The radar tropopause above Svalbard 2008–2012: Characteristics at various timescales

Authors


Abstract

[1] Determinations of the tropopause altitude over Adventdalen, Svalbard (78°N, 16°E) have been assembled for the period 2008–2012. These reveal characteristics at various timescales: interannual, seasonal, and stochastic. First, it is established that the inclusion of considerably more data does not alter earlier findings of good temporal agreement with radiosonde measurements, that the radar tropopause is located at a fairly constant altitude above the meteorological tropopause, and that the seasonal variation lags the surface temperature variation by approximately 1 month, indicating a degree of top-down control from the stratosphere. Over the longest timescale available, i.e., 5 years, we find an increase in tropopause altitude and identify this with increasing solar flux during the growth phase of solar cycle 24, and contrary to the seasonal variation, there is evidence that this is controlled by troposphere warming. At stochastic timescales, the scaling characteristics of the signal are examined, and in an exploratory investigation, the Hurst exponent is found to be ~0.82, which is in excellent agreement with independent findings for high latitude surface temperature data and therefore supportive of a complexity matching approach to identification of causal relationships between atmospheric metrics.

1 Introduction

[2] Increasing greenhouse gas (GHG) concentrations have been identified as the cause for a globally averaged warming of the troposphere in recent decades [Solomon et al., 2007]. Although it might be debated whether action to reduce chlorofluorocarbon (CFC) emission is the cause, there is now evidence for a recovery of ozone concentration [World Meteorological Organization, 2006; Zeng et al., 2010; Hu et al., 2010]. More recently, public attention has been brought to increasing nitrogen oxides' (NOx) quantities, although this has been known to the scientific community for some years [e.g., Hauglustaine et al., 1994]. NOx plays a role in the production of tropospheric ozone, but at the same time, other GHGs may be reduced [e.g., Hauglustaine and Koffi, 2012]. The interplay between various anthropogenic emissions, not to mention the chemistry between them, and the rest of the atmosphere's thermal and dynamic structure and aeronomy is becoming increasingly more difficult to understand as more effects are revealed. Furthermore, these processes are compounded by solar fluxes which vary at many timescales but may or may not be regarded as constant at climatic timescales. Bringing many different types of observation to bear on the atmospheric column as a whole should hopefully enhance our understanding of processes affecting climate change. The tropopause, the demarcation between the troposphere below and the stratosphere above, is generally thought of as the first minimum in temperature gradient disregarding any local inversion layers near the Earth's surface and occurs typically between 8 and 18 km altitude depending on latitude [e.g., Salby, 1996; Brasseur and Solomon, 2005]. The troposphere is roughly characterized by a constant temperature lapse rate, whereas in the stratosphere, temperature increases with altitude due to absorption of solar UV-radiation by ozone [e.g., Brasseur and Solomon, 2005]. The temperature minimum—the “cold-point”—is therefore formed at the intersection of these two profiles. If the temperature of the troposphere increases, the intersection will be displaced upward, provided the stratospheric temperature profile remains unchanged. Similarly, if the tropopause temperature profile is fixed, then an increase in stratospheric temperature will displace the intersection downward. Such stratospheric temperature increases can be caused by increasing ozone concentrations (i.e., increasing the capacity for absorption of solar UV radiation), by increases in solar UV flux, or a combination of these. Characterized by a constant temperature lapse rate, the troposphere temperature profile is thus in general directly related to the surface temperature (again disregarding any local effects such as inversions): a higher surface temperature can be expected to be indicative of a higher tropopause and vice versa. In particular, greenhouse warming and thus expansion of the troposphere are expected to displace the tropopause upward; at the same time, cooling of the middle atmosphere by the very same GHGs [Roble and Dickinson, 1989; Rishbeth and Roble, 1992; Rishbeth and Clilverd, 1999] would also serve to displace the tropopause upward which suggests the tropopause altitude could well be a highly sensitive metric for climate change. With this motivation, we attempt to assist understanding of tropopause altitude at high latitude by presenting observations from the last 5 years from the Scandinavian sector Arctic at 78°N, 16°E, together with tentative inferences on controlling processes.

[3] The SOUSY Svalbard radar, located in Adventdalen at (78°N, 16°E) has been operating since 1999. Originally, the system operated at high power with a steerable beam but could only be used when personnel were present and therefore on an observation-campaign basis. In later years, changes in political and scientific foci dictated that it was preferable to operate unattended and with a simpler, cheaper, and more reliable system. This solution is described by Hall et al. [2009] and essentially consists of a single vertical beam produced by an array of 356 4-element Yagi antennas, with an operating frequency of 53.5MHz and only 1kW peak power. As a consequence of the simplification of the system, we currently have at our disposal 5 years of continuous troposphere and mesosphere measurements: these two height regions are sounded alternately, 10 minutes at a time and otherwise 24/7. Temporary system problems have interrupted operations during the 5 year (hitherto, viz. 2008–2012) period, but generally, good continuity has been achieved.

[4] In this study, we build upon the method of tropopause altitude determination described in detail by Hall et al. [2009] and the subsequent application of the results by Hall et al. [2011]. In the latter paper, monthly climatologies were determined which were then validated by and compared with independent data including radiosonde measurements of the meteorological tropopause [World Meteorological Organization, 1996] and various surface air temperature time series. We shall not reproduce the method of determination of the radar tropopause here, suffice to say that it manifests itself as a layer of enhanced echoes arising from a local maximum in Brunt-Väisälä frequency [e.g., Gage and Green, 1979; Röttger and Hall 2007]. Due to the low power of the SOUSY radar, the maximum in echo power is used to identify the radar tropopause rather than the vertical gradient in power beneath the peak [again, Gage and Green, 1979]. Even so, on occasion, this layer is poorly defined, and such cases are treated as missing data, but otherwise, tropopause altitude is determined every hour. This 1-hour resolution dataset is thus the starting point here. Subsequently, we employ different smoothing schemes to facilitate identification of features predominant at different timescales. The basic data set is shown in Figure 1: daily (dots) and monthly (thick line) values are approximated by applying appropriate Lee filters [Lee, 1986] to the underlying hourly values (which in themselves are intermittent and noisy since the tropopause identification algorithm, although robust and objective occasionally fails to find a tropopause and/or detects some spurious echo/noise spike instead). Prima facie, one can see a chaotic nature at short timescales, a seasonal variation with spring minima and summer maxima, and also a suggestion of increasing overall altitude during the 5-year period. In the following sections, shall address these timescales in detail. First, we shall examine the seasonal variation and revisit the comparison with independent measurements performed by Hall et al. [2011]; here, we shall primarily check that the inclusion of an extra 2 years of data has not altered the conclusions regarding top-down and bottom-up influences on the tropopause position. Next, armed with the seasonal climatology, we examine the superimposed change in altitude over the entire observational period. Third, we examine the signal normally removed from such time series, namely, the noise, or as we prefer to call it here, the stochastic component; this is analysed in terms of complexity. The signature we identify can then be compared with similar studies of other atmospheric metrics as an exercise in complexity matching as a tool for searching for linking and causality in atmospheric processes.

Figure 1.

Tropopause heights (radar tropopause) since November 2007, as determined by SOUSY. Daily values approximated by applying a 1-day Lee filter are indicated by black dots; the thick solid line shows result of applying a 30-day Lee-filter.

2 Monthly Climatology

[5] The investigation by Hall et al. [2011] entailed comparing tropopause altitudes assimilated over the first 3 years of operation with corresponding meteorological tropopause heights obtained by radiosonde soundings from Ny-Ålesund approximately 120 km to the NW. Given the anticipated height offset, the results were in excellent agreement with the temporal variation matching almost exactly. The agreement is demonstrated in Figure 2 (adapted from the study by Hall et al. [2011]). The upper panel shows the almost constant offset but with a weak seasonal dependence well within the 1-sigma variability of the radar measurement. The lower panel shows the result of a linear regression of the radar measurement on the radiosonde one, together with 95% confidence intervals [Working and Hotelling, 1929], which accentuate a high degree of linearity. By averaging monthly values (i.e., all January values, all February etc.) in the same way as previously, we arrive at a monthly climatology shown in the top panel of Figure 3. This figure is almost identical to that in Hall et al. [2011] (intentionally). More values are used to determine each point in the radar tropopause climatology, but otherwise, the conclusion remains unchanged: the seasonal variation of both observations is still the same and furthermore indicates a degree of inadequacy in the somewhat dated semi-empirical prediction of World Meteorological Organization [1996], the latter predicting the minimum height to occur as late as May. The panel also includes a 3-month smoothing, this extending to the edges of the plot because a wrap-around is used to ensure continuity from December to January. In the middle and bottom panels of Figure 3, we reproduce the ozone column density (above Ny-Ålesund) [Vogler et al., 2006] and surface air temperature data (at the nearby Kjell Henriksen Observatory (KHO) approximately 2km SE of the radar at an altitude of 520m on Breinosa) previously given in the study by Hall et al. [2011] to illustrate how the spring tropopause altitude minimum coincides with the maximum in ozone concentration rather than the local temperature minimum. Furthermore, the temperature maximum occurs (although marginally) in July, whereas there is a marked tropopause maximum height in August. The current results support the earlier inference, i.e.., of Hall et al. [2011], that there is a top-down driving of tropopause position: the stratosphere and troposphere responses to seasonal variation in insolation compete with each other.

Figure 2.

Top: difference between tropopause determined by SOUSY and by radiosonde (vertical lines indicate 1-sigma variability in radar measurements). Bottom: regression of SOUSY observations on radiosonde observations (data points are shown by +'s and 95% confidence limits are indicated by dashed hyperbolae). Adapted from Hall et al. [2011].

Figure 3.

Top: monthly climatologies of tropopause altitude obtained by SOUSY since November 2007 (solid lines: monthly values with standard deviations and 3 monthly smoothing). The lower dashed line shows the corresponding radiosonde measurement. Middle: total column ozone monthly climatology. Bottom: surface air temperature monthly climatology.

3 Change Over Entire Observation Period to Date

[6] We shall now further investigate the prima facie increase in tropopause altitude perceivable in the 5 years of data shown in Figure 1. Given the pronounced seasonal variation, a natural approach would be to separate out different seasons and look for changes over the five available years in each one. However, this would mean relatively large uncertainties and inconclusive results—it will be better to wait until rather more years of data have been obtained. Instead, therefore, we subtract the seasonal variability determined in the previous section from a time series of monthly means. The residual time series represents month-to-month variation but with the intra-annual deterministic component removed, this following the guidelines of, for example, Weatherhead et al. [2002]. The residual is shown in Figure 4. A least squares linear fit is then performed on these points resulting in the thick line in the figure. Dashed hyperbolae either side indicate the 95% confidence limits for the fit, using the method of Working and Hotelling [1929]. The slope of the regression corresponds to 0.75 ± 0.25 km decade-1, but we must be careful not to refer to this as a trend. Since the 2-σ uncertainty (i.e., 0.5 km decade-1 is less than the absolute slope, the finding is significant to 90% following Tiao et al. [1990].

Figure 4.

Tropopause height monthly means following subtraction of seasonal climatology. The linear least-squares fit (solid line) is superimposed together with 95% confidence intervals (dashed hyperbolae).

[7] The 5 years of results presented here coincide with the growth phase of solar cycle 24 and are influenced by increasing solar fluxes, which can be parameterized by total solar irradiance (TSI) and UV. The UV flux is measured at the surface as the “f10.7” flux [Covington, 1948]. TSI and f10.7 data can be obtained via http://www.ngdc.noaa.gov/stp/solar/solaruv.html and http://omniweb.gsfc.nasa.gov/html/ow_data.html, respectively. If several solar cycles of data were available as is the case for surface air temperature and certain ionospheric parameters [e.g., Ulich and Turunen, 1997; Hall et al. 2007], it would be a simple matter to apply a filter to remove seasonal variation and arrive at a possible trend, which could then be subsequently investigated for possible anthropogenic forcing. This is obviously not the case here, but even so, one can investigate the relationship between solar input and tropopause height and the degree to which this is responsible for the change over the 5 years. A fundamental difference between the two indicators of solar flux, f10.7 and TSI, is that the former is measured at the Earth's surface and the latter by satellite. Comparing with f10.7 first, we arrive at the scatter plot in the upper panel of Figure 5. A linear regression of tropopause height on f10.7 (thick line) indicates a dependence of 5 ± 1 m sfu-1. Again, the 95% confidence limits are indicated by the dashed hyperbolae. A tropopause height time series corresponding to the original data is then constructed from the f10.7 data and using the coefficients of the linear regression. This is now considered as the tropopause response to the solar UV flux. A residual tropopause height time series is then obtained by subtracting the response to f10.7 from the original, as now shown in the bottom panel of the figure. Clearly, the 5-year change seen in the original data has all but disappeared in the residual; the uncertainty is over three times larger than the slope itself, and therefore, the slope is insignificantly nonzero.

Figure 5.

Upper panel: dependence of tropopause height on f10.7 (UV) flux. +s indicate individual monthly data points from the 5 year observation period; the linear regression is shown by the thick line, and the dashed hyperbolae indicate the 95% confidence limits. Lower panel: residual after subtracting the upper panel dependence from the original tropopause height time series; the linear regression is shown by the thick line and the dashed hyperbolae indicate the 95% confidence limits.

[8] Initially, it may appear that the positive dependence of tropopause height on f10.7 is a paradox because it is the very UV that is absorbed by O3 in the stratosphere resulting in a heating that causes the inversion of the vertical temperature gradient and defining the tropopause. If absorption by O3 and subsequent heating of the stratosphere was dominant, then we would observe the opposite dependence in the upper panel of Figure 5. However, f10.7 is representative of short wave radiation arriving at the Earth's surface causing subsequent heating of the surface and re-emission in the IR and then warming of the troposphere. Thus, over solar cycle timescales, we can infer that increasing short-wave radiation, indirectly leading to increasing tropospheric temperature is the dominant mechanism for positioning the tropopause.

[9] We further investigate this by making a parallel comparison with TSI in Figure 6. As in Figure 5, linear regressions are shown by thick lines accompanied by 95% confidence limits as dashed hyperbolae. Again, in the upper panel, we see a positive dependence of tropopause height on TSI: 385 ± 56 m (Wm-2)-1 and a significant dependence. Akin to the treatment for f10.7 dependence, we subtract this response from the original tropopause height time series to arrive at the residuals in the lower panel. As before, the 5-year change now disappears, this time, the uncertainty amounting to an order of magnitude more that the slope, the latter thus being insignificantly nonzero. TSI integrates all wavelengths but is determined outside the atmosphere; even so it is representative of the radiation arriving at the surface (of which, f10.7 is a part) and so as expected, an increase imparts a corresponding increase in tropopause height which in turn supports our interpretation of the positive dependence on f10.7.

Figure 6.

Upper panel: dependence of tropopause height on total solar irradiance. +s indicate individual monthly data points from the 5 year observation period; the linear regression is shown by the thick line, and the dashed hyperbolae indicate the 95% confidence limits. Lower panel: residual after subtracting the upper panel dependence from the original tropopause height time series; the linear regression is shown by the thick line, and the dashed hyperbolae indicate the 95% confidence limits.

[10] As a final check on this response, the same procedures as for Figures 4 and 6 have been performed on radiosonde data. These results are shown in Figure 7. In the top panel, the monthly average tropopause altitudes obtained from the radiosonde soundings are shown, together with the corresponding radar determinations presented earlier in this study, for comparison. As already established, the temporal agreement is good, and there is an anticipated height offset essentially due to the difference in definitions of cold-point and radar tropopauses. Recall that the radar measures all day, but the situ measurements are at fixed times during the day, and that the underlying daily measurements are therefore not necessarily representative of daily means. As was done for Figure 4, the next panel shows the deseasonalized (residual) tropopause heights from the radiosonde soundings. The trend over the 5 year period is 0.3 km decade-1 but from the uncertainty of 0.3 km decade-1 we see that this is not significantly non-zero. Moreover, this trend, if it is to be believed is only half that detected by the radar. More data and in-depth analysis (outside the scope of this study) would be required to ascertain the cause of this difference, but it is not unreasonable to hypothesize that the mechanisms controlling the radar echo, up to 2 km above the cold-point, and the cold-point itself vary differently with changing solar flux. In some respects, at such high latitude, summer and winter data correspond to day and night measurements, respectively, and therefore with the availability of more data, it would be interesting to investigate trends for summer and winter individually. The bottom two panels of Figure 7 correspond to the analysis shown in Figure 6: the dependence of the radiosonde-determined tropopause height on TSI followed by the trend obtained after removal of this dependence. Again, the dependence on TSI is only half that found for the radar tropopause height (in agreement with the hypothesis that the two tropopause measurements or rather definitions respond slightly differently to solar forcing). The final result is not surprisingly, that the radiosonde tropopause fails to exhibit any trend over the 5 year period after the solar forcing is subtracted—exactly as the corresponding finding for the radar measurements.

Figure 7.

Top panel: Tropopause altitudes determined by radiosonde from Ny-Ålesund (solid line) with the radar measurements reported elsewhere here reproduced for comparison. Upper middle panel: results corresponding to Figure 4, but for the radiosonde measurements, i.e., deseasonalized tropopause height variation together with trendline and 95% confidence limits. Bottom two panels: results corresponding to Figure 6, but again for radiosonde measurements, i.e., dependence of the height residual on total solar irradiance, and the new residual dataset with the dependence removed and with derived trend and confidence limits.

4 Short Timescale Characteristics

[11] As a final analysis approach, we shall examine the statistical characteristics of the short timescale variability of the dataset. The following analysis represents an exploratory investigation into the stochastic nature of the tropopause, and the main focus is to illustrate the potential for identifying, for example, causality and linking between atmospheric processes by complexity matching. Subsequently, one should consider more dedicated studies on nonlinearity in tropopause data. Such possibilities are reviewed by Suteanu and Mandea [2012], and the reader is referred to this work for a plethora of references. Specific comparison to the results of Suteanu and Mandea [2012] will indeed be made later in this section.

[12] It was explained earlier that the tropopause altitude time series can be somewhat intermittent. This is because, if echo power or signal-to-noise ratio (SNR) is low, the detection algorithm may fail, either as a consequence of a weak tropopause for meteorological reasons or inadequate radar power for the task. Other reasons can be purely meteorological, for example, frontal passages [e.g., May et al., 1991] or tropopause folding [e.g., Nastrom et al., 1989, Sprenger et al., 2003] or quite simply occasional failure of the radar for some reason. We employ the underlying hourly data on the assumption that any data gaps are most likely radar problems or failure of the tropopause identification algorithm, as opposed to intermittence in the presence of the tropopause itself. First, the hourly data are smoothed by applying a running median filter to eliminate all periodicity under 1 month; we have already examined the seasonal variation and can qualitatively see month-to-month determinism in the data, for example, the weak secondary peak in tropopause altitude in mid-winter. The resulting smoothed time series, which also includes any trend, is then subtracted from the original, both having 1-hour resolution, to arrive at a residual dataset containing submonthly variability, shown in the top panel of Figure 8. As stated above, this investigation into complexity and nonlinearity of tropopause data is exploratory. For convenience, however, we shall refer to these residuals as the stochastic component of the tropopause altitude time series, although more in-depth analyses would be needed to ascertain the signal is truly stochastic. Indeed, from the upper panel of Figure 8, one might suppose that there is still a periodicity in the data, manifested in the lowest tropopause detections. Several attempts were made to remove this (the aforementioned median filter giving best results), but with no success, and the bottom panel of Figure 8, in which higher densities of data points are indicated by more intense shading, provides the answer to the problem. There is no seasonal variation remaining in the data concentrated around zero, i.e., the majority of measurements; however, there are outlying patches corresponding to quasi-seasonally varying larger fluctuations. Such bursts of high variability occurring spasmodically are typical of non-Gaussian distributions. Characteristics of and analysis strategies for, such time series are described by Lennartz and Bunde [2009]. Different approaches are possible including the detrended fluctuation analysis (DFA) genre and spectral density function (SDF) methods, comparisons of which can be found in Heneghan and McDarby [2000]. Having read both Lennartz and Bunde [2009] and Heneghan and McDarby [2000], however, the reader should be aware of nuances in nomenclature: the Hurst exponent, α, is referred to in the former paper, and in both papers, α is the slope of the SDF in log-log space. The exponent α is related to the Hurst parameter or coefficient, H. In other words, the Hurst exponent referred to by Lennartz and Bunde [2009] is not the same as the Hurst parameter referred to by Heneghan and McDarby [2000]. Furthermore, Suteanu and Mandea [2012] derive H, the Hurst parameter. Therefore, we assume that the residuals shown in Figure 8 are a good representation of the stochastic component of the tropopause altitude and derive the power spectral density of the signal as a function of scale (the basic unit of which being 1 hour).

Figure 8.

Representations of the 1-hour time resolution residual data after removal of seasonal variation. Top panel: data points. Bottom panel: smoothed representation, such that data point density, is indicated by more intense shading. Comparison of the panels reveals that the apparent periodicity in the lower tropopause altitude determinations is actually due to periods of increased variance. Reduced scatter toward the end of the observation period is due to a problem with the radar, but does not change the end result.

[13] Since this is an exploratory exercise in complexity analysis of such data, we shall employ the SDF method rather than the somewhat nonintuitive DFA approach, the latter requiring a degree of a priori knowledge of the signal to select an optimal detrending polynomial and window limits. It is easier to assess the quality of the result when using SDF because the power-law spectrum is more familiar (to the physicist). The SDF analysis is shown in Figure 9. Qualitatively a power-law dependence can be seen at all, but the shortest scales α is found (using the whole spectrum) to be 0.64, giving H = 0.813. Strangely, it does not appear normal practice in mathematics to quote uncertainties for derivations of H, but this may be a consequence of more popular use of DFA methods. Here, however, while the standard method yields H = 0.813, varying the frequency limits for derivation of the slope indicates H = 0.818 ± 0.04. This corresponds to α = 0.64 ± 0.08. From Heneghan and McDarby [2000], we learn that for a shallow dependence defined by α < 1, H is related to α by α = 2H-1. The signature of the stochastic process is one of short autocorrelation (since the spectral dependence is shallow) and instead, rapid switching between adjacent data points and indicative of fractional Gaussian noise—a so-called “fGn” process. To further test the confidence in the result, we follow the approach prescribed by Theiler et al. [1992]. Surrogate data sets (100 in this analysis) are constructed from the stochastic component of the tropopause observations using the amplitude adjusted Fourier transform algorithm (AAFT) method, also explained by Theiler et al. [1992]; these data sets then have the same Fourier spectrum as the original but with randomly shuffled phases. The same statistic, H, is then determined for each surrogate and the mean, μH, and standard deviation, σH, of the surrogates' statistics are determined. The significance in units of standard deviation is given by |H- μH| / σH and is found to be 4.7—in other words, well distinct from the uncertainty in μH, supporting our hypothesis that the stochastic component of tropopause altitude is indeed a nonlinear process.

Figure 9.

Spectral density function (SDF) as a function of scale in log-log space. The fitted dependence is shown by the straight line giving a Hurst coefficient of 0.813.

5 Discussion and Conclusions

[14] At intra-annual timescales, this study has confirmed the observations by Hall et al. [2007]. First, the tropopause observed by radar lies at a virtually constant 2 km above that measured by radiosondes as explained by Gage and Green [1979]; here, we identify the radar tropopause as the peak in echo power. This importantly means that temporal variations of the radar tropopause can subsequently be compared with corresponding radiosonde determinations with some degree of confidence. Second, the mean seasonal variation is characterized by a minimum in, primarily, of a late-winter/early-spring minimum in April, a maximum in August, and a weak secondary maximum in December-January. Third, while corresponding to the surface temperature variation as intuition would dictate, there is also evidence for influence from the stratosphere since the spring minimum in tropopause altitude corresponds with the maximum in ozone column density rather than the minimum in surface air temperature. These concepts, founded on the basic principles underpinning the formation of the temperature structure of the atmosphere as outlined in the Introduction, are not new [e.g., Santer et al., 2003; Zängel and Hoinka, 2001] but are confirmed here for this particular location in the high latitude Scandinavian sector. Furthermore, this study supplements the growing information on tropopause structure [e.g., Bèque et al., 2010] but in high latitude regions where observations are sparse [e.g., Highwood et al., 2000; Alexander et al., 2012; Suteanu and Mandea, 2012]. The results also agree well with the findings (for 50°N-90°N) of Feng et al. [2012]. Using a more sophisticated radar, Alexander et al. [2012] report tropopause variability, but for the Antarctic and more correctly estimate the radar tropopause height from the maximum in vertical gradient in echo power. This results in their tropopause altitudes being slightly lower than those measured by radiosondes and therefore not in disagreement with the results shown here. Moreover, Alexander et al. [2012] report similar responses of radar and ozone tropopauses to horizontal dynamics, also supporting the scenario of top-down influence via ozone heating.

[15] As mentioned earlier, the tropopause region has attracted attention due to its possible sensitivity to global warming of the troposphere combined with global cooling of the middle atmosphere and therefore as a potentially useful metric for climate change. Currently, the SOUSY dataset is too short for climate studies, but the 5-year time series available allows an investigation of the response of the tropopause during the growth phase of solar cycle 24. In comparing solar fluxes—both the UV arriving at the surface and the total solar irradiance outside the atmosphere—the detected change in tropopause altitude can be attributed wholly to increasing troposphere thickness due to increasing solar heating. It is important, however, to note that this conclusion is purely based on 5 years of observations and in no way contradicts the findings of Feng et al. [2012]. Four decades of radiosonde measurements for four latitude bands have been analysed by Feng et al. [2012] demonstrating positive trends in tropopause altitude, results which demonstrate the value of prolonged tropopause observations. Although the seasonal variation we and others report can be shown to be driven by temperature variation from both troposphere and stratosphere, at interannual timescales, absorption by the surface and subsequent re-emission at longer wavelengths appears to dominate.

[16] This study also explores the potential for identifying signatures in the stochastic component of the tropopause height, on timescales ranging from 1 hour to 1 month. The rationale for this is to compare these signatures with those of possible coupled processes—complexity matching. Complexity matching is somewhat in its infancy, however, and is notoriously prone to misinterpretations. An example of the technique is the study by Rypdal and Rypdal [2011] who find identical multifractal noise signatures in both the auroral electrojet index and the z-component of the interplanetary magnetic field suggesting the existence of a physical mechanism linking intermittency in the two parameters. On the other hand, Scafetta and West [2003] proposing the linking of solar flare intermittency to terrestrial temperature anomalies via a Lévy process has sparked heated discussion in the literature. At the risk of provoking similar reactions, it is interesting to compare the Hurst component H = 0.82 from this study with those obtained by Suteanu and Mandea [2012]. Exactly the same values are reported for surface air temperature time series at high latitude Canadian stations in 2008. The earliest Canadian data presented by Suteanu and Mandea [2012] is pre-1950, and their analysis demonstrates H varies with both latitude and year, but for latitudes closest to that of SOUSY, H converges on values of ~0.82 by 2008.

[17] To conclude, tropopause altitudes are presented for a 5-year period from the end of 2007 to autumn 2012 for 78°N, 16°E, obtained by radar. Over the 5 years, the tropopause is characterized by a significant increase in altitude, and this can be explained by the similarly increasing solar flux corresponding to the growth of solar cycle 24. In addition, radiosonde soundings indicate a similar although weaker response in cold point tropopause height; this corroborates the radar measurements but at the same time indicates differing degrees of solar forcing on the different characterizations of the tropopause, indicating a direction for further study. Although the 5-year change in tropopause is explicable by solar heating of the troposphere, this does not disprove that climatic change can result from increases in greenhouse gas concentrations over much longer timescales. Over intra-annual timescales, results are in agreement with similar observations elsewhere and by other techniques and strengthen the opinion that the tropopause location is influenced, not only by tropospheric temperature but also by ozone-driven stratospheric temperature. At short timescales, the stochastic component of the signal has been identified, albeit in an exploratory study, as fractional Gaussian noise with a Hurst component of 0.82, in very close agreement with similar results for surface air temperature, and thus showing promise for future complexity matching investigations, a technique still in its infancy.

[18] The radar observations presented here represent a useful supplement to traditional radiosonde soundings. The radar has the advantage that it can operate unmanned and deliver tropopause altitude and, potentially, other tropospheric information at relatively short timescales. In addition, quasi simultaneous upper atmosphere echoes can be obtained which could be used to provide information on coupling between different regimes in the atmosphere. On the other hand radiosondes provide more detailed information from dedicated sensors and so radar techniques can never be used to replace these in situ observations. Since the SOUSY Svalbard radar is located at some distance from the nearest radiosonde stations, combining information from the different instruments can provide tropopause altitudes at timescales from hours to solar cycles. Such combined observations open up possibilities for examination of spatial variation and comparison of traditional and innovative analysis approaches.

Acknowledgments

[19] The authors would like express gratitude for the help and support of the Jicamarca staff, in particular Karim Kuyeng and the members of the Electronics and Instrumentation group. We thank the Alfred Wegener Institute, in particular Marion Maturilli, for tropopause data from their Ny Ålesund station. Temperature data have been provided by Fred Sigernes and Stefan Claes from the Kjell Henriksen Observatory. We also acknowledge the occasional maintenance of SOUSY by the staff of the EISCAT Svalbard Radar.