REVIEW Tropical Temperature Variability in the UTLS: New Insights from GPS Radio Occultation Observations

: Global positioning system (GPS) radio occultation (RO) observations, ﬁrst made of Earth’s atmosphere in 1995, have contributed in new ways to the understanding of the thermal structure and variability of the tropical upper troposphere–lower stratosphere (UTLS), an important component of the climate system. The UTLS plays an essential role in the global radiative balance, the exchange of water vapor, ozone, and other chemical constituents between the troposphere and stratosphere, and the transfer of energy from the troposphere to the stratosphere. With their high accuracy, precision, vertical resolution, and global coverage, RO observations are uniquely suited for studying the UTLS and a broad range of equatorial waves, including gravity waves, Kelvin waves, Rossby and mixed Rossby–gravity waves, and thermal tides. Because RO measurements are nearly unaffected by clouds, they also resolve the upper-level thermal structure of deepconvection andtropicalcyclones aswell asvolcanicclouds.Theirlowbiasesandstabilityfrommissionto missionmake RO observations powerful tools for studying climate variability and trends, including the annual cycle and intraseasonal-to-interannual atmospheric modes of variability such as the quasi-biennial oscillation (QBO), Madden–Julian oscillation (MJO), and El Ni ñ o–Southern Oscillation (ENSO). These properties also make them useful for evaluating climate models and detection of small trends in the UTLS temperature, key indicators of climate change. This paper reviews the contri- butions of RO observations to the understanding of the three-dimensional structure of tropical UTLS phenomena and their variability over time scales ranging from hours to decades and longer.


Introduction
The thermal structure and variability of the tropical upper troposphere-lower stratosphere (UTLS) are fundamental to many aspects of the climate system. Tropical temperatures decrease with height from the surface to the tropopause near 17 km (Fig. 1) and then increase with height in the stratosphere. The UTLS region refers to the broad altitude layer over ;12-22 km, spanning the transition from the relatively well-mixed convective troposphere to the stably stratified lower stratosphere. The chemical composition also changes sharply across this layer, including species such as ozone, carbon monoxide, and water vapor. A somewhat narrower layer centered around the tropical tropopause, that is, ;14-19 km, is also referred to as the tropical tropopause layer (TTL; e.g., Fueglistaler et al. 2009;Randel and Jensen 2013). Temperatures in the UTLS are strongly coupled to large-and small-scale circulations and deep convection (with likely two-way coupling) and are important for controlling UTLS cirrus and aerosol behavior and, hence, radiative balances. Further, temperatures at the tropical Denotes content that is immediately available upon publication as open access. tropopause exert a strong control on global stratospheric water vapor via dehydration of air entering the stratosphere across the cold tropopause (Brewer 1949;Mote et al. 1996;Randel and Park 2019).
Tropical UTLS temperatures are characterized by large variability in space and time, especially near the tropopause, as illustrated by the example profiles in Fig. 1. Much of the variability occurs with relatively small vertical scales, associated with a broad spectrum of large-and small-scale waves and various atmospheric circulations with a range of diurnal-tointerannual time scales ( Fig. 1) (e.g., Salby and Garcia 1987;Alexander 1996). This thermal variability is clearly observed in high-resolution radiosonde observations (Tsuda et al. 1994;Kim and Alexander 2015) but is less well resolved in nadirsounding satellite measurements or meteorological analyses/ reanalyses with typical vertical resolution of 1 km or greater. In contrast, Global Navigation Satellite System (GNSS) radio occultation (RO) measurements provide high-verticalresolution (of order 100 m) soundings and thus sensitivity to short-vertical-scale UTLS features (Noersomadi and Tsuda 2017;Zeng et al. 2019). The dense sampling and two-decadelong RO record provide novel information on UTLS thermal behaviors and their links to processes spanning the wide range of scales indicated in Fig. 1. RO observations, with their small biases, also serve as ''anchor'' observations in forecast models and reanalyses and are assimilated in several reanalyses without bias corrections, unlike satellite radiance measurements (Fujiwara et al. 2017). This review is intended to summarize this information and demonstrate the added value of RO data with regard to the tropical UTLS in combination with other measurements and analyzed meteorological products.

Radio occultation data
a. Occultation geometry and retrieval RO is an active remote sensing technique that exploits electromagnetic signals transmitted by a GNSS satellite at two (or more) L-band microwave frequencies with wavelengths of about 20 cm. GNSS includes GPS, Globalnaya Navigatsionnaya Sputnikovaya Sistema (GLONASS), and other systems, but so far mainly GPS signals have been used for RO measurements. These signals travel through the atmosphere, where they are refracted due to ionospheric and neutral atmospheric vertical density gradients (e.g., Melbourne et al. 1994;Syndergaard 1999;Yunck et al. 2000). A receiver on a satellite in low-Earth orbit (LEO) records the signals in an oblique vertical scan of the atmosphere due to the relative motion of the two satellites . RO geometry is shown in Fig. 2.
Basic observables are the GNSS signals' excess phases (excess compared to propagation in vacuum), which can be converted to bending angle profiles. Linearly combining the bending angles that are derived from the two L-band frequencies (L1 and L2 bending angles) removes most of the ionospheric contribution to the measurement. Atmospheric refractivity is obtained from the neutral atmospheric bending angle. This step requires accurate information about the atmospheric state at high altitudes (e.g., above 55 km; Ao et al. 2006), which is usually not available from RO measurements as noise overshadows the weak RO signal at high altitudes. Therefore, some a priori information about the atmospheric state at high altitudes is used in this step. Refractivity in the neutral atmosphere depends mainly on temperature, pressure, and water vapor pressure. In the UTLS region, however, the amount of water vapor is negligibly small. There, temperature can be computed directly from refractivity (Kursinski et al. 1997;Leroy 1997;Scherllin-Pirscher et al. 2017b). In the moist atmosphere, temperature and humidity are retrieved with additional background information (Ware et al. 1996;Rieckh et al. 2018;Li et al. 2019). Details about the retrieval process are given by Melbourne et al. (1994), Kursinski et al. (1997), Syndergaard (1999), Hajj et al. (2002), and Steiner et al. (2011).
Mean fields of atmospheric parameters are obtained by gridding and averaging (e.g., Foelsche et al. 2008a) or statistical Bayesian interpolation (Leroy et al. 2012) of individual profiles. Alternatively, mean profiles of bending angle are propagated through the retrieval Gleisner and Healy 2013;Danzer et al. 2014bDanzer et al. , 2018. Climatological geopotential fields can then be used to derive averaged wind fields outside the tropics using geostrophic or gradient wind approximations (Scherllin-Pirscher et al. 2014;Verkhoglyadova et al. 2014;Healy et al. 2020).

b. Data characteristics
Advantages of RO measurements are manifold. Data characteristics are summarized in Table 1. The occultation geometry (Fig. 2) yields high-vertical-resolution profiles (e.g., Noersomadi and Tsuda 2017) with an effective vertical resolution ranging from 0.1 km near the tropical tropopause to about 1.4 km in the polar lower stratosphere (Zeng et al. 2019). A caveat is that the horizontal averaging inherent in RO (over ;200 km) and altitude variations along the ray path can combine to limit the effective vertical resolution to ;1 km, which is important for atmospheric wave analyses. Due to the use of microwave signals, measurements can be performed during day and night and penetrate through clouds. Since RO measurements are based on precise atomic clocks, they are traceable to the International System of Units (Leroy et al. 2006b). Exploitation of relative rather than absolute phase measurements additionally ensures self-calibration, long-term stability, and high precision of measurement profiles. Measurements are therefore independent of the satellite and data from different satellites can be combined to create consistent long-term RO records. In the UTLS region, long-term stability and high precision are also valid for derived atmospheric RO variables as shown by Hajj et al. (2004), Schreiner et al. (2007), Foelsche et al. (2011b), Steiner et al. (2011), Angerer et al. (2017), and Steiner et al. (2020a. Uncertainty (random errors) of individual RO profiles is smallest in the UTLS region, where it amounts to approximately 0.7 K for dry temperature (Scherllin-Pirscher et al. 2011b). Comparisons to other measurements, analyses, and reanalyses have demonstrated the high accuracy (low biases) of RO atmospheric profiles in the UTLS (e.g., Schrøder et al. 2003;Wang et al. 2004;Kuo et al. 2005;Gobiet et al. 2007;Ho et al. 2007;Steiner et al. 2007;Foelsche et al. 2008a). In most cases, a large part of the difference was attributed to biases of the comparison datasets, such as the varying quality of different radiosonde types (e.g., He et al. 2009;B.-R. Wang et al. 2013), biases in daytime radiosonde measurements (Sun et al. 2010;Ladstädter et al. 2015;Tradowsky et al. 2017) or changes in analysis or reanalysis systems (Angerer et al. 2017;Shangguan et al. 2019;Ho et al. 2020). Important for all these comparison studies is to appropriately minimize representativeness and sampling errors of all correlative datasets (e.g., Ladstädter et al. 2011;Feltz et al. 2014;Gilpin et al. 2018). On a long-term basis, residual biases from ionospheric contributions to the RO measurement, which depend on solar activity, cannot be neglected as they can be as high as 2 K at 30 km during high solar activity (Danzer et al. 2013). Application of an ionospheric error model , however, can significantly reduce this error (Danzer et al. 2015).
For climate applications, consistent long-term data records are required. RO measurements are stable over time and intersatellite biases are small (,0.1 K between 8 and 25 km; Angerer et al. 2017). However, background information from the high altitude initialization enters the retrieval at the refractivity level and background biases can propagate down into the stratosphere and affect processing results. Early RO intercomparison studies of Ao et al. (2003), von Engeln (2006, Staten andReichler (2008), andLöscher et al. (2009) found considerable differences among RO products from different processing centers. Close cooperation among major RO processing centers started in 2007 to systematically investigate these differences and to improve data processing. Studies of Ho et al. (2009Ho et al. ( , 2012 and Steiner et al. ( , 2020a significantly increased knowledge and understanding of these differences and revealed larger differences above approximately 25 km resulting from varying high-altitude initialization methods. Even though differences between processing centers have decreased over time, merging data from different processing schemes for climate applications is not recommended, but rather one should use data from an ensemble of RO records.

c. RO missions and processing centers
The first RO measurements of Earth's atmosphere were obtained by the proof-of-concept mission GPS/Meteorology  Self-calibration and long-term stability Kursinski et al. (1997), Danzer et al. (2014a), and Angerer et al. (2017) The measurement and the early retrieval stages are based on phase changes rather than on absolute phases; therefore, the measurements are stable over the long term and no data calibration is needed; data from different satellites can be combined to a single record Limitations: (i) Long-term stability of retrieved data is degraded by the use of background information in the retrieval process; (ii) variations of solar activity, which can leave a nonnegligible ionospheric residual; and (iii) the amount of humidity, which affects RO dry atmosphere parameters / Data stability is highest from 8 to 30 km Random error variance or standard deviation Significantly decreases with a large number of profiles for mean atmospheric fields / Highest precision/smallest uncertainty (;0.7 K for individual profiles) in the UTLS (GPS/MET) in 1995 (Ware et al. 1996;Hocke 1997;Rocken et al. 1997;Feng and Herman 1999). After GPS/MET successfully demonstrated the capability of RO, a number of RO satellite missions were launched in the following decade, significantly increasing the available number of measurements. Single satellites Satellite de Aplicaciones Cientifico-C (SAC-C; Hajj et al. 2004) and Challenging Minisatellite Payload (CHAMP; Wickert et al. 2001aWickert et al. ,b, 2003 (Luntama et al. 2008;von Engeln et al. 2009). COSMIC increased the global number of daily profiles from about 200 to more than 2000 (Jin et al. 2011;Angerer et al. 2017). Additional measurements from the Terra Synthetic Aperture Radar at X Band (TerraSAR-X), TerraSAR-X Add-On for Digital Elevation Measurement (TanDEM-X), and Communications/ Navigation Outage Forecast System (C/NOFS) (see https:// earth.esa.int/web/eoportal/satellite-missions for a summary of these missions) increased measurement density in the late 2000s. Recently launched MetOp-C, Fengyun-3C (Sun et al. 2018), and COSMIC-2 (Anthes and Schreiner 2019; Schreiner et al. 2020) ensure the continuity of the RO record to be used in atmospheric and climate sciences for the next five years at least.
Depending on the RO mission and retrieval product, RO data are made available by the Danish Meteorological Institute (DMI; http://www.romsaf.org), EUMETSAT (http://www.eumetsat.int), the German Research Centre for Geosciences (GFZ; http:// www.gfz-potsdam.de/en/section/space-geodetic-techniques/topics/ gnss-radio-occultation/), Jet Propulsion Laboratory (JPL; https:// genesis.jpl.nasa.gov/genesis/), COSMIC Project Office, University Corporation for Atmospheric Research (UCAR; http://cdaacwww.cosmic.ucar.edu), and the Wegener Center for Climate and Global Change, University of Graz (WEGC; http://www. wegcenter.at). Information on the consistency and structural uncertainty of multisatellite RO data from the different processing centers is given by Steiner et al. ( , 2020a including a detailed description of the RO retrieval and of the center-specific processing steps.

Tropopause structure
A detailed understanding of the processes that determine the temperature structure of the tropical tropopause is important for our understanding of stratospheric water vapor content and vertical wave propagation [with potential effects on the associated wave forcing of stratospheric winds, e.g., related to the quasi-biennial oscillation (QBO)]. The RO data are especially helpful in characterizing the tropical tropopause where highvertical-resolution data (e.g., radiosondes) are sparse, meteorological analyses are not well constrained Tegtmeier et al. 2020), and climate models show large uncertainties (Gettelman et al. 2010;Kim et al. 2013).

a. Thermal tropopause
In the tropics the transition from the troposphere to the stratosphere does not take place across a single interface but rather across a layer, the so-called tropical tropopause layer (Fueglistaler et al. 2009). The TTL roughly extends from the top of main convective outflow near 150 hPa (;14 km in altitude) to the highest level of convective tops near 70 hPa (;18.5 km in altitude). Near the bottom of the TTL the temperature lapse rate starts to depart more strongly from a moist adiabat, whereas near the top of the TTL stratospheric temperature lapse rates are observed and static stability maximizes.
Although the concept of the TTL is physically more appropriate than a single tropopause interface, various tropopause definitions (i.e., thermal, dynamical, and chemical tropopauses) are still widely used in the literature. The thermal tropopauses, including the lapse-rate tropopause (LRT) and coldpoint tropopause (CPT), are based on the temperature structure of the TTL and well defined across the tropics, with the CPT perhaps being the more physically relevant level because of its direct connection to the amount of water vapor that may enter the stratosphere. A dynamical tropopause based on a potential vorticity (PV) isosurface, however, is not applicable near the equator Structural uncertainty (uncertainty arising from different choices in processing of the same raw data) Ho et al. (2009Ho et al. ( , 2012 and Steiner et al. ( , 2020a Smallest structural uncertainty in the UTLS; ,0.1 K from 8 to 30 km where PV switches sign from negative values in the Southern Hemisphere to positive values in the Northern Hemisphere. The LRT is conventionally defined as the lowest level at which the temperature lapse rate drops below 2 K km 21 , provided that its average between this level and all higher levels within 2 km remains below this threshold (WMO 1957). Multiple LRT levels have been diagnosed in the tropics (e.g., Schmidt et al. 2006;Randel et al. 2007) but are generally rare when compared with the subtropics and midlatitudes (e.g., Rieckh et al. 2014;Wilhelmsen et al. 2020).
In the case of RO data, a distinct tropopause level may also be identified based on the structure of the bending angle profile (Narayana Rao et al. 2007;Lewis 2009;Vespe et al. 2017), which has the advantage that it is based on raw data independent of uncertainties in the retrieval procedure.
Unlike throughout the free tropical troposphere, temperature near the tropical tropopause shows strong spatial and temporal variability (e.g., Randel et al. 2003;Scherllin-Pirscher et al. 2017a). This may be heuristically understood based on the long radiative time scales (;60 days) that are observed near the tropical tropopause: a seemingly unimportant heating rate of 0.1 K day 21 could result in a temperature perturbation of ;6 K over 60 days (cf. Birner and Charlesworth 2017). The RO data have been instrumental in quantifying these sources of variability from observations with quasi-homogenous coverage throughout the tropics. This was first demonstrated by GPS/ MET Randel et al. 2003), extended to CHAMP , and later further extended to COSMIC (Son et al. 2011). The CPT is simply the level corresponding to the minimum temperature in a given profile and is generally located somewhat higher than the LRT (e.g., Munchak and Pan 2014). The RO data reveal that CPT temperatures show a pronounced minimum over the western Pacific warm pool with values often dropping below 190 K, especially during boreal winter (Fig. 3). This minimum is consistent with the general tendency of a reduction in CPT temperature above regions of active deep convection [Randel et al. 2003; cf. low outgoing longwave radiation (OLR) values in Fig. 3]. Likewise, CPT temperatures are higher above regions of suppressed deep convection, such as above the eastern Pacific Ocean. In general, the distribution of CPT temperature reflects the large-scale quasi-stationary temperature response to convective heating (Matsuno 1966;Gill 1980, see also section 6). The RO data have also been used as a baseline for large-scale spatial variations as well as subseasonal, seasonal, and longer time-scale variations in CPT temperature and height.
Much of the temporal variability in CPT characteristics on subseasonal time scales is associated with wave perturbations, such as those that arise due to equatorial gravity and Kelvin waves (Randel et al. 2003). The Madden-Julian oscillation (MJO) also gives rise to variations in CPT characteristics Virts and Wallace 2014). Waves have been further shown to cause enhanced CPT temperature minima affecting cirrus formation (e.g., Mehta et al. 2011Mehta et al. , 2013Kim and Alexander 2015;J.-E. Kim et al. 2016;Son et al. 2017).
On interannual time scales the main drivers of CPT variability are El Niño-Southern oscillation (ENSO), QBO, and the occurrence of sudden stratospheric warmings (via corresponding modulation of tropical upwelling within the Brewer-Dobson circulation), all of which have been quantified using RO data (Rieckh et al. 2014). Estimating decadal-scale tropopause variations from RO data is limited by its comparatively short record (about 20 years). Shorter-term trends have nevertheless been considered, such as for the CHAMP period from 2001 to 2007 (Schmidt et al. , 2010bW. Wang et al. 2013).

b. Tropopause inversion layer
The tropopause inversion layer (TIL) is characterized by strongly increasing temperature with height and therefore strongly enhanced values of static stability just above the local tropopause. In the tropics, a broad layer of enhanced static stability exists between the tropopause and ;23 km altitude (Grise et al. 2010) that is fairly well resolved in reanalyses and models (Birner et al. 2006). However, a strong static stability maximum, which is confined to a thin layer just above the local tropopause (;1 km), can only be detected in high-verticalresolution data such as RO and radiosondes (Bell and Geller 2008). Unlike radiosondes, which are sparsely distributed in the tropics, RO data allow a quantification of characteristics of such a thin TIL throughout the tropics (Grise et al. 2010;Schmidt et al. 2010a;Son et al. 2011). Figure 4 shows that the zonal-mean tropical TIL exists throughout the year, is strongest close to the equator, and shows strong seasonality at the edge of the tropics with enhanced values in the respective summer hemisphere. During boreal winter secondary stability maxima occur centered near 19 km over 108-208N and 108-208S, which are related to seasonally varying quasi-stationary tropical waves (e.g., Dima and Wallace 2007). On interannual time scales, Kumar et al. (2014) have shown that the tropical TIL is stronger during the westerly phase of the QBO. Pilch Kedzierski et al. (2016) have furthermore documented a secondary thin static stability maximum just below the zero wind line associated with the descending westerly shear phase of the QBO and have found the enhanced TIL strength associated with deep convective outflow and equatorial waves. Noersomadi et al. (2019) have quantified TIL strengthening due to enhanced convective activity associated with the MJO, ENSO, and monsoon circulations.

Clouds and convection
The high vertical resolution, insensitivity to weather conditions and the global coverage make RO well suited to study temperature changes due to extreme weather and volcanic clouds. These phenomena were not the objectives of the first RO missions, but quickly became one of the focuses when scientists understood the importance of the high vertical resolution to study cloud structure and dynamics.

a. Deep convection
During convective development, the environmental temperature decreases from the surface to the midtroposphere with a lapse rate close to the climatological temperature profile. In the cloud, the lapse rate increases and the temperature reaches a minimum near the cloud-top height, where a rapid inversion may reestablish the standard lapse rate (Biondi et al. 2012;Bonafoni and Biondi 2016). The largest negative temperature anomaly corresponds to the cloud-top height of the deepest convection (reaching heights greater than 17 km) with the magnitude decreasing for lower cloud-top heights down to 15 km (Paulik and Birner 2012). When deep convection does not reach the standard tropopause height, the presence of the cloud can create a secondary tropopause associated with the inversion at the cloud top (Biondi et al. 2012;Shi et al. 2017) as shown in Fig. 5. In particular cases, such as deep convection developing in the western Pacific during an active MJO, a thermal vertical dipole appears with large horizontal scale characterized by warming in the upper troposphere and cooling near the tropopause and lower stratosphere, which enhances dehydration of the lower stratosphere (Kim et al. 2018). However, the temperature anomalies in the lower stratosphere can sometimes differ in case of tropopausepenetrating convection or nonpenetrating convection (Xian and Fu 2015) and in different regions of the globe (Johnston et al. 2018). Colocation with rainfall measurements from the Tropical Rainfall Measuring Mission (TRMM) has shown that tropopause-penetrating convection is characterized by warm anomalies in the midtroposphere, cold anomalies at the tropopause and lower stratosphere with a rapid lift of the LRT height. In contrast, nonpenetrating convection shows cooling at the tropopause and warming in the lower stratosphere (Xian and Fu 2015;Johnston et al. 2018).
A diurnal temperature anomaly due to the nonmigrating tides and overshooting turrets is usually evident during strong convection (Khaykin et al. 2013;Johnston et al. 2018) with amplitudes proportional to the strength of the convection and more evident over land in the southern tropics (Khaykin et al. 2013). Land convection cools the lower stratosphere more than ocean convection (Johnston et al. 2018). The temperature of the lower stratosphere in summer over land shows a diurnal cycle collocated with the most intense convective activity, with afternoon cooling also corresponding to the maximum frequency of overshooting in the area.

b. Tropical cyclones
The use of RO also allows improved understanding of tropical cyclone (TC) thermal structure, including the differences between TCs developing in different areas (Bonafoni et al. 2019). Although the RO profiles have a horizontal averaging scale of approximately 250 km, most of the information in the RO profile represents smaller horizontal scales closer to 50 km (Anthes et al. 2000), and hence capture an important part of the TC temperature anomalies. The RO profiles confirm the already well-known TC warm core, but also highlight the cold TC cloud top (Biondi et al. 2011a) and the presence of a double tropopause in some specific cases (Biondi et al. 2011b). The collocation of RO profiles with radiosonde profiles during TCs (Biondi et al. 2013) shows a cooling at the TC cloud-top height with a strong inversion layer over the cloud top. Rivoire et al. (2016) showed that the cooling at the tropopause level precedes the formation of the TC warm core and speculated about a positive feedback for the intensification of the storm.
A comprehensive study by Biondi et al. (2015) characterized the TC thermal structure by ocean basin and intensity collocating the RO profiles with the TCs' center coordinates in a time window of 6 h and a space window of 600 km. The results showed different features according to the ocean basins (see Some studies limited to the north Indian Ocean (Ravindra Babu et al. 2015;Ratnam et al. 2016) evaluated the impact of TCs on the tropopause parameters at different distances from the TC center. They found the largest effect on the tropopause height and temperature within 500 km from the center. The tropopause height was a minimum over the center and gradually increased outward, while the tropopause temperature increased outward from the center to beyond 500 km. The closest profiles to the TC centers also show multiple tropopauses, confirming the results of previous studies (e.g., Biondi et al. 2015).
A comprehensive archive of collocations between TCs and RO profiles is provided by Lasota et al. (2020).
c. Volcanic cloud impacts  were the first using RO data for studying the impact of the Mount Chaiten 2008 eruption on atmospheric temperature. Volcanic clouds in the UTLS modify the thermal layer structure immediately after the eruption, producing a cooling at the cloud-top height (Okazaki and Heki 2012;Biondi et al. 2017;Cigala et al. 2019). The buoyancy perturbation produced by the eruption pushes up the tropopause level relative to the standard atmosphere in the same area and season (Biondi et al. 2017;Prata et al. 2020) and creates conditions in which the altitude distance between the CPT and the LRT increases by a few kilometers (Prata et al. 2020) instead of a few hundred meters as in standard conditions. The volcanic effect in the UTLS can be different for different eruptions: Nabro in 2011 produced a quick warming of the lower stratosphere (Biondi et al. 2017), whereas Anak Krakatau in 2018 warmed the upper troposphere and cooled the lower stratosphere (Prata et al. 2020).
It has been recently suggested that even relatively small volcanic eruptions can impact the climate (Solomon et al. 2011). Mehta et al. (2015)

Gravity waves
Gravity waves (GWs) have scales too small to be realistically described by data assimilation in global reanalysis datasets. However, collectively they contribute substantially to the global circulation momentum budget, particularly at levels in the upper troposphere and above (Alexander and Ortland 2010;Geller et al. 2013). GW temperature anomalies also lower the cold-point temperature of air rising through the TTL, modulating cirrus cloud occurrences and lowering the concentration of water vapor entering the stratosphere (Jensen and Pfister 2004;Kim and Alexander 2015). While GW breaking and cloud seeding processes are localized in space and time, the collection of many GW events has global effects on general circulation and surface temperature (Richter et al. 2010;Kim and Alexander 2015). Therefore, while research work on a wide variety of GW processes requires detailed local measurement of wave properties such as temperature anomalies, wavelengths, phase speeds, and momentum fluxes, global coverage is also needed to address these issues.
RO temperature profiles contain information on shortvertical-scale GWs. Many early studies attempted to isolate GWs from the large-scale temperature structure by performing a polynomial fit to individual RO temperature profiles to define a ''background temperature profile'' and then interpreting the residual temperature anomalies as those caused by GWs (de la Ratnam et al. 2004;de la Torre et al. 2006;Alexander et al. 2008a;Schmidt et al. 2008a;de la Torre et al. 2010;Luna et al. 2013;Chane Ming et al. 2014;Nath et al. 2015). This definition of background temperature allows for retrieval of information on shortvertical-scale temperature anomalies, which include all waves with a vertical wavelength typically in the range ;2-10 km (Steiner and Kirchengast 2000;Tsuda et al. 2004Tsuda et al. , 2011Wright et al. 2011). Tsuda et al. (2000) mapped the variance of small-verticalscale temperature anomalies with this method using GPS/ MET RO temperature retrievals. The global map of November-February averaged wave potential energy (PE) at 20-30 km derived from these temperature anomalies showed a clear peak in energy at low latitudes, as well as a winter season enhancement compared with the summer season at extratropical latitudes. Tsuda et al. (2000) suggested that the tropical peak shows waves generated by convection and recognized that these small-vertical-scale anomalies could include contributions from global-scale tropical wave modes as well as smaller-scale GWs. Indeed, Holton et al. (2001) subsequently found global-scale Kelvin waves with very short vertical wavelengths of 3-4.5 km and temperature amplitudes up to 6 K. Alexander and Ortland (2010) found tropical Rossby and mixed Rossby-GW (MRGW) modes with vertical wavelengths as short as 4 km and amplitudes near 2 K. de la  performed an analysis similar to Tsuda et al. (2000), but reported short-verticalwavelength (,10 km) variances in multiple years of SAC-C and CHAMP RO temperature profiles. The same tropical enhancement is evident, but with a clear interannual variation related to the QBO in tropical winds. These results are consistent with a mixture of global-scale tropical waves and smaller-scale GWs interacting with QBO wind shear zones, since vertical wavelengths grow small as waves approach a critical level where the wave phase speed equals the wind speed.
GWs may develop very long vertical wavelengths when stratospheric winds become strong enough. A simplified form of the GW dispersion relation shows this: where m is the vertical wavenumber [2p/(vertical wavelength)], N is the buoyancy frequency, U is the horizontal zonal wind speed upstream of the direction of wave propagation, and U 2 c is the intrinsic phase speed. The equation indicates that, with N ' 0.02 s 21 , vertical wavelengths may exceed 10 km (the limit often imposed by a vertical fit to define a ''background temperature'') whenever the intrinsic phase speed exceeds 32 m s 21 . Zonal winds commonly exceed 32 m s 21 during midwinter. In the tropics, studies applying a vertical polynomial fit to define the background temperature for RO will include planetary-scale waves, and may exclude important convectively generated GW signals with high phase speeds, particularly those propagating upstream against QBO winds where jU 2 cj is large. Relating vertically high-pass-filtered PE to wave sources has sometimes indicated correlations with tropical deep convection (Kawatani et al. 2003;Ratnam et al. 2004;Chane Ming et al. 2014) or topography (Noersomadi and Tsuda 2016;Rapp et al. 2018). However, quantitative information about wave properties that could improve parameterization of GW effects on circulation has been difficult to acquire with these methods . The parameterizations need gravity wave momentum fluxes, which require determination of the three-dimensional structure of gravity waves to derive it from temperature measurements (Ern et al. 2004).
One estimate of the contributions to the short-verticalwavelength signal of GWs versus longer-period planetaryscale waves appears in Kim and Alexander (2015). Analyzing tropical radiosondes, they found that vertical wavelengths including all wave types in the TTL peak at ;4 km, but only ; 1 /4 of the temperature variability is contributed by GW frequencies, while ; 3 /4 is contributed by waves with periods longer than 3 days. Comparing COSMIC and radiosonde subseasonal-to-interannual variations in wave-induced CPT variability showed that the two datasets were quantitatively very similar, suggesting that RO captures most or all the important short vertical wavelength waves contributing to CPT variability.
Other methods for defining the large-scale, zonal-mean, and planetary-scale wave temperature structure have since been developed, and can be applied to improve the separation of GW signals from other planetary-scale waves, while also avoiding the low-pass vertical filtering that was eliminating many important long-vertical-wavelength GWs. Horinouchi and Tsuda (2009) studied unique sets of near-instantaneous quasi-linear groupings of COSMIC temperature profiles during the first December-February following the satellite launch (2006-07). Focusing on groupings at midlatitudes in boreal winter, they were able to identify propagation directions, and found a preference for northward over southward propagation indicating more wave sources at lower latitudes. J.-E. Kim et al. (2016) found clear relationships between wave-induced cooling and thin cirrus occurrence. Studies comparing RO density perturbations to temperature perturbations in Sácha et al. (2014) andPi soft et al. (2018) showed that both methods represent a high-pass filter eliminating vertical wavelengths of ,8 km.
A number of studies used RO temperatures to define a background for analyses of residual variance as GWs (Randel and Wu 2005;Alexander 2009, 2010;McDonald 2012;Alexander et al. 2013;Faber et al. 2013;Schmidt et al. 2016a;Xu et al. 2018). These methods appropriately distinguish GWs and planetary-scale waves by horizontal wavelength and/or period rather than by vertical wavelength. With this method, Xu et al. (2018) showed that GW PE (Fig. 7) does not peak directly over the equator but shifts to the tropical summer hemisphere over continental landmasses where convection is strongest.
Success in identifying GW horizontal wavelengths, propagation directions, and momentum fluxes have been derived by removing planetary-scale waves, followed by selection of special subsets of RO profiles that are close in space and time.  and Schmidt et al. (2016a)  limb-sounding temperature profiles that are close in space and time to retrieve three-dimensional properties of GWs, dramatically improving estimates of GW horizontal wavelengths and momentum fluxes.  combined pairs of high-resolution dynamics limb sounder (HIRDLS) profiles together with a nearby COSMIC profile, while Schmidt et al. (2016a) used RO from multiple missions to find close triads. Both obtained very similar zonal-mean values of GW momentum fluxes that peak in the winter extratropics. Schmidt et al. (2016a) considered line of sight angle in their triad analysis, and de la Torre et al. (2018) highlighted potential effects of the slant angle of RO profiles on horizontal wavelength retrievals. These effects are particularly important for short-horizontal-wavelength GWs, and relatively less important for the longer-horizontal wavelengths (.250 km) that are typically retrieved by RO and other limb-sounding methods.
Future higher density RO temperature profile measurements can be exploited to improve knowledge of GW properties, but the optimal measurement density requires multiple simultaneous closely spaced profiles. RO observed from long duration stratospheric balloon platforms (Haase et al. 2018) are expected to probe the resolution limits of the RO technique for tropical GWs.

Atmospheric planetary waves a. Atmospheric thermal tides
Atmospheric thermal tides are persistent planetary-scale global oscillations, with periods of some integer fractions of a solar day. They are mostly excited by the absorption of solar radiation, latent heat release, or nonlinear wave-wave interactions. Tides that are generated in the lower atmosphere propagate upward with exponential growth in amplitude, dominate the motion of the mesosphere, and deposit their energy and momentum in the lower thermosphere Forbes 2002, 2003). Tides can be separated into two components: migrating and nonmigrating. Migrating tides are sunsynchronous, moving westward with the apparent motion of the sun, while nonmigrating tides do not. Measuring the tides from the source regions (close to the UTLS) is essential for understanding, monitoring, and predicting changes in the whole atmosphere.
However, our observational picture is based on limited data: (i) tidal amplitudes in the UTLS are relatively small (;0.5 K), which requires highly accurate measurements; (ii) tidal analysis is best for observations with full local-time sampling, which is generally not achievable from ground-based observations (which cannot separate the migrating and nonmigrating tides) or a single satellite measurement (which often has specific local-time sampling).
RO data have been used to study tides due to their high quality in the UTLS (Zeng et al. 2008;Pirscher et al. 2010;Xie et al. 2010;Khaykin et al. 2013;Sakazaki et al. 2015). Zeng et al. (2008) first used CHAMP RO dry temperature at 10-30 km from 2001 to 2005 to estimate the migrating diurnal tidal structures in the tropics. The problem of local-time sampling from a single satellite has been overcome by considering the slow local-time drift of CHAMP and merging multiyear data. Pirscher et al. (2010) advanced the study by using COSMIC data, which sampled all local times within one month equatorward of 508N/S. Both studies revealed that diurnal amplitudes in temperature increase with altitude and reach ;1 K at 30 km in the tropics but exhibit strong annual and latitudinal variations. These variations are clearly linked to the movement of the intertropical convergence zone. The tides propagate upward with a vertical wavelength of about 20 km. Khaykin et al. (2013) found differences of diurnal tidal variations between land and ocean in the lowermost stratosphere, which are associated with nonmigrating tides generated by the landocean contrast and/or the local effects of deep overshooting convection. Sakazaki et al. (2015) extracted tropical nonmigrating tides that are excited by diabatic heating over Africa and South America and propagate zonally away from their sources.

b. Equatorially trapped waves
Planetary-scale disturbances propagating parallel to the equator have been observed since the 1960s (Yanai and Maruyama 1966;Wallace and Kousky 1968) and theoretically understood as the equatorially trapped waves (Matsuno 1966;Lindzen 1967). In the troposphere, the waves that are primarily forced by deep tropical convection are convectively coupled and therefore visible in OLR data (Wheeler and Kiladis 1999). Transient deep convection also forces equatorial waves in the UTLS, but these waves have faster phase speeds that represent the remote response . Corresponding temperature perturbations can be observed throughout the UTLS (Tsai et al. 2004;de la Torre et al. 2004). These waves are important as they drive the QBO, lower the tropical CPT temperature, and modulate the transport of trace constituents and water vapor from the troposphere into the lower stratosphere (Kim and Alexander 2015).
Understanding of equatorially trapped waves has advanced through high-temporal-and spatial-resolution satellite data. RO-focused analyses have been complemented by analyzing data from a range of satellite instruments including Sounding of the Atmosphere Using Broadband Emission Radiometry (SABER) (Ern et al. 2008) and HIRDLS (Alexander and Ortland 2010). Vertical resolution of RO data is higher than that of SABER and similar to HIRDLS. While comparisons with radiosonde measurements revealed a limited representation of waves with short vertical wavelengths in RO data (Kim and Alexander 2015), wave structures and their propagation can be studied globally with RO data.
RO data sampling is optimal to resolve eastward propagating Kelvin waves. These waves are symmetric about the equator and propagate eastward in the stratosphere in regions with easterly background winds (e.g., Sato and Dunkerton 1997). Wavenumbers range from k 5 1 to 5, periods from approximately 3 to 30 days, and vertical wavelengths from 2 to 24 km (phase speeds from 7 to 75 m s 21 ) (Alexander and Ortland 2010). Kelvin waves with zonal wavenumbers of one and two were resolvable in the coarser-resolution CHAMP data , which revealed wave amplitudes of 2-4 K, vertical wavelengths of 4-8 km, and the characteristic eastward phase tilt with height (Tsai et al. 2004;Randel and Wu 2005).
The large increase in the number of RO profiles from the COSMIC satellite constellation enabled studies of Kelvin waves with higher zonal wavenumbers and shorter periods. Alexander et al. (2008b) observed higher wavenumber Kelvin waves with periods of about 5 days and maximum amplitudes of 2 K. Fast Kelvin waves observed by COSMIC had wave amplitudes smaller than 1.2 K and periods of less than 10 days.
The Kelvin wave periods reported in these RO studies are consistent with theory (i.e., wave periods confined within the region of large power in the symmetric wavenumber-frequency power spectra). The RO observations of wave periods are consistent with outgoing longwave radiation observations, which by definition are coupled to the top of the generating cloud systems (Wheeler and Kiladis 1999); and the RO-derived periods and temperature amplitudes are consistent with radiosonde observations (e.g., Wallace and Kousky 1968;Shimizu and Tsuda 1997;Suzuki et al. 2013).
Kelvin waves dominate subseasonal variability in the tropical tropopause region (Kim and Son 2012) and their activity increases in the upper troposphere when deep, transient convection is present (Randel and Wu 2005). In the TTL the seasonal structures and interannual variability of Kelvin wave activity are dominated by the background wind fields and tropospheric convection likely plays a minor role (Flannaghan and Fueglistaler 2013). However, relationships between Kelvin wave activity and wind near the tropopause are still not clear as shown in Fig. 8 In the lower and middle stratosphere, Kelvin wave activity is strongly modulated by the QBO with enhanced activity observed during the descending westerly wind shear QBO phase (Randel and Wu 2005;Ratnam et al. 2006;Scherllin-Pirscher et al. 2017a). Kelvin waves propagate upward in QBO easterlies, with a vertical group velocity proportional to their intrinsic phase speed squared (U 2 c) 2 . In upper-level regions of westerly shear the waves slow down and amplify (hence the large temperature variance), and thermal damping results in momentum deposition and eastward acceleration of the zonal flow.
RO data have been used to validate Kelvin wave activity of general circulation models and atmospheric reanalyses (Kawatani et al. 2009;Flannaghan and Fueglistaler 2013). It was found that wave activity in the TTL increased in the ERA-Interim reanalysis with the assimilation of COSMIC RO data (Flannaghan and Fueglistaler 2013).
Westward-moving MRGWs are antisymmetric about the equator and have periods of less than 5 days. While individual phases of MRGWs propagate westward, the wave packets propagate eastward. The sporadic nature of MRGWs contributes to their low variance in wavenumber-frequency spectra, yet they are at times clearly visible in the data Alexander and Ortland 2010). There remain very few RO studies of MRGWs, although Alexander et al. (2008b) showed MRGWs with amplitudes around 1 K that vanished once the background mean flow became westward.

Intraseasonal-to-interannual atmospheric modes of variability
In addition to fast equatorial waves and various tropical phenomena in the troposphere and stratosphere, temperatures in the tropical UTLS reveal a wide range of variabilities spanning intraseasonal through interannual time scales. Owing to the high accuracy and fine vertical resolution of RO, these sources of atmospheric variability are better quantified on an observational basis.

a. Intraseasonal variability
The intraseasonal variation in the tropical UTLS is mostly related to the equatorial waves (Wallace and Kousky 1968;Tsuda et al. 1994) and the MJO (Kiladis et al. 2001). As described in the previous section, detailed features of intraseasonal variability related to the equatorial waves are well captured by RO due to its high vertical resolution and global coverage (e.g., Alexander et al. 2008b;Kim and Son 2012;Scherllin-Pirscher et al. 2017a).
The MJO is another important source of intraseasonal variability in the tropics. It is generally characterized by organized mesoscale convection in the tropics propagating from the Indian Ocean to the central Pacific (around the date line, Zhang 2005). The circulation and convective heating of the MJO produces a unique temperature anomaly structure showing tropospheric warming covered by a thin layer of cooling near the tropopause (Kiladis et al. 2001; also shown in Fig. 9). Particularly the near-tropopause cooling has a strong temperature anomaly of ;2-4 K in a narrow vertical depth of 2-3 km, which impacts water vapor content and cirrus clouds in the tropical tropopause layer (Virts and Wallace 2014;Kim et al. 2018). The temperature anomaly increases the static stability above the tropopause and affects the strength of the tropopause inversion layer. The detailed features are clearly captured by RO measurements and they show coherent features with satellite composition and cloud measurements during MJO events (Kim and Son 2012;Virts and Wallace 2014;Noersomadi et al. 2019). Furthermore, the tropospheric and stratospheric temperature anomalies propagate slowly to the east following the organized convection in the MJO (Fig. 9). These transient characteristics are observed well by RO data because of its enhanced sampling rate and spatial coverage over the tropics (e.g., Tian et al. 2012;Zeng et al. 2012). Recent work has highlighted that MJO UTLS structure is strongly modulated by the QBO (Hendon and Abhik 2018).

b. Annual cycle
The annual cycle is the dominant source of variability in the tropical UTLS over altitudes ;16-22 km with the maximum annual difference of 8 K at ;18 km (Kim and Son 2012;Randel and Wu 2015). Particularly, a significant temperature variation is caused by the Brewer-Dobson circulation (BDC) inducing more vigorous tropical upwelling (and adiabatic cooling) in the tropical stratosphere during the boreal winter than in the boreal summer (Fueglistaler et al. 2009). This temperature variation is further amplified by local radiative forcing from ozone and water vapor (Fueglistaler et al. 2011;FIG. 9. RO-based temperature anomalies in the deep tropics (58N/S) correlated with the real-time multivariate Madden-Julian oscillation (RMM) indices. This represents one-half of an MJO cycle (Virts and Wallace 2014). Ming et al. 2017). This temperature variability is accurately quantified at the tropopause and in the lower stratosphere by RO measurements Son et al. 2011;Kim and Son 2012;Rieckh et al. 2014). Also, RO, with its fine vertical resolution, revealed that the annual cycle of temperature is maximized in the lowermost stratosphere (e.g., Randel et al. 2003;Schmidt et al. 2004;Kim and Son 2012), which became the key observational evidence supporting the importance of the lower branch of the BDC for driving the annual cycle in the tropical UTLS (Ueyama et al. 2013;J. Kim et al. 2016).
With an increased sampling of temperature profiles from the COSMIC mission, spatial structures of the UTLS temperature were also examined for different seasons, revealing regional details of seasonal anomalies in near-tropopause temperature and inversion layer (Nishi et al. 2010;Grise et al. 2010;Kim and Son 2012;Scherllin-Pirscher et al. 2017a). For example, RO clearly captures quasi-stationary cold anomalies located over the western Pacific in boreal winter and South Asia in boreal summer related to seasonal migration of deep convection. A frequent upper tropospheric inversion is also observed over the equatorial Indian Ocean during boreal summer (Nishi et al. 2010). Newton and Randel (2020) have shown that these quasistationary inversions are a climatological feature, linked to transient convectively forced Kelvin waves that are Doppler shifted in the background UTLS easterly winds of the Asian monsoon anticyclone.

c. Interannual variability
Interannual variability in the tropical UTLS is associated with tropospheric and stratospheric processes. The QBO is a well-known source of stratospheric variability that is identified as alternating westerly and easterly winds in the deep tropics (within 108N/S; Baldwin et al. 2001). A warm (cold) anomaly accompanies westerly (easterly) shear layers due to thermalwind balance in the tropics, and the temperature anomalies affect the tropical UTLS down to the tropopause region with an approximate 28-month cycle. The maximum QBO-related temperature anomalies measured by RO are approximately 64 K over altitudes of ;22-31 km (Fig. 10), which is consistent with the radiosonde observations at Singapore (Baldwin et al. 2001). Although the amplitude of the temperature variation diminishes downward, its impact near the CPT is still crucial as it modulates the amount of the stratospheric water vapor (Randel et al. 2004;Randel and Park 2019). The temperature impact has been clearly detected in the lower stratosphere and at the tropopause by early records of RO (e.g., Randel et al. 2003;Schmidt et al. 2005). The QBO-induced temperature variation at the cold point appears uniformly in the deep tropics (158N/S), and it is quantified as approximately 60.5 K by longer RO records (e.g., Son et al. 2017;Wilhelmsen et al. 2018).
Another source of interannual variability in the tropical UTLS originates from ENSO-related changes in organized convection (Yulaeva and Wallace 1994). Notably, a positive phase of the ENSO enhances convection in the tropical central to eastern Pacific while suppressing it over the western Pacific (vice versa in the negative phase). This strongly influences the three-dimensional UTLS temperature structure, including a large longitudinal shift in seasonal temperature anomalies and a zonal-mean response (e.g., Calvo Fernández et al. 2004;Scherllin-Pirscher et al. 2012). The zonal-mean influence of positive ENSO is warming in the tropical upper troposphere and cooling in the lower stratosphere ( Fig. 10b) with maximum zonal-mean response approximately 3 months after ENSO forcing (Randel et al. 2009;Lackner et al. 2011b;Scherllin-Pirscher et al. 2012). The most pronounced zonal-mean tropospheric warming (8-15 km) reaches up to 2 K. Longitudinal UTLS response to ENSO effects occur more rapidly (within 1 month) and feature maximum amplitude in the upper troposphere (near 11 km) and with opposite polarity in a shallow layer near the tropopause (Scherllin- Pirscher et al. 2012).  Wallace et al. (1993), and ENSO is based on the multivariate ENSO index (MEI; https://psl.noaa.gov/enso/mei/). Contours interval is 0.2 K, with negative values dashed. Shading shows statistically significant values at 99% level. The heavy dashed lines denote the thermal and cold-point tropopauses (Randel and Wu 2015).
ENSO dynamics is also associated with a higher number of double tropopauses during the negative ENSO phase, which is clearly visible in the tropical eastern Pacific (Wilhelmsen et al. 2020).
In addition to the QBO and ENSO, interannual variations in the BDC (Randel and Wu 2015) and enhanced MJO convection during the easterly phase of the QBO (Son et al. 2017) contribute to the interannual temperature variability in the tropical UTLS by directly affecting temperature or modulated convection in the tropics. The high-resolution temperature observations from RO significantly contribute to understanding the dynamic mechanisms involved in these processes.

Climate applications
a. Climate monitoring and benchmarking RO has been identified as a key component for the Global Climate Observing System (GCOS) owing to its potential as a climate benchmark measurement (GCOS 2011(GCOS , 2016. Benchmark measurements can be related to absolute standards and may be repeated at any subsequent time with precise comparability (Goody et al. 1998(Goody et al. , 2002, such that the accuracy of the record archived today can be verified by future generations. Traceability to standards of the International System of Units (SI) is the foundation for benchmark measurements (Leroy et al. 2006b;Ohring 2007). For RO, the fundamental measurement is the GNSS signal phase change as function of time, based on highly precise and stable atomic clocks (see section 2).
RO observations can be assimilated without bias correction in numerical weather prediction models. Most modern reanalyses assimilate RO observations producing more consistent temperature analyses in the upper troposphere and stratosphere (e.g., Fujiwara et al. 2017;Hersbach et al. 2020).
For climate signal detection, the uncertainty in the observations must be smaller than the expected change signals (Ohring et al. 2005;Bojinski et al. 2014). In the UTLS, RO observations are of highest accuracy and the structural uncertainty arising from different processing schemes is smallest (see section 2). The RO record from 2002 to the present is found suitable for trend detection (Steiner et al. 2020a). The structural uncertainty of temperature between 8 and 25 km altitudes is ,0.05 K decade 21 for the global mean and ,0.1 K decade 21 for individual latitude bands, conforming to GCOS requirements. Yuan et al. (1993) were among the first who suggested the use of RO for trend detection. By simulating the propagation of GPS signals in a climate model with doubled carbon dioxide concentration, they found an increase in the signal phase path. Melbourne et al. (1994) and Ware et al. (1996) further stated that RO could detect trends over short time periods. More quantitatively, Leroy et al. (2006a) estimated climate change signal detection times of only 7-13 years for a scenario with 1% yr 21 growth in carbon dioxide.
For the same forcing scenario, clear signals of bending angle trends with detection times of 10-16 years were shown in climate simulation experiments (Ringer and Healy 2008). Vedel and Stendel (2003), Stendel (2006), and Leroy et al. (2006a) investigated the value of RO geopotential height and refractivity for climate studies. The complementary sensitivity for monitoring the UTLS was demonstrated with observing system simulation experiments for a medium emission scenario Foelsche et al. 2008b) and climate change indicators (Lackner et al. 2011a).

b. Detection of short-term trends
The detection of trends in short-term records requires appropriate accounting of natural variability. Steiner et al. (2009) were the first to demonstrate the trend detection capability of actual RO temperature observations using GPS/MET and CHAMP observations from 1995 to 2008. They showed an emerging warming trend in the tropical troposphere and a significant cooling trend in the tropical lower stratosphere.
In an optimal fingerprinting study, Lackner et al. (2011b) investigated climate change patterns in several RO variables for the period of 2001-10. They detected emerging trend signals in geopotential height (90% confidence level) and temperature (95% confidence level). In the tropics, a warming of ;0.3 K decade 21 was detected in the upper troposphere and a cooling of ;0.6 K decade 21 in the lower stratosphere. Warming of the tropical tropopause layer was confirmed by Wang et al. (2015). The most recent study of Steiner et al. (2020b) gives an update on UTLS temperature trends from a range of observations. They provide trend results for RO observations from 2002 to 2018 detecting a significant warming in the troposphere of ;0.3 K decade 21 (95% confidence level) and a cooling from 20.1 to 20.2 K decade 21 in the tropical lower stratosphere (Fig. 11). Schmidt et al. (2008bSchmidt et al. ( , 2010b and W.  performed early studies on tropopause height and temperature trends. Due to significant natural variability, tropopause trends were inconclusive for short periods (Rieckh et al. 2014). A widening of the tropical belt deduced from RO tropopause statistics was found in the Northern Hemisphere from 2002 to 2011 (Ao and Hajj 2013;Davis and Birner 2013;Gleisner et al. 2015 Short-term trends of RO observations have been compared with different satellite observations such as those from the Advanced Microwave Sounding Unit (AMSU) (Gleisner et al. 2015) and the Advanced Infrared Sounder (AIRS) (Leroy et al. 2018) as well as reanalyses (Shangguan et al. 2019). Khaykin et al. (2017) showed good agreement in lower stratospheric temperature trends between RO and Aqua AMSU records. A comparison of trends from RO observations and those from reanalyses assimilating various satellite observations, however, revealed significant biases in the reanalyses due to inherent discontinuities in assimilated observations and methods (Shangguan et al. 2019). c. Climate model evaluation Leroy and North (2000) investigated the application of RO for climate model evaluation. They showed that climate signal detection can be a useful method to test whether climate models reproduce trends and anomalies as revealed by the observations. Leroy et al. (2006a) tested future projections and climate change fingerprints in models from phase 3 of the Coupled Model Intercomparison Project (CMIP3), based on simulated RO pressure trends. Ao et al. (2015) reported good agreement between models from phase 5 of CMIP (CMIP5) and RO observations in terms of the annual cycle and interannual variability of the tropical geopotential height at 200 hPa. In the extratropics, the agreement became poor with the largest model biases in high latitudes. By comparing UTLS temperature fields at selected pressure levels with COSMIC observations from 2006 to 2013, Kishore et al. (2016) concluded that the CMIP5 models reproduce the overall annual and interannual temperature variations, but have a large bias of about 2 K at 100 hPa on average.
Data from RO can be used for more detailed assessments of atmospheric climate models, such as on the effect of different model resolutions (e.g., Schmidt et al. 2016b).  investigated the representation of tropical convection in the CMIP5 atmosphere-only models (AMIP), which are available at higher vertical resolution. They focused on the vertical structure in moist and dry regimes. Results showed that the tropical tropopause is poorly captured in most atmosphere models, with model temperature biases larger than 4 K in the tropopause region and the lower stratosphere. Results also revealed that some models do not fully capture the thermodynamic structure in regions of enhanced convection. In moist convection regions, models tend to underestimate moisture over the oceans, whereas in dry downdraft regions, they overestimate moisture up to 100%. These shortcomings of models were also discussed by Kursinski and Gebhardt (2014).
The RO observations were also used to evaluate the width of the tropical belt based on the tropopause and subtropical jet core metrics, showing that models tend to overestimate the mean width and underestimate its seasonal cycle amplitude (Davis and Birner 2016).
In the multimodel assessments of coupled chemistry-climate models (CCMs) by Gettelman et al. (2010) for the tropics and by Hegglin et al. (2010) for the extratropics, RO observations were used to evaluate the representation of the tropical and extratropical tropopause transition layers. The findings showed that a tropopause inversion layer is present in most models, but it is much weaker and thicker than observed in RO data, even when the observations are regridded to model levels (Fig. 12). Gettelman et al. (2010) explain the limited representation of the tropopause inversion layer and buoyancy frequency by coarse vertical model resolution and less adiabatic cooling associated with weak upwelling in the models.
9. Summary, conclusions, and outlook GNSS radio occultation observations provide unique information on the thermal structure and variability of the tropical UTLS because of their high accuracy (low biases/systematic errors) and precision (low uncertainty) as well as high vertical resolution, global availability, and long-term stability. The UTLS region is fundamental to many aspects of the climate system, showing strong spatial and temporal variability ranging from subseasonal to interannual time scales. RO data have been instrumental in quantifying these sources of variability due to their quasi-homogeneous coverage in time and space throughout the tropics.
Subseasonal variability associated with perturbations from gravity waves, atmospheric Kelvin waves, Rossby and mixed Rossby-gravity waves is well resolved by RO data. Even small-amplitude atmospheric thermal tides can be investigated due to complete local-time sampling at low latitudes. Insensitive to clouds and precipitation, RO observations reveal novel details of the structure of deep convection, tropical cyclones, and volcanic clouds. Data have also been exploited to study the UTLS response of the MJO, ENSO, and the QBO. Long-term stability and consistency make the almost twodecade-long RO record well suited to evaluate climate models and to study climate change.
Highlights of added value of RO for improving knowledge on tropical UTLS characteristics and behavior include the following: With the successful launch of COSMIC-2 in July 2019, the proliferation of commercial LEO satellites, and the use of more GNSS satellites besides GPS such as the Russian GLONASS or the European Galileo system, tropical sampling density will further increase, opening new potential to study fine-scale structures of the tropical UTLS. This will yield new insights on the morphology of the tropical UTLS and improve understanding of the atmospheric climate system.