1. Introduction
The overarching scientific objective of the Deep Propagating Gravity Wave Experiment (DEEPWAVE) was to observe and understand the end-to-end dynamics of gravity waves from the ground to the edge of space at
To achieve this science, DEEPWAVE was organized around a major field measurement campaign based out of Christchurch, New Zealand (43.49°S, 172.54°E), during May–July 2014. During austral winter, this location was identified as a natural laboratory for observing these dynamics due to abundant local orographic and nonorographic sources of gravity waves (e.g., Reeder et al. 1999; Lane et al. 2000; Guest et al. 2000; Zink and Vincent 2001; Smith et al. 2013) and a stable vortex circulation, which, as shown in Fig. 1a, maintains eastward mean winds climatologically from the surface to
(a) July zonal winds (m s−1) as a function of pressure (altitude) and latitude averaged over the years 2007–09 and the longitude zone 140°–190°E near New Zealand from the high-altitude global reanalysis of Eckermann et al. (2009). Dotted white line marks the DEEPWAVE operational base in Christchurch (43.5°S). (b) Altitude ranges of the (left) NGV measurements and (right) operational NWP models available during DEEPWAVE [cf. Tables 2 and 3 of Fritts et al. (2016)].
Citation: Monthly Weather Review 146, 8; 10.1175/MWR-D-17-0386.1
The primary observational platform was the National Science Foundation (NSF)/National Center for Atmospheric Research (NCAR) Gulfstream V research aircraft (NGV; Laursen et al. 2006). As shown in Fig. 1b, during DEEPWAVE, the NGV was equipped with in situ and remote sensing instruments with the necessary vertical range, space–time resolution, and measurement precision to observe gravity wave dynamics over most of the 0–100-km altitude range. NGV observing missions were planned and supported by a suite of gravity wave–resolving numerical weather prediction (NWP) systems (Fig. 1b) and by an extensive network of temporary and permanent ground-based observations; for details, see Fritts et al. (2016).
Some of the most spectacular and unanticipated gravity wave events observed during DEEPWAVE occurred in the MLT at
This common observational restriction is typically overcome through the use of analysis or reanalysis products, which provide an estimate of the state of the atmosphere based on assimilation of available heterogeneous observations using data assimilation systems (DASs). However, as depicted in Fig. 1b, the suite of NWP DASs used during DEEPWAVE all had upper boundaries that did not extend into the MLT. Indeed, at present, no NWP center provides either near-real-time or retrospective analysis products above 60–80-km altitude operationally.
Recognizing this analysis gap above 60–80 km, several groups are developing research prototypes that extend global NWP capabilities into the MLT and higher (e.g., Polavarapu et al. 2005; Wang et al. 2012). The first such prototype to successfully assimilate MLT observations and generate global reanalysis products through the MLT was based on the Navy Operational Global Atmospheric Prediction System (NOGAPS), described in Eckermann et al. (2009); reanalysis winds from that system are plotted in Fig. 1a. In 2012, NOGAPS was replaced by the next-generation Navy Global Environmental Model (NAVGEM; Hogan et al. 2014), and similarly motivated research that begins to extend NAVGEM capabilities into the MLT has been reported by Hoppel et al. (2013) and McCormack et al. (2017).
This paper describes a new 0–100-km reanalysis system based around NAVGEM and its use in generating 0–100-km global atmospheric reanalysis products for DEEPWAVE scientific research. The properties of this system and the MLT observations it assimilates are described in section 2. Reanalysis experiments for the 2014 austral winter are outlined in section 3. Reanalyzed temperatures and winds in the MLT are validated against independent observations in section 4. The 0–100-km reanalysis products are applied in section 5 to delineate aspects of planetary-wave dynamics specific to the greater New Zealand region that potentially impacted MLT gravity waves observed during DEEPWAVE. Major scientific conclusions derived from this reanalysis research are summarized in section 6.
2. High-altitude NAVGEM
a. System overview
The forecast-assimilation cycle of NAVGEM is depicted schematically in Fig. 2. On the outer loop, the global forecast model, depicted with a red box at the top of Fig. 2 and described in section 2b(1), issues a high-resolution deterministic forecast. The 0–9-h forecasts provide a background trajectory
Schematic depiction of the NAVGEM reanalysis system. See text for explanation of pathways, nomenclature, and mathematical symbols.
Citation: Monthly Weather Review 146, 8; 10.1175/MWR-D-17-0386.1
As depicted by the teal arrows in Fig. 2, these 6-h analyses are both archived offline and fed back to the forecast model as atmospheric initial conditions for the next update cycle, closing the outer loop, which repeats every 6 h and generates a new analysis every 6 h. To better resolve tides in the MLT (see section 4), here, we supplement the 6-h analysis with outer-loop forecast backgrounds from the next cycle at 1-h intervals from +1 to 5 h after initialization to provide a seamless global time series of 1-h resolution.
b. System components
1) Forecast model
Hogan et al. (2014) provide detailed descriptions of the operational configuration of the forecast model, which is structured around a global, three-time-level (3TL), semi-implicit, semi-Lagrangian (SISL) dynamical core.
In the vertical, the model uses the NEWHYB2 hybrid σ–p coordinate of Eckermann (2009). For operational NWP at the Fleet Numerical Meteorology and Oceanography Center (FNMOC), 60 vertical layers (L60) are currently adopted with a rigid upper boundary at ptop = 0.04 hPa (see Fig. 1 of Eckermann et al. 2014). As shown in Fig. 3b, in extending NAVGEM through the MLT, we mirrored those operational L60 levels at pressure altitudes
(a) The 74 NAVGEM σ–p levels (L74) shown as interface pressures along a 43.5°S latitude circle from 140°E to 180°. Note upward displacement of levels over resolved T425 terrain of South Island of New Zealand (
Citation: Monthly Weather Review 146, 8; 10.1175/MWR-D-17-0386.1
Hogan et al. (2014) document the suite of physical parameterizations used in the L60 NAVGEM. Physical parameterizations above 70 km in the L74 model will be described more fully in forthcoming publications and so are only briefly summarized here, given this paper’s focus on data assimilation. Although new fast parameterizations of exothermic chemical heating and radiative heating and cooling modified by breakdown in local thermodynamic equilibrium were available, these schemes are still being tested and refined. For this work, we incorporated simpler temporary lookup-table-based parameterizations of these rates as a function of season, latitude, and height, derived by archiving and averaging rates from a 25-yr simulation of the specified dynamics version of the Whole Atmosphere Community Climate Model (SD-WACCM), which incorporates detailed parameterizations of UV absorption, chemistry, and species transport (see section 2 of Marsh et al. 2007). Orographic gravity wave drag (OGWD) and flow-blocking drag were parameterized following Webster et al. (2003). Subgrid-scale nonorographic GWD (NGWD) was parameterized using the stochastic scheme of Eckermann (2011). No attempt was made to retune NGWD parameters separately for the different reanalysis runs described in section 3a. The ensemble forecasts described in section 2b(2) employed stochastic kinetic energy backscatter (SKEB), as described in section 2b of Reynolds et al. (2011), but with an additional convective dissipation mask based on moisture convergence (see section 3b of Reynolds et al. 2011) that enhances kinetic energy by introducing vorticity perturbations in areas where convective processes are likely to occur.
2) Data assimilation algorithm
(i) Formulation
The current NRL Atmospheric Variational DAS (NAVDAS) is based around a four-dimensional variational (4DVAR) algorithm solved in observation space using an accelerated representer (AR) method. An overview of NAVDAS-AR relevant to the 0–100-km NAVGEM is provided here; more complete descriptions of specific aspects are provided elsewhere (see, e.g., Daley and Barker 2001a; Xu et al. 2005; Kuhl et al. 2013; Allen et al. 2014).























As depicted in Fig. 2, this AR solution converts the input innovations
(ii) Numerical solvers












Multiple sweep cycles are involved in solving iteratively for
(iii) Digital filter
Analysis errors lead to spurious imbalances, which, when passed to the forecast model as atmospheric initial conditions, can trigger spontaneous emission of resolved gravity waves. In high-altitude models, this gravity wave noise propagates deep into the MLT, where it attains large amplitudes and often breaks, driving deleterious upscale impacts on analyzed mean and tidal structures (e.g., Sankey et al. 2007; Nezlin et al. 2009b). Standard methods for removing spurious imbalances, such as nonlinear normal-mode initialization (NNMI) and digital filters, perform poorly when applied directly to high-altitude analyses, since they distort tides and suppress geophysical gravity waves (Wergen 1989; Sankey et al. 2007). Conversely, applying these filters to analysis increments effectively removes noise while better preserving geophysical mean, tidal, and gravity wave features at all altitudes (Ballish et al. 1992; Seaman et al. 1995; Courtier et al. 1998; Sankey et al. 2007; Buehner et al. 2015).
While a previous version of our high-altitude analysis system used incremental NNMI (Eckermann et al. 2009), NNMI has been superseded in NAVGEM by a digital filter of the Lanczos form (see section 3a of Lynch and Huang 1992), which is applied to time-evolved increments from the TLM forward sweep of the final
(iv) Specifying 



Aspects of the static background error covariances
Citation: Monthly Weather Review 146, 8; 10.1175/MWR-D-17-0386.1
The matrix
One off-diagonal correlation of particular relevance to this study is the coupling between rotational and geostrophic winds. Since geostrophic winds are specified diagnostically from horizontal gradients in geopotential (temperature), this correlation allows assimilated temperature information to produce rotational wind increments. Following Lorenc (1981), the strength of this coupling is prescribed by a correlation coefficient μ, where









c. Assimilated MLT observations
The heterogeneous tropospheric and stratospheric observations assimilated by NAVGEM for operational NWP (e.g., Hoover and Langland 2016; Campbell et al. 2017) are also assimilated here. Here, we focus on additional observations assimilated above
1) MLS and SABER limb retrievals
We assimilate temperatures from the Microwave Limb Sounder (MLS) on NASA’s Aura satellite (Schwartz et al. 2008) and the Sounding of the Atmosphere using Broadband Emission Radiometry (SABER) instrument on NASA’s Thermosphere, Ionosphere, Mesosphere Energetics and Dynamics (TIMED) satellite (Remsberg et al. 2008). In earlier work, we assimilated temperatures from version 2 (v2) MLS and v1.06 and v1.07 SABER retrievals (Hoppel et al. 2008; Eckermann et al. 2009; Hoppel et al. 2013). Improved data from new retrieval versions have since appeared. The v3 and v4 MLS retrievals mainly improve the upper troposphere and lower stratosphere, as well as provide temperatures on a finer-altitude grid (Yan et al. 2016). SABER v2.0 retrievals incorporate better radiometric calibration and improved nonlocal thermodynamic equilibrium (LTE) retrievals that lead to less data rejection (higher data rates) and extend temperatures to higher altitudes (Rezac et al. 2015). Thus, here, we assimilate v4 MLS temperatures from 100 to 0.002 hPa and v2.0 SABER temperatures from 100 to 2 × 10−4 hPa. Initial NAVGEM reanalysis runs described in Eckermann et al. (2016) assimilated v3 MLS temperatures because the newer v4.2 retrievals were not yet available for all observation days in 2014.
Previous comparisons of v2 MLS and v1 SABER temperatures revealed height-dependent mean biases (e.g., Schwartz et al. 2008; Remsberg et al. 2008; Hoppel et al. 2008; Eckermann et al. 2009). Thus, we recomputed mean biases between MLS and SABER temperatures for all available retrieval versions. Differences were studied via MLS–SABER coincidences, defined as profile pairs separated in local time by ≤1 h and horizontally by great circle distances ≤200 km. SABER temperatures were linearly interpolated onto the MLS pressure grid, and coincident difference profiles were averaged over the years 2005–12 for v1.07 SABER, 2005–14 for v2.0 SABER and v2 and v3 MLS, and 2005–15 for v2.0 SABER and v4 MLS. Global-mean biases are plotted in Fig. 5, showing a reproducible mean bias profile over all MLS and SABER retrieval versions. Consistent with previous studies (e.g., Schwartz et al. 2008), we found little seasonal or latitudinal variation in this bias. Note in particular from Fig. 5 the large increases in coincidence data with v2.0 SABER, due to higher raw data acceptance rates in v2.0 temperature retrievals relative to earlier versions.
Global-mean temperature biases between v2, v3, and v4 MLS and v1.07 and v2.0 SABER temperatures, based on coincidence criteria in local time of
Citation: Monthly Weather Review 146, 8; 10.1175/MWR-D-17-0386.1
We use the v4 MLS-v2.0 SABER mean bias profile (red curve in Fig. 5) within NAVGEM to correct SABER temperatures from 68 to 5 hPa, given evidence of a SABER warm bias at these altitudes (Remsberg et al. 2008; García-Comas et al. 2014). After bias correction, three-point vertical smoothing is applied to SABER profiles, with additional three-point smoothing at levels above 1 hPa to produce profile resolutions more similar to layer thicknesses in Fig. 3b. From 5 to 0.002 hPa, the v4 MLS-v2.0 SABER bias profile in Fig. 5 is used to correct MLS temperatures, given evidence of large MLS cold biases at higher altitudes (Schwartz et al. 2008; French and Mulligan 2010; García-Comas et al. 2014).
The MLS and SABER contributions to the observation-error covariance matrix
We also assimilate v4.2 MLS ozone retrievals from 100 to 0.6 hPa and v4.2 MLS water vapor retrievals from 50 to 0.002 hPa. Since MLS water vapor precision degrades rapidly with height at upper levels [e.g., Table 2 of Lambert et al. (2007)], as in Eckermann et al. (2009), adjacent data are smoothed at heights above 0.05 hPa by applying three-point along-track smoothing prior to assimilation. From 0.012 to 0.005 hPa, this along-track smoother is applied twice; from 0.005 to 0.003 hPa it is applied three times; and from 0.003 to 0.002 hPa it is applied four times.
2) SSMIS UAS radiances
The Special Sensor Microwave Imager/Sounder (SSMIS; Kunkee et al. 2008) is a conical-scanning, nadir-viewing, 24-channel radiometer deployed on operational satellites of the Defense Meteorological Satellite Program (DMSP). With the launch of the DMSP F19 satellite on 3 April 2014, SSMIS radiances from four DMSP satellites (F16–F19) were potentially available for assimilation during the 2014 austral winter. However, in April 2013, SSMIS temperature sounding channels on F16 suffered hardware failures of the 56.4-GHz phase-locked oscillator (PLO) controlling precise positioning of central channel frequencies. A resulting failsafe PLO mode that attempts to regain frequency lock via continuous frequency sweeps allowed temperature information to be recovered only from the single passband lower-atmosphere sounding (LAS) temperature channels on F16 after April 2013.
Unless otherwise noted below, our assimilation of microwave radiances from the six SSMIS upper-atmosphere sounding (UAS) channels 19–24 (Swadley et al. 2008) on F17, F18, and F19 follows procedures described in sections 3a and 4a of Hoppel et al. (2013). Since the SSMIS UAS channels have extremely narrow frequency bandwidths, additional spatial averaging is necessary to lower the effective scene noise required by NWP assimilation systems. Thus, unlike data from other satellite nadir sensors, which are averaged and thinned by the NAVGEM preprocessor (PP) and quality control (QC) algorithms, SSMIS radiance averaging is performed prior to transfer of the data to NAVGEM by onboard flight software and then by the SSMIS Unified Preprocessor (UPP; Bell et al. 2008). The original UPP has been modified to increase the spatial averaging of UAS radiances, as well as adding information necessary for the UAS-channel components of the Community Radiative Transfer Model (CRTM; Han et al. 2010), such as the upwelling radiation propagation vector and geomagnetic field vectors (Maurer et al. 2015). Within NAVGEM, systematic radiance biases are identified and removed using variational bias correction (varBC; Dee and Uppala 2009), with the LAS and UAS channels treated separately, replacing earlier SSMIS bias-correction procedures described in section 4a of Hoppel et al. (2013). Bias-corrected UAS radiances are assimilated here using version 2.2.1 of the CRTM that incorporates Zeeman splitting of O2 lines by geomagnetic fields and frequency shifts due to Earth’s rotation. Prognostic temperature inputs to the CRTM are capped at 10−3 hPa and replaced by climatology at higher levels.
As shown in Fig. 1 of Hoppel et al. (2013), SSMIS UAS radiances provide observational temperature information from
3. 0–100-km reanalyses for 2014 austral winter
a. Experiments
As summarized in Table 1, we conducted four separate L74 NAVGEM reanalysis experiments.
The four NAVGEM L74 DEEPWAVE reanalysis experiments.
Our control 4DVAR runs (hereafter labeled “4DVAR”) deactivated the inner-loop ensemble capability such that background error covariances were specified statically
Our research reanalysis runs (hereafter labeled “HYBRID”) sought to exploit new hybrid-4DVAR assimilation capabilities recently accommodated within NAVGEM that, as discussed in section 2b(2), offer potentially large improvements in reanalysis skill in the MLT. While for purely MLT applications
Both the 4DVAR and HYBRID experiments were performed at two different horizontal resolutions. “Synoptic” runs (hereafter denoted “119”) adopted outer-loop triangular spectral truncation at total wavenumber 119 (T119) and used a full quadratic Gaussian grid within the forecast model, yielding grid cells of
All four experiments were initialized from a previous reanalysis at 0000 UTC 20 March 2014, with the first
b. Ensemble error covariances 

Figures 4c and 4d show zonal-mean standard deviations of temperature and horizontal wind perturbations, respectively, averaged over all HYBRID119 inner-loop (T47) ensemble forecasts during July 2014. Relative to corresponding static values in Figs. 4a and 4b, these errors in July 2014 were generally comparable in the stratosphere but larger in the MLT.
Covariances were formed relative to an observation point located over the South Island of New Zealand (45°S, 170°E) and averaged over all ensemble forecasts in July. Vertical temperature correlations in Fig. 4g show similar properties to the static values in Fig. 4e.
Solid and dotted lines in Fig. 4h show variation with height of mean correlation coefficients between zonal and meridional components, respectively, of geostrophic and rotational wind perturbations in both HYBRID119 and HYBRID425 inner-loop ensemble forecasts. Through the free troposphere and stratosphere up to
c. Observational diagnostics
Figure 6 presents a color-coded checkerboard summary of observations assimilated during the 4DVAR119 experiment every 6 h from 10 June to 11 July 2014. Although
Assimilated data counts for 4DVAR119 run every 6 h from 10 Jun to 11 Jul 2014, listed by sensor and ordered vertically according to mean data counts over the month, for a mean total of 3.3 × 106 assimilated observations per cycle. Data sources used in operational NWP are listed using standard acronyms; see Table 1 of Hoover and Langland (2016). Color codes show instantaneous observation count as a percentage departure below or above sensor means (see color key at bottom; white indicates missing data). Note large data rates from the SABER and MLS sensors and SSMIS UAS channels (highlighted in red), which provide data above
Citation: Monthly Weather Review 146, 8; 10.1175/MWR-D-17-0386.1
The largest number of assimilated observations in Fig. 6 comes from hyperspectral infrared nadir sensors—Infrared Atmospheric Sounding Interferometers (IASIs) on the European MetOp-A and MetOp-B satellites and Atmospheric Infrared Sounder (AIRS) on NASA’s Aqua satellite. NAVGEM assimilated radiances from 51 IASI and 50 AIRS channels in the temperature-sensitive
Our three sensors providing MLT observations, marked in red on the left and right of Fig. 6, provide
Figure 7 shows the global distribution of the major assimilated satellite observations for three consecutive NAVGEM update cycles during 1 July 2014. In the troposphere and stratosphere (bottom row), dense global pole-to-pole observational coverage is provided every cycle. In the MLT (top row), while observations are less dense, the combined coverage is also global. Just as important for MLT assimilation, nearly all longitudes are sampled at each latitude with few gaps in coverage, implying complete local time coverage in the MLT observations entering the analysis. This in turn allows the combined MLT observations to provide information on migrating and nonmigrating solar tides to the analysis. Note also that unlike the other sensors in Fig. 7, which are in polar orbits, TIMED orbits at a 74° inclination and undergoes a yaw cycle every
Geographic sampling of observations assimilated within successive 6-h NAVGEM assimilation windows on 1 Jul 2014 centered at (a),(d) 0600; (b),(e) 1200; and (c),(f) 1800 UTC. (bottom) Sampling of satellite nadir radiances observed by IASI on MetOp-A and MetOp-B (green), ATMS on Suomi NPP (blue), AMSU-A on Aqua, MetOp-A/B, NOAA-15–19 (pink), and AIRS and AMSU-A on Aqua (red). (top) Assimilated MLT observations from the SSMIS UAS channels on DMSP F17, F18, and F19 (green), SABER on TIMED (blue), and MLS on Aura (red).
Citation: Monthly Weather Review 146, 8; 10.1175/MWR-D-17-0386.1
Our MLT observations in Figs. 7a–7c are a combination of nadir-scanned radiances with high horizontal resolution but poor vertical resolution (SSMIS UAS) and limb-scanned temperature profiles with poorer horizontal coverage but high vertical resolution (MLS and SABER). Figures 7d–7f show only the coverage of assimilated data from nadir sensors of the troposphere and stratosphere. Figure 8 shows an example of the high vertical resolution profile data entering the tropospheric and stratospheric analysis from global positioning system radio occultation (GPSRO) data available operationally from GPS-enabled satellite platforms, as well as observations from the worldwide radiosonde network (RAOBs). Included within the RAOBs are dropsonde observations of opportunity from aircraft sorties (e.g., into hurricanes) made available via near-real-time transmission through WMO’s Global Telecommunication System (GTS) to operational centers. These include AVAPS dropsonde data acquired from the NGV during DEEPWAVE (Fritts et al. 2016; Young et al. 2016); assimilated AVAPS profiles from DEEPWAVE research flight number 25 (RF25) are highlighted in black in Fig. 8.
Observational coverage during update cycle at 1200 UTC 18 Jul of GPSRO (red) and radiosonde/dropsonde observations (raob; blue). Assimilated NGV AVAPS observations from DEEPWAVE RF25 are highlighted with black/yellow circles.
Citation: Monthly Weather Review 146, 8; 10.1175/MWR-D-17-0386.1
Figure 9 shows an example of the bias-corrected SSMIS UAS radiance innovations provided as observational MLT inputs to NAVGEM. The plots show results for channels 19 and 21, which peak at
NAVGEM HYBRID425 bias-corrected radiance innovations for F18 SSMIS UAS channels (a) 19 and (b) 21 at 0000 UTC 14 Jun 2014, expressed as a brightness temperature (K). Histograms on color bar show probability distribution of innovation values in each map.
Citation: Monthly Weather Review 146, 8; 10.1175/MWR-D-17-0386.1
4. Observational validation of MLT reanalysis
a. MLT temperatures
1) SOFIE
The Solar Occultation for Ice Experiment (SOFIE; Gordley et al. 2009) acquires
Figure 10a plots the geographic distribution of all the SOFIE temperature profiles acquired in the Southern Hemisphere from May to August 2014 (1624 in all). We interpolated NAVGEM reanalyses to the latitude, longitude, and time of each SOFIE limb profile, then interpolated SOFIE temperatures from the version 1.03 vertical retrieval grid (Marshall et al. 2011) onto the NAVGEM vertical model levels in Fig. 3. Figures 10b and 10c show resulting means and standard deviations, respectively, of temperature differences between SOFIE and each of the four NAVGEM MLT reanalyses.
(a) Geographic distribution of 1624 SOFIE solar limb occultation profiles acquired during May–August 2014. Colored curves in remaining panels show mean (b) bias and (c) standard deviation of temperatures from all four NAVGEM reanalyses (see color key at top) with respect to these SOFIE temperature profiles acquired at locations in (a). Error bars in (b) are standard errors of the mean derived from standard deviations in (c). Solid curves show results from 1-h NAVGEM fields, and dotted curves show results from 6-h analysis only. Gray curves in (c) show zonal-mean background errors at 67°S: zonal-mean static errors (solid curve; see Fig. 4a) and flow-dependent errors (broken curves) from the inner-loop ensemble forecasts in July 2014 for HYBRID119 (T47 inner; see Fig. 4c) and HYBRID425 (T119 inner).
Citation: Monthly Weather Review 146, 8; 10.1175/MWR-D-17-0386.1
Mean biases in Fig. 10b show a systematic NAVGEM 4DVAR warm bias that is absent from the HYBRID reanalyses. This MLT warm bias of 4DVAR relative to HYBRID is observed at other latitudes. The HYBRID MLT temperatures are unbiased to
Standard deviations of the temperature differences
These error variances also help to explain the warm bias of 4DVAR MLT reanalysis relative to HYBRID evident in Fig. 10b. The 4DVAR reanalysis adopts a
Dotted and solid curves in Figs. 10b and 10c compare results from interpolating NAVGEM fields at 6- and 1-h time cadence, respectively. For HYBRID in particular, the addition of +1–5-h forecasts into the 6-h reanalysis stream improves reanalysis skill by reducing mean biases above 80 km and standard deviations above 60 km. We show below that a major source of this improvement is through improved temporal resolution of MLT tidal temperature structure.
2) DLR lidar
The German Aerospace Center (DLR) deployed a Rayleigh–Raman lidar in Lauder, New Zealand (45.04°S, 169.68°E), that took measurements from 25 June to 3 November 2014. Kaifler et al. (2015) describe the instrument and data processing used to derive vertical temperature profiles extending into the MLT. Various temperature retrieval products were provided for DEEPWAVE with different space–time resolutions (Kaifler and Kaifler 2016). Here, we use retrieved temperatures with an effective vertical resolution of
We first interpolated temperature reanalyses from full model levels onto an equispaced 1-km grid of geometric heights z above sea level spanning 0–110 km, derived from reanalyzed geopotential heights
Figures 11a and 11b plot weighted means and standard deviations, respectively, of differences between the lidar and NAVGEM temperatures over Lauder. Weights were the inverse of the squared measurement error associated with each lidar temperature value; the gray curve in Fig. 11b shows the mean of those errors. From 30 to 50 km, the biases of the NAVGEM HYBRID and 4DVAR reanalyses are all very similar and also compare closely to the bias of ECMWF operational analysis temperatures to these lidar data reported by Gisinger et al. (2017). Above
(a) Mean temperature bias between NAVGEM reanalyses and DLR lidar profiles acquired from Lauder [see map inset in (b)] from 25 Jun to 30 Sep (1864 profiles in all). Color keys for each NAVGEM reanalysis are provided above each plot panel, with solid and dotted curves indicating use of analysis fields with 1- and 6-h resolution, respectively. Dark solid curve shows mean bias between lidar and ECMWF operational analysis temperatures from July to September 2014 [after Fig. B2 of Gisinger et al. (2017)]. (b) Corresponding error-weighted standard deviations between NAVGEM and DLR lidar profiles. Gray curve shows mean lidar measurement error.
Citation: Monthly Weather Review 146, 8; 10.1175/MWR-D-17-0386.1
Above
To study local time variations over Lauder, we interpolated lidar and NAVGEM profile pairs from the irregular measurement times on each night onto a common regular local-time grid of 30-min resolution. We computed means within each time bin, then extracted the time-mean profile to study anomalies versus local time. Figure 12a plots the lidar temperature anomalies as a function of local time. The panel beneath it shows the total number of 1-km layers from 30 to 100 km containing a lidar temperature measurement within each time interval, revealing fairly uniform measurement coverage from
(a) Mean lidar temperature anomalies vs UTC over Lauder (local time is 12 h ahead of UTC) based on averaging profiles from 25 Jun to 30 Sep 2014 and (e) total number of individual lidar temperature points in each 30-min time bin going into this mean. Remaining panels show corresponding results for NAVGEM reanalysis over Lauder: HYBRID119 at (b) 1-, (c) 3-, and (d) 6-h time cadence and 1-h reanalysis from (f) 4DVAR119, (g) 4DVAR425, and (h) HYBRID425. Note color bars on right for range and units.
Citation: Monthly Weather Review 146, 8; 10.1175/MWR-D-17-0386.1
Remaining panels in Fig. 12 show corresponding mean temperature anomalies over Lauder using the NAVGEM reanalyses. The top row shows results from the HYBRID119 reanalyses at different time cadences. At 1-h time cadence (Fig. 12b), there is excellent reproduction of the amplitude and phase of the observed temperature anomalies at all times and heights. Using a 3-h time cadence (analysis and +3-h background forecasts; Fig. 12c), there is still quite good agreement with the lidar data, although amplitudes are reduced. For a 6-h time cadence (Fig. 12d), the comparison is much poorer, most notably in the MLT where downward-propagating phase structure is lost, and local maxima and minima are both reduced and shifted in time. This aliasing problem highlights why the 6-h time cadence of analyses generated by standard forecast-assimilation update cycles at operational centers proves problematic when extending these systems through the MLT, given that, as in the DEEPWAVE region, MLT dynamics can often be dominated by large-amplitude semidiurnal tides.
Remaining panels on the bottom row of Fig. 12 show results from the other three reanalysis experiments at 1-h time cadence: 4DVAR119 (Fig. 12f), 4DVAR425 (Fig. 12g), and HYBRID425 (Fig. 12h). Again, the time–height amplitude and phase structure is captured well in the other reanalyses, although MLT amplitudes are somewhat weaker in the 4DVAR425 results.
b. MLT winds
The University of Adelaide and the Australian Antarctic Division deployed a portable meteor radar to measure MLT winds during DEEPWAVE. An “all sky” 55-MHz antenna transmitted at a peak power of 40 kW. Meteor echoes were received using a nearby five-antenna interferometer. Radial drift velocities of meteor ionization trails were derived from returned signals and used to derive wind velocities (see sections 2 and 3 of Holdsworth et al. 2004). As summarized in Fig. 13, the system was installed in Kingston, Tasmania, and observed winds in the MLT from
(a) Location of Kingston meteor radar (red) and 60-km radius within which NAVGEM fields are averaged. Orange and purple squares show locations of model grid points at T119 and T425, respectively. Meteor wind counts are plotted as (b) means vs height, (c) diurnal means vs day number, and (d) normalized means vs time of day (local time is +10 h ahead of UTC). Values at 70, 80, 90, and 100 km are color coded in each panel. DEEPWAVE deployment period of 5 Jun–21 Jul is marked in green.
Citation: Monthly Weather Review 146, 8; 10.1175/MWR-D-17-0386.1
To compare to the MLT winds observed hourly over Kingston, we reinterpolated 1-h NAVGEM reanalyses onto a 1-km geometric height grid then averaged MLT wind estimates at each height from 70 to 100 km within a 60-km great circle radius over Kingston, as shown in Fig. 13a, to mimic all-sky meteor detection out to off-zenith angles of 50°–60° [see Fig. 3 of Holdsworth et al. (2004)].






Black curves in Figs. 14a–14l show mean zonal (left column) and meridional (right column) meteor radar winds at
Time series of mean zonal and meridional winds over Kingston at heights
Citation: Monthly Weather Review 146, 8; 10.1175/MWR-D-17-0386.1
Mean biases with height are plotted in the bottom panels of Fig. 14. Results are shown using both the
Corresponding standard deviations of the wind differences are plotted in Fig. 15. The results show that the HYBRID119 reanalysis substantially outperforms the other three reanalyses in reducing MLT wind errors. To test the robustness of this result, we performed an identical standard deviation calculation using meteor radar MLT winds from a second system at Buckland Park in southern Australia (Holdsworth et al. 2004; Reid et al. 2006). The results (not shown) were very similar to those over Kingston in Fig. 15, with HYBRID119 errors systematically smaller than those from the other analyses.
Standard deviations of (a) zonal and (b) meridional wind differences between radar and reanalysis for 2-day harmonic fits (broken colored curves) and original (unfitted) 1-h fields (solid colored curves). Gray lines show zonal-mean static and flow-dependent background wind errors in NAVGEM at 43°S for July 2014.
Citation: Monthly Weather Review 146, 8; 10.1175/MWR-D-17-0386.1
Reduced errors in the HYBRID119 run likely originate from the ensemble error covariance
To assess this idea very preliminarily, we first recomputed standard deviations using reanalysis time series averaged within a 400-km great circle radius over Kingston rather than 60 km (see Fig. 13a). The corresponding plots (not shown) led to small but significant reductions in standard deviation such that HYBRID425 wind errors were less than 4DVAR errors at all heights below
5. Planetary-scale MLT dynamics
The 0–100-km NAVGEM reanalyses are used here to provide insights into planetary-scale dynamics of the MLT relevant to DEEPWAVE science. Our focus here is mostly on winds, given their primacy in controlling gravity wave propagation and breakdown through the MLT, and on the region in and around New Zealand where the core DEEPWAVE observations were acquired.
a. Split stratopause jet in July
Figure 16 profiles monthly mean zonal winds at latitudes (20°–70°S) and longitudes (140°–190°E) in and around New Zealand during June (top row) and July (bottom row).
Latitude–pressure cross sections of reanalysis zonal winds (m s−1; see underlying color bars) averaged over months of (top) June and (bottom) July. (a),(f) MERRA2 averaged from 1998 to 2017 within a DEEPWAVE zone from 140° to 190°E; (b),(g) MERRA2 for 2014 averaged from 140° to 190°E; (c),(h) NAVGEM HYBRID119 for 2014 averaged from 140° to 190°E; (d),(i) NAVGEM HYBRID119 for 2014 averaged from 0° to 360°; and (e),(j) NAVGEM HYBRID119 zonal wind anomalies at 140°–190°E relative to zonal-mean (difference of adjacent two panels to left). Region above Christchurch is shown by dashed vertical line in each panel.
Citation: Monthly Weather Review 146, 8; 10.1175/MWR-D-17-0386.1
Left panels show climatological zonal winds averaged from 140° to 190°E using 20 years of reanalysis (1998–2017) from the Modern-Era Retrospective Analysis for Research and Applications version 2 (MERRA2; Gelaro et al. 2017). The July climatological means in Fig. 16f resemble the 3-yr high-altitude reanalysis means in Fig. 1a. Adjacent panels to the right show the corresponding 140°–190°E MERRA2 zonal winds in 2014. These plots show a stratopause jet over the DEEPWAVE area that was stronger than climatology during June 2014, then split into two separated stratospheric jets in July 2014. However, since both of these features occur near the 0.1-hPa upper boundary of the MERRA2 reanalysis, it is difficult to assess the reliability of these wind features or how they might evolve into the MLT to affect propagation and filtering of gravity waves observed during DEEPWAVE.
Figures 16c and 16h show the corresponding zonal winds in 2014 from the NAVGEM HYBRID119 reanalysis. At altitudes below 0.1 hPa, the NAVGEM winds agree closely with MERRA2. However, only the NAVGEM reanalysis fully resolves the jet structure, revealing in Fig. 16c a stratopause jet peaking near 140 m s−1 at
Corresponding longitude–height cross-sections of local anomalies from the zonal mean are shown in Fig. 17, as computed from NAVGEM HYBRID119 and MERRA2 reanalyses within a 5° latitude belt centered over Christchurch; the longitude of Christchurch is shown by the red dotted line. The plots reveal that this characteristic monthly mean anomaly structure originates from quasi-stationary wave-1 Rossby wave dynamics. Top panels show monthly mean zonal-wind and geopotential-height anomalies in July 2014 from the HYBRID119 reanalysis, with the corresponding 20-yr mean anomalies from MERRA2 shown in panels beneath (middle row). All four panels reveal a sloping wave-1 pattern with similar phasing, but with the amplitudes in NAVGEM during July 2014 enhanced by factors of
Longitude–pressure cross sections of NAVGEM HYBRID119 anomalies (deviations from zonal mean) averaged from 41° to 46°S for (a) zonal wind and (b) geopotential height in July 2014, with (c),(d) corresponding mean July anomalies for years 1998–2017 from MERRA2. (bottom) Corresponding HYBRID119 anomalies for July 2014 in (e) meridional wind and (f) temperature. Region above Christchurch is shown by dashed vertical red line in each panel.
Citation: Monthly Weather Review 146, 8; 10.1175/MWR-D-17-0386.1
The corresponding wave-1-induced anomalies in meridional wind and temperature are plotted in the bottom panels of Fig. 17. Figure 17e shows that stratospheric wave-1 activity produced anomalous equatorward transport at
The NAVGEM anomalies in Fig. 17 reveal that wave-1 amplitudes attenuated significantly with height above
To study this wave-1 MLT feature in more depth, Fig. 18 shows maps of the mean wave-1 anomalies in HYBRID119 reanalysis at a representative stratospheric level of 1 hPa and an MLT level of 7 × 10−3 hPa. The wave-1 anomalies in zonal wind and geopotential height in the MLT have a very similar meridional structure to the large-amplitude wave-1 Rossby wave in the stratosphere, but are
Mean zonal-wind anomalies (departures from zonal mean) for July 2014 in the midlatitude Southern Hemisphere (20°–70°S), mapped at (a) a stratospheric pressure level of 1 hPa and (b) an MLT pressure level of 7 × 10−3 hPa, from the NAVGEM HYBRID119 reanalysis (contour labels in m s−1). (c),(d) Corresponding results for geopotential height (contour labels in meters). Coastlines are plotted in green.
Citation: Monthly Weather Review 146, 8; 10.1175/MWR-D-17-0386.1
b. Large-amplitude semidiurnal tide
Our earlier intercomparison of NAVGEM reanalysis with DLR lidar temperatures in the MLT (see Fig. 12) and some previous studies (e.g., Eckermann et al. 2016; Gisinger et al. 2017) suggested appreciable semidiurnal variation of the MLT during DEEPWAVE.
To investigate further, we computed two-dimensional space–time spectra from longitude–time (Hovmöller) cross sections of NAVGEM MLT reanalysis fields at a given latitude and height. Reanalysis fields were remapped onto a common geometric height and longitude grid, means removed, and two-dimensional power spectral densities (PSDs) formed and averaged within 5° latitude belts equatorward and poleward of the South Island for June and July. The resulting mean space–time PSDs of horizontal MLT winds for June and July are plotted in Fig. 19 at
Mean two-dimensional power spectral densities (see color bars for units) of horizontal winds at (bottom)
Citation: Monthly Weather Review 146, 8; 10.1175/MWR-D-17-0386.1
While these spectra show evidence of weak migrating diurnal and terdiurnal tides, nonmigrating tidal modes, and (at
To assess how reliable these reanalyzed MLT tidal wind fields are for DEEPWAVE science applications, Fig. 20 plots time series of the amplitudes and phases of the semidiurnal component resulting from the harmonic fitting procedures applied to time series of meteor radar and reanalysis winds over Kingston at
Plots of semidiurnal tidal features in MLT winds at
Citation: Monthly Weather Review 146, 8; 10.1175/MWR-D-17-0386.1
The colored curves show corresponding results from the various reanalysis time series over Kingston. The NAVGEM reanalyses generally perform well in capturing both the amplification and attenuation of the semidiurnal tidal winds evident in the radar observations. Likewise, the tidal phases in each wind component are captured impressively in the reanalyzed MLT winds.
Time series of the differences between each reanalysis and the radar observations reveal that the HYBRID reanalyses consistently outperform the 4DVAR reanalyses in reproducing observed properties of semidiurnal tidal winds in the MLT, in terms of both phase and amplitude. This is consistent with the better background error covariances in the MLT provided in the HYBRID runs. For example, static geostrophic coupling of observational temperature increments into rotational wind increments in the 4DVAR runs (Fig. 4f) will perform poorly whenever unbalanced (divergent) tidal motions dominate the MLT temperature variability (Fig. 4h).
We conclude from these radar–wind comparisons, as well as the lidar temperature comparisons in Fig. 12, that the NAVGEM HYBRID reanalyses provide reliable estimates of MLT winds and temperatures for assessing how gravity waves observed in the MLT during DEEPWAVE were impacted by migrating semidiurnal tides.
6. Summary and conclusions
This work has described a vertically extended configuration of NAVGEM developed specifically to address a
Separate reanalysis experiments activated 4DVAR and hybrid-4DVAR (HYBRID) data assimilation, with each configuration run at both a synoptic (T119/47) and operational (T425/119) horizontal resolution. The four global 0–100-km atmospheric reanalyses that resulted were each compared to independent observations of winds and temperatures in the MLT. Finally, the HYBRID reanalyses were used to study aspects of deep planetary-wave dynamics salient to the gravity wave–focused science goals of DEEPWAVE.
Major scientific findings of this work are as follows:
4DVAR reanalyses exhibited systematic warm biases and larger temperature errors in the MLT relative to HYBRID. These differences resulted from prespecified static errors in background MLT temperatures that were too small relative to objective estimates based on standard deviations of reanalyzed temperature differences with respect to independent MLT measurements (SOFIE and DLR lidar). By contrast, HYBRID runs weighted MLT observations and backgrounds more realistically via more representative errors in MLT temperature backgrounds from ensemble forecasts (see Fig. 10c). These findings, while specific to the greater New Zealand region during the 2014 austral winter, nevertheless suggest a need to reinvestigate and recalibrate static error variances in NAVGEM for future MLT-focused reanalysis applications.
HYBRID119 reanalyses revealed substantial reductions in MLT wind errors relative to 4DVAR and HYBRID425. These improvements originated in more realistic coupling of background wind and temperature errors via background error covariances formed from inner-loop ensemble forecasts. Sensitivity tests suggest lesser impacts in HYBRID425 may originate from unpredictable MLT dynamics at high wavenumbers that affect ensemble error covariances.
Local MLT dynamics during DEEPWAVE were dominated by large-amplitude migrating semidiurnal tides. MLT reanalyses reproduced salient aspects of observed tidal amplitudes and phases, including observed 10–15-day vacillations in tidal wind amplitudes. HYBRID reanalyses outperformed 4DVAR in reproducing observed amplitudes and phases of MLT tidal winds.
NAVGEM output at 1-h time cadence (i.e., 6-h analysis and +1–5-h outer-loop forecasts) substantially increased reanalysis skill relative to use of 6-h analysis alone, due primarily to distortion of semidiurnal tidal structures in the MLT via temporal aliasing at 6-h time cadence.
Reanalysis winds revealed splitting of the stratopause jet in and around New Zealand in July 2014 due to a large-amplitude, quasi-stationary, wave-1 Rossby wave. The high-altitude reanalysis reveals that while this wave-1 disturbance dissipated in the upper stratosphere, it reintensified in the MLT, probably via in situ generation via zonal variations in MLT GWD produced by wave-1-induced zonal variations in stratospheric gravity wave filtering.
Based on this work, we have identified HYBRID119 reanalyses as our most reliable MLT reanalysis for distribution to the wider DEEPWAVE science community, where it is already aiding modeling studies of MLT gravity waves observed during DEEPWAVE (Eckermann et al. 2016; Fritts et al. 2018). All NAVGEM reanalysis versions are available to the DEEPWAVE and wider community upon request, and HYBRID119 wind and temperature fields along DEEPWAVE flight tracks are in the process of being uploaded to the DEEPWAVE data archive housed at NCAR’s Earth Observing Laboratory (EOL).
The high-altitude NAVGEM described here extends reliable global reanalysis products into the MLT, providing new research insights into global MLT dynamics driven by wave forcing from below. More generally, validated 0–100-km global DAS capabilities within NAVGEM permit future upward extension of the operational NAVGEM to the edge of space at
Acknowledgments
NRL authors acknowledge support of the Chief of Naval Research via the base 6.1, 6.2, and platform support programs. Generation of NAVGEM reanalyses was made possible by the DoD High-Performance Computer Modernization Program via grants of computer time at the Navy DoD Supercomputing Resource Center. Collection of the Kingston meteor wind data was supported by ATRAD Pty. Ltd. and through Australian Antarctic Science Project 4025. Lidar temperatures, meteor winds, and AVAPS data were made available by NCAR/EOL under the sponsorship of the National Science Foundation via the DEEPWAVE Data Archive Center (https://www.eol.ucar.edu/field_projects/deepwave/). We thank Chris Kruse and two anonymous reviewers for helpful comments on earlier drafts.
REFERENCES
Allen, D. R., K. W. Hoppel, and D. D. Kuhl, 2014: Wind extraction potential from 4D-Var assimilation of stratospheric O3, N2O, and H2O using a global shallow water model. Atmos. Chem. Phys., 14, 3347–3360, https://doi.org/10.5194/acp-14-3347-2014.
Allen, D. R., K. W. Hoppel, and D. D. Kuhl, 2018: Extraction of wind and temperature information from hybrid 4D-Var assimilation of stratospheric ozone using NAVGEM. Atmos. Chem. Phys., 18, 2999–3026, https://doi.org/10.5194/acp-18-2999-2018.
Baker, N. L., and R. Langland, 2017: Assessing the impact of observations on the prediction of effective atmospheric angular momentum from NAVGEM. Fifth Symp. on the Joint Center for Satellite Data Assimilation, Seattle, WA, Amer. Meteor. Soc., 2.6, https://ams.confex.com/ams/97Annual/webprogram/Paper304159.html.
Ballish, B., X. Cao, E. Kalnay, and M. Kanamitsu, 1992: Incremental nonlinear normal-mode initialization. Mon. Wea. Rev., 120, 1723–1734, https://doi.org/10.1175/1520-0493(1992)120<1723:INNMI>2.0.CO;2.
Bell, W., and Coauthors, 2008: The assimilation of SSMIS radiances in numerical weather prediction models. IEEE Trans. Geosci. Remote Sens., 46, 884–900, https://doi.org/10.1109/TGRS.2008.917335.
Bossert, K., and Coauthors, 2015: Momentum flux estimates accompanying multiscale gravity waves over Mount Cook, New Zealand, on 13 July 2014 during the DEEPWAVE campaign. J. Geophys. Res. Atmos., 120, 9323–9337, https://doi.org/10.1002/2015JD023197,https://doi.org/10.1002/2015JA021568.
Buehner, M., and Coauthors, 2015: Implementation of deterministic weather forecasting systems based on ensemble–variational data assimilation at Environment Canada. Part I: The global system. Mon. Wea. Rev., 143, 2532–2559, https://doi.org/10.1175/MWR-D-14-00354.1.
Campbell, W. F., E. A. Satterfield, B. Ruston, and N. L. Baker, 2017: Accounting for correlated observation error in a dual-formulation 4D variational data assimilation system. Mon. Wea. Rev., 145, 1019–1032, https://doi.org/10.1175/MWR-D-16-0240.1.
Chua, B. S., L. Xu, T. Rosmond, and E. D. Zaron, 2009: Preconditioning representer-based variational data assimilation systems: Application to NAVDAS-AR. Data Assimilation for Atmospheric, Oceanic and Hydrologic Applications, S. K. Park and L. Xu, Eds., Springer, 307–319, https://doi.org/10.1007/978-3-540-71056-1_16.
Courtier, P., and Coauthors, 1998: The ECMWF implementation of three-dimensional variational assimilation (3D-Var). I: Formulation. Quart. J. Roy. Meteor. Soc., 124, 1783–1807, https://doi.org/10.1002/qj.49712455002.
Daley, R., and E. Barker, 2001a: NAVDAS: Formulation and diagnostics. Mon. Wea. Rev., 129, 869–883, https://doi.org/10.1175/1520-0493(2001)129<0869:NFAD>2.0.CO;2.
Daley, R., and E. Barker, 2001b: The NAVDAS source book 2001. NRL Publ. NRL/PU/7530–01-441, 163 pp., http://www.dtic.mil/dtic/tr/fulltext/u2/a396883.pdf.
Dee, D. P., and S. Uppala, 2009: Variational bias correction of satellite radiance data in the ERA-Interim reanalysis. Quart. J. Roy. Meteor. Soc., 135, 1830–1841, https://doi.org/10.1002/qj.493.
Eckermann, S. D., 2009: Hybrid σ–p coordinate choices for a global model. Mon. Wea. Rev., 137, 224–245, https://doi.org/10.1175/2008MWR2537.1.
Eckermann, S. D., 2011: Explicitly stochastic parameterization of nonorographic gravity wave drag. J. Atmos. Sci., 68, 1749–1765, https://doi.org/10.1175/2011JAS3684.1.
Eckermann, S. D., J. Ma, D. L. Wu, and D. Broutman, 2007: A three-dimensional mountain wave imaged in satellite radiance throughout the stratosphere: Evidence of the effects of directional wind shear. Quart. J. Roy. Meteor. Soc., 133, 1959–1975, https://doi.org/10.1002/qj.187.
Eckermann, S. D., and Coauthors, 2009: High-altitude data assimilation system experiments for the northern summer mesosphere season of 2007. J. Atmos. Sol.-Terr. Phys., 71, 531–551, https://doi.org/10.1016/j.jastp.2008.09.036.
Eckermann, S. D., J. P. McCormack, J. Ma, T. F. Hogan, and K. A. Zawdie, 2014: Stratospheric analysis and forecast errors using hybrid and sigma coordinates. Mon. Wea. Rev., 142, 476–485, https://doi.org/10.1175/MWR-D-13-00203.1.
Eckermann, S. D., and Coauthors, 2016: Dynamics of orographic gravity waves observed in the mesosphere over the Auckland Islands during the Deep Propagating Gravity Wave Experiment (DEEPWAVE). J. Atmos. Sci., 73, 3855–3876, https://doi.org/10.1175/JAS-D-16-0059.1.
French, W. J. R., and F. J. Mulligan, 2010: Stability of temperatures from TIMED/SABER v1.07 (2002–2009) and Aura/MLS v2.2 (2004–2009) compared with OH(6-2) temperatures observed at Davis Station, Antarctica. Atmos. Chem. Phys., 10, 11 439–11 446, https://doi.org/10.5194/acp-10-11439-2010.
Fritts, D. C., and Coauthors, 2016: The Deep Propagating Gravity Wave Experiment (DEEPWAVE): An airborne and ground-based exploration of gravity wave propagation and effects from their sources throughout the lower and middle atmosphere. Bull. Amer. Meteor. Soc., 97, 425–453, https://doi.org/10.1175/BAMS-D-14-00269.1.
Fritts, D. C., and Coauthors, 2018: Large-amplitude mountain waves in the mesosphere accompanying weak cross-mountain flow during DEEPWAVE research flight RF22. J. Geophys. Res., in press.
García-Comas, M., and Coauthors, 2014: MIPAS temperature from the stratosphere to the lower thermosphere: Comparison of vM21 with ACE-FTS, MLS, OSIRIS, SABER, SOFIE and lidar measurements. Atmos. Meas. Tech., 7, 3633–3651, https://doi.org/10.5194/amt-7-3633-2014.
Gauthier, P., and J. N. Thépaut, 2001: Impact of the digital filter as a weak constraint in the preoperational 4DVAR assimilation system of Météo-France. Mon. Wea. Rev., 129, 2089–2102, https://doi.org/10.1175/1520-0493(2001)129<2089:IOTDFA>2.0.CO;2.
Gelaro, R., and Coauthors, 2017: The Modern-Era Retrospective Analysis for Research and Applications, version 2 (MERRA-2). J. Climate, 30, 5419–5454, https://doi.org/10.1175/JCLI-D-16-0758.1.
Gisinger, S., and Coauthors, 2017: Atmospheric conditions during the Deep Propagating Gravity Wave Experiment (DEEPWAVE). Mon. Wea. Rev., 145, 4249–4275, https://doi.org/10.1175/MWR-D-16-0435.1.
Gordley, L. L., and Coauthors, 2009: The Solar Occultation for Ice Experiment. J. Atmos. Sol.-Terr. Phys., 71, 300–315, https://doi.org/10.1016/j.jastp.2008.07.012.
Guest, F. M., M. J. Reeder, C. J. Marks, and D. J. Karoly, 2000: Inertia–gravity waves observed in the lower stratosphere over Macquarie Island. J. Atmos. Sci., 57, 737–752, https://doi.org/10.1175/1520-0469(2000)057<0737:IGWOIT>2.0.CO;2.
Han, Y., P. van Delst, and F. Weng, 2010: An improved fast radiative transfer model for special sensor microwave imager/sounder upper atmosphere sounding channels. J. Geophys. Res., 115, D15109, https://doi.org/10.1029/2010JD013878.
Hogan, T. F., and Coauthors, 2014: The Navy Global Environmental Model. Oceanography, 27, 116–125, https://doi.org/10.5670/oceanog.2014.73.
Holdsworth, D. A., I. M. Reid, and M. Cervera, 2004: Buckland Park all-sky interferometric meteor radar. Radio Sci., 39, RS5009, https://doi.org/10.1029/2003RS003014.
Hoover, B. T., and R. H. Langland, 2016: NAVGEM forecast and observation impact experiments with assimilation of ECWMF analysis data in the global domain. NRL Memo. Rep. NRL/MR/7531–16-9693, 37 pp., https://doi.org/10.13140/RG.2.2.32127.82083.
Hoppel, K. W., N. L. Baker, L. Coy, S. D. Eckermann, J. P. McCormack, G. E. Nedoluha, and D. E. Siskind, 2008: Assimilation of stratospheric and mesospheric temperatures from MLS and SABER into a global NWP model. Atmos. Chem. Phys., 8, 6103–6116, https://doi.org/10.5194/acp-8-6103-2008.
Hoppel, K. W., S. D. Eckermann, L. Coy, G. E. Nedoluha, D. R. Allen, S. D. Swadley, and N. L. Baker, 2013: Evaluation of SSMIS upper atmosphere sounding channels for high-altitude data assimilation. Mon. Wea. Rev., 141, 3314–3330, https://doi.org/10.1175/MWR-D-13-00003.1.
Hortal, M., and A. J. Simmons, 1991: Use of reduced Gaussian grids in spectral models. Mon. Wea. Rev., 119, 1057–1074, https://doi.org/10.1175/1520-0493(1991)119<1057:UORGGI>2.0.CO;2.