Moisture source assessment and the varying characteristics for the Tibetan Plateau precipitation using TRMM

Precipitation over the west Tibetan Plateau (TP) was always being accused of lacking observations that limits the precipitation moisture attribution and quantitation over the whole TP. By introducing precipitation from the Tropical Rainfall Measuring Mission (TRMM) and other observation-based data, moisture sources for the whole TP and their variations from 1998 to 2018 are explored using an Eulerian model. It is found that the Southwest subregion from TP to the western Indian Ocean, is the largest moisture contributor. It contributes around 147.6 ± 13.0 mm yr−1 in water-depth of the TP in climatology, accounting for 31.9 ± 1.9% of the annual precipitation. The TP, the West (TP to Europe), and the Southeast (TP to Indochina Peninsula) follow by contributing 23.6 ± 2.3, 21.8 ± 1.5, and 2.6 ± 0.6%, respectively. Circulations dominate the TP in different seasons. Take spring for example, the westerlies prevail over the TP and the West contributes the most moisture, which accounts for 38.6 ± 2.9% of the spring precipitation. In summer, with the breakout of the Indian monsoon, contribution from the Southwest reaches the highest of 91.1 ± 11.5 mm JJA−1, accounting for 34.6 ± 2.6% of the summer precipitation. The interannual variability (IAV) of the TP precipitation is mainly influenced by the moisture IAVs from the Southwest and the TP, contributing around 36.6% and 31.7%, respectively. Moisture contributed from the Southwest decreases significantly from 1998 to 2018 at a rate of −10.6 mm yr−1 dec−1, but moisture from the local increases significantly at 12.1 mm yr−1 dec−1. Further analyses reveal that the local increase in moisture contribution (and ratio) is primarily due to intensified evaporation of the TP, but the Southwest decrease is mainly caused by reduced moisture transport from the Indian monsoon.


Introduction
The Tibetan Plateau (TP), also known as the 'Third Pole' , exerts great influence on the climate of the surrounding and the northern hemisphere (Ye et al 1979. The high elevated landmass (more than 5 km above sea level on average) serves as a large heating source that strengthens and amplifies the Asian monsoons (Duan and Wu 2005, Zhao et al 2018. The large landmass also hinders the transport of the westerlies (Wu and Zhang 1998) and blocks the warm humid air in the south from entering into the north (Boos and Kuang 2010).
In addition to the crucial role in climate, the giant TP is also birthplace to many large rivers in Asia (Immerzeel et al 2010, Sun and Wang 2019, Zhang et al 2020, such as the Yellow River, the Yangtze River, and the Brahmaputra River etc. Since precipitation is a major source of the TP rivers and lakes (Su et al 2016, Lei andYang 2017), finding the sources of the TP precipitation is of great importance and has invoked the interest of many scientists.
In early studies, two main moisture transport channels were identified for the TP, i.e. the Indian monsoon channel and the westerlies channel (Sugimoto et al 2008, Feng and. Through the moisture trajectory analyses with HYSPLIT (Ma et al 2020), more diverse moisture paths were found, but they can still be classified into the two major channels aforementioned. For some regions in the eastern and southern TP, there exits another moisture corridor from the east, which is possibly influenced by the East Asian summer monsoon (EASM) (Ma et al 2020). By adopting an 'areal source-receptor attribution' method, Sun and Wang (2014) quantified the moisture contributions from different sources to the eastern TP. The results indicate that moisture released over the eastern TP mostly comes from the Eurasian continent. Although moisture uptake from ocean is considerable, much is lost in route. Recently, similar conclusions are reached that terrestrial originated moisture is dominant for precipitations in the central-western TP (Zhang et al 2017) and the Sanjiangyuan region (Zhang et al 2019b).
It is also being known that the moisture origin over the Plateau is not spatially homogeneous. Based on isotope observations, Yao et al (2013) divided the TP into three climate zones and claimed that moisture in the northern TP is mainly influenced by the westerlies, while moisture from the southern TP is mainly influenced by the monsoons. This viewpoint is supported by Zhang et al (2019a) and Pan et al (2019) through models. Zhang et al (2019a) further gave an estimation that 38.9% of precipitation moisture for the northern TP comes from the westerlies, and 51.4% of precipitation moisture for the southern TP comes from monsoons. Besides, as different circulations influence them, their precipitations display different change trends with a north increase and south decrease pattern in recent decades. The northern and central-western TP precipitation increase can be attributed to more moisture contributions from the TP and the monsoons (Zhang et al 2017(Zhang et al , 2019a. Precipitation change in the southern TP is, however, more complicated. Through a further division of the southern TP, Chen et al (2019) found that large differences exist in the sub-seasonal moisture source evolution between the southeastern TP and the southwestern TP, which suggests internal differences within the southern TP.
Despite the progress made so far, the moisture sources of the TP are still in a lack of an accurate estimation. The reasons vary with different methods. For studies that apply general circulation model (GCM) (e.g. Pan et al 2019), the GCM generally overestimates the precipitation over the TP, causing large uncertainties in attributing the precipitation moisture. For studies that apply the Lagrangian moisture diagnostics, the tracked moisture is released moisture in the air, which does not match well with the observational ground precipitation (e.g. figure 10(a) in Chen et al (2019)). For studies that apply the Eulerian model (e.g. Zhang et al 2017), the input of precipitation is mandatory, but the precipitation gauges on the west TP are extremely sparse, which limits the applicability of the gauge-based precipitation at the TP scale. Fortunately, with the rise of the satellite era, more sensors are being designed to collect precipitation data, such as the Tropical Rainfall Measuring Mission (TRMM). The TRMM Multisatellite Precipitation Analysis (Huffman et al 2007), which has already been widely applied (e.g. Chen et al 2013, Zhang andTang 2015), provides a unique opportunity to explore precipitation over sparsely gauged or ungauged regions. As a reliable precipitation dataset is a prerequisite for accurate assessment of the precipitation origins, the application of TRMM over the TP can be significant in unveiling the moisture sources for the whole TP for the first time. Based on the moisture traceback from 1998 to 2018, the moisture contribution, the interannual variability (IAV), the changing trend, and the seasonal variation of the sources are all explored, respectively.

Data and study area
The primary data include precipitation, evaporation, and atmospheric data. They serve as model input for moisture tracking. For precipitation, the gauge-calibrated TRMM research product 3B43 (V7) is chosen, which covers the latitude of 50 • N-S at 0.25 • × 0.25 • grids monthly from 1998 to 2018. For evaporation over the land, the 3 h 1 • gridded evaporation product from the community land model (CLM) in the Global Land Data Assimilation System (GLDAS; Rodell et al 2004) is chosen. The CLM is a physically-based model that is subject to vigorous evaluation. The forcing data, including precipitation, temperature, radiation, etc, are observational. In general, GLDAS outperforms other reanalyses on surface variables Zeng 2012, Gao et al 2014). Over the ocean, the monthly 1 • gridded evaporation from the Objectively Analyzed Air-Sea Fluxes (OAFlux, Yu and Weller 2007) is adopted. OAFlux has assimilated satellite data since 1985. According to Trenberth et al (2011), OAFlux evaporation agrees well with in situ buoy data and appears to be best among the available products. In addition to these observation-based data, the 3 h 1 • gridded evaporation and precipitation fields from the European Centre for Medium-Range Weather Forecasts (ECMWF) Interim (ERA-I hereafter; Dee et al 2011) are also applied. Precipitation and evaporation from ERA-I are pure model output (Dominguez et al 2006), which are used for comparison.
For the atmospheric data, the reanalysis of ERA-I at 1 • × 1 • grids is chosen because it has a better performance concerning the atmospheric water budget among the available reanalyses (Trenberth et al 2011, Lorenz andKunstmann 2012). ERA-I provides a variety of atmospheric data, including 6 hourly model-level zonal winds, meridional winds, and specific humidity; 6 hourly surface pressures; and a set of vertically integrated moisture and flux variables (vertically integrated water and vertically integrated northward/eastward water fluxes in the forms of vapor, liquid, and ice).
The geographic location of the TP is shown in figure 1(a) . The topographic height data are provided by the Global Land One-km Base Elevation Project (GLOBE). With TRMM, the TP precipitation is around 462.7 ± 24.4 mm yr −1 from 1998 to 2018. With ERA-I, this value is extremely large as 813.3 ± 50.7 mm yr −1 ( figure 1(b)). The ERA-I series also shows a strong declined trend, while there is no obvious trend for TRMM. The two series show similar variabilities with the correlation parameter between the two linearly detrended series being 0.69 at the 0.01 significant level, implying that the precipitation variability of ERA-I is somewhat reliable. one major difference is that WAM-2layers tracks the moisture of real precipitation fallen on the ground, while the Lagrangian methods track the moisture released in the air, which is not the actual precipitation observed on the ground (Huang and Cui 2015, Chen et al 2019). Besides, WAM-2layers is able to keep the 'tagged water' conserved (Zhang et al 2017). In other words, all the precipitated moisture can be tracked back to the evaporative sources. These features make WAM-2layers more suitable for groundprecipitation-based research. The tracking algorithm is introduced in the supplementary material.

WAM-2layers and experiment design
Two groups of experiments are performed. The first group is entirely based on the ERA-I data, referred to as ERA-Suite. The advantage of this suite is that the inner water cycle within ERA-I is more selfconsistent. The second group is based on observationconstrained data, i.e. the atmospheric data of ERA-I and surface fluxes of TRMM precipitation and GLDAS/OAFlux evaporation. Compared with observations, ERA-I precipitation over the TP is far too abundant (Tong et al 2014; figure 1(b)). Thus, the first group is regarded as supplementary. Limited use is made of it such as providing the variabilities and uncertainties. The second group is primary and more preferred. It is also referred to as the observation group. The data processing details for the two groups are put in the supplementary material.

Variability contribution index
The index proposed by Ahlstrom et al (2015) is applied to evaluate the variability contribution of the annual and seasonal moisture from different subregions to the annual precipitation of the TP. It is expressed as where x jt is the moisture anomaly (departure from a long-term trend) for subregion j at time t (in years), and X t is the TP precipitation anomaly (X t = Σ j x jt ). f j can be referred to as the average relative anomaly x jt /X t for subregion j, weighted with the absolute precipitation anomaly |X t |. Subregions with higher and positive scores are considered as contributing more in governing the TP IAV, as opposed to subregions with smaller or negative scores. It enables a comparison of their relative importance in governing the IAV (Ahlstrom et al 2015).
Besides, the results are generally attached with one standard deviation to indicate the uncertainty.

Annual moisture contribution
The annual moisture contribution (figure 2) indicate that influence from the west is long and wide. The contribution intensity from the west tends to be mild but extended. The contribution from the southwest is substantial and intense. Moisture from the west is mainly transported by the westerlies, while moisture from the Southwest is basically transported by the Indian monsoon. The circulation turning along the east coast of Africa leaves a long tail of moisture trace over the western Indian Ocean. In contrast, the contribution from the southeast is narrow and short in distance. This region is seated downwind of both the Indian monsoon and the westerlies. Moisture downwind is hard to influence the upwind region directly. However, in summer, when the EASM forms, moisture there has a chance to be transported upwind to the Plateau by the EASM. It is also worth noting, sources from the north barely have any influence on the TP.
Moisture distribution generally follows the rule that the nearer of the source in the upwind, the more it contributes. Within the target region, the core of moisture contribution is seated along the south side of the TP. This area is humid as opposed to the dry north (Yao et al 2013). The humid area produces more evaporation. The more it evaporates, the more it is recycled. Besides, moisture evaporated from the south is more likely to be recycled as local precipitation as it stays longer in the Plateau along with the southwesterly flux, while moisture from the north is quickly transported outside. These spatial features are generally consistent with those using ERA-Suite.
Keys et al (2014) proposed a concept of 'precipitationshed' , i.e. atmospheric watershed, where the target's precipitation is basically originated from within.
Following this idea, this study set a threshold of 5.0 mm yr −1 and extracted grids with higher moisture contributions that contribute around 80.0% of the TP precipitation. This watershed scope is robust in the control experiment, as it contributes around 82.3% of the TP precipitation using ERA-Suite. To further quantify the contribution from different sources, the precipitationshed is divided into four subregions according to respective dominating circulations, i.e. the West representing the westerlies, the Southwest representing the Indian monsoon, the Southeast representing the EASM, and the TP representing the local circulation ( figure 2(c)).
Moisture contribution series ( figure 3(a)) indicate that the Southwest contributes the most of about 147.6 ± 13.0 mm yr −1 (in water-depth of the TP). The TP, the West, and the Southeast follow by contributing 109.5 ± 13.1, 100.6 ± 7.5, and 12.3 ± 2.8 mm yr −1 , and accounting for 31.9 ± 1.9, 23.6 ± 2.3, 21.8 ± 1.5, and 2.6 ± 0.6%, respectively, of the annual precipitation. There are some quantitative evaluations on the sources for some particular regions over the TP, such as the northern TP (Zhang et al 2019a) and the Sanjiangyuan Region (Zhang Y et al 2019). The results differ in the largest moisture source. This is not hard to comprehend as they are influenced by different climates. For the northern TP where the westerlies prevail, the westerlies are the main moisture source. For the southern TP, the monsoon dominates and provides the most  moisture. As the TP precipitation decreases rapidly from south to north (figure S1 (available online at https://stacks.iop.org/ERL/15/104003/mmedia)), precipitation in the south weighs much larger than that in the north. When the TP is put as a whole, it is no wonder that the largest source is still the monsoon region, i.e. the Southwest representing the Indian monsoon.
There are no apparent trends in moisture contributions from the West and Southeast. Moisture contributed from the Southwest decreases significantly at a rate of −10.6 mm yr −1 dec −1 , while moisture contributed from the local increases significantly at 12.1 mm yr −1 dec −1 . The two items offset the moisture change trend, resulting in a weak trend in the precipitation. Since the ERA-I precipitation trend differs too much from TRMM, the moisture trend using ERA-Suite is not analyzed. For the TP precipitation IAV, the Southwest and the TP contribute the most, which account for 36.6% and 31.7%, respectively. The West makes an unparalleled contribution of 17.3% in comparison with the TP. The Southwest contributes only 2.9%. The results are generally consistent with those using ERA-Suite.

Seasonal moisture contribution
As circulations affect the TP differently in different periods of the year, it is necessary to explore the seasonal change in moisture contribution. As winter precipitation in the TP accounts for only 4.3% of the annual precipitation, it is thus omitted for analysis. The seasonal change in moisture contribution for the TP precipitation is shown in figure 4. Moisture varies remarkably with seasons. In spring, moisture for the precipitation mainly comes from the west by the westerlies. The West contributes around 38.6 ± 2.9% of the spring precipitation. Moisture contributed from the south is somewhat limited. In summer, both the contributions from the west and south are strengthened. Moisture from the south bursts due to the breakout of the Indian monsoon. Contribution from the Southwest reaches the highest among the subregions in all seasons, which is 91.1 ± 11.5 mm JJA −1 accounting for 34.6 ± 2.6% of the summer precipitation. The EASM also acts actively as contributing more moisture from the southeast. In autumn, moisture contributed from the west weakens substantially, while the precipitation moisture comes mainly from the south with the Southwest contributing 23.9 ± 3.0% of the autumn precipitation.
Moisture from the subregions varies with seasons, so does the moisture IAV from 1998 to 2018. To be more specific about the seasonal contributions in the annual precipitation IAV, the subregions' moisture IAV is deposed at the seasonal scale. Moisture from the Southeast explains only 2.9% (2.4% with ERA-Suite) of the precipitation IAV, and it is not considered. As a result, figure 5 shows the variability contributions from the West, Southwest, and the TP in spring, summer, and autumn. In general, the sources explain the variability most in summer except for the West, where there is no distinct difference between spring and summer. The Southwest in summer contributes the most of 27.5%, followed by the TP of 22.1%. The results with ERA-Suite also indicate that the Southwest and TP in summer are the largest two in contributing to the precipitation IAV among the sources in different seasons.

Validation of TRMM
The TRMM 3 h precipitation product of 3B42 (V6) was once evaluated over the TP (Tong et al 2014). They found that 3B42 was among the best estimates over the TP basins, just slightly worse than the gauged-based products. The 3B43 (V7), which is the monthly sum of 3B42 (V7), is applied in this study. To be more cautious about its applicability over the Plateau, a comparison is made between TRMM-3B43 and a gauge-based precipitation dataset by the China Meteorological Administration (CMA, Zhao et al 2014 ). The CMA precipitation is monthly at 0.5 • × 0.5 • grids, which is available from 1961 to 2017. The precipitation stations are densely distributed in the east while they drop to nearly none in the west ( figure 6(a)). Thus, the east part of the TP (ETP) is chosen for comparison over the overlapped period of 1998-2017. The ETP precipitations are 676.6 ± 40.4 for TRMM and 729.8 ± 56.4 mm yr −1 for CMA. The TRMM precipitation is a little smaller, which accounts for 92.9 ± 2.9% of the CMA precipitation. In light of that there are also biases in gauge measurement due to losses from wind or others (Goodison et al 1998), a difference between two gauge-based precipitation products of over 10% would be common depending on whether gauge corrected (Adam andLettenmaier 2003, Tong et al 2014). As a satellite product that the difference from the ground observation is less than 10%, it is better than average. In addition, the correlation coefficient of the two series ( figure 6(b)) is 0.92, which further demonstrates TRMM's excellent performance. If TRMM performs well over the gauged TP, it is inferred that it also does well over the ungauged TP due to similar retrieval and calibration algorithm.

Increased precipitation recycling ratio
The precipitation recycling ratio is defined as the contribution of local evaporation to local precipitation (Brubaker et al 1993, Eltahir andBras 1996). It is an important indicator that measures the potential of interactions between land surface processes and atmospheric processes (Goessling and Reick 2011). During the study period, the ratio shows a strong increasing trend (figure 7(a)), indicative of a strengthened role of local moisture. The recycling ratio is expressed as r = P loc /P, where P loc is recycled precipitation, i.e. local moisture contribution from evaporation. When P keeps stable, the increased r must be caused by the increased P loc . The recycled precipitation shows an increasing trend of 12.1 mm yr −1 dec −1 at the 0.01 significant level ( figure 7(b)). If precipitation keeps steady, the increased local contribution may be due to intensified local evaporation, i.e. the more it evaporates, the more it contributes. The evaporation series from GLDAS as presented in figure 7(b) indeed shows a significant increasing trend. However, is this increasing trend reliable? Under global warming, the temperature over the TP increases more rapidly due to the high elevation effect (Pan andLi 1996, Yang et al 2014). More glaciers and snow are being melted and retreated, thus releasing more liquid water (Li et al Besides, due to more water and proper temperature, the TP is becoming 'greener' (Zhong et al 2019, Li et al 2019. All these evidences support an increased evaporation/evapotranspiration over the TP, just as the GLDAS evaporation series reflects.

Decreased moisture contribution from the Southwest
A significant decreasing trend in moisture contributed from the Southwest is observed during the study period ( figure 3(a)). To further explore the decrease, the annual trend is deposed into seasonal. From spring to autumn, moisture from the Southwest decreases at rates of −2.4, −7.2, and −1.5 mm yr −1 dec −1 , respectively. Moisture decrease in summer accounts for a major part of the annual decrease of 67.9%. Thus, a specific focus is put on the moisture change in summer. As precipitation during the study period remains stable, the reduced moisture contribution from the Southwest is probably caused by the weakened moisture transport by the Indian monsoon. To prove this, an index is constructed as the moisture transported from the Southwest into TP at the adjacent boundary. The standardized moisture transport index is shown in figure 8 along with the standardized moisture contribution from the Southwest. The transport index shows similar fluctuations with that of the moisture contribution. The correlation parameter of the two linearly detrended series is 66.6% at the 0.01 significant level. Besides, the index decreases as the Southwest moisture contribution at the 0.1 significant level. It is worth noting that moisture transport is not the only factor that guarantees precipitation. Other factors such as coupling with the local system, air uplifting, are also indispensable for precipitation (Gustafsson et al 2010). Thus, the moisture transport index cannot be in a full linear relationship with the contributed precipitation. That said, the reduction in moisture transport is considered to be the main reason for the decreased moisture contribution from the Southwest.

Summary and conclusion
Moisture sources for precipitation over the whole TP are identified and assessed by WAM2layers using the TRMM precipitation and other observation-based data. Through comparison with the control experiment, these findings are considered to be important.
(a) The Southwest subregion is the largest moisture contributor to the TP, which contributes around 147.6 ± 13.0 mm yr −1 in waterdepth of the TP in climatology, followed by the TP, the West, and the Southeast, contributing 109.5 ± 13.1, 100.6 ± 7.5, and 12.3 ± 2.8 mm yr −1 , respectively. They account for 31.9 ± 1.9, 23.6 ± 2.3, 21.8 ± 1.5, and 2.6 ± 0.6%, respectively, of the annual precipitation. Circulations dominate the TP and the precipitation according to the seasons. In spring, the westerlies prevail over the TP and the West subregion contributes the most moisture among the subregions, which accounts for 38.6 ± 2.9% of the spring precipitation. In summer, the contribution from the Southwest amplifies due to the burst out of the Indian monsoon, which equals 91.1 ± 11.5 mm JJA −1 , and accounts for 34.6 ± 2.6% of the summer precipitation. In autumn, the Southwest contributes the most moisture that accounts for 23.9 ± 3.0% of the autumn precipitation.  a rate of −10.6 mm yr −1 dec −1 , while moisture contributed from the local increases significantly at 12.1 mm yr −1 dec −1 . The two major sources counteract each other to result in a trivial trend in annual precipitation. Further analyses reveal that the local increase in moisture contribution and recycling ratio are mainly caused by intensified evaporation of the TP. The decrease in moisture contribution from the Southwest is mainly caused by decreased moisture transport from the Indian monsoon in summer.