We report Rayleigh lidar measurements of nightly temperature profiles in the 40–80 km altitude region and 15 min relative density profiles in the 40–50 km altitude region at Poker Flat Research Range, Chatanika, Alaska (65°N, 147°W), in December, January, and February over three winters (2002–2003, 2003–2004, and 2004–2005). We characterize the gravity wave activity in terms of the measurements of buoyancy period and relative density fluctuations and estimate the potential energy density and growth of potential energy density with altitude. We use satellite and reanalysis data to analyze the synoptic structure of the stratospheric vortex and the Aleutian anticyclone, the planetary wave activity, and the mean winds. These three winters have a major stratospheric warming in 2002–2003, an extreme warming event in 2003–2004 resulting in an elevated stratopause, and no warming in 2004–2005. The gravity wave activity shows significant interannual variability, with an average potential energy density of 2.1 J/kg in 2002–2003, 1.1 J/kg in 2003–2004, and 5.7 J/kg in 2004–2005. We find a positive correlation of 0.74 between the gravity wave activity in the upper stratosphere and the winds in the lower stratosphere where the winds are lightest. The reduction in gravity wave activity in 2002–2003 relative to 2004–2005 reflects the influence of the Aleutian anticyclone that is present over Chatanika in 2002–2003 but absent in 2004–2005. The occurrence of lower gravity wave activity in 2003–2004, when the Aleutian anticyclone is present less often than in 2002–2003, supports recent modeling studies that indicate that the elevated stratopause is formed due to a reduction of gravity waves propagating upward into the mesosphere.
 The general circulation of the middle atmosphere includes two wave-driven circulation patterns: the planetary wave-driven equator-to-pole circulation (often referred to as Brewer-Dobson circulation after Brewer  and Dobson ) in the lower stratosphere and the gravity wave-driven pole-to-pole circulation in the mesosphere [Houghton, 1978]. These circulations have a variety of impacts on the atmosphere, including the transport of critical species (e.g., ozone, water vapor) in the middle atmosphere, the lifetime of minor species (e.g., chlorofluorocarbons) in the stratosphere, the cooling of the polar summer mesopause region, and the warming of the polar winter stratopause region (see reviews by Holton et al. , Solomon , Holton and Alexander , Fritts and Alexander , and references therein). Changes in the Arctic wintertime circulation with warming of the stratosphere, cooling of the mesosphere, and breakdown of the polar vortex have been documented and understood in terms of planetary wave activity since the 1970s [Matsuno, 1971; Labitzke, 1972]. The breakdown of the polar vortex during these stratospheric warming events inhibits the formation of polar stratospheric clouds and the subsequent depletion of ozone (see review by Schoeberl and Hartmann ). Stratospheric warmings have attracted attention in recent years due to an increase in frequency and strength, with disruptions of the circulation in the winters of 2003–2004 and 2005–2006 and little disruption in 2004–2005 [Manney et al., 2005, 2006, 2008]. Rex et al.  have documented the extensive ozone loss in the Arctic middle atmosphere during the winter of 2004–2005 when the Artic vortex was not disturbed. Recent studies indicate that increases in greenhouse gases could increase the tropospheric wave forcing and strengthen the planetary wave activity in the stratosphere [Butchart et al., 2006; Deckert and Dameris, 2008].
 Two recent studies of transport of NOx from the upper mesosphere/lower thermosphere to the stratosphere and the subsequent interaction of NOx with stratospheric ozone have raised questions about the relative role of planetary and gravity waves in the polar circulation [Randall et al., 2006]. Hauchecorne et al.  and Siskind et al.  have studied the downward transport of NOx in the wintertime Arctic middle atmosphere associated with sudden stratospheric warmings in 2004 and 2006, respectively. During these winters, the disruption of the circulation resulted in a cooling of the upper stratosphere and warming of the mesosphere and the formation of an elevated stratopause. Hauchecorne et al.  used satellite observations in January–February 2004 to study a large increase of NO2 in the Arctic polar mesosphere with simultaneous depletion of O3. Hauchecorne et al.  attributed this enhancement of NOx to the blocking of planetary waves that allowed the propagation and breaking of gravity waves and tidal waves in the middle and upper mesosphere and enhanced the descent of NOx into the polar vortex. Siskind et al.  used satellite observations and a general circulation model to study tracer descent in February 2006 and addressed the observations of Randall et al. . Siskind et al.  concluded that the elevated stratopause was coupled to a disturbed stratosphere owing to the occurrence of stratospheric warming that suppressed vertical propagation of orographic gravity waves. Subsequent vertical propagation and breaking of a planetary wave 1 in the mesosphere facilitated the downward decent of NOx. Both Hauchecorne et al.  and Siskind et al.  emphasized the coupled role of planetary and gravity waves in the variability of the Arctic wintertime stratosphere and mesosphere but presented opposing views of the dynamical interaction. While both studies identified gravity waves as a key physical process in these events, neither study presented direct observations of the gravity wave activity.
 Rayleigh lidar studies of the Arctic middle atmosphere by Gerrard et al.  have shown that the synoptic scale movement of the stratospheric vortex influenced the stratospheric and lower mesospheric temperatures over three sites (i.e., Kangerlussuaq, Greenland (67°N, 51°W); Eureka, Canada (80°N, 86°W); and Andoya, Norway (69°N, 16°E)) located in the eastern Arctic where the vortex was present and the Aleutian anticyclone was absent. Lidar studies of gravity wave activity over Eureka, Canada [Whiteway and Carswell, 1994; Duck et al., 1998; Duck et al., 2001], have shown a reduction of gravity wave activity in the upper stratosphere associated with movement of the stratospheric vortex overhead. However, these lidar studies were conducted during winters when there were no major stratospheric warmings (i.e., 1992–1993 through 1997–1998).
 In this paper we present Rayleigh lidar measurements of gravity wave activity in the upper stratosphere and lower mesosphere during the winters of 2002–2003, 2003–2004, 2004–2005 over Poker Flat Research Range (PFRR), Chatanika, Alaska (65°N, 147°W). The location of Chatanika provides an opportunity to study gravity wave variability under the influence of both the Aleutian anticyclone and the stratospheric vortex using direct measurements. Dunkerton and Butchart  have shown by ray tracing studies that gravity wave propagation is modulated by the stratospheric vortex and Aleutian anticyclone with the lighter winds in the anticyclone blocking the upward transmission of orographic gravity waves. This paper is arranged as follows. In section 2 we describe the Rayleigh lidar technique and the methods used to determine and characterize the gravity wave activity. In section 3 we present lidar measurements of the temperature profiles and the gravity waves in December, January, and February (DJF) of 2002–2003, 2003–2004, and 2004–2005. In section 4 we present the synoptic structure of the Arctic middle atmosphere during these winters using meteorological global analyses data. We discuss the evolution of the circulation over the whole Arctic as well as over Chatanika. In section 5 we analyze the observed gravity wave activity in terms of the synoptic structure and horizontal winds. In section 6 we present our summary and conclusions.
2. Rayleigh Lidar Technique
 The National Institute of Information and Communications Technology (NICT) Rayleigh lidar has been operated at PFRR, Chatanika, Alaska, since 1997 [Mizutani et al., 2000; Thurairajah et al., 2009; Thurairajah, 2009]. The NICT Rayleigh lidar is a zenith pointing system and consists of a Nd:YAG laser, a 0.6 m receiving telescope, and a photomultiplier based photon-counting receiver system. The laser operates at 532 nm with a pulse repetition rate of 20 pps. The laser pulse width is 7 ns Full Width Half Maximum, and the average laser power is 9 W. The lidar signal is integrated over 0.5 μs yielding a 75 m range sampling resolution. The raw signal profiles are acquired every 50 s representing the integrated echo from 1000 laser pulses. The raw signal profiles are then integrated in time to yield 15 min profiles for analysis of wave-driven density fluctuations and over the whole observation period to yield a nightly temperature profile. In this study we use 29 nights of lidar measurements from DJF of 2002–2003, 2003–2004, and 2004–2005. These measurements last between 4 and 14 h for a total of 255 h of observations.
 The lidar observations yield measurements of the stratospheric and mesospheric temperature profile (∼40–80 km) under the assumptions of hydrostatic equilibrium and an initial temperature at the upper altitude of 80 km. This initial temperature, from the Stratospheric Processes And their Role in Climate (SPARC) reference atlas [Randel et al., 2004; SPARC, 2002], contributes to 100% of the temperature estimate at 80 km, 21% of the temperature estimate at 70 km, 4% at 60 km, and 1% at 50 km. The uncertainty in the temperature due to the initial temperature estimate decreases with decreasing altitude. The Rayleigh lidar technique and inversion method are described in detail as given by Thurairajah . The mean temperature profile is also used to determine the buoyancy frequency for the observation period and thus characterize the atmospheric stability. We characterize the gravity wave activity in the upper stratosphere and lower mesosphere in terms of the relative density fluctuation, ρ′/ρo, from the 15 min resolution Rayleigh lidar data. We use the relative density fluctuation rather than the temperature fluctuations to characterize the wave activity to avoid biases in the fluctuations due to the constant initial temperature at 80 km. To estimate the gravity activity, we first calculate the logarithm of the nightly average density profile. A third-order polynomial is fitted to this profile, which is then subtracted from the logarithmic average density profile. The residual is filtered by a low-pass filter of wavelength of 6 km to remove photon noise and then added back to the third-order fitted density profile. The antilog of the resulting density profile forms the background density, ρo(z). The background density is subtracted from the 15 min density profiles (obtained by binning the raw photon count profiles acquired every 50s) to form the perturbation density, ρ′(z, t). The perturbation is then normalized by dividing it by the background density profile to obtain the relative density perturbation, ρ′(z, t)/ρo(z). The mean-square vertical displacement () is derived from the mean-square relative density fluctuation using the gravity wave polarization relations [e.g., Gill, 1982],
where g is the acceleration due to gravity, and N is the buoyancy frequency defined as,
and where To is the mean temperature profile, and Cp is the specific heat at constant pressure. The potential energy density (Ep) is given by [e.g., Gill, 1982],
 The photon counting process is statistical in nature, and thus, the recorded profiles include both the statistical fluctuations due to the measurement technique as well as fluctuations due to geophysical variations. These statistical fluctuations are an inherent source of uncertainty in the measurement. To reduce the statistical uncertainty in the measurement and characterize short period waves the perturbations are spatially band limited between vertical wave numbers 0.5 km−1 and the required vertical altitude range and temporally band limited by the Nyquist frequency of 2 h−1 and the low frequency 0.25 h−1. We determine the variance of the statistical fluctuations from the average vertical wave number spectrum of the perturbations. We estimate the spectrum using the periodogram method [Koopmans, 1974]. We subtract the variance of the statistical fluctuations from the total variance to yield an estimate of the variance of the relative density fluctuations, ( [Wang, 2003]. The signal-to-noise ratio (SNR) is then calculated as the ratio of the variance of the relative density fluctuations to the uncertainty in the estimate of the variance of the statistical fluctuations. The Rayleigh lidar measurements are made under clear sky conditions, and lidar signal levels are relatively constant. Thus, the statistical variance is relatively constant while the variance of the relative density fluctuations, and hence the SNR, represents the variations in the wave activity. In this study the values of the SNR range between 1.3 and 62.4 with an average value of 14.7.
 We plot two examples of the relative density perturbations measured with the lidar as contour plots in altitude and time in Figure 1. In Figure 1 (top), we show relative density perturbations derived from lidar measurements taken over a ∼11 h period (2014–0649 LST (LST = UT − 9 h)) on the night of 15–16 January 2004. In Figure 1 (bottom), we show relative density perturbations derived from lidar measurements taken over a ∼14 h period (1820–0820 LST) on the night of 10–11 January 2005. The relative density perturbations show periodic variations with downward phase progression typical of upwardly propagating gravity waves reported in other lidar studies [e.g., Wilson et al., 1991]. The different tilt of the relative density perturbations on 15–16 January 2004 and 10–11 January 2005 reflects differences in the vertical wavelength of the dominant waves. The relative density perturbations on 15–16 January 2004 have a dominant vertical wavelength of 45 km and time period of 2.5 h. The relative density perturbations on 10–11 January 2005 have a dominant vertical wavelength of 14 km and period of 2.9 h.
 We present the buoyancy period and wave parameters for these observations in Table 1. To provide a reference for these values, we compare the values of the potential energy density (1.6 J/kg on 15–16 January 2004 and 13.3 J/kg on 10–11 January 2005) with the extensive Rayleigh lidar measurements of potential energy density in the 30–45 km and 45–60 km altitude ranges at the midlatitude sites of Observatoire de Haute Provence (OHP, 44°N, 6°E) and the Centre d'Essais des Landes at Biscarosse (BIS, 44°N, 1°W) in France [Wilson et al., 1991]. To reduce the noise in the measurements at Chatanika, we smooth the logarithm of the raw data profiles with a running average over 2.0 km before computing the temperature and relative density profiles. We have compared the value of the mean-square density fluctuations for smoothed and unsmoothed data. We find that this smoothing reduces the mean-square density fluctuations by a factor of 1.7. Thus, for comparison between the lidar measurements at Chatanika and that reported by Wilson et al. , we first take the geometric mean of the measurements at 30–45 and 45–60 km to yield an equivalent measurement at 45 km and then divide these values by a factor of 1.7. The compensated monthly mean for January at OHP and BIS are 10 J/kg and 7 J/kg, respectively. Thus, we see that these two measurements at Chatanika span the monthly mean values reported from midlatitudes. We will compare the measurements from Chatanika with the measurements reported by Wilson et al.  in more detail later in the paper.
Table 1. Buoyancy Period and Gravity Wave Activity at 40–50 km at Chatanika, Alaska (65°N, 147°W)
15–16 January 2004
10–11 January 2005
Calculated from nightly average temperature profile.
 We plot the individual and monthly mean profiles of 14 nighttime measurements for January 2003, 2004, and 2005 in Figure 2 (left). These measurements have yielded a total of 122 h of data. We average the nighttime profiles for each month to form the monthly mean profile. We also compare (Figure 2, right) the monthly mean profile for January 2003, 2004, and 2005 to the monthly mean January profile calculated from 22 nights of lidar measurements over Chatanika from 1998 to 2005 [Thurairajah et al., 2009] and to the zonal mean temperature climatology from the SPARC reference atlas. The Rayleigh lidar measurements at Chatanika show a high degree of interannual variability in the monthly mean temperature profile over these 3 years. In January 2003, the upper stratosphere and mesosphere were colder than the Chatanika and SPARC averages with the stratopause located at 54.0 km with a temperature of 237.3 K. In January 2004, the upper stratosphere was much colder and the mesosphere was much warmer than both the Chatanika and SPARC averages, and the stratopause was vertically displaced to 70.3 km with a temperature of 245.0 K. In January 2005, the upper stratosphere was warmer than the Chatanika and SPARC averages with the stratopause located at 47.5 km with temperature of 261.4 K.
 In Figure 3, we plot buoyancy period averaged over the 40–50 km altitude range as a function of day during DJF of 2002–2003, 2003–2004, and 2004–2005. On average, the buoyancy period was lower in the winter of 2003–2004 than in 2002–2003 and 2004–2005 (Table 2). In 2003–2004, the background atmosphere was less stable with higher buoyancy period during the first half of 2003–2004 winter (320 s (±5 s)) and more stable with lower buoyancy period after 15 January 2004 (273 s ((±4 s)). The increase in stability in 2003–2004 coincided with the formation of the elevated stratopause. In 2002–2003 and 2004–2005, the stability decreased through the winter.
Table 2. Buoyancy Period and Gravity Wave Activity at 40–50 km at Chatanika, Alaska (65°N, 147°W), in December, January, and February
Calculated from nightly average temperature profile.
Mean value and uncertainty in mean (i.e., standard error).
 In Figure 4, we plot the gravity wave activity in terms of rms relative density fluctuation, rms vertical displacement fluctuation, and potential energy density averaged over 40–50 km as a function of day during DJF in the three winters. We tabulate the average values for DJF in Table 2. The gravity wave fluctuations had lower rms relative density, vertical displacement, and potential energy density in 2003–2004 compared to the other two winters. The average potential energy density in 2004–2005 of 5.7 J/kg was 5.2 times larger than in 2003–2004 (1.1 J/kg) and 2.7 times larger than in 2002–2003 (2.1 J/kg). The interannual differences were more pronounced in the root mean square (rms) displacements and the potential energies than the rms relative density fluctuations due to the interannual differences in the atmospheric stability (equation (2)).
 In Figure 5, we plot the rms relative density fluctuation as a function of buoyancy period averaged over 40–50 km for all three winters. We calculate linear fits to the data for the entire data set as well as by year. The correlation coefficient for all 29 nights is 0.21, with a value of 0.24 for the 2002–2003 data (15 nights), 0.17 for the 2003–2004 data (8 nights), and 0.69 for the 2004–2005 data (6 nights). There is no obvious variation of rms relative density fluctuation with buoyancy period except in 2004–2005, where the rms amplitude of the gravity wave fluctuations decreased as buoyancy period increased (i.e., the stability decreases). Examination of the variation of rms vertical displacement and potential energy density with buoyancy period shows similar behavior albeit with lower correlation coefficients. This behavior suggests that the amplitude of the gravity wave fluctuations does not increase as the stability decreases, and we conclude that the measured fluctuations represent gravity waves that are propagating through the 40–50 km altitude region rather than being generated by the local atmospheric stability conditions. The decrease of the rms amplitude of the gravity wave fluctuations with increase in buoyancy period in 2004–2005, when the gravity wave amplitudes were largest, is consistent with saturation of the gravity wave amplitudes due to convective and dynamic instabilities [Fritts and Rastogi, 1985]. The limiting wave amplitudes are inversely proportional to the buoyancy period and thus decrease as the buoyancy period increases [Smith et al., 1987]. Thus, in 2004–2005, the gravity wave amplitudes appear large enough to have generated internal instabilities, while in 2002–2003 and 2004–2005, they are not large enough to do so.
 To better understand how the waves are propagating with altitude, we calculate the vertical growth length, or scale height, of the potential energy. We calculate the growth length from the ratio of the potential energy densities over the 45–50 km to the 40–45 km altitude regions. In 2002–2003, the average ratio of the nightly energy densities at both altitudes was 2.0 (±0.4) corresponding to a growth length of 7.2 km. In 2003–2004, the average ratio was 1.1 (±0.3) corresponding to a growth length of 52 km. In 2004–2005, the ratio was 1.7 (±0.4) indicating a growth length of 9.4 km. We expect that freely propagating waves would have a growth length in their energy equal to the density scale height of 7.0 km that corresponds to a ratio of 2.0. Thus, the gravity waves in 2002–2003 appear to have been propagating freely, while the waves in 2003–2004 and 2004–2005 appear to be have been loosing energy as they propagated upward.
 In summary, the Rayleigh lidar observations at Chatanika show that the middle atmosphere thermal structure and gravity wave activity during the winter of 2003–2004 was different from the winters of 2002–2003 and 2004–2005. The temperature profile had an elevated stratopause during January 2004 that was not observed in 2003 and 2005. The upper stratosphere (40–50 km) was more stable in January and February 2004 relative to 2003 and 2005, with shorter buoyancy periods. The gravity wave activity was reduced in 2003–2004 relative to the other winters with lower rms relative density fluctuations, lower rms vertical displacements, and lower potential energy densities. The gravity waves in 2003–2004 had less growth with altitude (i.e., longest growth length) than in the two other winters indicating stronger dissipation of the waves with altitude during that winter.
4. Arctic Planetary Wave Activity and Synoptic Structure
 In this section we analyze the synoptic structure in the Arctic stratosphere and mesosphere during the three winters. Our goal is to understand how the interannual variations in the gravity wave activity are related to the variations in the synoptic structure in the different winters. To characterize the synoptic structure of the middle atmosphere, we analyze the 3-D structure and temporal evolution of the stratospheric vortex and anticyclones calculated using the United Kingdom Meteorological Office (MetO) global analyses data. Harvey et al.  presented a methodology to identify vortices in terms of evolving 3-D air masses. The calculations are done on 22 potential temperature surfaces from 240 K (∼6 km, 800 hPa) to 2000 K (∼48 km, 0.6 hPa). The vortex edges are identified by integrating a scalar measure of the relative contribution of strain and rotation in the wind field around the scale stream function that characterizes the large-scale flow. The analysis of Harvey et al.  provides a full view of the 3-D structure of the vortex and anticyclones from which we can follow the evolution of the vortex and anticyclones through the winter. This analysis allows us identify vortex anticyclone interactions, vortex displacement events, and vortex splitting events. We can also determine the synoptic conditions over Chatanika (i.e., below the vortex, below the Aleutian anticyclone, or neither). We also calculate the wind speeds from MetO analyses data and define the wind speed as the magnitude of the horizontal wind, by combining the zonal, u, and meridional, v, wind ().
 We characterize the planetary wave activity using geopotential heights measured by the Sounding of the Atmosphere using Broadband Emission Radiometry (SABER) instrument data (Level 2A version 1.07) (K. Beaumont, SABER: Sounding of the atmosphere using broadband emission radiometry, The TIMED Mission Data Center, John Hopkins University Applied Physics Laboratory, Laurel, Maryland, 2008, available at http://saber.gats-inc.com/) aboard the Thermosphere Ionosphere Mesosphere Energetics Dynamics (TIMED) satellite [Mertens et al., 2004; Russell et al., 1999]. Because of the satellite's 2-month yaw cycle, wintertime measurements are available for a 2-month period from mid-January through mid-March. We use standard Fourier techniques to determine the wave 1 and wave 2 components in the geopotential along latitude circles for a given day of satellite observations similar to the method of Riggin et al. . We characterize the amplitude of the planetary wave as the magnitude of the corresponding Fourier component of the geopotential. While the Fourier analysis yields the amplitudes of the planetary waves, it does not yield a measure of the planetary wave breaking or whether the planetary waves disrupts the vortex and causes mixing of midlatitude and polar air. Harvey et al.  showed that during planetary wave breaking events the vortex and anticyclone become tilted in altitude, intertwine, and mix. If the stratospheric vortex is displaced from the pole, there will be a large Fourier wave 1 component indicating a large amplitude wave 1 planetary wave, but the vortex may remain intact. Similarly, a wave that breaks and disrupts the vortex in two would contribute to a large wave 2 component indicating a large amplitude wave 2 planetary wave [Shepherd, 2000]. Charlton and Polvani  have recently analyzed the planetary wave activity associated with displacement and splitting of the vortex. We also determine the gradient winds from the SABER data and hence calculate the Eliassen-Palm flux divergence using established techniques [e.g., Andrews et al., 1987].
4.1. Pan Arctic Perspective
 The 2002–2003 winter was characterized by a cold early winter and warm mid winter to late winter [Singleton et al., 2005]. By mid-December, a robust vortex had formed (Figure 6a). The lack of planetary wave activity in mid-December is evidenced by the barotropic structure of the vortex where the vortex does not tilt with height (Figure 6a). A minor warming occurred in late December 2002 in the upper stratosphere while the lower stratospheric temperatures remained undisturbed [Manney et al., 2005]. During the first half of January 2003, the Aleutian anticyclone intensified, moved eastward, and by the January 13 had developed a westward tilt with height (not shown) consistent with increase in planetary wave activity. At the same time, the anticyclone elongated and displaced the vortex off the pole. By January 18, a major stratospheric warming was in progress and the anticyclone had split the vortex in the lower and the middle stratosphere (Figure 6b and Figure 7, top). The anticyclonic winds found in the Aleutian anticyclone were stronger than the cyclonic winds in the vortex during this splitting event. During this major warming, the planetary wave 1 and wave 2 geopotential amplitude increased and the Eliassen-Palm flux divergence at 65°N decreased and had negative values. During the minor warming in mid-February, when the upper stratospheric vortex had again strengthened and entwined around the Aleutian anticyclone and the mid and lower stratospheric vortices were distorted (Figure 6c), there was an increase of planetary wave 1 and wave 2 amplitudes and the Eliassen-Palm flux divergence at 65°N decreased and had negative values.
 The 2003–2004 winter was characterized by one of the most prolonged midwinter warming event on record [Manney et al., 2005]. By mid-December, a robust vortex had formed and the stratospheric vortex and Aleutian anticyclone remained quasi-stationary (Figure 6d). A major warming with strong disruption of the vortex occurred in early January 2004 when the upper stratospheric vortex had entwined around the Aleutian anticyclone while the mid and lower stratospheric vortices were distorted and had small areas (Figure 6e and Figure 7, middle). The winds in the anticyclone were much weaker and the winds in the vortex are stronger than during the 2003 vortex splitting event. By mid-January, the upper stratospheric vortex had reformed into a stronger cyclone. The lower and middle atmospheric vortices remained small and disrupted through late February 2004 (Figure 6f). This disruption of the polar vortex was accompanied by repeated periods of strong planetary wave 1 and wave 2 amplitudes in January, February, and early March. There were five periods of large wave 1 amplitudes in late January and in mid and late February. There were several periods of large wave 2 amplitudes in mid and late January and mid and late February. The Eliassen-Palm flux divergence at 65°N was negative between scaled heights of 5 and 9 (∼35 and ∼63 km) throughout the period from mid-January through early March 2004.
 The 2004–2005 winter was the coldest winter in the lower stratosphere on record at that time, and no warming events satisfied the zonal mean definition of reversed meridional temperature gradients and/or easterly flow at midlatitudes [Manney et al., 2006]. During December 2004, the vortex in the upper stratosphere remained warm and strengthened while in the lower and midstratosphere the vortex was cold. The vortex remained robust and well established throughout the winter (Figures 6g, 6h, and 6i). The area of the vortex was larger, the Aleutian anticyclone was displaced further equatorward, and the vortex winds were stronger and more uniform than in the previous two winters (Figure 7, bottom). There were four periods of large wave 1 amplitudes in February. There are several periods of large wave 2 amplitudes in mid-January and February. However, despite these repeated periods of large amplitude planetary wave 1 in February and wave 2 in mid-January and February, where the upper stratospheric vortex had entwined with the Aleutian anticyclone, the vortex remained robust and well established in the mid and lower stratosphere (Figures 6g, 6h, and 6i). The Eliassen-Palm flux divergence at 65°N between scaled heights of 5 and 9 (∼35 and ∼63 km) is negative in only two periods in late January and from the middle to end of February.
 In all three winters, we find tilting and intertwining of the vortex and anticyclone, peaks in the planetary wave amplitudes, and negative Eliassen-Palm flux divergence that characterize periods of planetary wave breaking. Our analysis of the synoptic structure and the planetary wave activity of the wintertime Arctic middle atmosphere also shows that during winters when the amplitudes of the planetary waves are similar, the planetary wave action and resultant synoptic structure can be markedly different.
4.2. Chatanika Perspective
 In Figure 8, we plot the temporal evolution of the stratospheric vortex (in green) and Aleutian anticyclone (in red) over Chatanika, Alaska, over the altitude range of ∼14–48 km (isentropic surfaces from 400 to 2000 K) for DJF of the three successive winters. These three winters show distinctive differences in the temporal evolution of the stratospheric vortex and Aleutian anticyclone over Chatanika. The planetary wave activity is evident as the vortex and Aleutian anticyclone cross over this site.
 During the 2002–2003 winter, the vortex appeared above 1000 K (∼33 km, 5 hPa) while the anticyclone appeared at lower altitudes and was present throughout the winter. During the January 2003, major stratospheric warming the Aleutian anticyclone extended to (at least) 2000 K (∼48 km, 0.6 hPa). During the 2003–2004 winter, the Aleutian anticyclone was over Chatanika during the second half of December reflecting the vortex and anticyclone are quasi-stationary in December. During the major stratospheric warming in January 2004, there was a transition to a regime where the vortex remained overhead above 800 K (∼30 km, 10 hPa), and the anticyclone appeared occasionally at lower altitudes. During the 2004–2005 winter, the large vortex extended over Chatanika for all of December and January. The appearance of the anticyclone overhead in the second half of February coincided with the increase in planetary wave activity during that period. In summary, the synoptic structure in 2002–2003 showed more variability over Chatanika, with the repeated movement of the vortex and anticyclone over the site, while in 2003–2004 and 2004–2005, there is less variability.
 We plot the horizontal wind and the associated cumulative distribution functions (CDFs) at the 500 K (∼19 km, ∼60 hPa), 800 K (∼30 km, ∼10 hPa), and 1600 K (∼44 km, ∼1 hPa) isentropic surfaces above Chatanika in Figure 9. The interannual variations in the wind speeds are clearly evident with significantly higher winds in 2004–2005. The CDFs highlight the difference in the winds in each year. The CDFs show that the median winds at 500 K were 8.7, 11.5, and 36.3 m/s in 2002–2003, 2003–2004, and 2004–2005, respectively. The median winds at 800 K were 11.0, 22.0, and 59.2 m/s in 2002–2003, 2003–2004, and 2004–2005, respectively. The median winds at 1600 K were 38.5, 40.2, and 62.2 m/s in 2002–2003, 2003–2004, and 2004–2005, respectively. In 2002–2003, the reduction in wind coincided with the presence of the Aleutian anticyclone except during the period of the major stratospheric warming in the second half of January when there was an increase in the winds during the period when the Aleutian anticyclone extended up to 2000 K. On 18 January 2003 at 800 K, the Aleutian anticyclone was over Alaska with stronger winds compared to the weak winds inside the split vortex (Figure 7) and the horizontal wind over Chatanika inside the anticyclone was 43 m/s. In 2003–2004, the decrease in the winds in the second half of December coincided with the appearance of the Aleutian anticyclone. The anticyclone appeared first at the upper altitudes, and the winds weaken first at these upper altitudes. On 5 January 2004, the Aleutian anticyclone at 800 K was over the western Arctic with weak winds (Figure 7) and the horizontal wind over Chatanika inside this anticyclone was 11 m/s. In 2004–2005, the winds remained strong from early December through early February as the vortex remained overhead (Figure 7) and the horizontal wind inside the vortex over Chatanika on 18 January 2005 was 59 m/s. The decrease in wind speeds in late February coincides with the appearance of the Aleutian anticyclone over Chatanika.
 The monthly mean wind profiles in all 3 years show that the winds increased in the troposphere with a local maximum near the tropopause (∼320 K), decreased in the lower stratosphere, and then increased through the stratosphere. The winds in 2004–2005 were clearly highest while the winds in 2003–2004 were lowest. The interannual variability in the vortex and anticyclone positions shown in Figure 8 was reflected in the interannual variability in monthly mean wind profiles. During the winter of 2002–2003, there was considerable month-to-month variability in the winds above 500 K. During the winter of 2003–2004, the wind profiles for January and February showed similar winds up to 800 K. During the winter of 2004–2005, the wind profiles for December and January showed similar winds up to 900 K. The disruption of the stratosphere in 2003–2004 was evident in that the expected upper stratosphere jet (i.e., wind maximum between 1000 K (∼33 km, 5 hPa) and 1800 K (∼45 km, ∼0.8 hPa) observed in 2002–2003 and 2004–2005) was not observed. At 500 K, the average winter wind speed in 2003–2004 was 10 m/s, a factor 1.2 and 3.3 times less than the average winds in 2002–2003 and 2004–2005, respectively. The interannual variation is greatest at 800 K where the average winter wind speed in 2003–2004 was 15 m/s, a factor 1.9 and 3.9 times less than the average winds in 2002–2003 and 2004–2005, respectively. This interannual variation reduced at 1600 K where the average winter wind speed in 2003–2004 was 43 m/s, similar to the value of the wind speed in 2002–2003 of 42 m/s and a factor of 1.4 times less than the average winds in 2004–2005. The weakest winds were found in January 2004, and the interannual differences were most pronounced in January when the monthly average wind speed at 500 K was 6 m/s, a factor 1.9 and 5.9 times less than the average winds in January 2003 and 2005, respectively. At 800 K, the difference is again more pronounced when the average wind speed for January was 9 m/s, a factor 3.5 and 7.5 times less than the average winds in January 2003 and 2005, respectively. The difference is less pronounced at 1600 K when the average wind speed for January 2004 was 28 m/s, a factor 1.2 and 2.5 times less than the average winds in January 2003 and 2005, respectively.
5. Variability of Gravity Wave Activity and Synoptic Structure
 To understand the relationship between the synoptic structure of the middle atmosphere and the gravity wave activity at Chatanika, we calculate the linear correlation between the potential energy density of the gravity waves at 40–50 km and the horizontal wind speeds at each of the 22 altitudes. For the combined set of 28 gravity wave measurements and MetO winds over the three winters, we find that the correlation coefficient was greater than 0.6 between 300 K (∼6 km, 420 hPa) and 700 K (∼27 km, 16 hPa). The maximum correlation of 0.74 is found at 400 K (∼14 km, ∼130 hPa) (Figure 10). This suggests that about 50% (r2 = 0.55) of the day-to-day variability of the gravity wave activity in the upper stratosphere is related to the variation of the mean wind speed in the lower stratosphere. This correlation of 0.74 at 400 K is similar to the value of 0.73 found by Wilson and colleagues from Rayleigh lidar measurements of gravity waves in the 30–45 km altitude range and winds at 50 hPa (∼20 km) at BIS and OHP [Wilson et al., 1991]. Our analysis indicates that the gravity wave activity in the upper stratosphere is modulated by the horizontal wind speed in the lower stratosphere with larger potential energies associated with larger wind speeds. The fact that the correlations are highest with the winds in the regions where the wind speeds are lowest indicates that the physical process underlying the modulation is critical layer filtering of low-frequency waves with low horizontal phase speeds in the altitude regions where the winds are weakest.
 The interannual variation in the energy growth lengths can also be understood in terms of the local winds at 1600 K (Figure 9, bottom) and internal wave instabilities. In 2002–2003, the gravity waves appear to have been growing freely with altitude in the 40–50 km altitude region and were neither limited by critical layer filtering in weak winds (average of 40 m/s on the nights of the lidar measurements) nor internal instabilities in the waves themselves. In 2003–2004, the reduced growth of the waves with altitude in the 40–50 km altitude region arose from continued critical layer filtering of waves in the weaker winds (average of 32 m/s on the nights of the lidar measurements) at these upper altitudes. In 2004–2005, the large amplitude gravity waves propagating in higher winds (average of 54 m/s on the nights of the lidar measurements) are limited by internal convective and dynamic instabilities and have growth lengths (9.4 km) that are shorter than in 2003–2004 (52 km) but longer than in 2002–2003 (7.2 km). Thus, the weak winds associated with the disruption of the circulation in 2003–2004 served to reduce the amount of wave energy within the upper stratosphere and lower mesosphere in two ways; weak winds in the lower stratosphere blocked the propagation of gravity waves from below, while weak winds in the upper stratosphere and lower mesosphere dissipated the waves at those altitudes.
 The interannual variations in the gravity wave activity (Figure 4) are also consistent with the observed variations in the monthly mean temperature profiles at Chatanika (Figure 2). In 2002–2003, the stratosphere had a typical structure and was colder than the multiyear averages with intermediate gravity wave activity. In 2003–2004, the stratosphere had an elevated stratopause and the stratosphere was colder than the multiyear averages with reduced gravity wave activity relative to 2002–2003. In 2004–2005, the stratosphere had a typical structure and was warmer than the multiyear averages with increased gravity wave activity relative to 2002–2003. These observations are consistent with the work of Kanzawa  and Hitchman et al.  who concluded that the heating of the polar stratopause resulted from adiabatic heating in the descending gravity wave-driven circulation of the winter pole. This adiabatic heating was greater than the radiative cooling. Thus, we conclude that, in 2004–2005, more abundant gravity waves in the upper stratosphere propagated upward into the mesosphere, breaking at mesospheric heights, and driving a stronger meridional circulation. The stronger meridional circulation had stronger descent at the winter pole and stronger adiabatic warming of the stratopause. In 2002–2003, less abundant gravity waves in the upper stratosphere propagated upward into the mesosphere, breaking at mesospheric heights, and driving a weaker meridional circulation that yielded a weaker descent at the winter pole and weaker adiabatic heating. In 2003–2004, the gravity waves were suppressed further, the disruption of the meridional circulation was greater, and because of the reduced adiabatic heating, the thermal structure of the stratosphere was dominated by radiative cooling and an elevated stratopause was formed.
 This physical scenario is the same as that presented in the study of Siskind et al.  of the elevated stratopause in February 2006. Siskind et al.  used a general circulation model to show that a major stratospheric warming resulted in the disruption of the stratospheric circulation, subsequent suppression of the gravity wave-driven circulation, and hence the formation of the elevated stratopause. Siskind et al.  modeled the gravity wave activity as an orographic gravity wave drag parameterization which can be turned on and off in the model. In their analysis of the Arctic middle atmosphere in February 2006, the two sets of model results where the model internally suppresses the gravity waves and where the gravity wave drag is turned off in the model both yield the observed elevated stratopause and the authors conclude that the elevated stratopause resulted from the suppression of orographic gravity waves.
6. Summary and Conclusion
 We have characterized the gravity wave density fluctuations in the upper stratosphere (40–50 km) from Rayleigh lidar observations at Chatanika over three winters (DJF 2002–2003, 2003–2004, 2004–2005). These winters have significant differences in their meteorological conditions which we have characterized using analysis of MetO and SABER data. Over all three winters (DJF), we find that the average gravity wave potential energy density (2.6 J/kg) was lower than that measured at midlatitude sites (6–11 J/kg) [Wilson et al., 1991]. The day-to-day variation in the gravity wave energy densities was positively correlated with the weakest winds in the lower stratosphere. The lidar measurements of low levels of wave activity that appears modulated by the mean winds is consistent with the study of Dunkerton and Butchart  and the recent study of Wang and Alexander . Dunkerton and Butchart  showed that synoptic conditions in the western Arctic, where winds were lighter, blocked the upward transmission of gravity waves. Wang and Alexander  reported that there was significant geographic variability in the gravity wave activity across the Arctic in winter 2007–2008, with higher amplitudes found near the vortex edge (as also reported by Duck et al. ) where winds were stronger and lower amplitudes in the North Pacific and Alaska where the winds were weaker.
 We find that there was also interannual variability in the vertical growth length of the gravity wave potential energy density. In 2002–2003, the waves propagated freely with altitude. In 2003–2004, the growth of the waves was limited by critical layer interactions with the wind. In 2004–2005, the growth of the gravity waves was limited by internal instabilities generated in the waves themselves.
 In each of the three winters, the gravity wave activity and temperature structure of the stratosphere and mesosphere were consistent in terms of the gravity wave-driven circulation and formation of the polar stratopause. In 2004–2005, when no stratospheric warming events occurred, the gravity wave activity was highest and the stratopause temperatures were higher than expected. In 2002–2003, when there were several stratospheric warming events, the gravity wave activity was lower and the stratopause temperatures were also lower than expected. In 2003–2004, when one of the most prolonged midwinter stratospheric warming events occurred, the gravity wave activity was lowest of all 3 years, the stratosphere was coldest, and the stratopause was displaced vertically to ∼70 km. These lidar observations provide direct evidence of the suppression of gravity wave activity during an elevated stratopause event and observationally support the recent modeling study by Siskind et al.  that attributed the formation of the elevated stratopause to the suppression of gravity wave activity, the reduction of adiabatic heating, and radiative cooling of the stratosphere and lower mesosphere.
 The authors thank the staff at Poker Flat Research Range for their ongoing support of the lidar program. The authors thank the following University of Alaska students for their assistance in making the ongoing lidar observations; S. Nadakuditi, M. Peshave, T. Stern, L. Su, W. Wang, and J. Yue. The authors thank K. Sakanoi of Komazawa University for her assistance with the lidar measurements in spring 2003. The authors acknowledge the SABER science and data processing teams for providing the SABER data presented in the paper. The authors acknowledge support from the United States National Science Foundation under grants ARC-0632387 and ATM-0640340. PFRR is a rocket range operated by Geophysical Institute-University of Alaska Fairbanks with support from the United States National Aeronautic and Space Administration. The authors thank three anonymous reviewers for their valuable comments.