Precipitable water vapor on the Tibetan Plateau estimated by GPS, water vapor radiometer, radiosonde, and numerical weather prediction analysis and its impact on the radiation budget



[1] Precipitable water vapor amounts (PW) determined by Global Positioning System (GPS), radiosonde and operational numerical weather prediction (NWP) system analysis at three stations (Naqu, Gaize, and Deqin) on the Tibetan Plateau are compared. PW measured by water vapor radiometer at Naqu and a low-elevation site, Xian, is used for calibration. The results show that the PW determined by NWP analysis in these regions is comparable with that of the radiosonde measurements but that they both are systematically smaller than those determined by the GPS measurements. The averaged difference of PW between GPS and radiosonde estimates is about 1.75 mm, and that between GPS and NWP analysis can be as large as 7.75 mm. These differences are relatively larger than those reported in the literature because the absolute PW in this region is much smaller. The effect of such large differences on the surface radiation budget is evaluated using a radiation model. The results show that both longwave and shortwave radiative fluxes at the surface determined using the model analysis profiles with the water vapor corrected by the GPS PW are closer to the observations compared with those without water vapor correction. The flux difference at the surface with and without water vapor correction is about 20 W m−2 in the shortwave and 30 W m−2 in the longwave. These differences are much larger than that caused by doubling the concentration of carbon dioxide in the atmosphere in this region.

1. Introduction

[2] Water vapor is an important element in numerical weather prediction (NWP) and climate studies because by its cycling through evaporation and condensation, water vapor transports latent heat energy between the surface and atmosphere. This process plays a crucial role in the global energy and hydrological cycles. Water vapor is a dominant greenhouse gas in the atmosphere and is a strong absorber of both solar and infrared radiation. Long-term variation in its total global content could potentially be used as an indicator of global climate change [Yuan et al., 1993].

[3] The distribution of water vapor is highly variable in both space and time. Lack of precise and continuous water vapor measurements is one of the major sources of error in short-range weather prediction [Kou et al., 1993, 1996]. The traditional technique for water vapor measurements is to launch radiosondes, normally twice a day. This method is not only expensive but also poor in both spatial coverage and temporal resolution. In recent years, a new observational technique, based on the Global Positioning System (GPS) that is sensitive to the spatial and temporal distribution of the water vapor content in the atmosphere has made it possible to retrieve precise and continuous estimates of water vapor with spatial density governed by the number of receivers deployed. Many GPS based estimates of water vapor have been compared with the estimates based on radiosondes, water vapor radiometers (WVR) and NWP analysis [Yang and Sass, 1999; Kopken, 2001; Liou and Teng, 2001] and the results indicate that the GPS estimates of water vapor are in generally good agreement with the measurements of radiosondes and WVR. Emardson et al. [1998] reported that the differences between WVR and GPS observations of precipitable water (PW) are about 1–2 mm. Tregoning et al. [1998] compared PW estimates from GPS, radiosondes and WVR, and their results show differences of about 1.4 mm between any two methods. Liou and Teng [2001] examined measurements carried out in the tropical region where the water vapor burden is higher and more inhomogeneous. They have shown that the difference in PW between WVR and GPS is about 2.2 mm. Penna et al. [2003] evaluated GPS PW data from the Australian national network and confirmed the results reported in the literature except over Antarctica where the difference between GPS and radiosonde-derived PW is relatively large. They also presented diurnal variability of PW at the Australian regional GPS network locations. A relatively large difference of 4.7 mm in PW between GPS and radiosonde in a mountainous area of Sumatra Island was also reported by Wu et al. [2003]. They compared PW determined using the two methods for 1–27 August 2001 and found that the GPS based estimates of PW were systematically higher than those determined using the radiosonde. They claimed the cause for such a large difference is a dry bias in the Vaisala radiosonde measurements. Nakamura et al. [2004] made a series of comparisons for two types of radiosondes and they confirmed a dry bias of 3–4 mm from the Vaisala RS80-A radiosonde. Takagi et al. [2000] reported a comparison between GPS PW and radiosonde measurements at Lhasa on the Tibetan Plateau. Their results also show that the GPS PW is systematically higher than the radiosonde estimate.

[4] All these studies have demonstrated the advantages of the application of GPS based estimates of PW. The most suitable atmospheric application of GPS is probably the assimilation of water vapor content estimates into numerical weather prediction and climate models. In this study, we present the comparison of PW over the Tibetan Plateau determined by GPS, radiosonde, and analyses from the Global Assimilation and Prediction System (GASP) [Bourke et al., 1995] used at the Australian Bureau of Meteorology and analyses from the European Centre for Medium-Range Weather Forecasts (ECMWF) operational NWP system. A short period of measurements by WVR is also used for calibration. The Tibetan Plateau is the largest topographic feature on the earth, covering a large area of central and western Asia. The average surface elevation here is about 4000 m above sea level. Along with the Indian Ocean, Bay of Bengal and South China Sea, the Tibetan Plateau is a key area for the development of the Asian summer monsoon [Xu et al., 2003]. The water vapor amount transported from the southwest boundary of the Plateau to this region facilitates the development of convective clouds under strong solar heating at the surface. When these cloud systems move eastward they may trigger the onset of the Asian monsoon and sometimes cause severe floods in southeast China [Wang et al., 2003]. An accurate estimation of water vapor distribution and transport in this region is thus very important for numerical prediction of the Asian monsoon. However, because of its high topography and remote nature the spatial coverage of radiosonde measurements in this region is poor. Application of GPS measurements in this area may improve moisture measurements and NWP model forecasts.

[5] In addition to the PW comparison, we further investigate the effect of PW determined by different methods on the surface radiation budget and compare the results with observations. Since the air mass in this region is only about half of that at sea level and the aerosol load in the atmosphere is lower, the downward solar irradiance at the surface may be largely determined by water vapor. Thus this comparison may provide an indirect validation for the water vapor measurements as an accurate water vapor amount used in the calculations should lead to results comparable to the observations.

2. Water Vapor and Radiation Observations

[6] A long-term observational project for the measurement of water vapor on the Tibetan Plateau has been conducted as a Japanese-China collaboration since 1999. The GPS receivers were installed at Gaize (32.3°N, 84.06°E, 4420 m) in the western part of the plateau, Naqu (30.48°N, 92.06°E, 4518 m) in the eastern part of the plateau and Deqin (28.65°N, 99.17°E, 3594.9 m) at the southeast edge of the plateau. The GPS receiving stations consist of Dorne-Margolin antenna-plus-choke ring assemblies with a radome. The GPS receivers recorded the data from seven to eight satellites in view every 30 s. The measurements at Gaize and Naqu started on 1 January 2000 and measurements at Deqin started in April 2002. The instruments are installed at meteorological observational sites for the sake of utility of operational measurements. The operational radiosonde measurements are available at Naqu. A water vapor radiometer WVR-1100 developed by Radiometrics Corporation, Boulder, Colorado ( was used to measure PW at Naqu for calibration purpose. This instrument has 12 channels operated from 22 to 59 GHz. The water vapor can be measured by the frequency channels 22 to 30 GHz.

[7] Duan et al. [1996] have suggested that the relative accurate values of PW can be derived directly from GPS observations if GPS stations with separations greater than 500 km are included in the analysis. For this reason, we have included seven other ground-based receivers in the network. The locations of these receivers as shown in Figure 1 are about 500 to 4000 km away from the Tibetan Plateau. WVR measurements are also available at Xian in this network. We used the precise orbit parameters available from the International GPS Service. The constraint on station coordinates is 1 cm. The elevation cutoff angle used is 15°.

Figure 1.

Geographical location of GPS sites used in this study.

[8] An Automatic Weather Station (AWS) was installed at Gaize and started operation in October 1997 as a part of the Japanese Experiment on the Asian Monsoon [Wang et al., 2004]. Total downward and upward solar radiation (0.3–2.8 μm) and infrared thermal radiation (3–50 μm) are measured at 5 seconds s intervals by upward-looking and downward-looking high-precision pyranometers (MS-802, EKO Instruments Trading Co., Ltd., Japan) and a precision pyrgeometer (MS-202) mounted on a 1.5-m-high horizontal platform. The data sets were averaged to a 1-hour time step to reduce random error. All radiation sensors are calibrated and repaired every year. The data collected in year 2001 are used in this study.

3. Determination of PW From GPS Measurements

[9] Radio signals transmitted from GPS satellites are delayed by the atmosphere before they are received on the ground. The delay due to the water vapor component of the troposphere provides an opportunity for the retrieval of PW with ground-based GPS. Several software packages are available for the retrieval of PW using GPS measurements. We used GAMIT developed by the Massachusetts Institute of Technology [King and Bock, 2003] to analyze the GPS phase observations for the determination of the total delay by tropospheric ZT. The tropospheric delay can be separated into two components:

equation image

where Zh represents the zenith hydrostatic delay due to the induced dipole moment [Duan et al., 1996] and Zw denotes a zenith wet delay due to the permanent dipole moment of water vapor [Bevis et al., 1994]. The zenith hydrostatic delay can be computed using the Elgered et al. [1991] method,

equation image

where ps is the surface pressure; ϕ the latitude; and h the altitude in km. The wet delay Zw is then obtained from the difference between ZT and Zh. Once Zw is determined it can be transformed to PW using the following expression [Bevis et al., 1994],

equation image


equation image

ρ is the density of liquid water, Rν is the specific gas constant for water vapor, k3 and k2 are constants, and Tm is the weighted mean temperature of the atmosphere defined [Davis et al., 1985] as

equation image

where e represents water vapor pressure and T temperature. To obtain Tm one needs the vertical profiles of temperature and water vapor pressure which are sometimes not available at GPS locations. Bevis et al. [1994] developed a linear relationship between Tm and the surface temperature Ts to solve this problem,

equation image

The coefficients a and b for the United States are 70.2 and 0.72 as given by Bevis et al. [1994]. Liou and Teng [2001] obtained a = −31.5, b = 1.07 for Taipei. The values we derived from the measurements on the Tibetan Plateau are 44.05 and 0.81. It is clear that the Tm and Ts relationship is geographic location-dependent.

[10] It is important to have accurate and simultaneous surface pressure and temperature measurements at the GPS location for accurate estimates of PW using the GPS measurements. This condition is satisfied at Gaize as pressure and temperature were measured by an AWS. At Naqu and Deqin stations, however, these variables need to be estimated by linear time interpolation using the routine meteorological measurements. We have investigated the errors in the determination of PW due to the uncertainties induced by the interpolation of surface temperature. Assuming the ‘true’ surface temperature is 25°C, we introduced temperature errors from −10 to 10°C and calculated the corresponding PW errors. The results are shown in Figure 2. It is seen that the PW is not sensitive to the surface temperature. For a 5°C Ts error (relative error of 20%), the PW error is only 1.39%, i.e., 1.39 mm if the true value is 100 mm. For the worst case scenario, i.e., the absolute surface temperature error is 10°C (relative error of 40%), the corresponding relative PW error is only 2.79%, or 2.79 mm if the true value is 100 mm. It is unlikely that linear interpolation would cause such a large error. We also tested other different ‘true’ surface temperatures and obtained very similar results. We can conclude that the uncertainty due to the surface temperature does not have a significant influence on PW estimates.

Figure 2.

Sensitivity of precipitable water to the changes in the surface temperature.

4. NWP Analyses

[11] Analyses of water vapor from the Bureau GASP and ECMWF operational global NWP systems have been compared with the GPS estimates of PW. The GASP analyses is based on the Australian Bureau of Meteorology global spectral model with 29 vertical levels and a triangular truncation of 239 waves, equivalent to about 1.5 × 1.5 longitude and latitude degrees of horizontal resolution. The model field analysis is carried out every 6 hours. PW values at the GPS locations were determined by linear interpolation using the nearby model grid values and vertical integration above the GPS antenna elevation. The horizontal resolution of the ECMWF analysis data is 2.5 × 2.5 degrees.

5. Radiation Model

[12] The irradiance at the surface is calculated using a modified version of the Edwards and Slingo [1996] code. The longwave code is the same version as described by Sun and Rikus [1999]. The shortwave code is a new version with a very high spectral resolution developed recently. There are total of 210 spectral bands ranged between 100 and 50,000 cm−1, 102 in the UV-VIS and 108 in the near IR. The gaseous transmission is treated in terms of the correlated k distribution method. Six absorbing gases (H2O, CO2, O3, CH4, N2O and O2) are included. The oxygen collision induced continuum (O2-O2, O2-N2) is also included. The binary absorption cross-section data for O2-O2 and O2-N2 were obtained from the European Space Agency (ESA) [Smith and Newnham, 2000]. The Global Aerosol Data Set (GADS) compiled by Koepke et al. [1997] are incorporated into the code. The code has been extensively validated against the results of line by line and observations, and the results show excellent agreement. The detailed development of the code and validation results will be discussed in a separate paper.

6. Comparison of PW Between GPS, Radiosonde, WVR, and Model Analysis

[13] We first compare PW measurements determined by GPS and radiosondes operated at the national meteorological observational station at Naqu. The results at the radiosonde launch times (0000 and 1200 UTC) for the year 2001 are shown in Figure 3a. The GPS data for a period between 25 July and 24 August are missing. One can see that the water vapor amount at Naqu is very low. Even in the rainy season (JJA) PW is only about 20 mm, less than half of that in tropical regions [Liou and Teng, 2001]. The majority of GPS estimates of PW are higher than those from radiosonde measurements. The mean difference and RMS difference between the GPS and radiosonde are 1.27 mm and 1.75 mm, respectively. Although these values are compatible with those reported in the literature the relative difference is larger than those in the literature as PW itself is relatively small. In the winter season, for example, the relative difference in PW between the GPS and radiosonde can be 100% as seen in Figure 3.

Figure 3.

(left) Comparison of precipitable water in 2001 at Naqu: (a) results determined by the GPS and radiosonde at 0000 and 1200 UTC, (b) comparison between GPS and GASP analysis, and (c) comparison between GPS and ECMWF analysis. (right) Corresponding differences.

[14] The comparisons of GPS based PW with that determined by the GASP analysis at 1100/2300 UTC and the ECMWF analysis at 0000/1200 UTC are shown in Figures 3b and 3c, respectively. These results are very similar to those shown in Figure 3a, with the PW from the model analysis being smaller than the GPS estimates. This is expected as the model analysis is largely dependent on the available radiosonde observations. It should be mentioned that Naqu is the national standard station and the radiosonde data at this station are used in the NWP assimilation systems.

[15] Gaize and Deqin do not have radiosonde observations. Thus we compare only the GPS PW with the analysis and the results for Gaize are shown in Figure 4. It is seen that the PW determined by the two NWP model analysis is again systematically smaller than the GPS estimate. The difference is even larger than that found at Naqu. The mean PW difference between the GPS and GASP is 1.8 mm with an RMS of 2.2 mm and that between GPS and ECMWF is 1.85 mm with an RMS of 2.4 mm. The GPS-PW data at Deqin were obtained from 26 April 2002 to 22 August 2003. A scatterplot for this period is shown in Figure 5. There are several valleys in this region through which the water vapor is transported to the plateau. As a result, the water vapor amount in this region is higher than that in the other two stations. PW from the two model analyses are significantly less than the GPS estimates. The mean difference between GPS and GASP is 7.1 mm with an RMS difference of 7.7 mm and that between the GPS and ECMWF is 6.7 mm with an RMS of 7.3 mm. The lack of observations in this region may be responsible for the large discrepancies.

Figure 4.

(left) Comparison of precipitable water in 2001 at Gaize: (a) results determined by the GPS and GASP analysis and (b) comparison between GPS and ECMWF analysis. (right) Corresponding differences.

Figure 5.

Comparison of precipitable water at Deqin determined by the GPS and analysis of the GASP and ECMWF operational models.

[16] The above results clearly show that the difference between the PW from the GPS measurements and those from radiosonde and NWP model analyses is significant. A natural question that can be raised is whether there is an error in the GPS retrieval or in the radiosonde measurements or in both. In order to clarify this issue, we further compare the results at the plain area in this network and the results determined from the WVR measurements. As mentioned earlier, the short-period measurements using WVR are available at Naqu and Xian for calibration purposes. The measurements using WVR have been used as standards by many authors [e.g., Duan et al., 1996; Tregoning et al., 1998]. Duan et al. [1996] have suggested that if GPS networks analyzed for PW contain at least one station with a collocated WVR, the comparison between the WVR and GPS at that station could provide a basis for controlling the quality of the network solution.

[17] Figure 6 shows a comparison of PW estimated by GPS, WVR, and radiosonde launches for the days of 26 June to 3 July 2004 at Xian located in the middle of China with an elevation of 500 m. The lines in Figure 6 represent the least squares fit to the data. It is seen that at this low-elevation site the agreement between these three observations is reasonably good but it is clear that the agreement between the GPS-PW and WVR-PW (Figure 6a) is better than that between GPS and radiosonde (Figure 6b). The RMS difference between the GPS-PW and WVR-PW is 2.5 mm. Whereas that between the GPS and radiosonde is 4.8 mm. The other distinct difference between Figures 6a and 6b is that the GPS-PW is slightly less than WVR-PW as the cross section of the fit line with the y axis is negative (Figure 6a), while it is larger than radiosonde PW because the cross-section point is positive (Figure 6b). This indicates that the PW determined by the radiosonde measurements at the low-elevation location may be also biased toward low values. Figure 7 shows the same comparison at Naqu but for the days 1 to 10 in May 2004. Both the WVR and GPS PW are 30 minute mean values and the radiosonde data are for launch times of 0000 and 1200 UTC. It is seen that the PW determined by GPS measurement in this period is in a good agreement with that determined by the WVR. The results from the radiosonde are again significantly smaller. These comparisons provide clear evidence that the radiosonde measurements on the Tibetan Plateau may suffer from a serious dry bias. The radiosondes used at Naqu and Xian are both made in Nanjing, China. The model 701A was used at Naqu and 701C at Xian. These two models were designed in the early 70s with a similar specification. The accuracy of this instrument is highly questionable because of the age of its design and no cross comparison has been done before. It is therefore necessary to have the instrument carefully tested and find out the reasons for the bias and fix the problem with an appropriate correction factor as has been done for the Vaisala radiosonde by Wang et al. [2002]. This task, however, is beyond the scope of this study.

Figure 6.

Comparison of precipitable water determined by the (a) WVR and (b) radiosonde from 26 June to 3 July 2004 at Xian (after Mao J. and Bi Y., personal communication).

Figure 7.

Comparison of precipitable water determined by the WVR, GPS, and radiosonde from 1 to 11 May 2004 at Naqu.

[18] Note that the above results are similar to the findings reported by Wu et al. [2003] for the mountainous area of Sumatra Island and by Takagi et al. [2000] for Lhasa on the Tibetan Plateau. Wu et al. explained the difference being due to dry bias in the Vaisala radiosonde based on a series of laboratory tests reported by Wang et al. [2002]. They found that the Vaisala humidity sensors contained a chemical contamination error and a temperature dependence error. The chemical contamination error generally increases with the age of the radiosondes and relative humidity. The temperature-dependent error mainly occurs at temperature below −20°C, and increases substantially with decreasing temperature below −30°C. Takagi et al., however, attributed the differences between the GPS-PW and radiosonde PW to the low vertical resolution of the radiosonde data at Lhasa. They drew this conclusion based on the result of a comparison study in Japan using sounding data including significant levels, which shows negligible systematic difference between the two types of measurement. Their conclusion is questionable as they did not identify whether the difference at the two sites is due to the use of different radiosonde. The vertical resolution of sounding data may effect the accuracy of PW, but it is unlikely to cause such a large difference on the Tibetan Plateau. The vertical resolution of the radiosonde data at Xian is the same as that at Naqu (the values are all given at standard pressure levels), but the results have shown that the agreement between GPS-PW and radiosonde PW at Xian is better than that at Naqu. Our model results do not support the Takagi et al. explanation either. The ECMWF analysis provides data at the standard pressure levels and there are only seven vertical levels between the surface and 200 hPa at Naqu and Gaize. In contrast, there are 20 levels between the surface and 200 hPa in the GASP analysis and therefore the vertical resolution of GASP is much better than ECMWF data, but as seen from Figures 3 to 5, the PW determined by the GASP analysis is not better than that of the ECMWF. The results from these two data sets show a similar level of dry bias compared with the GPS PW. Although we do not know the exact reason for the dry bias of sounding data on the Tibetan Plateau, the similarity to the findings at other high-elevation areas may indicate that the problem is related to the cold temperature.

7. Variability of GPS PW

[19] The water vapor variation over the Tibetan Plateau is an important factor influencing convective activities. However, a study of water vapor variability is difficult without continuous water vapor measurements. The radiosonde measurements only allow determination of water vapor variation at large spatial scales and the daily temporal scale. The GPS measurements are performed at higher frequencies, making it possible to analyze the diurnal variation of water vapor.

[20] In order to explore the seasonal and diurnal variation of water vapor, we calculated monthly hour mean PW using the GPS measurements at the two sites and the results are shown in Figure 8. It is seen that there is a clear seasonal variations of PW at these two stations. Large PW values appear in the summer from June to August with small values in winter from December to February. This seasonal distribution reflects the typical monsoon characteristics on the Tibetan Plateau. It is a rainy season from June to August and therefore the mean PW in this period is generally higher than in the other seasons. There is also some diurnal variation but the magnitude of diurnal change is smaller than that of seasonal change. Figure 9 presents the monthly hourly mean PW anomaly (difference between hourly values and daily mean), which may give a clearer diurnal variation pattern. It is seen that in summer months the PW at Gaize is higher during the nighttime than during the day. This is also a typical monsoon characteristic on the Tibetan Plateau. The precipitation in the summer months usually occurs at night and is known as plateau night rain. The higher PW at night is obviously due to the convective activities. In the other seasons, especially in winter, however, the higher PW appears in the daytime between 1000 and 1600 LT. Since this is relative dry season, the high PW during daytime is mainly due to evapotranspiration associated with strong radiative heating. A similar pattern also occurs at Naqu as shown in Figure 9 (bottom).

Figure 8.

Seasonal and diurnal variation of precipitable water at Naqu and Gaize determined by the GPS measurements.

Figure 9.

Anomaly of precipitable water at Naqu and Gaize determined by the GPS measurements.

8. Impact on Radiation Budget at the Surface

[21] The previous comparisons have shown that both the PW determined by radiosonde and NWP model analysis systems on the Tibetan Plateau are significantly smaller than those determined by GPS and WVR. If we trust the PW from the GPS measurements then our study implies that the initial moisture fields on the Tibetan Plateau used in current NWP models are subject to a serious dry bias. It is clearly worthwhile to investigate how these biases could effect the simulations of water vapor cycle and energy balance using NWP and climate models. Before doing this study, we need to know whether the problem exists on the whole Tibetan Plateau and if so we need to work out a method to fix the problem. This is an ongoing study and will be discussed in a separate paper. Here we perform simple radiative transfer calculations to examine the impact of these water vapor differences on the surface radiation budget and compare the results with radiation measurements at the surface. As mentioned earlier, the atmosphere over the Tibetan Plateau is relatively clean with much less air pollution and aerosol loading in the atmosphere compared with other regions. Therefore the water vapor should play a dominant role on the surface radiation budget under cloud free conditions. It is expected that the modeled radiation using more accurate water vapor should be closer to the observations.

[22] The calculations were first performed for Gaize as the radiation observations from the AWS are available. The GASP analysis profiles at 1030 LT were used with the ozone profile scaled by the total column amount from TOMS and the surface albedo from the AWS measurements. The analysis water vapor profile is then scaled by the GPS PW to force the analysis PW to match the GPS values. The effect of cloud is not included in the calculations as cloud information is not available. The rural aerosol optical properties from GADS [Koepke et al., 1997] were included in the calculation as background. Figure 10a shows a plot of the time series of the total downward solar irradiance at the surface at 1030 LT for the year 2001. The observations show a large day-to-day variation due to the effect of clouds. The upper limit of the observational envelope can be regarded as irradiance under clear sky. The modeled results generally follow this envelope but not close enough. This implies that the model atmosphere is not optically thick enough, possibly due to uncertainties in water vapor and aerosols. The results from using the scaled PW are closer to the observations, indicating that the GASP moisture analysis in this region may be too dry. Figure 10a (right) shows the irradiance difference between the two modeled results. It is seen that with the scaling of water vapor amount the downward solar irradiance at the surface can be reduced by about 20 W m−2 (about 3% of total value), which is large and will in turn affect the surface energy balance.

Figure 10.

(left) Comparison of modeled and observed radiation in 2001 at Gaize. (a) Downward solar radiation at the surface and (b) downward longwave radiation at the surface. The observations at 1030 LT are plotted as solid curve. The corresponding modeled results are obtained by performing the calculations twice: once using the GASP analysis profiles and once with the GASP water vapor profile scaled by the GPS PW. (right) Difference between the two modeled results (flux without scaling the water vapor minus that with scaling).

[23] Figure 10b shows the comparison for the downward longwave irradiance at the surface. Again the large observed values are due to the presence of clouds. Of the two modeled results, the values determined with the water vapor scaled by the GPS PW are clearly closer to the observations. The differences in the longwave irradiance due to the PW differences in the calculations range up to 40 W m−2, as shown in Figure 10b (right).

[24] The same calculations are also performed for Naqu and the results are shown in Figure 11. It is seen that the shortwave irradiance at the surface determined using the GASP analysis profile is also about 20 W m−2 higher than that when the water vapor profile is scaled by the GPS PW and the reverse is true for the longwave. These results are generally the same as in Gaize.

Figure 11.

(left) Comparison of modeled shortwave (a) and longwave (b) radiation at Naqu. The solid curve represents the result determined using the GASP analysis profiles, and the dashed curve denotes the result using the GASP water vapor profile scaled by GPS PW. (right) Flux difference (flux without GPS scaling minus that with GPS scaling).

[25] Chen et al. [2003] have recently performed a sensitivity study using the NCAR regional climate model (RegCM2) to examine the effect of doubling CO2 on the surface warming on the Tibetan Plateau region. Their results show that doubling CO2 in this region can have a shortwave radiative forcing at the surface in the range from −10 to +10 W m−2, and a longwave forcing of about 5–15 W m−2, depending on the season and surface elevation. The changes in both shortwave and longwave radiation due to the uncertainty of water vapor found in this study are larger than the forcing due to the doubling of CO2. This means that this effect is important in this region and its influence on water vapor and heat transfer needs to be further investigated. It should be emphasized that the results of Chen et al. were obtained using a climate model. If an offline radiation code is run for the case of doubling CO2 the change in radiative forcing at the surface is smaller than these values. On the other hand, the results shown in Figures 10 and 11 were obtained at the relatively high solar elevation angle (local solar time 1030) and the results for the daily average will be smaller. Therefore it may not be appropriate to directly compare these two findings. A similar study to Chen et al. should be performed for water vapor using a regional climate model to appropriately assess the impact of water vapor uncertainty in this region. Nevertheless, the problem of the uncertain water vapor measurements in this region is considered to be significant and caution is needed when discussing model results associated with water vapor in this region.

[26] The reduced solar irradiance at the surface is actually due to radiation absorbed by the water vapor in the atmosphere and is therefore associated with a heating effect in the atmosphere. We further calculated the shortwave heating and longwave cooling differences caused by the difference in PW to explore this effect. Figure 12 shows the 1 year time series of heating/cooling differences calculated for Gaize. The solar zenith angle for the shortwave calculations corresponds to a local solar time of 1030. As expected, the solar heating is increased with increasing PW in the atmosphere so the difference in solar heating rate is positive and in a range between 0.1 and 1.0 K d−1. The increased solar heating is quite large and must have a large impact on convective activity in this region. It should be noted that Figure 11 is from instantaneous calculations. When averaged for the whole day, the values should be smaller. Figure 12 (middle) shows the longwave cooling difference. The negative values indicate that the increase in PW in the atmosphere has a further cooling effect. One can see that the cooling effect dominates in the longwave when the GASP water vapor profile is corrected by the GPS PW except in the boundary layers in summer months where it shows less cooling (positive difference). This is due to the vertical distribution of water vapor mixing ratio in the boundary layer. It happens if the water vapor in an upper layer is higher than for a layer below. The net heating difference is shown in Figure 12 (bottom). The positive difference is quite strong in the boundary layer in summer months, which obviously facilitates the development of convection during daytime. The positive heating difference is also seen in the middle and upper troposphere but not as large as that in the boundary.

Figure 12.

(top) Solar heating difference at Gaize between the calculation using the GASP analysis profile with and without correction by GPS PW. (middle) Longwave cooling difference and (bottom) net heating difference.

9. Conclusion

[27] The precipitable water over the Tibetan Plateau determined by GPS, WVR, radiosonde and NWP operational model analysis are compared in this study. It is found that the PW from the radiosonde measurements is systematically smaller than both the GPS and WVR estimates. Similarly, the PW determined by the GASP and ECMWF operational model analysis are also smaller than the GPS values. These systematic differences are likely due to the old type of radiosonde used in the Tibetan Plateau which may be subject to a dry bias at cold temperatures.

[28] The annual and diurnal variations of water vapor are analyzed using the GPS estimated PW. The annual variation of PW reflects the monsoon characteristics in this region with higher PW values occurring during June to August when the Plateau is in the summer monsoon season. There are also some diurnal variations but the amplitude of the diurnal cycle is small. In the summer months the PW is higher at nighttime due to frequent precipitation at night whereas in winter the PW is higher during daylight due to daytime evaporation.

[29] The effects of the uncertainty in water vapor on the surface radiation budget and the heating/cooling rate in the atmosphere in this region were investigated using the radiation model with GASP model analysis profiles. The results show that the downward solar radiation at the surface can be reduced by about 20 W m−2 and the downward longwave radiation at the surface can be increased by up to 40 W m−2 if the GASP water vapor profiles are corrected to have PW matching the GPS measurements. These changes are much larger than those due to the effect of doubling CO2 in this region.

[30] Many studies have shown that the water vapor amount in this region is an important factor influencing the Asian summer monsoon. The large uncertainty in water vapor found in this study is a clear concern. It is necessary to examine the accuracy of the radiosonde used in this region and install more GPS receivers to improve the moisture analysis. This is an ongoing study of this research project. We not only try to install more GPS receivers at the other sites on the Tibetan Plateau we also started to process the MODIS satellite water vapor data. This will help us to identify whether the dry bias exists in the whole Tibetan Plateau and if so we will work out the method to correct bias. We will then perform numerical experiments using a NWP or climate model to investigate the effect of the water vapor uncertainty on NWP forecasts and climate simulations in this and related regions.


[31] The authors would like to acknowledge W. P. Bourke for useful discussion and encouragement, L. J. Rikus and Tomasz for reviewing the manuscripts. Two anonymous reviewers made many valuable comments and suggestions that led to improving the paper quality. This study was supported by the National Natural Science Foundation of China under grant 40375035 and National 863 Project under grant 2002AA135360.