Using Constellation Observing System for Meteorology, Ionosphere, and Climate (COSMIC) radio occultation data from June 2006 to December 2009, the temperature structure and the tropopause height over the Tibetan Plateau (TP) is studied in this paper. The temperature over the TP is warmer and experiences a lager magnitude of change than that over the same latitudinal plain area (hereinafter referred to as the Plain) at the same level of lower troposphere. The lapse rate tropopause (LRT) shows a strong correlation with thermal properties of the atmosphere. Its height variation is anticorrelated with its temperature, highest at ∼19 km and coldest at approximately −72°C in boreal summer, but lowest at ∼13 km and approximately −56°C in boreal winter. Those of the cold point tropopause (CPT) exhibit a positive correlation but barely seasonal variation all year long. As an outstanding heat source in boreal summer, the TP thermally pushes the LRT upward by ∼2 km compared with the Plain, while the LRT drops below that over the Plain in boreal winter. The LRT height is also strongly dependent on the subtropical jet that is associated with the tropopause fold and/or multiple tropopauses, leading to fairly bimodal distribution of the LRT probability density function. As the “roof of the world,” the elevated topography of the TP dynamically lifts the CPT to a higher altitude without significantly seasonal variations. Along the latitude of 32.5°N the CPT is located at ∼18 km over the main body of the TP and drops to ∼15 km over the Plain. Given the area in the TP (30°–35°N, 87°–95°E) as well as in the Plain (30°–35°N, 112°–120°E), the CPT primarily resides at 17–18 km, 72.52% over the TP and 69.22% over the Plain. Their monthly mean difference can reach 1.4 km for the complete analysis period.
 The tropopause is the transition between the troposphere and stratosphere; both have fundamentally different characteristics with respect to chemical processes and dynamical regimes [Schmidt et al., 2004]. The tropopause acts as a “two-way gate” for the exchange of mass, water vapor, and chemical species between troposphere and stratosphere [Fueglistaler et al., 2009]. Since the vertical transport of water vapor and chemical compositions from the troposphere-stratosphere exchange affects the radiative flux balance in the upper troposphere and lower stratosphere, the tropopause plays an important role in the dynamical, radiative, and climate change [Holton et al., 1995].
 As a sensitive indicator of climate variability and global change, the tropopause height itself has received much attention in recent years. Using reanalysis data and climate model simulations, Santer et al. [2000, 2003] and Sausen and Santer  pointed out that tropopause height is closely associated with tropospheric warming and stratospheric cooling. Thus, continuous identification and monitoring of the tropopause become an important goal in atmospheric and climate research. During the last decades, the global mean tropopause height has shown an increase in reanalysis and radiosonde observations [Randel et al., 2000; Seidel et al., 2001; Gettelman and de F. Forster, 2002; Santer et al., 2003]. However, either radiosonde or reanalysis data suffers from lower resolution and attendant biases. What is worse, the radiosonde, as the most important observation for the tropopause-related research, bears an uneven spatial resolution on Earth. Over the Tibetan Plateau (TP hereafter), radiosonde stations are sparsely located in the eastern half where the average altitude is pretty low compared with the average elevation of the TP that is over 4500 m above sea level and sometimes called “the roof of the world.” In addition, the temporal frequency of radiosonde observations is also fairly low with 12 h intervals per day as well as substantial changes in instrumentation over time. The spatial and temporal scarcity of available observations limits the studies on the characteristics of the tropopause over the TP.
 As “the global water tower,” the TP is the main pathway for water vapor cross-tropopause transport [Fu et al., 2006], which exerts a major influence on the energy balance of the Earth-atmosphere system despite its low abundance in the stratosphere [de F. Forster and Shine, 1999]. Also, the TP is one of the most sensitive regions to global climate change [Liu and Zhang, 1998; Liu and Chen, 2000]. Liu and Chen  revealed that the recent warming over the TP began early and appeared intense with respect to the global warming compared with its neighboring areas or the same latitudinal zone.
 In this study, we attempt to compare the tropopause height over the TP with that over the same latitudinal plain (hereinafter referred to as the Plain) on the basis of the Formosa Satellite 3 (FORMOSAT-3)/Constellation Observing System for Meteorology, Ionosphere, and Climate (COSMIC) data. The COSMIC data are derived from the Global Positioning System (GPS) radio occultation (RO) limb-sounding technique that is an effective self-calibrating instrument and a new promising remote sensing technique. The RO technique offers much higher vertical resolution (∼100 m) than radiosonde observations and model reanalyses. It has the same order of precision as radiosonde observations but with the global coverage over continents as well as over oceans. To take advantage of the COSMIC data with the high vertical resolution and global coverage, one of the efficient approaches is to directly assimilate the GPS refractivity data into high-resolution numerical models [Zou et al., 1995; Liu et al., 2007, 2008; Ma et al., 2009]. The experiments conducted by Huang et al.  showed that the assimilation of the RO refractivity in the Mesoscale Model (MM5) causes the pressure and accumulative rainfall fields to adjust in a mutually consistent way, leading to a more accurate forecast of typhoon cases. In addition to mesoscale applications, the high-resolution, global GPS RO observations are useful in climate studies by providing accurate temperature distributions and allow a more accurate identification of the tropopause over the TP. Examination of the tropopause using the GPS RO observations by comparing the temperature structures over the TP and over the Plain opens a new application of the COSMIC data.
 Marked by changes in the thermal and dynamical structure and the chemical tracers of the atmosphere, the location of the tropopause has been identified in a variety of ways in previous studies, including using pressure, equivalent potential temperature, neutral buoyancy, potential vorticity (PV) and its gradient, and ozone and its gradient [Holton et al., 1995; Highwood and Hoskins, 1998; Seidel et al., 2001; Wang, 2003; Pan et al., 2004; Liu and Zipser, 2005; Fueglistaler et al., 2009]. Hence, the precise location of the tropopause can be defined in a variety of ways, according to the dynamic, thermal, and chemical properties of the atmosphere. In observational studies, all these definitions are used depending on the research topic. Model simulations often use a PV-based dynamical tropopause, ranging from 1.5 to 3.5 PVU [Holton et al., 1995; Pan et al., 2004]. It has been shown that these definitions agree qualitatively well on the large scale, but present different details of the tropopause locations [Highwood and Hoskins, 1998; Seidel et al., 2001; Randel et al., 2007; Fueglistaler et al., 2009]. The focus here is to compare the tropopause over the TP and the Plain rather than to put forward a better definition. According to the temperature structure of the atmosphere, the cold point tropopause (CPT) and the temperature lapse rate tropopause (LRT) are therefore applied in this study.
 The structure of the present paper is as follows. Section 2 briefly introduces the COSMIC RO data with the comparisons to radiosonde data. The definitions of tropopause derived from the cold point and the temperature lapse rate algorithms are also described in section 2. In section 3, the temperature structures over the TP and the Plain are examined. Using the CPT and LRT definitions for the tropopause, section 4 presents the comparison of the tropopause over both the TP and the Plain regions. We then present discussions and conclusions of the results in section 5.
2. Data and Methods
 The FORMOSAT-3/COSMIC is a joint U.S./Taiwan mission for weather, climate, space weather and geodetic research. It includes six GPS RO receivers (identical micro satellites) developed by the National Aeronautics and Space Administration's (NASA's) Jet Propulsion Laboratory (JPL) in low Earth orbit, and successfully tracks roughly 2000 profiles per day [Rocken et al., 2000; Anthes et al., 2008]. By measuring the phase delay of radio waves from GPS satellite as they are occulted by Earth's atmosphere, accurate and precise vertical profile of the bending angles of radio wave trajectories are obtained, and profiles of atmospheric refractivity are in turn obtained from these bending angles. By inverting profiles of atmospheric refractivity, high-resolution and vertical profiles of pressure, temperature, moisture, and other constituents are estimated [Kursinski et al., 1997; Kuo et al., 2004].
where N is refractivity, n is refractivity index, P is pressure in hPa, T is temperature in K, Pw is water vapor pressure in hPa, ne is electron density in number of electrons per cubic meter, W is liquid water in g/m3, f is transmitter frequency in Hz.
 The four refractivity terms in equation (1) are referred to as the dry, moist, ionospheric, and scattering terms. In the troposphere, the ionospheric and scattering terms can be neglected and equation (1) reduced to
In the lower troposphere, the contribution of water vapor concentration increases up to 30% to net refractivity and can locally dominate the vertical refractivity gradients and bending near the surface [Kursinski et al., 1997], as a consequence, the presence of significant tropospheric water vapor complicates the interpretation of refractivity.
 The GPS RO measurement has higher vertical resolution than radiosonde observations, reanalysis data, and model simulations, which could improve both the precision and accuracy of derived LRT heights [Seidel and Randel, 2006]. The temperature measurements retrieved from GPS RO data are very accurate with the precision of 0.25 K between 10 and 20 km [Anthes et al., 2008]. In this study, the COSMIC temperature profiles from June 2006 to December 2009 with 100 m vertical resolution recorded in level 2 postprocessed wetPrf product provide by the UCAR COSMIC Data Analysis and Archive Center (CDAAC) are used. These profiles are based on one-dimensional variation method using the European Centre for Medium-Range Weather Forecasts (ECMWF) low resolution analysis to estimate the relative contributions of moisture and temperature to the measured bending angle. Staten and Reichler  examined that the postprocessed wetPrf profiles have negligible difference with the “dry” profiles (atmPrf) in the tropopause region. We therefore focus on the wetPrf product to provide the closest possible validation with radiosonde data.
 In addition, the zonal wind data from the National Centers for Environmental Prediction (NCEP)/National Center for Atmospheric Research (NCAR) reanalysis project [Kalnay et al., 1996] are employed to examine the relationship between the location of the subtropical jet and the tropopause.
2.2. Tropopause Definitions
 Promulgated by World Meteorological Organization (WMO) , the LRT is defined as the lowest level at which the temperature lapse rate is less than 2°C/km and the lapse rate average between this level and the next 2 km does not exceed 2°C/km. To avoid boundary layer inversions, this algorithm is applied only to altitudes above 500 hPa level. Using the temperature lapse rate definition, temperature profiles in the regions we focus often exhibit multiple tropopause as defined [Schmidt et al., 2006; Randel et al., 2007; Añel et al., 2008]. Añel et al.  indicated that the third tropopause of triple tropopause there reaches ∼40 hPa, no matter in June–July–August (JJA) or in December–January–February (DJF). Hence, we determine the LRT within 500∼40 hPa which can be translated to 5.5–22 km, and all the outliers have been omitted to avoid the unreal tropopause arising from the tropopause definition. The CPT is the location of the temperature minimum, or cold point, in a vertical temperature profile. The study of Selkirk  implied that the CPT is probably consistent with the LRT in radiosonde ascents, but commonly lies above it. Subsequently, Highwood and Hoskins  suggested that the CPT is only a reliable definition when the lower stratosphere is not close to being isothermal (i.e., within the deep tropics).
 The COSMIC ROs are used to create time series for the comparison of the temperature structure and tropopause over the TP and the Plain regions. We only consider those profiles that pass quality control and all the valid profiles have been applied unweighted running average among five vertically adjacent points to reduce uncertainty stemming from RO measurements.
 In section 3, the climatological means, differences, and anomalies of temperatures over the TP (30°–35°N, 87°–95°E) and the Plain (30°–35°N, 112°–120°E) regions are compared. Mean temperatures during boreal summer and winter are calculated by averaging all profiles during JJA and DJF, respectively. Subtracting the long-term average temperature from each monthly mean, the temperature anomalies are derived.
 In section 4, a collocation requirement is employed to match the ROs in the Plain region with those in the TP region for further comparison. The ROs in the TP region are compared with those in the Plain that are recorded within a distance of less than 300 km and a time difference less than 3 h from those in the TP. During the entire period, there are 353 and 343 collocated tropopause measurements over the TP and the Plain, respectively. In the TP region, there are 98 profiles in DJF and 82 in JJA; in the Plain, there are 104 in DJF and 67 in JJA. This tightening collocated requirement yields smaller collocation error but reduces the robustness of the statistical analysis. On the basis of these matchups, the probability density functions (PDFs), climatologically monthly mean, and seasonal cycle of the tropopause on the basis of the CPT and LRT algorithms are compared between the TP and the Plain. In addition, we average the daily RO temperature profiles and then interpolate them to 0.5° × 0.5° grid boxes of which centers are located at 32.5°N from 75°E to 125°E to present the zonal variation of the tropopause along 32.5°N to examine the impact of topography forcing on the tropopause height.
2.3. Validation of COSMIC ROs With Radiosondes
 As the primary source of validation of GPS retrievals, other satellite technique, and reanalysis products, radiosondes in the TP and the Plain regions are first compared with the FORMOSAT-3/COSMIC GPS ROs that are recorded within ±300 km and ±3 h window from the radiosonde launch during the period from June 2006 to December 2009. Radiosonde soundings in this study are provided by the Meteorological Information Center of China, and the report includes atmospheric pressure, temperature, and relative humidity with an average interval of 12 h per day.
 Owing to an increasing meridional temperature gradients over the extratropics [Fueglistaler et al., 2009], radiosonde stations at the TP and the Plain are selected in the same latitudinal zone listed in Table 1. Two of these radiosonde stations with the average altitude of 4951 m are located in the TP region, and seven are located in the Plain region with the average altitude of 107 m.
Table 1. Radiosonde Stations Over the Tibetan Plateau and Plain
Latitude (deg N)
Longitude (deg E)
Plateau (30°–35°N, 87°–95°E)
Plain (30°–35°N, 112°–120°E)
 Using the valid data, we interpolate COSMIC RO retrieved temperatures to the radiosonde stations at the same levels. The comparisons of the temperatures from COSMIC and radiosondes are then made over the TP and the Plain, respectively. The comparison results are shown in Figure 1. The diagonal lines represent the points where the interpolated RO temperature is exactly equal to the radiosonde temperature. Figure 1 shows that the RO temperature is highly correlated with that from the radiosonde observation. The correlation coefficients are 0.994 and 0.996 over the TP and Plain, respectively. During the whole analysis period, the total amount of matched samples is abundant with 1118 over the TP and 3710 over the Plain. Collected all 4828 matchups from June 2006 to December 2009, Figure 2 shows the root-mean-square (RMS) differences between the interpolated COSMIC temperature profiles and the radiosonde temperature for both the TP and Plain regions. Similar variations of RMS differences with altitude were shown for both regions, although the RMS difference for the TP is smaller than that for the Plain. Their maximum RMS differences are both located at 10–11 and 16 km owing to the occurrence of multiple tropopause and/or the resulting strongly discontinuous nature of the first tropopause in the subtropical regions (see section 4) [Schmidt et al., 2006; Randel et al., 2007; Staten and Reichler, 2008]. Besides, since water vapor becomes the dominant factor in the GPS retrieved algorithm, their RMS differences also reach maxima near the surface, 2.5°C for the TP and 4.25°C for the Plain, which is in a good agreement with Staten and Reichler .
 Using the COSMIC RO measurements, the average temperature profiles over the TP and the Plain are shown in Figure 3. The temperature structures for both regions differ slightly, yet the variability is somewhat lower over the TP than over the Plain (Figure 3a). The atmospheric structure becomes stable above the cold point, and the temperature during JJA has a much lower variability than during DJF. The temperature profiles become more stable above the cold points and near the surface. The former is a feature common in numerous studies as a strong indication that this is a layer of limited convective mixing [Selkirk et al., 2010], but the latter may result from that frozen glaciers and snow covers make the near surface air of the TP pretty stable, which leads to the temperature within ∼300 m from the surface increases upward, especially in boreal winter (see Figure 3b).
 In terms of the minimum temperatures (cold points) and the corresponding altitudes in Figure 3, to a good approximation, the CPT resides at 17–18 km, somewhat higher in DJF (18.3 km over the TP and 17.8 over the Plain) than in JJA (17.5 km over the TP and 17.3 km over the Plain). Also, the temperature of the CPT is higher in DJF (−67°C for the TP and −67.1°C for the Plain) but lower in JJA (−75.7°C for the TP and −72.5°C for the Plain), with the mean value of approximately −70°C.
 With more details, Figure 4 compares the temporal variations of the temperature structures over the TP and the Plain. It shows that the layer between the troposphere and stratosphere is thicker and colder over the TP than over the Plain. For the complete analysis period from June 2006 to December 2009, the atmosphere over the TP is colder above 14 km that is the typical location of the subtropical jet (see Figure 5), except the months of January 2007 and February 2009 (Figure 4c), but warmer for almost the whole period below the subtropical jet. The temperature reaches the cold point at about 17 km (Figure 4a). Analyzed using ERA40 reanalysis data, Duan and Wu  and Duan et al.  argued that the recent climate warming over the TP was primarily the result of the increasing anthropogenic greenhouse gases emissions. The impacts of the increased greenhouse gases emissions on the climate change in the TP region are probably more serious than the rest of the world. However, this topic is beyond the scope of this study.
 Heated by strong solar radiation, the temperature lapse rate is close to supermoist adiabatic characteristics near the TP surface, which to a large extent stimulates the development of convective activities [Luo and Yanai, 1984]. This strong convective activity leads to a deep mixture of the lower troposphere and even bring water vapor to the upper troposphere where air is subjected to condensation heating to lead the maintenance of higher temperature and humidity over the TP [Ye and Wu, 1998], compared to the same level over the Plain. Given similar patterns of the monthly mean temperature anomaly for both the TP and the Plain regions (Figure 6), the temperature change over the TP has a larger magnitude, which provides support that the TP is one of the most sensitive areas to the climate change [Liu and Zhang, 1998; Liu and Chen, 2000].
4. Tropopause Comparison
 To address if the tropopause is higher over the TP than over the Plain, the PDFs of the tropopause derived from the cold point and lapse rate definition are depicted in Figure 7. Either over the TP or over the Plain, the CPT is climatologically overall located at 17–18 km with the total PDF of 72.52% and 69.22% and reach maximum at 18 km with the PDF of 38.24% and 37.9% (Figure 7b), while the LRT shows an bimodal distribution with maxima at 11 km (11.07% over the TP and 10.26% over the Plain) and 16–17 km (44.3% over the TP and 42.05 over the Plain) in Figure 7a, which is associated with the multiple tropopauses and/or tropopause fold around the interface between the tropics and extratropics as a result of the subtropical jet stream aloft (see Figure 5) [Kochanski, 1955; Holton et al., 1995; Chang et al., 1998; Schmidt et al., 2006; Randel et al., 2007; Añel et al., 2008]. There is an excellent agreement between the wind speed and the multiple tropopause occurrences. In DJF, the subtropical jet becomes stronger and the climatological location is around 30–35°N which is the region we focus. The statistics for LRT show a fairly bimodal distribution with 11 km (18.1%) and 16 km (13.8%) for the TP and 11 km (21.4%) and 15 km (13.3%) for the Plain, reflecting the fact that double tropopause occurrences are often observed during winter [Hoinka, 1998]. With the rapidly weakened subtropical jet moving to 40°–45°N in boreal summer, the strong surface heating does not be carried away quickly owing to the weakened subtropical jet. The altitude of the lower LRT approaches the altitude of the higher LRT in summer [Randel et al., 2007], its PDF shows a single maximum centered at 17 km. Additionally, the fierce surface heating of the TP is directly added to the middle troposphere and used only to heat half of the atmosphere compared to the Plain region [Ye and Wu, 1998], the maximum PDF of the LRT located at 16 km reaches 58% for the TP region but 39.2% for the Plain.
 Note that the LRT height ranging from 8 to 22 km in this study is with a wider range than the previous findings (11–17 km) derived from the radiosonde profiles [Highwood and Hoskins, 1998; Seidel et al., 2001; Fueglistaler et al., 2009]. There are three reasons to explain this difference. First, it is related to the different vertical resolution of the respective measurement. Davis and Emanuel  pointed out that the high-resolution data in the vicinity of the tropopause is crucial for resolving features near the tropopause. Second, it is due in part to the cooling in the lower stratosphere in recent years [Randel et al., 2007]. The tropopause height variations are anticorrelated with stratospheric temperature variations of which per degree cooling raises the tropopause up to 2–3 km [Seidel and Randel, 2006]. Randel et al.  found that temperature changes in the lower stratosphere are cooling of ∼1.0 K/decade over the latitudinal zone of this study for 1979–2007. Last but not least, the multiple and/or discontinuous tropopause can also lead to the difference in the LRT location between our study and previous studies which is difficult to be accurately detected by the low vertical resolution of radiosonde observations.
 To reduce the noise and increase the statistical significance of the trends, monthly mean and seasonal cycle of the tropopause parameters is calculated. Figure 8 shows the individual monthly means of tropopause height and temperature for the TP and the Plain regions on the basis of the cold point and the lapse rate definitions. The variability of the LRT and the CPT heights are opposite but respectively consistent for each region. The LRT is highest and coldest during boreal summer, and lowest and warmest during boreal winter; the CPT is highest and warmest during boreal winter, and lowest and coldest during boreal summer. Another obvious feature seen in Figure 8 is that the height and the temperature of LRT shows considerable up-down variability, with values ranging from ∼13–19 km and approximately −72°C to −56°C, while the variability of the CPT exhibits only ∼1.5 km in height and ∼10°C in temperature over time. For the period from June 2006 to December 2009, the LRT over the TP region is higher than that over the Plain region in boreal summer with the maximum of 2.5 km, but lower in boreal winter. However, the CPT over the TP is fairly higher than that over the Plain, and their difference can increases to 1.4 km in the winter half of a year.
 On the basis of these monthly means, the seasonal cycle of the tropopause height and temperature for the TP and the Plain regions are shown in Figure 9, clearly depicting their seasonal variations and differences. The CPT height shows barely seasonal variations all year long, whereas the LRT is higher in boreal summer and lower in boreal winter. The temperature of CPT between the TP and the Plain regions has a relative large difference, increasing up to 4.2°C, to the height that just maintains within 0.6 km all year long. Resulting from the CPT has a relative large interannual variability, the seasonal cycle of the CPT height between the TP and the Plain region exhibit smaller differences than the monthly mean. However, the maximum difference of the LRT height occurs in August with 1.1 km, when their temperatures differ barely.
 The variation of the LRT has a strong connection with the thermal properties of the atmosphere [Highwood and Hoskins, 1998; Hoinka, 1998; Randel et al., 2000; Gettelman and de F. Forster, 2002; Bischoff et al., 2007; Selkirk et al., 2010]. It reaches the maximum height at 18.4 km with lowest temperature at −71.4°C in August and drops to the minimum height at 13.1 km with highest temperature at −57.4°C in February over the TP, while it reaches the maximum height at 17.4 km with −68.7°C and the minimum height at 13.4 km at −58.7°C over the Plain. The seasonal variations of the LRT height and temperature show anticorrelation, which has a great agreement with previous studies [Randel et al., 2000; Santer et al., 2003; Schmidt et al., 2004; Schmidt et al., 2006]. For both regions, however, those of the CPT seems have positive correlation. They are somewhat higher in the winter half of a year and somewhat lower in the summer half of a year. This result is inconsistent with the results of Gettelman and de F. Forster , which suggested that the CPT is the lowest during winter and the highest during summer in the tropical zone (25°N–25°S). However, the opposite seasonal variations of the CPT height were demonstrated by Seidel et al. , when they designate the zone of 10°N–10°S as the tropics. This indicates that the CPT presents different seasonal variation with latitude.
 As an outstanding heat source in boreal summer, the TP thermally raises the LRT to a higher altitude, which is clearly shown in our study. To explore the topographic effects of the TP, the climatological tropopause parameters along 32.5°N are displayed in Figure 10. In the western half of the TP where the average elevation exceeds 5000 m, the CPT height resides at ∼18 km, whereas it drops to ∼15 km in the eastern plain of China where the average altitude is ∼100 m. The TP plays a dynamical role in the CPT lift since the CPT height ascends with the increasing elevation from the Plain to the TP plus the small in situ seasonal variation along 32.5°N (Figure 10a). This kind of topographic lift is hardly seen in the climatological mean and winter mean of the LRT (Figure 10b). In boreal summer, however, forced by the surface heating, the altitude of the LRT increases rapidly not only over the TP but also over the Plain. What is more, the LRT is boosted to a much higher altitude over the TP that has much stronger surface heating, which makes the difference of the LRT altitude between the TP and the Plain reaches ∼2 km.
5. Discussions and Conclusions
 This work reveals the observational characteristics of the tropopause over the TP from COSMIC RO data. The evidence is from our analyses based on PDF, monthly and climatological means, and relation to the vertical and temporal variation of the temperature profiles from COSMIC data. Comparisons are made between the COSMIC RO measurements over the TP as well as over the Plain in the data set covered from June 2006 to December 2009.
 As the strong heating near the TP surface triggers intense convective mixing in the lower troposphere, the temperature over the TP is warmer than that over the Plain at the same level of lower troposphere. The temperature change over the TP is also larger, which proves that the TP is one of the most sensitive regions to climate change [Liu and Zhang, 1998; Liu and Chen, 2000].
 Statistics show that there is an anticorrelation in the height and temperature variations of the LRT while a positive correlation in those of the CPT. The LRT is highest at ∼19 km and coldest at approximately −72°C in boreal summer, but lowest at ∼13 km and approximately −56°C in boreal winter. Although the CPT is somewhat higher and warmer in the winter half of the year, and somewhat lower and colder in the summer half, the variability of CPT exhibits only ∼1.5 km in height and ∼10°C in temperature. The LRT has a strong correlation with thermal properties of the atmosphere. Attributed to the strong convective mixing that stems from the surface heating in boreal summer, the LRT over the TP then shows a higher altitude than that over the Plain with the maximum of 2.5 km, while it exhibits a lower altitude in boreal winter. However, the CPT over the TP is subtlety higher than that over the Plain all year long. Resulting from a relative large interannual variability, the differences of seasonal variation of the CPT parameters between both regions have smaller values than their monthly means that can increase up to 1.4 km.
 In addition, the LRT height is strongly dependent on the subtropical jet that plays a crucial role in the formation of tropopause fold or/and multiple tropopauses, which makes fairly bimodal distribution in the LRT PDF. With the intensification of the subtropical jet in boreal winter, the peaks of the LRT PDF are at 11 km and 16 km over the TP as well as 11 km and 15 km over the Plain. Since the altitude of the lower LRT approaches the higher one in boreal summer, the PDF of the LRT located at 16 km increases to 58% over the TP and to 39.2% over the Plain. For the complete analysis period, the CPT stably resides at 17–18 km with the PDF of 72.52% over the TP and 69.22% over the Plain.
 Along the latitude of 32.5°N, the TP presents a great impact on the ascended tropopause. It dynamically lifts the CPT owing to its elevated topography, and thermally pushes the LRT upward in boreal summer owing to its outstanding surface heating. Along 32.5°N, the CPT is located at ∼18 km over the main body of the TP (elevation, ∼5000) and drops to ∼15 km over the Plain (∼100 m) without in situ seasonal variations. However, the LRT over the TP is thermally pushed up by ∼2 km compared with that over the Plain in boreal summer. Note that the tropopause difference between the TP and the Plain along 32.5°N greatly differs with that of monthly mean or seasonal cycle. There are two reasons. One is due in part to the meridional temperature gradients increase over the extratropics. The climatological mean tropopause difference between the TP and the Plain based on a 5° × 8° area (Figures 8 and 9) has been smoothed compared with that along 32.5°N (Figure 10). Another is likely to result from the different distribution of the ROs between the 5° × 8° area and the 0.5° × 0.5° along 32.5°N. Añel et al.  found that the distribution of the soundings and the possible influence must be taken into account.
 The results of the tropopause-related researches, such as the overshooting convection and the stratosphere-troposphere exchange, may vary significantly depending on the tropopause definition. Highwood and Hoskins  suggested that the LRT is an arbitrary definition for limitations in physical relevance. The CPT definition is more important for describing the tropopause characteristics because it correlates better with the dynamical process [Schmidt et al., 2004], which is demonstrated well in our study. Recently, it has become clear that there exists a layer that has properties of both the troposphere and the stratosphere. Therefore, to constitute a clear distinction between troposphere and stratosphere is inadequate [Gettelman and de F. Forster, 2002; Liu and Zipser, 2005; Fueglistaler et al., 2009]. Emphasizing the tropical tropopause layer (TTL) associated with temperature and the circulation, Fueglistaler et al.  synthetically set the bottom of the TTL at 150 hPa, 355 K, 14 km (pressure, potential temperature, and altitude) and the top at 70 hPa, 424 K, 18.5 km. However, a synthesis definition for the subtropical or extratropical tropopause has still not well presented by far for a complicate circulation there.
 As an all-weather system with global coverage and high vertical resolution, practically unaffected by clouds, precipitation, and aerosols, the COSMIC RO observation offers a new insight into the characteristics of the tropopause-related studies. The COSMIC RO data provides the basis for thermal definitions, but is not appropriate to present dynamical definitions (such as based on the potential vorticity) because of no wind information in the COSMIC data. However, proper specification of the multivariate forecast error correlations in data assimilation systems should improve the analysis of wind increment when assimilating RO data [Liu et al., 2007], which should be encouragingly benefit for the estimation of the characteristics of dynamical tropopause. We will pursue the data assimilation approach for better analysis of the tropopause structure in the future.
 The authors are greatly appreciative to the Meteorological Information Center of China for providing radiosonde observations, to the COSMIC Data Analysis and Archive Center (www.cosmic.ucar.edu) for providing the COSMIC RO data, and to the NCEP/NCAR reanalysis project at the NOAA/ESRL Physical Sciences Division for providing the zonal wind data. This work was supported by the National Basic Research Program of China (973 Program, grant 2010CB428601), the Knowledge Innovation Program of the Chinese Academy of Sciences (grant KZCX2-YW-Q11-04), and the NSFC (40730950, 41075041).