TIMED Doppler Interferometer (TIDI) measurements of zonal and meridional winds in the mesosphere/lower thermosphere are analyzed for diurnal nonmigrating tides (June 2002 to June 2005). Climatologies of monthly mean amplitudes and phases for seven tidal components are presented at altitudes between 85 and 105 km and latitudes between 45°S and 45°N (westward propagating wave numbers 2, 3, and 4; the standing diurnal tide; and eastward propagating wave numbers 1, 2, and 3). The observed seasonal variations agree well with 1991–1994 UARS results at 95 km. Comparisons between the TIDI results and global scale wave model (GSWM) and thermosphere-ionosphere-mesosphere-electrodynamics general circulation model (TIME-GCM) tidal predictions indicate that the large eastward propagating wave number 3 amplitude is driven by tropical tropospheric latent heat release alone. In contrast, latent heating and planetary wave/migrating tidal interactions are equally important to westward 2 and standing diurnal tidal forcing. There is good quantitative agreement between TIDI and the model predictions during equinox, but the latter tend to underestimate the westward 2 and standing diurnal tide during solstice. Neither model reproduces the observed seasonal variations of the eastward propagating components.
 The upward propagation of atmospheric tides is one of the key processes in lower/upper atmosphere coupling, e.g., by transporting energy and momentum from the troposphere and stratosphere to upper altitudes. Tides are global-scale waves in temperature, winds, and density with periods that are harmonics of a solar day. They are among the most striking dynamical features in the mesosphere and lower thermosphere (MLT). Tidal winds are on the order of the time-averaged zonal wind and dominate the meridional wind field. Tides also modify the upward propagation of gravity waves (GWs) and their momentum deposition in the upper atmosphere via critical layer filtering mechanisms [e.g., Fritts and Vincent, 1987] and via modulation of the buoyancy frequency [e.g., Preusse et al., 2001]. They play a major role in the diurnal cycle of chemically active species [Coll and Forbes, 1998; Marsh and Russell, 2000; Zhang et al., 2001]. Temperature oscillations may change reaction rates with simultaneous transport of air parcels some 1000 km in the horizontal and some kilometers in the vertical [Ward, 1999]. As a result, tides have a major impact (up to 40% from the migrating diurnal tide alone) on the total heating in the MLT [Smith et al., 2003].
 The nonmigrating or non-Sun-synchronous tides are far less understood due to a limited number of observations, model deficiencies, and some ignorance of their role in upper atmosphere dynamics, chemistry, and energetics. Nonmigrating tides do not follow the apparent westward motion of the Sun but may propagate westward, eastward, or remain standing. Their zonal wave numbers do not equal their frequencies (in cycles per day) and they may be forced by a variety of quite different mechanisms. The aggregate effect of nonmigrating tides introduces a longitudinal variability of the amplitudes and phases of the total tidal fields [Khattatov et al., 1996; Ward et al., 1999; Manson et al., 1999].
 In this paper, only nonmigrating tides of diurnal frequency are considered. Nonmigrating tidal components (propagation direction/wave number pairs) are identified by using a letter/number combination indicating both propagation direction (w: westward, e: eastward, s: standing) and zonal wave number s ≥ 0; i.e. w2 is the westward propagating nonmigrating diurnal component of zonal wave number 2, s0 is the standing oscillation (having zonal wave number 0), and e3 is the eastward propagating diurnal component of zonal wave number 3. With the same nomenclature, the migrating diurnal tide is w1.
 The two leading nonmigrating tidal sources are latent heat release in the tropical troposphere [e.g., Hagan and Forbes, 2002, 2003] and nonlinear interactions between Quasi-Stationary Planetary Waves (QSPWs) and the migrating tide [Hagan and Roble, 2001; Lieberman et al., 2004]. Additional sources are longitudinal variations in the heating from ozone and water vapor due to land-sea differences and topography [e.g., Kato et al., 1982; Tsuda and Kato, 1989], and nonlinear interactions between the migrating tide and gravity waves [e.g., McLandress and Ward, 1994]. Oberheide and Gusev  suggested nonlinear interactions between latent heat forced nonmigrating tides and QSPWs as an additional mechanism. The role of absorption of solar radiation in forcing the nonmigrating tides has not yet been fully resolved. Model results of Hagan et al.  point to a rather small contribution, but revised radiative heating rates by Lieberman et al.  indicate that the radiative source may need to be revisited.
 The amount of data suitable for nonmigrating tidal analysis has so far been quite limited, owing to shortcomings of both satellite-borne and ground-based instruments. Satellite instruments suffer from their lack of local time coverage, but they provide the longitudinal coverage essential to extract the nonmigrating components from the tidal signal. The 30–60 day composites of data from instruments onboard slowly precessing satellites are required to provide a local time coverage of 24 hours that can be Fourier analyzed [e.g., Forbes et al., 2003; Oberheide et al., 2005]. Such satellite results must therefore be interpreted in a rather climatological sense. A recently developed non-Fourier method [Oberheide et al., 2002] allows tidal analysis on a daily basis, but the method critically depends on data quality and orbit geometry. It has been successfully applied to Cryogenic Infrared Spectrometers and Telescopes for the Atmosphere (CRISTA) and Limb Infrared Monitor of the Stratosphere (LIMS) data [Oberheide and Gusev, 2002; Lieberman et al., 2004]. Many ground-based instruments provide 24-hour local time coverage every day, but it is impossible to dealias nonmigrating and migrating tides from the observations. Networks of ground-based stations may help, but to avoid tidal aliasing, numerous stations must be equally spaced in longitude. Such a requirement is difficult to meet and basically limited to the analysis of semidiurnal components in high northern and southern latitudes [Wu et al., 2003; Murphy et al., 2003].
 The few global observations of nonmigrating tides that are available therefore originate from satellite instruments. Using LIMS data, Lieberman  showed the combined nonmigrating tidal amplitudes equal to or in excess of the migrating diurnal amplitude at altitudes below 80 km with vertical phase progression suggesting a tidal forcing below the stratosphere. These findings were confirmed by Oberheide and Gusev  and Oberheide et al.  who derived nonmigrating tides between 50 and 120 km altitude from CRISTA temperatures. The CRISTA results could be understood by a combination of both latent heat and QSPW/tidal interaction forcing with the sources of the observed large w3, w4, and w5 amplitudes remaining unknown. Lieberman et al.  revisited the LIMS data and found evidence for wave-wave interaction forcing in the stratosphere. Forbes et al. , Manson et al. [2002, 2004], and Huang and Reber  were able to retrieve nonmigrating tidal amplitudes and phases from the two wind instruments onboard the Upper Atmosphere Research Satellite (UARS): the High Resolution Doppler Imager (HRDI) and the Wind Imaging Interferometer (WINDII). Although the specifics of the UARS instruments confined their analyses to an altitude of 95 km, they clearly showed the month-to-month variability of the nonmigrating tides as well as the resulting longitudinal modulation of the diurnal tidal amplitude. With a different analysis approach, Talaat and Lieberman  retrieved nonmigrating tidal information between 60 and 120 km from HRDI but without being able to clearly identify the zonal wave numbers of the tidal components. It is interesting to note that the results of the UARS analyses are partly contradictory in terms of tidal amplitudes. Differences reach 50%, but in all these studies the nonmigrating tides partly exceeded the migrating tide. This emphasizes their important role in MLT dynamics. Recently, Forbes and Wu  established internal consistency between the nonmigrating tides in Microwave Limb Sounder (MLS) temperatures at 86 km and the UARS tidal winds at 95 km within the context of classical tidal theory.
 State-of-the-art tidal models [Hagan and Forbes, 2002, 2003; Grieger et al., 2004] as well as general circulation models like the thermosphere-ionosphere-mesosphere-electrodynamics general circulation model (TIME-GCM) [Hagan and Roble, 2001], the whole atmosphere chemistry climate model (WACCM) (R. Garcia, private communication, 2003), and the extended Canadian middle atmosphere model (CMAM) [Ward et al., 2005] can now reproduce nonmigrating tides but with a number of deficiencies in terms of predicted amplitudes, the presence of specific nonmigrating tidal components, and seasonal and interannual variability. Observation-based tidal definitions and climatologies are therefore needed to aid modelers in tuning source functions and dissipative parameters to match observations.
 With the launch of the thermosphere-ionosphere-mesosphere-energetics and dynamics (TIMED) satellite in 2001, data sets amenable to tidal analysis over a range of MLT altitudes are now becoming available. The orbit precession rate of TIMED requires that tidal parameters obtained using a Fourier method be accumulated over 60 days. The wind measuring instrument, the TIMED Doppler Interferometer (TIDI) [Killeen et al., 1999], has been operating since early 2002. TIDI is characterized by dual-sided viewing and high sensitivity to nighttime as well as daytime O2 emissions in the 75–105 km layer. These characteristics render the TIDI winds highly amenable to tidal analysis.
 This paper presents climatologies of monthly mean diurnal amplitudes and phases for seven tidal components (w4, w3, w2, s0, e1, e2, e3) at altitudes between 85 and 105 km and latitudes between 45°S and 45°N. The tidal climatologies were derived from TIDI zonal and meridional wind measurements taken between 1 June 2002 and 15 June 2005. They represent the first evidence of the vertical structure of nonmigrating tides over a 20-km range in the mesopause region. Owing to the above-mentioned limitations of previous satellite-borne wind instruments, such climatologies so far only exist at 95 km altitude from UARS data. A comparative analysis of the TIDI climatologies with global scale wave model (GSWM) and TIME-GCM tidal predictions provides further insight into the latent heat and the QSPW/tidal interaction forcing contributions to the observed tides and their seasonal variability.
 The paper is organized as follows. The TIDI data, the analysis method, and the error analysis are described in section 2. Monthly averaged amplitudes and phases are presented in section 3. Section 4 provides a comparison of the TIDI tides to the 95 km UARS climatology from Forbes et al. . The tidal sources responsible for the observed tidal winds are discussed in section 5 by comparing the TIDI data to GSWM and TIME-GCM tidal predictions. Concluding remarks are given in section 6.
2. Data and Analysis
2.1. TIDI Instrument and Sampling
 TIDI is the wind measuring instrument on board the TIMED satellite. It was developed and built by the University of Michigan [Killeen et al., 1999]. Daytime and nighttime neutral winds are measured by limb scanning various upper atmosphere airglow layers and monitoring the Doppler shift. The TIDI data used here are O2 (0-0) band P9 vector winds (level3, data versions 00_01 , 01_01 to 01_03 , 03_03 , 03_04 ) between 85 and 105 km that were produced by the National Center for Atmospheric Research (NCAR, see http://timed.hao.ucar.edu/tidi).
 TIDI has four telescopes that are orthogonally oriented. This allows the instrument to measure full wind vectors on both sides of the satellite track (cold and warm sides). The viewing directions of the two telescopes on the same side of the spacecraft are perpendicular to one another such that the same locations are observed with a time delay of a few minutes as the satellite moves forward. The samples in the two directions are then used to form the neutral wind vector in terms of the zonal (eastward) and meridional (northward) components.
 Data are taken from pole-to-pole with a vertical resolution of 2.5 km and an along track resolution of about 800 km. Simultaneous measurements on both sides of the satellite track provide four local solar time (LST) samples per orbit equatorward of ±60° and two at latitudes poleward of ±60°. For a given latitude, the LSTs of measurements taken on the ascending (asc) and descending (dsc) orbit nodes can be considered to be independent of longitude for warm and cold side data, respectively. Ascending (descending) orbit nodes are the instrument footprints when the satellite moves from south to north (north to south). The daily LST variation for a given latitude, side, and orbit node is 12 min toward earlier LST as time progresses. Complete (24 hours) LST coverage is obtained every 60 days which corresponds to one satellite yaw cycle (satellite orientation with respect to the orbital flight direction). More details about the instrument, its measurements, and recent results are given by Killeen et al. .
 TIDI has continuously taken data since March 2002 with one larger data gap in early 2003. Unfortunately, the instrument suffered from a light leak that resulted in a higher signal background than expected. Skinner et al.  describe the effect on optical performance: a decrease in throughput due to ice deposition on some parts of the optics. Efforts to sublimate the frost have led to an improved instrument performance since April 2003. Finding the zero wind position for space-borne Fabry-Perot interferometers has always been a challenging task. The TIDI zero wind determination is improving with every new version of the data products but some uncertainty still remains. In this analysis, however, this issue is accounted for in the analysis procedure (see section 2.2). The light leak also increased the noise level of the inverted wind data (30 m/s during the day, double that during the night) but this does not affect the monthly tidal climatologies. This is discussed in more detail later in this section.
2.2. Tidal Analysis
 The tides are derived as described by Oberheide et al. . However, the specifics of the analysis method are reviewed herein because they are essential for understanding the present paper. It is basically a two-dimensional Fourier transform of a 60-day composite data set. The composite data set is composed and analyzed as follows:
 1. For each measuring day, split the TIDI wind data into four subsets of ascending (asc) and descending (dsc) orbit node measurements for the TIDI warm and cold sides, respectively. All four data subsets have different local times for a given latitude.
 2. For each subset, combine 5 days of consecutive TIDI wind measurements to produce zonal and meridional winds in each of the latitude and longitude bands that saw satellite overpasses. Combining several days of data increases the number of data points per longitude/latitude band and thus reduces the noise level. It also closes data gaps. Note that combining 5 days of data results in 1 hour LST smoothing but this does not affect the analysis. Combining 3 days of data yielded almost identical amplitudes and phases (not shown).
 3. Map the four subsets separately onto a horizontal grid of 5° × 5° using a two-dimensional triangular filter function with a full width of 7.5° in the north-south direction and 45° in east-west direction. This yields 36 grid points in latitudinal and 72 grid points in longitudinal direction. Each grid point represents the average of about 40 TIDI wind measurements.
 4. Remove the zonal mean from the four mapped subsets. The zonal mean removal accounts for potential zero wind line inconsistencies. It also removes the migrating tides from the mapped data because migrating tides are observed as zonally symmetric features. They would therefore be aliased by temporal variations of the background (i.e., diurnal mean) winds. See Oberheide et al.  for details of tidal sampling issues in the data from slowly precessing satellites.
 5. Repeat steps 1 to 4 as a 5-day running mean for a 60-day period. The TIMED orbit geometry is such that 60 days of combined TIDI asc/dsc, warm/cold side data subsets have 24 hours of LST coverage.
 6. Combine 60 days of mapped data subsets and sort them in local time. Interpolate the sorted data set onto a fixed LST grid, including averaging if more than one data subset is available for a given LST bin. The resulting composite data set is therefore evenly spaced in LST (24 hours LST coverage) and longitude (360° coverage) for each latitude and altitude.
 7. Compute wave number/frequency pairs with two-dimensional Fourier transform. Owing to the 60-day averaging, the results must be interpreted in a climatological sense. Short-time variations will be smoothed out. Account for the fact, that the observed wave number is shifted by −1 as compared to the real zonal wave number (i.e., diurnal w2 is observed as w1). This is due to the satellite sampling [Oberheide et al., 2003]. Assign the amplitudes and phases to the day in the middle of the 60-day period.
 8. Repeat steps 1 to 7 as a 60-day running mean for the analysis period (June 2002 to June 2005) and average the derived amplitudes and phases into monthly bins.
 As an example for the composite data, Figure 1a shows a longitude/LST plot at 95 km for the meridional (northward) wind at 20°N (step 6 above). Fourier analysis then provides amplitudes and phases for different frequencies. The resulting wave numbers must, however, be shifted by 1 to account for the satellite sampling (step 7 above). Figure 1b shows the shifted amplitudes for diurnal and semidiurnal frequencies as function of latitude. It should be emphasized again that the removal of the zonal means (step 4 above) also results in the removal of the migrating tides. The migrating components cannot be analyzed with the analysis method and are not present in the figure. Semidiurnal tides are not discussed further in this report, but Figure 1b indicates their presence. They will be analyzed more closely in future work.
 The analysis is carried out for about 3 years of TIDI data (1 June 2002 to 15 June 2005). In early 2003, the satellite was moved such that TIDI was looking toward the Earth's surface. This prevented the tidal analysis during the first 3 months of the year. Apart from this gap, the TIDI data have an almost continuous temporal coverage.
2.3. Error Analysis
 There are basically three error sources that introduce some uncertainty in the derived tidal amplitudes and phases: measurement noise, artifacts from the asynoptic satellite sampling, and the analysis method itself. The analysis method accounts for the absolute error (accuracy) while the measurement noise and the asynoptic satellite sampling govern the relative error (precision).
 The analysis method itself introduces a considerable damping of the derived amplitudes because of the large widths of the triangular filter function in the horizontal mapping routine (analysis step 3 in section 2.2). This damping becomes more serious for larger wave numbers and also depends on the latitudinal structure of the tides. Components with a broad amplitude and phase distribution as function of latitude will be less affected than components with sharp amplitude minima/maxima and phase transitions. All derived tidal amplitudes and phases must therefore be corrected for the damping. The uncertainty of the correction must be considered as the accuracy. The correction and its uncertainty is determined using both model simulations and the measured data.
 First, monthly tidal wind amplitudes and phases from the GSWM model (see section 5 for details) are linearly interpolated to each day of the year. From these data, tidal wind perturbations for the LST, altitude, longitude, and latitude of the TIDI measurements are extracted for each day of an equivalent 3-year period (with data gaps). They are the synthesis of 13 diurnal and 13 semidiurnal tidal components (w6 to e6). The model mapping provides a data set identical to the TIDI data but with the measured zonal and meridional winds replaced with the model diurnal and semidiurnal tidal wind perturbations (“flying the satellite through the model”). This model data set is then analyzed as the measured TIDI data. The derived amplitudes are compared month by month to the full model output (monthly means of 60-day running mean averages). The comparison provides a correction for each tidal component that can be described by a scaling factor independent of month, latitude, and altitude. Phases remain unaffected but with some scatter (1 to 2 hours depending on the component) introduced by the asynoptic satellite sampling.
 Next, the scaling factors derived from the model simulations are applied in the sense of an initial guess to the monthly TIDI climatologies (called run 1). The scaled amplitudes and the (unscaled) phases from run 1 are then used to compute tidal wind perturbations for the TIDI footprints, as for the model simulation described above. Analyzing the data again as described in section 2.2 provides new monthly climatologies (called run 2) that are compared to the (unscaled) results from run 1. With perfect scaling of run 1, no systematic differences should occur. However, because GSWM and TIDI amplitude and phase distributions partly differ, run 1 and 2 results also differ. These differences are used to further improve the scaling of run 1 and the whole procedure is repeated. The systematic difference between the resulting monthly climatologies (run 3) and run 1 is on average smaller than 3%, indicating good scaling factors. The standard deviation of scaling factors derived month-by-month gives the uncertainty of the mean scaling factors provided in Table 1 and therefore the amplitude accuracy of the monthly climatologies. All tidal amplitudes presented in the remainder of the paper are from run 1 and have been scaled with the values given in Table 1. Phases do not require scaling.
Table 1. Scaling Factors Applied to the Derived Tidal Amplitudes and Meridional (v) and Zonal (u) Wind Errors of the Monthly Climatologiesa
Given values are for the scaled amplitudes.
Table 1 also provides the amplitude and phase precisions. They are the mean standard deviations between the (scaled) amplitudes and phases from runs 1 and 3 deduced for each month, latitude, and altitude. The precisions therefore include both the measurement noise and the noise introduced by the asynoptic satellite sampling. Propagating the measurement noise separately through the analysis yielded a consistent amplitude error of about 1 m/s. The inherent smoothing of the measurements due to the horizontal gridding and the use of 60 days of composite data significantly reduces the noise level.
3. Monthly Climatologies
 The tidal analysis covers the nonmigrating tidal components w4, w3, w2, s0, e1, e2, and e3 for the zonal (eastward) and meridional (northward) winds. For space reasons, it is impossible to include monthly amplitude and phase distributions for all of these 14 tidal components in the paper. Components w2, s0, e2, and e3 (zonal and meridional winds) will be presented in the following, while components w4, w3, and e1 are provided as electronic supplements to this paper.
 Electronic data files with the numerical values for all 14 tidal components are available on the web (http://www.atmos.physik.uni-wuppertal.de/cawses/nmt_mlt/). In the following, all amplitudes are given in m/s and phases are given in universal time (hours) of maximum amplitude at 0° longitude.
 The largest amplitudes (18 m/s) are found in January, February, and September to December, maximizing at about 20°S and 20°N around 95 km altitude. During these months, this component is antisymmetric with respect to the equator, as indicated by the phase jumps at 0° latitude. The phase behavior in June, when the amplitudes are smallest (8 m/s), is rather symmetric about the equator. Phases decrease with increasing altitude. This indicates an upward propagation in the observed altitude range and thus tidal forcing from below.
 The latitudinal structure of the s0 meridional component exhibits a large seasonal variation. Peaking in August (16 m/s), this component has two maxima at 20°S and 20°N in March, April, May, and again in August, September, and October with the Southern Hemisphere (SH) maximum much more pronounced than the Northern Hemisphere (NH) maximum. The peak altitude is between 90 and 95 km with a rather antisymmetric phase distribution with respect to the equator. During the remainder of the year, the phases suggest a more symmetric behavior of the s0 component with larger vertical wavelengths. A single amplitude maximum that varies with height is located at about 10°S to 20°S. Phases generally decrease with increasing altitude during all months of the year thus indicating tidal forcing from below.
 The e2 meridional amplitudes are generally symmetric about and centered on the equator with the symmetry mirrored in the phases. Exceptions are March and April, where a second amplitude maximum occurs at 40°S and above 100 km. These secondary maxima are out-of-phase with the equatorial peaks that are usually located between 95 to 100 km. All the equatorial amplitude maxima are between 6 to 8 m/s without much seasonal variation. The vertical phase distributions again suggest a tidal forcing from below.
 The e3 meridional component is always symmetric with respect to the equator with the largest amplitudes (10 m/s) occurring between November and March. The altitude of maximum amplitude varies between 95 km (November) and ≥105 km (April) with tidal forcing coming from below. Latitudinal phase jumps are observed at altitudes with small amplitudes which might indicate an increasing contribution of antisymmetric modes at higher latitudes.
 Like its meridional counterpart, the w2 zonal component also shows largest amplitudes (10 m/s) in January, February, and September to December. Amplitude maxima are found at about 95 km but at slightly higher latitudes (30°S and 30°N). In contrast to the meridional wind, the w2 zonal component is symmetric with respect to the equator. The amplitudes between March and August are relatively small, with the exception of June, where an equatorial maximum is observed above 100 km altitude. Tidal forcing is also from below, as indicated by the decreasing phases with increasing altitude.
 Similar to the s0 meridional component, the s0 zonal component is quite variable. Amplitude maxima (6–10 m/s) are observed between 30°–40°S and 30°–40°N but their altitude varies from ≤85 km (April, SH) to ≥105 km (March, NH) with the SH maximum usually found at lower altitudes. The phase distributions are also highly variable. They indicate a rather symmetric behavior of the s0 zonal component at altitudes below 95 km with increasing antisymmetric contributions toward higher altitudes. Although the phases usually decrease with increasing altitude (upward propagation, tidal forcing from below), there are some exceptions, e.g., in November, NH, which may indicate an in situ forcing or even a forcing from above.
 The e2 zonal component is symmetric about the equator with a long vertical wavelength. Amplitude maxima occur at about 20°S and 20°N with the altitude of maximum amplitude usually located above the upper boundary of the analysis. Tidal forcing is from below with maximum observed amplitudes of 6 m/s.
 The e3 zonal component is the single largest component observed in the TIDI data. Amplitudes reach almost 20 m/s in August. From April to November, this component is symmetric about the equator with a long vertical wavelength, peaking at altitudes ≥105 km. This behavior, however, is completely different during the remainder of the year. The vertical wavelength is much smaller and the equatorial amplitude maximum changes into two antisymmetric maxima located at about 20°S and 20°N. These maxima peak at lower altitudes (95 to 100 km). The vertical phase distributions nevertheless suggest that the tidal forcing is always from below.
3.9. w4, w3, e1; zonal and meridional
 The w4, w3, and e1 zonal and meridional monthly climatologies are provided as electronic supplement. Their basic features may be summarized as follows: The measured w4, w3, and e1 zonal components have relatively small amplitudes (2–4 m/s) which also applies to the w4 meridional component. However, the w3 and e1 meridional amplitudes can be up to 10 m/s (September) with both components generally being antisymmetric with respect to the equator. Tidal forcing is always from below.
4. Comparison With UARS
 A validation of the TIDI climatology is difficult because observations of nonmigrating tides have so far been very sparse. This is, however, different at 95 km altitude where climatological amplitude information deduced from HRDI and WINDII on UARS is available. Forbes et al.  provide monthly mean zonal and meridional wind amplitudes for the s0, w2, and e3 diurnal components based upon data taken between 1991 and 1994. The specifics of the UARS instruments prevented an analysis at other altitudes. Fortunately, 95 km is a good altitude for comparing the TIDI and UARS results. As shown in the previous section 3, many tidal components peak or already have large amplitudes near 95 km.
 It should nevertheless be noted that comparing the UARS data to the TIDI results does not meet the hard requirements of validation. The measurements are almost 11 years apart in time. This might be acceptable for climatologies that are taken during the same phase of the solar cycle, but there may or may not have been long term changes in the middle atmosphere. Furthermore, both TIDI and HRDI use the same technique to measure MLT winds. Validation would require comparisons with data obtained using different measurement techniques. This work will be carried out when, if any, such data become available, e.g., from chains of radar instruments.
 In order to get a general view of the TIDI and UARS results, as a first step it is helpful to compare the 3-year average climatological mean amplitudes and phases. Figure 10 shows the comparison for the meridional wind amplitudes (Figure 10a) and phases (Figure 10b). The comparison for the zonal wind is provided in Figure 11. Both instruments measure an identical latitudinal distribution for all three components but with the UARS amplitudes being roughly 50% smaller than the TIDI amplitudes. The corresponding climatological mean phases agree within the combined error bars, except for s0 in the NH. There is also a slight offset between UARS and TIDI in the SH s0 meridional phases. While such a deviation might be expected, considering the 11-year time lag between both data sets, the very stable w2 and e3 phases are remarkable.
 There may be several reasons for the amplitude differences between TIDI and the UARS analysis. Recent results by Huang and Reber  of nonmigrating tides in HRDI winds also indicate larger amplitudes (30%–50%) than those reported by Forbes et al. . In contrast, the analysis of Manson et al.  of the same data set is more consistent with the Forbes et al.  results. These inconsistencies have yet to be resolved, but it has been speculated [Huang and Reber, 2004] that the specifics of the horizontal data binning might be an issue (similar to those in section 2.3 above). Hence a reanalysis of the UARS data in the same way the TIDI data have been analyzed might be worthwhile. It is of course also possible that the amplitude differences between TIDI and HRDI come from the 11-year time difference of the measurements.
Forbes et al.  also provide time series for the diurnal components s0, w2, and e3. They are shown in Figure 12 for the meridional and in Figure 13 for the zonal wind together with the corresponding TIDI amplitudes. The largest differences between both data sets occur in the s0 zonal component. The NH maxima in April and August are not present in the UARS analysis which in turn shows a sharp amplitude peak in September at 30°S that is much broader in the TIDI data. Nevertheless, the agreement between TIDI and the UARS analysis is quite encouraging. Both data sets show an almost identical seasonal variation for several components. Considering the 11-year time lag, this points to a remarkable seasonal cycle stability of the nonmigrating tides.
5. Model/Observation Comparison
 The interpretation of the TIDI tidal diagnostics and the elucidation of the associated processes on the MLT requires complementary modeling efforts. One question to be answered in this context is what tidal sources and forcing mechanisms are responsible for the observed relative strength and seasonal variation of the tides. The TIDI climatologies are therefore compared to the tidal predictions from the global scale wave model (GSWM) [Hagan et al., 1995, 1997] and from the thermosphere-ionosphere-mesosphere-electrodynamics general circulation model (TIME-GCM) [Roble et al., 1988; Roble and Ridley, 1994; Roble, 1996].
 As a climatological, linear tidal model, GSWM does not account for nonlinear processes such as wave-wave interaction forcing and it does not produce interannual variations. The model version used here is described by Hagan and Forbes . The only tidal source included is latent heat release due to tropical deep convection. GSWM provides monthly amplitudes and phases for 13 diurnal and 13 semidiurnal tidal components (w6 to e6).
 Nonlinear wave-wave interaction forcing is the dominant source of nonmigrating tides in TIME-GCM that on the other hand, does not include the latent heat source. Hagan and Roble  show that the w2 and s0 diurnal components in TIME-GCM are predominantly forced by the nonlinear interaction between the migrating tide and quasi-stationary planetary waves (QSPWs). For this study, TIME-GCM was run for the years 2002 and 2003 with the migrating tides at the lower boundary specified by GSWM (radiative forcing only) and 10 hPa temperature and geopotential data from the National Center for Environmental Prediction (NCEP). The simulations also included realistic solar and geomagnetic forcing based upon the conditions that prevailed in 2002 and 2003. The model includes solar radiative excitation of migrating and nonmigrating tides above the lower model boundary. Daily model output was generated with 1 hour time resolution. Fast Fourier transform then provides daily tidal amplitudes and phases that were averaged into monthly bins.
 Time series of GSWM and TIME-GCM tidal predictions at 95 km have been compared to preliminary TIDI tides in an earlier paper [Oberheide et al., 2005]. Neither model alone could reproduce the seasonal variation of the w2 and s0 components. The combined model results described the observed amplitudes well during equinox, but they underestimated the w2 and s0 tides during winter solstice. The e3 component was solely forced by latent heat release. Because these basic findings still apply and for space reasons, comparisons of model and observation time series are not shown again. Model/observation comparisons are provided for two exemplary months: January and September.
 The GSWM and TIME-GCM amplitudes and phases of the w2 meridional component for January (Figure 14a) and September (Figure 14b) must be compared to the TIDI fields from Figure 2. With 2–4 m/s, the January response from both models is much smaller than the TIDI result (14–16 m/s). However, latitudes and altitudes of maximum amplitude compare well with the TIDI data. Both models predict an antisymmetric phase behavior which is in agreement with the observation. The September model results are between 8 m/s (GSWM) and 16 m/s (TIME-GCM) which comes close to observation. Latitude/altitude structure of the model tides as well as their antisymmetric phase behavior are in good agreement with the TIDI data. Both latent heat release and wave-wave interaction forcing are equally important.
 The model data must be compared to the TIDI data in Figure 3. The GSWM response in January is below 2 m/s but TIME-GCM predicts a broad amplitude distribution centered around the equator with a symmetric phase behavior and an amplitude maximum above 105 km (≥10 m/s). This is also found in the TIDI data although the peak altitude is lower (95 to 100 km). This suggests that s0 in January is basically forced by wave-wave interaction. The situation in September looks quite different. TIME-GCM has a response similar to that in January but with a larger amplitude (up to 20 m/s). In contrast, the GSWM response is antisymmetric with respect to the equator with a maximum amplitude of 6 m/s around 100 km. The TIDI result is less unequivocal. The long wavelength in the SH along with the large amplitudes below 100 km are rather consistent with TIME-GCM but the phase jump at 0° latitude and the small amplitudes in the NH and above 100 km are more consistent with GSWM. Both latent heat release and wave-wave interaction forcing therefore appear to be almost equally important in September.
 The e2 meridional model results must be compared to the TIDI data in Figure 4. TIME-GCM response is very small with GSWM predicting amplitudes of 4–6 m/s, symmetric about the equator. This is also seen in the TIDI data, but with the peak (95 km, 6 m/s) located at lower altitudes. Latent heat release in the tropical troposphere appears to be the dominant source for this component.
 The model data must be compared to the TIDI data in Figure 5. As for e2, the TIME-GCM response is negligible with GSWM predicting equatorial amplitudes of about 6 m/s peaking around 95 km (January, Figure 17a) and 105 km (September, Figure 17b). The model phase distribution is basically symmetric with respect to the equator but with some slope in it. The TIDI amplitudes agree well with the model predictions although they are slightly larger (10 m/s in January, 8 m/s in September). TIDI data also reveal a symmetric phase behavior about the equator. The comparison indicates that latent heat release in the tropical troposphere is the dominant source for this component.
 The w2 zonal model results must be compared to the TIDI data in Figure 6. Both models predict a symmetric phase behavior which agrees with the observation. The model responses in January are between 2 to 4 m/s and, similar to the w2 meridional component, much smaller than the observation. Consistent with the findings for the meridional wind, the larger (6 to 8 m/s) model responses in September come close to the TIDI observations but with the models peaking at a slightly higher altitude. Both latent heat release and wave-wave interaction forcing are equally important.
 The model data must be compared to the TIDI data in Figure 7. GSWM response in January is small, with TIME-GCM amplitudes at about 6–8 m/s. TIME-GCM amplitudes maximize at higher altitudes as compared to TIDI. The antisymmetric phase distribution is also observed in the TIDI data. Obviously, wave-wave interaction forcing contributes significantly to the observed s0 zonal component. This situation looks different in September. The large TIME-GCM amplitudes basically agree with the TIDI amplitudes but instead of the observed symmetric phases, the model predicts an antisymmetric phase behavior. Such a symmetric phasing is reproduced by GSWM but with small amplitudes. The phases therefore point to an important latent heat release contribution but GSWM obviously underestimates this source in September.
 The model data must be compared to the TIDI data in Figure 8. Both model results in January are almost symmetric with respect to the equator with a broad amplitude distribution. The GSWM response (10 m/s) is much larger than TIME-GCM with a slight shift toward northern latitudes below 100 km. This is similar to the observation although the TIDI amplitudes are only half of the model amplitudes. The e2 zonal component in January therefore appears to be largely dominated by the latent heat source. This is different in September. The antisymmetric phase behavior predicted by GSWM is not present in the TIDI data. Amplitudes and phases are more consistent with the TIME-GCM although the model response is too small. It is unclear, whether the e2 component in TIME-GCM is excited by wave-wave interaction or whether the small model response is more due to radiative forcing.
 The e3 zonal model results must be compared to the TIDI data in Figure 9. TIME-GCM response is insignificant such that from the model point of view, this component is solely forced by latent heat release. GSWM response in January and September is well above 20 m/s with symmetric phase behavior. This agrees well with the TIDI results in September but the observed January tides are rather antisymmetric about the equator. It remains unclear whether this means that an antisymmetric wind expansion mode is dominating the observational response (compared to a symmetric one in the model) or whether other mechanisms may play a role in explaining the observed e3 response during January. It should be noted that the TIDI January structure is also present in the UARS data shown in Figure 13.
 The model/observation comparisons indicate that for the w2 and s0 components both latent heat release in the tropical troposphere and wave-wave interaction forcing need to be considered. Their relative contributions may differ from month to month but these sources can basically explain the observations. The eastward propagating components, however, appear to be mostly governed by the latent heat source, although some differences between the model prediction and the observation exist.
 The quantitative agreement between the model predictions and the TIDI observations is good for some months and rather bad for other months. This is not very surprising because even the combined models do not account for all tidal sources (latent heat release tidal/planetary wave interaction forcing is missing) and interactions with the background atmosphere (GSWM is a linear model). The model/observation comparisons also indicate that tidal dissipation may be underestimated. Many of the observed components show a lower peak altitude than predicted by the models.
6. Summary and Conclusions
 TIDI winds provide a data set that is unprecedented in that it is amenable to global nonmigrating tidal analysis over a range of MLT altitudes. Monthly climatologies of zonal and meridional diurnal tides for seven components have been derived between 45°S and 45°N and between 85 and 105 km altitude (westward propagating wave numbers 2, 3, and 4; the standing oscillation s0; and eastward propagating wave numbers 1, 2, and 3). Comparisons with UARS results at 95 km yield a good agreement and provide additional confidence in the TIDI results. Amplitudes of a single nonmigrating tidal component can reach 20 m/s. Their aggregate effects easily exceed the amplitude of the radiatively forced migrating tide. Nonmigrating tides therefore introduce a considerable amplitude and phase modulation of the tidal fields in the MLT. This is particularly important when comparing ground-based observations with satellite data or with models. The variability of the tidal wind fields is also a very challenging task for data assimilation approaches in the MLT.
 A comparative analysis with the tidal predictions from two models, the GSWM and the TIME-GCM, provides insight into some of the tidal forcing mechanisms. Latent heat release in the tropical troposphere and nonlinear wave-wave interaction are equally important in forcing the w2 and s0 tides. The eastward propagating components are mostly governed by the tropospheric latent heat source. There are nevertheless numerous shortcomings in our present understanding of diurnal nonmigrating tides. Parts of the observed seasonal variations and some monthly structures remain unexplained by the models. There is an obvious need to further improve tidal forcing and dissipation schemes in the future. The TIDI climatologies will provide the necessary guidance for such efforts.
 Future work will focus on providing a similar climatology of the semidiurnal tides. Interannual variations and how they are related to variations in the atmospheric background state, tidal source variations, and solar cycle effects will also be investigated. This is particularly interesting because TIDI will be the only instrument measuring MLT winds for some time to come.
 The authors wish to thank H.-L. Liu for comments on an initial draft of this paper. J. Oberheide's work is supported by the Deutsche Forschungsgemeinschaft (DFG), Bonn, Germany, through its priority program CAWSES, grant OB 299/2-1. M. Hagan's efforts were partially supported by a TIMED/CEDAR grant. The National Center for Atmospheric Research is sponsored by NSF.
 Lou-Chuang Lee thanks the reviewers for their assistance in evaluating this paper.