High- and low-frequency 11-year solar cycle signatures in the Southern Hemispheric winter and spring

Authors


Abstract

We have studied the characterization of the 11-year solar cycle (SC) signals in the Southern Hemisphere (SH) during the winter and spring using European Centre for Medium-Range Weather Forecasts (ECMWF) daily and monthly data from 1979 to 2009. By separating the response into high (<6 months) and low (>36 months) frequency domains, we have found that spatially different 11-year SC signals exist for high- and low-frequency domains. In the stratosphere, the high- and low-frequency responses tend to enhance each other near the Equator and Subtropics, while they oppose one another at high latitudes. The high-frequency response is marked by a strengthened stratospheric jet during winter and the response is not static but tracks with the centre of the polar vortex. In the lower stratosphere, the positive response of temperature to the 11-year SC is dominated by its low-frequency component, which extends from the North Pole to the South Pole. The low-frequency tropospheric response is latitudinally symmetrical about the Equator and consistent with the modelled responses to temperature perturbation in the lower stratosphere. The signals are found to be sensitive to contamination from the 2002 sudden stratospheric warming event and major volcanic eruptions but the general spatial pattern of the responses remains similar. A significant projection of the 11-year SC onto the Southern Annular Mode (SAM) can only be detected in the stratosphere and in the high-frequency component. The signature is marked by a strengthening of the stratospheric SAM during winter and a weakening of the SAM in the uppermost stratosphere during spring. Copyright © 2011 Royal Meteorological Society

1. Introduction

Studies have shown that changes associated with the 11-year solar cycle (SC) have detectable effects on the stratospheric and tropospheric circulation (e.g. Gleisner and Thejll, 2003; Haigh, 2003; Coughlin and Tung, 2004; Salby and Callaghan, 2006; Lu et al., 2007; van Loon et al., 2007). The 11-year SC signature in stratospheric temperature is characterized by positive correlation at low latitudes with a vertical double-peaked structure: one in the lower stratosphere and another in the upper stratosphere (Crooks and Gray, 2005; Keckhut et al., 2005; Claud et al., 2008; Frame and Gray, 2010; Gray et al., 2010). In zonal-mean zonal wind, it is marked by a strengthening of the subtropical jet in the upper stratosphere and a poleward and downward movement of westerly anomalies (Kuroda and Kodera, 2002; Kodera et al., 2003; Gray et al., 2004, 2010; Matthes et al., 2004).

As the total solar irradiance varies by only ∼0.1% over an 11-year SC, it has been suggested that larger variations of solar ultraviolet (UV) radiation (∼5–8% over an 11-year SC) and its absorption by stratospheric ozone could provide a detectable solar link to variation of the global circulation (Haigh, 1994, 2003). Thermal structure change caused by the associated heating in the upper stratosphere may lead to a modulation of the stratospheric polar vortex and cause a dynamical feedback in the lower stratosphere (Kodera and Kuroda, 2002). The enhanced equatorial heating due to solar UV–ozone interaction results in anomalously stronger westerlies in the upper stratosphere/lower mesosphere, in thermal wind balance with an enhanced pole-to-Equator temperature gradient. The stronger westerlies may deflect planetary waves poleward and cause a further strengthening of the polar vortex (Kodera et al., 2003). The solar UV-induced wind anomalies move poleward and downward as the winter progresses (Kodera and Kuroda, 2002; Kodera et al., 2003; Matthes et al., 2004).

Perturbation of the stratosphere by UV radiation variations followed by downward propagation of the resulting circulation anomalies to the surface has been proposed to explain the observed tropospheric solar signals (Haigh et al., 2005; Hameed and Lee, 2005; Matthes et al., 2006). Changes to the winter stratospheric polar vortex may influence the underlying tropospheric circulation. In addition, idealized model simulations have suggested that a solar modulation of the equatorial lower stratosphere, which affects the Equator-to-pole temperature gradient, may also modulate the synoptic scale wave activity (Haigh and Blackburn, 2006; Simpson et al., 2009). Gas chromatographic–mass spectrometric studies also suggest that solar UV-induced change in the Brewer–Dobson (BD) circulation in the upper stratosphere can cause a suppression of tropical convection during solar maximum (Kodera, 2004; Kodera and Shibata, 2006; Matthes et al., 2006). One thing in common with the above proposed mechanisms is that they all involve a stratospheric influence on the tropospheric circulation and they are consequently referred to as ‘top-down’ mechanisms (Gray et al., 2010).

Another mechanism of solar influence over an 11-year cycle is through air–sea radiative coupling at the ocean surface in the Tropics, whereby the spatial asymmetries of solar forcing, induced by cloud distributions, result in greater evaporation in the Subtropics and consequent moisture transport into the tropical convergence zones (Meehl et al., 2003). Higher solar forcing may cause stronger upward motion in the winter hemispheric Subtropics and enhanced downward motion in the summer hemispheric Subtropics (van Loon et al., 2004; Meehl et al., 2008). As it involves near-surface processes and because significant responses have been detected in models without a stratosphere, it is consequently referred to as the ‘bottom-up’ mechanism (Gray et al., 2010). For a recent review of all proposed mechanisms of solar influence on the climate, see Gray etal. (2010).

It is possible to study each individual mechanism separately using models but it is very difficult to separate the responses associated with different mechanisms using observational data. This presents a substantial challenge to interpreting the solar signals of different origin/cause based on the observational signature. Proposed mechanisms involve fast processes such as wave mean flow interaction and/or slow processes such as temperature change induced by redistribution of ozone in the lower stratosphere and differential ocean heating. For instance, the wave-driven dynamic response to the 11-year SC tends to change on sub-monthly time-scales (e.g. Kodera, 2004; Matthes et al., 2006; Lu et al., 2009), while the radiative response is more likely to be static over an extended period, especially in the lower stratosphere, and hence can be detected in the annual mean (e.g. Keckhut et al., 2005; Claud et al., 2008; Frame and Gray, 2010). It is natural to think that the responses associated with fast processes may show signals with a high-frequency component within the atmospheric data, while processes with slow responses may be detected better in the low-frequency domain.

In an attempt to refine the traditional way of analysing the atmospheric response to the 11-year SC, here we separate the atmospheric data into high- and low-frequency domains, so that we are able to examine the characterization of the solar signal in terms of fast versus slow processes, direct versus indirect effects, radiative versus dynamic responses and their possible links to ‘top-down’ and/or ‘bottom-up’ mechanisms. We focus on the Southern Hemispheric (SH) winter and spring, where the polar vortex is stronger and longer-lived than its Northern Hemispheric (NH) counterpart due to weaker planetary wave forcing. This implies that radiative interaction is likely to play a relatively larger role in the SH winter than in the NH winter, and the solar signals in the SH have been found to be different from those in the NH (Labitzke, 2002; Salby and Callaghan, 2006). The dynamically ‘calm’ SH winter may in fact present a better environment to differentiate high- and low-frequency responses as the signals are less likely to be affected by transient events such as stratospheric sudden warmings (SSWs). For example, the modelling study of Cnossen et al. (2011) showed that the high-latitude stratospheric and tropospheric responses differ significantly for the SSW condition and non-SSW conditions. By focusing on the SH winter and spring, we can understand more how the 11-year SC signal behaves under non-SSW conditions.

Although our main focus is on detecting high- and low-frequency solar responses in the Earth's atmospheric temperature and zonal wind and studying how those signals might be linked to previously proposed mechanisms, we also study whether or not the 11-year SC may project onto the Southern Annular Model (SAM) more clearly if the data are separated into high- and low-frequency components. Because the stratospheric effects on the troposphere may be realized through the SAM and the coupling depends primarily on wave activity (Thompson et al., 2005; Limpasuvan and Hartmann, 2000), it is expected that the high-frequency component plays a more important role than the low-frequency component. In this study, we will check if that is also the case for the 11-year SC signal on the SAM.

Previous studies have shown that the large-scale structure of the late winter and spring SAM is modulated by the 11-year SC (e.g. Kuroda and Kodera, 2005; Kuroda et al., 2007; Kuroda and Yamazaki, 2010). These authors have shown that the 11-year SC modulates the spatial structure of the SAM from October through to December. In solar maximum years, the solar influence on the SAM extends vertically from the surface to the upper stratosphere, while it is capped within the troposphere during solar minimum years. Using a coupled chemistry–climate model, Kuroda and Shibata (2006) found that the SAM signal in the Antarctic stratosphere is noticeably stronger in the high UV runs than in the low UV runs. Unlike those previous studies, our main focus here is detecting the 11-year SC signals in the atmospheric variables including the SAM rather than the modulating effect of the 11-year SC on the SAM signature. More precisely, we do not study how the spatial extent of the SAM signature in wind or temperature changes under high and low solar conditions. Instead, we study how much of the interannual variability of the SAM may be directly linked to the 11-year SC variation and at what frequency and altitude such a linkage may be statistically significant.

The rather sparse observational network before the satellite era (i.e. pre September 1978) in the SH limits the accuracy of the ERA-40 reanalysis before 1979. As a result, few observational studies of the atmospheric response to the 11-year SC have been undertaken for the SH winter and spring. With a relatively short record length, it is important to consider the influence of other major physical disturbances. SSWs have so far been recorded in the SH on only one occasion, which was in 2002. Consequently, in the 2002 winter and spring, the stratospheric wind and temperature are distinctly anomalous to the other years (Charlton et al., 2005; Scaife et al., 2005). In addition to the primary focus of this study, the dynamic features of 2002 will be briefly discussed in the context of the atmospheric response to solar forcing. Also, the possible contamination of results by the temporary warming associated with volcanic aerosols will be examined by comparing results after excluding and including two years of data following two major eruptions (i.e. El Chichón in March 1982, and Pinatubo in June 1991). Owing to the limited sample size, the modulating effect of the equatorial Quasi-Biennial Oscillation (QBO) will not be explicitly studied here as statistical significance cannot be reliably established with any further subsampling, though the possible effect of the QBO on the detected solar signals will be briefly discussed.

2. Data and methods

Data used by this study are daily mean zonal winds and temperatures blended from the European Centre for Medium-Range Weather Forecasts (ECMWF) ERA-40 Reanalysis (January 1979 to August 2002) and ECMWF Operational analyses (September 2002 to December 2009), as employed by Frame and Gray (2010). The ERA-40 dataset was assimilated using direct radiosonde and satellite measurements and has a horizontal resolution of 1.125° in both latitude and longitude on 23 pressure levels from 1000 to 1 hPa (Uppala et al., 2005). The ECMWF Operational data were output from the ongoing analyses produced by the most recent ECMWF Integrated Forecasting System (IFS) model. Data from September 2002 to the present day are available on the same 1.125° grid but on 21 pressure levels (before 7 November 2007) and 25 pressure levels (since 07/11/2007). The ERA-40 and the Operational datasets share 21 of the pressure levels, the exceptions being four levels in the lower troposphere (i.e. 600, 775, 900 and 950 hPa). For simplicity, only the data for those 21 pressure levels are used here. Both ERA-40 and Operational datasets extend to 1 hPa (∼50 km), thus allowing an examination of the solar signals throughout the stratosphere. While it is not ideal to merge two datasets derived from different data assimilation models, we have done this here in order to maximize the length of the dataset, as needed for the study of 11-year SC signals.

The daily data have not previously been used for this purpose; therefore this work is complementary to earlier analyses by providing a more detailed temporal resolution on the timing of the solar signals and their propagation pathways. Before the satellite era (i.e. before September 1978), the scarcity of SH radiosonde measurements and lack of direct measurement at altitudes above 10 hPa result in unreliable estimations in ERA-40, particularly in the upper stratosphere. To better capture solar perturbation of the stratosphere by UV variations and the downward propagation of the resulting circulation anomalies, only data from January 1979 onwards are used. The entire period of 1979–2009 covers 31 years and about three 11-year SCs.

Daily observed 10.7 cm solar radio fluxes were obtained from the National Geophysical Data Center (NGDC) website and are used as a proxy for the 11-year SC. Their power spectrum shows that the variability in the daily 10.7 cm solar radio fluxes is dominated by two frequency bands: one around 26–28 days and another around 9.5–11.5 years. In order to focus on the 11-year SC, a 365-day low-pass filter was applied to the solar radio fluxes. The filtered daily time series is denoted as Fs hereafter, and is used in the analysis. Qualitatively similar results can be obtained if a running average window of 1–12 months is applied to the daily solar data. Noisier but qualitatively similar results can also be obtained by using the raw daily fluxes. Similar to Lu et al. (2007, 2009), we use equation image and equation image to define low solar (LS) and high solar (HS) activity when deriving composites, where equation image stands for the normalized values of Fs; transition years where equation image are excluded. Under this definition, 11 years (i.e. 1979–1982, 1989–1992 and 1999–2001) were HS years and 15 years (i.e. 1984–1987, 1994–1998 and 2004–2009) were LS years for June to October mean.

A possible projection of solar signal onto the SAM is studied by using the SAM index derived from two different sources. The first is the daily SAM index derived as the leading empirical orthogonal function (EOF) of daily zonal-mean zonal wind over 20–90°S, from the blended ECMWF ERA-40 Reanalysis and Operational data for the same period. Details about this method can be found in Baldwin and Thompson (2009). The second type of SAM index is the station-based SAM index, which is estimated as a monthly mean difference between the mean sea-level pressure anomaly at six stations close to 40°S and six stations close to 65°S (Marshall, 2003, available at www.nerc-bas.ac.uk/icd/gjma/sam.html). By using only long-term stations and surface pressure measurements, this index avoids the possible bias problems when combining the ERA-40 and ERA Operational datasets. Thus the principal advantages of the Marshall (2003) index are its simplicity and temporal consistency across its entire time-span and between different seasons. As the station-based SAM extends back to 1957, it is used to test the robustness of the signals near the surface. For simplicity, these two SAM indices are referred to hereafter as daily ECMWF-SAM and monthly Marshall-SAM, respectively.

The main diagnostic tools used are composite analysis and linear regression, with standard Student t-test being used to test their significance. The analyses were carried out for a range of frequency bands where Chebyshev filters (Smith, 1997) are used to separate one band of frequencies from another. Chebyshev type II is used as high-pass filter, while Chebyshev type I is used as low-pass filter. The main criteria in designing the filters are a small (i.e. <0.5%) peak-to-peak ripple and stability for all geographical locations. Through a large number of computational experiments, we found that a high-pass filter with an order of 15–17 and peak-to-peak ripple of 2–5 dB in the normalized passband was optimal for monthly data and an order of 3–5 was sufficient for daily data. For the low-pass and bandpass filter, filter orders in the range of 4–9 for the monthly mean and 1–3 for daily data are sufficient. In the sections below, a ripple magnitude of 2 dB is used throughout. Similar results can be obtained if a Butterworth filter is used, though higher-order filters are required.

3. Solar signal during SH winter and spring

3.1. Winter and spring mean signals

Figure 1 shows composite differences (HS—LS) of June–August averaged zonal-mean temperature (ΔTHS–LS, first column) and zonal wind (ΔUHS–LS, second column), in meridional–height cross-section. When the entire record is used (Figure 1(a) and (e)), significant responses to the 11-year SC are detected primarily in temperature and in the stratosphere. In temperature, significant positive ΔTHS–LS are found to extend from 90°N to 40°S latitudinally and from 7 hPa to 100 hPa vertically. Near the Equator, ΔTHS–LS is statistically significant at a 95% confidence level only at ∼70 hPa and is not significant at other altitudes. In the polar SH, ΔTHS–LS displays an alternating vertical negative and positive pattern, which is most likely to be an artifact of the dataset rather than a real response. Significant ΔUHS–LS is found in the mid-latitude stratosphere in the NH. In the SH, although ΔUHS–LS is marked by westerly anomalies at 15–45°S and easterly anomalies poleward of 55°S, they are not significant at a 95% confidence level.

Figure 1.

Composite differences (high solar - low solar (HS—LS)) of zonal-mean temperature (a–d) and zonal wind (e–h) for June–August averages. In cases (a, e), all data from 1979 to 2009 are included. In cases (b, f), data from year 2002 are excluded. In cases (c, g), years affected by the major volcanic eruptions (i.e. 1982, 1983, 1992 and 1993) are excluded. In cases (d, h), data from 2002 and years affected by the major volcanic eruptions are excluded. Statistical confidence levels above 95% are shown as dashed grey contours.

A similar response pattern can be obtained if selected anomalous data are excluded, i.e. 2002 major SSW (Figure 1(b) and (f)), the major volcanic eruptions (Figure 1(c) and (g)) and both the major SSW and the volcanic eruptions, although the magnitude and significance of the differences change slightly. Possible contaminations by the major SSW and the volcanic eruptions can be viewed as enhancement or weakening of the solar signals. When data from 2002 are excluded, the low-latitude temperature response is enhanced and the westerly anomalies at 15–45°S become significant. When the years affected by the major volcanic eruptions are excluded, the temperature response becomes weaker (<0.3 K) in the lower stratosphere and stronger (∼0.5 K) in the upper stratosphere. Thus the positive solar signature in lower-stratospheric temperature is amplified by the bias of the eruptions of El Chichón and Mount Pinatubo. Almost no significant signals can be found in the troposphere except for the westerly anomalies near the Equator in the SH.

It has been suggested that the lack of statistically significant solar signals at 5–50 hPa in the Tropics is likely due to a contamination effect of the QBO (Smith and Matthes, 2008). To examine such a possibility in a simple manner, the same composite analysis is carried out at two distinct frequency domains distinctly away from the typical period of the QBO (∼28 months). Figure 2 shows the June–August zonal-mean temperature (a–c) and zonal wind (d–f) responses to the 11-year SC when a 6-month high-pass filter (a, d), 36-month low-pass filter (b, e) and linearly detrend plus 36-month low-pass filter are applied to the atmospheric data, respectively. Where a 6-month high-pass filter is applied to the data, the primary feature of the temperature response to the 11-year SC is positive ΔTHS–LS in the equatorial and subtropical stratosphere and at high latitude in the NH upper stratosphere. In the high-latitude SH there are alternating positive and negative anomalies, but only the negative ones are significant at the 0.05 level (i.e. a confidence level of 95% or above), implying a dominated cooling effect in the region under HS condition. A small region near the Equator in which a 95% confidence level cannot be achieved is found at ∼10–20 hPa, though a positive temperature difference remains there. The accompanying solar signals in zonal wind are marked by westerly anomalies up to 12 m/s in the SH subtropical stratosphere extending downward from 1 to 70 hPa and 3 m/s in the NH subtropical stratosphere extending downward from 1 to 20 hPa. Easterly anomalies are found in the high-latitude upper stratosphere but are only significant in the NH at 55–75°S, 1–2 hPa (see Figure 2(d)). A similar pattern of solar response is obtained if the high-pass cut-off period is in the range of 3–12 months, and the magnitude and significance of the responses reduce when the cut-off period is either smaller than 3 months or greater than 12 months. With a 36-month low-pass filter (see Figure 2(b) and (e)), the temperature response is marked by a general positive ΔTHS–LS in the stratosphere where a significant 11-year SC signal appears near the Equator, in the Subtropics and also in the polar region. The corresponding response in the zonal wind is broadly symmetric about the Equator. In the upper stratosphere, it is characterized by statistically significant westerly anomalies in the Subtropics and partly significant easterly anomalies poleward of 60°. In the middle to lower stratosphere, it is marked by easterly anomalies near the Equator and in the region poleward of 60°, and by westerly anomalies at mid latitudes. In the troposphere, it is marked by westerly anomalies near the Equator and by easterly anomalies near 30° and in the region poleward of 60°. The clearest vertical connection of the signals is present in the westerly anomalies originating in the subtropical upper stratosphere. The spatial pattern remains similar when the lower cut-off period is anywhere in the range of 12–36 months. However, the equatorial-symmetric signal becomes noticeably stronger and clearer when the cut-off period is greater than 28 months because the contamination of the QBO is suppressed.

Figure 2.

Composite differences (high solar—low solar (HS—LS)) of zonal-mean temperature (a–c) and zonal wind (d–f) for June–August averages. In cases (a, d), a Chebyshev type II high-pass filter with a cut-off period of 6 months is applied. In cases (b, e), a Chebyshev type I low-pass filter with a 36-month cut-off period is used. In cases (c, f), temperature and zonal wind are linearly detrended first and then a Chebyshev type I low-pass filter with a 36-month cut-off period is applied. Statistical confidence levels above 95% are shown as dashed grey contours.

It is known that there has been a steady decrease in temperature in the stratosphere and increase in temperature in the troposphere over the last three decades and the diagnosis of the solar signal is sensitive to the presence of trends in the data (Lu et al., 2007; Kodera et al., 2008; Frame and Gray, 2010). To account for the effect of the long-term trends, the atmospheric data are first detrended linearly before applying a 36-month low-pass filter (Figure 2(c) and (f)). The results of this show that the effects of removing the linear trend give a generally weaker and less significant response in the stratosphere and an enhanced response in the troposphere. In particular, at 30–50° in both hemispheres, a significant positive temperature response (∼0.4 K) in the troposphere appears only when the long-term trend is removed. Conversely, only marginal changes in zonal wind response are found to be associated with long-term trend. There is an indication that the high- and low-frequency responses in tropospheric temperature are of the opposite sign, though the high-frequency responses are hardly significant at a 95% confidence level. More data are needed to confirm such a cancellation effect. Unlike temperature, the signature in zonal wind is not sensitive to the long-term trends.

3.2. Seasonal progression of high-frequency signals

It has been suggested that the temperature responses in the lower stratosphere are likely to be indirect responses either through a modification of the BD circulation, changes in ozone transport or both (Kodera and Kuroda, 2002; Hood and Soukharev, 2006; Marsh and Garcia, 2007; Gray et al., 2009; Eyring et al., 2010). In this section, we focus on the temporal evolution of the high-frequency signal, which helps the diagnosis of dynamical responses. We also show how the high-frequency signals in the upper stratosphere might contribute to those in the lower stratospheric temperature, through a downward propagating mechanism which is likely to be associated with changes in BD circulation. Daily data are used to demonstrate that solar UV forcing in the upper stratosphere may be transferred dynamically through a downward and poleward movement of the signals.

3.2.1. Signals in temperature

Figure 3 shows the monthly zonal-mean temperature (Tclim) for June to October (first column) and the corresponding 6-month high-pass-filtered temperature differences ΔTHS–LS (second and third columns), all displayed in meridional–height cross-section with latitude extending from 30°N to 90°S. The corresponding zonal wind signals are described in the next section. In June, a significant positive ΔTHS–LS signal emerges in the equatorial and subtropical upper stratosphere (5–10 hPa) and in the lower stratosphere (70 hPa). The upper stratospheric signature enhances in July and latitudinally expands and descends towards the lower stratosphere in August. Significant negative values of ΔTHS–LS are found in the high-latitude SH in July and August. Such a response pattern in temperature suggests an enhanced Equator-to-pole temperature gradient in both the upper and lower stratosphere. The high-latitude and lower stratospheric tropical responses would be expected from a weakened BD circulation in solar maximum years or from a strengthened BD circulation during solar minimum years. The temperature responses both at low and high latitudes weaken in spring (September and October) and the high-latitudinal negative signals are replaced by positive temperature responses.

Figure 3.

Left: monthly climatology zonal-mean zonal temperature (in units of K) for June to October (top to bottom), displayed as lined contour plots in meridional–height cross-section from 30°N to 90°N and from 1000 to 1 hPa. Right: same as the left panels but displayed as coloured contour plots for the composite differences of the temperature between high solar and low solar conditions (HS—LS). The areas enclosed within grey lines indicate that the differences are statistically significant from zero with a confidence level of 95% or above, calculated using a Monte Carlo trial-based non-parametric test. Data which might be affected by the major volcanic eruptions and 2002 stratospheric major warming are excluded.

Figure 4 shows the daily running regression (in K per 100 solar units) of zonal-mean temperature on Fs averaged over 5°S to 5°N (a, c, e), and 55°S to 65°S (b, d, f), where data from 2002, 1982, 1983, 1991 and 1992 are excluded. In Figure 4(a) and (b), the regression is performed based on raw zonal-mean temperatures at those two latitude bands and daily Fs. It shows that a downward descent of the positive temperature anomalies with maximum magnitude of 3 K per 100 solar units is found at ∼2 hPa near the equator in late June (Figure 4(a)). The signal descends to 20 hPa in late August. There is a weaker but significant (i.e. at a confidence level of 95% or above) temperature response at 70 hPa for the same period which does not show any direct vertical connection with that originating in the equatorial upper stratosphere. A negative temperature response (∼6 K per 100 solar units) appears at various altitudes at 55–65°S in July and August (Figure 4(b)).

Figure 4.

Time versus pressure level plot of linear running regression (in K per 100 units of F10.7 cm solar flux) between Fs and equatorial temperature averaged over 5°N to 5°S (a, c, e, where the contour interval is ±0.5 K) and temperature averaged over 55–65°S (b, d, f, where the contour interval is ±2 K). In (a, b), raw daily temperature is used. In (c, d), a high-pass filter with a cut-off period of 183 days is applied to the daily temperature. In (e, f), a bandpass filter with a bandwidth of 31–183 days is applied to the daily temperature. An order 4 Chebyshev type II with passband ripple of 2 dB is used for (c–f). For all cases, the temperatures are linearly detrended first and the data from 2002, 1982, 1983, 1991 and 1992 are excluded from the regression. Statistical confidence levels above 90% and 95% are shown as light and dark shadings.

Figure 4(c) and (d) shows the same as Figure 4(c) and (d) but with a Chebyshev type II high-pass filter having a cut-off period of 183 days applied to the temperature data. An apparent descent from 20 to 100 hPa in the case of equatorial temperature and from the stratosphere to the troposphere in the case of high-latitude temperature can now be observed. Similarly, Figure 4(e) and (f) shows the same as Figure 4(c) and (d) but with a Chebyshev type II band-pass filter having a bandwidth of 31–183 days applied to the temperature data. Downward descent of the solar signals becomes clearer, though only the high-latitude signal manages to cross the tropopause and move into the troposphere.

We found that the downward movement from the upper stratosphere to the lower stratosphere is robust and becomes clearer when either a Chebyshev type I or type II or Butterworth filter is applied to the temperature data. Together, the signals suggest that during these seasons the high-frequency response in the stratosphere is characterized by warming in the Tropics and cooling at high latitudes, implying a slowing down of the BD circulation. They also suggest that the effect moves gradually from the upper stratosphere to the lower stratosphere. However, the downward movement of the signal into the troposphere loses statistical significance when either a Chebyshev type I or Butterworth filter is applied. When the cut-off period of the filter is increased beyond 6 months, the high-latitude signals become noticeably weaker (not shown). Thus it remains to be verified whether or not the high-latitude temperature signal actually descends from the stratosphere into the troposphere in reality.

To examine the variability of the solar signals in stratospheric temperature for HS and LS years more closely, time series of the daily temperatures in the upper and lower stratosphere for each individual year are shown in Figures 5 and 6, respectively. In Figure 5, daily mean temperature at 10 hPa averaged over the latitude ranges of 10–15°S (top row) and 55–65°S (bottom row) are compared for HS (first column) and LS (second column) conditions, where the averages are weighted by the cosines of the latitudes.

Figure 5.

Daily averaged zonal-mean temperatures (K) at 10 hPa, averaged over 0–5°S (first row) and over 55–65°S (second row), for HS (first column), and LS (second column) together with daily mean values for HS (line with red shading) and LS (line with grey shading) conditions (third column). The daily mean values for HS and LS at high frequency (<183 days) estimated using a Chebyshev type II filter (fourth column) are also shown. The shaded regions represent 95% confidence intervals of the mean. In the HS group, the years in which data might be affected by major volcanic eruptions are shown as blue lines and the 2002 data are shown as a purple line. The daily mean values are calculated by excluding years affected by major volcanic eruptions and 2002 data.

Figure 6.

As in Figure 5 but for temperatures at 50 hPa, averaged over latitude ranges of 10–15°S (first row) and 55–65°S (second row).

Figure 5 shows that, in the upper stratosphere, considerable temperature variation exists and the variance is larger for HS years than for LS years. The larger temperature variation under HS may be partially due to the contaminating effects of the major SSW (year 2002; purple line) and the major volcanic eruptions (years 1982, 1991 and 1992; blue lines). The third and fourth columns of Figure 5 display the mean daily temperatures under HS (red) and LS (grey) conditions for no filtering and183-day Chebyshev type II high-pass filter. The shaded regions represent 95% confidence intervals for the mean. When the shaded regions do not overlap, this indicates that the average temperature differences between the HS and LS groups are significant at the 95% confidence level. In order to focus on the 11-year SC effect, years affected by the SSW and major volcanic eruptions are excluded from the group mean in HS. For equatorial temperature, statistically significant differences between HS and LS conditions are evident for both the raw data and the high-frequency component. For high-latitude temperature, the difference is only significant for the high-frequency component. When the unfiltered data are used, it is barely differentiable.

Figure 6 shows the corresponding temperature time series for the lower stratosphere at 50 hPa and for the same latitude ranges as in Figure 5. It shows that higher temperatures existed for the volcanic eruption affected years near the Equator and Subtropics, due to the warming effect of the volcanic aerosols (Robock and Mao, 1992). In the 2002 winter, substantially lower temperatures are found at low latitudes, accompanied by higher temperatures at high latitudes. In the Subtropics (top row), significant positive temperature differences between HS and LS exist whether or not a filter is applied. The response obtained by using raw unfiltered daily data is found from April to September, while the high-frequency response occurs briefly only in August and September. The magnitude of the response is larger in the unfiltered data than in the high-frequency component. Conversely, the high-latitude temperature differences are only significant in the high-frequency domain in July to early September. The lack of significance in unfiltered high-latitude temperature in the lower stratosphere is due to a cancellation effect between high- and low-frequency components (not shown).

3.2.2. Solar signals in zonal wind

Figure 7 is similar to Figure 3 except that the temperature is replaced by the zonal-mean zonal wind. The composite differences are the high-frequency zonal-mean zonal wind between HS and LS conditions (ΔUHS–LS). The climatology of the wind field is characterized by a strengthening and poleward movement of the stratospheric westerly jet from June to mid August and a gradual weakening and downward movement of the jet thereafter. The high-frequency ΔUHS–LS signature is clearly marked by a strengthening of the stratospheric jet as the significant westerly anomaly almost tracks the centre of the jet. The signals originate in the subtropical upper stratosphere and lower mesosphere and then move poleward and downward. There is a symmetric, but weaker, response in the NH Subtropics (not shown), which shows little poleward movement. As a whole, the overall structure of high-frequency ΔUHS–LS in the stratosphere agrees exceedingly well with that proposed earlier by Kodera and Kuroda (2002) and is consistent with the thermal wind balance in the presence of the anomalous latitudinal temperature differences at 5–10 hPa shown in Figure 3.

Figure 7.

As in Figure 3 but the zonal-mean temperature is replaced by the zonal-mean zonal wind in units of m s−1.

To examine the high-frequency responses and variability of the wind in more detail, Figure 8 shows the daily evolution of the zonal wind at 30–40°S, 5 hPa (upper panels), and at 40–50°S, 50 hPa (lower panels) for HS (first column) and LS (second column) conditions. Significant positive ΔUHS–LS for the unfiltered data exists only in austral winter and only in the upper stratosphere winds. The difference is noticeably enhanced both in the upper and lower stratosphere if a high-pass filter is applied. ΔUHS–LS becomes insignificant if data from 2002 are included.

Figure 8.

As Figure 5 but for zonal-mean zonal wind (m/s) at 30–40°S, 5 hPa (first row) and at 40–50°S, 50 hPa (second row).

3.3. Solar signal in the SAM

It is known that the changes in zonal-mean zonal wind in the tropospheric westerly jet during the SH winter are affected by changes in wave forcing (Limpasuvan and Hartmann, 2000; Lorenz and Hartmann, 2001) and the effects may be projected onto the large-scale atmospheric mode, i.e. the SAM (Thompson et al., 2005). Figure 9 shows the running linear regression (in SAM index per 100 solar units) using raw daily SAM (a) and 183-day high-pass filtered SAM (b), where the linear trend is removed and the data from 2002 and two years following the major volcanic eruptions are excluded. An almost identical response pattern can be seen in (a) and (b), suggesting that the SAM response to the 11-year SC is dominated by the high-frequency response. In the stratosphere, a positive solar response originates at 1 hPa and above in early July and descends towards the lower stratosphere in July and August. A negative response is obtained from October near the stratopause. Noisier but significant responses are also found in the troposphere. They are positive in June–July preceding the stratospheric signal, negative during mid-winter, and positive again from October onward. No consistent signals can be found in the low-frequency domain (not shown).

Figure 9.

As in Figure 4 but for the SAM derived from ECMWF daily data: (a) the raw daily SAM is used; (b) a 183-day high-pass filter is applied to the SAM. The treatment of data and use of filter, lines and shadings are the same as for Figure 4 and the contour interval is 0.25 SAM index per 100 solar units.

Figure 9 suggests that the 11-year SC may project onto the tropospheric SAM in early winter or late spring. To test whether there is any definite projection of the 11-year SC on the SAM in the troposphere, scatter-plots of Fsversus the station-based monthly Marshall-SAM from 1957 to 2009 are shown in Figure 10. We found that in both June and November the correlations are not significantly separable from zero either using the raw data or in the high- or low-frequency domains. Similar results were obtained for other months or seasonal averages. We also found that there is no significant difference in the variance and the histogram of the SAM under HS and LS conditions in the SH winter and spring.

Figure 10.

(a, b) Scatter-plot of Fs and July Marshall-SAM index. (c, d) Scatter-plot of Fs and November Marshall SAM index. For (a, c) no filtering is applied to the data; for (b, d) a high-pass filter of 6 months is applied to the data. This figure is available in colour online at wileyonlinelibrary.com/journal/qj

4. Discussion

The most significant solar signature is the positive response of stratospheric temperature, consistent with the theory that the interaction between solar UV and ozone photochemistry increases temperature there (Haigh, 1994). However, the stratospheric signals of the 11-year SC are sensitive to the contamination of the 2002 SSW event and the major volcanic eruptions, and zonal winds are more sensitive than temperature. We have noted that the zonal wind response becomes noticeably clearer and more significant by excluding the data from year 2002 or by separating the analysis into high- and low-frequency components. The fact that the solar signal in zonal wind is more sensitive to sampling methods than that in temperature implies that temperature is more directly affected by the variation of solar output during the 11-year SC. This is consistent with the known mechanism of solar UV–stratospheric ozone interaction. In the troposphere, however, the solar signature in temperature is more sensitive to the long-term trends than to the SSW or the major volcanic eruptions. Linearly removing the trends enhances the signal in the troposphere and weakens the signals in the stratosphere. Without removing a linear trend, up to 0.3 K of the temperature response in the low-latitude lower stratosphere might be the result of aliasing of a long-term trend onto the solar signal, and this is in good agreement with Lu et al. (2007) and Frame and Gray (2010).

Here we have found that, during the SH winter, the 11-year SC signals become stronger and statistically more significant when the analysis is carried out at high and low frequencies than when the raw data are used. The difference is more noticeable at high latitude as the high- and low-frequency responses tend to cancel each other there. The cancelation partly explains why high-latitude responses are hard to detect statistically when raw data are used.

The difference in high- and low-frequency responses highlights new aspects of the underlying physical processes. The static low-frequency responses in the stratosphere may represent the slow varying temperature and circulation change due to ozone transport induced by solar UV–stratospheric ozone interaction and changes in BD circulation (Kodera and Kuroda, 2002; Gray et al., 2009, 2010). The non-static high-frequency responses in temperature, zonal wind and the SAM to the 11-year SC, and the fact that the signals are statistically significant only during SH winter, point to a dynamic effect of the 11-year SC on the stratospheric circulation. These high-frequency signals are generally consistent with the dynamic response studied previously (Kodera and Kuroda, 2002; Matthes et al., 2004; Kuroda et al., 2007). It may be interpreted as a possible association between the 11-year solar cycle and the temporal or seasonal persistence of the polar vortex and the BD circulation. Apart from the known mechanism of in situ solar UV and ozone photochemistry interaction, other mechanisms may also be at play to produce those stratospheric signals. For instance, the signature of the solar wind streaming out from the Sun has also been found in various climate records (Lu et al., 2008b; Seppälä et al., 2009; Lockwood et al., 2010a, 2010b; Woollings et al., 2010). Observational studies have shown that the solar wind-induced geomagnetic activity may alter stratospheric chemistry indirectly through energetic particle precipitation (EPP) (Solomon et al., 1982; Randall et al., 2007; Seppälä et al., 2007; Siskind et al., 2007). Chemical–dynamical coupled general circulation models (GCMs) have indicated that odd nitrogen (NOx) induced by energetic charged particle precipitation during geomagnetic storms may cause temperature changes in both the polar and equatorial regions and in the stratosphere and the troposphere (Langematz et al., 2005; Rozanov et al., 2005). Lu et al. (2007) found that the magnitude of temperature response in the low-latitude lower stratosphere to geomagnetic activity was slightly larger than that associated with the 11-year SC and that the temperature response to the 11-year SC and geomagnetic Ap index tend to enhance each other. Lu et al. (2008b) have shown that there is a robust relationship between solar wind dynamic pressure and the zonal wind and temperature in the northern polar winter. Stratospheric wind and temperature variations are positively projected onto the Northern Annular Mode (NAM) when the 11-year SC is at its maximum phase, and are negatively projected onto the NAM during the 11-year SC minimum phase. A weakening of the BD circulation with reduced upwelling into the lower stratosphere at low latitude under high solar wind forcing is consistent with the behaviour of the high-frequency response shown here. Our rationale is that, at solar maximum, a warming effect in the low-latitude stratosphere due to solar UV–stratospheric ozone interaction, together with a cooling effect in the high-latitude stratosphere due to solar wind-driven geomagnetic activity, would enhance the Equator-to-pole temperature gradient. When both mechanisms work together, this would result in a stronger dynamic effect on the stratospheric circulation than that resulting from just one of these two mechanisms. Given that the 11-year SC signals are found to be opposite at high latitudes, it may be speculated that the high-frequency responses represent a combined effect of enhanced solar UV, which causes warming near the equatorial stratosphere, and solar wind-driven processes, which cause cooling at high-latitude stratosphere. This may also explain the static high-frequency temperature response in the lower stratosphere (∼70 hPa) which seems to be independent of those originating in the upper stratosphere (see Figures 3 and 4). Nevertheless, it remains to be understood what mechanism has caused the cooling effect in the high-latitude stratosphere in relation to solar wind activity. Recent studies have suggested that the EPP-NOx and its downward descent during winter and spring play an insignificant role in stratospheric ozone and circulation (Lu et al., 2008a; Salmi et al., 2011). Lu et al. (2008a) suggested that changes observed in stratospheric winds and temperatures were unlikely to have been caused by photochemistry associated with EPP-NOx and stratospheric ozone, but were more likely due to an indirect dynamical link, e.g. changes in wave activity. They found that the spring temperature and wind variations in relation to changes of geomagnetic Ap index have a sign that is opposite to that expected from the NOx–ozone photochemistry mechanism.

The modulating effect of the QBO has not been explicitly studied here owing to the limited data records. Nevertheless, our analysis indicates that the QBO may add noticeable contamination to the signals in the temperature near the equatorial stratosphere at 20–30 hPa if no high- and low-frequency separation is made for the solar signature. This agrees well with the findings of Smith and Matthes (2008). Here, we also found that separating the response into high- and low-frequency components not only allows isolation of the fast and slow responses but also partially eliminates the QBO contamination at those pressure levels. However, unlike the Northern Hemisphere, where a significant solar signature can only be detected when the data are subgrouped according to the QBO phases (e.g. Labitzke and van Loon, 1988; Lu et al., 2009), it is not essential to separate the data according to the QBO phases in order to obtain a significant 11-year SC signal in the Southern Hemisphere.

The low-frequency responses (>36 months) are found in the lower stratosphere temperature extending from the North Pole to the South Pole, with the signal becoming weaker at mid latitudes. These low-frequency responses could be related to slowly varying transported ozone anomalies as suggested by Austin etal. (2008) and Gray et al. (2009). They may also be partially associated with a downward movement of ozone–UV interaction in the upper to middle stratosphere due to a change in the BD circulation, and such a process could account for up to 1 K per solar unit in the observed increase in temperature near the equatorial lower stratosphere (see Figure 4(a), (c) and (e)). Other slow processes such as thermal inertia of the oceans may also play a role (White et al., 2003; Austin et al., 2008; White and Liu, 2008).

In the troposphere, we found that the zonally averaged solar signals are generally more fragmented than those in the stratosphere. At high frequency, there are some indications of downward propagation of solar signals from the stratosphere into the troposphere at high latitudes (see Figures 4 and 7). However, the most significant signal in temperature is a positive response at 30–50° and this can only be obtained when the long-term trend is removed (see Figure 2) and is consistent with previous studies (van Loon and Labitzke, 1998; Gleisner and Thejll, 2003; Crooks and Gray, 2005; Haigh et al., 2005; Lu et al., 2007; Frame and Gray, 2010). A significant signal in tropospheric zonal winds was found only in the low-frequency domain. The signal is characterized by a weakening of the subtropical jet and the effect is symmetric about the Equator, consistent with Haigh etal. (2005). The magnitude and spatial structure of the signal in tropospheric zonal wind (∼1–2 m s−1) is in general agreement with the model simulations of Haigh et al. (2005) and Simpson et al. (2009), where a positive Equator-to-pole temperature gradient was applied in the lower stratosphere. Nevertheless, we have noted that the positive signal in the tropical lower stratosphere/troposphere is much weaker in ECMWF data than those imposed in the model simulations.

It has also been suggested that higher solar forcing may cause stronger upward motion in the winter hemispheric Subtropics and enhanced downward motion in the summer hemispheric Subtropics (van Loon et al., 2004; Meehl et al., 2008). This response is associated with the ‘bottom-up’ mechanism that is not symmetric about the Equator and is longitudinally varying and confined mostly to the Pacific Ocean (Meehl et al., 2003; van Loon et al., 2007). As our analysis here has mainly focused on the zonal pattern of the signals, our results cannot be directly compared to those reported by Meehl etal. (2003, 2008). Using a stratosphere–troposphere coupled GCM, Meehl etal. (2009) show that both ‘top-down’ and ‘bottom-up’ mechanisms may work together to induce positive feedbacks in the ocean–atmosphere system that may amplify the response to solar irradiance variations. Further studies are needed to understand how the stratospherically originating changes may amplify the longitudinal variations associated with air–sea interaction in the tropical troposphere.

Linear regression between the 11-year SC and the SAM shows that the solar signal is marked by a strengthening of the stratospheric SAM during winter and a weakening of the SAM in the uppermost stratosphere during spring. However, an 11-year solar cycle influence in the stratosphere as a cause of change in the tropospheric SAM cannot be established directly or linearly for the extended period of 1957–2009; other modulating factors, e.g. the QBO, must be taken into account to reveal any significant solar effect on the tropospheric SAM. This is consistent with the results of Roscoe and Haigh (2007) based upon multi-regression analysis, where they also used station-based Marshall-SAM and found that the 11-year SC signal is not significant. However, a QBO modulation of the polar vortex may itself be modified by the 11-year cycle of solar activity (Labitzke, 2003; Salby and Callaghan, 2006) and this combined solar–QBO (SQBO) effect has been detected in the SAM by using a composite SQBO index for the period of 1958–2006 (Roscoe and Haigh, 2007). They showed that SQBO was able to capture more variance in the SAM than the QBO and the 11-year SC alone, and suggested that the effect of the 11-year SC on the tropospheric SAM is nonlinear and is modulated by the QBO. Indeed, the modulation effects of the QBO and the 11-year SC on the SAM have been studied by Kuroda and Kodera (2005), Kuroda and Shibata (2006) and Kuroda and Yamazaki (2010). As our primary goal here is to study whether or not there is a direct projection of the 11-year SC on the SAM and also as we are limited by the amount of data available, we have not studied the modulation effects either by the 11-year SC or by the QBO on the SAM. Therefore, our results are not directly comparable to either the signal of SQBO shown by Roscoe and Haigh (2007) or the QBO or solar-modulated SAM patterns studied by Kuroda and co-authors. In addition to aliasing effects of the QBO, long-term trend, volcanoes and the 2002 SSW, the signals could also be affected by El Niño–Southern Oscillation, which may be also related to nonlinear effects (e.g. Marsh and Garcia, 2007; Calvo et al., 2009). Again, owing to the limited data available, those effects were not studied here.

To check whether or not the signals presented here may have been affected by merging two different datasets together, we have also performed the same analysis by excluding the ECMWF Operational data from our analysis and have found that the results remain qualitatively similar. For the high-frequency response, both temperature and zonal wind responses in fact become larger in magnitude, though the significant regions remain similar due to a smaller sample size. For the low-frequency response, the spatial pattern of the response remains similar but the regions covered by the 95% confidence levels reduce. We also performed the same analysis by including data back to 1968, as Kuroda and Yamazaki (2010) had done. Again the spatial patterns of the solar signals are generally similar, though the magnitude of the high-frequency response is reduced significantly. Thus the results presented here should be verified in the future, when longer datasets become available. This applies in particular to the upper stratosphere, where the impacts of discontinuities are more marked, and in the troposphere, where the magnitude of the response is small and comparable to the uncertainty range associated with the ECWMF Reanalysis data (Uppala et al., 2005).

5. Conclusions

We have studied the characterization of 11-year SC signals in the SH atmosphere during the winter and spring using ECMWF daily and monthly data from 1979 to 2009. By separating the response into high (<6 months) and low (>36 months) frequency domains, we have found:

  • 1.The 11-year SC signal is generally more robust in temperature than in zonal wind.
  • 2.Spatially different 11-year SC signals exist at high (<6 months) and low (>36 months) frequency domains. In the stratosphere, the high- and low-frequency responses tend to enhance each other near the Equator and Subtropics, while they oppose one another at high latitudes.
  • 3.The high-frequency response in the stratosphere is marked by an enhanced latitudinal temperature gradient in HS years from the Equator to pole and a strengthened stratospheric jet during winter. The response moves poleward and downward from the upper stratosphere to the lower stratosphere by tracking the movement of the stratospheric jet, as noted by Kodera and Kuroda (2002).
  • 4.In the lower stratosphere, the positive response of temperature to the 11-year SC is dominated by its low-frequency component.
  • 5.In the stratosphere, the magnitude and the statistical significance of temperature and zonal wind responses to the 11-year SC are sensitive to the contamination of the 2002 SSW event and major volcanic eruptions, but the general spatial pattern of the responses remains similar with or without these data.
  • 6.In the troposphere, the responses in the high- and low-frequency domains are of opposite signs but only the low-frequency response with warmer temperatures in HS years is significant (Figure 2). The low-frequency latitudinal temperature gradients and the resulting zonal wind anomalies are consistent with the results of Haigh et al. (2005).
  • 7.In the troposphere, the solar signature in temperature is more sensitive to the long-term trend than to the SSW or the major volcanic eruptions. Linearly removing the trend enhances the signal in the troposphere and weakens the signature in the stratosphere.
  • 8.A significant projection of the 11-year SC onto the SAM can only be detected in the stratosphere and in the high-frequency component. The signal is marked by a strengthening of the stratospheric SAM during winter and a weakening of the SAM in the uppermost stratosphere during spring.

Acknowledgements

HL and MJJ were supported by the UK Natural Environment Research Council (NERC). LJG was supported by the UK NERC National Centre for Atmospheric Science (NCAS). MPB was funded by NSF under the US CLIVAR program and the Office of Polar Programs. Finally, we thank two anonymous reviewers for their detailed and constructive comments on the original version of the manuscript.

Ancillary