The Global Positioning System (GPS) radio occultation (RO) technique has proven to be a powerful tool with which to analyze meteorological tropospheric features with a moderate/high spatial and temporal resolution in essentially any meteorological condition. Its ability to penetrate clouds allows retrieval of temperature (T) and water vapor pressure (e), amongst several other atmospheric variables with high vertical resolution near the surface. Vertical profiles of T are provided with an accuracy better than than 1 K (e.g., Kursinski et al., 1997; Steiner and Kirchengast, 2005; Scherllin-Pirscher et al., 2011, 2017; Kursinski and Gebhardt, 2014) in the troposphere to lower stratosphere and specific humidity with an accuracy of about 0.1 to 0.3 g kg-1 in the lower to middle troposphere. Although measurements are taken irregularly in time and space, they provide global coverage. Data are available from several missions – such as the CHAllenging Minisatellite Payload (CHAMP); Satélite de Aplicaciones Científicas-C (SAC-C); Gravity Recovery and Climate Experiment (GRACE); or the Formosa Satellite mission-3/Constellation Observing System for Meteorology, Ionosphere, and Climate mission (hereafter referred to as COSMIC) – and have been found to be of high quality and consistency in the troposphere and lower stratosphere (e.g., Steiner et al., 2011, 2013; Angerer et al., 2017; this special issue).

Biondi et al. (2011) recognized double tropopause events using bending angle (BA) anomalies derived from GPS RO measurements from different missions. Later, Biondi et al. (2015) found that the GPS RO technique is useful for understanding the thermal structure of tropical cyclones and possible overshootings into the stratosphere. The complexity of the relationship between deep convection and flow convergence over mountains has been widely studied. Demko et al. (2009) showed that, during days with deep convection, the convergence over mountains is weaker than on days when deep convection does not occur.

Over the Alps – considering the kinematic and dynamic features of divergence, flow splitting, or mesoscale vortices – it is possible to find regions which initiate or intensify the storms (Bica et al., 2007). Thermally driven flows over the Alps are associated with convergence caused by large-scale topographic heat flows (Langhans et al., 2013). These flows supply moisture from source regions close to the surface, which in turn stimulates the initiation of deep convection (e.g., Barthlott et al., 2006). Southward-oriented reliefs receive more solar radiation, resulting in a warmer atmosphere, in comparison with flat terrain. Also, the orography presents a negative energetic balance as compared to flat regions.

Extratropical regions in the Southern Hemisphere show strong wave activity close to the Andes and to the Antarctic Peninsula (e.g., Eckermann and Preusse, 1999; de la Torre et al., 2012; Hierro et al., 2013). The eastern side of the Andes at midlatitudes, between the subtropical and polar jets (Houze, 2012), is a natural laboratory for gravity waves (GWs), in particular mountain waves (MWs). The dynamic processes involved in convection over this region have been analyzed, e.g., by de la Torre et al. (2004), who found that anabatic winds act as a triggering mechanism in the presence of moist enthalpy under unstable conditions. A relationship between MWs and the development of deep convection was found by de la Torre et al. (2011). Through the design of several non-dimensional numbers related to storms development and MW energy, Hierro et al. (2013) found that MWs are able to provide the necessary energy to overcome a surface stable layer. Vertically propagating short-period GWs strongly affect the general circulation as well as the structure of the middle atmosphere (e.g., Dutta et al., 2009).

Convective activity is one of the most important sources of GWs through the release of latent heat, contributing to the interaction between waves and mean flow in the middle atmosphere (e.g., Alexander, 1995; Pandya and Alexander, 1999). When a convective cloud reaches the mean flow, waves which propagate upstream are generated (Beres et al., 2002, and references therein). Convective instability, in turn, yields oscillatory movements, which give place to GWs that propagate vertically as a harmonic oscillator (e.g., Fritts and Alexander, 2003). Several authors have analyzed the main mechanisms which describe the possible sources of gravity waves generated by convection. In the “obstacle effect”, the background finds a barrier provided by the convective flow (Clark et al., 1986). Fovell et al. (1992) proposed a mechanism where updrafts and downdrafts reach the tropopause, generating high-frequency GWs. Röttger (1980) studied penetrating cumulus convection, which generates GWs by transferring kinetic energy from the troposphere to the lower stratosphere.

Evan et al. (2012) showed that the Weather Research and Forecasting (WRF) mesoscale model (Skamarock et al., 2008) is able to simulate stratospheric GWs when it is run under actual boundary conditions. It was also possible to resolve GWs generated by convection in the tropics. Stephan and Alexander (2014), in turn, showed that WRF physics parameterizations are not decisive to obtain good results from GW simulations. From WRF simulations above the Southern Andes, de la Torre et al. (2012) detected systematic large-amplitude, stationary, nonhydrostatic GW structures, forced by the mountains up to the lower stratosphere and persisting for several hours. Their dominant modes were characterized by horizontal wavelengths (λH) of around 50 km. The vertical wavelengths (λZ) were estimated to be between 2 and 11 km. Over the Andes region, de la Torre et al. (2011) detected two main modes of mountain waves with large amplitude and high intrinsic frequency. Over the same region, Hierro et al. (2013) found stationary modes with λH between 40 and 160 km and λZ of around 7 km. De la Torre et al. (2015), analyzing storms in the presence of MWs, distinguished two different structures in vertical wind simulations. Both of them seem to be fixed to the mountains, defining systematic updraft and downdraft sectors. GW parameters were analyzed from band-pass and wavelet analysis, indicating for the cases analyzed the presence of short λH and long λZ, as expected for high-intrinsic-frequency GWs.

The motivation of the present work is twofold: first, to find a set of collocations among GPS RO BA and T profiles and mesoscale midlatitude convective systems under reasonable conditions of proximity in space and time over orographic regions (Alps and Andes). We use this data set to test the utility of GPS RO observations for the detection of convective systems. To detect the cloud top altitude from the RO profiles, we apply a technique based on the anomaly in the BA. Secondly, the GW structures are analyzed and discussed for two selected case studies detected in the absence of jets and fronts, from high-resolution mesoscale model simulations and reanalysis data. The possible determination or misleading interpretation of GW parameters from GPS RO is discussed in detail for these case studies. Section 2 outlines the RO data used and the methodology applied and describes the two subsets of RO events retrieved during convective activity close to the Alps and Andes mountain ranges. In Sect. 3, one case study at each region is selected, and relevant GW features are analyzed from the simulation, the reanalysis data, and the measurement of both RO events. In Sect. 4, conclusions are given.

The utility of RO observations for the investigation of convective systems (Biondi et al., 2012, 2015) and GWs over orographic regions in Europe and South America (Alps and Andes Mountains) is tested. It is known that a sharp spike in RO BA is highly correlated with the top of the cloud, corresponding to anomalously cold temperatures within the cloud. Above the cloud, the temperature returns to background conditions, and a strong inversion appears above the cloud top. For usual tropospheric cloud tops, the T lapse rate within the cloud often approaches a moist adiabat, consistent with rapid undiluted ascent within the convective systems.

We built a collocation database between RO observations and mesoscale convective systems over subtropical to midlatitude mountain regions. The selected regions for the Alps and Andes are 40–55∘ N, 0–20∘ E and 20–40∘ S, 60–74∘ W, respectively (Fig. 1). We use RO data processed at the Wegener Center for Climate and Global Change (WEGC) with the Occultation Processing System (OPS) version 5.6 (Schwärz et al., 2016), based on excess phase and orbit data version 2010.2640 from the University Corporation for Atmospheric Research (UCAR) from the CHAMP, SAC-C, GRACE, and COSMIC missions. We analyze BA and T profiles which are available from near the surface up to 40 km altitude with 100 m vertical sampling.

Convective systems are located in time and space using the global deep convective tracking database of the International Satellite Cloud Climatology Project (ISCCP) (Rossow et al., 1996), from January 2006 to July 2008. This period was chosen due to the constraints and limitations imposed by the ISCCP database and the COSMIC data. The first source, available between 1983 and 2008, is currently incomplete and being re-processed. The global ISCCP data set, with a horizontal grid resolution of 30 km and a nominal time resolution of 3 h, is based on brightness temperatures from geostationary satellite measurements. It provides information on the location and extent of mesoscale deep convective cloud systems and their properties. The parameters extracted from ISCCP data are time of occurrence, center (mass center), and radius of the storm. The COSMIC mission started in June 2006.

The selection criterion applied in the present work considers the position of the RO observation with respect to the center of the storm, thus providing 294 and 50 collocations in the Alps and Andes regions, respectively. According to this criterion, it is observed whether the latitude and longitude corresponding to the mean tangent point (TP) belonging to each RO profile are located within a radius of 100 km with respect to the center of the storm, as provided by ISCCP. A maximum time difference of 3 h was allowed between each RO event and the data from ISCCP. The collocated events were selected using cloud data from geostationary satellites METEOSAT (Europe) and GOES (South America) (Fig. 2).

The collocated RO BA and T profiles were also used to determine the vertical structures of subtropical convective systems over orographic regions. In order to detect the cloud top altitude with RO, we applied the anomaly technique developed by Biondi et al. (2013) atmospheric BA profiles for cloud top detection of convective systems. Each BA and T profile collocated with a storm was referenced against a RO background reference climatology profile, which was extracted for the same location and the same month from the global RO BA and T reference climatology, respectively (for details see Biondi et al., 2017). We then subtracted the collocated RO reference climatology profile from the individual RO profile. BA was normalized with respect to the reference climatology profile in order to obtain a fractional anomaly profile. The cloud top altitude is represented as a pronounced spike in the vertical BA anomaly structure and, correspondingly, in the T anomaly profile.

Taking into account the wave signatures observed in the RO T profiles (see below), GWs were analyzed from two different data sources: mesoscale model simulations and European Centre for Medium-Range Weather Forecasts (ECMWF) Reanalysis Interim (ERA-Interim) data. The WRF simulations (Skamarock et al., 2008) were performed using 1∘ × 1∘ National Center of Environmental Prediction (NCEP) Global Final Analysis (FNL) as boundary conditions. They are conducted in four nested domains (27, 9, 3, and 1 km) with 60 vertical levels. A sponge layer was applied in the upper 3 km. The size of the inner domain in the Alps region is about 300 km × 200 km, and that of the Andes region is about 300 km × 300 km. For each domain, the microphysical schemes used were the following: WRF single-moment 6-class microphysics (WSM6; Hong et al., 2004); Yonsei University (YSU; Hong et al., 2006) to represent the planetary boundary layer (PBL) physics; Rapid Radiative Transfer Model Longwave (RRTM; Mlawer et al., 1997) and MM5 Dudhia Shortwave (Dudhia scheme; Dudhia, 1989) for radiation processes; and the Noah Land Surface Model (developed jointly by NCAR and NCEP; Skamarock et al., 2008) and Monin–Obukhov scheme (Monin and Obukhov, 1954) for surface physics and thermal diffusion processes, respectively. The cumulus parameterization used was the new Grell scheme (Grell3; Grell and Devenyi, 2002) for the first two domains, while non-cumulus parameterization was selected for the two inner ones.

In the present study, a GW climatology from the limited number of available collocated cases is not intended. Instead, by focusing on selected large-amplitude RO T profiles, wave features in both mountain subtropical regions are compared. In doing so, five collocations were pre-selected in each region (Fig. 3a–b). All of the pre-selected collocations show some spikes in the BA profile, from which one is correlated with the cloud top of the corresponding convective structure. Large-amplitude oscillations possibly associated with hydrostatic and/or nonhydrostatic GW structures are evident. The analysis is further conducted in two peculiar case studies (central panels in Fig. 3a–b), defined by the simultaneous absence of jets and fronts.

To confirm this scenario, we analyzed a possible imbalance in the flow between mass and momentum, able to generate inertia-gravity waves through geostrophic adjustment, as the atmosphere tries to restore the equilibrium (see, e.g., Zhang et al., 2000; Zhang, 2004; Plougonven and Zhang, 2014). In doing so, we considered different available methods. Each of them involves the calculation of a specific parameter, with its advantages and disadvantages (cross-stream component of the Lagrangian Rossby number (Ro-), Psi vector, generalized omega equation, nonlinear balance equation). Following de la Torre at al. (2006), we analyzed the Ro- distribution from reanalysis and from the simulated geopotential and velocity data. When a geostrophic imbalance (e.g., Fritts and Alexander, 2003) coexists with other sources, Ro-, defined by the ratio of the component of the ageostrophic wind normal to the flow to the observed wind speed, is expected to be greater than 0.5, and the analysis should be more intricate. In the present case studies, Ro- remains < 0.5 (not shown) at jet pressure levels, and the GW structures are conceivably limited to orographic and/or convective sources. The RO case studies exhibit attractive oscillatory features and correspond, in the Alps region, to 2 February 2008 at 17:24 UTC (47.29∘ N, 12.02∘ W) and in the Andes region to 19 December 2006 at 16:56 UTC (25.35∘ S, 67.37∘ W).

The GWs in the two selected study cases in the Alps and Andes are investigated from WRF numerical simulations, ERA-Interim reanalysis data, and collocated GPS RO observation profiles.

From an initial set of collocations between convective systems and GPS RO observations, the applicability of these data sets for the detection and investigation of convective systems and GWs over orographic regions was analyzed. In doing so, mountain regions of Europe and South America, over subtropical to midlatitude regions, the Alps and Andes mountain ranges, were selected. A collocation database was built up. We used RO bending angle and T profiles retrieved at the Wegener Center with processing version OPSv5.6. The storm systems were located in time and space according to the global deep convective tracking database ISCCP. From an initial set of 294 and 50 collocations at Alps and Andes regions, respectively, a subset of 10 collocations was pre-selected. This pre-selection was based on the observation of large amplitudes, presumably GWs, in the retrieved T RO profiles. Two case studies were finally studied in detail, one in the Alps region and one in the Andes region. The case studies were investigated using mesoscale WRF simulations, ERA-Interim reanalysis data, and measured RO T profiles. The case studies considered were selected based on the absence of jets and fronts, in order to be able to filter out two relevant possible sources of GWs. Similar GW regimes and dominant vertical and horizontal wavelengths, from convective and orographic origin, were found at both regions. MWs reach higher altitudes above the Alps than close to the Andes. The background mean wind above the latter region imposes a critical level for mountain wave propagation, preventing the propagation of GW packets up to stratospheric heights.

In the Alps, mostly bi-dimensional GW structures with constant phase surfaces are seen. The mean horizontal wind causes the apparent forcing of MWs. Prevailing amplitudes and λH ranging from 1 to 2 m s-1 and 20 to 60 km, respectively, are distinguished. GW amplitudes are weaker below than above the cloud tops, and two different structures are visible. One structure is stationary and the other is not stationary. The GW sources seem to be orographically forced or associated with cloud development. δw amplitude values up to 2 m s-1 correspond to MWs with short horizontal wavelengths. These δw perturbations exhibit, as expected, the presence of MWs more clearly than δT. By contrast, the non-stationary GWs with longer λH and amplitude values above 2 K are more evident in δT than in δw. Systematic λZ values close to 8 km, associated with these longer λH, are observed. The orographic amplitudes are more significant early in the morning, exhibiting a general decrease with increasing local time. They reach large amplitudes at stratospheric heights beyond the tropopause, located at 11 km in the Alps case. No critical levels or reflection effects are observed. The strongest orographic structures are observed until the early afternoon. The non-stationary GW packets are generated during the convection development after mid-afternoon and are radiated above the cloud tops. The longer λH values are not well defined, suggesting the coexistence of two or more non-stationary modes.

The observed RO T profile was first filtered, and then a CWT was applied to the remaining δT. A clear GW signal throughout the tropo–stratospheric region, with prevailing λZ = 4 km, was found. The correspondence between these two modes and the structures observed in the simulations was investigated, considering the expected amplitude attenuation effects in the RO sounding.

It is concluded that the LOS and the wavelength ratio should prevent the observation of vertical oscillations corresponding to short λH structures. This clearly does not benefit GW detection during the RO event. The horizontal averaging of RO retrievals produces an amplitude attenuation and phase shift in any plane GWs, which may lead to significant discrepancies with respect to the original values. An estimation of the expected attenuation in the stationary and non-stationary structures during the RO sounding was performed. The amplitude attenuation factor defined as the ratio between derived and original amplitudes was deduced. Ideal conditions that lead to no attenuation are a null ratio between vertical and horizontal wavelengths and a null angle between LOS and the fronts. The case study analyzed shows the corresponding values of both parameters and the attenuation factor. In this Alps case study, considerable attenuation of mountain waves below and above the tropopause is expected. The resulting estimation of a negligible attenuation factor confirmed that these GWs cannot be captured during the RO event. As the mesoscale simulations are not enough to capture and explain the observed GW structure, corresponding ERA-Interim reanalyses data were investigated. From these data, a defined non-stationary oscillation propagating in a NW–SE direction was observed, explaining the oscillation seen in the RO δT profile.

At the Andes region, coherent bi-dimensional GW structures with constant phase surfaces oriented from SW to NE, of a type already reported in a previous work, are seen. Additional non-orographic GW structures situated in the NE and SE of the domain are observed. They propagate from lower altitudes until at least the upper limit of the simulations. These structures exhibit stationary circular wave fronts and penetrate beyond the critical layer for orographic GWs, situated at an almost zero wind level, near the tropopause at about 18 km height. As in the Alps case, δw perturbations exhibit the presence of MWs more clearly than δT. The presence of two different GW structures, separated by a critical layer at 18 km, is identified. The MWs propagate upwards, and just below the critical layer their amplitude increases; at the same time λZ decreases. Inversely to the Alps case study, the orographic amplitudes in the Andes region are, as usual, more significant in the afternoon, exhibiting a general increase with increasing local time. Above the critical layer, non-stationary GW packets are generated during the convection development after mid-afternoon and are radiated above the cloud tops. In this case, there is almost no spatial coexistence between GWs from both GW sources.

In the retrieved RO δT profile a signal appears also propagating throughout the tropo–stratospheric region with a prevailing λZ = 4.5 km. An evaluation of the partial attenuation coefficient reveals that, in the troposphere, MWs can be captured during the RO event. In this estimation, λX, λY, and λZ values equal to 30, 100, and 10 km, respectively, are considered. For the Andes case study, the mesoscale simulations explain the observed GW structure above the critical layer also. Clear signatures of λX, λY, and λZ equal to 350, 400, and 3 km, respectively, may be seen. These values yield a very high attenuation factor (0.99). It is concluded that the oscillation is entirely due to orographic waves below the tropopause and due to convective GWs above it. In both cases, mesoscale simulations were able to capture the GW structures. In this case study, as in the Alps case, additionally, an expected distortion in the measured λZ value must be foreseen. This is due to the slanted nature of the sounding and also depends on the GW aspect ratio.