Atmospheric gravity waves have a substantial impact on weather and climate. They transport energy and momentum, contribute to turbulence and mixing, and influence the mean circulation and thermal structure of the middle atmosphere . Low-frequency and long wavelength gravity waves can be explicitly resolved in mesoscale model simulations, whereas global circulation models typically require parametrization schemes to represent effects of gravity waves on subgrid scales . The development of gravity wave parametrization schemes is challenging, because gravity waves are excited by various sources, each having individual characteristics. Two prominent sources of gravity waves are orographic generation and convection . Other sources include adjustment of unbalanced flows in the jet streams and frontal systems . Another source, body forcing accompanying localized wave dissipation, is likely to occur commonly in the middle atmosphere . The individual characteristics of the gravity wave sources and the alterations of the gravity wave spectrum with altitude-dependent wind and stability variations are important research topics.

In the stratosphere gravity waves from convective sources are generally most important in the summer hemisphere, where planetary wave activity is weak . In the winter hemisphere orographic and jet sources play a more important role, and small-scale orographic hotspots may provide a significant contribution to the total gravity wave drag that is currently not well-represented in global climate models . More comprehensive observations may help to develop and improve parameterizations to better incorporate the wave drag even for such small sources. In this study we analyze satellite observations of stratospheric gravity wave activity at 18 orographic hotspots located in the Southern Hemisphere. The study closely follows recent work of , which analyzed the seasonal cycle of orographic gravity wave occurrence above remote islands in the southern oceans. Further motivation to study stratospheric gravity wave activity at mid- and high latitudes during winter arises from the fact that gravity waves play an important role in the formation of polar stratospheric clouds (PSCs). Localized temperature fluctuations associated with gravity waves can yield stratospheric temperatures below the threshold values for PSC formation, even if synoptic-scale temperatures are too high . , , , , and used comprehensive satellite observations to study the impact of mountain waves at high latitudes on PSC formation.

Satellite instruments offer excellent opportunities to study gravity waves on a global scale. In this study we focus on nadir scanning observations of AIRS aboard NASA's Aqua spacecraft. The main advantage of nadir sounders such as AIRS is good horizontal resolution and coverage. The disadvantage is that the nadir measurement geometry limits the observations to gravity waves with rather long vertical wavelengths (λz≳ 15 km for AIRS) due to the “observational filter” effect . However, observations of gravity waves with long vertical and short horizontal wavelengths are of particular interest, because these waves can potentially carry large momentum flux and excite significant wave drag . AIRS radiance measurements have successfully been exploited in a number of gravity wave studies. For instance, , , , , , and demonstrated the capabilities of AIRS to observe mountain waves at orographic hotspots such as the Antarctic Peninsula, the Andes, the Greenland topography, or the Himalayas. and also analyzed global long-term records of stratospheric gravity wave activity from AIRS observations. By September 2015 AIRS had completed 13 years of measurements and gathered about 13.8×109 infrared radiance spectra, which can be used to explore the climatological variability of stratospheric gravity wave activity.

This study focuses on stratospheric gravity wave activity from orographic sources in the Southern Hemisphere, which is of particular interest in relation to the dynamical behavior of the Southern Hemisphere polar vortex. The analysis is based on a 12-year record (January 2003–December 2014) of 4.3 µm radiance observations of AIRS/Aqua. Stratospheric gravity wave signals in terms of brightness temperature perturbations and variances are extracted by applying a number of standard techniques developed for nadir sounders . We introduce a simple and effective new method to detect orographic gravity wave signals from infrared nadir sounder measurements. To infer the orographic wave signals this method analyzes brightness temperature variance differences between two boxes located up- and downstream of an orographic hotspot. The method is used to estimate the occurrence frequencies of mountain waves at 18 orographic hotspots in the Southern Hemisphere based on the long-term AIRS record. Furthermore, interactions between the mountain wave activity and tropospheric and stratospheric background winds are studied. To predict the occurrence of mountain wave events in the AIRS observations we propose a simple model based on zonal wind thresholds in the lower troposphere and in the mid-stratosphere. Our approach uses similar criteria as a model presented by that was used to quantify stratospheric gravity wave activity above Scandinavia. However, the present model does not consider wind turning with height as we focus on southern hemispheric conditions with generally weaker planetary wave activity diverting the stratospheric winds from nearly pure westerlies. The main purpose of our model is to provide a means of separating upper level wind effects, like the observational filter, from low-level effects, like those related to the gravity wave sources. This will allow the model to be used to discuss whether waves are likely present or affecting the atmosphere even though they are only weakly observed or invisible in the AIRS observations.

In Sect.  we provide a brief description of the AIRS instrument and the methods used to extract brightness temperature perturbations related to stratospheric gravity waves from the radiance measurements. In Sect.  we introduce the method to detect and discriminate between gravity wave signals from orographic or other sources. Seasonal mean occurrence frequencies of orographic gravity waves at various hotspots based on the 12-year AIRS record are discussed in Sect. . Correlations between gravity wave activity and tropospheric and stratospheric background winds are discussed in Sect. . In Sect.  we also introduce the threshold model to predict the occurrence of mountain wave events in the AIRS observations. Section  focuses on inter- and intraseasonal variability of mountain wave activity at the hotspots and discusses the performance of the threshold model in explaining this variability. In Sect.  we provide conclusions and an outlook on how the results of this study might be used in future research.

AIRS is one of six instruments aboard NASA's Aqua satellite. Aqua was launched in a nearly polar, low earth orbit (705 km altitude, 100∘ inclination, 100 min period) in May 2002. Nearly global coverage is achieved during 14.4 orbits per day. The Aqua orbit is sun-synchronous, with Equator crossings at 01:30 LT (descending orbit nodes) and 13:30 LT (ascending orbit nodes). AIRS measures infrared radiance spectra from the Earth's atmosphere in the nadir and sub-limb geometry. Each across-track scan covers 1780 km ground distance and consists of 90 footprints. The scans are separated by 18 km along-track distance. The footprint size varies between 14×14 km2 at nadir and 21×42 km2 at the scan extremes. AIRS measurements cover the 3.74–15.4 µm spectral range in three bands, with a resolving power of λ/Δλ=1200. We analyze measurements from multiple channels in the 4.3 µm spectral region, with a noise equivalent delta temperature (NEDT) of 0.13–0.15 K at 250 K scene temperature.

We infer information on stratospheric gravity wave activity directly from the AIRS radiance measurements following the approach of and . We analyze spectral mean brightness temperatures in the 4.3 µm CO2 fundamental band (2322.5–2346.0 and 2352.5–2367.0 cm-1), which gets optically thick in the mid-stratosphere. Temperature kernel functions for the 4.3 µm channels show a broad maximum in sensitivity of the radiances to stratospheric temperatures at 30–40 km altitude and have a full-width at half-maximum of about 25 km . The broad kernel functions limit the AIRS observations to gravity waves with long vertical wavelengths. We found that the 5, 20, and 50 % response levels to wave amplitude are first exceeded at 16, 32, and 48 km vertical wavelength, respectively . The observed brightness temperatures are mainly composed of three contributions: (i) gravity wave signals, (ii) slowly varying background signals, and (iii) measurement noise. Background signals associated with large-scale temperature gradients or planetary waves are removed with the detrending procedure of , , and , i.e., brightness temperature perturbations are calculated as differences from a 4th-order polynomial fit for each across-track scan. This limits the amplitude response to 90, 50, and 20 % at 800, 1200, and 1650 km across-track wavelength, respectively . The short wavelength limit of the observations is at about 30 km, based on the Nyquist theorem and a sampling distance of 14 km at nadir. The noise of the spectral mean brightness temperatures is about 0.059 K at 250 K scene temperature . The 4.3 µm brightness temperature variances shown in this paper have been corrected for noise, by subtracting noise variances scaled to scene temperature.

Climatological studies based on AIRS and other satellite observations revealed that stratospheric gravity wave activity at mid- and high latitudes during the winter season is closely linked to orographic hotspots and jet sources . Figure  shows the 2003–2014 multi-annual seasonal mean of detrended and noise-corrected AIRS 4.3 µm brightness temperature variances in the Southern Hemisphere. Here we focus on the time period from mid-fall to mid-spring (April–October), when stratospheric gravity wave activity in the Southern Hemisphere is largest. Figure  also shows terrain variability from a 2-min gridded global relief data set (ETOPO2v2; ). The standard deviation of terrain altitudes is one of the parameters considered in gravity wave parametrization schemes for subgrid-scale orographic sources . The AIRS and ETOPO2v2 maps show local maxima or “hotspots” of stratospheric gravity wave activity being clearly associated with orographic features. The strongest hotspots are found at large mountain ranges, such as the Andes, the Antarctic Peninsula, and New Zealand. Many small-scale hotspots are also evident, e.g., at some of the remote islands in the southern oceans. The small-scale hotspots are visible due to the high horizontal resolution of the AIRS observations. Based on these maps we selected 18 hotspots of stratospheric gravity wave activity that are more closely examined in this study (Table ). Note that some prominent hotspots at the border of East Antarctica are not considered here. In these places gravity waves are triggered by katabatic winds from mainland Antarctica , which is a rather different source mechanism from those in the other places.

In addition to the orographic hotspots, the variance map in Fig.  shows a broad zonal band of stratospheric gravity wave activity around 50–70∘ S. A pronounced maximum of gravity wave activity within this latitude band is found leeward of the Andes and the Antarctic Peninsula, extending as far as 150∘ E. The origin of this broad maximum is not entirely clear as the region of enhanced activity extends well beyond the reach of the direct effect of orography. It may be caused by propagating mountain waves , but also by non-orographic sources in winter storm tracks such as spontaneous adjustment, frontogenesis, and convection . Figure  shows 2003–2014 April–October seasonal mean winds from the ERA-Interim reanalysis at the AIRS observational level (3 hPa, about 40 km) and at low level (750 hPa, about 2 km). The stratospheric gravity wave activity observed by AIRS is closely linked to the winds at both levels. The activity of the orographic sources is directly coupled to the strength of the surface winds, as strong surface winds are needed for waves to be launched. For AIRS to be able to observe the waves, strong background winds in the stratosphere are needed to foster the propagation of gravity waves with long vertical wavelengths, to which AIRS is most sensitive due to its observational filter.

In this paper we introduce a simple and effective approach, referred to as the “two-box method”, to detect gravity wave activity at orographic hotspots from the AIRS measurements. In this method we examine the variance of detrended 4.3 µm brightness temperature perturbations in two boxes, located upstream and downstream of an orographic hotspot. We assume primarily westerly winds, so the western edge of the downstream box includes the hotspot and the box then extends to the east. The variance σe2 of this box is considered to primarily be influenced by signals from orographic gravity waves. The upstream box is located to the west of the hotspot and is not placed directly adjacent to the downstream box, but is slightly separated to reduce the likelihood of capturing orographic wave activity in this box. Orographic waves typically propagate downstream, so the variance σw2 in this box should not be affected by waves from the hotspot. The upstream box provides information on the background levels of gravity wave activity, being related to other sources. The presence of orographic wave activity is then determined from the difference in variance between these two boxes, calculated as σoro2=σe2-σw2. The transfer of background variances from the upstream to the downstream box introduces some uncertainties in this analysis. However, large variance differences σoro2 most likely relate to the occurrence of orographic waves. We cope with the uncertainties of the method by introducing a variance threshold σ02 and by considering only those events exceeding the threshold, σoro2≥σ02, as being related to the orographic source. Note that we applied the method to noise-corrected brightness temperature variances σe2 and σw2, but due to the difference approach it also bears the potential to provide effective noise correction itself.

Figure  shows examples of orographic wave events at selected hotspots detected with the two-box method. The events shown here are among those with the largest σoro2 values that we found in the 12-year record of AIRS data and in all cases the wave patterns clearly indicate orographic wave activity at the hotspots. As can be seen from the maps, we have chosen the box positions and sizes individually for each hotspot. Common box sizes for all hotspots may be desirable, in principle, regarding the wavelength sensitivities of the method. However, we found that individual optimization of the box sizes to the typical size of the wave patterns at the hotspots improves the detection rates. Large boxes were used for strong hotspots producing extensive wave patterns such as the Andes and New Zealand (with box sizes of 10∘×8∘ in longitude × latitude) and the Antarctic Peninsula (15∘×5∘). Mid-size boxes were used for Kerguelen and Tasmania (6∘×6∘) as well as Crozet and South Georgia (5∘×5∘). Small boxes were used for Balleny and Peter I Island (5∘×3∘), located at high latitudes, and the remaining hotspots from Table  (3∘×3∘), located at mid-latitudes. For most of the hotspots the mean latitude of the boxes was chosen to match the latitude of the hotspot. However, for Heard we applied a latitudinal shift, to stay away from orographic waves created at Kerguelen (cf. Fig. ). The longitudinal separation between the western and eastern boxes was 1∘ for all hotspots.

In order to validate the two-box method we performed two tests that compare the results of this automatic detection of orographic wave events with statistics from visual inspection of AIRS brightness temperature maps. In order to identify orographic waves in as objective and consistent a manner as possible, the visual inspections follows criteria defined by . In particular, there must be a clear difference in the wave pattern near the hotspot to distinguish orographic waves from waves from other sources, i.e., the location of the hotspot should be clearly indicated by the position of the wave pattern. Furthermore, if the observation includes both an orographic wave and a larger-scale background wave pattern, there must be a distinct change in the pattern directly adjacent to the hotspot. For the first test we used the automatic detection method to select the three events for each year and each hotspot which had the largest σoro2 values. This gave us 3×12×18=648 individual events, for which we inspected the AIRS images to verify that orographic wave activity was visible at the hotspot. The performance of the two-box method varied between the hotspots. The largest success rates were found for the Andes (100 %) and the Antarctic Peninsula (100 %), followed by Kerguelen (93 %), New Zealand (91 %), Balleny (88 %), Heard (78 %), South Georgia (77 %), and Tasmania (74 %). For the remaining hotspots the success rates were below 54 % and became as low as 6 % for Macquarie. This test indicates that the two-box method performs best for strong hotspots with frequent wave activity and large values of σoro2. The success rates clearly correlate with the peak altitude of the hotspots (see Table ). Results for weak hotspots with low success rates should be considered more carefully, because those are more likely to be influenced by gravity waves from non-orographic sources.

As a second test we compared the detection results from the two-box method with the event statistic of . The study of analyzed gravity wave activity at Auckland, Heard, Kerguelen, Prince Edward, South Georgia, and Tasmania during the years 2003 and 2004. Orographic wave events were identified by visual inspection of AIRS 15 µm brightness temperature perturbation maps. The vertical coverage of the AIRS channel analyzed in that study (667.8 cm-1) is at slightly higher altitudes (around 35–45 km) than in this work. Noise levels of the 15 µm data are about a factor of 7 larger than the 4.3 µm data used here. This introduces some uncertainty when we compare the detection results. Considering the data of as “observations” and the results from the two-box method as “predictions”, we calculated a set of skill scores to assess the performance of the two-box method. In particular, we analyze the Gilbert skill score (GSS), which is also known as “equitable threat score” . This is a standard verification method for dichotomous (yes/no) model predictions. It takes into account the probability of detection (POD) and the false alarm rate (FAR) of the model and is adjusted for hits associated with random chance. The GSS analysis allows for method verification and can help to establish the variance threshold σ02. Here we analyzed the GSS for two choices of the variance threshold, σ02=0.1 K2 and σ02=1 K2. These values define a reasonable range of thresholds. The total number of events decreases significantly for thresholds much larger than 1 K2 and the method would not be applicable for some of the smaller hotspots with weaker wave activity at all. Choosing thresholds much lower than 0.1 K2, this would include many events with rather low wave amplitudes that may not be too important overall or that are possibly affected by measurement noise. Using a variance threshold of σ02=1 K2, we found a bias (ratio of predictions / observations) of 12 %, a POD of 11 %, a FAR of 5 %, and a GSS of 7 %. With such a large variance threshold the two-box method missed many of the weaker events identified by . This leads to a low POD, but also to a good FAR. The fact that the GSS is larger than zero indicates that the method still does have skill compared to a random prediction. Choosing a threshold of σ02=0.1 K2 improves the skill scores of the method substantially. The bias is then 69 %, the POD is 57 %, the FAR is 18 %, and the GSS is 33 %. Future work may focus on fine-tuning of the variance threshold, including possible optimization for the individual hotspots. However, for this study we decided to focus on events characterized by a globally constant variance threshold to allow us to compare the results of the different hotspots to each other. We selected σ02=0.1 K2 since with this value more events are included and the method has better skill than when using σ02=1 K2.

In this section we discuss the seasonal mean occurrence frequencies of stratospheric gravity waves at the orographic hotspots. As a first step we calculated histograms of the variances in the eastern and western boxes at each hotspot (Fig. ). In these histograms increased numbers of events in the eastern boxes point to more frequent orographic wave activity. To quantify the increase, we calculated the ratio ne/nw of the numbers of events below and above the identity line, respectively. A large ratio ne/nw indicates that the occurrence of orographic waves is more likely. Figure  and Table  show that ne/nw is largest for the Andes (20.8), followed by the Antarctic Peninsula (6.9), Kerguelen (6.0), Heard (3.4), Prince Edward (3.4), Tasmania (2.8), South Georgia (2.6), New Zealand (2.4), Auckland (2.3), Balleny (2.1), and Tristan (2.1). For most of the remaining hotspots ne/nw ranges from 1.3 to 1.9, indicating that orographic wave activity is less frequent. For South Orkney the ratio is 0.9, i.e., gravity wave activity in the upstream (western) box exceeds that in the downstream (eastern) box. This is due to the western box often being influenced by orographic waves from the Antarctic Peninsula. In principle, the ratio ne/nw provides a simple way to select hotspots that are well suited to study orographic wave activity. However, this selection can be further optimized by considering uncertainties in measurement coverage and uncertainties in the confidence level at which orographic waves are detected, as will be discussed below.

The total number of events in the histograms in Fig.  depends on the number of satellite overpasses at each hotspot. Usually there are two satellite overpasses per day at each location, but this varies with latitude. At the equator there are regular data gaps between the AIRS swaths from neighboring overpasses, these gaps become narrower with increasing latitude. The AIRS swaths start to overlap at ±45∘ latitude. At high latitudes there is significant overlap of the swaths so that there may be four, or even more, overpasses per day, which can be analyzed. The area observed during an overpass varies with each orbit and does not always cover the entire area of a box. In our analysis we considered only those overpasses that covered at least 50 % of the area of the boxes, to ensure the variance calculations were robust. To cope with the variability in measurement coverage, we focus on occurrence frequencies, i.e., fractions of overpasses showing gravity wave activity with respect to the total number of overpasses, rather than event counts in the rest of the paper. Table  provides the 2003–2014 April–October seasonal mean occurrence frequencies fgw of all observed gravity waves at each hotspot. These were calculated using only the information in the eastern box, and applying a variance threshold of σe2≥0.1 K2. The occurrence frequencies fgw vary greatly between the hotspots, from 4.8 % for Tristan to 59 % for the Andes. Table  also presents the occurrence frequency foro of orographic waves determined with the two-box method with σ02=0.1 K2 for each hotspot. The occurrence frequencies foro vary from 1.5 % for Tristan to 53 % for the Andes. Furthermore, Table  presents the ratio foro/fgw as a measure of the contribution of orographic wave activity to total gravity wave activity as observed by AIRS at each hotspot. The ratio foro/fgw varies from 24 % for Gough to 90 % for the Andes. It is important to note that the absolute values of fgw and foro largely depend on the choice of σ02. For instance, raising σ02 from 0.1 to 1 K2, fgw and foro typically decrease by a factor of 5–15. However, while the frequencies changed, the ratio foro/fgw remained nearly constant and we found that the rankings between different hotspots in terms of any of these measures – foro, fgw, and the ratio foro/fgw – are largely independent of the choice of σ02. The ratio foro/fgw provides a good way to select hotspots that are best suited to study orographic wave activity.

Note that the occurrence frequency foro of orographic waves is expected to grow with the terrain peak altitude, because taller mountains cause larger vertical displacements in the flow and so will generate larger amplitude waves. A statistical association between terrain peak altitudes and gravity wave occurrence frequencies was also found from Table . The peak altitudes listed here are local maxima of the ETOPO2v2 data in the eastern boxes considered for the two-box method. The maxima are representative for 2-min horizontal grid resolution and can be lower than actual mountain peak heights cf. Table 1 of. We calculated the Spearman rank-order correlation coefficient ρs between the terrain peak altitudes and the seasonal mean occurrence frequencies fgw and foro of the hotspots. We found a medium degree of correlation (ρs=0.39) using fgw and a high degree of correlation (ρs=0.70) using foro. This indicates that the two-box method effectively identifies orographic wave events, for which occurrence frequencies are closely linked to terrain altitude. Another important factor controlling the occurrence frequencies fgw and foro are the background winds in the troposphere and stratosphere, which will be discussed in the next section.

In this section we discuss yearly and monthly variations of the orographic wave activity at the hotspots. We focus on results for the Antarctic Peninsula and Kerguelen Islands, being representative examples of a mountain ridge and a peak, respectively. Figure  shows 2003–2014 monthly mean occurrence frequencies of the orographic waves from AIRS observations and the prediction model. In addition, Fig.  shows monthly occurrence frequencies of the ERA-Interim zonal winds exceeding the thresholds at the 2 and 40 km levels at the hotspots. The occurrence frequencies of the zonal winds were calculated using the wind thresholds defined in Sect.  and Table . A clear seasonal variation is found in the monthly occurrence frequencies, with minima of 1–12 % in April and October and maxima as large as 62 % in July. For the Antarctic Peninsula we found a rather long season, with occurrence frequencies exceeding the 50 % level from May to September. At Kerguelen the 50 % level is exceeded only in June and July. The intraseasonal variations of the orographic waves clearly follow the occurrence frequencies of the zonal winds at the observational level. Those cover ranges of 14–96 % at the Antarctic Peninsula and 0–88 % at the Kerguelen Islands. For both the Antarctic Peninsula and Kerguelen the occurrence frequencies of the low-level winds are in the range of 60–80 % from April to October on average, which indicates a high chance for orographic waves being excited at all times. The prediction model reproduces the monthly variations of the observed occurrence frequencies with a mean absolute error of 5 percentage points. A larger error of 15 percentage points was found only for Kerguelen Island in June, and seems to be related to an overestimation of the influence of the observational level wind on the wave activity.

Figure also presents the seasonal mean occurrence frequencies at the Antarctic Peninsula and Kerguelen Islands for individual years from 2003 to 2014. The time series reveal substantial interannual variations of the occurrence frequencies, covering ranges of 33–52 % at the Antarctic Peninsula and 10–38 % at Kerguelen Islands. The annual variations of the gravity wave occurrence frequencies are again found to be closely correlated with the occurrence frequencies of the zonal winds. The example for the Antarctic Peninsula indicates that even though winds at the observational level are often most influential, the low-level winds are still important. This becomes most evident during the years 2005 to 2010. Given that the wind at the observational level remains high during this time, the occurrence of gravity waves then clearly follows the low-level winds. This shows that both levels are required to predict AIRS observations of orographic waves. The prediction model reproduces the interannual variations of the seasonal occurrence frequencies with a mean absolute error of 4 percentage points. The absolute errors became as large as 12 percentage points (Antarctic Peninsula) and 9 percentage points (Kerguelen) in individual years. The large interannual variability indicates that to get statistically meaningful results the occurrence frequencies should be calculated based on long-term records such as those provided by AIRS.

In this paper we introduced the simple and effective two-box method that can be used to detect orographic gravity waves in infrared nadir sounder imagery. The method was applied to 12 years of AIRS/Aqua observations to analyze mountain wave activity during April to October at 18 orographic hotspots in the Southern Hemisphere. The seasonal mean mountain wave activity was most frequent over the Andes (with an occurrence frequency of 53 %), followed by the Antarctic Peninsula (43 %), Kerguelen (25 %), South Georgia (24 %), Heard (23 %), Balleny (17 %), and less than 13 % in other places. At many hotspots mountain waves contribute significantly to the total gravity wave activity as observed by AIRS. Contributions are as large as 90 % at the Andes, followed by the Antarctic Peninsula (76 %), Kerguelen (73 %), Tasmania (70 %), New Zealand (67 %), Heard (60 %), and other hotspots (24–54 %). Mountain wave occurrence frequencies are closely correlated with terrain peak altitudes (ρs=0.70). Orographic gravity wave variances are also strongly correlated with the zonal background wind at 40 km altitude (with ρs varying between 0.6 and 0.8), which we attributed to the AIRS observational filter. However, in a recent study of gravity wave measurements by a Raleigh/Raman lidar at Lauder, New Zealand, also found enhanced correlation of the observed stratospheric gravity wave activity with the stratospheric winds. Weaker correlations are found with respect to low-level winds at 2 km altitude (with ρs varying between 0.2 and 0.4), but this may be mostly due to the fact that the low-level winds at the hotspots were rarely below the threshold required for launching waves. The range of low-level winds for New Zealand found here agrees well with the range of tropospheric winds for which found the largest gravity wave energies at mesospheric altitudes, supporting the findings of the present paper. We developed a simple model that predicts the occurrence frequencies of mountain waves in AIRS observations based on zonal wind thresholds at 2 and 40 km altitude. This prediction model has significant skill (GSS of 10–42 %). It reproduces yearly and monthly variations of the mountain wave occurrence frequencies at the Antarctic Peninsula and Kerguelen which vary from near zero to over 60 % with mean absolute errors of 4–5 percentage points.

Our results on the seasonal cycle of gravity wave activity at Southern Hemisphere hotspots and correlations with the background winds agree well with those by . Use of the orographic wave detection algorithm developed here, and the thresholds found for upper and lower level wind, could permit extension of their wave flux estimates to include geographic and interannual variability more comprehensively. This would allow us to better characterize the collective effect of these waves on the circulation of the Southern Hemisphere stratosphere. Although terrain peak altitudes and zonal background winds are most closely correlated with mountain wave occurrence frequencies, there are many other factors that influence the excitation, propagation, and observability of these waves. These include the following: (i) source variations such as terrain roughness, slope, and orientation with respect to the surface winds, (ii) wind variations between different height levels and between the zonal and meridional components, and (iii) observational effects related to the AIRS measurement geometry, e.g., the orientation of the wave fronts with respect to the line of sight. Future work should aim for improved understanding of these effects. The two-box method and the prediction model based on wind thresholds introduced here can be used to identify interesting case studies in the vast amount of AIRS data, improving the usefulness of the data for future research on mountain waves and their impact on atmospheric dynamics.

AIRS data products are distributed by the NASA Goddard Earth Sciences Data Information and Services Center . ERA-Interim data are provided by the European Centre for Medium-Range Weather Forecasts . The ETOPO2v2 data set was obtained from the US Department of Commerce, National Oceanic and Atmospheric Administration, National Geophysical Data Center .