We use meteorological fields provided by the High Asia Refined analysis (HAR). The HAR is the result of the dynamical downscaling of the global gridded data set, the Operational Model Global Tropospheric Analyses (final analyses, FNL; data set ds083.2). These final analyses are available every 6 h and have a spatial resolution of 1 ∘. The model used for this purpose is the advanced research version of the Weather and Research Forecasting model (WRF-ARW, Skamarock and Klemp, 2008) version 3.3.1. The HAR provides products at a spatial resolution of 30 km and temporal resolution of 3 h for the first domain, covering most parts of central Asia (HAR30). A second nested domain (HAR10) covers high Asia and the TP with a spatial resolution of 10 km and temporal resolution of 1 h (Fig. 1). The data set is available online at http://www.klima.tu-berlin.de/HAR and described in detail by Maussion et al. (2011, 2014). The HAR provides meteorological fields at the surface and on 28 terrain-following vertical sigma levels. The data set covers a period of more than 12 years from October 2000 to December 2012 and is updated continuously. HAR products are available for different time aggregation levels: hourly (original temporal model resolution), daily, monthly and yearly.

The HAR precipitation data were compared to rain gauge observation and precipitation estimates from the Tropical Rainfall Measuring Mission (TRMM) by Maussion et al. (2011, 2014). The accuracy of the precipitation data is described more in detail in Sect. 4.2.

Atmospheric water transport happens through WV transport and as cloud particle (CP) transport. In this study, we are interested in cloud particle transport but do not further distinguish between liquid (water droplets) and solid (ice) cloud particles which are both resolved by the model microphysics. WV and CP fluxes are calculated for each of the 28 original sigma levels, which are terrain following, based on the original temporal model resolution of 1 h (HAR10) and 3 h (HAR30) using the formula Q=vhρqΔz, where Q is the water vapour flux (kg m-1 s-1) or cloud particle flux, vh denotes the horizontal wind vector (m s-1), ρ is the dry air density (kg m-3) and q is the specific humidity (kg kg-1). Δz is the thickness of each sigma level (m), this value is not constant but increases with increasing height above ground. Since the WRF model just provides mixing ratios (r) for the three atmospheric water components (water vapour, liquid water and ice), we first calculated the specific humidity for each component using the relationship q=r1+r. Additionally, we integrated the fluxes over the whole atmospheric column to obtain the vertically integrated atmospheric water transport fluxes. The vertical integration is performed along the metric z coordinate along the model sigma levels from surface to top using the rectangle method. Q=∫z=zsfcz=ztopvhρqΔz We calculated these fluxes for the original model levels and did not interpolate them to pressure levels to avoid information loss due to the interpolation. For the analyses, 10 and 5 grid points from the HAR30 and HAR10 domain boundaries, respectively, are removed to avoid lateral boundary effects.

To analyse the AWT towards the TP, we compute vertical cross sections along transects following the border of the TP. To be able to calculate a moisture budget, we defined a region which we henceforth call “the inner TP”, with 14 transects attempting to follow the highest elevations in the mountain ranges and to cut across the high mountain valleys which we assume to be pathways for atmospheric moisture. A map with the transects is shown in Fig. 1. The u and v components of AWT are then rotated to the transect coordinate system to compute the normal fluxes towards the cross section.

To examine if our data set is able to reproduce the general characteristics of the WV flux, we compare the WV fluxes derived from HAR30 with ERA-Interim Reanalysis data (Dee et al., 2011). The ERA-Interim WV fluxes are available online as an integral over the atmospheric column for the eastward (u) and northward (v) components as monthly means. ERA-Interim has a horizontal resolution of 0.75∘. To calculate the differences between the HAR30 and ERA-Interim WV fluxes, we transformed HAR30 data to the ERA-Interim grid by averaging the HAR30 grid points below each ERA-Interim grid point. The u and v components of the HAR fluxes were rotated to Earth coordinates first.

The general patterns and the magnitude of the WV transport amounts of HAR30 and ERA-Interim are in agreement (Fig. 2). Figure 2c shows the differences between HAR30 and ERA-Interim for July when the largest differences were found. The main differences between the two data sets are visible south of the eastern and central Tibetan Plateau along the southern slopes of the Himalayas. The WV transport through the Brahmaputra Channel towards the Tibetan Plateau is higher for ERA-Interim than for HAR30. This is probably due to differences in the representation of the orography, caused by different horizontal resolutions. HAR30 produces more transport westward along the Himalayas (upstream Ganges River), which is caused by more WV blockage. When the WV flux hits the Himalayas from the south it is mostly redirected to the west and follows the southern slopes of the Himalayas. Additionally there are differences in the Arabian Sea and the Bay of Bengal. Over the Arabian Sea, more water vapour is transported further to the south in the HAR30 data set. The transport direction in the ERA-Interim data set is more from the south-west to north-east. Therefore, the transport amount over the southern part of the Indian peninsula is also higher for HAR30. Because of that southward shift, the transport amount over the southern part of the Bay of Bengal is higher for HAR30 than for ERA-Interim and thus has a stronger northward component in the eastern part of the bay. So the South Asian Monsoon circulation has a modified shape in the HAR, possibly due to the influence of the southern branch of the mid-latitude westerlies which is more pronounced in HAR30. In winter, the differences are in general less pronounced (not shown).

Figure 3 displays the December–February (DJF) (left) and June–August (JJA) (right) decadal average of the vertically integrated atmospheric water transport (AWT) derived from HAR30. In winter (DJF), the AWT from west to east is dominant over the TP and most parts of high Asia. In the tropical ocean region, there is transport from east to west with the trade winds. The TP and the regions north of the plateau show small transport amounts, below 40 kg m-1 s-1. For comparison, AWT reaches up to 450 kg m-1 s-1 over the South Chinese Sea off the coast of Vietnam. In summer (JJA), the pattern in the southern part of the domain is completely different from winter due to the circulation change related to the Indian summer monsoon (ISM). The largest amount of atmospheric water is now transported from west to east over the Arabian Sea and the Bay of Bengal. Over the Bay of Bengal the flow gets a larger southerly component, and atmospheric water is directly transported to the southern slopes of the Himalayas. Over the TP the transport amount is still low in comparison.

We display the decadal average of AWT for selected vertical levels in Fig. 8. We selected the levels 1 (∼ 25 m above ground in Tibet), 5 (∼ 450 m above ground in Tibet), 8 (∼ 1200 m above ground in Tibet), 10 (∼ 2200 m above ground in Tibet), 12 (∼ 3200 m above ground in Tibet) and 15 (∼ 5500 m above ground in Tibet) because these levels show the most interesting features of the WV transport. The WV transport near the ground (level 1) is generally low in the TP and just a little bit higher south of the Himalayas, probably because of the lower surface wind speeds and of the stronger mixing in the boundary layer. The transport amount increases strongly up to level 12, where the largest transport occurs, due to higher wind speeds, higher moisture availability or both. Above this level, the WV transport starts to decrease and above level 17 (not shown) the transport amount is close to zero.

The general atmospheric circulation at different levels is visible in the transport patterns, but we have to consider that AWT is a complex mixture of wind and moisture availability. At the lower levels (1 and 5) we see the cyclonic circulation around the Tibetan heat low, with its centre in the central TP. At level 8, this structure shifts to the central northern TP. Level 12 and 15 (and levels in between) show an anticyclonic circulation around a centre in the southern TP, directly north of the Himalayas. This is the high-tropospheric Tibetan anticyclone that forms during May or early June (Flohn, 1968). In level 15 and 16, a second anticyclonic circulation is visible in the south-east of the domain, directly south of the Himalayas. Above these levels the anticyclonic circulation slowly weakens and a division into a northern part with transport from west to east, where the westerlies are dominant, and a southern part with transport from east to west is visible. Level 10, which lays between the cyclonic (level 1 and 5) and anticyclonic (level 12 and 15) circulation features, could be called the equilibrium level.

At level 5, monsoonal air and moisture is transported relatively far to the western (north-western) parts of the TP. Air from the south which originates in the tropical oceans (Indian summer monsoon) is included in the cyclonic circulation over the TP, but the WV does not seem to originate from the East Asian monsoon. In the higher levels (10–15), this cyclonic circulation is replaced by the westerlies and therefore extra-tropical air masses are transported to this region. Therefore, we find air masses and consequently moisture from different sources at one place. These results should be considered for the analysis of stable oxygen isotopes in precipitation samples, lake water, sediment and ice cores. Precipitation originating in the boundary layer will result in a monsoonal signal in the isotopes while precipitation originating from deep convection could have an isotope signature dedicated to the westerlies.

In the dry Tarim Basin north of the TP, we can see an anticyclonic circulation above the boundary layer at level 10 and transport of WV from north-east to south-west following the northern boundary of the TP. Therefore, the air at this level tends to descend. This means that just below this level clouds in the boundary layer are possible. These clouds can only provide small amounts of precipitation due to their low vertical extent. Above this level, the transport of WV is admittedly higher and in the opposing direction, but does not result in precipitation. Deep convection is inhibited by the subsidence tendency at the lower level.

In the south-western parts of the domain, in the border region between India and Pakistan, we can see the same feature, but there it leads to higher differences in the precipitation patterns and affects a region with a higher population density. We see an anticyclonic circulation of the WV transport in the levels 10 and 12 (and in between). The transported amount is nearly as high as the WV transport associated with the ISM along the southern slopes of the Himalayas. However, if we look at the precipitation patterns, we see that there is a precipitation minimum in this region (Fig. 6). The development of deep convection is suppressed by subsidence. In the lower levels, the heat low over Pakistan (Bollasina and Nigam, 2010) is visible in the transport patterns (Fig. 8) and in the 10 m wind field (Fig. 6). Saeed et al. (2010) point out that the heat low over Pakistan connects the mid-latitude wave train with the Indian summer monsoon. In the surrounding region where we do not see this anticyclonic movement in the levels above the boundary layer, the large amounts of transported WV result in high amounts of precipitation. These results can provide an indication of the processes, which lead to the risk of droughts and floods (e.g. in July 2010) in Pakistan, as already analysed by Galarneau et al. (2012)

We calculated the WV input towards the TP through its borders (Fig. 9 and Table 1). For the southern boundary cross sections (CS 2–6) the highest transport amounts occur in summer (Fig. 9a) in the lower layers and decreases with increasing height. The largest fluxes occur in the regions where the elevation is lower compared to the direct surroundings, in the large meridionally orientated valleys (eastern Himalayas, in the region of the Brahmaputra channel). There, the areas of lower elevation are wider and the AWT from the Indian Ocean hits the mountain ranges directly from the south.

The AWT from the TP to the south is negligible in summer (Fig. 9a). It occurs mainly in CS 2 and a little in CS 3. This kind of recirculation may reflect northward transport. It takes place in the lower levels above a layer with high northward transport amounts. In winter (Fig. 9b), AWT is lower but the input of atmospheric moisture is still dominant.

The AWT through the western boundary (CS 1) (Fig. 9a and b, and Table 1) is higher in winter than in summer, as it is for CS 2 in the westernmost region of the Himalayas. These regions are dominated by atmospheric water input by extra-tropical air masses, transported to this region by the westerlies. In sum, the AWT in CS 2 is still almost as high as the AWT in CS 6. CS 6 contains the Brahmaputra Channel, which is often referred to as one of the main input channels for atmospheric moisture (Tian et al., 2001).

Table 1 shows the monthly decadal average of AWT input to the TP through the individual cross sections converted to a theoretical equivalent precipitation amount on the inner TP. We picked CS 1 and CS 6 for a closer comparison because CS 6 includes the Brahmaputra Channel and CS 1 is the western boundary, and they are of the same length. From November to April the transport through the western boundary (CS 1) is distinctly greater than that through CS 6. From May to October CS 6 shows higher transport amounts, but the differences are less than for the wintertime. In July the input of CS 1 is 90.47 % that of CS 6. This means that the input through the western boundary is around 90 % of the transport through the Brahmaputra Channel region. This is the month where the differences are smallest. We can see that for almost every month, CS 2–3 exhibits the largest input amounts. AWT through this cross section is controlled not only by the ISM but also by the southern branch of the mid-latitude westerlies.

The cross section for the eastern boundary (CS 7) shows that the TP is a source of atmospheric water for the downstream regions east of the TP for all months (Table 1 and Fig. 9c and d). In January, there is only eastward transport through the eastern boundary (Fig. 9d). In the other months (not shown except for July, Fig. 9c), we find additionally transport towards the TP in the lower layers near the surface. The transport towards the TP through eastern cross section has its peak in July (Fig. 9c) in the northern parts of the boundary where the elevation is distinctly lower than in the southern parts. Above this region, there is still eastward AWT away from the TP. However, if we look at the total of the AWT amount through the eastern boundary, we see that the transport from the plateau towards the east is also dominant in summer.

The transport through the northern boundary (CS 14–8) towards the TP (input) is lower than from the west and south in January and July (Fig. 9d and c, Table 1), although the circulation is directed to the boundary of the TP especially in summer. There, we find a strong gradient in altitude and fewer passages through which the atmospheric water could enter the TP than in the Himalayas. For the westernmost northern cross sections (14–12) the transport from the TP to the north is dominant. The reason for this is the north-eastward transport in the western TP, which also explains the lower transport amounts towards the TP. The AWT within the northern branch of the westerlies north of the TP is blocked by the high-altitude TP. AWT then follows the northern border of the TP to the east, where the elevation is lower in some regions (CS 9–11), e.g. at the border to the Qaidam Basin (CS 9). There, the input of atmospheric water to the TP is dominant for all months and the maximum input takes place in spring. AWT from the north to the Qaidam Basin is also visible in Fig. 4 for all months. This transport takes place in the lower layers of the atmosphere. The transport from the plateau northwards has its peak at the easternmost northern CS (CS 8) in July, August and September, when the TP can provide large amounts of atmospheric water, as shown in Sect. 3.2.1 in Fig. 4.

Since we analysed AWT transport in and to the TP and quantified the input and output, the question arises of which amount of precipitation falling on the inner TP results from external moisture supply and which amount is provided by the TP itself from local sources and moisture recycling.

Table 1 displays the monthly decadal average of AWT through the individual cross sections and the sum for all cross sections, the precipitation falling on the inner TP and the ratios between them. To make the comparison with the precipitation easier, we converted the net atmospheric water input to a theoretical precipitation equivalent (mm month-1). We obtain an annual mean AWT input of 206.0 mm yr-1 for HAR10. For the mean annual precipitation falling on the inner TP, a value of 559.2 mm yr-1 results. The ratio of net input of atmospheric water to the precipitation falling on the inner TP reveals that, on average, AWT through the borders accounts for 36.8 % of the precipitation during the year. According to this, the remaining 63.2 % of atmospheric water needed for precipitation must be provided by the TP itself. This moisture supply probably takes place via moisture recycling from local sources, e.g. evaporation from numerous large lakes, soil moisture, the active layer of permafrost, snow melt and glacier run-off. The ratio is highest in winter when the TP cannot provide moisture for precipitation by itself, followed by summer, where the largest net input occurs. In October and November the ratio is negative, which means that the TP provides more moisture than it receives from external sources. These are the 2 months where the output of moisture from the TP is larger than the input; this is possible because the moisture imported to the TP in summer is available for export in autumn. On a monthly basis, there is certainly a time lag between the moisture entrance and the precipitation, making the analysis of the monthly ratios difficult. The standard deviations for HAR10 (HAR30) in Table 1 (2) show that the atmospheric water input varies more between the years than the precipitation falling on the TP. This implies that the evaporation from local sources stabilises the precipitation falling on the inner TP.

The main WV input to the TP takes place through the southern and western boundaries, confirming the results of Feng and Zhou (2012), even if their western boundary is further east. The WV entering the TP through the eastern part of the southern boundary originates from the monsoonal air masses, while the WV entering the TP through the western boundary originates in the mid-latitude westerlies. The relatively high input through the western boundary shows that the westerlies are not fully blocked by the TP and not all moisture transported with them is redirected north or south. The magnitude of this WV input is similar to that of the input through the Brahmaputra channel. This agrees with the findings of Mölg et al. (2013), who found that the westerlies play a role for precipitation and glacier mass balance in summer also, and they challenged the assumption that the westerlies contribute moisture only in winter or just in the northernmost parts of the TP in summer (Hren et al., 2009; Tian et al., 2007). Our results show that there is direct atmospheric water transport through the western boundary by the mid-latitude westerlies in summer, which shoes that the westerlies are not fully blocked by the TP. The westerlies also contribute moisture to the TP through valleys in the western parts of the southern boundary by the southern branch of the westerlies. This implies that moisture entering the TP from the south-west can be transported there either by the westerlies or the monsoon, depending on the how far these systems extend eastward or westward along the southern slopes of the Himalayas, respectively. The examination of this structure will be the subject of a subsequent study. The main WV output from the TP takes place through the eastern border, as also found by Feng and Zhou (2012). The TP is a source of moisture for the downstream regions in the east throughout the year like the Yangtze River valley in China, agreeing the results from Chen et al. (2012) and Luo and Yanai (1983), that the TP contributes precipitation to its downstream areas in summer. Thus we can confirm the importance of the findings from Bin et al. (2013), Xu et al. (2011) and Chen et al. (2012), who called the TP a transfer or re-channel platform of moisture for the downstream regions in East Asia.

In our study, we could not find any contribution of the East Asian Summer Monsoon to the WV transport towards the TP, although it has thus far been assumed to have an influence on the TP (Yao et al., 2012; Bolch et al., 2012). The WV transported from east to the TP in the lower levels in summer (CS 7, Fig. 9c), also detected by Luo and Yanai (1983) and Feng and Zhou (2012), is not transported to this region by the East Asian Monsoon flow but by the eastern branch of the Indian summer monsoon flow. This is clear if we look at the transport patterns for HAR30 (Fig. 3). It is interesting that the HAR WV flux has a stronger westward component east of the TP than ERA-Interim (Fig. 2c) and still does not show any transport from east to west in the climate mean state. Since we focus on the mean climatology during our period of investigation, we cannot exclude the contribution of moisture from the East Asian summer monsoon (EASM) to the TP for single years or events, but at the same time we expect it to be visible in the mean if it is significant. Our hypothesis is that the transient weather systems could bring moisture from the east to the TP; however they are not visible in the means. It will be the goal of a future study to examine such weather systems in detail.

Prior studies focused on the WV transport and did not consider the CP flux. The assumption so far was that the CP transport is so small compared to the WV flux, that it does not have a significant influence on the atmospheric moisture transport. Our results show that the contribution of the CP flux to the entire AWT is not negligible in winter in the Pamir and Karakoram ranges (the western and south-western border of the TP), where it contributes up to 25 % of the entire AWT. The fact that the CP transport plays a role in the Karakoram and western Himalayas, the regions which are controlled mainly by the westerlies, lets us conclude that in this region moisture advection presumably plays a strong role. The horizontal motion is dominant in advective processes towards convection where the vertical motion is dominant. This leads to the fact that clouds developed in advective processes, for example frontal processes, can be transported further away from their origin than convective clouds.

The moisture supply from external sources provides around 36.8 % of the atmospheric water needed to produce the mean annual precipitation on the inner TP while the remaining part originates from the TP itself by local moisture recycling. These results highlights the importance of local moisture recycling as already emphasized by Kurita and Yamada (2008), Joswiak et al. (2013) and Chen et al. (2012). For the northern Tibetan Plateau, Yang et al. (2006) detected that 32.06 % of the precipitation is formed by water vapour from ocean air mass and 46.86 % is formed by water vapour evaporated from local sources. They found that at least 21.8 % of the precipitation is formed by water vapour evaporated on the way and then transported by the monsoon circulation. Yang et al. (2007) also showed that for two flat observation sites in the central eastern part of the TP, the evaporation is 73 and 58 % of the precipitation amount. This is in a good agreement with our results that 63.3 % is provided by local moisture recycling. This moisture is provided by evaporation from numerous large lakes, soil moisture, the active layer of permafrost, snow melt and glacier run-off. The question arises over what will happen with the atmospheric water, which is transported to the TP. Does it remain in the TP or is it lost as run-off. Could this moisture input be an explanation for the observed lake level rises?

The comparison of the net atmospheric water input to the TP through the cross sections, the precipitation falling on the inner TP and the ratio between them for HAR10 (Table 1) and HAR30 (Table 2) show that the different horizontal resolutions result in differences between the two data sets. On an annual basis, the different horizontal resolutions result in an AWT input difference of 57.3 mm yr-1 and a precipitation difference on the inner TP of 23.3 mm yr-1, where HAR30 has the lower AWT input but the higher precipitation amount. This leads to a difference of 11.3 percentage points in the ratio of AWT to precipitation between the HAR10 (36.8 %) and HAR30 (25.5 %) data sets. Nevertheless, HAR30 has, for almost every cross section and month, higher transport amounts for both input and output. Due to the fact that the output values are also higher, the annual net input for HAR30 is lower than for HAR10. On an annual basis HAR30 shows just three-fourths of the HAR10 AWT input. A possible explanation for lower transport amounts for individual cross sections in HAR10 could be that the higher horizontal resolution is overridden by higher orographic barriers due to better representation of the topography. Shi et al. (2008) showed that a higher horizontal resolution and more realistic representation of the topography is important for the development of disturbances leading to precipitation events in the downstream regions of the TP like the Yangtze River valley. Our study shows that for a quantification of AWT and the spatiotemporal detection of its major pathways and sources it is important to examine the complex topography of high Asia with high spatial resolution.

The uncertainty of the results depends on the accuracy of the data itself, the position of the cross sections used to calculate the budget and the vertical resolution of the data set.

Maussion et al. (2014) compared HAR precipitation with rain-gauge observations and the TRMM precipitation products 3B42 (daily), 3B43 (monthly) and 2B31 (higher resolution). They found an improvement when increasing the horizontal resolution from 30 to 10 km in comparison to the gauges. A slight positive bias could be detected against the same stations (0.17 mm day-1 for HAR10 and monthly precipitation values, their Fig. 3), comparable to that of TRMM 3B43 (0.26 mm day-1). Converted to annual values (62 mm yr-1) and compared to the value of HAR precipitation averaged over the inner TP (559 mm yr-1), this bias remains significant. In a simple first-order approach, by assuming this bias to be constant over the region (and assuming no rain gauge undercatch), this would increase the part of moisture needed for precipitation coming from the outer TP from 36.8 to 41.4 %.

To determine if the position of the cross sections has an influence on our results, we replicated our budget analyses with the cross sections moved around 60 km towards the centre of the TP for the HAR10 data set. This results in new budget values for net atmospheric water input of 202.6 mm yr-1 (206.0 mm yr-1 with the old cross sections), and 506.3 mm yr-1 (559.2 mm yr-1) precipitation falling on the inner TP. This result in a change of the percentage from 36.8 % (found for the original position of the cross sections) to 40 %. This is caused mainly by a change of the precipitation amount of -52.9 mm yr-1, while the atmospheric water input is nearly the same (-3.4 mm yr-1). This change is smaller than the standard deviation (6.3 %) of this ratio.

We analysed if the vertical resolution could have an influence on precipitation and moisture transport by creating a new series of simulations for the entire year of 2010, whereby we increased the number of vertical levels from 28 to 36 (all other settings kept unchanged). Precipitation patterns in the HAR10 domain remain very similar. Large absolute differences are found in the monsoonal affected regions south of the TP and the Himalayas. The largest relative differences occur in regions with very low precipitation rates, in the arid regions (e.g. Tarim Basin) north of the TP. The amounts and patterns of the vertically integrated atmospheric water transport match well with one another, relative differences are around ±5 % in the TP and only slightly higher in small regions south and north of the TP. The computation of the water budget for the year 2010 for 36 (28) vertical levels results in a net input of atmospheric water of 191.9 mm yr-1 (195.2 mm yr-1), with 623.3 mm yr-1 (621.8 mm yr-1) precipitation falling on the inner TP and an annual proportion of 30.8 % (31.4 %) for 2010. This shows that the results obtained with 28 vertical levels are reliable.

There is no other possible way to estimate the uncertainty of the computed fluxes than the rough comparison of ERA-Interim provided in Sect. 3.1. Certainly, the choice of the model set-up and the reinitialization strategy also influence our results. Put together, these uncertainties are not negligible but there is no indication that our core conclusions are significantly affected.

Due to the temporal averaging of the model results to monthly means, we ascertain the mean climatology, but lose the ability to analyse the data process based. The processes regarding the interplay of atmospheric water transport and precipitation will be the subject of a subsequent study at a higher temporal resolution (e.g. daily).