TP-PROFILE: Monitoring the Thermodynamic Structure of the Troposphere over the Third Pole

Over the past decade, meteorological applications of ground-based microwave radiometers (MWRs) have remarkably increased since MWRs have proven to be effective in various fields relative to conventional radiosonde observations. Automated and continuous thermodynamic profiling of the atmosphere up to 10 km above ground level (AGL) with a high temporal resolution using MWRs is critical for monitoring real-time thermodynamic states (Gaffen and Elliott, 1993; Güldner and Spänkuch, 1999, 2001), as well as for improving short-range forecasts of rapidly changing weather phenomena (Knupp et al., 2009; Madhulatha et al., 2013; Kim and Lee, 2015; Caumont et al., 2016; Illingworth et al., 2019; He et al., 2020; Martinet et al., 2020; Temimi et al., 2020). MWRs have been used in previous studies of boundary layer thermodynamics, clouds, and precipitation (Güldner and Spänkuch, 2001; Knupp et al., 2009; Madhulatha et al., 2013; Kim and Lee, 2015). Ground-based MWRs can obtain continuous temperature and humidity profiles with a high temporal resolution and can provide frequently available thermodynamic information, which is very beneficial for severe weather nowcasting (Madhulatha et al., 2013). Ground-based MWRs can close the observational gap in the atmospheric boundary layer (Illingworth et al., 2019). Therefore, the consortium of European Meteorological Services mandated a program to produce a European network of existing MWRs (Rüfenacht et al., 2021). The Atmospheric Radiation Measurement (ARM) program also operates a network of MWRs in various locations around the world.

Due to its topography, the Tibetan Plateau (TP) surface absorbs a large amount of solar radiation energy (Ye, 1981), which makes the TP an extremely large elevated heating source (Duan and Wu, 2005). Because of the height of the TP, the heat is added directly to the middle and upper troposphere and is used by only half of the total mass of the atmosphere. Thus, the same amount of heat will be used more effectively over the plateau than over the adjacent low-level terrain. There are also many mountains on the plateau, which are the sources of convective and precipitation activities (Ye, 1981; Son et al., 2019). Luo and Yanai (1984) suggested that dry (unsaturated) thermal convection rising from the heated surface can penetrate the upper troposphere; hence, there is a deep layer of turbulent convection occupying the entire troposphere over the western TP (Ueda et al., 2003; Chen et al., 2016; Lai et al., 2021). Such convective development can effectively warm the atmosphere (Taniguchi and Koike, 2007). Intense convective activity not only maintains a particular large-scale circulation pattern over the TP but also transports large quantities of sensible heat, moisture, chemical pollutants, and air with a low ozone concentration from the near-surface layers to the upper layers (Chen et al., 2013). Sensible heating tends to destabilize an air column (Ueda et al., 2003), permitting the downward transfer of westerly momentum in the vicinity of the jet stream (Shen et al., 1986; Lai et al., 2021). In winter, the plateau is situated in the latitudes of the westerly jet stream; and in summer, it is located at the juncture of the westerlies and South Asian monsoon. Because of the low air density and strong surface heating on the TP, the air is mixed up to 8-9 km above sea level by dry thermal updrafts (Yang et al., 2004; Chen et al., 2016). Routine ground-based tropospheric observations are sparse on the TP, which makes thermodynamic studies on the TP challenging, consequently is difficult to make reliable estimates of the TP’s heating effects (Seto et al., 2013). All of these elements have constrained the development and validation of weather forecasting models for the TP region.

Ground-based MWR enables the quasi-continuous monitoring of the thermodynamic state of the lower troposphere, which is only interrupted during moderate and heavy precipitation events. MWRs measure the downwelling of the thermal emission originating from Earth’s atmosphere in the microwave band. The radiance observations are commonly expressed as an equivalent brightness temperature (TB), from which estimates of atmospheric temperature profiles (from oxygen absorption at 55–60 GHz) and humidity profiles (from water vapor absorption around 22 GHz), as well as the column-integrated water vapor (CWV, unit: cm), can be inferred under non-precipitation conditions. The MWR profiling capability in the lowest atmosphere according to the observation geometry and scanning strategy (Rüfenacht et al., 2021) has been proven to be valuable because of the poor sampling inherent to satellite measurements (Illingworth et al., 2019). MWR retrievals have been found to capture near-surface temperature inversions very well (Martinet et al., 2017). The high temporal resolution of an MWR allows it to resolve detailed mesoscale thermal and humidity profiles for analysis of rapidly changing mesoscale and/or hazardous weather phenomena (Knupp et al., 2009). The MWRs have reached a high level of maturity (Güldner and Spänkuch, 2001; Illingworth et al., 2019), and therefore, they are suitable for operational networks and can contribute to closing the observational gaps on the TP.

Recently, the Tibetan Plateau atmospheric profile (TP-PROFILE) project installed a ground-based MWR network on the TP (Fig. 1). Nine MWR stations (Ma et al., 2023) were used to cover the different regions of the TP. Short-period (2 min) temperature and humidity soundings up to a height of 10 km are retrieved from the ground-based 22-channel MWR observations recorded by this network. In contrast to radiosondes, these radiometric retrievals provide very high temporal resolution (2 min) thermodynamic profiles. MWRs have the advantage of continuous monitoring of the atmosphere, which covers the temporal and spatial gaps of radiosonde measurements over the TP. The primary objective of this paper is to introduce this observation system and demonstrate its usefulness for studying the basic characteristics of convective activity over the TP region.

Figure 1.  The location of the nine microwave radiometer sites and observation environment around the instrument. MWR sites are marked by the blue hexagram. Radiosonde sites that were used to compare with the microwave radiometer retrievals are marked with a “+”. The red “+” radiosonde data is observed by drone sounding. The black “+” represents the CMA radiosonde site.

The remainder of this paper is organized as follows. Section 2 introduces the MWR system and section 3 demonstrates the radiosonde measurements collected by this study. Section 4 evaluates the MWR accuracy and outlines our preliminary findings from the MWR dataset and section 5 provides a discussion and conclusion.

The MWR instrument used in this study was an MWP967KV temperature, humidity, and liquid profiler (Lei et al., 2021), which operates in eight K-band (22.235, 22.50, 23.035, 23.835, 25.00, 26.235, 28.0, and 30.0 GHz) and fourteen V-band (51.25, 51.760, 52.28, 52.8, 53.34, 53.85, 54.4, 54.94, 55.50, 56.02, 56.66, 57.29, 57.96, and 58.80 GHz) microwave channels. It measures the radiation intensity of the sky in 22 different frequency channels. A neural network retrieval method has been constructed to retrieve the temperature, water vapor density, and liquid water density profiles, as well as the CWV, from the brightness temperature measurements. The neural network methods have been trained by the manufacturer based on historical radiosonde data of the Chinese Meteorological Agency and the Monochromatic Radiative Transfer Model (MonoRTM) (Clough et al., 2005) to simulate the observations of a radiometer. The neural network algorithm was trained by creating a high-resolution radiosonde dataset for 2006–2009 from a location with a similar altitude and climatology to the MWR site. During the field experiment, the nine MWRs were configured to conduct zenith scans. Only the zenith scan data were used in this study to retrieve the thermodynamic profiles of the troposphere for levels as high as 10 km, which were normally available every two minutes. A separate infrared radiometer installed on each of the nine MWRs received downwelled infrared radiation in the 9.6–11.5 μm band, which was converted to a zenith infrared temperature and the height of the cloud base. The surface temperature, humidity, wind, and pressure sensors on the MWRs also provided the surface meteorological conditions. Absolute calibration was performed every year using liquid nitrogen. The profiles associated with the presence of rainfall were detected by the rain detector of the radiometer and were removed in this study.

The TP-PROFILE MWR system has nine MWR sites, including the Muztagh Ata Westerly Observation and Research Station, Chinese Academy of Sciences (MAWORS), the Ngari Desert Observation and Research Station, Chinese Academy of Sciences (NADORS), the Qomolangma Atmospheric and Environmental Observation and Research Station, Chinese Academy of Sciences (QOMS), the Nagqu Station of Plateau Climate and Environment, Chinese Academy of Sciences (Nagqu), the Southeast Tibet Observation and Research Station for the Alpine Environment, Chinese Academy of Sciences (SETS) (Ma et al., 2008, 2020), and the Mangya, Qamdo, Leshan, and Kabu stations. The MAWORS, NADORS, and Mangya stations were located in the westerly-dominated area and the SETS, Kabu, Qamdo, and Leshan were located in the monsoon-dominated area. The QOMS and Nagqu stations were located in the westerly–monsoon transition domain. Kabu was also located in the water vapor channel within the Yarlung Zangbo Grand Canyon. The locations of the MWR sites are listed in Table 1.

This paper has collected radiosonde observations from the eight sites nearest to the nine MWR sites. The radiosonde information used in this paper is presented in Table 1. We collected radiosonde observed temperature and humidity profiles at six sites (Kashi, Mangya, Nagqu, Qamdo, Wenjiang, and Dingri), from 1 November 2020 to 20 May 2021. The vertical temperature and humidity soundings at the six sites were obtained by radiosonde balloon flights launched at around 0800 and 2000 (Beijing Standard Time) daily. The radiosonde data used for Kabu and SETS MWR evaluation were collected by an intensive experiment supported by the Second Tibetan Plateau Scientific Expedition and Research Program (Chen and Xu, 2022). The temperature and humidity profiles at Kabu and SETS were observed by drone soundings. The height of the drone sounding is less than 10 km. Consequently, they are not used to evaluate the freezing level height and column water vapor obtained from the MWR, but they are used to verify the vertical profiles of the MWR. Notably, there is no radiosonde observation around NADORS.

Kashi is about 130 km away from MAWORS. Wenjiang is around 140 km away from Leshan. Dingri is about 82 km from the QOMS site. There are mountains and hills between the paired MWR and radiosonde sites. Consequently, these three radiosonde sites were not used to check freezing level height and column water vapor retrievals from the MWRs. The Mangya, Chandu, and Nagqu stations have a radiosonde site quite close to the position of their MWRs. Therefore, their respective nearby radiosonde sites were selected to evaluate the performance of freezing level height and column water vapor retrieved from the MWR. Radiosonde data at different heights have different sampling times and the MWR has a temporal resolution of two minutes, so it is possible to select MWR retrieval values that have the same time and height as that of the radiosonde. Upon comparing the MWR against the radiosonde, a height-time linear interpolation was conducted to ensure proper sampling between the MWR and radiosonde. This method guarantees that we have compared the MWR and radiosonde at the same time and height.

The predictability of precipitation occurrence and the estimation of rainfall intensity were investigated by Won et al. (2009) using the brightness temperature (TB) of the MWR. Their work demonstrates that the 30 GHz brightness temperature and rainfall intensity have a non-linear relationship. They further found that the 30 GHZ brightness temperature can be used to derive rainfall intensity. Won et al. (2009) demonstrated that the observed and estimated rainfall intensity from the MWR at low-elevation areas showed substantially high correlation coefficients. However, the TP has different rainfall microphysical characteristics in low-altitude areas (Xu et al., 2023), a fact that motivates us to analyze the relationship between the brightness temperatures of the MWR and the rainfall intensity on the Tibetan Plateau. Figure 2 demonstrates the variation of TBs before and after rainfall events during 20–24 September 2019 at the Kabu site. Before rainfall, a pronounced increase of TB was observed in water vapor channels (22–30 GHz), while the TBs in 54–59 GHz, the oxygen channels, remained nearly constant, except for 51–52 GHz. Therefore, forecasts of precipitation occurrences could be accomplished by using the preceding increases of TBs at 22.2 GHz, 30.0 GHz, and 51.2 GHz. Figure 2 also demonstrates a similar result to that of Won et al. (2009), showing that the TBs of 22–30 GHz were the highest when rainfall intensity that occurred during the night of 23 September was stronger than for other periods during 20–24 September 2019. The rainfall intensity on the TP can be further estimated by the MWR TBs using a nonlinear regression analysis method. The great increase of TB at 51–52 GHz during rainfall time is related to cloud liquid water. Therefore, these sensitive bands of cloud liquid water could be also further used to improve the cloud liquid water profile estimation on the TP.

Figure 2.  Brightness temperature variation of 22 channels (22.235−58.8 GHz) during 20–24 September 2019 at the Kabu site. The blue bars show the hourly rainfall rate (units: mm h^–1) and its y-axis is 0–6 mm h^–1.

Available radiosonde sites and MWR sites are not colocated. This makes it impossible to perform a comprehensive evaluation of the MWR data. But the climatology of MWR and radiosonde sites can be compared, due to the continuous-medium nature of the atmosphere. Therefore, we collected radiosonde data from eight sites near the MWR stations to compare the climatology of the atmosphere profiles. There is a 1–3 K systemic bias between the paired MWR and radiosonde site. Figure 3 presents the comparative results at eight MWR sites, and Fig. 4 compares their mean temperature and mixing ratio profiles after we have done the systemic bias calibration. The mean bias (MB) of the temperature at Mangya, Nagqu, Qamdo, Leshan, and QOMS is lower than 1 K. The MB of humidity mixing ratio at MAWORS, Mangya, Qamdo, and Leshan is lower than 0.1 g kg^–1. The mean bias of the seven MWRs is <1.0 K near the surface and <2.2 K up to 6.5 km. For water vapor, the mean bias values are 0.1–0.3 g m⁻³ near the surface and −0.15 g m⁻³ at altitudes of 2–4 km. The accuracy results for the MWRs are in agreement with the conclusions of a previous study (Güldner and Spänkuch, 2001). These results confirm that the MWRs are potentially useful in applications requiring high temporal resolution data or data for data-sparse regions of the TP.

Figure 3.  A comparison of (a1–h1) air temperature (Ta, units: K) and (a2–h2) mixing ratio (MR, units: g kg^–1) between the MWR and nearby radio sounding data, 10 km above ground level, after systematic bias calibration.

Figure 4.  A vertical profile comparison of (a1–h1) averaged air temperature (Ta, units: K) and (a2–h2) mixing ratio (MR, units: g kg^–1), as derived from radiosonde (Rs) and ground-based microwave radiometers (MWR)

Figure 4 shows the mean profiles for the respective radiosonde observation period, which serve as a test if the retrieval algorithm uses an appropriate climatology for the radiosonde period. There is less variance in the temperature climatology than that of the mixing ratio. The MAWORS and SETS stations have a high bias in the layer 5-km AGL layer. QOMS has a relatively high bias in its mixing ratio estimation in the 2-km AGL layer. These greater differences could be explained due to the valley location of the MWRs at these three sites. The local topography at these sites may influence the accuracy of neural network retrievals, which have been trained by the radiosonde observations in a different surrounding environment. Thus, it is better to use more radiosondes in the valleys of the MAWORS, SETS, and QOMS stations to re-train the neural network retrieval algorithms for the three sites. Unfortunately, the radiosonde observations at the three sites is quite limited due to the high cost. The Mangya, Nagqu, and Qamdo sites perform better in the climatology estimation. This means that the neural network method is appropriate at the three sites.

The TP forms an isolated atmospheric warm-humid island (Wu et al., 2007; Xu et al., 2008). The atmosphere over the plateau in summer is constantly in an unstable convective state. There is a high rain rate center 6 km above the plateau, which forms a tower penetrating the middle troposphere against the nearby background (Fu et al., 2006), implying that a unique latent heat source is being injected directly into the middle atmosphere. There are more isolated rain cells over the plateau than over nearby regions, and the strongest diurnal cycle of rainfall over the plateau reaches a peak at around 1600, and over the valley at around 0500 local time, indicating the dominance of convective clouds caused by solar heating (Fu et al., 2006). Figure 5 shows our observations of the atmospheric warm-humid island above the TP. The warm-humid island is obvious in 590–400-hPa layer (Xu et al., 2008). Thus, the mean air temperature and humidity in the atmosphere at 4.5–7.5 km AGL was taken as an indicator of the intensity of the warm-humid island. The warm-humid island is clearly captured by the MWRs at NADORS, Nagqu, Qamdo, and QOMS stations. Other sites have shown relatively weak warming and humidity variations in the middle troposphere.

Figure 5.  The mean diurnal variation of (a1–i1) mean air temperature (4500–7500 m AGL, units: K), and (a2–i2) mean water vapor density (4500–7500 m AGL, units: g m^–3) at the nine sites for each day of 2021. Its x-axis is day of the 2021 year, y-axis is the hour.

The freezing level height (FLH) is the altitude of the location where the air temperature is 0°C. The CWV is defined as the depth of liquid water that would be present in a column of unit cross section if all of the water vapor in that column throughout the depth of the atmosphere were condensed (Madhulatha et al., 2013). The monitoring of FLH and CWV using ground-based MWRs is of particular interest since there is presently no alternative method of routinely measuring these parameters. An MWR can provide us with FLH, and CWV variations at minute temporal resolution. Before analyzing their variations, we used the radiosonde results to check their accuracy against what was obtained from the MWR. Figure 6 demonstrates the performance of the MWR retrievals of FLH and CWV. The determination coefficient for both FLH and CWV is higher than 0.75, and the biases for FLH and CWV are lower than 164 m and less than 2 mm, respectively. This demonstrates that the retrievals of FLH and CWV from the MWR correspond to those of radiosondes very well. Thus, they could be used for further analysis.

Figure 6.  Comparison of (a, b, c) FLH and (d, e, f) CWV as derived from the MWR and nearby radiosonde data at the (a, d) Qamdo, (b, e) Mangya, and (c, f) Nagqu sites.

Whether the accuracy of the monthly mean CWV values obtained using low-temporal resolution monitoring systems (satellites, radiosondes) is strongly or only slightly affected by their diurnal trends is an important scientific question for research on the TP. Based on two years of measurements with a temporal resolution of 2 min, the monthly and diurnal mean findings regarding the CWV over the TP are presented in Fig. 7. Leshan has the highest CWV among the nine sites, followed by Kabu, SETS, Qamdo, QOMS, Mangya, Nagqu, NADORS, and MAWORS. The daily average of the CWV remained within the range of 10–60 mm (Fig. 7a). The diurnal cycle of the CWV was weak at MAWORS, NADORS, Mangya, Nagqu, Qamdo, and Leshan. The Kabu and SETS sites have a slightly stronger diurnal variation. As the averaging procedure suppresses advectively induced CWV changes to some extent, the mean inter-hourly variations reflect dew formation and evapotranspiration around the site (Güldner and Spänkuch, 1999). The nearly constant CWV during the night and day indicates insignificant dew or rime formation and surface evapotranspiration at the nine sites.

Figure 7.  The mean (a, b, c) diurnal and (d, e, f) monthly variations of column water vapor (CWV, units: mm), freezing level height (FLH, units: km, AGL), and cloud base height (units: km, AGL) at the nine sites.

Figure 7d shows the monthly mean CWV values for the nine sites. The CWV values were highest in summer and lowest in winter. The smooth seasonal variation of CWV demonstrates that the accuracy of the monthly mean CWV values based on the monitoring systems with a low-temporal resolution (i.e., satellites and radiosondes) was only slightly affected by their diurnal trends over the TP. However, the low-temporal resolution may lead to large uncertainties at the sites that do have a clear diurnal variation, such as the SETS and Kabu sites that are located in the mountain valleys on the TP.

The FLH has a clear diurnal and seasonal variation at all nine sites with peak values occurring in the afternoon. Summer months show the highest FLH among the seasons. The Leshan site demonstrated the highest FLH among the nine sites, which also had the highest CWV. The MAWORS, NADORS, Mangya, and Nagqu sites feature a flat variation in their diurnal changes of cloud base height. Meanwhile, the remaining five sites have a clear diurnal variation in their cloud base height. The seasonal variation of cloud base height is quite chaotic, yet the seasonal variation of CWV and FLH are consistent among the nine sites.

We assume that a cloudy sky characterizes the situation when the cloud base height is below 10 km AGL, and a clear sky condition occurs when the cloud base height is higher than 10 km AGL. All sky includes cloudy sky and clear sky conditions. Figure 8 shows the probability distributions of CWV and freezing level height (FLH) under clear and cloudy conditions during the past two years. The maximum percentage of the FLH at seven sites (MAWORS, NADORS, Mangya, Nagqu, Qamdo, QOMS, and Kabu) was 1–3 km under cloudy conditions (Fig. 8b1). A peak percentage of 1–3 km AGL cloud base height also exists for six sites (MAWORS, Nagqu, Qamdo, Leshan, QOMS, and SETS) (Fig. 8c1). The FLH was lower at SETS than at the other sites. Kabu had a lower cloud base height than the other sites. The maximum percentage of the CWV under cloudy was in the range of 3–12 mm (Fig. 8a1), while under clear conditions it was in the range of 3–10 mm (Fig. 8a2). A cloudy sky condition has a wider percentage distribution than clear sky conditions. The MAWORS, NADORS, and Mangya sites on the northern TP had peak percentage values of cloud base temperature at −55°C, −67°C and −45°C, respectively (Fig. 8c2), while Leshan, SETS, and Kabu on the southeastern TP had their highest percentage close to 0°C. This means that there were frequent supercooled clouds over the westerly-wind-dominated northern TP, but such clouds were rare over the monsoon-dominated southeastern TP. This analysis of cloud indicators is expected to improve satellite remote sensing precipitation data, which uses cloud temperature to classify cloud patch groups (Huffman et al., 2019).

Figure 8.  The occurrence probability distribution functions (PDF) comparisons of (a1, a2) column water vapor (CWV, units: mm), (b1, b2) freezing level height (FLH, units: km, AGL), (c1, c2) cloud base height, and temperature (units: °C). The PDF results for the cloudy (clear) sky period are demonstrated in a1 and b1 (a2 and b2). All sky (cloudy and clear sky) results are shown in c1 and c2.

The amount of water vapor in the column available for rainfall should affect the amount of precipitation generated. The CWV under clear conditions often becomes the antecedent condition for the development of heavy precipitation. An MWR provides useful indicators of the accumulation of water vapor before the occurrence of heavy rainfall and can provide useful data for the nowcasting of intense convective weather (Wei et al., 2021). To examine the temporal variations in the water vapor indicators of rainfall, lagged composites of the CWV are calculated. The lag composites of the CWV for the identified rainfall events are shown in Fig. 9. The time of occurrence of the storm is considered to be lag 0, while lag −12 corresponds to the pre-environment, 12 h prior to the storm occurrence. In this analysis, the hour corresponding to the storm occurrence is set to zero. The CWV twelve hours before the storm occurrence were superposed and averaged to produce one composite curve. By observing the variations in the composite time series of the CWV, it was possible to explain the prerequisites necessary for the genesis of thunderstorm activity. Most of the stations recorded an increased CWV before the rainfall event occurred, except for the Kabu station. Kabu recorded only a slight change in the CWV during the twelve hours before the rainfall event. Meanwhile, Chen et al. (2023) reported that both CAPE and water vapor conditions play a role in rainfall intensity at the Kabu site. This may be explained on the premise that other factors may be playing a more important role in generating rainfall at the Kabu site since water vapor is sufficient in the Yarlung Zangbo Grand Canyon.

Figure 9.  The CWV variation in the twelve hours before rainfall. CWV is normalized by its values at rainfall time. Zero lag time indicates the rainfall time. Negative-hour values represent the number of hours before the rainfall event.

The construction of the infrastructure for the TP-PROFILE MWR monitoring network was completed, and the observation network ran smoothly throughout 2021. To illustrate the capabilities of the TP-PROFILE MWR and its potential benefits for research and operational activities, we present an analysis example of the radiometric retrievals. The TP-PROFILE MWR not only provides valuable in situ observations of the troposphere over the TP region but also improves our understanding of the influence of the TP on the atmosphere. There are no significant diurnal variations in the CWV, except in the deep valley on the southeastern TP, which is dominated by the general circulation of synoptic-scale weather. The radiometric retrievals of the TP-PROFILE MWR system compare fairly well to the corresponding values observed from operational sounding sites. Since the measurement principles of an MWR and a radiosonde are different (volume integral above a fixed location on the ground for radiometers vs. point measurement of a drifting balloon for radiosondes), there are biases and spreads of the data points, but the two datasets exhibit similar trends for the evolution of the temperature and humidity within the troposphere.

The use of radiosonde measurements as ground truth reference data is also worthy of debate. Notably, a radiosonde may sample a different air mass than a zenith-pointing ground-based MWR. Some of the paired radiosonde sites are more than 100 km away from the MWR sites. There are mountains and hills between the paired MWR and radiosonde sites, which makes it unreasonable to use the nearest radiosonde data to evaluate the MWR retrievals. The variability of water vapor and temperature is relatively low above the boundary layer because the flow is determined by large-scale advection which tends to promote homogeneity. Therefore, radiosonde data are included in our evaluation.

The MWR has a temporal resolution of two minutes. The radiosonde flies with the balloon for more than one hour before falling to the ground. When we compare the MWR against radiosonde data, we selected MWR retrieval values that have the same time and height as those of radiosonde. Meanwhile, reanalysis data usually provide us with hourly atmosphere profiles. It is not possible to choose the reanalysis values at the same time and same height as that of a radiosonde. These considerations present complications for comparing the accuracy of MWR and reanalysis data, based on the current availability of the radiosonde observation. The available radiosonde sites and MWR sites are not collocated. This is the reason why a strict evaluation of the MWR data with these available radiosonde data is also not possible. However, the climatology of paired MWR and radiosonde sites can be compared since the atmosphere is a continuous medium.

In addition, the data quality of MWR still needs enhancement when more ground truth observations have been collected. Massaro et al. (2015) formulated a novel and promising method of improving MWR profile retrievals in a mountainous region by adding further information to the retrieval, such as the surface temperature at fixed levels along a topographic slope or on nearby mountain tops. Other observational data collected during separate TP atmospheric experiments may potentially be used, e.g., the third atmospheric scientific experiment over the TP (Zhao et al., 2018), to improve the TP-PROFILE retrieval accuracy.

MWRs are mainly used under non-precipitation conditions because radiometer measurements become less accurate in the presence of a water film on the outer housing (radome) of the equipment. They are equipped with a hydrophobic radome surface and a blower that blows heated air over the radome to minimize the effect of rain on the radiometer measurement. Thus, a MWR can generally provide profiles under cloudy and light-precipitation conditions. The cloud base height is assumed to be the height at which the IR temperature is equal to the temperature of the retrieved profile, further noting that bias in the retrieved temperature profiles could influence the accuracy of the cloud base height. A radiosonde determines the cloud base height using a relative humidity threshold [e.g., Karstens et al. (1994) used a threshold value of 95%]. When the humidity exceeds this threshold value, the appearance of liquid water or the formation of a cloud is expected. We found that a 95% threshold value is not accurate for the Tibetan Plateau. The cloud base height retrieved from radiosonde also may not be reliable. Thus, it is not suggested to use cloud base heights from radiosondes to verify MWR-derived cloud base heights until the radiosonde threshold method is improved. Lidar data could provide a more reasonable cloud base height. Unfortunately, lidar data is currently not available on the TP. However, we are planning to employ lidar to verify the cloud base height in the future.

We have collected data for three years, and more data are expected to become available in the future after careful equipment maintenance and updates. The datasets from the TP-PROFILE MWR will be updated annually and made freely available to the scientific community. Researchers who are interested in this dataset can send their request to the corresponding author.

Acknowledgements. This research was supported by the Second Tibetan Plateau Scientific Expedition and Research Program (STEP) (Grant Nos. 2019QZKK0103 and 2019QZKK0105) and the National Natural Science Foundation of China (Grant Nos. 41975009, 42230610, 41840650 and U2242208), the Youth Innovation Promotion Association of the Chinese Academy of Sciences (Wang Binbin, 2022069).