Beijing National Observatory of Space Environment, Institute of Geology and Geophysics, Chinese Academy of Sciences, Beijing, China
Graduate University, Chinese Academy of Sciences, Beijing, China
Corresponding author: J. Xiong, Beijing National Observatory of Space Environment, Institute of Geology and Geophysics, Chinese Academy of Sciences, 19 Beitucheng Western Rd., Chaoyang District, 100029, Beijing, China. (email@example.com)
 From 8 years' SABER/TIMED temperature profiles between January 2002 and December 2009, we studied the activity of gravity waves in the stratosphere globally. Global distribution of stratospheric gravity wave potential energy was calculated from the temperature perturbations. Seasonal comparison of gravity wave potential energy Ep shows an annual variation in middle and high latitudes and a semiannual variation in the tropics. Around the equator, gravity wave interannual enhancements are identified just below the zonal wind zero (u = 0) contours corresponding to descending eastward shear phase of the QBO. Furthermore, we provide observation evidence to support the conclusion that the deep convection is a major source for the observed tropical gravity wave activity. The considerable longitude variations of largest potential energy around the equator are related not only to the specific topography and tropical convections but also to many other factors. We can infer that topography and tropical deep convection are the important sources of the gravity waves in the stratosphere, but the observed gravity waves in the tropical/subtropical stratosphere are strongly affected by winds with different QBO phases.
 Atmospheric gravity wave (GW) activity has been one of the intense research subjects in recent years because it plays a crucial role in determining the circulation, structure, and variability of the atmosphere. GWs generated from tropospheric sources propagate upward and their amplitudes increase exponentially with decreasing air density. The waves break, deposit energy and momentum, and decelerate or accelerate the background winds after reaching the saturation amplitudes [e.g., Andrews et al., 1987; Fritts and Alexander, 2003]. In situ, ground-based, and space-based observations, together with theoretical and numerical studies, have remarkably improved our understanding of GWs recently. It is widely agreed that GWs play an important role in driving the global-scale general circulation of the middle atmosphere and lower thermosphere. For example, the breaking GWs are the major driving force of the quasi-biennial oscillation (QBO) in tropical lower stratosphere [Dunkerton, 1997; Ern and Preusse, 2009; Kawatani et al., 2010]. By studying the primary characteristics of GW activity in the troposphere and lower stratosphere where the waves are mainly excited, we may obtain a more comprehensive and detailed understanding of the global atmosphere dynamics.
 A long-term period observation is very significant for GWs study. In recent years, people have studied the distribution of GWs in the troposphere and lower stratosphere from long-term in situ and ground-based observations.Moffat-Griffin et al.  examined gravity wave activity in the lower stratosphere over Antarctica by calculating potential and kinetic energy from 8 years of radiosonde data from Rothera (67°S, 68°W). The gravity wave energy was shown to have a seasonal variation with peaks at the equinoxes; the largest peak was around the spring equinox in the lower stratosphere. Furthermore, comprehensive information about the activity of GWs at midlatitudes was extracted from more than 230 nights of temperature measurements with two lidars between 2002 and 2006 by Rauthe et al. . The seasonal variability of vertical wavelengths, amplitudes, and potential energies and a comparison between the monthly mean temperature fluctuations and temperatures to study the transition between the seasons and the different scales of variability have been shown in their research.
 Climatology from global observations is essential to understand the role that GWs play in atmosphere circulation. However, the in situ and ground-based data is limited to specific regions or a certain latitude band. Fortunately, satellite observation can provide a global morphology. Basing on different observation data and research methods, a lot of analyses have been performed on global GW distribution by now.Tsuda et al.  evaluated the global wave potential energy in the stratosphere from radio occultation data retrieved from the Global Positioning System/Meteorology (GPS/MET) experiment. Their results showed that enhanced potential energy regions were around the equator with considerable longitude variations as well as at middle latitudes in the winter hemisphere. On the basis of the global wave potential energy retrieved from Challenging Minisatellite Payload (CHAMP) satellite, de la Torre et al. found that a general weaker (stronger) wave activity is associated to vertical wavelengths shorter (longer) than 4 km. Also, the tropical/extra tropical signatures decreased/increased with increasing altitude. In addition, by using temperature data retrieved from radio occultation measured by CHAMP and Satelite de Aplicaciones Cientificas-C (SAC-C),Fröhlich et al.  made an estimation of the GW momentum flux. The distributions of the momentum flux agreed well with simulations using the Warner and McIntyre parameterization scheme not only in global distribution but also in absolute values. The most prominent feature was the strong GW activity around the subtropics.
 Decades of observation and theoretical studies suggest that the GWs in lower atmosphere are primarily controlled by their excitation sources and wind filtering. It implies that the wave characteristics may be an indicator of the wave excitation source characteristics. For instance, aircraft observations have revealed that the generation of small-scale disturbances in the tropical stratosphere was closely related to deep convection, and their primary behavior was consistent with a theoretical GWs model [Alexander and Pfister, 1995]. As we have known, the main GW sources include topography, convection, and wind shear. There are probably many other important sources such as adjustment of unbalanced flows in the vicinity of jet streams and frontal systems [Fritts and Alexander, 2003].
 The basic aim of the present paper is to study the climatological features of global GW distribution in the stratosphere by using an 8-year continuous data set of temperatures retrieved from the SABER/TIMED. The multiyear time series, which is much longer than previous observations, provides the opportunity to make a systematic survey of long-term variations in global GW activity. SABER data has been used byKrebsbach and Preusse  and Ern et al.  to study gravity waves. Using SABER/TIMED temperature data from 29 January 2002 to 31 January 2006, Krebsbach and Preusse  presented a spectral analysis of time series of weekly zonal root mean square GW amplitudes for systematic intraannual, annual, and interannual GW activity, while Ern et al.  presented an investigation of absolute values of GW momentum flux derived from global temperature measurements in the stratosphere and the whole mesosphere. We will use a different method to analyze 8 years' SABER data in order to statistically identify the dominant wave sources, their spatial and temporal variability, and the general characteristics of GWs arising from them.
 In section 2 a short description of the observational data and the calculation method of relevant GW parameters is presented. Subsequently, we describe the data analysis procedure for extracting potential energies Ep from temperature profiles. The seasonal mean global morphologies of stratospheric GW activity are shown in section 3. For the purpose of profound understanding of the underlying processes, we make comparisons between the global distributions of GW potential energy and topography, winds, and convection, respectively. In section 4, a discussion of the results and their possible implications is shown. Our summary and conclusions are provided in section 5.
2. Data and Method
2.1. Data Analyses
 The data we used here are temperatures observed by the Sounding of the Atmosphere using Broadband Emission Radiometry (SABER) instrument on board the NASA Thermosphere-Ionosphere-Mesosphere-Energetics and Dynamics (TIMED) spacecraft. The TIMED satellite was launched on 7 December 2001 and the SABER instrument began to make observations in January 2002. By using limb-scanning, broadband infrared radiometry, SABER can measure temperature, carbon dioxide, nitric oxide, ozone, water vapor, and so on over a broad range of altitude from near the tropopause to the lower thermosphere (approximately 20 km to 120 km) [Garcia et al., 2005; Remsberg et al., 2008]. These continuous measurements are very suitable for studying the long-term variability and the global spatial structure of GWs. The data presented here are based on Version 1.07, which is publicly available. We use “level 2A” retrievals of temperature in this study, which are consisting of vertical profiles registered in pressure as functions of latitude, longitude, and universal time (more information available online athttp://saber.gats-inc.com.). The viewing direction of the instrument changes every 60 days so that the latitude coverage on a given day extends from about 53° in one hemisphere to 83° in the other. We use temperature data from late January 2002 to December 2009, almost 8 full years, in order to study the temporal variability and global spatial structure of GW monthly mean potential energy. The long-term continuous data record allows for statistical identification of systematic intraannual (seasonal), annual, and interannual structures of GW activity.
 Following the theory of GW energy, the wave energy per unit mass (energy density) E0 is chosen as a measure of GW activity. It is defined as
where Ek and Ep are kinetic and potential energy per unit mass, respectively, which are
 Here u′, v′, and w′ correspond to the zonal, meridional, and vertical perturbation components of wind velocity, respectively. and T′ are the background atmosphere temperature and temperature perturbation; g is the acceleration due to gravity; Nis the Brunt-Väisälä frequency. The Brunt-Väisälä frequencyN is calculated from the filtered background temperature in order to eliminate the influence of the waves on N itself.
 As the linear theory of GW predicted, the ratio of Ek to Ep is a constant under the condition that atmospheric GW intrinsic frequency ωis much smaller than the Brunt-Väisälä frequencyN and sufficiently larger than the inertial frequency f (i.e., f2 ≪ ω2 ≪ N2) [VanZandt, 1985]. Therefore we can assume the total gravity wave energy to be proportional to potential energy, which can be calculated from temperature variance alone. It should be noted that the Ep may be smaller than the Ek and consequently Ep may not adequately present the full GW energy if the atmospheric GW intrinsic frequency ω is close to the inertial frequency f. For example, this case may occur at middle and high latitudes because the inertial frequencies at middle and high latitudes are considerably larger than those at low latitudes [Xiao and Hu, 2010].
 A very few of the vertical profiles which have negative vertical gradient of potential temperature in the lower stratosphere are rejected in the data set. In this study, we consider the height range between 20 and 45 km in which the method of separating the GWs from the global-scale atmospheric background can be applied dependably [Tsuda et al., 2000].
 The observations from January 2002 to December 2009 are used to derive temperature profiles and subsequently temperature disturbances. Each temperature profile is assumed to be the sum of the background and perturbation of atmospheric temperature. In the first place, each temperature profile is interpolated into altitude intervals of 1 km by cubic spline interpolation. Then in order to obtain the background temperature structure, we use a least squares cubic polynomial fitting method. By subtracting the fitted background temperature profile from the original profile, we get the temperature residuals. These temperature residuals are then vertically high-pass filtered using a cutoff wavelength of 10 km in order to extract the fluctuationsT′ with vertical wavelengths ranging from 2 to 10 km. T′2 was integrated using a sliding window with a width of 2 km and a step size of 1 km. The potential energy Ep of GWs can be evaluated from equation (3) with these fluctuations. We have eliminated some profiles (about 2.5% of total data), when T′2 exceeded 15 K2 [Tsuda et al., 2000].
 The Ep values are binned into grid cells of size 10° longitude × 5° latitude for each month of measurements. We define a monthly mean potential energy Ep value by averaging all available data in the surrounding area of a specific region or in a certain latitude band.
2.2. About Kelvin Waves
 Global-scale waves often have longer vertical wavelengths than GWs so that short vertical wavelength GWs can be separated from global-scale waves by high-pass filtering. However, some global equatorial waves also have short vertical wavelengths (mainly slow Kelvin waves) are likely to be included in our study since the data is not filtered by their zonal wave numbers.
 Slower Kelvin waves have a narrow amplitude distribution. On the basis of Andrews et al. , the width of the amplitude distribution in degree latitude is
where ΓE is the rotation period of the Earth, i.e., 1 day; Γ is the period of the Kelvin wave; and K is the zonal wave number, i.e., 1, 2, .... The width of the intensity distribution σI is half of the amplitude distribution, i.e., σI = σφ/2. We can determine Γ directly from the dispersion relation
where ΓN is the buoyancy period, λX and λZ are the horizontal and vertical wavelengths, respectively, and RE is the radius of the Earth (∼6371 km). Substituting this into equation (4), we find that σφ depends solely on λZ and ΓN. But ΓN is almost constant in the stratosphere (about 5 min). On the basis of our vertical wavelength filtering we can assume λZ = 6 km to be a representative wavelength. This results in Γ = 23 days for wave number 1 and a width of σI = 4.2° latitude.
 In addition, basing on theoretical considerations, Eckermann  studied the meridional extent of equatorial waves from rocketsonde data and reported that the wave energy in the 2–20 km vertical scale was dominated by equatorial waves in the 20–40 km height range. However, in the 2–10 km wavelengths band, equatorial waves were not obviously prevalent; instead, it was suggested that GWs significantly contribute the wave energy.
 On the basis of the analysis above, we may draw a conclusion that in the tropics potential energies represent a superposition of GWs and Kelvin waves. Owing to the vertical wavelength filter, only slow Kelvin waves are contained in this analysis, restricting their influence to a narrow band of approximately 4.2° latitude around the equator. While the equatorial maximum persistent during all seasons is likely due to Kelvin waves, tropical and subtropical maxima at some distance from the equator and in particular their longitudinal structure are reliable.
3.1. Seasonal Gravity Wave Activity
 In order to investigate the behavior of GWs during a year, variations of Epin different seasons of the 8-year average are shown inFigure 1. Apart from some gaps due to the cyclical changes of viewing direction of the instrument, the observations are well distributed throughout the years and allow us to systematically study the seasonal behavior of GW activity.
 From Figure 1 we can find an annual variation of the stratospheric GW activity at middle and high latitudes with stronger Ep values occurring in the winter hemispheres and weaker values in the summer hemispheres. The most significant feature is a semiannual variation in the tropics with the strong GW occurring in northern winter and summer (Figures 1a and 1c).
Figure 1a shows that large Ep values in November–January are generally centered round the equator with considerable longitude variations. Large values are observed over South America, Africa, and the Pacific, which may be related to the specific topography and convection. Moreover, at middle and high latitudes, the Ep values over the continents are larger than those over the oceans, which might be caused by topography and geostrophic adjustment.
 In the low latitudes, the latitude range of the enhanced Ep in November–January is wider than in other seasons. This feature has also been mentioned in the study of GPS/MET experiment data [Tsuda et al., 2000]. They considered that the Ep values in the 15–25 km region could have been affected by the temperature structure near the tropopause because the tropopause was generally located at 16–17 km in the tropics. However, in our study the height range of major variations is above the tropopause. Therefore the large Ep values are not related to the tropopause structure but due to the stronger wave activities at low latitudes.
 Different from the distribution in November–January (Figure 1a), three obvious regions appear over the eastern Pacific, Africa, and India in the low latitudes in May–July (Figure 1c). In addition, larger Ep values in the Southern Hemisphere appear at the edge of the polar vortex, decreasing toward the north. The enhanced Ep corresponds to the steep descend east of the Antarctic plateau, the Antarctic Peninsula [e.g., Wu and Jiang, 2002; Baumgaertner and McDonald, 2007; Moffat-Griffin et al., 2011], and the southern tip of South America at high latitudes. The relation between enhanced Ep and geographic location indicates that topography is a strong source for GW activity. The mountain waves with small phase velocities will shift to longer vertical wavelengths in the strong background wind and reach larger amplitudes before saturation. In addition, the jet in this area is also likely to be one of the major causes for this enhancement. These results are consistent with a scenario whereby the stratospheric gravity wavefield over Rothera (situated on the Antarctic Peninsula) is determined by a combination of wind flow over topography, generating waves from below, and sources such as the edge of the polar stratospheric vortex, generating waves from above [Moffat-Griffin et al., 2011].
 Moreover, during equinoxes, larger Ep values dominate around the tropics latitudes. As discussed in section 2.2, Kelvin waves restrict their influence to a narrow band of 4.2° latitude around the equator. The width is very well compatible with the stripe at the equator observed in Figure 1 (i.e., all seasons). This is a clear indication that Kelvin waves are the main contribution to this equatorial stripe. In addition, in February–April (Figure 1b) the latitudinal distribution of Ep is symmetric in amplitudes and structure between the two hemispheres; however, for August–October (Figure 1d) the Southern Hemisphere activity dominates especially with the distinct enhance Ep appearing over the southern tip of South America.
 Compared with the previous studies, Figure 1 shows some interesting additional features. For instance, the strong activity appears around the South Pole especially over the Drake Passage in May–July is with much higher values than those in the previous studies [e.g., Fröhlich et al., 2007; Wu and Eckermann, 2008; Moffat-Griffin et al., 2011]. The vertical wave wavelengths in our study are limited to shorter than 10 km, so the dominant waves are slow GWs (refer to the equation in section 4.1) which tend to propagate far downstream. The topography-generated waves of Tierra del Fuego and the Antarctic Peninsula, with southwest and northwest pointing wave vectors, respectively, are likely to form the maximum east of the Drake Passage. This process is suggested by the particular triangular shape, the onset of GW activity slightly downstream of the mountain ridges, and the drawn-out tail.
3.2. Relation of Potential Energy Ep and Wind
3.2.1. Spectral Analysis
 In order to identify the periodic properties of GW activity in the tropics, we perform a spectral analysis. Monthly zonal means of GW potential energy Ep amplitudes are calculated and then merged to a time series covering 8 years. The time series is analyzed by applying a Fast Fourier Transformation (FFT). Figure 2 shows the spectrum of the performed FFT for vertical wavelengths between 2 km and 10 km for 10°S–10°N latitude and 21–26 km altitude. The significant spectral components with much larger spectral amplitudes are marked. Their frequencies are 0.5, 1, and 2 year−1 respectively, corresponding to biennial, annual, and semiannual periods.
 Our results are similar to the spectrum for vertical wavelengths between 5 km and 30 km in the latitude band 30.0°S ± 2.5°and altitude range 60.0 ± 1.0 km shown by Krebsbach and Preusse . This comparison is intended as a general hint for the reliability of our method. In the following section, we focus on the period of 2 years corresponding to the significant magnitude of almost 2.1 J/kg which occur in the equatorial stratosphere.
 By applying the spectral analysis, we make a further investigation to other latitudes about the dominant modes of variability consisting of annual, semiannual, and quasi-biennial oscillations (SAO and QBO).
 As shown in Figure 3, the annual cycle consists in each latitude bands. At middle and high latitudes, the annual cycle is by far the largest variation, with stronger values occurring in the winter hemispheres and weaker values in the summer hemispheres (as shown in Figure 1), while SAO and QBO are mainly observed in the low latitude band (centered on the equator). In the tropics, there are Kelvin, Mixed Rossby, and other equatorial waves. These waves carry energy and momentum from the troposphere and impart eastward and westward accelerations to the stratospheric atmosphere. The large period atmospheric oscillations such as QBO and SAO are generated from these wave-wave and wave-mean flow interactions. Consequently, the dynamics of the low-latitude atmosphere is characteristically different from those of midlatitude and high-latitudes atmosphere.
 It is noteworthy that the QBO and SAO patterns are not completely symmetric in each hemisphere with somewhat larger amplitude values in the Northern Hemisphere than those in the Southern Hemisphere. These features may be due to more continental regions as well as more convectively triggered waves in this hemisphere.
3.2.2. Potential Energy Ep and QBO
 In the tropical stratosphere, interannual variability of zonal winds and temperatures is dominated by a quasi 2 years periodicity termed the quasi-biennial oscillation (QBO). Characteristics of the QBO in tropical zonal winds and temperatures have been first derived from long time series of rawinsonde observations byAngell and Korshover .
 We analyze tropical space-time character of the GW and QBO using output of the United Kingdom Meteorological Office (UKMO) stratospheric assimilation. The UKMO stratospheric analyses used here are output of a data assimilation system. This system uses a global numerical model of the atmosphere, with fields continuously adjusted toward available wind and temperature observations as the model is integrated forward in time [Randel et al., 1999]. Output of these analyses includes three-dimensional winds whose latitudinal resolution is 2.5°, with 22 levels spaced uniformly in log pressure from 1000 to 0.316 mb (∼0–57 km). The time period analyzed here covers January 2002–December 2009. The UKMO data has been proved to be reliable for identifying tropical zonal wind QBO variations as shown bySwinbank and O'Neill .
 The monthly zonal mean averaged Ep between 10°S and 10°N from 21 km to 34 km is plotted in Figure 4b. The equivalent monthly mean zonal wind from the UKMO averaged is shown for comparison in Figure 4a.
 There is remarkable agreement between Ep and zonal wind in Figure 4. Besides seasonal variations, we also find a QBO signal in Ep distribution in the altitude between 21 and 45 km. However, the signal is obvious under 34 km (Figure 4b) and becomes weaker with increasing altitude (not shown here).
 There are almost three complete QBO cycles in our time range. The downward movement of the u = 0 m/s phase line associated with westward QBO shear is marked with black bold line in Figure 4b, for example from 32 km in April 2002 to 21 km in March 2003. In the same way, the descending eastward QBO shear u = 0 m/s phase line is marked with red bold line.
 The enhanced GWs are identified just below the zonal wind zero (u = 0) contours corresponding to descending eastward shear phase of the QBO marked with the red bold line. To be more precise, the maximum Ep values descend with time closely following the evolution of the zero (red) line in zonal mean wind. This coincidence suggests an interaction between the vertically propagating waves and the winds around zero. Randel and Wu  interpreted the shear region acting as a wave sink. It can be summarized as two processes. First, when gravity waves propagate vertically in background shear flow, there is a functional relationship EP ∼ |u − c|−1 (c is wave phase speed). For a spectrum of tropical inertia gravity waves with phase speeds centered around c = 0, EP will increase as the u = 0 (critical) level is approached. EP will maximize below the level due to nonlinear effects and dissipation. Second, the vertical group velocity slow down (Cgz ∼ |u − c|2). For intermittent wave sources and “snapshot” observations, the probability of observing wave perturbations increases. In a word, the enhanced values near the critical line are attributed to the intrinsic growth of the perturbations and the increased probability of observing perturbations.
 Furthermore, the distribution of the enhanced Ep in these years is coincident with the westward phase of the QBO and the Ep values are considerably larger than during the eastward phase. From the distribution of zonal mean wind, westward winds are observed throughout the lower and middle troposphere (not shown in Figure 4a), which can filter westward propagating GWs at lower levels. As a result, eastward propagating GWs are assumed to be the main waves at higher altitudes. When the winds switch from westward to eastward in stratosphere, the eastward propagating GWs are filtered out partly and the Ep shows a decrease in the QBO eastward phase. For example, the Ep value at 24 km is 4 J/kg and at 27 km drops to 1.5 J/kg in January 2004.
 In order to further investigate the QBO effect, latitude-time section of the monthly zonal mean wind and the zonal meanEp from 30°S to 30°N are shown in Figure 5. The data is averaged between the height of 21 and 26 km. The distribution of GWs can be considerably changed in the propagating upward under the influence of background winds and other factors, so we choose a smaller altitude range than in Figure 4. Around the equator, GWs Ep interannual enhancements, related to QBO, are observed in Figure 5. Clear enhancements are identified in winters at the beginning of 2002, 2004, 2006, and 2008 around the equator just as shown in Figure 4. Because the SABER data was not available until 25 January 2002, the feature is not obvious in this year. Additionally, a significant weaker Ep is seen where the zonal wind is the strongest westward. Moreover, the tropical GW Epseems to stronger in July than in January in 2003, 2005, and 2007. These features may be due to the stronger westward zonal wind in July which is helpful for the convection-generated GWs to propagate upward into the stratosphere [Jiang et al., 2004].
3.3. Relation of Tropical Potential Energy Ep and Convection
 GWs in the tropics are mostly generated by deep convection. In most of the studies before, outgoing longwave radiation (OLR) data has been used as a proxy for deep convective activity. The NOAA Climate Prediction Center Merged Analysis of Precipitation (CMAP) index is a better indicator of deep convection compared with the traditional OLR data because it includes more small-scale details besides the common patterns found in OLR data [Jiang et al., 2005]. In order to estimate global precipitation amounts, the CMAP blends station rain gauge observations and five different types of satellite products [Huffman et al., 1997]. The CMAP data can be obtained from http://www.esrl.noaa.gov/psd/. The distributions of monthly mean CMAP indices during the 8 years are very similar, resulting in similar effects on GWs as a wave source in every year when we mainly consider convection generated waves. In order to study the relationship of tropical GWs with the convective activity, the 8-year mean seasonal mean precipitations derived from CMAP in different seasons are shown by white contours inFigure 6.
 The seasonal structure of equatorial Ep averaged between the height of 21 and 34 km which varies with latitude and longitude as well as the seasonal mean CMAP indices are illustrated in Figure 6. We can see that convection events in the tropics appear in every season but their distributions are quite different.
 It is noteworthy that the distribution of Ep values in May–July (Figure 6c) shows a strong correspondence with deep convection. As described in section 2 (Figure 1c), large Ep values mostly appear over the eastern Pacific, Africa, and India at the equatorial latitudes. Deep convection is generally observed close to the three regions.
 During November–January (Figure 6a), the deep convective regions are centered at south of the equator. The enhanced Ep at 30°S–5°N occurs approximately above the deep convective regions in Figure 6a. While the centers of enhanced Ep are shifted in latitude from the convection centers, the longitudinal structures are well matched.
 During August–October (Figure 6d), deep convection activity occurs above the Bay of Bengal, the northern Indonesian region, Western Africa, and the northern tip of South America, which not always coincide with enhanced Ep. In Figure 6b, the largest Ep is observed from 60 to 80°W and 70–140°E in February–April. There is only little indication for convectively generated GWs during equinoxes. This feature was also shown in the study of average horizontal distribution of GW momentum flux absolute values [Ern et al., 2011]. They stated that in these two seasons mountain waves were stronger sources.
 To summarize, Figure 6 indicates that the GWs at low latitudes are mostly generated by tropical deep convection. Convectively generated gravity waves might have a significant direct impact on the tropics, and they are considered to be an important source of the QBO of the tropical stratosphere [Chun and Baik, 2002].
 However, several potential energy enhancements are not coincident with strong convection in Figure 6. For instance, in November–January, the stripe of large Ep around 20°N cannot be explained by convection, and not all convection maxima around 20°S produce enhanced Ep. Similar discrepancies exist in the other figures. These discrepancies underline the existence of other effects enhancing GW activity except convection. This enhanced Ep may be interpreted by other sources such as wind shear and geostrophic adjustment. Nevertheless, these sources do not necessarily produce enhanced Ep, for the distribution of GWs can be changed by background winds and other factors in the propagation. We can get the conclusion that the considerable longitude variations of largest potential energy Ep values around the equator are mainly due to the topography and tropical convection, but many other effects also contribute to variations of Ep.
4.1. Comparison of Individual Years
 The behaviors of Ep in January of 2007 and 2008 are compared in Figures 7a and 7b in order to verify the conclusions above. As mention in section 2, the SABER yaw cycle limits the data poleward of ∼50° to half of the month. The data is averaged in the altitude range of 26–30 km. And the zonal winds are averaged in the altitude range of 23–27 km. Because we choose the altitude range in which the distribution of GW has been influenced by background winds in the propagating upward. Although under the similar convective conditions and the same season, the tropical Ep values are found to be much larger in 2008 than in 2007. The different QBO pattern and the induced wind filtering should be responsible for this remarkable difference. The overall average results of the 8 years may eliminate most of the influence of QBO. As a result, the rest of the structure is mainly related with convection and topography as shown in Figure 1. But in the interannual variability, the influence of the QBO is important.
 It had been pointed out that the strong winds would act as a directional filter [Yeh et al., 1972]. GWs propagating close to the direction of zonal wind may be reflected or trapped. Therefore only those waves propagating in the opposite direction of winds can reach higher altitudes.
 The zonal winds were westward around equator in the troposphere both in January 2007 and 2008, so then the predominant component of GWs propagated to stratosphere must be eastward. In 2007, the zonal wind in the stratosphere was eastward. Consequently, the eastward propagating GWs from troposphere would encounter critical levels and would be reflected or absorbed and blocked from entering the higher stratosphere. But in 2008, the zonal wind in the stratosphere was westward, so waves from below can propagate in the stratosphere. The opposite directions winds near equator have different effects on the gravity waves propagation. Generally, stratospheric GWs are stronger in years with QBO westward phases than in years with QBO eastward phases as shown in Figure 4.
 At middle and high latitudes, the longitudinal variations of the GW Ep are quite visible. Large Ep is observed just below the strongest zonal winds. The vertical wavelength of gravity waves is λz = 2π|c − Ucosθ|/N, where c is its horizontal phase speed, U is the background wind speed, θ is the angle between the wind vector and the direction of wave propagation, and Nis the background Brunt-Väisälä frequency. According to this equation, as the wind increases with height, GWs are refracted to long vertical wavelengths. At middle and high latitudes, the most important mountain waves with small phase velocities shift toward longer vertical wavelengths in the strong wind. The wave energy attenuation is less when the vertical wavelength becomes longer. Therefore the GW energy at middle and high latitudes below the strong winds becomes large as shown inFigure 7. However, in the north of 50°N region, the enhanced Ep has a much larger extension than the mountainous region. Those GWs related to the polar night wind jets are likely generated by other wave sources, such as geostrophic adjustment and jet instabilities [Ern et al., 2011].
4.2. Comparison of Individual Months
 The behaviors of Ep over south Asia and southeast Asia in July and August of 2009 are shown in Figure 8. The data is averaged in the altitude range of 26–30 km. Remarkably, the distribution of enhanced Ep, shows a strong eastward shift from 70°–100°E in July to 80°–130°E in August. Between July and August there is a shift of the Asian monsoon to the east. This coincident motion between enhanced Ep and monsoon indicates that monsoon is an importance influence factor for GW activity.
 The global morphology of gravity wave activity in the stratosphere has been presented as a function of latitude, longitude, season, altitude, and year. GW activity has been estimated by calculating potential energy, Ep, from temperature profiles obtained by the SABER/TIMED experiment from January 2002 to December 2009. The multiyear time series which cover a time longer than the previous studies enable us to identify the systematic intraannual (seasonal), annual, and interannual structures of GW activity. The global distribution of Ep data has been provided with topography, winds, and convection in order to understand the underlying processes. Main results are as following.
 1. At middle and high latitudes, seasonal variation of gravity wave activity shows an annual variation with stronger values occurring in the winter hemispheres and weaker values in the summer hemispheres. And there is a semiannual variation in the tropics with the strong GW occurring in November–January and May–July.
 2. The relationship between GW activity and the geographic location indicates that topography is one of the important sources for wave activity.
 3. The relation of monthly zonal mean gravity wave potential energy Ep and the zonal mean wind in these years has been studied. The enhanced GWs are identified just below the zonal wind zero (u = 0) contours corresponding to descending eastward shear phase of the QBO. This coincidence suggests an interaction between the vertically propagating waves and the zero wind line regions.
 4. Comparisons of stratospheric QBO and GW activities show that stratospheric GWs are stronger in years with QBO westward phases than years with QBO eastward phases. QBO winds act as a significant filter in gravity wave propagation.
 5. The distribution of Ep values shows a strong longitudinal correspondence with tropical deep convection. Our results suggest that enhancement Ep at low latitudes is partly generated by deep convection.
 From the above results, we can infer that topography and tropical deep convection are the important sources of the gravity waves in the stratosphere, but the gravity waves in the tropical/subtropical stratosphere are strongly affected by winds with different QBO phases. However, there are many other wave sources such as wind shear and geostrophic adjustment. And the distribution of GWs could be influenced by background winds and other factors in the propagation. We need to study all of these in more detail in the future.
 We are grateful to the SABER team for the access to the data on http://saber.gats-inc.com. The CMAP Precipitation data was provided by the NOAA/OAR/ESRL PSD, Boulder, Colorado, from their Web site at http://www.esrl.noaa.gov/psd/. UKMO data was obtained from the British Atmospheric Data Center. The research was supported by the National Natural Science Foundation of China (40974086) and the National Important Basic Research Project of China (2011CB811405).