Discharge regime and simulation for the upstream of major rivers over Tibetan Plateau


  • Leilei Zhang,

    1. State Key Laboratory of Hydrology-Water Resources and Hydraulic Engineering, Hohai University, Nanjing, China
    2. Key Laboratory of Tibetan Environment Changes and Land Surface Processes, Institute of Tibetan Plateau Research, Chinese Academy of Sciences, Beijing, China
    Search for more papers by this author
  • Fengge Su,

    Corresponding author
    1. Key Laboratory of Tibetan Environment Changes and Land Surface Processes, Institute of Tibetan Plateau Research, Chinese Academy of Sciences, Beijing, China
    • Corresponding author: F. Su, Key Laboratory of Tibetan Environment Changes and Land Surface Processes, Institute of Tibetan Plateau Research, Chinese Academy of Sciences, Beijing 100101, China. (fgsu@itpcas.ac.cn)

    Search for more papers by this author
  • Daqing Yang,

    1. National Hydrology Research Center, Environment Canada, Saskatoon, Saskatchewan, Canada
    Search for more papers by this author
  • Zhenchun Hao,

    1. State Key Laboratory of Hydrology-Water Resources and Hydraulic Engineering, Hohai University, Nanjing, China
    Search for more papers by this author
  • Kai Tong

    1. Key Laboratory of Tibetan Environment Changes and Land Surface Processes, Institute of Tibetan Plateau Research, Chinese Academy of Sciences, Beijing, China
    Search for more papers by this author


[1] The hydrological regimes for the major river basins in the Tibetan Plateau (TP), including the source regions of the Yellow (UYE), Yangtze (UYA), Mekong (UM), Salween (US), Brahmaputra (UB), and Indus (UI) rivers, were investigated through a land surface model and regression analyses between climate variables and runoff data. A hydrologic modeling framework was established across the TP to link the Variable Infiltration Capacity (VIC) land surface hydrology model with a degree-day glacier-melt scheme (VIC-glacier model) at a 1/12° × 1/12°. The model performance was evaluated over the upper basins of the six rivers. The heterogeneity and scarcity of the meteorological stations are the major limitation for hydrological modeling over the TP. The relative contributions to streamflow from rainfall, snowmelt, and glacier melt for the six basins were quantified via the model framework and simulation. The results suggest that monsoon precipitation has a dominant role in sustaining seasonal streamflow over southeastern regions, contributing 65–78% of annual runoff among the UYE, UYA, UM, US, and UB basins. For the UI, the runoff regime is largely controlled by the glacier melt and snow cover in spring and summer. The contribution of glacier runoff is minor for the UYE and UM (less than 2% of total annual flow), and moderate for the UYA and US basins (5–7% of yearly flow), while glacier melt makes up about 12% and 48% of annual flow for the UB and UI basins, respectively.

1 Introduction

[2] The Tibetan Plateau and its surrounding regions (TP, including the Himalayas), also known as the Third Pole [Qiu, 2008] and the “roof of the world,” with an average elevation of over 4000 m above sea level (asl) and a total area of about 2.5 × 106 km2 [Zheng et al., 2001], are the most extensive highland and highest plateau in the world. The TP is characterized by extensive snow cover, glaciers, permafrost, and mountain lakes. About 67% of the plateau is covered by permafrost [Zhou et al., 2000], and the glacier area (about 91,822 km2) [Kotlyakov et al., 2012] over the TP is the third largest on the earth, after the Arctic/Greenland and Antarctic regions (1.7 × 106–12.3 × 106 km2) [Lemke et al., 2007]. As a unique geological and geographical unit, the TP exerts a profound influence on the East Asian and global climate through mechanical and thermal forcing mechanisms [Kutzbach et al., 1993; Yanai et al., 1992; Zheng and Li, 1999]. Given the uniqueness of the area, the TP has been designated as an experimental region for the Global Energy and Water Cycle Experiment (GEWEX) — the Asian Monsoon Experiment on the Tibetan Plateau (GAME/Tibet, 1996–2000) [Ma et al., 2003] and the Coordinated Enhanced Observing Period (CEOP) Asia-Australia Monsoon Project on the Tibetan Plateau (CAMP/Tibet, 2001–2006) [Ma et al., 2006].

[3] The TP is also the source of many major Asian rivers (Figure 1), e.g., Brahmaputra (Yaluzangbu), Salween (Nu), Mekong (Lancang), Yellow, and Yangtze Rivers, and is considered to be the water tower of Asia [Immerzeel et al., 2010]. These rivers sustain the ecosystems to support hundreds of millions of people living downstream. Studies of meteorological observations, reanalysis data, and ice core records have suggested a warming trend (0.16°C/decade–0.36°C/decade) over the TP in recent decades [Frauenfeld et al., 2005; Liu and Chen, 2000; Thompson et al., 2000; Wang et al., 2008; Wu et al., 2007; Xu et al., 2008; Yao et al., 2000]. Global climate models also project a strong warming of 3.8°C over the TP by the end of the 21st century, which is well above the projected warming for the global mean (2.5°C) [Christensen et al., 2007]. Along with the rising temperature, major climate-induced changes have occurred across the TP, such as glacier melt [Yao et al., 2004, 2007, 2012; Bolch et al., 2012], permafrost degradation [Cheng and Wu, 2007; Wu et al., 2005], and lakes changes [Lu et al., 2005; Wang et al., 2011; Wu and Zhu, 2008].

Figure 1.

Topography of the Tibetan Plateau (TP) and boundaries of six source river basins.

[4] Glaciers across the TP release large amount of water in summer to many rivers, especially the Indus and Brahmaputra basins with glacier contributing up to 45% of the river flow [World Resources Institute, 2003]. Glacier melt is an important hydrologic process in cold regions, as glaciers significantly modify streamflow regime, including the quantity, timing, and variability of flows over space and time [Kaser et al., 2003; Jansson et al., 2003]. Changes in temperature and precipitation are expected to seriously affect the melt characteristics of mountain glaciers [Immerzeel et al., 2009; Barnett et al., 2005]. In a warming climate, glacier melt will be enhanced due to the disappearance of fresh snow and decline of the albedo [Barnett et al., 2005]. The glacier melt especially affects streamflow in warm and dry seasons [Kaser et al., 2010] and will increase the water supply to downstream rivers for a given time period [Yao et al., 2004]. However, this trend is unstable and may reverse when glaciers continuously retreat [Yao and Yao, 2010]. Over a long time period, mountain glaciers will disappear and lead to a shortage in river flow in the summer season [Immerzeel et al., 2009, 2010], posting serious and unprecedented threats to water resources for TP and downstream regions [Kehrwald et al., 2008].

[5] Climate changes may exert substantial impacts to the hydrological cycle over the TP. However, the relevance and contribution of snow and glaciers to the major TP rivers remain largely unknown [Immerzeel et al., 2010; Kaser et al., 2010]. Quantitative assessment of the impacts of glacier change to streamflow is restricted mostly to modeling approaches due to the lack of direct large-scale observations. Most modeling studies have focused on branches of the upper Indus River, which mainly originate from the western TP region where snow and glacier runoff dominates river flow [Hasnain, 1999; Singh and Kumar, 1996, 1997; Singh et al., 1995, 2000]. Singh and Kumar [1997] used the University of British Columbia (UBC) watershed model to simulate different streamflow components and to examine hydrological response to climate change over the Spiti River (10,071 km2). Singh and Jain [2002] and Singh and Bengtsson [2004] applied a conceptual snowmelt model (SNOWMOD) to simulate flow components and predict the effects of warming climate on Satluj River hydrology (22,275 km2). Akhtar et al., [2008], using the conceptual HBV model (Hydrologiska Byråns Vattenbalansavdelning model) [Lindström et al., 1997], reported streamflow changes associated with climate change in three subbasins (area ranging from 3700 km2 to 14,000 km2) of the upper Indus River. Immerzeel et al. [2009] applied the Snowmelt Runoff Model (SRM) [Martinec, 1975] in the upper Indus basin above the Tarbela dam (200,677 km2) to study snowmelt runoff. These studies chiefly concerned basin hydrology over the upper Indus watershed. In the TP, a variety of basins exist with different climate and glacier conditions (Figure 1). Hydrological processes differ due to the difference in basin climatic and physical characteristics. Literature survey suggests a comprehensive and systematic hydrological study for the TP rivers is still lacking. Recently, Immerzeel et al. [2010] investigated the effects of climate change on the water supply for the upstream regions of the Indus, Ganges, Brahmaputra, Yangtze, and Yellow Rivers with the SRM model. This study is probably the first large-scale hydrologic simulations for most major TP basins.

[6] All the modeling studies above used conceptual models to investigate hydrological process. However, the physical basis of the Land Surface Models (LSMs) allows better understanding of the land surface-vegetation system and the interaction between land surface and atmosphere. In this study, we provide a broad picture of the hydrologic regimes for the major upstream TP rivers and quantify the runoff sources by a distributed land surface hydrologic model suitable for cold region processes. We establish a hydrologic modeling framework over the entire TP with an LSM and evaluate the model performance over the source regions of the Yellow (UYE), Yangtze (UYA), Mekong (UM), Salween (US), Brahmaputra (UB), and Indus (UI) Rivers (Figure 1 and Table 1). With this model framework, we aim to: (1) investigate runoff regimes and their linkage to climatic variables, 2) quantify the contributions of rainfall, snowmelt, and glacier melt to the streamflow of major TP river basins, and 3) assess the feasibility and limitation of land surface hydrology model in examination of major hydrologic processes over the TP.

Table 1. Characteristic of the Six Upstream River Basins in the Tibetan Plateau (TP)a
  1. a

    Annual precipitation and runoff are calculated for the period of streamflow observations.

  2. b

    The precipitation is from the APHRODITE data sets.

LocationLatitude (°)35.3033.0231.1130.5129.2734.92
Longitude (°)100.0997.1397.1196.1294.3472.88
Period of streamflow (Obs)1961–19991963–20051961–20001980–19851961–19991969–1997
Drainage area (km2)121,972137,70453,80067,740201,200162,000
Percent of total basin area (%)16.227.616.7720.9130.8913.91
Glacier area (km2)134.161308.19225.961151.584225.2015,325.2
Percent of drainage area (%)0.110.950.421.702.109.46
Discharge (m3/s)66940247367818452396
Percent of total discharge (%)37.711.332.9613.909.5636.30
Annual precipitation (mm)515333527607405425b
Annual runoff (mm)17492278362291470
Runoff ratio0.340.280.530.600.721.11

2 Study Region

[7] Our study area is limited to the domain between 68°–106°E and 22°–40°N, with a boundary defined as elevation above 2000 m asl (Figure 1). The TP is renowned for its numerous high mountain chains, with the Kunlun Mountains in the north, the Tangula Mountains in the central, the Hengduan Mountains in the east, the Gangdise Mountains stretching from the west across center toward southeast, and finally the renowned Himalayas on the far south margin (Figure 1). The elevation is gradually sloping from the northwest downward the southeast. Vegetation types vary gradually from forest in the southeast region to temperate shrubland/meadow in the middle region, to temperate desert, alpine desert, ice/polar desert in the northwest region [Cui and Graf, 2009].

[8] The TP climate is characterized by a wet and warm summer and a cool and dry winter. Winter temperatures exhibit large spatial variation, ranging from around −25°C in the west to −15°C in the east. In summer, mean temperature over the entire TP rises above 0°C. The east parts warm up to 5–10°C, and the west regions with higher elevations reach to 0–5°C [Cui and Graf, 2009]. Annual precipitation has an east-west gradient, ranging from approximately 1500 mm in the southeast to less than 100 mm in the northwest [Tong et al., 2013]. In the summer months, the south-east monsoon produces heavy precipitation, predominantly in the southeast region. The monsoon weakens in a western direction. In the western part, westerlies winds bring winter precipitation, mostly in the form of snow [Rees and Collins, 2006].

[9] The six selected upstream basins occupy 744,000 km2 and cover 30% of total TP (Table 1). Four basins (UYE, UYA, UM, and US) are located in the monsoon-dominated southeast. The UI lies in the west and largely affected by the westerlies and the UB in the south affected by both monsoon and the westerly system (Figure 1). Among the six basins, the UI has the largest glacier coverage (about 15,325 km2, close to 9.5% of the basin area), and the UYE has the smallest ice area (about 134 km2 and 0.11% of the UYE basin; Table 1). The UB has the second largest glacier area (4225.20 km2), about 2.1% of basin (Table 1), while the UYA, UM, and US have the moderate glacier fraction (0.95%, 0.42%, and 1.7% of the basins, respectively).

3 Model Implementation and Data Sources

3.1 Model Implementation

[10] The hydrology model used in this study is the Variable Infiltration Capacity (VIC) model. VIC is a grid-based land surface scheme [Liang et al., 1994, 1996], which parameterizes the dominant hydrometeorological processes taking place at the land surface-atmosphere interface. The model solves both surface water and energy balances over a grid mesh. The VIC model uses a mosaic representation of land cover and the variable infiltration capacity curve to account for subgrid heterogeneity in saturated extent, meaning that the soil does not need to be saturated for the entire grid before runoff generation. It assumes that surface runoff from the upper two soil layers is generated by those areas for which precipitation, when added to soil moisture storage at the end of the previous time step, exceeds the storage capacity of the soil. Base flow was generated according to an empirical nonlinear soil moisture relationship [Liang et al., 1994]. Surface runoff and base flow for each cell were routed to the basin outlet through a channel network [Lohmann et al., 1998]. The critical elements in the model that are particularly relevant to cold land implementations include: (1) a two layer energy balance snow model [Cherkauer and Lettenmaier, 1999; Storck and Lettenmaier, 1999] which represents snow accumulation and ablation and an overlying forest canopy, where present, and (2) a frozen soil/permafrost algorithm [Cherkauer and Lettenmaier, 1999; 2003] that solves for soil ice contents within each vegetation type and represents the effects of frozen soils on the surface energy balance and runoff generation. The updated VIC model has been applied and extensively evaluated in many large Arctic watersheds [Su et al., 2005]. Unfortunately, the current version of the VIC model does not have a glacier-melt module.

[11] Glacier-melt models generally fall into two categories: energy balance models to calculate melt as residual in the heat balance equation, and temperature-index models to quantify an empirical relationship between air temperature and melt rate [Hock, 2003]. Temperature index models have been the most common approach for snow and ice melt modeling due to wide availability of air temperature data and computational simplicity. Many studies have used a conceptual hydrological model coupled with a degree-day model to analyze the hydrological process [Schaefli et al., 2005; Huss et al., 2010] and determine the effect of climate change on streamflow [Stahl et al., 2008] in diffident regions/basins from Switzerland to Canada. In this study, a simple degree-day glacier algorithm [Hock, 2003] was linked with the VIC model to simulate glacier runoff. The VIC model was set up over the entire TP at a 1/12° × 1/12° spatial resolution.

[12] When there is a snowpack on the glacier, the snow melts first before the glacier starts melt, with the same degree-day approach but different degree-day factors. The temperature over the glacier area within each grid was adjusted by a commonly used temperature lapse rate (0.65°C/100 m). The degree-day factors of snow and ice for the UYE, UYA, UM, US, and UB were adopted from Zhang et al. [2006] based on the glacier observations in the TP. Snow factor is 4.1 mm°C−1 day−1 for most basins, and the ice factors are 4.7 mm°C−1 day−1 for UYE, 7.1 mm°C−1 day−1 for UYA and US, 13.8 mm°C−1 day−1 for UM, and 9.0°C−1 day−1 mm for UB, respectively. For the UI, the snow factor is set to be 4.0 mm°C−1 day−1, and ice factor is 7.0 mm°C−1 day−1 according to Immerzeel et al. [2010]. The glacier calving and debris cover are not taken into account in our model.

[13] The total runoff including the glacier meltwater from each grid can be calculated as:

display math(1)

where R(i) is the total runoff (mm) in grid i; f is the glacier area fraction in grid i; Rvic is the runoff (mm) in grid i calculated from VIC for the ice-free area; and Mi is the melt water (mm) for the glacier area in grid i. The VIC model combined with the glacier algorithm is referred as “VIC-glacier” model thereafter.

[14] Glacier area affects the estimation of melt water. In this paper, we use the volume-area scaling relation [Bahr et al., 1997; Radić et al., 2007, 2008; Radić and Hock, 2010] to derive glacier volume at the regional scales. We use an exponential form [equation (2)], derived from the glacier observation in the western China [Liu et al., 2003], to convert the glacier area to volume for each basin:

display math(2)

where V is glacier volume and S is glacier area. The initial ice volume for each basin was determined using glacier area from the glacier distribution dataset (See section 3.2 for the data source). This volume was updated every 10 years with the snowfall accumulation and simulated ice melt from all the glacier cells for each basin. The updated glacier area was determined by inverting equation (2) with the updated ice volume. This procedure was repeated for all the VIC-glacier simulation periods.

[15] The parameters most often adjusted during calibration of the VIC model include, the infiltration parameter (b_inf), the depth of the first and second soil layers (d1, d2), and three base flow parameters (Ds, Ws, Dsmax) [Nijssen et al., 2001; Su et al., 2005]. The parameter b_inf, with a typical range of 0–0.4, defines the shape of the variable infiltration capacity curve; and an increase of b_inf tends to enhance runoff production, while a decrease of b_inf tends to reduce runoff. The maximum moisture storage capacity is dynamically determined by the change in soil thickness. The thicker soil depths have higher moisture storage capacity, thus less runoff production. Three base flow parameters determine how quickly the water storage in the third layer evacuates and are generally less sensitive than parameters b_inf and d2. The topsoil layer depth (d1) for each grid cell was set to 5–10 cm after Liang et al. [1996], which indicated that a thin top layer significantly improved evapotranspiration predictions in the arid climates. The three base flow parameters and the third layer depth (d3) described in Nijssen et al. [2001] were used with minor adjustment during the calibration. Therefore, only the infiltration parameter (b_inf) and the second soil depth (d2) were targeted for intensive calibration in this study. More information on the typical range of these parameters can be found from http://www.hydro.washington.edu/Lettenmaier/Models/VIC/Documentation/CalibrateSoil.shtml.

3.2 Data Sources

[16] The data and information of the land surface characteristics required by the VIC model include soil texture, topography, and vegetation types. The meteorological inputs to the VIC model are daily precipitation, maximum temperature (Tmax), minimum temperature (Tmin), and wind speed. Vapor pressure, incoming shortwave radiation, and net longwave radiation are calculated from daily temperature and precipitation [Kimball et al., 1997; Thornton and Running, 1999], which have also been utilized in earlier application of the VIC model [Maurer et al., 2002].

[17] The daily metrological data from 176 stations over the TP and adjacent regions (Figure 1) during 1961–2009 were collected and quality controlled by the China Meteorological Administration (CMA). All station data were gridded to the 1/12° × 1/12° resolution by using the inverse distance weighting method. A common temperature lapse rate of 0.65°C/100 m has been applied during the interpolation from points to grids. For precipitation, the influence of topography on precipitation distribution has not been considered in the interpolation mainly due to lack of reliable precipitation lapse rate information. The model time step is 3 hourly. In the hydrological simulation, daily precipitation was apportioned equally to a 3 h time step, and temperature at each time step was interpolated by fitting an asymmetric spline function through the daily maximal and minimal. The 3 hourly wind speeds were made identical to the daily mean values.

[18] Most of the meteorological stations lie in the southern and southeastern TP, with very few in the middle and western parts (Figure 1). Also, all the 176 stations are located within China, and precipitation data for the UI are not available to this study. Therefore, we use the APHRODITE (Asian Precipitation-Highly Resolved Observational Data Integration Towards Evaluation of Water Resources) project (http://www.chikyu.ac.jp/precip/) [Yatagai et al., 2009, 2012] for the simulation for the international river basin UI. The APHRODITE is a daily gridded precipitation data set at 1/4° resolution. It is the only long-term (1961 onward) continental-scale daily product with a dense gauge network for Asia. In this analysis, the 1/4° APHRODITE precipitation data have been regridded to the 1/12° grids by the nearest neighbor approach.

[19] Topography data over the TP were obtained from GTOP30 (resolution: 1 km × 1 km) (http://eros.usgs.gov/#/Find_Data/Products_and_Data_Available/gtopo30_info). GTOP30 was used in this study to create digital river networks (flow direction file) at 1/12° × 1/12° grids over the study basins. A routing scheme [Lohmann et al., 1998] was run off-line using daily VIC surface and subsurface runoff as inputs to calculate streamflow at the basin outlets. The daily runoff was accumulated from the 3 hourly VIC outputs. Soil texture and vegetation information are from the same global data sets as used in Su et al. [2005].

[20] River discharge observations at the control stations of the selected river basins (Table 1) were used to evaluate the VIC model simulations. The discharge data at the Besham station for the UI River were collected from the Pakistan Water and Power Development Authority (WAPDA). Discharge data for all the other rivers (see Table 1 for data periods) were collected from Qinghai and Tibetan Hydrological Bureau, and these data can be regarded as the natural flow because water use is very small in quantity and can thus be ignored in these catchments.

[21] Glacier distribution data set was from the “Environmental & Ecological Science Data Center for West China” (http://westdc.westgis.ac.cn/data/ff75d30a-ee7d-4610-a5a3-53c73964a237) and the Randolph Glacier Inventory [Arendt et al., 2012] which is complete for UI and includes partly more recent data (http://www.glims.org/RGI/). These data are used for calculating the glacier fraction within each gird.

[22] The gridded precipitation from the 176 stations (CMA data hereafter) was used for calibration for all the basins except for the UI, where the APHRODITE data were used. The calibration is made manually, and two statistical criteria were used to evaluate the model performance: relative error (Er) and Nash-Sutcliffe efficiency (NSE)[Nash and Sutcliffe, 1970]. The NSE describes the prediction skill of the simulated streamflow as compared to observations.

4 Hydrological Regime

[23] It is useful to define the hydrological regimes and to understand their controlling mechanisms for the six upstream river basins (UYE, UYA, UM, US, UB, and UI, Figure 1). Long-term flow data were used to quantify the hydrological characteristics, and a regression analysis was conducted to investigate the linkage between climatic variables and river flow for these basins.

[24] Figure 2 shows the seasonal cycle of basin precipitation and streamflow. For the UI, with the influence of the westerly and monsoon, precipitation pattern is characterized with two peaks in winter months from November through March (NDJFM, about 37% of yearly total) and summer months from June to September (JJAS, 44% of annual total). However, more than 70% of the annual runoff occurs in the summer months (JJAS), with nearly 50% in July and August, and less than 10% in the winter months (NDJFM). A possible explanation for the inconsistency between runoff and precipitation in summer is that winter snowfall and snow cover and glacier melt contribute to the summer flows [Archer, 2003; Mukhopadhyay and Dutta, 2010].

Figure 2.

Mean monthly streamflow and precipitation for the six basins.

[25] For the other basins, 70–80% of annual precipitation falls in the summer months and less than 10% occurs in the winter months. The runoff pattern is similar to that for precipitation for these basins, with 60–80% of annual runoff in summer, and 10–15% in winter. The good correspondence between the streamflow and precipitation regimes suggests that the monsoon precipitation plays a dominant role in runoff generation over these basins.

[26] Regression analysis is performed between seasonal runoff and precipitation and temperature to investigate the main controlling factors in runoff generation over the regions. The time periods used in the analysis are the same as those of streamflow observations (Table 1). Correlation coefficients are shown in Table 2 for the six basins. The regression results with r > 0.3 and passing significance test (5% or 10%) are bold in Table 2. We found some interesting features from the correlation analysis:

  1. [27] Summer runoff (JJAS) is significantly correlated with summer precipitation (JJAS) for the UYE (r = 0.80), UYA (r = 0.80), UM (r = 0.85), US (r = 0.82), and UB (r = 0.88), but not for the UI (r = 0.29) which, in contrast, shows a significant positive correlation between summer runoff and summer mean temperature (r = 0.51).

  2. [28] Runoff in spring months April–May (AM) is significantly correlated with winter precipitation (NDJFM) for the UYE (r = 0.39) and UM (r = 0.36). For the US, the correlation coefficient between the spring runoff and winter precipitation is also positive (r = 0.52) but of low significance. Such correlations do not seem to exist for the UYA and UB (r = 0.23–0.27).

  3. [29] For the UI, there is a significant positive correlation between temperature and runoff (r = 0.43), and a significant inverse relationship between precipitation and runoff (r = −0.37) in April–May. The inverse relationship is likely due to spring clouds and precipitation at high altitudes that reduce energy input to snow ablation, thus slowing down the melt rate and river flow [Archer, 2003].

  4. [30] Unexpectedly, spring runoff has a significant negative correlation with spring temperature (r = −0.50) in the UYE. No significant correlations exist between runoff and temperature in spring for the UM, UYA, US, and UB.

  5. [31] Strong and somewhat puzzling inverse relationships occur between winter precipitation and both winter and summer runoff for the UM (r = −0.50, −0.37) and the UB (r = −0.51, −0.41). A significant correlation (r = 0.39) between spring precipitation (AM) and summer runoff (JJAS) for the UYA is also unexpected, given only 11% of annual precipitation occurring in April and May.

Table 2. Seasonal Correlation Between Runoff and Precipitation and Temperature for the Six Source River Basins in the TP (**: Significance 0.05; *: Significance 0.01)a
RegionAttributeClimate PeriodRunoff Period
  1. a

    Data period for each basin is the same as that of streamflow observation shown in Table 1. Note: JJAS represent months June–September, NDJFM represent November–March, and AM represent April and May.

AM0.27 0.30
AM−0.18 0.50**
AM0.39** 0.27
AM−0.26 −0.28
AM0.00 0.22
AM−0.00 −0.27
AM0.18 0.37
AM0.36 −0.34
AM0.07 −0.11
AM0.08 0.29
AM0.00 0.37*
AM−0.08 0.43**

[32] The results show that the runoff during June–September is affected by monsoon rainfall for the southeastern basins, including the UYE, UYA, UM, US, and UB. This is consistent with the good correspondence between precipitation and streamflow regimes for those basins (Figure 2). Spring (April–May) runoff for the UYE and UM, which accounts for 10–12% of the annual total flow, is influenced by winter (November–March) snow accumulation and spring melt process. However, the influence of glacier or snow melt on the flows over the southeastern basins (UYA, US, and UB) was not indicated from the regression results (Table 2).

[33] For the UI, spring and summer runoff largely depends on the energy input, represented by temperature for the concurrent seasons, suggesting that spring and summer flows in the UI are strongly controlled by snow and glacier melt. Despite 44% of annual precipitation occurring in summer season (JJAS), no good correspondence between summer runoff and precipitation was detected (r = 0.29). Archer [2003] suggests that, for the basins with large glacier cover, seasonal peak flows do not necessarily coincide with high precipitation, but with the combined availability of energy input and snow and ice storage.

5 Streamflow Simulations

[34] In this section, we present VIC-glacier model simulation and evaluate its performance with the monthly streamflow records at the basin outlets.

[35] Figure 3 compares the simulated and observed mean monthly streamflow for the six basins. The monthly time series of simulated streamflow are presented in Appendix A. Table 3 shows the values of the infiltration parameter (bi_inf), the second soil depth (d2), NSE, and Er for each basin. The NSE was based on monthly streamflow, and Er was based on the mean annual runoff. The period for the statistics varies among the basins, mostly depending on the length of streamflow data (five basins with flow records more than 28 years; Table 1).

Figure 3.

Mean monthly simulated and observed streamflow for the six source river basins in the TP. The data periods are the same as in Figure 2.

[36] The UYE accounts for about 16% of the Yellow River basin area, but it produces about 38% of the total annual runoff. The seasonal flows for the UYE exhibit a broad shape with flow increasing slowly in April and a slow recession after September. Flow regime is characterized by double peaks, with the first one in July and the second one in September (Figure 3a). The VIC-glacier model can generally reproduce the flow pattern; however, the model tends to overestimate the flow for June, July, and October (Figure 3a). The 39 year (1961–1999) monthly streamflow was reasonably simulated by the VIC-glacier model, with apparent overestimation of the summer flows especially for the low flow years (e.g., 1994–1998, Figure A1). The model efficiency based on the monthly streamflow (1961–1999) is 0.80 for the UYE, and the mean relative error is less than 1% (Table 3).

[37] The runoff contribution to the basin total flows is limited from the UYA (1%) and UM (3%) mainly due to large downstream areas, limited upstream precipitation, and wet monsoon-dominated downstream climates (Table 1). The streamflow of UYA and UM show similar seasonal patterns, with a quick rise in May and reaching a maximum in July, then recessing slowly to September (Figures 3b and 3c). The simulations in Figures 3b and 3c and Figures A2 and A3 cover the period 1963–2005 for the UYA (137,704 km2), and 1961–2000 for the UM (53,800 km2). The VIC-glacier simulation generally shows consistency with the observations at seasonal and monthly scales. The overall overestimation of the peak flows in UM results in a moderate model efficiency of 0.74 and a positive bias of 1.7%. The model has a negative bias of 1.7% for the UYA, mostly due to the large flow underestimation in the cold seasons.

[38] There were only six years of flow records for the US (1980–1985, Figure A4). The mean seasonal streamflow is characterized by a sharp rising from May and a single peak in July (Figure 3d). The simulated streamflow tends to overestimate the observations during August to October; however, the model closely simulated the timing and amount of the July peak flow (Figure 3d), resulting in a high model efficiency of 0.88 and a positive bias of 2.6% for the 6 years.

[39] The UI in the west of the TP has the largest glacier area of 15,325.2 km2 among the six basins. The long-term basin-average annual precipitation (425 mm) is less than the annual runoff (470 mm) (Table 1), suggesting other important sources of water (e.g., glacier melt) to sustain river flow. The seasonal flow pattern at the UI outlet (Besham station, 162,000 km2) is characterized by low flows (400–800 m3/s) from November to April, and a sharp peak in July (7442 m3/s), followed by a second peak in August (6567 m3/s). There is a good agreement between the flow simulation and observation for both monthly and seasonal hydrographs (1969–1997) (Figure 3f and Figure A5), resulting in a high efficiency of 0.86 and a positive bias of 3.6% for the annual flow.

[40] The UB, above the Nuxia station (201,200 km2), has the second largest glacier area (about 4225 km2 or 2% of the UB basin) among the six basins (Table 1). Based on the climate records (Figure 1), the basin-mean annual precipitation is 405 mm and annual runoff is 291 mm (Table 1). The peak flow in the UB occurs in August, consistent with the highest precipitation in July and August (Figure 2). Figure 3e and Figure A6 suggest that the simulated streamflow with the CMA data are greatly underestimated from 1961 to 1999, with a model efficiency of 0.64 and a large negative bias of 42%. One possibility of the underestimates is related to the model input. The analyses in section 4 suggest that annual runoff in the UB is mostly from monsoon rainfall and the annual streamflow is highly correlated with annual precipitation (r of 0.74). In another words, the streamflow simulation is highly dependent on the accuracy of precipitation input for the monsoon-dominated basins. The model performance cannot considerably improve through calibration over the UB. Therefore, the large underestimates in flow simulation are most likely due to the underestimation of precipitation over the basin. However, how much precipitation might be underestimated for the UB is actually unknown.

[41] To estimate the “real” mean annual precipitation for the UB, we assume a linear relationship between annual precipitation and simulated runoff (y = k*x + b, where y is annual runoff (mm), x is annual precipitation (mm), k and b are parameters to be determined). To obtain the parameters of k and b, we need two sets of variables of x and y. The corresponding simulated annual runoff (y) is 168 mm with the annual precipitation input (x) of 405 mm. We increase the annual precipitation amount by 20% and force the VIC-glacier model with the same parameters. Then we get another pair of variables, i.e., annual precipitation (x = 481 mm) and simulated runoff (y = 238 mm). The parameters k of 0.91 and b of −200 were obtained by solving the linear equation. The estimated “real” annual precipitation is about 540 mm by the linear equation for the observed runoff (y = 291 mm). We rescale the daily gauge-based grid precipitation (CCMA, hereafter) with the ratio of estimated to observed precipitation (540 mm/405 mm) to force the VIC-glacier model over the UB.

[42] The mean and monthly time series of simulated streamflow forced by the CCMA data for the UB at Nuxia are shown in Figure 3e and Figure A7. The Er is very small (−0.45%, Table 3) for the simulation mostly due to the cancellation of overestimation in summer seasons (June–September) and underestimation in cold seasons (October–March) (Figure 3e). The model efficiency is high (0.86), resulting from the good agreement between the simulated and observed monthly streamflow variations (Figure A7).

[43] For comparison, the VIC model simulations without the glacier model were also included in Figure 3 and Table 3. The simulated runoff without glaciers generally tends to underestimate flow (Table 3), although the difference varies among the basins. There is almost no difference for the UYE (Figure 3a), and the glacier effects are moderate for the UM, US, and UYA (Figures 3b–d), with the difference being 1.4–6.7%. The impacts of glacier on river flow became larger for the UB (Figure 3e), since simulated annual runoff with glaciers is higher by 11%. The largest difference between the two simulations was observed in the UI (almost 50%, Figure 3f). The importance of glacier contribution to the river runoff for each basin is further discussed in section 6.

[44] The seasonal patterns of simulated evaporation and temperature are shown in Figure 4 for the basins. Evaporation patterns are consistent with those of temperature for the eastern basins, with 72–80% of the total evaporation occurring in June–September. However, the ratio of annual evaporation to total precipitation is variable among the basins, with large ratios of 0.66–0.71 for the UYE and UYA, medium ratios of 0.52–0.55 for the UM and UB, and a low ratio of less than 0.4 for the US. For the UI, 60% of evaporation occurs in the summer season (June–September), and the annual evaporation only takes up about 37% of annual precipitation, suggesting that the glacier melt has consumed large portion of energy and little energy is left for latent heat.

Figure 4.

Seasonal patterns of simulated evaporation and temperature for the six basins. The periods covered by the data are the same as in Figure 2.

6 Flow Composition and Separation

[45] Evaluations of the VIC model performance show acceptable results for most basins, including the UYE, UYA, UM, and US. Based on this result, it is possible to discuss rainfall, snowmelt, and glacier-melt contributions to river flow for the selected basins.

[46] The seasonal and annual runoff components for the six basins are shown in Figures 5 and 6. The regime of rainfall runoff is consistent with that of total runoff for the southeastern basins (UYE, UYA, UM, US, and UB), especially during June–August (Figure 5). The rainfall runoff contributions range from 65% to 78% among the five southeastern basins (Figure 6), suggesting that the monsoon rainfall plays an important role over these basins. However, UI basin is different from other basins (Figure 5f). Rainfall runoff makes up about 18% of flow during June–August and contributes about 21% of yearly flow (Figure 6). These results are consistent with the regression analysis, i.e., summer runoff highly correlated with the monsoon precipitation for the UYE, UYA, UM, US, and UB, but no relationship found between summer precipitation and runoff in the UI (Table 2).

Figure 5.

Variable Infiltration Capacity (VIC)-glacier model simulated seasonal distribution of rainfall runoff, snowmelt runoff, glacier-melt runoff, and total runoff for the six basins. The periods covered by the data are the same as in Figure 2.

Figure 6.

Contribution of rainfall, snowmelt, and glacier-melt runoff to the total annual runoff for the six basins simulated with the VIC-glacier model.

Table 3. Values of the Infiltration Parameter (b_inf) and the Second Soil Depth (d2, m), and the Nash-Sutcliffe Efficiency (NSE) and Relative Error (Er) of the Simulated Monthly Streamflow With or Without Glacier Model (Variable Infiltration Capacity (VIC)-Glacier or VIC) Relative to the Observation for the Six Source River Basinsa
BasinHydrological Stationsbd2Mean Annual Runoff mmNSEEr (%)Model Used
  1. a

    The period involved in the statistics for each basin is indicated in Table 1.


[47] Snowmelt runoff for the UYE, UYA, UM, and US is characterized by two peaks in March–June and September–November, respectively (Figures 5a–d). The first peak is in May for the UYE and UYA, but in June for the UM and US; the second peak is in September except for the UYE in October. Snowmelt runoff accounts for 60–80% of total flow during March–May for the UYE, UM, UYA, and US. What is interesting is the simulated snowmelt in the autumn over the four eastern basins. Yang et al. [1993] found snowfall in summer and quantified its runoff contribution over the upper Urumqi river basin; they called this component “snowfall runoff.” Analysis of hourly precipitation and temperature records for the UYE suggests two snowfall peaks in May and October, respectively. Therefore, the second snowmelt peak in autumn simulated by the model is likely due to fresh snowfall, i.e., very short-term snow cover up to a few days, and it melts.

[48] Snowmelt runoff in the UB exists mostly during April–November, with the peak in August (Figure 5e). Similar to the other four southeastern basins, snowmelt in summer and autumn mostly comes from the snowfall in the same seasons, while the snow melt in spring months (before June) is associated with the accumulated snow during winter. The contribution of snow melt to the annual flow is about 23% for the UB (Figure 6). The seasonal pattern of snow runoff in the UI is different from other basins, with only one peak in June (Figure 5f). Snowmelt before June depletes both winter snowpack and spring snowfall. Snowmelt contribution to the annual flow is about 31% for the UI, which is much larger than the other basins (Figure 6).

[49] The calculated glacier runoff mostly occurs during June–September for all the basins, with the peak in July and August (Figure 5). The contribution of glacier melt to the annual runoff varies greatly among the basins, with about 1%, 7%, 1%, 5%, 12%, and 48% for the UYE, UYA, UM, US, UB, and UI, respectively (Figure 6). This result is consistent with the glacier area for each basin (Table 1). For the UI, glacier runoff during June–September alone accounts for 54% of the annual runoff. These results are similar to the previous regression analysis (Table 2), i.e., a high correspondence between the mean temperature and runoff in June–September for the UI. It is important to recall that the long-term annual runoff (470 mm) is larger than the yearly precipitation (425 mm) in the UI (Table 1). This imbalance suggests that the glacier melt is an important source to sustain the streamflow in the UI. On the other hand, glacier melt is much less important than precipitation in runoff production for the UYE, UYA, UM, and US.

7 Discussion on Model Uncertainties

[50] The results in Figure 6 are generally consistent with other studies [Kaser et al., 2010; Immerzeel et al., 2010]. They all suggest that glacier melt is extremely important for the Indus, and it plays a modest role over the southeastern TP basins (e.g., UYE, UYA, UM, US, and UB) with the monsoon domination. This analysis is different in the study domain and approach from Kaser et al. [2010] and Immerzeel et al. [2010]; they mostly used conceptual approaches to quantify the glacier contribution to downstream discharge. In this work, we apply a physically based LSM model to specifically focus on the source region of six major rivers in the TP. We provide a clear picture for the hydrologic regimes and quantify the runoff sources via VIC-glacier model simulations.

[51] Land surface hydrology model simulations are highly sensitive to the accuracy of the precipitation input. The VIC-glacier model shows an acceptable performance over the UYE, UYA, UM, US, and UI basins, with the annual bias of less than ±5% (Table 3). While for the UB, the model significantly underestimated streamflow by 42% (Table 3) when forced with the gridded gauge precipitation data (yearly total of 405 mm). We have carried out a separate study on the precipitation issues [Tong et al., 2013] and found that the precipitation estimates from different data sets for the UB have the largest variance among the study basins. We derived the mean annual precipitation for the UB (540 mm) by assuming a linear relationship between annual runoff and precipitation, and further validations are needed. Precipitation data uncertainties and its impact to glacier hydrology research, including melt water estimations, have not been sufficiently discussed in previous studies [e.g., Immerzeel et al., 2010; Kaser et al., 2010]. The heterogeneity and scarcity of meteorological stations over TP are probably the biggest limitation for hydrological modeling. Uncertainties in precipitation may also come from the biases of gauge measurements, such as wind-induced gauge undercatch, wetting and evaporation losses, and underestimation of trace precipitation amounts [Ye et al., 2004; Yang et al., 2005]. Ye et al. [2004] estimated an overall mean of 19% increase in yearly precipitation due to bias corrections at the 710 climate stations over China. A comprehensive program of precipitation measurement and analysis is necessary for the TP. It should include intensive observations at selected key locations to collect precipitation, temperature, humidity, wind speed, wind direction, and radiation data, especially in the west of UB. The network should also consider vertical gradient of temperature and precipitation in difference altitude zones. In addition, more investigations of satellite precipitation products for hydrological modeling [e.g., Su et al., 2011] over the TP are also useful.

[52] Model uncertainties may also result from parameters, such as temperature lapse rate (TLR), degree-day factor (D), and the parameters governing runoff generation, especially b_inf and d2 in the VIC model. Figure 7 shows the VIC model sensitivity for the parameters TLR and degree-day factor for ice (D) over the UI (Table 1). The change in glacier runoff is within 3% when the TLR changes from 0 to 1.1°C/100 m (Figure 7a). The model becomes less sensitive when TLR is above 0.5°C/100 m, with the values between 0.6 and 0.8°C/100 m producing the highest NSE and lowest Er (Figure 7b). The TLR used in this study (0.65°C/100 m) is within this range. The uncertainties resulting from TLR in the glacier runoff simulation should be within 3% for the other basins, since the UI receives the largest effects of glacier meltwater.

Figure 7.

Model sensitivity to the parameters of temperature lapse rate (TLR) and degree-day factor in terms of glacier runoff contribution, Nash-Sutcliffe efficiency (NSE), and relative error (Er) for the UI basin.

[53] Another important source of uncertainties in glacier runoff estimation is the determination of parameter D, which has considerably spatial and temporal variability, and no universal value exists [Hock, 2003]. Since, it is unknown how D may vary spatially and temporally, in our application, the degree-day factors represent the average conditions for each basin, and theoretically can be treated as calibration parameters [Hock, 2003]. The model is more sensitive to the degree-day factor than TLR, with about 5% increase in glacier runoff for every 1 unit (mm°C−1 day−1) increase in D (Figure 7c). The Er changes from −15% to 45% when D increases from 4.1 to 13.1 mm°C−1 day−1 (the range used among the six basins), and the range of NSE can reach 0.6 for the same D changes (Figure 7d). The D values between 6.1 and 7.1 mm°C−1 day−1 produce the highest NSE and lowest Er for the UI (Figure 7d). The D value we use in this work (7.0 mm°C−1 day−1) is within this range. Since the UI is the most glacier-affected basin, the sensitivity of D and TLR found here could be the upper limit for the basins.

[54] The VIC model parameters and their sensitivities have been extensively studied in other studies [e.g., Nijssen et al., 2001; Su et al., 2005; Xie et al., 2007; Yong et al., 2010]. In this work, the infiltration parameter (b_inf) and the second soil layer depth (d2) were identified as most sensitive among the VIC parameters for calibration. Figure 8 shows model sensitivity to the parameters b_inf and d2 in terms of NSE and Er for the UYE basin with the least glacier impact. The Er changes are within 4% when b_inf varies from 0 to 0.4 (Figure 8a), while Er is more sensitive to d2, with Er reaching −50% when d2 increases to 3.0 m (Figure 8b). Since all parameters were calibrated in terms of NSE and Er, the final parameters shown in Table 3 generally have the highest NSE and lowest Er. However, these parameters are highly dependent on the precipitation data used for calibration. When the precipitation input changes, the parameters may change accordingly in order to match streamflow data. Therefore, reliable precipitation input is the premise to acquire reasonable model parameters and simulations.

Figure 8.

Model sensitivity to the parameters of infiltration parameter (b_inf) and the second soil layer depth (d2) in terms of Nash-Sutcliffe efficiency (NSE) and relative error (Er) for the UYE basin.

[55] The glacier area vs. volume relationship used in this work was mostly derived from the field observations in the western China. It is not clear whether this relationship is suitable to all the study basins. More observations are needed to verify or test the area to volume scaling approach over the basins with different characteristics of glacier type.

[56] The current glacier scheme has a lower complexity than the other processes in the VIC model. Also, the off-line linkage between the degree-day module and the VIC model means a weak interaction between the glacier and nonglacier areas within each grid. However, this simple approach allows the modeling of large-scale river basins with limited meteorological data and basic glacier information. Estimation of glacier melt using the energy balance approach requires meteorological data, such as radiation, cloudiness, wind speed, etc., and such meteorological data are scarce in most TP regions. The VIC model inputs of vapor pressure, incoming shortwave radiation, and net longwave radiation were calculated from daily temperature and precipitation [Kimball et al., 1997; Thornton and Running, 1999]. For large river basins with sparse observation stations, the estimated meteorological forcings may not be accuracy to determine glacier melt with the energy balance approach. Our effort is currently under way to test a more sophysiticated glacier module in the VIC model over a small experimental basin in the TP.

8 Conclusion

[57] In this study, the hydrological regimes of six source rivers in the TP were investigated through regression analyses between climatic and hydrological variables. A hydrological modeling framework was established by linking the VIC land surface hydrology model with a degree-day glacier-melt scheme across the TP. The VIC-glacier model performance was evaluated over six basins, and the relative importance of the rainfall, snowfall, and glacier runoff to the total streamflow was quantified through model simulations. The main results are summarized as the follow:

  1. [58] Monsoon precipitation plays a dominant role in sustaining streamflow in the southeastern basins of TP, including UYE, UYA, UM, US, and UB. For the western basin of UI, the runoff regime is largely controlled by the melt of glacier and snowfall in summer and seasonal snowpack in spring. The contribution of rainfall runoff varies greatly among the basins in the monsoon season (June–September). Rainfall runoff accounts for 18% of summer flow (or about 21% of yearly runoff) for the UI, while it makes 68–90% of the summer runoff (or 55–60% of total flow) among the UYE, UYA, UM, US, and UB. Future precipitation changes therefore would irectly impact the timing and amplitude of streamflow for the monsoon-dominated basins, while temperature should be largely responsible for the seasonal changes in streamflow for the UI with the largest ice coverage.

  2. [59] The importance of glacier melt is low for the UYE and UM (less than 2% of annual runoff) and moderate for the UYA and US (5–7%). Glacier water becomes relatively more important for the UB (around 12% of annual flow), although this basin is mostly affected by the monsoon precipitation. Glacier water is extremely important for the UI, contributing approximate 48% of annual total runoff. This understanding will be very useful for the adaptation to hydrologic consequences of climate change across the TP.

  3. [60] The seasonality and importance of snowmelt differ greatly between UI and other basins. For the UI, 57% snowmelt (accounting for 31% of annual runoff), occurs in the pre- and early-monsoon season (April–June), and 39–43% snowmelt (20–23% of annual runoff) takes place in April–June for the UYE, UYA, UM, US, and UB. It is clear that snowmelt is more important for the UI than for the monsoon-dominated basins in water supply for the dry seasons (April–June). In addition to snow cover melt in spring, we found snowfall runoff contribution to autumn flows over the southeastern rivers.

  4. [61] Uncertainties exist in this study regarding data analysis and model simulations. The lack of reliable precipitation data creates major challenges and uncertainties in our model simulations and is probably the largest limitation for the hydrologic model applications over the TP, especially for the west and the middle part of the region. Model uncertainties associated with temperature lapse rate are generally within 3%, while the degree-day factor for ice melt may cause 5% increase in glacier runoff for every 1 unit (mm°C−1 day−1) increase. Our effort continues to improve land surface model performance in this important region.

Appendix A

[62] Figures A1A6 and A7 show monthly time series of simulated and observed streamflow with the various periods shown in Table 1 forced by the CMA and CCMA for the six basins.

Figure A1.

Monthly time series of simulated and observed streamflow for the Yellow River at Tangnaihai for 1961–1999.

Figure A2.

Monthly time series of simulated and observed streamflow for the Yangtze River at Zhimenda for 1963–2005.

Figure A3.

Monthly time series of simulated and observed streamflow for Mekong at Changdu for 1961–2000.

Figure A4.

Monthly time series of simulated and observed streamflow for Salween at Jiayuqiao for 1980–1985.

Figure A5.

Monthly time series of simulated and observed streamflow for the Indus at Besham for 1969–1997.

Figure A6.

Monthly time series of simulated and observed streamflow for Brahmaputra at Nuxia for 1961–1999.

Figure A7.

Monthly time series of simulated streamflow forced by the CCMA for the Brahmaputra at Nuxia for 1961–1999.


[63] This work was supported by the National Basic Research Program of China (973 program) (2010CB951702, 2010CB951101),National Natural Science Foundation of China (41190081, 41171051), and the Chinese Academy of Sciences “100-Talents” Program to the Institute of Tibetan Plateau Research, Chinese Academy of Sciences. It was also partly supported by the Fundamental Research Funds to the Central Universities (2010B13814). We would like to thank Richard Armstrong for his helpful comments and three anonymous reviewers for their constructive comments.