Journal of Geophysical Research: Atmospheres

Water vapor transport for summer precipitation over the Tibetan Plateau: Multidata set analysis

Authors

  • Lei Feng,

    1. LASG, Institute of Atmospheric Physics, Chinese Academy of Sciences, Beijing, China
    2. Public Meteorological Service Centre, China Meteorological Administration, Beijing, China
    Search for more papers by this author
  • Tianjun Zhou

    Corresponding author
    1. LASG, Institute of Atmospheric Physics, Chinese Academy of Sciences, Beijing, China
      Corresponding author: T. Zhou, LASG, Institute of Atmospheric Physics, Chinese Academy of Sciences, Beijing 100029, China. (zhoutj@lasg.iap.ac.cn)
    Search for more papers by this author

Corresponding author: T. Zhou, LASG, Institute of Atmospheric Physics, Chinese Academy of Sciences, Beijing 100029, China. (zhoutj@lasg.iap.ac.cn)

Abstract

[1] The atmospheric water vapor transport for summer precipitation over the southeastern Tibetan Plateau (hereafter TP) during 1979–2002 is examined by using five precipitation data sets and three reanalysis data sets. The multidata ensemble mean shows that under climate mean conditions, TP is a moisture sink in summer, having a net moisture convergence of 4 mm/day. The climatological water vapor transport from the southern boundary, which originates from the Indian Ocean and the Bay of Bengal, dominates the summer precipitation over the southeastern TP. It is estimated that the water vapor from the western boundary along the southern edge of the TP is about 32% of that from the southern boundary. The summer precipitation over the southeastern TP exhibits strong interannual variability, with a standard deviation of 1.3 mm/day, but no significant long-term trend. The water vapor transport for the interannual variability of summer rainfall over the southeastern TP mainly comes from the western boundary of the TP, which is originally from lower latitudes. An excessive rainfall anomaly of 1 mm/day over the southeastern TP is associated with an anomalous water vapor input of 138 (104) kg/m/s from the western (southern) boundary. It is worth noting that the quantitative analysis in this study is determined by the setting of the domain. The interannual variability of summer precipitation over the southeastern TP is dominated by an anomalous anticyclone over the northern Indian subcontinent and the Bay of Bengal, which intensifies the water vapor transport along the southern edge of the TP and leads to more water vapor convergence over the southeastern TP, thus the excessive rainfall in the area.

1. Introduction

[2] The Tibetan Plateau (TP) impacts global weather and climate through both dynamical and thermal effects [Wu et al., 2007]. It also plays a pronounced role in the regional water cycle. For example, many South and East Asian rivers originate from the TP, including several of the world's major river systems, i.e., the Indus, the Ganga-Brahmaputra, the Yangtze, and the Yellow River. The TP is usually called the “water tower of Asia” owing to its importance in the hydrological cycle [Xu et al., 2008]. Since a huge amount of water flows away from the TP, a supply of water vapor from surrounding regions is needed to maintain the regional water balance. Thus, the TP is one of the most active centers of hydrological cycle in the world, and understanding the processes of the hydrological cycle over the TP is a subject of crucial importance.

[3] Given the importance of the hydrological cycle over the TP, much effort has been devoted to this issue. A distinct diurnal cycle of precipitation has been found over the TP [Barros et al., 2004; Bhatt and Nakamura, 2005; Zhou et al., 2008a]. Generally, the precipitation activity over the hilly region is strongest during late afternoon, while valleys and lakes show dominant late-evening peaks, and a secondary morning rainfall peak is distinctly evident over large lakes [Singh and Nakamura, 2009]. The diurnal variation of circulation is revealed based on satellite data [Bai et al., 2008; Liu et al., 2009]. A strong daytime wind speed accompanied by increasing relative humidity prevails along deep valleys in the Himalayas [Ueno et al., 2008]. At night, the southerly wind continues but with weaker intensity, and water vapor stagnates in front of the Himalayas. Therefore, nocturnal precipitation is evident at the foot and in the valley of the Himalayas [Barros and Lang, 2003]. Chow and Chan [2009] proposed a possible mechanism for the diurnal variation of precipitation over the TP associated with the TP's anomalous heating.

[4] Intraseasonal variability of precipitation accompanied by convective activity is also evident over the TP. Yamada and Uyeda [2006] have pointed out the difference in the synoptic condition, which is dominated by the Tibetan High or the passage of synoptic disturbance. The passing of the synoptic trough is expected to contribute strongly to water vapor transport from the Indian Ocean to the TP during the monsoon season [Sugimoto et al., 2008]. The heavy precipitation during the 1993 monsoon was caused by moisture intrusion through the synoptic trough with meandering westerlies [Ueno, 1998]. On a sub-seasonal timescale,Chang [1981] has found an inverse correlation of precipitation amount in central India and that in the central TP, with a timescale of about 15 and 40 days. The plateau's summer monsoon break corresponds with an active phase of the Indian summer monsoon. The cyclonic circulation over India prevents water vapor intrusion into the TP [Sugimoto et al., 2008].

[5] In addition to the analysis of the diurnal and intraseasonal variability of the water cycle, many studies have focused on the sources of water vapor transport to the TP under climate mean states. However, the results remain inconclusive. The sparse availability of accurate instrumental data may be one reason for this. Some studies emphasized the southern channel, through which the water vapor originating from the Bay of Bengal flows to the southeastern TP after passing by three big valleys, i.e., Brahmaputra, Nujiang, and Jinshajiang [Xu et al., 1996, 2002]. The water vapor transport through the Brahmaputra channel is of central importance to the rainfall over the southeastern TP [Yang et al., 1987]. Other studies, however, stressed the importance of the water vapor transport from the northern Indian subcontinent, which passes by the Middle Himalaya through some passages and finally arrives at the western TP [Gao et al., 1985; Xu et al., 1996, 2002]. The western channel has been demonstrated by the trajectory of satellite clouds, which moves eastward from the western TP [Yang et al., 1992; Xu et al., 1996]. Which channel is more important? Unfortunately, there is no consensus up to now except for an ambiguous compromise that “both are important” [Lin and Wu, 1990; Xu et al., 2002].

[6] The uncertainties in previous studies of the hydrological cycle over the TP are mainly due to the limitation of observational data. Thanks to the availability of more and more satellite products, field observations, and reanalysis data in recent decades, it is now possible to reexamine the fundamental hydrological process of the TP using multi data sets, such as the sources of water vapor transport under climate mean states and interannual variability. This is the main motivation of this study. In addition, due to global warming, the surface air temperature over the TP has increased by about 1.8°C during the past 50 years [Wang et al., 2008]. Thus, one issue of both social and scientific interest is the potential changes in the hydrological process over the TP. We attempt to answer the question by examining the changes in the hydrological cycle over the TP at both long-term and interannual time scales.

[7] The remainder of the paper is organized as follows. Section 2 describes the data set and analysis procedures. The climatological summer precipitation characteristics over the TP and the corresponding water vapor transport are analyzed in Section 3. The variation of summer precipitation over the TP and the anomalous water vapor transport are presented in Section 4. A summary of the results, along with a discussion, are provided in Section 5.

2. Data and Method Description

2.1. Data

[8] The following data sets are used in our analysis:

[9] 1. The monthly precipitation data set at 97 stations, with no missing data, was provided by the National Meteorological Information Center and is used to examine the precipitation characteristics over the TP. The data cover the period 1961–2005. This precipitation data set has been homogenized and thus is reliable. As shown in Figure 1a, most of the stations are located over the eastern TP. There are only a few stations over the western TP, where the altitude is high. These data have been widely used in research on climate change over China [Li et al., 2010].

Figure 1.

(a) Terrain height (shaded) and 97 station locations (black dots) over the Tibetan Plateau; (b–i) Spatial distributions of climate mean summer (June–July–August) precipitation during the period 1979–2002 (unit: mm/day).

[10] 2. Two satellite-merged precipitation data sets are compared with the station data. The first is the Climate Prediction Center (CPC) Merged Analysis of Precipitation (CMAP) data set. These data merge rain gauge data over land; satellite IR, OLR, MSU, and SSM/I estimates over ocean; and model data (mostly over polar regions) together, with oceanic rainfall calibrated by rain gauge data from coastal and atoll stations [Xie and Arkin, 1997]. The second is the Global Precipitation Climatology Project (GPCP) data set. The IR estimates were calibrated by microwave estimates and then adjusted by rain gauge data [Adler et al., 2003]. The resolution of the two data sets is 2.5° × 2.5°, and the time period covered is 1979–2002. Both CMAP and GPCP data have been used in global monsoon studies [Zhou et al., 2008b].

[11] 3. Two high-resolution gridded precipitation data sets over Asia are also used in this study. The first was developed byXie et al. [2007]. It covers the period 1962–2002 and uses a resolution of 0.5° × 0.5°. It was constructed using gauge observations at over 2200 stations, collected from several independent sources. The precipitation field has been adjusted using Parameter-elevation Regressions on Independent Slopes Model (PRISM) monthly precipitation climatology to correct the bias caused by orographic effects. The second was created by collecting rain gauge observations across Asia through the activities of the Asian Precipitation—Highly Resolved Observational Data Integration Toward the Evaluation of Water Resources (APHRODITE) project [Yatagai et al., 2009]. The data cover the period 1961–2004 and has a high resolution of 0.25° × 0.25° (hereinafter APHRO data). Up to now, it is considered as the only long-term (1961 onward), continental-scale daily product that contains a dense network of daily rain gauge data for Asia, including the Himalayas and mountainous areas in the Middle East.

[12] 4. The National Centers for Environmental Prediction/National Center for Atmospheric Research (NCEP/NCAR) reanalysis data during the period 1979–2002 are used to estimate the water vapor transport over and around the TP [Kalnay et al., 1996]. The quality of the data in estimating the water vapor cycle, including moisture transport, has been evaluated [Trenberth and Guillemot, 1995], in particular over the Asian monsoon area [Zhou et al., 1999; Zhou and Yu, 2005]. Another two reanalysis data sets, the European Centre for Medium-Range Weather Forecasts 40-year reanalysis (ERA40) [Uppala et al., 2005] and the Japanese 25-year reanalysis (JRA25) [Onogi et al., 2007], are also used to assess the results obtained from the NCEP/NCAR reanalysis. Both the NCEP/NCAR and the ERA40 reanalysis data are available on a 2.5° × 2.5° grid. The horizontal resolution of the JRA25 is 1.25° × 1.25°. Table 1 presents detailed information on the three reanalyses, i.e., the data sources. In our intercomparison, we use the common time period 1979–2002.

Table 1. Detailed Information on Three Reanalysis Data Sets
 NCEP/NCARERA40JRA25
Time period1948-present1957.9–2002.81979–2004
Horizontal resolutionT62/∼209 kmTL159/∼125 kmT106/∼125 km
Vertical layer286040
Satellite dataretrievalsradiancesradiances
Upper-air dataUpper-air wind, temperature, and humidity from radiosondes; wind and temperature from dropsondes; wind from pilot balloonsUpper-air wind, temperature, and humidity from radiosondes; wind and temperature from dropsondes; wind from pilot balloons and wind profilersSame as ERA40
Surface dataStation observations; surface pressure and temperature from buoy and ship reports; ocean data: pressure, temperature, humidity, and wind.Station observations; surface pressure and temperature from buoy and ship reports; land data: pressure, temperature, humidity, and snow cover; ocean data: pressure, temperature, humidity, and speed.Same as ERA40
Experiment dataPAOBs, FGGE, TOGA, COARE, and so onPAOBs, FGGE, TOGA COARE, ALPEX, GATE, and so on; O3 from TOM and SBUVChinese snow data, GEWEX GAME data, wind from TCR
Boundary fieldsReynolds SST, sea ice from ECMWF, snow coverage from NESDISHADO SST, NOAA/NCEP 2DVAR SST, and sea iceCOBE SST and sea ice

[13] In order to examine the extent to which the reanalysis precipitation captures the information from the station and the gridded precipitation data sets, we also make an intercomparison among the precipitation data sets, including reanalysis products.

[14] Since the data sets are available at different horizontal resolutions, according to Xie et al. [2007], the three reanalysis data sets are re-gridded on a 2.5° × 2.5° mesh by using the weighted average interpolation method.

2.2. Analysis Method

[15] The total water vapor flux can be separated into two parts: the stationary and the transient components [Trenberth, 1991]. We have calculated the stationary and transient water vapor transport using the NCEP daily reanalysis data set and found that the latter is about one magnitude weaker than the former (figure omitted). Analysis using the ERA40 daily reanalysis showed the same results [Bothe et al., 2010, Figures 1d and 1e]. A previous quantitative comparison of Zhou et al. [1999] also found that the transient water vapor transport contributes little to the total flux for East Asia. Therefore, in our analysis we mainly focus on the stationary component, which is calculated using the monthly mean data as Zhou and Yu [2005] and Bothe et al. [2010]. According to Trenberth [1991], the vertically integrated water vapor flux over and around the TP is calculated as follows:

display math
display math

where q is the specific humidity, u is the zonal wind, v is the meridional wind, and g is the acceleration due to gravity. The pressure of top layer pt is equal to 300 hPa in our calculation as there is negligible water vapor above 300 hPa, and the specific humidity in the NCEP/NCAR reanalysis was set to zero [Xu et al., 2008; Zhou, 2003]. The water vapor budget analysis over the complex terrain is sensitive to treatment of topography [Rasmusson, 1968; Trenberth, 1991]. In our analysis, we first interpolate the variables from three reanalyses into the same pressure levels with an interval of 50 hPa and then integrate the water vapor from surface pressure ps to 300 hPa.

[16] The divergence of water vapor flux can also be divided into two parts [Huang et al., 1998]: the moisture advection term and the wind divergence term:

display math

where inline image is the horizontal wind vector, and q is the specific humidity. The first term on the right is the moisture advection. When the wind flows from areas with higher (lower) specific humidity to areas with lower (higher) specific humidity, this term is negative (positive), which is called wet (dry) advection. It has a positive (negative) contribution to the total water vapor convergence (divergence). The second term on the right represents the contribution of wind divergence. The divergence in the wind field corresponds to the divergence in the water vapor transport field.

3. Climate Mean Water Vapor Transport

3.1. Precipitation and the Corresponding Water Vapor Transport

[17] This study focuses on areas (26°–42°N, 75°–105°E) with an altitude above 3,000 m (hereafter the TP region), as illustrated in Figure 1. The precipitation from the station data is used as the reference precipitation. To make the data comparable, four gridded and three reanalysis precipitation data sets are interpolated into 97 stations over the TP (black dots in Figure 1a) by bilinear interpolation. The annual cycles of precipitation (Figure 2) averaged over the 97 stations all exhibit their peaks in July. The precipitation during June–July–August accounts for more than 50% of the annual rainfall. There is only a slight difference in the percentage of precipitation in June and August among the station and the four gridded data sets (Figures 2a–2e). The three reanalysis products generally capture the annual cycle of precipitation over the TP (Figures 2f–2h) except that the percentage of spring (summer) precipitation is more (less) than that in the other five data sets. The precipitation from the reanalysis also peaks in June or July, but the peaks are not as evident as those derived from the station data. Based on the annual cycle shown in Figure 2, it is reasonable to focus on summer precipitation (June–July–August) over the TP.

Figure 2.

Annual cycle of precipitation averaged for all 97 stations over the TP from eight precipitation data sets during the period 1979–2002 (unit: %).

[18] The spatial patterns of summer rainfall over the TP from different data sets all show a strong southeast-northwest gradient (Figures 1b–1i). The rainfall decreases from the southeastern (with rainfall amount above 5 mm/day) to the northwestern (with rainfall amount below 1 mm/day) TP. The results derived from the Xie (Figure 1e) and APHRO (Figure 1f) data sets resemble the spatial pattern derived from the station data (Figure 1b). The similarity is expected since more station rainfall data were merged into these two data sets. The other precipitation data sets show a lot of uniform spatial patterns, which may be due to their low horizontal resolutions (Figures 1c, 1d, 1g, and 1i). Three reanalysis products overestimate the precipitation amount over the TP, with the largest precipitation amount being more than 10 mm/day (Figures 1g and 1i). A large discrepancy is seen among the three data sets in the rainfall center, which is located in the Sichuan basin (22°–28°N, 98°–103°E) in both the NCEP/NCAR and the JRA25 reanalyses but in the southern TP (24°–28°N, 88°–98°E) in the ERA40.

[19] The ensemble mean climatological water vapor transport over and around the TP derived from the three reanalysis data sets is shown in Figure 3d. A box covering 26°–40°N, 80°–102°E (shown in Figure 3a) is used to perform budget analysis on the water vapor transport across the four boundaries. There are two main water vapor channels for the climatological summer precipitation over the TP. The first is the transport by the Indian summer monsoon, which has been documented in previous studies [Zhuo et al., 2002; Xu et al., 2002, 2008]. The strong southwesterly water flow associated with the Somali jet brings moisture from the Arabian Sea and the Bay of Bengal into the southeastern TP. The importance of the Somali jet to the Asian climate has been stressed in many studies [Huang et al., 1998; Zhang, 2001; Zhou and Li, 2002; Wang and Xue, 2003; Zhou and Yu, 2005]. Water vapor from the southern boundary enters into the TP through Grand Canyon areas [Sugimoto et al., 2008]. There are several meridionally oriented valleys over the southeastern TP, such as the Jinsha River, Nujiang River, and Lancang River valleys, which facilitate the transport of water vapor into the TP [Xu et al., 2008].

Figure 3.

Difference in climatological summer water vapor transport between (a) NCEP/NCAR, (b) ERA40, and (c) JRA25, and the ensemble mean from three reanalysis data sets for the period 1979–2002; (d) Ensemble mean climatological summer water vapor transport (unit: kg/m/s). The corresponding black solid isolines represent the terrain height of 3,000 m (as in Figures 4, 11, and 12); the box (26°–40°N, 80°–102°E) in Figure 3a is used to examine the water vapor transport budget quantitatively.

[20] The second channel is the transport by the midlatitude westerly. The westerly in summer is located to the north of the TP. It branches into two when it encounters the barrier of the western TP. Water vapor carried by the southern branch of the westerly, goes straight southward, and then turns to the east at about 28°N. Combined with the southwesterly from the Indian Ocean, this branch acts as the western channel of water vapor transport to the TP. The west-east direction of the Brahmaputra valley over the southern TP facilitates the water vapor transport eastward [Gao et al., 1985].

[21] In addition to the above two main channels, water vapor transport from the northern boundary is also evident, but its amplitude is very weak and nearly negligible. This channel only has an impact on summer precipitation over the northern TP. The main water vapor output is from the eastern boundary. The ensemble mean result (Figure 3d) bears a close similarity to that of previous studies based purely on the NCEP/NCAR reanalysis [Xu et al., 2003].

[22] Compared with the ensemble mean results, the NCEP/NCAR (ERA40) data show much weaker (stronger) southwest water vapor transport from the Indian Ocean and western water vapor transport from middle latitudes (Figures 3a and 3b). The JRA25 result shows the largest similarity with the ensemble mean, albeit with stronger southern water vapor transport from the Indian Ocean (Figure 3c).

[23] Water vapor convergence is an important precondition of monsoon precipitation. The relative contribution of the moisture advection term and the wind divergence term to water vapor convergence differs between over East Asia and over Indian monsoon domains; i.e., the water vapor convergence over East Asia is mainly due to the water vapor advection caused by the monsoon flow, while the water vapor convergence over Indian monsoon domains is mainly caused by wind convergence [Huang et al., 1998]. The TP is located at the western (northern) periphery of the East Asian (Indian) monsoon. The contribution of moisture advection and wind divergence to the TP rainfall is unknown. Figure 4 shows the total water vapor convergence (Figure 4a), the moisture advection term (Figure 4b), and the wind divergence term (Figure 4c) over and around the TP, derived from the ensemble mean of three reanalysis data sets. It is evident that water vapor converges over most parts of the TP, especially over the southern and eastern TP (Figure 4a). Hence, the TP is a sink of the water vapor budget in summer. In the surrounding areas, water vapor diverges over the Indian Ocean and upstream of the TP and converges over the Bay of Bengal and East Asian monsoon domain. The pattern of the moisture advection term over the TP resembles that of the total water vapor convergence (cf. Figures 4a and 4b). The convergence center of the advection term is located over the southern edge of the TP (Figure 4b). The wind divergence term converges over the central TP but diverges along the TP (Figure 4c). Therefore, water vapor convergence along the TP is mainly caused by the moisture advection term, while water vapor convergence within the central TP is caused by both the moisture advection term and the wind convergence term.

Figure 4.

Ensemble mean divergence of climatological water vapor transport over and around the Tibetan Plateau during 1979–2002 from three reanalyses; (a) total divergence, (b) moisture advection term, and (c) wind divergence term (units: mm/day). The solid (dashed) lines indicate areas where the value is positive (negative); (d) different components of the area-averaged (26°–40°N, 80°–102°E) water vapor divergence derived from three reanalyses and the ensemble mean.

[24] We further compare the water vapor convergence averaged over the TP (26°–40°N, 80°–102°E), derived from the three different reanalysis data sets shown in Figure 4d. The magnitude of total water vapor convergence derived from the NCEP/NCAR reanalysis is nearly equal to that derived from the ERA40 but is about 20% weaker than that from the JRA25. It is highly consistent among the three reanalysis data sets that the total water vapor convergence over the TP is dominated by the moisture advection term. The main difference among the three data sets is the effect of the wind divergence term, which is negative in the NCEP/NCAR reanalysis but positive in both the ERA40 and the JRA25. This difference suggests that previous results solely based on the NCEP/NCAR reanalysis, such as Huang et al. [1998], should be carefully re-examined. What causes the difference among the three reanalysis data sets deserves further study. Previous studies found that the quality of the NCEP/NCAR reanalysis data over the East Asian domain was questionable [Inoue and Matsumoto, 2004; Wu et al., 2005]. The total water vapor transport over the East Asian summer monsoon domain derived from the NCEP/NCAR reanalysis is stronger in magnitude than that from the ERA40 [Zhou and Yu, 2005].

3.2. Vertical Structure of Water Vapor Transport

[25] A quantitative measure of water vapor transport across the four boundaries of the TP is shown in Figure 5. The lower right panel is the reference map. Note that the coverage of the box used to estimate water vapor transport is shown in Figure 3a. To reveal the vertical distribution of water vapor transport, we divide the whole air column into lower (1000–700 hPa), middle (700–400 hPa), and upper (400–300 hPa) layers. The common features of the water vapor transport derived from the three reanalysis data sets for the period 1979–2002 are as follows:

Figure 5.

Vertical distribution of climatological water vapor transport (unit: kg/m/s) in the three layers across the four boundaries of the Tibetan Plateau (1,000 hPa-700 hPa, 700 hPa-400 hPa, and 400 hPa-300 hPa) during the period 1979–2002; (a) NCEP/NCAR, (b) ERA40, and (c) JRA25. (d) The reference map. The blue (red) arrows represent the vertically integrated water vapor input (output) across the four boundaries.

[26] First, the main water vapor inputs to the TP are from the southern and western boundaries, and the main water vapor output is from the eastern boundary. The total water vapor transport from the western boundary is 47% (28%, 26%) of that from the southern boundary in the NCEP/NCAR (ERA40, JRA25) reanalysis. Previous estimation based on station data indicated that the water vapor transport from the western boundary is about 40% of that from the southern boundary [Gao et al., 1985], although the domain setting used then is slightly different from this study.

[27] Second, the water vapor transport from the southern boundary is concentrated in the lower layer and decreases from the surface to the upper levels. As the height increases up to 400 hPa, the northward water vapor transport turns back to the south.

[28] Third, the water vapor transport across the eastern boundary is complicated. Under 700 hPa, the westward water vapor transport has a weak magnitude. In the middle layer, the eastward transport of water vapor is far stronger in magnitude than that in the lower layer. The water vapor transport in the upper layer is weak due to low humidity.

[29] Fourthly, the water vapor transport from the northern boundary also decreases from the surface to the upper layer. Because it comes from drier high latitudes, the southward transport is smaller in magnitude than that from the other three boundaries.

[30] The main difference among three data sets is that the magnitude of the water vapor transport from the southern and eastern boundaries in the JRA25 reanalysis is about 50% larger than that in the NCEP/NCAR and ERA40 reanalyses. This may be due to the fact that the specific humidity in northern India in the JRA25 is much larger than that in the other reanalyses [Sugimoto et al., 2008]. A difference is also evident in water vapor transport from the western boundary. In the NCEP/NCAR reanalysis, the transport across the western boundary decreases from the surface to the upper levels, but in both the ERA40 and the JRA25 reanalyses, the strongest transport is evident in the middle layer.

[31] The average altitude of the TP is about 4,500 m, especially in the Himalayan system over the southern TP, which extends from Kashmir in the west to Assam in the east, with more than 100 mountains higher than 7,200 m. Can water vapor from the western and southern boundaries climb up the TP? To answer this question, the multidata set ensemble mean vertical cross section of the zonal water vapor transport and wind field averaged between 28°–35°N is shown in Figure 6a. The water vapor transport decreases to about 5 kg/m/s when it arrives at 85°E and then increases to more than 25 kg/m/s when it reaches 95°E. There are two zonal water vapor transport centers. One is upstream of the TP at around 80°E, and the other is in the eastern TP around 95°E. The prevailing west wind facilitates the eastward transport of water vapor.

Figure 6.

(a) Longitude-pressure (hPa) cross section of the climatological zonal water vapor transport (shaded) and wind field (vectors) composed of zonal wind and vertical velocity (enlarged by 100 times) averaged over 28°–35°N; (b) latitude-pressure (hPa) cross section of the climatological meridional water vapor transport (shaded) and wind field (vectors) composed of meridional wind and vertical velocity (enlarged by 100 times) averaged over 80°–95°E; and (c) same as Figure 6b but averaged over 95°–102°E. The data represent the ensemble mean from three reanalyses during 1979–2002. Positive values mean the direction is to the east in Figure 6a or to the north in Figures 6b and 6c; the histogram shows the in situ vertical profile of topography; dark (light) shading indicates areas where the value is greater (less) than 5 (−5) kg/m/s.

[32] As the terrain height in the western TP is higher than that in the eastern TP, the meridional water vapor transport is averaged within 80°–95°E and 95°–102°E, respectively. As shown in Figure 6b, in the western part, the water vapor transport from the southern boundary decreases rapidly with height. The northward water vapor transport decreases to nearly zero at 30°N. In the eastern part (Figure 6c), the northward water vapor transport penetrates into 35°N due to the existence of many meridionally orientated big valleys. The water vapor transport reaches 32°N with a magnitude of 15 kg/m/s. The wind field coincides well with the water vapor transport. The strong sensible heating and topography-forced upward motion over the TP forms an anti-Hadley circulation over the TP longitudes in summer, which is usually called the monsoon meridional cell, and the northern edge of the upward motion is located at 35°N [Ye and Gao, 1979; Zhou and Li, 2002; Chen et al., 2010]. The upward motion prevents the water vapor from moving further northward. Therefore, the water vapor from the southern boundary enters the TP mainly through channels east to 95°E below the level of 500 hPa. For the areas west to 95°E, the water vapor transport mainly affects the summer precipitation to the south of 30°N.

4. Variation of Water Vapor Transport

4.1. Variation of Summer Precipitation

[33] The summer precipitation time series averaged over the 97 stations are shown in Figure 7. The variabilities of precipitation derived from the five data sets match well with each other, although the amplitudes of precipitation change derived from four gridded data sets are larger than that from station data. A discrepancy is evident between the three reanalysis products and the other data sets.

Figure 7.

The summer precipitation time series averaged over 97 stations and normalized by its standard deviation.

[34] The detailed statistics for each precipitation time series are listed in Table 2. The CMAP and the APHRO are much closer to the station data in terms of both climate mean value (mean) and root-mean square error (RMSE). The reanalysis products overestimate the precipitation amount over the TP by more than 40%, especially for the NCEP/NCAR data, with the monthly mean precipitation about twice that from station data. The variability of precipitation in the Xie data set is highly consistent with that from station data, having a correlation coefficient of 0.99. The correlations of the three reanalysis products with the station data are lower than those in the other four gridded data sets, and the JRA25 data show a better performance compared with the other two reanalyses. The linear trend is not statistically significant at the 5% level based on the nonparametric Mann-Kendall test in all precipitation time series. Under the TP warming, the change in precipitation is complicated and does not simply exhibit a linear trend. This is consistent with previous studies based on station data over different sites of the TP [Lin and Zhao, 1996; Du and Ma, 2004; Li and Kang, 2006; Wu et al., 2007; You et al., 2008]. In the following analysis, we focus on the interannual variability of summer precipitation over the TP.

Table 2. The Mean Value, RMSE, Correlation Coefficient, Linear Trend, and Standard Deviation for the Summer Precipitation Time Series Over the TP Derived From Eight Precipitation Data Setsa
 StationCMAPGPCPXieAPHRONCEPERA40JRA25
  • a

    The gridded precipitation data sets (CMAP, GPCP, Xie, APHRO, NCEP/NCAR, ERA40, and JRA25) were interpolated into the 97 stations using bilinear interpolation for inter-comparison (units: mm/month for the mean and RMSE, mm/month/year for the linear trend and standard deviation).

Mean86.6483.2495.4098.2681.95153.14122.57121.29
RMSE__6.7511.7911.666.5292.89103.3861.33
Cor__0.750.720.990.850.560.520.66
Trend0.090.18−0.070.12−0.16−1.261.310.99
Std7.598.9511.638.378.8217.7116.4612.68

[35] The summer precipitation over the TP has larger spatial variations than the temperature over the TP. Therefore, it is reasonable to perform regionalization. We performed a REOF (Rotated Empirical Orthogonal Function) analysis on the summer precipitation from the 97 stations. Figure 8 shows the two leading modes and the corresponding PCs. The first PC shows strong interannual variability, with the center located over the southeastern TP (Figures 8a and 8b). The northeastern TP exhibits completely different temporal variability with the southeastern TP (Figures 8c and 8d). Therefore, to avoid mixing and missing many different climate signals, we chose to study the southeastern TP, which has the largest precipitation amount.

Figure 8.

(a and c) The first two REOF modes and (b and d) the corresponding time series for summer precipitation over the Tibetan plateau based on station data.

[36] The interannual variability of summer rainfall over the TP exhibits larger regional differences, as shown in the standard deviation (hereafter STD) of JJA mean rainfall (Figure 9). The STD values from the three reanalysis data sets are much larger than those from the station and the other four gridded precipitation data. All data sets exhibit a southeast-northwest decreasing gradient (Figures 9a–9h). The strongest variability is evident over the southeastern TP, with the STD value larger than 1 mm/day. Thus, we further define the TP JJA rainfall index as the regional average rainfall within 28°–34°N, 90°–102°E. The normalized time series of the index based on the station data is shown in Figure 9i. Based on the threshold of one STD, the data can be classified into wet and dry years. The wet years include 1982, 1985, 1991, and 1998, and the dry years include 1986, 1988, 1994, and 2001. We further compare the water vapor transport between wet and dry years in the next section.

Figure 9.

(a–h) The spatial distribution of the standard deviation for summer precipitation over the Tibetan Plateau from eight precipitation data sets (unit: mm/day); (i) the normalized time series of summer precipitation over the southeastern TP (28°–34°N, 90°–102°E) during 1979–2002 based on station data.

4.2. Water Vapor Transport for Wet and Dry Years

[37] The water vapor transport anomalies during wet/dry years are shown in Figure 10. The composites are performed with reanalysis data sets according to the precipitation index (Figure 9i) based on the station data. In wet (dry) years, the water vapor input from the western and southern boundaries increases (decreases), mostly in the lower and middle layers. Although the water vapor output from the eastern boundary in the middle and upper layers also increases (decreases), the total water vapor budget over the TP shows an anomalous convergence (divergence). For the regional average of JJA rainfall over the southeastern TP, if the difference in water vapor transport between wet and dry years from each boundary is divided by the difference in precipitation, we can obtain that an excessive rainfall anomaly of 1 mm/day is associated with an anomalous water vapor input of about 138 (104) kg/m/s from the western (southern) boundary. Compared with the magnitude of the climate mean transport, the relative change in water vapor transport from the southern boundary is smaller than that from the western boundary, especially in dry years.

Figure 10.

Vertical distribution of the ensemble mean water vapor transport change (unit: kg/m/s) in the three layers across the four boundaries of the Tibetan Plateau (1,000 hPa-700 hPa, 700 hPa-400 hPa, and 400 hPa-300 hPa) during the period 1979–2002: (a) wet years and (b) dry years (the size and position of the cube is the same as inFigure 5). Anomalies are defined as the difference between the water vapor transport in wet/dry years and the climate mean water vapor transport.

[38] To examine the consistency of the different reanalysis data sets in measuring the water vapor budget, the change in vertically integrated water vapor transport for individual reanalyses and the ensemble mean are shown in Table 3. The main results are consistent with that in Figure 10. Both the water vapor input from the western and southern boundaries and the water vapor output from the eastern boundary increase (decrease) in wet (dry) years. The change in water vapor transport from the western and eastern boundaries exhibits the largest consistency among the three reanalysis data sets, although the magnitude of change in water vapor transport from the western boundary derived from the JRA25 is about twice larger than that from the NCEP/NCAR and ERA40 reanalyses. The change in water vapor transport from the southern boundary is more consistent in wet years and shows uncertainty in dry years.

Table 3. Vertically Integrated Water Vapor Transport Anomalies From the Four Boundaries for TP Wet/Dry Years, Derived From Three Reanalyses and the Ensemble Mean (Unit: kg/m/s)a
 Wet YearDry Year
NCEPERA40JRA25ENSNCEPERA40JRA25ENS
  • a

    “ENS” refers to the ensemble mean from three reanalysis data sets. Anomalies are defined as the difference between the water vapor transport in wet/dry years and the climate mean water vapor transport.

west42.6435.00115.6364.42−26.09−45.75−114.94−62.26
east37.6313.3926.1725.73−40.43−23.95−68.98−44.46
south93.354.5344.7647.54−25.5414.00−44.31−18.62
north0.24−5.071.72−1.041.13−7.13−3.06−3.02

[39] A large spread is seen from the northern boundary. For example, the ERA40 data show a large southward anomaly in wet years, but both the NCEP/NCAR and the JRA25 reanalysis show weaker northward anomalies; the transport anomaly in dry years in the NCEP/NCAR is also different from that in the ERA40 and the JRA25 reanalysis. However, this discrepancy does not have significant impact on the summer precipitation over the southeastern TP as the main water vapor transport is from the western and southern boundaries.

4.3. Regression Analysis

[40] The regression of vertically integrated water vapor transport anomalies with the precipitation index (shown in Figure 9i) is presented in Figure 11. The shading indicates the regions that show statistical significance at the 5% level based on t test. The results derived from three reanalysis data sets all feature a zonally orientated anticyclonic anomaly across the Indian subcontinent and the Bay of Bengal, extending from the northern Indian subcontinent to the South China Sea and with the center located at 25°N. The strengthened western water vapor transport at the northern edge of the anticyclone leads to more summer precipitation over the southeastern TP. Although quantitative budget analysis exhibits a spread among the three reanalyses, as shown in Table 3, these data sets are highly consistent in featuring the anticyclone, which dominates the water vapor supply to the excessive rainfall over the southeastern TP. This circulation pattern explains why water vapor transport across the western and southern boundaries can modulate the interannual variability of rainfall over the TP.

Figure 11.

Regression of the summer precipitation index over the southeastern TP on the vertically integrated water vapor transport (unit: kg/m/s) during the period 1979–2002. (a) NCEP/NCAR, (b) ERA40, (c) JRA25, and (d) Ensemble mean from three reanalysis data sets. Shading indicates statistical significance at the 5% level based on the t test.

5. Summary and Discussion

5.1. Summary

[41] Although the TP is usually called the “water tower of Asia” owing to its importance in the hydrological cycle, the summer water vapor transport over the TP at both the climate mean state and the interannual timescale is still poorly understood, mainly due to the sparse data available. In this paper, we attempt to provide a complete picture by using eight precipitation data sets and three reanalysis circulation data sets. Budget analysis on the water vapor transport over the TP is done. The climate mean water vapor transport derived from the three reanalysis data sets are quantitatively compared, and the interannual variability of the water vapor transport is revealed. The major findings are summarized below.

[42] 1. Multidata set analysis indicates that precipitation over the TP peaks in summer and exhibits a southeast-northwest decreasing gradient. The summer precipitation intensity over the southeastern TP is larger than 5 mm/day.

[43] Among the gridded precipitation data sets, both the CMAP and the APHRO data sets are highly consistent with the station data on the magnitude of climatological rainfall distribution. Three reanalysis data sets, including the NCEP/NCAR, the ERA40, and the JRA25, capture well the annual cycle and spatial pattern of summer precipitation over the TP. However, they all overestimate the precipitation amount by more than 40%, especially the NCEP/NCAR data. The JRA25 data show a comparatively better performance than the NCEP/NCAR and the ERA40.

[44] 2. During the rainy season in the TP, the climatological water vapor sources are mainly from the southern and western boundaries. The former originates from the Indian Ocean and the Bay of Bengal, and the latter comes from the Indian Ocean and midlatitudes. The water vapor transport from the western boundary is about 32% of that from the southern boundary, indicating that the southern boundary is the main water vapor channel for TP rainfall.

[45] The multidata set ensemble mean results indicate that the TP is generally a moisture sink during summer at climate mean states, having a net moisture convergence of 4 mm/day. The water vapor convergence along the TP is dominated by the moisture advection term, while that within the central TP is dominated by both the moisture advection and the wind convergence term.

[46] Three reanalysis products are highly consistent on the spatial pattern of climatological water vapor transport. The main water vapor input is from the southern boundary, which decreases from the surface to the upper levels. The main water vapor output is from the eastern boundary, which outflows from the TP in the middle level (about 700–400 hPa).

[47] However, the water vapor transport from the southern boundary in the JRA25 is about 50% larger than that in the NCEP/NCAR and the ERA40. The water vapor input from the western boundary is mainly concentrated in the lower (1000–700 hPa) level in the NCEP/NCAR, while in the ERA40 and the JRA25, it is mainly concentrated in the middle (700–400 hPa) level. Although the three reanalyses all exhibit a moisture sink over the TP in summer, their results on the contribution of the wind divergence term to the total convergence differ, being negative (positive) in the NCEP/NCAR (ERA40 and JRA25).

[48] 3. The summer rainfall over the southeastern TP exhibits robust interannual variability, with STD of 1.3 mm/day during 1979–2002. A comparison with the station record indicates that the Xie and APHRO data sets show the best performances on the interannual variability, having temporal correlation coefficients larger than 0.8. The three reanalysis data sets, including the NCEP/NCAR, the ERA40, and the JRA25, show comparatively lower correlations with station data (<0.7) than the other four gridded precipitation data sets.

[49] 4. The ensemble mean results from the three reanalysis data sets indicate that the excessive rainfall over the southeastern TP on interannual variability follows increased water vapor inputs from both the western and southern boundaries, mostly from the surface to 400 hPa in vertical layers. An excessive rainfall anomaly of 1 mm/day over the southeastern TP is associated with an anomalous water vapor input of 138 (104) kg/m/s from the western (southern) boundary.

[50] The three reanalysis products are consistent with each other in measuring the change in water vapor transport from the western and eastern boundaries, although the magnitude in the JRA25 is 50% larger than that in the NCEP/NCAR and the ERA40. The main discrepancy is seen in the change in water vapor transport from the southern and northern boundaries.

[51] 5. The anomalous water vapor transport patterns from three individual reanalysis data sets (NCEP/NCAR, ERA40, and JRA25) all show that the interannual variability of summer rainfall over the southeastern TP is dominated by an anomalous anticyclonic water vapor transport over the northern Indian subcontinent and the Bay of Bengal. The anticyclone enhances water vapor transport along the southern edge of the Himalayas and thereby produces excessive rainfall over the southeastern TP.

5.2. Discussion

[52] In addition to analyzing rain gauge measures, satellite products, and rain gauge merged data sets, we have also done analysis on precipitation data derived from reanalysis. Since observational circulation information has been assimilated into the model during the reanalysis processes, we may partly regard the circulation fields of reanalysis data as real. The precipitation data from reanalysis were purely predicted by the current most advanced model which is forced by the analyzed initial state. The high correlation coefficients between the reanalysis precipitation and the observed interannual variation demonstrate the importance of the realistic simulation of the atmospheric circulation over the TP, while the fact that all reanalysis products overestimate the climate mean precipitation amount by at least 40% indicates the importance of model physics. How to improve precipitation simulation over the TP thus deserves further study.

[53] Since three data sets used in our analysis ended in 2002, i.e., the two precipitation data sets (CMAP, GPCP) and one reanalysis data set (ERA40), we chose a uniform time period 1979–2002 for multidata set comparison. Since both the NCEP/NCAR and the JRA25 reanalysis products have extended to recent years, to examine whether the relationship between TP precipitation and water vapor transport has changed in recent years, we further investigate the anomalous water vapor transport patterns using updated reanalysis products, i.e., the NCEP/NCAR (1979–2010) and the JRA25 (1979–2008) (Figure 12). A significant anticyclonic water vapor transport is still evident over the northern Indian subcontinent and the Bay of Bengal in the two reanalyses. Therefore, the results based on the period 1979–2002 are correct for the period up to the present.

Figure 12.

Regression of the summer precipitation index over the southeastern TP on the vertically integrated water vapor transport (unit: kg/m/s). (a) NCEP/NCAR (1979–2010) and (b) JRA25 (1979–2008). Shading indicates statistical significance at the 5% level based on the t test.

Acknowledgments

[54] This work was partly supported by the National Natural Science Foundation of China under grants 41125017 and 40890054, and by the National Key Technologies R&D Program under grant 2007BAC29B03.