Runoff Regime, Change, and Attribution in the Upper Syr Darya and Amu Darya, Central Asia

Jingheng Huang aState Key Laboratory of Tibetan Plateau Earth System, Resources and Environment, Institute of Tibetan Plateau Research, Chinese Academy of Sciences, Beijing, China
cUniversity of Chinese Academy of Sciences, Beijing, China

Search for other papers by Jingheng Huang in
Current site
Google Scholar
PubMed
Close
,
Fengge Su aState Key Laboratory of Tibetan Plateau Earth System, Resources and Environment, Institute of Tibetan Plateau Research, Chinese Academy of Sciences, Beijing, China
bCAS Center for Excellence in Tibetan Plateau Earth Sciences, Beijing, China
cUniversity of Chinese Academy of Sciences, Beijing, China

Search for other papers by Fengge Su in
Current site
Google Scholar
PubMed
Close
,
Tandong Yao aState Key Laboratory of Tibetan Plateau Earth System, Resources and Environment, Institute of Tibetan Plateau Research, Chinese Academy of Sciences, Beijing, China
bCAS Center for Excellence in Tibetan Plateau Earth Sciences, Beijing, China
cUniversity of Chinese Academy of Sciences, Beijing, China

Search for other papers by Tandong Yao in
Current site
Google Scholar
PubMed
Close
, and
He Sun aState Key Laboratory of Tibetan Plateau Earth System, Resources and Environment, Institute of Tibetan Plateau Research, Chinese Academy of Sciences, Beijing, China

Search for other papers by He Sun in
Current site
Google Scholar
PubMed
Close
Free access

Abstract

The upper Syr Darya (USD) and Amu Darya (UAD) basins are the two biggest flow formation zones in central Asia and the only water supply sources for the Aral Sea. Upstream snow and ice reserves of those two basins, important in sustaining seasonal water availability, are highly sensitive and prone to climate change, but their importance and changes are still uncertain and poorly understood due to data scarcity, inaccessibility, harsh climate, and even geopolitics. Here, an improved forcing dataset of precipitation and temperature was developed and used to drive a physically based hydrological model, which was thoroughly calibrated and validated to quantify the contributions of different runoff components to total flow and the controlling factors for total runoff variations for 1961–2016. Our analysis reveals divergent flow regimes exist across the USD and UAD and an ongoing transition from nival–pluvial toward a volatile pluvial regime along with rising temperatures. Annual total runoff has weakly increased from 1961 to 2016 for the entire USD and UAD, while the subbasins displayed divergent flow changes. Spring runoff significantly increased in all the USD and UAD basins primarily due to increased rainfall and early snow melting, tending to shift the peak flow from June–July to April–May. In contrast, distinct runoff changes were presented in the summer months among the basins primarily due to the trade-off between the increase in rainfall and the decrease in snowmelt and glacier runoff. These findings are expected to provide essential information for policymakers to adopt strategies and leave us better poised to project future runoff changes in ongoing climate change.

© 2022 American Meteorological Society. For information regarding reuse of this content and general copyright information, consult the AMS Copyright Policy (www.ametsoc.org/PUBSReuseLicenses).

Corresponding author: Fengge Su, fgsu@itpcas.ac.cn

Abstract

The upper Syr Darya (USD) and Amu Darya (UAD) basins are the two biggest flow formation zones in central Asia and the only water supply sources for the Aral Sea. Upstream snow and ice reserves of those two basins, important in sustaining seasonal water availability, are highly sensitive and prone to climate change, but their importance and changes are still uncertain and poorly understood due to data scarcity, inaccessibility, harsh climate, and even geopolitics. Here, an improved forcing dataset of precipitation and temperature was developed and used to drive a physically based hydrological model, which was thoroughly calibrated and validated to quantify the contributions of different runoff components to total flow and the controlling factors for total runoff variations for 1961–2016. Our analysis reveals divergent flow regimes exist across the USD and UAD and an ongoing transition from nival–pluvial toward a volatile pluvial regime along with rising temperatures. Annual total runoff has weakly increased from 1961 to 2016 for the entire USD and UAD, while the subbasins displayed divergent flow changes. Spring runoff significantly increased in all the USD and UAD basins primarily due to increased rainfall and early snow melting, tending to shift the peak flow from June–July to April–May. In contrast, distinct runoff changes were presented in the summer months among the basins primarily due to the trade-off between the increase in rainfall and the decrease in snowmelt and glacier runoff. These findings are expected to provide essential information for policymakers to adopt strategies and leave us better poised to project future runoff changes in ongoing climate change.

© 2022 American Meteorological Society. For information regarding reuse of this content and general copyright information, consult the AMS Copyright Policy (www.ametsoc.org/PUBSReuseLicenses).

Corresponding author: Fengge Su, fgsu@itpcas.ac.cn

1. Introduction

The headwaters of the Syr Darya (USD) and Amu Darya (UAD) originate in the high mountains of the western Tien Shan, Pamir, and the Hindu Kush. The two rivers are the only sources of streamflow into the Aral Sea Basin, constituting a crucial freshwater source for central Asia (Micklin 2010; Unger-Shayesteh et al. 2013; Alford et al. 2015). The catchments comprise the six countries of Kyrgyzstan, Tajikistan, Turkmenistan, Uzbekistan, southern Kazakhstan, and small parts of Afghanistan. The high mountain domain of the river catchments is considered central Asia’s water towers for their importance as freshwater suppliers (Immerzeel et al. 2020), with about 60 million people in this geopolitically important region depending on water from rainfall and snowmelt and glacier melt in the USD and UAD for agriculture, hydropower generation, drinking, and other purposes (Jarsjö et al. 2012; Alford et al. 2015; Chen et al. 2016). The Syr Darya and Amu Darya have been one of the most complex transboundary rivers in the world, with recurrent international seasonal water allocation conflicts ever since the collapse of the former Soviet Union (Smith 1995; Siegfried et al. 2012; Bernauer and Siegfried 2012; Sorg et al. 2014a).

Central Asia’s climate is warming, with a mean annual temperature rise of 0.37°–0.43°C decade−1 from 1979 to 2011 (Hu et al. 2014; Yao and Chen 2015). Along with the rapid warming, significant climate-induced changes have occurred across the USD and UAD basins, such as glacier retreat (Hagg et al. 2013a,b; Sorg et al. 2014a,b), permafrost degradation (Marchenko et al. 2007; Schöne et al. 2012), and decrease of snowfall (Khan and Holko 2009), which may exert substantial impacts on the hydrological cycle of central Asia. Anthropogenic climate change is projected to bring considerable changes to both the timing and volume of river flow through rising temperatures, increased snowmelt and glacier melt, and a more variable rainfall regime (Siegfried et al. 2012; Immerzeel et al. 2012; Kure et al. 2013; Luo et al. 2018; Gulakhmadov et al. 2020; Shafeeque et al. 2020), potentially exacerbating the current water stress and seasonal water allocation conflicts between the upstream and downstream riparian countries (Siegfried et al. 2012; Sorg et al. 2012; Unger-Shayesteh et al. 2013; White et al. 2014; Russell 2018). The Aral Sea, which has already lost up to 90% of its pre-1960 volume caused by overexploitation of the regions’ renewable water resources for irrigation (Badescu and Schuiling 2010), may retreat further (Micklin 2016) and consequently could cause more environmental problems, such as soil pollution (Singer et al. 2003), land salinization (Spoor 1998; Saiko and Zonn 2000), and dust storms (Orlovsky et al. 2004). All these climate-induced changes highlight the importance of a better understanding of the processes governing surface flow in the region, which can offer important insights for regional ecological restoration, water resources management, and adaptation strategies downstream in the Aral Sea Basin.

However, the annual and seasonal hydrological regimes across the USD and UAD are poorly understood due to data scarcity, inaccessibility, harsh climate, and geopolitics (Unger-Shayesteh et al. 2013; Sorg et al. 2014a,b; Alford et al. 2015), hindering a comprehensive assessment of the future water availability in the region (Miller et al. 2012; Hagg et al. 2013a,b). Statistical approaches were widely used to study the runoff regime (e.g., annul runoff changes), while those studies provide a puzzle of contradicting statements (Nezlin et al. 2004; Khan and Holko 2009; Unger-Shayesteh et al. 2013). To close the knowledge gap, distributed cryospheric–hydrological models have been used as a feasible approach to quantify runoff regimes and the relative importance of rainfall, snowmelt, and glacier runoff to river flow when large-scale observations are absent (e.g., Zhang et al. 2013; Kan et al. 2018; Meng et al. 2019; Sun and Su 2020) and to assess hydrological responses in the upper mountainous basins of Tibetan Plateau to projected climate change (Lutz et al. 2014; Su et al. 2016; Luo et al. 2018; Khanal et al. 2021). However, quantitative modeling studies on climate and runoff changes are quite limited in the USD and UAD. Studies either primarily focused on small and highly glacierized upstream branches (Kure et al. 2013; Gan et al. 2015; Pohl et al. 2017; Radchenko et al. 2017; Shafeeque et al. 2020) or at regional-scale explorations with conceptual models, which usually deal with hydrological processes in a simplified way (Siegfried et al. 2012; Immerzeel et al. 2012; Hagg et al. 2013a; Kure et al. 2013; Alford et al. 2015; Tarasova et al. 2016; Armstrong et al. 2019).

Physically based hydrological models with a higher degree of process complexity allow for a better understanding of the interaction between the land surface and atmosphere and, therefore, are crucial to assess the hydrological impacts in high mountain basins with complex topography (Zhang et al. 2013; Ragettli et al. 2016). Modeling studies in the USD and UAD are hindered by the lack of reliable calibration data and model precipitation input. The latter is the key and fundamental forcing data of hydrological models and is still the most problematic meteorological parameter in the Tien Shan, Pamirs, and the Himalayas due to the scarcity of metrological observations (Refsgaard 1997; Immerzeel et al. 2014; Pohl et al. 2015; Tarasova et al. 2016; Kan et al. 2018; Wortmann et al. 2018). This is especially true in the USD and UAD (Duethmann et al. 2013; Hu et al. 2018), as the gauged data quality has deteriorated or measurements were discontinued after the collapse of the former Soviet Union in 1991, and most of the available precipitation gauge data are only at monthly scale (Unger-Shayesteh et al. 2013). Variations in input precipitation data have produced divergent and even contradictive results, resulting in highly uncertain hydrological regimes (Immerzeel et al. 2012; Duethmann et al. 2013; Gan et al. 2015; Wang et al. 2016; Khanal et al. 2021). In addition, different hydrological regimes coexist among the subregions of high mountain central Asia due to their different climatic and land surface conditions (Unger-Shayesteh et al. 2013). However, a comprehensive assessment of the regional differences in the hydrological characteristics and regimes across the USD and UAD is still lacking.

In this work, we attempt to provide a broad picture of the hydrologic regimes and quantify the runoff sources across the USD and UAD basins by using a well-established physically based cryospheric–hydrological model with ably controlled precipitation input. To allow more robust parameter estimation and further constrain the model uncertainty, we calibrate and validate the model in a two-step approach, with extensive runoff, snow cover, and glacier mass balance observation data. With this well-calibrated modeling framework, we specially aim to 1) identify the runoff regimes across the USD and UAD subbasins and quantify the contribution of the three major runoff components (glacier melt, rainfall, and snowmelt runoff) to total runoff and 2) investigate the role of runoff components and climate factors in historical runoff changes at both annual and seasonal scales in the USD and UAD basins. To the best of our knowledge, this work is the first to systematically examine the attribution of recent long-term runoff changes in timing and magnitude for all subbasins in this region. The following novel components are expected to advance our understanding of the hydrological process in the USD and UAD: 1) well-controlled and validated forcing data based on more gauges and 2) a well-developed and well-validated hydrological model, with extensive observed runoff data, snow cover, and glacier mass balance. We expect these findings will provide essential information for policymakers and water managers to adopt strategies and leave us better poised to project future runoff changes in ongoing climate change.

2. Study area

Here, the USD and the UAD were defined as all the basins upstream of the Chardara (R5) and Kerki (A8) hydrological stations (also upper basin outlets, Fig. 1), with areas of 200 269 and 284 831 km2, respectively. To understand the basin heterogeneities, the USD and UAD were divided into four and five subbasins, respectively, from upstream to downstream in terms of the river network and availability of observed streamflow data. They are subbasins of Naryn (NRB), Karadarya (KRB), Chirchik (CCB), and Fergana (FGB) in the USD (Fig. 1) and Vakhsh (VKB), Pyanj (PJB), Kafirnigan (KFB), Surkhandarya (SKB), and Kunduz (KDB) in the UAD (Fig. 1). The glacier cover varies among the basins (Table 1), with a total glacierized area of 10 440 km2 (3.7%) in the UAD and 2 747 km2 (1.2%) in the USD. In the subcatchments, glacier coverage ranges between 0.9% and 2.5% in the USD with the largest in the very upper basin NRB, and between 0.9% and 10.1% in the UAD with the most glacierized in the subbasins of VKB and the PJB (Table 1, Fig. 1). Extensive snow cover characterizes both the USD and UAD with mean annual snow cover fractions (SCFs) of 21%–51% in the subbasins according to Moderate Resolution Imaging Spectroradiometer (MODIS 10C2) (Hall et al. 2002), resulting in a mean SCF of 33% in the USD and 37% in the UAD (Table 1).

Fig. 1.
Fig. 1.

Location of the watersheds in the USD and the UAD. Naryn (NRB), Karadarya (KRB), Fergana (FGB), and Chirchik (CCB) are four subbasins of the USD. Pyanj (PJB), Vakhsh (VKB), Kafirnigan (KFB), Surkhandarya (SKB), and Kunduz (KDB) are five subbasins of the UAD. The sequence numbers (S1–S3, A1–A7, and R1–R6) denote the hydrological stations and major dams in the USD and UAD basins.

Citation: Journal of Hydrometeorology 23, 10; 10.1175/JHM-D-22-0036.1

Table 1

Characteristics of nine subbasins in the USD and the UAD.

Table 1

The climate in central Asia is primarily influenced by the intricate interaction between the air masses originating from the Atlantic–Mediterranean region, the westerlies, and topography (Schiemann et al. 2008). Precipitation variability in the upper mountainous regions is mainly influenced by the location and intensity of the westerlies (Chen et al. 2011; Hu et al. 2017). Airflow of the westerlies is deflected by the topographical boundaries of the Pamir and Tian Shan, resulting in a strong precipitation gradient with annual precipitation amounts exceeding 1000 mm at the western flanks (windward) and very arid conditions toward the direction of the Tarim Basin with annual precipitation amounts as low as 100 mm (Baldwin and Vecchi 2016). Seasonally, the regions in the west of the Tian Shan and Pamir have winter–spring precipitation maxima, whereas those in the east have summer precipitation maxima (e.g., NRB of the USD) due to the deflection of incoming winter precipitation and an increase in the importance of summer precipitation caused by northern intrusions (Aizen et al. 2006; Schiemann et al. 2008; Baldwin and Vecchi 2016). There are large seasonal variations in air temperature in central Asia, with very cold winters (as low as −37°C) and hot summers (as high as 32°C) due to the extreme continental climate in the region (Williams and Konovalov 2008; Hu et al. 2014).

3. Data and methods

a. Hydrological model

The Variable Infiltration Capacity (VIC) model (Liang et al. 1994, 1996) is a semidistributed macroscale hydrological model, parameterizing the main hydrometeorological processes at the land surface–atmosphere interface. The model solves both surface water and energy balances within each grid cell. Surface runoff and baseflow for each cell are routed to the basin outlet via a channel network (Lohmann et al. 1998). A detailed model description can be found in Gao et al. (2010) and https://vic.readthedocs.io/en/master/. The model simulates surface water balance terms such as evapotranspiration, surface runoff, baseflow (subsurface drainage into the local stream channel network, as opposed to groundwater recharge), and total soil moisture, including liquid and ice content in each soil layer. The formulation of subsurface runoff follows the Arno model conceptualization (Todini 1996).

The current version of VIC does not have a glacier melt module. Here, we used the VIC model linked to a simple degree-day glacier melt algorithm (Hock 2003), termed “VIC-glacier,” which has been previously used to simulate the flow regime of several major river basins in the Tibetan Plateau (Zhang et al. 2013; Su et al. 2016). VIC-glacier was used here to simulate the hydrological processes in the USD and UAD basins at a 10 km × 10 km (1/12° × 1/12°) spatial resolution and 3-hourly time step.

The simulated total runoff of each grid is the sum from both glacierized and nonglacierized areas, i.e.,
Rj=f×Rg+(1f)×Rs,
where Rj is the total runoff (mm) of grid cell j; f is the percentage of glacier area; Rg is the glacier runoff in mm, including all sources of runoff from glacierized areas; and Rs is the sum of surface runoff and baseflow runoff (mm) for nonglacierized areas calculated by the VIC model. The Rs is divided into rainfall and seasonal snowmelt runoff by subtracting the standard VIC output of snowmelt from Rs.
The glacier runoff Rg is calculated as
Mi={DDF×(TTbase);T>Tbase0;TTbase,
Rg=M1++Mi;i=1,2,3,,n,
where Mi is the meltwater (mm) from elevation band i and n is the total number of elevation bands; DDF is the degree-day factors of glacier or snowmelt (mm °C−1 day−1); T (°C) is the daily average air temperature above the glacier surface; Tbase (°C) is the temperature threshold between rain and snow, which is a constant value (0°C) in this study. In a precipitation event, it rains when the temperature is above 0°C; otherwise, it snows. The melting process follows the principle that snow starts to melt in front of the glacier when there is a snow cover on the glacier, and different degree-day factors are used to calculate the melting of snow and glacier ice.

In normal operation, VIC disaggregates daily forcings to a subdaily time step as user defined, with the MTCLIM algorithms (Kimball et al. 1997; Thornton and Running 1999). The parameter “off_gmt” (from the soil parameter file) determines how VIC interprets subdaily time steps relative to the model start date and time.

b. Data

1) Climate data

(i) Reanalysis gridded data

The PGMFD (Princeton Global Meteorological Forcing Dataset for Land Surface Modeling) is a global, merged dataset of meteorological forcing based on the National Centers for Environmental Prediction–National Center for Atmospheric Research (NCEP–NCAR) reanalysis, with a spatial resolution of 0.25° and daily temporal resolution for 1948–2016 (Sheffield et al. 2006). Daily precipitation, maximum (Tmax) and minimum temperature (Tmin), and wind speed from the PGFMD were used in this work (Table 2). Precipitation from NCEP–NCAR reanalysis is downscaled in space by the Global Precipitation Climatology Project (GPCP) precipitation product (Huffman et al. 2001). Temperature and wind speed from NCEP–NCAR reanalysis are downscaled using bilinear interpolation but with adjustments for differences in elevation between two grids (Kistler et al. 2001).

Table 2

Data used in this work. Note. NSDIC = National Snow and Ice Data Center; APHRODITE = Asian Precipitation- Highly-Resolved Observational Data Integration Towards Evaluation of the Water; PGMFD = Global Meteorological Forcing Dataset for land surface modeling; IGBP-DIS = International Geosphere-Biosphere Program Data and Information System; GLCM = global land cover map; ICWC = Interstate Commission for Water Coordination of Central Asia. MODIS = The Moderate Resolution Imaging Spectroradiometer; USD = upper Syr Darya; UAD = upper Amu Darya.

Table 2

Although the PGMFD provides a benchmark forcing dataset that combines one of the state-of-the-art reanalysis products, systematic biases exist in the dataset (Sheffield et al. 2006; Sidike et al. 2016; Shafeeque et al. 2019). The PGMFD precipitation does not account for topographical effects, which tends to underestimate precipitation, especially in the high mountain regions (Adam et al. 2006). For example, Sidike et al. (2016) found that the PGMFD underestimates observed precipitation by 40% at Fedchenko Glacier station (elevation of 4169 m) of the UAD. Therefore, before using the data for the hydrological model, the PGMFD precipitation and temperature data were corrected based on gauge observations (see section 3c).

(ii) Gauge observations

Monthly precipitation, maximum temperature (Tmax), and minimum temperature (Tmin) data from 270 meteorological stations (Fig. 1, purple circles) were obtained from the National Snow and Ice Data Center (NSIDC; https://nsidc.org/). However, the data are mostly restricted to the period 1951–90 due to the degradation of observation networks after the dissolution of the Soviet Union in 1991, with about 30% of the data having data gaps of more than five consecutive years. We selected 71 stations in the USD, covering the period 1962–66, and 41 stations in the UAD for 1968–90 that show less than 1% of missing data in their consecutive monthly precipitation and temperature records (Table 2).

These monthly gauged data were used to generate a gridded reference climatology for precipitation and temperature at 10 km × 10 km grids, which, in turn, was used to correct the PGMFD reanalysis data in the USD and UAD (see section 3c).

(iii) Gauge-based gridded data

The APHRODITE V1101 (Asian Precipitation–Highly Resolved Observational Data Integration Toward Evaluation of the Water; http://www.chikyu.ac.jp/precip/) is a daily gridded precipitation dataset at 0.25° resolution with a time span from 1957 to 2007 (Yatagai et al. 2012). It is the only long-term (1951 onward) continental-scale daily product with a dense gauge network (including some unpublished data) for Asia. However, the sharp decline in the number of meteorological stations after the collapse of the Soviet Union in 1991 has resulted in a dramatic drop in the magnitude of APHRODITE precipitation around 1991 in the UAD (Fig. S1 in the online supplemental material). Here, the APHRODITE was only collected for 1951–90 and together with the gauged data to correct the PGMFD reanalysis precipitation product (see section 3c).

2) Discharge data

Table 3 lists the hydrological stations used in this study for the VIC-glacier model calibration and validation obtained from the Interstate Commission for Water Coordination of Central Asia (ICWC), the Hydrographic Bureau of Tajikistan and Kyrgyzstan, and some anonymous local hydrographers.

Table 3

Hydrological stations used in the VIC-glacier model calibration and validation.

Table 3

3) Other data

In addition to the meteorological forcing data, the required input data for the VIC-glacier model also include soil texture, vegetation types, and the initial percentage of glacierized areas (Table 2). The soil data were derived from International Geosphere–Biosphere Program Data and Information System (IGBP-DIS), which provides carbon density, nutrient status, water-holding capacity, and heat capacity for modeling and inventory purposes (Global Soil Data Task 2014). Vegetation types and parameters were taken from a global land cover map (resolution of 1 km) generated by the University of Maryland (Defries et al. 2000). Topography data (resolution of 10 arc s) were obtained from the Advanced Spaceborne Thermal Emission and Reflection Radiometer Global Digital Elevation Model Version 2 (ASTER GDEM V2). The initial glacier coverage data of the USD in the 1970s were adopted from Kriegel et al. (2013), which are based on satellite images from Landsat MSS, TM, and ETM+ (http://glovis.usgs.gov). The initial glacier coverage data in the 1970s of the UAD were adopted from Wang et al. (2016), which are based on the merged data of Pamir Glaciers Dataset V.1 (PGD) as published by Aizen et al. (2007) and the World Glacier Inventory (WGI) released by the NSIDC. The glacier mass balance observations are adopted from Hoelzle et al. (2017). Based on available historical mass balance data and newly installed glaciological mass balance measurement networks, Hoelzle et al. (2017) applied a mass balance model as an extrapolation tool to obtain from mass balance point measurements to the glacier-wide mass balance data, which were subjected to rigorous quality control and validated with snow line measurements (Hoelzle et al. 2017). All the data used in this work are summarized in Table 2.

c. Methods

1) Bias correction of precipitation and temperature

According to the precipitation climatology and gradients observed in the gauged data (Figs. S3–S6), the topographical and linear correction approaches (Kan et al. 2018; Shafeeque et al. 2020; Sun and Su 2020) were applied to correct the PGMFD reanalysis precipitation data in the USD and UAD (see Fig. S2 for the schematic). Generally, there are three main steps for the correction procedures: 1) generating reference monthly gridded (10 km × 10 km) background precipitation based on gauge-based observations; 2) generating spatial correction factor fields (10 km × 10 km) for the PGMFD based on the reference background precipitation; and 3) applying the correction factors to daily PGMFD precipitation to generate corrected daily precipitation in the USD and UAD for 1951–2016. The details of the correction procedure are described in Text S2 in the online supplemental material.

Air temperature is the second key variable in hydrological model applications in high mountain basins because of its crucial role in simulating snowmelt, glacier melt, and other hydrological processes (Hock 2003). Monthly temperature gradients (TLRs) (Fig. S7) obtained from gauge observations were used to generate monthly gridded (10 km × 10 km) temperature background reference data in the USD and UAD. With this reference background, corrected daily gridded (10 km × 10 km) temperature (Tmax and Tmin) data were generated using the same method as the precipitation correction (Text S2).

2) Model calibration

The performance of the VIC-glacier model depends mainly on two categories of model parameters: 1) the degree-day factors (DDFs) for snow (DDFsnow) and ice (DDFice) for the simulation of meltwater in glacierized regions; and 2) the parameters of the VIC model for runoff simulation in nonglacierized areas. The most commonly calibrated parameters for the VIC model include the infiltration shape parameter (binfilt), the depth of the first and second soil layers (d1 and d2), and three baseflow parameters (Ds, Ws, and Dsmax), that determine the shape of the variable infiltration capacity curve, maximum moisture storage capacity, and how quickly the water in the third layer evacuates, respectively (Gao et al. 2010). The DDFsnow is based on the relationship between DDFsnow and DDFice:
DDFice=t×DDFsnow,
where t is a scaling factor usually between 1.3 and 2.9 (Kumar et al. 2016; Singh et al. 2000).

To avoid the pitfalls of model equifinality, we use a systematic two-step modeling strategy to calibrate the model. The Nash–Sutcliffe efficiency (NSE), relative error (bias, %), and correlation coefficient (CC) were used to describe the prediction skill of the modeled variables. First, the initial values of parameters related to glacier melt were derived from previous studies based on observed glacier mass balance data in central Asia (Singh et al. 2000; Gardelle et al. 2013; Chen et al. 2017). Then the DDF parameters were further constrained with the glacier inventory data of the 1970s and 2000s in the USD and UAD (Konovalov and Shchetinnicov 1994; Narama et al. 2010; Kriegel et al. 2013; Lambrecht et al. 2014; Wang et al. 2016) (Table 2) by iteratively running the model to minimize (within ±5%) the differences between the simulated and the observed glacier area changes. In this study, the DDFice and DDFsnow for the subbasins of the USD and the UAD were eventually calibrated to 6.2–10.4 mm °C−1 day−1 and 3.6–5.2 mm °C−1 day−1, respectively. The volume-area scaling approach (Bahr et al. 1997, 2015) was used to update the calculated glacier area and volume in the VIC-glacier model every year.

Second, the model parameters for nonglacierized model pixels were calibrated and validated with observed discharge data at 15 hydrological stations (Table 3) at monthly, seasonal, and annual scales. The reservoir regulations heavily affect runoff observations at subbasins of KRB, FGB, CCB, and the outlets of the entire USD and UAD. The measured runoff at those gauging stations is thus not suited for hydrological model calibration and validation. The reference discharge data at R3 (Fig. 1 and Table 3) was reconstructed based on linear relationships with the observed inflows from upper branches. The reference discharge at A8 (Fig. 1 and Table 3) was reconstructed using information from flow measurements in artificial channels and drainage systems upstream (see Text S1 and Fig. S8 for the details). For the ungauged area within the subbasin, we transferred the parameters from gauged to the ungauged area based on the theory of similarity-based parameter regionalization, which presumes that the catchments with similar characteristics have the same hydrological response, such as the spatial proximity and the physical similarity (Beck et al. 2016). The VIC parameters were optimized until the NSE and bias between observed and simulated discharge were within 0.75–1 and ±10%, respectively, which can be classified as excellent performance for the calibration period (Moriasi et al. 2007). In this study, the binfilt was eventually adjusted to 0.01–0.4. The d1 for each grid cell was set to 5–10 cm following Liang et al. (1996). The values of d2 were calibrated to 0.1–1.2 m within the subbasins. The Ds, Ws, and Dsmax were set to 0.1–0.25, 0.6–0.8, and 7–24, respectively.

4. Results

a. Corrected precipitation and temperature data

To understand the reliability of the reconstructed precipitation and temperature data, observations from 10 rain gauges and 10 temperature sensors (black points with the purple cross in Fig. 1, basic information are shown in Table S2), which were not involved in the correction process, and were used to evaluate the reconstructed precipitation and temperature (Fig. 2). The reconstructed precipitation successfully reproduced the seasonal pattern of observed precipitation from the 10 rain gauges (CC of 0.83–1, bias from −10.45% to 32.81%) (Figs. 2a–j). The original PGMFD precipitation without correction generally underestimates (bias of 2.8–36.4%) precipitation at stations at high elevations (>1400 m) and overestimates (bias of 7.56–62.5%) precipitation at stations at low elevations (<1400 m). Notably, the corrections significantly improve the seasonal distribution of the original precipitation (e.g., the NRB’s precipitation peak was corrected from April to June). For temperature (Figs. 2k–t), monthly variations in the corrected data agree well with observations, with CC of 0.91–1.00 and monthly mean delta (PGMFD − observed) of 0.31°–2.71°C among the validation points. Although bias still exists in the reconstructed temperatures, the TLRs adjustments generally have improved the overall representations of basin temperatures. The original PGMFD temperatures without TLRs adjustments would have a larger hot bias with a mean delta of 0.55°–3.68°C at seven validation points and a cold bias with a mean delta from −0.09 to −4.71°C at the other three validation points.

Fig. 2.
Fig. 2.

Monthly validation of the interpolated (a)–(j) annual precipitation and (k)–(t) temperature in the USD and the UAD basins. The location of the validation points is indicated in Table S2 and Fig. 1 (purple cross with black circle outside).

Citation: Journal of Hydrometeorology 23, 10; 10.1175/JHM-D-22-0036.1

Figure 3 shows the spatial patterns of averaged annual precipitation and temperature before and after the correction for 1961–2016. For precipitation (Figs. 3a,b), the corrections increase the spatial variability and also the magnitude of mean annual precipitation of the original PGMFD, with basinwide mean annual precipitation increasing from 414 to 510 mm in the USD and from 294 to 362 mm in the UAD (Fig. S9). The most significant increases generally appear at high altitudes (2000 m) where glaciers are abundant (e.g., NRB). There is large spatial variability in mean annual precipitation across the USD and UAD, resulting from complex topography. For the USD, about 53% of the mean annual total precipitation occurs in the mountainous region (KRB, CCB, and NRB), while only 19% occurs in the lowland (FGB). For the UAD, nearly 64% of precipitation is concentrated in the windward sloping, high mountain domain of the eastern catchments (VKB and PJB), and about 21% occurs on the leeward sloping domains PJB and KDB.

Fig. 3.
Fig. 3.

Spatial patterns of average annual precipitation and temperature (a),(c) before and (b),(d) after the corrections for 1961–2016.

Citation: Journal of Hydrometeorology 23, 10; 10.1175/JHM-D-22-0036.1

For temperature (Figs. 2c,d), the corrections tend to decrease the original temperature in regions with high elevation (>2000 m, e.g., VKB) and increase in regions with low elevation (<2000 m, e.g., FGB), as a result of the TLRs corrections. About 60% and 39% of the basin areas are below 2000 m in the USD and UAD, respectively (Fig. S6). The difference in hypsometry results in an increased basin-averaged annual temperature from 6.2° to 6.8°C in the USD but a decreased mean temperature from 5.9° to 4.7°C in the UAD (Fig. S10). The spatial distribution of temperature with TLR adjustments depicts the topographic variability with greater detail and accentuates the temperature gradient from east to west.

Precipitation and temperature seasonality are preserved after the correction (Figs. S9, S10). About 44%–83% of annual total precipitation occurs in winter and spring (December–May) across the subbasins of USD and UAD (Fig. S9), reflecting the dominant effects of westerlies in this region. One exception is the upper basin NRB in the USD (Fig. S9), where about 60%–83% of annual precipitation falls in the spring and early summer (March–July) due to the moisture blocking by the Tian Shan. The warmest and coldest months appear in July/August (9.8°–21.7°C) and December to February (from −2.4° to −16.0°C), respectively, in the USD and UAD basins, reflecting the typical temperate continental arid and semiarid climate in this region (Fig. S10).

b. VIC-glacier hydrological model validation

1) Glacier area and glacier mass balance changes

The bias between simulated and observed glacier area changes (the 1970s–2000s) is within ±5% (Fig. 4a) with the calibrated DDF parameters. We further validate the glacier-related parameters and the overall performance of the VIC-glacier model in simulating glacier runoff with the available annual cumulative glacier mass balance data from three independent glacier observation sites: Glacier No. 354 (2004–16), Abramov Glacier (1968–2014), and Batysh Sook Glacier (1971–2016) (Fig. 1 and Fig. 4b). The modeled annual glacier mass balance changes show good consistencies with the observations for the three glaciers, with absolute errors of 0.02–0.04 m water equivalent per year during the study periods (Fig. 4b) (note: assuming the mass balance at Batysh Sook decreases linearly during the gap years).

Fig. 4.
Fig. 4.

(a) Initial glacier coverage (%) in the 1970s (left axis) and comparison between VIC-glacier model-simulated and observed glacier coverage change (%) from the 1970s to 2000s (right axis). (b) Cumulative glacier mass balance from the VIC-glacier model simulations and observations (Abramov, Batysh Sook, and No. 354 glacier).

Citation: Journal of Hydrometeorology 23, 10; 10.1175/JHM-D-22-0036.1

2) Streamflow

According to the observed data, at seasonal scales, in heavily glacierized basins (e.g., VKB), about 73%–81% of annual flows occur in April–September with the peak appearing in July–August (e.g., S1 or A3 in Fig. 5). By comparison, in less glacierized basins (e.g., KFB), about 67%–80% of annual flows occur in March–August (67%–80%), with the peak appearing in May–June (e.g., R2 or A7). The simulations can ably reproduce the mean seasonal pattern of measured discharge, with CC > 0.95 and bias < 5.2% (Fig. 5). The monthly variations in observed discharge (1961–2010) are also well reproduced, with the NSE of 0.73–0.92 and bias < 1%–12% among the 14 hydrological sites (Figs. S11a–n).

Fig. 5.
Fig. 5.

Mean monthly simulated, observed, and reconstructed discharge (m3 s−1) at 14 hydrological stations in the USD and UAD. The monthly and annual time series are shown in Fig. S11.

Citation: Journal of Hydrometeorology 23, 10; 10.1175/JHM-D-22-0036.1

At annual scales (Figs. S11o–r), the VIC-glacier model simulated discharge showed consistent variations (with bias of –4.6% to 13.5% and CC of 0.7–0.9) and annual change tendencies (absolute error < 1.8 mm yr−1) with the observed inflows at four reservoirs (R1–R4) in 1961–2016.

3) Snow cover fraction

To further validate the VIC-glacier model performance, the remotely sensed SCF data from MODIS was compared with the VIC model-simulated SCF for 2001–16 in 15 selected subbasins of the USD and UAD (Fig. 6). There is substantial seasonal variability in the MODIS SCF, with the highest SCF in October–March (15%–55%) and the lowest in June–August (<6%–8%) (Fig. 6). The comparison of modeled SCF and MODIS-derived SCF showed reasonable agreement in seasonal variation, with CC of 0.89–1, while the VIC model tends to underestimate the MODIS SCF by −22% to 7.1% in the USD and 3.7–52.6% in the UAD (Fig. 6). The bias may result from uncertainties in precipitation (see uncertainty in section 5).

Fig. 6.
Fig. 6.

Mean monthly SCF (%) from the MODIS and VIC-glacier estimates for 2001–16 in 15 basins (upstream of the hydrological stations are shown in Fig. 1).

Citation: Journal of Hydrometeorology 23, 10; 10.1175/JHM-D-22-0036.1

Overall, the well-constrained climate input data and well-calibrated and validated VIC-glacier model give us confidence to investigate the runoff regime, changes, and attribution with the model simulations in the USD and UAD.

c. Runoff changes and the attribution

1) Components and characteristics of runoff

In this section, we quantify the seasonal distribution of three major runoff components (rainfall, snowmelt, and glacier runoff) and their distribution along with elevations based on the VIC-glacier model results (Figs. 7, 8). In the USD (Fig. 7), seasonal snowmelt (36%–45%) and rainfall (48%–55%) dominate the flow regime in all subbasins, with peak discharge occurring in late spring and early summer primarily induced by snowmelt (>63%) in April–May and both snowmelt and rainfall in June–July (>80%). The contribution of glacier runoff to total flow in the USD is about 8%, associated with relatively low glacier coverage (1%, Table 1). However, glacier runoff is more significant in the heavily glacierized NRB (2.5%) subbasin in the very upper USD, providing 15% of annual and 30% of July–August monthly total flows. For the downstream subbasins of CCB, FGB, and KRB, 91%–96% of annual total flows are from nonglacierized areas (seasonal snow and rainfall), while glaciers are still seasonally important by providing 15%–40% of August–September monthly totals when precipitation is lowest.

Fig. 7.
Fig. 7.

Mean monthly runoff components (left axis) in the subbasins of the USD and UAD for 1961–2016. The pie chart indicates the average annual contribution of rainfall, snowmelt, and glacier runoff to total flow in 1961–2016. The seasonal regime of precipitation is also included.

Citation: Journal of Hydrometeorology 23, 10; 10.1175/JHM-D-22-0036.1

Fig. 8.
Fig. 8.

Percentage of total runoff and three runoff components (rainfall, snowmelt, and glacier runoff) at three elevation bands in (a) the USD basin and (b) the UAD basin. The number in the parentheses indicates the number of 10-km grids in each elevation band.

Citation: Journal of Hydrometeorology 23, 10; 10.1175/JHM-D-22-0036.1

In contrast, the UAD (Fig. 7) exhibits divergent flow regimes among the subbasins largely associated with different glacier distribution and basin morphology (Table 1). In the most glacierized VKB and PJB (glacier coverage of 5.3%–10.1%, Table 1), the annual flow regime is dominated by seasonal snow (24%–32%) and glacier melt (∼30%) (nival–glacial regime), with peak flow appearing in July–August mainly from glacier runoff (>51%). In the SKB and KFB (glacier coverage of 0.7%–0.9%), annual flow is dominated by snowmelt (∼39%) and rainfall (∼56%) (nival–pluvial regime), with flow peaking in May–June. In the KDB (glacier coverage of 0.3%), annual flow is mainly driven by rainfall runoff (74%) (pluvial regime), with flow peaking in May, while glacier runoff is a major water source (46%) in the driest months of July–September.

The USD and UAD catchments are divided into high (>3500 m), middle (2000–3000 m), and low elevation bands (<2000 m), respectively. There are 3294 and 4398 grids in the USD and UAD, respectively.

Figure 8 shows the percentage of total runoff and its components at the three elevation bands. For the USD, about 71% of basin runoff is from high altitudes, where 85% of the total runoff is from rainfall (49%) and snowmelt runoff (36%), and only 15% is from glaciers. About 26% of the basin runoff is generated from middle altitudes, nearly entirely from rainfall and seasonal snow. The elevations below 2000 m barely generate runoff (3%), although these zones account for about 50% of the USD area.

For the UAD, about 49% of total runoff is from high altitudes, where glacier runoff accounts for about 40% of total runoff. Unlike the USD, about 21% of total basin runoff is from low elevations with rainfall dominant (83%). The middle altitudes contribute about 30% of total runoff, where 6% of the flow is from glacier runoff and the rest from rainfall (53%) and snowmelt (39%). These runoff distributions with elevation are quite consistent with the three identified runoff regimes in the UAD–nival–glacial regime (>3500 m), nival–pluvial regime (2000–3000 m), and pluvial regime (<2000 m).

2) Changes and the attribution

(i) At annual scales

From 1961 to 2016, all the USD and UAD basins exhibited significant warming trends (0.2°–0.3°C decade−1) (Table 4), while precipitation tended to be either stable or weakly increasing (from −0.7 to +18.8 mm decade−1) among the subbasins (Table 4). This section investigates total runoff changes and the controlling factors in the USD and UAD basins at both annual and seasonal scales during 1961–2016 from the aspects of runoff components and climate factors. For the USD, simulated total runoff at the basin outlet (R5, Fig. 1) exhibited a weakly increasing trend (3.4 mm decade−1) in 1961–2016, as an integrated result of significant increases in rainfall runoff (3.7 mm decade−1) and decreases in glacier runoff (0.9 mm decade−1), and light increases in snowmelt runoff (0.5 mm decade−1) (Figs. 9a,c). Annual variation of total runoff was highly correlated with that of precipitation, rainfall, and snowmelt runoff, with CC of 0.85–0.98 (Table 5), while it was negatively correlated with temperature and glacier runoff, suggesting a dominant role of precipitation-induced runoff from nonglacierized areas and minor impacts from glacier runoff on annual runoff. Meanwhile, the contribution of rainfall runoff to annual flow increased significantly (1.6% decade−1) during 1961–2016, while the contribution from meltwater (including from both seasonal snow and glaciers) decreased, implying an ongoing transition from a nival–pluvial regime toward a pluvial regime in the USD along with rising temperatures (Fig. 9i). For the subbasins, the annual runoff generally displayed similar change patterns to that of the entire USD in trends and variations (Tables 4, 5), suggesting the dominant role of precipitation-induced runoff across the USD at annual scales.

Fig. 9.
Fig. 9.

Annual variations of (a),(b) simulated total runoff; (c),(d) three runoff components (rainfall, snowmelt, and glacier runoff) and (e),(f) their contribution to total runoff; and (i),(j) basin precipitation and temperature for 1961–2016. Linear trends are indicated by dashed lines (two stars represent significance at the 0.01 level and one star at the 0.05 level).

Citation: Journal of Hydrometeorology 23, 10; 10.1175/JHM-D-22-0036.1

Table 4

Trends in each runoff component (mm yr−1), runoff contribution to total runoff (% yr−1), precipitation (mm yr−1), and temperature (°C yr−1) for 1961–2016 (* = significant at the 0.05 level; ** = significant at the 0.01 level).

Table 4
Table 5

Correlation coefficients between annual total runoff and three runoff components, precipitation, and temperature for the period 1961–2016 (* = significant at the 0.05 level; ** = significant at the 0.01 level).

Table 5

For the UAD, total runoff also showed an insignificant increasing trend (1.1 mm decade−1) during 1961–2016 (Fig. 9b), primarily due to the balancing between the increases in rainfall runoff and decreases in snowmelt and glacier runoff (Fig. 9d). The total flow variations of UAD were generally affected by rainfall and snowmelt runoff, with CC of 0.96 and 0.91, respectively (Table 5). Similar to the USD, the contribution from rainfall runoff to total annual flow increased, and those from meltwater decreased in 1961–2016 (Fig. 9f). Also, annual flow exhibited divergent changes among the UAD subbasins (Table 4), with significant decreases in the PJB and SKB (4.0 and 5.5 mm decade−1, respectively) and insignificant increases (0.4–3.7 mm decade−1) in the other subbasins (Table 4). The decreasing total runoff in the PJB was primarily due to declining glacier runoff (−0.4 mm decade−1) associated with significant warming but with little change in precipitation input (Table 4). In contrast, the decreasing total runoff in the SKB, with its rainfall and snowmelt-dominated runoff regime (Table 4), was mostly due to the decreased snowmelt runoff associated with a reduction in precipitation (Table 4). For the other subbasins (Table 4), the increases in precipitation input were the dominant driver in a general increasing annual runoff during 1961–2016.

(ii) At seasonal scales

Changes in meltwater characteristics are expected to modify flow regimes significantly (e.g., both quantity and timing) (Barnett et al. 2005; Khanal et al. 2021; Kraaijenbrink et al. 2021), and glacier melt especially affects water availability in warm and dry seasons for glacierized basins (Kaser et al. 2010; Pohl et al. 2017; Pritchard 2019). The intensive validation of the VIC-glacier model with different variables and at multitime scales across the USD and UAD (Figs. 46; Figs. S11, S12) encourages us to investigate how the runoff source might affect flow seasonality by using the model simulations. Figure 10 displays the seasonal change of simulated total runoff and the components in 1998–2016 relative to 1968–79. For the entire USD, runoff seasonality was generally unchanged (Fig. S13a), while total runoff tended to increase in all seasons (Fig. 10a) with the largest increases appearing in the spring (March–May, 14%–28%) and winter (December–February, 25%–36%) months, resulting in a mean annual runoff increase of about 12% relative to 1968–1979. The large winter–spring flow increases were primarily due to increased rainfall and snowmelt runoff associated with a general increase in precipitation and temperature in the winter–spring seasons (Table S3). While the general decrease in snowmelt and glacier runoff, along with the warming, largely offset the rainfall runoff increases in summer months (June–August), resulting in a moderate total flow rising (6%) in the summer. For the subbasins (Figs. 10b–e), some common features were observed: 1) an overall runoff increase in annual total flow (by 5%–8%) associated with annual precipitation increases (9%–17% among the subbasins; Table S3), 2) a general increase in winter–spring flows (3%–24%) mainly due to increased snowmelt, and 3) a consistent decrease in glacier runoff in the summer (18%–56%) along with the warming (Table S3). However, distinct runoff changes were present in the summer months, which were related to the different runoff regimes of the basins (Fig. 7). In the most glacierized NRB (Figs. 10b), the large decrease of glacier runoff (23%) resulted in a light total flow reduction in July (1%). In the CCB (with the smallest glacier area of 160 km2, Table 1), there was a peak flow shift from June to May and a large flow reduction in summer months (by 10%–13% in June–July) due to the combining effects of earlier and increased snowmelt in the spring and decreased glacier runoff in the summer (37%). In the KRB and FGB, there was generally an increased summer flow mainly due to the increased rainfall runoff, while the decreased glacier runoff contributed to the flow reductions in August–September in these two basins (7%–53%).

Fig. 10.
Fig. 10.

Changes in the mean monthly runoff for the period 1998–2016 relative to the period 1961–79.

Citation: Journal of Hydrometeorology 23, 10; 10.1175/JHM-D-22-0036.1

For the UAD subbasins (Figs. 10f–k), a general runoff decrease in summer months (June–September) and an increase in winter and spring months (November–April) were observed, but with divergent drivers among the basins characterized by distinct runoff regimes. For the most glacierized PJB and VKB, the summer flow decrease (2%–7% in VKB and 8%–17% in PJB) was primarily due to decreased glacier runoff (contributing ∼80% of the total decrease in July) together with a reduction in snowmelt runoff. For the SKB and KFB with a nival–pluvial regime (Fig. 7), the decreased summer runoff (4%–29%) was mainly due to the reduction in both rainfall and snowmelt runoff (6%–16%) associated with the warming (Table S3). For the KDB with a pluvial regime (Fig. 7), the decreased summer runoff (3%) is mainly attributed to rainfall runoff reductions (48%) (Table S3).

For the spring months (Fig. 10), the flow increases in the VKB and KDB (4%–16%) were mostly due to the increased snowmelt (3%–12%) and rainfall runoff (5%–32%). In the PJB and KFB, increased rainfall (17%–19%) and decreased snowmelt runoff (2%–9%) resulted in a moderate flow rising in the spring (2%–7%). In contrast, the SKB exhibited a reduction in spring runoff (6%), which was closely related to a decrease in spring precipitation (20%) (Table S3). The general increase in winter flows (10%–39%) resulted from both increased rainfall (22%–303%) and snowmelt runoff (1%–6%) in the VKB, PJB, and KDB, while entirely due to the increased rainfall runoff (22%–290%) in the SKB and PJB (Table S3).

For the entire UAD (Fig. 10f; Fig. S13f), seasonal runoff changes and their contributions generally follow the characteristics of the glacierized basins, with flow decreases in summer (6%) mainly a result of both reduced glacier and snowmelt runoff (17%–21%), and flow increases (11%–18%) in winter and spring resulting from either increased rainfall or snowmelt runoff.

5. Discussion

a. Uncertainty

Overall, with corrected precipitation and temperature, our model performs well when compared with observed runoff, glacier mass balance, and SCF. However, uncertainties still exist in the modeling results, primarily originating from 1) precipitation and temperature inputs, 2) strategy and data used for degree-day model calibration, and 3) the uniform set of VIC parameters which might affect the runoff simulation at small scales.

(i) Precipitation

Melting water from glaciers and seasonal snow play a crucial role in central Asia’s hydrological cycle (Unger-Shayesteh et al. 2013; Pohl et al. 2017; Saks et al. 2022). The accuracy of the model’s meteorological inputs (precipitation and temperature), especially precipitation, corresponds to major uncertainty, which is how to quantify the meltwater contribution (Zhang et al. 2013; Pohl et al. 2015; Tarasova et al. 2016). For example, the underestimates in precipitation can be compensated for in the model by high glacier melt and vice versa (Lutz et al. 2014; Pohl et al. 2015; Tarasova et al. 2016). Among current studies, the contribution of each runoff component in the USD and UAD vary considerably (Table 6) associated with different precipitation inputs. In our study, precipitation was overall well constrained in terms of both point- and basin-scale validations, although uncertainties in precipitation still exist mainly due to the sparse gauge stations involved in the correction approach. For example, precipitation amounts tend to be underestimated with a decreased rain gauge density when using a linear correction approach (Sun and Su 2020), potentially leading to precipitation underestimates in high-altitude areas with few gauge observations. In this work, the large negative bias in the simulated SCF in the UAD basins (e.g., A1, Fig. 6) may be partly due to precipitation associated with the linear correction approach. However, it is worth noting that the changes in snowmelt runoff contribution are relatively sensitive only when the precipitation bias exceeds a certain limit (Fig. 11). For example, in the VKB, a 20% change in precipitation results in only a 1%–3% change in the contribution of snowmelt runoff, while a 40% change in precipitation will cause a 10%–16% change in snowmelt contribution (Fig. 11). Point validations (Fig. 2) show that the bias of corrected precipitation is generally within 20%. However, all the validation sites are below 4200 m in elevation, and the precipitation uncertainties above 4200 m are not really known. There is still much room for improvement in characterizing precipitation in complex mountain terrain by increasing monitoring networks (Immerzeel et al. 2015; Ji et al. 2020; Dahri et al. 2018; Kan et al. 2018).

Fig. 11.
Fig. 11.

Sensitivity of glacier, snowmelt, and rainfall runoff contributions to the changes of precipitation in the VKB for 1961–2016.

Citation: Journal of Hydrometeorology 23, 10; 10.1175/JHM-D-22-0036.1

Table 6

Relevant studies on simulated runoff contributions and trends in the entire USD and UAD basins.

Table 6

(ii) Temperature

Besides precipitation, the temperature is also crucial to hydrological model simulations but with fewer uncertainties than precipitation (Kan et al. 2018; Tarasova et al. 2016). Even with few stations, interpolating temperature with TLRs can sufficiently represent the spatial variability of the runoff generation controls in mountainous basins (Tarasova et al. 2016). In this study, most of the stations used to derive TLRs in the UAD lie in elevations lower than 4200 m, which may not fully represent regional TLRs given the complex topography (with the highest elevation of 6500 m) within the basin. TLRs at high altitudes may differ significantly from those at low altitudes (Immerzeel et al. 2014; Kan et al. 2018; Li et al. 2013). For example, Immerzeel et al. (2014) reported that TLRs at the Langtang basin (south Himalayas) are generally shallower than commonly used lapse rates at lower elevations (−0.0065°C m−1). In contrast, Kan et al. (2018) suggested that TLRs of the upper Yarkant (east Pamir) derived from high altitudes are higher than those from stations at low altitudes by about 0.02°–0.56°C (100 m)−1. In this work, TLRs at high altitudes (>4200 m, accounting for 20% of the total basin area) are likely to be underestimated, resulting in an overestimation of temperature leading to fast melting of snow cover (e.g., Figs. 6h–o). At this time, little is known about the spatial variability of TLRs in regions with extreme topography. More field campaigns are needed to further our understanding of TLR estimation. Besides uncertainties in the TLR, some uncertainties exist in the temperature (Ts) threshold. We employed a spatially uniform air temperature (Ts) threshold, in this study 1°C, to separate the rain and snow. However, the Ts threshold shows significantly spatial variation and is highly uncertain in areas with a complex topography and climate (Berghuijs et al. 2014; Li et al. 2020). Some high-altitude areas in central Asia (e.g., the Pamir) exhibit high observed Ts thresholds, approaching 4.5°C (Jennings et al. 2018), much higher than ours, meaning our model may underpredict snowfall in these areas. With more available snow depth observations (e.g., Liu et al. 2017), the underestimation of snow cover may be improved with the calibration of the Ts thresholds (Storck and Lettenmaier 1999), thus achieving better snowmelt runoff simulations.

(iii) Degree-day factors

The accuracy of the degree-day model primarily relies on the DDF parameters, which predominantly affect the glacier runoff simulation at regional scales (Huss and Hock 2018; Rounce et al. 2020a,b). The DDF parameters in this work have been overall well constrained through comparison with observed historical glacier area changes and mass balance data (Fig. 4). However, uncertainties in the degree-day model may arise from the initial glacier condition and the data used for calibration. First, since we only use regional average glacier area loss rates in each subbasin to calibrate the DDF parameters, the uncertainty of individual glacier’s runoff amount and year-to-year variations is considerable. However, these errors should be lower at the basin scale and over long time periods (Huss 2012). Second, our simulations use a reference glacier surface representing conditions in the early 1970s and nearly 15% are from the 1980s, during which the image is relatively complete (Bolch 2007; Unger-Shayesteh et al. 2013; Wang et al. 2016). However, since this period is not toward the beginning of our model period, our icefield geometry may be too low in elevation and too small in extent for the early years of our simulation. The former would likely result in the overestimation of glacial runoff due to higher temperatures at lower elevations, while the latter would cause underestimation due to insufficient glacier extent. The excessive melting of low elevation glaciers partly explains the slightly negative simulated glacial mass balance in the first decade compared with the measured data (Fig. 4b). However, given our principal goal of examining changes for regional runoff components with an emphasis on the long-term trends, we accept this cost.

The uncertainty associated with the VIC model parameters is much smaller than the uncertainty associated with the climate forcing (Zhang et al. 2013; Tarasova et al. 2016; Rounce et al. 2020a). In this study, a uniform set of calibrated model parameters was used at the subbasin scale, while the values of most calibrated parameters vary spatially and temporally. However, as discussed by Tarasova et al. (2016), increasing model complexity by using spatially distributed input values or semidistributed parameter values does not increase model performance in data-scarce regions, while additional calibration data (e.g., snow cover fraction) can improve model performance in quantifying the flow components (Unger-Shayesteh et al. 2013; Duethmann et al. 2014; Tarasova et al. 2016). In this study, the model was extensively calibrated and validated with runoff data and the snow cover fraction, glacier area change, and mass balance data, increasing the confidence in the results of this study.

b. Underneath mechanism linked to different hydrological regimes

The ultimate source of river flow in the USD and UAD is the occurrence of precipitation. However, river flow’s time distribution and magnitude are significantly modified by three factors: 1) glacier storage, 2) precipitation regime, and 3) topography.

First, glaciers act as frozen reservoirs, with the ability to temporarily store water over seasonal and decadal time scales (Miles et al. 2021). Glacier runoff typically shows a distinct seasonality with a minimum in the winter (accumulation season) and a pronounced maximum in summer (melting season). Therefore, meltwater from glaciers plays a crucial role in the hydrological regimes. In our results, three subbasins (NRB, VKB, and PJB) with relatively high (>2.5%) glacier coverages have a single peak flow in July–August associated with large seasonal glacier contributions (58%–76% in July–August). Other subbasins with relatively low glacier fractions (<2.5%) have peak flows mainly in May–June, associated with the highest spring snowmelt water and/or rainfall runoff. However, glaciers are still seasonally important over those subbasins by providing 15%–40% of August–September monthly totals. Our results indicate that glacier runoff decreases in all subbasins during the study periods. For the USD and UAD, the most negative glacier runoff decreasing trend occurs in the CCB and VKB subbasins, respectively, while the reasons differ. The CCB has a relatively high fraction (∼55%) of small glaciers (<1 km2) in the region, which are prone to disappear due to fast melting (Narama et al. 2010; Kriegel et al. 2013); thus, the total glacier flow cannot be compensated by the increase of melt rate, eventually leading to a decreased glacier runoff. In contrast, the VKB is reported to have a relatively less negative mass balance (Knoche et al. 2017; Hugonnet et al. 2021) while having the highest glacier runoff decreasing trend over the UAD subbasins. One possible explanation could be the cooling summer temperature in the glacier region, which has been reported in the northern Pamir region during recent decades (e.g., Wang et al. 2017). We suggest, therefore, that there is potential for future increases in glacier runoff volumes in the VKB. Although there are some disputes about whether or not a particular glacierized basin or region has passed the tipping point (peak water) (Sorg et al. 2014a; Luo et al. 2018; Huss and Hock 2018), the glacier volume is reported to decrease in the future and will eventually result in a decrease in glacier runoff (e.g., Bolch et al. 2012; Rounce et al. 2020a,b). Moreover, our results indicate that total annual runoff during the historical period is not significantly correlated with glacier runoff, suggesting an overwhelming precipitation signal. This is especially true for downstream basins (e.g., KFB, KDB), with a negative correlation between glacier runoff and total flow. The precipitation signal will likely be further enhanced in the future, given the projected change in rain/snow fraction toward rain. The buffering role of glaciers may disappear with a transition to a more volatile pluvial regime, thus exacerbating current water-related issues (relationship between irrigation, hydropower, and water availability).

Second, the precipitation regime can largely affect the runoff process by seasonal precipitation pattern and precipitation form (e.g., Li et al. 2018). Except for the NRB, all the subbasins possess a consistent seasonal precipitation pattern, with the winter–spring precipitation maxima and massive snowmelt in the spring and summer seasons. However, the NRB of the USD has a summer precipitation maximum due to the deflection of incoming winter precipitation and an increase in the importance of summer precipitation caused by northern intrusions (Aizen et al. 2006; Schiemann et al. 2008; Baldwin and Vecchi 2016). In the future, basins with winter–spring precipitation maxima could be more vulnerable than those with summer precipitation maxima. Those basins with winter–spring precipitation maxima generally show a shift of peak runoff and an increase in spring runoff due to rising temperature, especially basins with high snow cover fraction (e.g., CCB and VKB), which will lead to a severe reduction in dry-season water availability when precipitation is lowest. However, for the NRB, the glacier melt with the maximum precipitation in summer can somewhat buffer the flow shift, partly explaining that NRB had the slightest seasonal flow shift in the historical period among all subbasins.

Third, hydrological regimes are also linked to different topography conditions (Bertoldi et al. 2006). The topography can be well reflected by the hypsometric curve, which largely determines the ratio of solid and liquid precipitation within a basin and is strongly related to the hydrograph peak and travel time (Luo and Harlin 2003). For example, the flow in CCB is characterized by a sharp rising from March to May among all the subbasins (Fig. S14a). The different hypsometric curves can partly explain the sharp rising flow during March–May. The CCB has a more significant portion of its area at higher elevations than other subbasins (Fig. S14c) Thus, more winter precipitation can be stored in the form of snow and released massive water with the rising temperature. Moreover, the elevation at the lowlands (toe region) is steep in the CCB, as less surface area (<10%) is covered in the lowlands for elevations of up to 30% (Fig. S14c). Therefore, any runoff generated in the lowlands tends to drain out quickly due to the steepness of the topography.

c. Reservoir dampening effects

We quantified natural runoff regime, composition, and flow changes over the USD and UAD. The natural surface waters in USD and UAD are heavily regulated by extended reservoirs, with five large reservoirs in the USD and one in the UAD (Table 1). Those reservoirs were built mainly for irrigation downstream and hydropower upstream. Therefore, the operations of those reservoirs have resulted in a shift in streamflow seasonality, primarily reducing spring and summer peak flows and increasing fall and winter low flows (e.g., Toktugol reservoir, Fig. S15a), while not significantly affecting the overall magnitude of mean annual flow (<1%) (Fig. S15b). Water conflicts have become more prominent between upstream and downstream countries since the collapse of the former Soviet Union, as the upstream countries favor withholding irrigation water during the spring and summer months for release in winter to meet internal electricity needs (e.g., Toktogul reservoir, Fig. S16). Our analysis showed that the most critical impacts of climate change emerge from significant changes in runoff seasonality, which may further exacerbate current water stress. In our results, the USD has an increase in both spring and summer runoff, and the UAD has an increase in spring runoff while with a decrease in summer runoff during the historical period. For the USD, the reservoirs in the regulated basins seem to have the potential to adapt the water allocation according to the future runoff changes. For example, Fig. S16 shows the monthly time series of inflow and outflow at the region’s largest reservoir (Toktogul) for 1980–95 and 2000–10. There is about a 23% water deficit during the growing season (usually from April to September) under current regulations (data from ICWC for 2001–10). To fill the water gap, a shift of about 2.5 km3 more water from the no-growing season to the growing season is required for meeting downstream irrigation demands. The Toktogul reservoir has full potential to withhold the increased spring water and release them in the summer months for irrigation, as the typical capacity from March–April is 7–8 km3 (ICWC) while its total storage facility is 19 km3. Although the USD shows an increased summer runoff due to the increased rainfall runoff, the reduced glacier volume may eventually reduce future summer runoff. Therefore, it remains questionable whether the increased water in spring can offset the decreased water in summer through reservoir regulations. For the UAD, the basin is less regulated by river-reservoir systems and thus may be subject to more risks. For unregulated basins (e.g., SKB), with more water available in spring, water may be wasted after the irrigation needs are fully met, increasing the shortage of available water in the summer. Our results observed a declining trend in summer runoff for the regulated basins (e.g., VKB). Therefore, upstream countries may tend to store more water in summer to meet power generation needs, further exacerbating the current water stress in the future.

Generally, although the USD and UAD will likely undergo substantial changes in runoff seasonality under future climate change, the reservoir in the regulated basins can play a crucial role in adaptation to climate changes, at least in short to medium term. The potential of those big reservoirs offers some room for optimism that policymakers of the riparian countries can set up an effective international water management system before the most severe climate change–related problems hit the region. However, the biggest concern is that the changes in runoff seasonality are likely to cause severe problems in unregulated subbasins and subbasins only with small reservoirs. In those areas, the options for adaptation to future changes in water availability are limited. Therefore, the new construction of reservoirs could be an effective way to manage discrepancies between the seasonal cycles of water supply and demand in those areas.

6. Conclusions

In this work, an improved forcing dataset of precipitation and temperature (daily, 10 km) was developed for the USD and UAD for 1951–2016 through topographical and liner corrections of PGMFD grid data. The well-constrained climate input data were used to drive the VIC-glacier model in the USD and UAD basins, and the hydrological model was intensively calibrated and validated with observed discharge, glacier area, and glacier mass balance changes, and MODIS SCF data. The runoff regime, composition, and flow changes were quantified across the USD and UAD basins based on the well-calibrated VIC-glacier model results for 1961–2016. The main results are summarized as the following:

  1. Seasonal snowmelt (36%–45%) and rainfall (48%–55%) runoff dominate the flow regime across all the subbasins of USD, while glacier runoff is crucial in summer seasons when precipitation is lowest by providing 15%–40% of August–September monthly totals. In contrast, the UAD subbasins exhibit divergent flow regimes among the subbasins, which can be grouped into nival–glacial regime (VKB and PJB), with snowmelt (24%–32%) and glacier runoff (29%–30%) dominating; nival–pluvial regime (SKB and KFB), with rainfall (56%–57%) and snowmelt (38%–39%) runoff dominating; and pluvial regime (KDB), with rainfall runoff (74%) dominating.

  2. About 71.4% of basin runoff in the USD is from high altitudes (>3500 m) and is mainly comprised of rainfall (49%) and snowmelt runoff (36%), while almost no flow occurs at low altitudes (3%), which accounts for nearly half of the basin area. In the UAD, runoff distribution is relatively uniform, with 49%, 30%, and 21% of total runoff for the high, middle, and low elevations, respectively.

  3. Annual runoff exhibited weakly increasing trends across the USD basins during 1961–2016. Precipitation-induced runoff from nonglacierized areas were the major factors dominating the flow trends and variations in the USD, with minor impacts from glacier runoff at annual scales. In the UAD, annual flow also showed an overall insignificant increase as a result of the trade-off between the increase in rainfall and the decrease in snowmelt and glacier runoff. In contrast, the subbasins displayed divergent flow changes, with the most glacierized PJB showing decreasing runoff due to the decreased glacier and snow meltwater. Furthermore, the ongoing glacier mass loss and glacier runoff reductions in the PJB and VKB were not apparently impacting the annual flow variations due to the overwhelming precipitation signal. Both the USD and UAD saw an increasing contribution from rainfall runoff to annual flow, but a decreasing contribution from snow and glacier meltwater during 1961–2016, suggesting a transition from a nival–pluvial regime toward a pluvial regime along with the warming. Seasonally, spring runoff significantly increased in all the USD and UAD basins primarily due to increased rainfall and early snow melting, tending to shift the peak flow from June–July to April–May. In contrast, distinct runoff changes were presented in the summer months among the basins. For the USD, the general decrease in snowmelt and glacier runoff along with the warming largely offset the rainfall runoff increases in the summer months (June–August), resulting in a moderate total flow rising (6%) in the summer. While for the UAD, a general runoff decrease (2%–48%) in the summer months (June–August) was observed, with divergent drivers among the subbasins characterized by distinct runoff regimes. The decreased glacier runoff was generally responsible for the summer flow reductions in the most glacierized basins (VKB and PJB) and the UAD as a whole.

Our results show that the most prominent signs of hydrological regime change in this region relate to the timing of water delivery downstream. Agriculture is by far the most significant water user in central Asia, while mountainous countries rich in water resources (Kyrgyzstan and Tajikistan) tend to save the water within their dams for generating hydroelectricity in winter, putting downstream agricultural water supplies (mainly in spring and summer) under pressure. A marked increase in spring runoff coincides with the growing season of downstream crops (mainly cotton), potentially mitigating current water conflicts. With more water available in spring, water may be wasted after the irrigation needs are fully met, increasing the shortage of available water in the summer. This is especially true for the UAD, where a declining trend in summer runoff was observed. The UAD is less regulated by river-reservoir systems than the USD. Therefore, new dams are needed to address the imbalance between seasonal water supply and demand. In the context of increasing regional water demand, there is also a need to improve water use efficiencies, such as shifting the agricultural production away from cotton to wheat, as wheat only uses half as much irrigation water compared to cotton. Our model provides a basic framework that can be used for seasonal runoff prediction, which can permit some quantification of the ongoing debates concerning questions of seasonal water supply.

Acknowledgments.

This work was financially supported by the National Natural Science Foundation of China (41988101, 41871057), the Second Tibetan Plateau Scientific Expedition and Research (STEP) Program (2019 QZKK0201), and the “Strategic Priority Research Program” of the Chinese Academy of Sciences (XDA20060202). We would like to thank Eric Phol for his great contribution to this work.

Data availability statement.

The PGMFD precipitation, temperature, and wind speed are available from the Terrestrial Hydrology Research Group (http://hydrology.princeton.edu/data.pgf.php). The APHRODITE data can be obtained from the APHRODITE project website (http://aphrodite.st.hirosaki-u.ac.jp/). The constructed gridded daily forcing dataset (10 km × 10 km) for 1951–2016 in the USD and UAD in this study can be downloaded from the National Tibetan Plateau Data Center (https://data.tpdc.ac.cn). Snow cover fraction data from MODIS 10C2 can be obtained from the MODIS website (https://modis.gsfc.nasa.gov/data/). Station data of precipitation and temperature are from National Snow and Ice Data Center (https://nsidc.org/). Runoff data of the hydrological stations are collected from the Scientific-Information Center of the Interstate Commission for Water Coordination in Central Asia (http://isepei.org/organization/sic-icwc) and Hydrographic Bureau of Tajikistan and Kyrgyzstan. Due to confidentiality agreements, runoff data of the reservoirs in this study can only be made available to bona fide researchers subject to a non-disclosure agreement. Details of the data and how to request access are available from http://www.cawater-info.net/bd/index_e.htm.

REFERENCES

  • Adam, J. C., E. A. Clark, D. P. Lettenmaier, and E. F. Wood, 2006: Correction of global precipitation products for orographic effects. J. Climate, 19, 1538, https://doi.org/10.1175/JCLI3604.1.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Aizen, V. B., E. M. Aizen, D. R. Joswiak, K. Fujita, N. Takeuchi, and S. A. Nikitin, 2006: Climatic and atmospheric circulation pattern variability from ice-core isotope/geochemistry records (Altai, Tien Shan and Tibet). Ann. Glaciol., 43, 4960, https://doi.org/10.3189/172756406781812078.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Aizen, V. B., E. M. Aizen, and V. A. Kuzmichonok, 2007: Glaciers and hydrological changes in the Tien Shan: simulation and prediction. Environ. Res. Lett., 2, 045019, https://doi.org/10.1088/1748-9326/2/4/045019.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Alford, D., U. Kamp, and C. Pan, 2015: The role of glaciers in the hydrologic regime of the Amu Darya and Syr Darya Basins. Rep. ACS12128, The World Bank, 115 pp., http://hdl.handle.net/10986/24082.

  • Armstrong, R. L., and Coauthors, 2019: Runoff from glacier ice and seasonal snow in High Asia: Separating melt water sources in river flow. Reg. Environ. Change, 19, 1249–1261, https://doi.org/10.1007/s10113-018-1429-0.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Badescu, V., and R. D. Schuiling, 2010: Aral Sea; Irretrievable loss or Irtysh imports? Water Resour. Manage., 24, 597616, https://doi.org/10.1007/s11269-009-9461-y.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Bahr, D. B., M. F. Meier, and S. D. Peckham, 1997: The physical basis of glacier volume-area scaling. J. Geophys. Res., 102, 20 35520 362, https://doi.org/10.1029/97JB01696.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Bahr, D. B., W. T. Pfeffer, and G. Kaser, 2015: A review of volume-area scaling of glaciers. Rev. Geophys., 53, 95140, https://doi.org/10.1002/2014RG000470.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Baldwin, J., and G. Vecchi, 2016: Influence of the Tian Shan on arid extratropical Asia. J. Climate, 29, 57415762, https://doi.org/10.1175/JCLI-D-15-0490.1.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Barnett, T. P., J. C. Adam, and D. P. Lettenmaier, 2005: Potential impacts of a warming climate on water availability in snow-dominated regions. Nature, 438, 303309, https://doi.org/10.1038/nature04141.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Beck, H. E., A. I. J. M. van Dijk, A. de Roo, D. G. Miralles, T. R. McVicar, J. Schellekens, and L. A. Bruijnzeel, 2016: Global-scale regionalization of hydrologic model parameters. Water Resour. Res., 52, 35993622, https://doi.org/10.1002/2015WR018247.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Berghuijs, W. R., R. A. Woods, and M. Hrachowitz, 2014: A precipitation shift from snow towards rain leads to a decrease in streamflow. Nat. Climate Change, 4, 583586, https://doi.org/10.1038/nclimate2246.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Bernauer, T., and T. Siegfried, 2012: Climate change and international water conflict in Central Asia. J. Peace Res., 49, 227239, https://doi.org/10.1177/0022343311425843.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Bertoldi, G., R. Rigon, and T. M. Over, 2006: Impact of watershed geomorphic characteristics on the energy and water budgets. J. Hydrometeor., 7, 389403, https://doi.org/10.1175/JHM500.1.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Bolch, T., 2007: Climate change and glacier retreat in northern Tien Shan (Kazakhstan/Kyrgyzstan) using remote sensing data. Global Planet. Change, 56, 112, https://doi.org/10.1016/j.gloplacha.2006.07.009.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Bolch, T., and Coauthors, 2012: The state and fate of Himalayan glaciers. Science, 336, 310314, https://doi.org/10.1126/science.1215828.

  • Chen, F., W. Huang, L. Jin, J. Chen, and J. Wang, 2011: Spatiotemporal precipitation variations in the arid Central Asia in the context of global warming. Sci. China Earth Sci., 54, 18121821, https://doi.org/10.1007/s11430-011-4333-8.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Chen, Y., W. Li, H. Deng, G. Fang, and Z. Li, 2016: Changes in Central Asia’s water tower: Past, present and future. Sci. Rep., 6, 39364, https://doi.org/10.1038/srep39364.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Chen, Y., W. Li, G. Fang, and Z. Li, 2017: Review article: Hydrological modeling in glacierized catchments of central Asia – Status and challenges. Hydrol. Earth Syst. Sci., 21, 669684, https://doi.org/10.5194/hess-21-669-2017.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Dahri, Z. H., E. Moors, F. Ludwig, S. Ahmad, A. Khan, I. Ali, and P. Kabat, 2018: Adjustment of measurement errors to reconcile precipitation distribution in the high-altitude Indus basin. Int. J. Climatol., 38, 38423860, https://doi.org/10.1002/joc.5539.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Defries, R. S., M. C. Hansen, J. R. G. Townshend, A. C. Janetos, and T. R. Loveland, 2000: A new global 1-km dataset of percentage tree cover derived from remote sensing. Global Change Biol., 6, 247254, https://doi.org/10.1046/j.1365-2486.2000.00296.x.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Duethmann, D., J. Zimmer, A. Gafurov, A. Güntner, D. Kriegel, B. Merz, and S. Vorogushyn, 2013: Evaluation of areal precipitation estimates based on downscaled reanalysis and station data by hydrological modelling. Hydrol. Earth Syst. Sci., 17, 24152434, https://doi.org/10.5194/hess-17-2415-2013.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Duethmann, D., J. Peters, T. Blume, S. Vorogushyn, and A. Güntner, 2014: The value of satellite‐derived snow cover images for calibrating a hydrological model in snow‐dominated catchments in Central Asia. Water Resour. Res., 50, 20022021, https://doi.org/10.1002/2013WR014382.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Gan, R., Y. Luo, Q. Zuo, and L. Sun, 2015: Effects of projected climate change on the glacier and runoff generation in the Naryn River basin, Central Asia. J. Hydrol., 523, 240251, https://doi.org/10.1016/j.jhydrol.2015.01.057.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Gao, H., and Coauthors, 2009: Water budget record from Variable Infiltration Capacity (VIC) model. Algorithm Theoretical Basis Doc., 56 pp., http://www.hydro.washington.edu/SurfaceWaterGroup/Publications/Water_Cycle_MEaSUREs_ATBD_VICmodel_submit.doc.

  • Gardelle, J., E. Berthier, Y. Arnaud, and A. Kääb, 2013: Region-wide glacier mass balances over the Pamir-Karakoram-Himalaya during 1999–2011. Cryosphere, 7, 12631286, https://doi.org/10.5194/tc-7-1263-2013.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Global Soil Data Task, 2014: Global Soil Data Products CD-ROM Contents (IGBP-DIS) ORNL DAAC, accessed 2 October 2017, https://doi.org/10.3334/ORNLDAAC/565.

    • Crossref
    • Export Citation
  • Gulakhmadov, A., X. Chen, N. Gulahmadov, T. Liu, M. N. Anjum, and M. Rizwan, 2020: Simulation of the potential impacts of projected climate change on streamflow in the Vakhsh river basin in Central Asia under CMIP5 RCP scenarios. Water, 12, 1426, https://doi.org/10.3390/w12051426.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Hagg, W., M. Hoelzle, S. Wagner, E. Mayr, and Z. Klose, 2013a: Glacier and runoff changes in the Rukhk catchment, upper Amu-Darya basin until 2050. Global Planet. Change, 110, 62–73, https://doi.org/10.1016/j.gloplacha.2013.05.005.

    • Crossref
    • Export Citation
  • Hagg, W., C. Mayer, A. Lambrecht, D. Kriegel, and E. Azizov, 2013b: Glacier changes in the Big Naryn basin, Central Tian Shan. Global Planet. Change, 110, 4050, https://doi.org/10.1016/j.gloplacha.2012.07.010.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Hall, D. K., G. A. Riggs, V. V. Salomonson, N. E. DiGirolamo, and K. J. Bayr, 2002: MODIS snow-cover products. Remote Sens. Environ., 83, 181194,