheric jet has intensified during the last decade, and for results in the 80–90 km height interval, the confidence is 95%. We now take this aspect a step further and endeavor to investigate the temporal and spatial development of the jet itself. We determine the centroids of the summer westward components (i.e., minima in zonal wind) hypothesizing the jet is a feature of the mesosphere between May and August inclusive and between 77 and 90 km. The peak in meteor echoes occurs at ~90 km (for the operating frequencies of NTMR and NSMR, 31 MHz), and even though the jet extends to lower altitudes [e.g., Brasseur and Solomon, 2005; Hall et al., 2003], we limit our search to altitudes above 77 km because echoes are very sparse further down. Figure 8 shows the zonal wind as a function of height and time for the periods of observation so far. Again, the seasonal variation is the prominent feature with the summer jet clearly visible. Outside the summer period, the winds are generally eastward, and (although not a subject of this study) it is interesting to note the late winter increases in strength below ~80 km, particularly in early 2004 and 2006, these being associated with SSWs. This now illustrates the interplay between stratospheric and mesospheric processes and our reasoning for defining winter months as being November and December in order to avoid SSWs influencing the winter trends. The jet centers we determine are indicated by asterisks. We are able to see the problems faced in determining the jet centers because while we have identified times/heights of westward centroids, for Figure 8 (top; 78°N), for example, fitting by inspection alone might well yield lower altitudes for the years 2004 and 2006, while for 2005 and 2007 the altitudes might be higher. We have, however, used a completely objective method in this study although, in future work, it may be productive to investigate more sophisticated approaches. Indeed, at lower latitudes the core of the vortex occurs at somewhat lower altitude (viz. more than 5 km according to Keuer et al. [2007]). It is difficult to identify any such difference in overall altitude between the two latitudes investigated here; the features above 90 km are largely different, although the reversal from westward to eastward obviously occurs at lower heights at 70°N. Rather, more data will be required to investigate this quantitatively. Furthermore, at ~50°N the vortex upper boundary may also be sinking by several kilometers per decade [Offermann et al., 2011], the opposite of which appears to be the case for both latitudes in this study. Returning to Figure 8, the solid line shows the linear regression on these jet centers and the dotted hyperbolae above and below show the 95% confidence limits following the method of Working and Hotelling [1929]. The 2 × 1σ uncertainties of 1.5 km decade−1 for 78°N and 0.8 km decade−1 are both greater than the slope of the regression line (again, using the least-squares method), 1.5 km decade−1 for 78°N and 1.3 km decade−1 for 70°N, indicating the trend failing to be significantly nonzero at 95% according to the Tiao et al. [1990] criterion. On the other hand, the same change is observed at both locations. There are obviously too few data to investigate whether the process conforms to AR(1) or contains longer-term memory, or indeed, whether the process is Gaussian and therefore whether or not the Tiao et al. [1990] criterion is truly applicable. However, we take this approach a little further bearing in mind the above assumptions. After removing the 1.5 km decade−1 from the data, for 78°N, we find the standard deviation of the residual to be 1.5 km and the autocorrelation at lag 1 (1 year) to be −0.34, the latter indicating little or no short-term memory; Weatherhead et al. [1998] would then predict that only 8 years of data are required for 90% confidence in our derived trend of 1.5 km decade−1. The 95% confidence interval for this 8 year requirement is ~6–10 years but, again, under the assumption of an AR(1) process. For 70°N, the requirement is only for 5.4 years with a 95% confidence interval of only 3.7–7.8 years, possibly due to the jet center being better defined due to the stronger westward winds at the lower latitude as seen in the figure. In fact, the above statistical analyses combined with a suggestion of an oscillation in the jet center altitude (since the first lags of the autocorrelation functions are slightly negative) indicates that the stochastic component of the time series presumably cannot be represented by a Gaussian process and that an analysis akin to that of Hall et al. [2011] should be applied as an extension to that of Weatherhead et al. [1998]. Again, though, a different/more sophisticated approach to jet center identification and/or more data could well eliminate the apparent oscillation. Therefore, although interesting and giving some quantification to the time series required to detect any possible change in jet height over the last decade, the data length requirement for confidence in any trend should be treated as an exercise only, pending more data.

Figure 8.

(top) Svalbard, 78°N. (bottom) Ramfjordmoen, 70°N. Zonal wind, positive eastward, as a function of time and height. The summer minima are the mesospheric summer jet, the centers of which are indicated by asterisks (see text). The solid line shows the linear regression on the jet centers and the dotted hyperbolae above and below show the 95% confidence limits.

[13] Again, it is important to check the implications of our data rejection method. In Figure 9, we show the derived trend obtained in jet center altitude at 70°N as a function of different degrees (standard deviations) of data rejection—three standard deviations being the maximum allowed wind amplitude for the derivation of Figure 8. Vertical lines in the figure indicate uncertainties in the trend line fits. We see that at ~80 km altitude, the approximate height of the jet center, a limit of 5σ, corresponds to a magnitude of ~150 ms−1 (the 70°N version of Figure 1 is not shown here but is very similar). Even allowing such wind speeds in our analyses and, indeed, independent of the degree of data rejection, there is still an indication of a weak decadal increase in jet altitude, despite the changes being statistically nonsignificant as seen by the vertical uncertainty bars in the figure. Finally, recalling the preceding discussion of the statistical nature of the stochastic component of the time series—both for winds and jet altitude—it must be stressed that significances of decadal changes and confidence limits should be considered provisional; dedicated studies of complex signatures akin to those performed for the ionosphere [Hall et al., 2011] and solar forcing [e.g., Rypdal and Rypdal, 2010] should perhaps be considered the norm in future studies. In addition, especially when the time series becomes longer, it will be appropriate to apply a trend break analysis [e.g., Tome and Miranda, 2005] akin to the approach of Jacobi et al. [2012].

Figure 9.

Ramfjordmoen, 70°N. Derived trends, together with uncertainties, of jet altitude as a function of data rejection. The independent variable is the number of standard deviations (absolute values of which are found from analogy to Figure 1 but for 70°N) within which data are assumed to be reliable.

3 Comparison With Other Results

[14] Elsewhere, where time series of atmospheric or ionospheric parameters have been examined for possible trends, it has been common practice to perform a regression on a metric of solar activity. The resulting dependence (if any) is then subtracted from the original time series to arrive at a residual; if the residual then exhibits a trend, this can be attributed to anthropogenic driving. A classic example of this approach is the study of ionospheric parameters by Ulich and Turunen [1997] who chose the F10.7 flux [Covington, 1948] as a suitable indicator of solar activity. As stated earlier, we refrain from attempting to identify anthropogenic forcing in this study. In Figure 10, however, we compare, qualitatively, the TSI (via with the 78°N-derived jet center altitudes and the 80 km summer zonal winds. Using a parameter such as the F10.7 flux, one makes an a priori assumption that UV is forcing the observable. This is reasonable when coupling to stratospheric temperatures is known or when ionization by UV is a mechanism directly affecting the observable. For time series, predating the start of F10.7 monitoring, it is possible to use sunspot number. We have, however, chosen TSI [Kopp and Lean, 2011] because this does not involve a priori assumptions about driving (viz. ozone heating), is measured outside the atmosphere, and furthermore, is not predated by our relatively short dataset. Examination of Figure 10 fails to give any impression whatsoever that there is a solar cycle dependence (parameterized by TSI) in the jet altitude or approximately corresponding to the zonal wind strength. This negative result is different to that of Keuer et al. [2007] who identified a negative dependence of summer zonal wind on Lyman-α (similar to the F10.7 flux) and a weaker positive dependence for winter at 55°N. However, this could be explained by the much larger solar zenith angles for Svalbard and that data by Keuer et al. [2007] cover the decline of solar cycle 22 and almost all of solar cycle 23. Jacobi et al. [2012] identified a breakpoint in 90 km winds in 1998: a little over half the data analyzed by Keuer et al. [2007] are from before this date, whereas all our data are from after.

Figure 10.

Svalbard, 78°N. Comparison between (top) total solar irradiance (TSI), (middle) jet center altitude, and (bottom) zonal wind; in this case, we selected 80 km summer values. There is no convincing indication of solar cycle dependence in the mesospheric dynamics.

[15] As stated earlier, wind data for the same geographic locations as in this study, and particularly long time series for 78°N, are sparse. Indeed, the only data from the proximity of Svalbard are from the in situ measurements from Heiss Island (81°N, 58°E) ~750 km east of NSMR [Portnyagin et al., 1993]. The data are from the period 1968–1985 and, again, preceding the breakpoint identified by Jacobi et al. [2012]. Observations are reported for 95 km altitude only. It is difficult to identify any trends in meridional winds: there is considerable variability in winter values, and summer data would require a careful breakpoint analysis, but there is a qualitative suggestion of increasing poleward flow during the equinoxes. For the zonal winds, on the other hand, there are indications of a weakening of the eastward flow in winter by ~5 ms−1 decade−1 and a weakening of the westward flow in summer by ~2–3 ms−1 decade−1. While keeping in mind the rather different periods of observation, from our Figure 4 (top row) above 95 km, we see the trends are almost exactly the same.

[16] For middle latitudes, rather, more studies using contemporary observations are available for comparison. Jacobi et al. [2003], in addition to modeling the gravity wave flux corresponding to different CO2 and O3 scenarios and the subsequent accelerations to lower and middle atmosphere wind fields, also reported change in the summer meridional wind at 95 km altitude over Collm (52°N, 15°E) between 1979 and 2001 [Jacobi et al., 2003, Figure 4]. The indication is of a 2.5 ms−1 decade−1 decrease in the equatorward flow. This may be compared to our Figure 7 (bottom left), i.e., for 70°N; we observe an ~1–2 ms−1 decade−1 increasing poleward flow at the same altitude, although our observation lacks significance. At 78°N, however, we observe an ~3 ms−1 decade−1 increasing equatorward flow. As we also see from Jacobi et al. [2003], poleward of 50°N, 95 km approximates to the transition height between poleward and equatorward flows, suggesting a latitudinal variation of the transition height that could be quantified by including more observations from more latitudes. Subsequently, Jacobi et al. [2012] performed breakpoint analyses for zonal and meridional components of the wind at nominally 90 km for three different latitudes. The Jacobi et al. [2003] summer meridional Collm data are reproduced and extended, and interestingly, they now exhibit a breakpoint in 1998 followed by a decrease in poleward flow of 4 ms−1 between 1998 and 2008, thus, in good agreement with our results for 78°N. A breakpoint in the zonal component in 2004 for both summer and winter renders comparisons with our results difficult, as do our large uncertainties for the meridional wind changes during winter. Portnyagin et al. [2006] also reproduced these data but only up to 2004. For 55°N and between 1990 and 2005, Keuer et al. [2007] (in particular, [Keuer et al., 2007, Figure 13]) presented trends in both wind components as functions of season and altitude. For the zonal component, the summer is characterized by intensification, by 10 ms−1 decade−1, of the westward flow at 75 and 80 km, i.e., the jet, in reasonable agreement with Figures 4 and 7 (top left panels). The reversal in trend at 55°N between 80 and 85 km is not evident in our results at 70°N but does appear at 93 km at 78°N. The poleward flow at 55°N is seen to be increasing at a rate ~3 ms−1 decade−1 during summer at 75 and 80 km and weakening by a similar amount as previously mentioned. This corresponds exactly with our result for 70°N at 80 and 85 km, but the same change is not seen at 78°N. While Offermann et al. [2011] primarily addressed the vertical structure in 2 day waves, a comparison of mean zonal winds for the periods 1990–1995 and 1997–2002 at 55°N and 73 km altitude is given. This comparison corresponds to a change of −7 ms−1 between 1992 and 1999 for summer. In turn, this represents a summer increase in the westward jet of 10 ms−1 decade−1. This summer result may be compared with our Figure 4 (top left), viz. 10 ms−1 decade−1 at 78 km and Figure 7 (top left) ~4 ms−1 decade−1 at 80 km. If one then takes into account that the vortex occurs at somewhat lower altitude at lower latitude (again, 5 km according to Keuer et al. [2007]), we arrive at a good agreement between our findings for both 70°N and 78°N (our Figures 7 and 4, respectively) and those of other works [Keuer et al., 2007, Figure 13; Offermann et al., 2011, Figure 14];.

[17] At low latitude, Sridharan et al. [2007, 2010] reported that between 1993 and 2007 and with a height regime of 80–98 km, zonal winds exhibit a strengthening of 5 ms−1 decade−1 in spring and a weakening of 5–10 ms−1 decade−1 during autumn and winter. Corresponding changes in the meridional component are slight. Interestingly, the figure of Sridharan et al. [2010, Figure 4d] (mean zonal trend altitude profile) is rather similar to our Figure 4 zonal results but with smaller magnitudes; however, considering the extreme latitude difference, it would be unwise to read too much into this. Even further south, over Antarctica at ~90 km, the winter zonal prevailing wind has decreased by the order of 2 ms−1 decade−1 between 1970 and 2006 [Merzlyakov et al., 2009], which compares well with our value of ~3 ms−1 decade−1 at 78°N. The summer case, on the other hand, exhibits a breakpoint in the early 1990s; between 1995 and 2005 (i.e. the portion best corresponding to our observations) the westward flow decreased by 4.7 ms−1 decade−1, the opposite of our result for summer. Perhaps by coincidence, Portnyagin et al. [2006] also identified a 1990 breakpoint but in the Northern Hemisphere meridional component.

[18] We see from the aforementioned comparisons that the relatively nearby site of Heiss Island [Portnyagin et al., 1993] has demonstrated trends similar to the 2001–2012 changes we find. Although encouraging, the results of Portnyagin et al. [1993] are from a rather earlier time interval, and other authors have identified breakpoints in time series spanning both data set, but all at lower latitude. Even so, the magnitudes of the changes we report are not dissimilar from many other findings. Regarding the mesospheric jet, aspects of the studies by Offermann et al. [2011] and Keuer et al. [2007] corroborate our findings (that the jet may be increasing in altitude) to some degree.

4 Conclusions

[19] Examination of mesospheric winds over Svalbard (78°N, 16°E) during the interval 2001–2012 and the Norwegian mainland (70°N, 19°E) during 2004–2012 reveals an intensification of the westward zonal wind between ~80 and ~90 km in summer (the polar summer mesospheric jet). For the meridional corresponding component, there is little evidence for change at 78°N, but at 70°N, the associated equatorward flow is gradually decreasing. Intensification of the jet can be interpreted as the zonal wind being allowed closer to the radiation balance due to reduced acceleration caused by diminishing deposition of horizontal momentum, in turn due to changes in gravity wave (GW) filtering in the underlying middle atmosphere. In winter, there is a retardation of the eastward flow above 95 km which can be explained by increasing GW drag. Trends significant at the 95% level cannot be detected in the meridional component in winter at either 70°N or 78°N, yet there is consistency between adjacent heights and between the two locations. This gives us a cause to believe that there may be a slight increase in poleward flow below 85 km and a slight decrease above 90 km in winter. Below 85 km, this is not consistent with the expectation because convergence on the pole should correspond to an increasing eastward circulation which we do not see. For 70°N, 22°E where the surface topology and therefore orographic forcing are qualitatively similar to 78°N, Hall et al. [2008] reported positive trends in the upper mesosphere/lower thermosphere turbulence compatible with increasing GW flux from below, supporting our change in GW-forcing hypothesis.

[20] Analyses in which trends are obtained for different seasons, such as those presented here, are furthermore demonstrated to be highly sensitive to inclusion of stratospheric warmings. Very different results are obtained if January is included as a winter month. This prompts us to ask whether “climatic change” should include SSWs (or similar phenomena) or not and whether, perhaps, as suggested by comparing Figures 4 and 6, SSWs exhibit a secular change in either intensity or frequency or both that should be treated separately. The relationship between global change in the upper atmosphere and how it is influenced by, for example, ozone recovery and the underlying atmosphere is, to say the least, complicated, as described by, for example, Laštovička, et al. [2008]. Modeling by Jacobi et al. [2003] has addressed this question to some extent by examining the responses of the wind components to changes in GW momentum deposition. The sense of GW induced acceleration caused by reduced ozone absorption combined with reduced CO2 cooling implies a weakening of the mesospheric jet (at least at 52°N); that this is the opposite of our observational finding may simply be related to latitude but could also be attributable to ozone recovery during the last decade [e.g., Zeng et al., 2010; World Meteorological Organization, 2007]. Whether or not the decade change we report here is of anthropogenic origin will not be addressed further in this study, but it is interesting to note that the expected difference in zonal wind between solar maximum and minimum, irrespective of latitude, at around 70 km does not exceed 4 ms−1 according to Gray et al. [2010]—similar to the change we observed 10 km higher up at 70°N but over one decade (i.e., approximately one solar cycle) and only half of the change we find at 78°N. The results reported by Gray et al. [2010] are only a rough indication since they are based on data up to 65 km and until 2001, and moreover, observations spanning only one decade do not form a solid base for inferences on solar influence. On the other hand, a qualitative comparison of total solar irradiance with wind characteristics during summer, when forcing would be expected to be greatest at 78°N fails to indicate any obvious causality. While the summer results we show here are in agreement with the scenario drawn by Keuer et al. [2007], who also reported a strengthening of the mesospheric jet, and comparable with the westward cell duration analysis of Offermann et al. [2010], they represent the first observations at such high latitude in the Scandinavian sector, the nearest otherwise being the early results from Heiss Island at 81°N, 58°E [Portnyagin et al. 1993]. The observations, over one decade, reported in this study (albeit for only two high-latitude geographical locations) are consistent with a scenario of progressive strengthening of the Brewer-Dobson circulation [Garcia and Randel, 2008] and therefore redistribution of pollutants and ozone between high and lower latitudes, affected by the dynamics of the polar vortex—an important information attainable by future evolution of such time series combined with increasing geographical coverage


[21] The authors wish to express their appreciation to T. Aso for the instigation of the two meteor radar systems that contributed data to this study.