Evaluation of an ice ablation model to estimate the contribution of melting glacier ice to annual discharge in the Nepal Himalaya


  • Adina E. Racoviteanu,

    Corresponding author
    1. Institute of Arctic and Alpine Research and Department of Geography, University of Colorado, Boulder, Colorado, USA
    2. Laboratoire de Glaciologie et Géophysique de l'Environnement, Saint Martin d'Hères, France
    • Corresponding author: A. E. Racoviteanu, Laboratoire de Glaciologie et Géophysique de l'Environnement, 54 rue Molière, BP 96, FR-38402 Saint Martin d'Hères CEDEX, France. (Adina.Racoviteanu@lgge.obs.ujf-grenoble.fr)

    Search for more papers by this author
  • Richard Armstrong,

    1. National Snow and Ice Data Center, University of Colorado, Boulder, Colorado, USA
    Search for more papers by this author
  • Mark W. Williams

    1. Institute of Arctic and Alpine Research and Department of Geography, University of Colorado, Boulder, Colorado, USA
    Search for more papers by this author


[1] This study focuses on the contribution of annual glacier ice melt to streamflow along two rivers in two watersheds situated in the monsoon-influenced part of the Nepal Himalaya (Trishuli and Dudh Kosi basins). We used a simple elevation-dependent ice ablation model based on glacier areas from Advanced Spaceborne Thermal Emission and Reflection Radiometer (ASTER) and IKONOS remote-sensing data combined with hypsometry from the Shuttle Radar Topography Mission (SRTM). Long-term hydrologic measurements were used to calculate the percent contribution of the glacier ice melt component of the water balance to discharge at various elevations and distances from glacier outlets. Glacier ice melt was positively correlated with the basin glacierized area and contributed 58.3% to annual flow in the small Langtang Khola watershed in the Trishuli basin (43.5% glacierized area) and 21.2% in the Hinku watershed in Dudh Kosi basin (34.7% glacierized area). Of this, 17.7% and 4.1% of streamflow, respectively, was due to the contribution of debris-covered glaciers in Langtang Khola and Hinku. The contribution of glacier ice melt to measured discharge decreased substantially toward lowland locations in both study sites, i.e., 9.5% of the streamflow measured at Betrawati (600 m) and 7.4% at Rabuwa Bazaar (470 m), about 50 km from glacier termini. Glacier ice melt contribution decreased to 4.5% of annual discharge further downstream in Trishuli basin (at 325 m, about 75 km from glacier termini). At low elevations, debris-covered tongues contributed a small percent (1.1% and 3.0%) to measured discharge at Betrawati and Rabuwa Bazaar stations, respectively. We independently evaluated the ice ablation approach with synoptic sampling of stable water isotopes (δ18O and δD) collected during the post-monsoon season to quantify the contribution of various sources of water to river flow. Mixing models showed groundwater to be an important component of river flow within only tens of kilometers of the glacier outlets in the post-monsoon season.

1. Introduction

[2] Mountain glaciers constitute a significant component of the hydrologic regime of high alpine catchments, including parts of the Himalayas [Immerzeel et al., 2010; Kaser et al., 2010; Thayyen and Gergan, 2010]. Understanding the timing and spatial patterns of ice melt is key for planning water resources for irrigation, hydropower generation, and consumption in areas where glaciers may constitute important water resources. A number of studies have reported glacier shrinkage and negative glacier mass balance trends in the central and eastern Himalaya in the last decades [Bhambri et al., 2011; Fujita and Nuimura, 2011; Fujita et al., 2001; Kääb et al., 2012; Nuimura et al., 2012]. These trends are in contrast with stable or slightly increasing glacier mass balance and area trends in the drier western Himalaya and the Karakoram [Gardelle et al., 2012; Scherler et al., 2011; Vincent et al., 2013]. Nevertheless, there remains a widespread concern about the potential socio-economic implications of glacier changes on regional water supplies in the Indus, Ganges, Brahmaputra, Yangtze, and Yellow rivers [Immerzeel et al., 2012]. A clear understanding of the contribution of glacier ice melt to streamflow at various scales across the Himalaya, including the role of debris-covered ice, is still needed.

[3] Estimates of the water balance vary significantly in literature due to inconsistencies in the methodology used to derive glacier runoff estimates, differences in terminology, various scales of analysis and limited hydrologic measurements. For example, as much as 70% of the annual flow of the Ganges River and its principal tributaries was attributed to “accelerated” glacier melt [Barnett et al., 2005; Singh and Bengtsson, 2004]. In the Indus basin, the role of the monsoon precipitation was found to be important, with 70% of the annual runoff measured at lowlands attributed to seasonal monsoon rains in the foothills and front ranges [Winiger et al., 2005]. At higher elevations, the role of snow and ice melt appears to be increasingly important. Immerzeel et al. [2009] attributed 72% of the streamflow in the upper Indus basin to ice and snow melt (32% and 40%, respectively), with the remaining (28%) due to rainfall. Some studies addressed the lack of field measurements by using hypothetical data and conceptual models to predict climate-induced streamflow changes [Fukushima et al., 1991], or to assist water resource planning [Braun et al., 1993; Rees and Collins, 2006]. Simple models such as the degree-day models have also been used in various studies to estimate glacier runoff in some Himalayan basins [Azam et al., 2012; Immerzeel et al., 2009, 2010; Kayastha et al., 2000; Li and Williams, 2008]. While easy to use and quite robust, applying these models over larger areas in the Himalaya is often hampered by the lack of field measurements needed to estimate key model parameters, most notably air temperature.

[4] In this study, we propose a simple “ice ablation” approach in combination with ASTER remote-sensing data (glacier area and elevation) to determine the contribution of clean and debris-covered ice melt to streamflow runoff in the central-eastern Himalaya (Trishuli and Dudh Kosi basins in Nepal). We complement this method with synoptic measurements of stable isotopes of oxygen and hydrogen for estimating the relative contributions of snow/ice melt to river discharge during the post-monsoonal time period. Stable isotope ratios of surface waters and precipitation have been used in several studies to identify sources of water on northern slopes of the Himalaya and the Tibetan plateau [Aizen et al., 1996; Hren et al., 2009; Pande et al., 2000; Shichang et al., 2002] and in the western Himalaya [Garzione et al., 2000; Karim and Veizer, 2002]. However, such studies remain scarce in the monsoon-influenced area of the Himalaya, where the hydrologic regime is complicated due to summer accumulation and ablation occurring concomitantly on the glaciers. By combining remote sensing and isotopic analysis, we address three specific questions: (1) What is the contribution of glacier ice melt to discharge on an annual basis, and how does this change as basin area increases to include lower-elevation areas where the majority of the population lives? (2) What is the contribution of debris-covered ice melt to streamflow? and (3) Are there sufficient differences in the isotopic content of potential source waters to parameterize hydrologic mixing models to evaluate the ice-melt component as the ablation model? The overarching goal of this study is to develop a simple approach for estimating the contribution of glacial melt to water sources at a regional scale in a data-sparse area, where the local hydrology is complicated by interaction of the Asian monsoon with complex topographic relief.

2. Study Area

[5] Our study areas are situated in the monsoon-dominated central-eastern Himalaya (Trishuli and Dudh Kosi basins) (Figures 1a and 1b). Trishuli basin is part of the Narayani river basin, located approximately 60 km north of Kathmandu, and includes the Langtang valley (Figure 1a). Elevations in Langtang valley range from 325 to ∼7345 m. The upper glacierized catchment is called Langtang Khola watershed, situated above the Kyangjin climate station (meteorological station: 3920 m and hydrologic station: 3800 m). The hydrology and climate of this watershed have been studied in the field by Japanese research teams for more than 30 years [Kayastha et al., 2005; Seko, 1987; Takahashi et al., 1993; Yamada et al., 1992], including intermittent mass balance measurements on Yala glacier since 1985 [Fujita et al., 1998].

Figure 1.

Study area and the scales of analysis: (a) Trishuli subbasin with the three scales of analysis corresponding to the three runoff stations: Trishuli, Betrawati, and Kyangjin (also known as Base House, BH). The Narayani runoff station is also shown. (b) Dudh Kosi subbasin with the nested Hinku watershed and the two stations with discharge measurements (P1 and Rabuwa Bazar). The sampling sites in the two basins are also shown. Elevations are based on SRTM DEM.

[6] Dudh Kosi basin is located in the Solu-Khumbu region of Nepal, approximately 150 km east of Kathmandu, and is part of the larger Sapta Kosi river basin (Figure 1b). Elevations range from 451 m at the valley bottom to 8848 m at the summit of Mt. Everest, with rugged, glacierized terrain. Mera glacier, a small debris-free glacier at the headwaters of the Hinku River, has been monitored for mass balance since 2007 by the French Institute for Research and Development (IRD) and their partners at the Department of Meteorology and Hydrology (DHM) in Nepal within the framework of the GLACIOCLIM project. Glacier topography in these study sites is characterized by the presence of debris-cover on the ablation area of many of the glacier tongues, reaching thicknesses of up to 2 m at the termini [Kayastha et al., 2000].

[7] Climatically, this part of the Himalaya is dominated by the south-west Asian summer monsoon circulation system, caused by mid-troposphere heating over the Tibetan plateau during the summer and the inflow of moist air from the Bay of Bengal to the continent [Benn and Owen, 1998; Yanai et al., 1992]. The warm air masses interact with the topography of the Himalaya and Tibetan plateau (HTP), causing maximum precipitation at low to moderate elevations on the south slopes of the Himalaya during the summer months (June to September) [Bookhagen and Burbank, 2006; Shrestha, 2000]. The central-eastern Ganges plains, where our study sites are located, receive heavy orographic rainfall during the summer months (∼80% of the annual rainfall) [Ageta and Higuchi, 1984; Bookhagen and Burbank, 2006]. This causes a characteristic “summer-accumulation” glacier regime, with accumulation and ablation occurring simultaneously in the summer [Ageta and Higuchi, 1984].

3. Methodology

3.1. Hydrometeorologic Data and Basin Delineation

[8] Stream gauge and meteorologic data needed to estimate the glacier contribution to streamflow were obtained from the DHM, Nepal. We averaged data for the period 1988–2006 for four stations: (a) daily precipitation data for the Kyangjin meteorological station and (b) runoff data from Kyangjin station, Betrawati, and Narayani hydrological stations in the Narayani river basin, and Rabuwa Bazar in the Dudh Kosi river basin (Table 1). We calculated long-term discharge averages (1988–2006) for the four gauged hydrologic stations listed in Table 1.

Table 1. List of DHM Hydrometeorological Stations Used in This Study
Station NameStation CodeRiver BasinLatitudeLongitudeElevation (m)Start RecordEnd RecordStation Type
Rabuwa Bazar670Koshi27.2786.6546010/3/6412/12/06Hydrologic

[9] Watersheds corresponding to each runoff station were delineated using the hydrologic modeling functions in GIS and elevation data from the Shuttle Radar Topography Mission (SRTM) v.4 (CGIAR), a hydrologically sound, void-filled DEM [CGIAR-CSI, 2004]. The vertical accuracy of the SRTM DEM in this area, calculated as root mean square (RMS) with respect to 25 field-based ground control points (GCPs) was 31 ± 10 m. The GCPs were obtained in the field in 2008–2009 on non-glacierized terrain such as roads and bare land using a Trimble Geoexplorer XE series GPS unit.

[10] In addition to the existing gauged stations, we digitized two additional runoff points denoted as “Trishuli” (for Trishuli wastershed) and “P1” (for Hinku watershed) on Figures 1a and 1b, respectively. The general relationship between discharge and drainage area can be empirically cast as equation (1):

display math(1)

where Q is river discharge (m3 s−1), k is a constant that has been shown to generally not be useful, A is upstream drainage area (km2), and c is the scaling power relating A and Q [Galster, 2009]. A c value of 1 or nearly 1 represents a linear relation between discharge and drainage area and is caused by different areas of a drainage basin that contribute runoff at similar rates. Galster [2009] has shown that for rivers evaluated in the United States, about half have c = 1, and many of the basins where c did not equal 1, there were extenuating factors such as dams and extensive urbanization. Assuming c = 1, then discharge at these ungauged points can be estimated using a ratio of the watershed sizes with the closest runoff stations, following Helsel and Hirsch [1992]. For Langtang, we averaged average annual discharge measurements from two neighboring runoff stations (equations (2) and (3)): one above the Trishuli point (Betrawati, 600 m) and one below it (Narayani, 150 m) (Figure 1a):

display math(2)
display math(3)

[11] where Q is in km3 and A is the watershed area in km2. For Dudh Kosi, we calculated discharge at P1 based on measured values at Dudh Kosi station (Rabuwa Bazar, 460 m) (equation (4)):

display math(4)

3.2. Remote Sensing of Glacier Parameters

[12] Glacier datasets for this study were derived from ASTER scenes from October 2003 (Trishuli basin) and December 2005 (Dudh Kosi basin) (Table 2). The scenes had high contrast over glaciers, minimal cloud cover, and were acquired at the end of the ablation season to minimize seasonal snow. We used the orthorectified products (ASTDMO14) and the atmospherically corrected surface reflectance products (AST_07XT) for these scenes.

Table 2. Summary of the Satellite Imagery Used for Glacier Delineation
SatelliteAreaScene IDDateSpatial ResolutionNotes
ASTERKhumbuSC:AST_L1A.003:203226863815 Dec 200515m VNIRGood contrast, no snow/clouds
LangtangSC:AST_L1A.003:201832148130 Oct 200330m SWIR
  90m TIR
IKONOS-2Khumbu20081103050432100000116174303 Nov 20081m PAN (nadir)0% clouds
  4m MSI 
Langtang200311100510259000001161525122 Oct 20031m PAN (nadir)7% clouds
200310220518062000001162164222 Oct 20034m MSI0% clouds
200310220518062000001162164222 Oct 2003 0% clouds

[13] Basin-wide clean ice was delineated using ASTER band ratios ¾ with a threshold of 2.0 (ASTER ¾ > 2 = ice) and a 3 × 3 median filter, following protocols described in Racoviteanu et al. [2009]. Debris-covered ice was delineated manually using on-screen digitizing on false color composite images (ASTER 321 and 543). We validated these outlines with high-resolution IKONOS-2 imagery obtained from Geoeye archives for the Trishuli and Dudh Kosi basins (Table 1). The IKONOS scenes have one panchromatic band (0.82 m pixel size) and four multispectral bands in visible-near infrared (VNIR), a 3.2 m pixel size, and a swath width of 11.3 km. The standard geometrically corrected IKONOS scenes were orthorectified using rational polynomial coefficients (RPCs) and the SRTM DEM. False color composites of VNIR IKONOS bands (FCC 432) were used for manual editing of ASTER-derived glacier boundaries in areas covered by shadow, clouds, and debris-covered ice, where glacier mapping is notoriously difficult [Racoviteanu et al., 2009]. Additionally, we cross-checked our glacier outlines against data from the Randolph inventory [Arendt et al., 2012].

[14] We averaged atmospherically corrected radiance values from the ASTER surface reflectance product to calculate a broadband albedo from ASTER surface reflectance bands 1, 3, 5, 6, 7, and 9 [Greuell and Oerlemans, 2004; Greuell et al., 2002]. We digitized 150 snow line altitudes (SLAs) for the Trishuli basin (2003) and 100 SLAs for Dudh Kosi basin (2005) on the basis of differences in albedo values for ice, snow, and debris cover [Paterson, 1994]. We used the assumption that the SLA visible on satellite images at the end of the ablation season can be considered an approximation of the yearly ELA [Paterson, 1994; Rabatel et al., 2005]. We averaged digitized points in two ways to estimate the regional ELA (average altitude at which mass balance is zero for a given glacier in particular climatic and topographic settings): (1) all digitized pixels along the snowlines and (2) all mean snowline altitudes on separate glaciers. We compared the ELAs derived by remote sensing with outputs from other existing methods (e.g., median elevation of glaciers and frequency distribution of ice elevations [Benn et al., 2005]). Spatial patterns of ELAs were examined for each basin using basic statistics and trend analysis in a raster GIS environment. We conducted a sensitivity analysis of the ELA values (±200 m) to quantify its potential effect on the glacier ice melt estimates.

[15] Glacier ice hypsometry (the area-altitude distribution of ice masses) was derived from the glacier outlines and the SRTM DEM by extracting ice elevations on a pixel-by-pixel basis from the DEM for each 100 m elevation band. These were converted to area [km2] and used as input in the ice ablation model (section 'Ablation Model Description'). Glacier parameters such as glacier termini elevation and median elevation of the ice masses were also extracted from the ice outlines and the SRTM DEM.

3.3. Ablation Model Description

3.3.1. Model Setup

[16] For any time period, discharge in a glacierized watershed can be described using the generalized hydrologic balance equation (equation (5)):

display math(5)

where Q is the mean discharge in the watershed [km3], Pr and Ps are inputs from rain and snow, Ms and Mi are runoff generated from snow and ice melt, ET is evaporation/sublimation, and ΔS is the change in groundwater/glacier storage, all in km3 [Fukushima et al., 1991; Yamada and Motoyama, 1988]. In this paper, we only focus on the glacial ice melt component (Mi) and do not quantify any additional basin runoff released from temporary storage of water as ice during years of negative mass balance (glacier wastage). Here we follow the approach of Rupper et al. [2012], acknowledging that this study does not account for mass loss due to sublimation (underestimates total mass loss), and it assumes that all ice melt leaves the glacier system (overestimates total mass loss).

[17] Since ablation measurements are scarce in the Himalaya, and limited to a few glaciers, we estimate glacier ice melt using a straightforward “ice ablation model”, initially developed by Alford [1992] and subsequently tested on nine glacierized basins in the Nepal Himalaya by Alford et al. [2010]. Parameterization of the ice ablation model relies on the three key glacier variables: ice area, hypsometry, the basin-wide ELA described in section 'Remote Sensing of Glacier Parameters' and an ice ablation gradient described in section 'Ice Ablation Gradient'. General steps in the ice ablation model (Figure 2) involved: (1) delineating the clean and debris-covered glacier boundaries; (2) calculating the basin-wide ELA and ice ablation gradient (section 'Ice Ablation Gradient'); (3) computing the ice hypsometry for 100 m altitudinal bands below the ELA using the glacierized area and the DEM; (4) multiplying the area of each altitudinal band by the ablation gradient values to obtain the ice melt volume per elevation band; and (5) summing up the ice melt volumes to obtain the basin-wide runoff volume (equation (6)):

display math(6)

where Q is the glacier ice melt volume [m3], bn is the specific ice melt [m] computed for each altitudinal band, and Ai is the area of each altitude band in the ablation zone of the glaciers [m2].

Figure 2.

Conceptual diagram for the workflow used in this study.

[18] Steps 3–5 were performed for clean ice and debris-covered ice separately, yielding the annual volume of water resulting from each of the two types of surface. We compared these amounts to the discharge measured at gauging stations along an elevation gradient to determine the percent contribution of clean and debris-covered ice to streamflow at downstream locations in the two selected basins and subbasins.

3.3.2. Ice Ablation Gradient

[19] The ice ablation gradient was initially referred to by Haefeli [1962] and subsequently denoted by various terms such as the “activity index” [Meier, 1962; Meier and Tangborn, 1965], “mass balance gradient” [Konz et al., 2006], or “vertical budget gradient” (VBG) [Kaser, 2001; Kaser and Osmaston, 2002]. In this study, we denote the ice ablation gradient as δb/δz, defined as the water equivalent of ice melt per one hundred meters vertical elevation change on the ablation zone of the glacier (m/100 m), starting with zero at the ELA. Ablation gradients are traditionally measured in the field using the glaciologic method [Braithwaite, 1984; Kotlyakov and Krenke, 1982; Rabatel et al., 2005; Wagnon et al., 2007], and they vary across a mountain range depending on local climate, size, and aspect [Wagnon et al., 2013]. For clean ice, the ablation gradient was chosen by averaging existing field measurements on Himalayan glaciers: Chhota Shigri in Himachal Pradesh, India (average δb/δz = 0.69 m/100 m, 2002–2006) [Wagnon et al., 2007]; Yala glacier, Langtang Khola watershed, Nepal (0.81 – 1.3 m/100 m, 1996) [Fujita et al., 1998; Konz et al., 2006]; Glacier AX010, Shorong Himal, Nepal (0.81 – 0.9 m/100 m, 1996–1999), Rikka Samba glacier (0.64 m/100 m, 1999) [Fujita et al., 2001], Mera glacier (0.48 m/100 m, 2007–2012), and Naulek glacier (0.85 m/100 m, 2007–2012), all in Khumbu Himalaya, Nepal [Wagnon et al., 2013]. Averaging these values yielded δb/δz = 0.8 m/100 m, which we used as a linear ablation gradient for clean ice (Figure 3). To quantify uncertainty and to examine the potential impact of future climate forcing on glacier runoff estimates, we conducted sensitivity analyses on δb/δz (boundary values of 0.6 m and 1.3 m/100 m).

Figure 3.

Ice ablation gradients used in this model (black triangles: linear ice ablation gradient of 0.8 m/100 m; gray circles: nonlinear debris covered ice ablation gradient, starting with 0.6 m/10 m below the ELA).

[20] The debris-covered ice ablation gradient is generally nonlinear due to the variable effect debris cover thickness on sub-debris glacier ice melt rates. Observations on several glaciers (Rakhiot glacier in the Indian Himalaya, Khumbu glacier in the Nepal Himalaya, Baltoro and Barpu glaciers in the Karakoram, Pakistan, and Miage glacier in the Italian Alps) showed the variable effect of debris cover on meltwater rates [Brock et al., 2010; Mattson et al., 1993; Mihalcea et al., 2008a; Zhang et al., 2011]. In general, very low ablation rates are observed at the glacier termini, where the debris cover is often thick and continuous and accelerated ice ablation rates are observed in the middle to upper part of the debris covered ice, where the debris cover is patchy and thin [Benn and Lehmkuhl, 2000]. In this study, we used a generalized mass balance curve for the debris covered ice, similar to those estimated by Benn and Lehmkul [2000] and Osmaston [2005]. We started with a gradient of 0.6 m/100 m below the ELA in the upper part of the debris cover, increased the ice melt until ∼5100–5200 m, then decreased the gradient nonlinearly until zero toward the terminus to reflect the insulating effect of debris (Figure 3). We conducted a sensitivity analysis using the same curves with values of 0.4–1.3 m/100 m to account for the high uncertainty in debris cover melt rates.

3.4. Stable Isotope Measurements

[21] Liquid water samples were collected as grab samples during two synoptic surveys in Trishuli and Dudh Kosi basins at the end of the ablation season (November 2008 and 2009, respectively). Samples were collected along an elevation gradient on the main trekking routes as well as side valleys in both basins. The sample locations were recorded with a handheld Trimble Geoexplorer XE GPS unit, with an average horizontal accuracy of ∼5 m and a vertical accuracy of ∼9 m (calculated based on the position of the satellites at the time of the sampling). We consider water samples collected at the snout of glaciers to represent an integrated measure of water emitting from the glacial system, which is some combination of melting glacial ice and melting snow. Water samples collected from springs were assumed to represent groundwater. Snow samples were collected in the Dudh Kosi basin only, along the climbing route on Mera Peak from the base camp (∼5300 m) to the summit (6350 m) (Tables 3 and 4 and Figures 1a and 1b).

Table 3. Water Samples Collected Along the Trishuli River and Its Tributaries
CodeSite DescriptionSampleSample Elevation (m)Headwater Elevation (m)δ18O (‰)δD (‰)
P-1TrishuliTrishuli river490.0−11.67−87.23
P-2Trishuli river at Syabrubesi after remixingTrishuli river1408.0−11.64−89.22
P-3Bhote Kosol villageTrishuli river1414.5−11.86−88.88
P-4Langtang KholaTrishuli river1432.9−12.83−96.78
P-5Spring sample (Groundwater)Trishuli river1663.05101−13.30−99.45
P-6Lama Hotel, Langtang KholaTrishuli river2447.35848−14.00−103.20
P-7Ghoratabela villge, Langtang KholaTrishuli river2987.47034−13.66−102.82
P-8Mixing from clean and debris at bridgeTrishuli river3853.76257−12.31−92.97
P-9Langtang River at Kyangin GompaTrishuli river3776.86302−14.64−109.47
P-10Lirung outlet at bridgeTributary3982.17183−12.50−90.11
P-13Langtang glacier, west outletTributary4660.56166−13.16−101.16
P-14Langtang glacier, main outletTrishuli river4464.25690−15.15−113.92
P-15Small clean glacier above Langtang RiverTributary4220.84570−14.36−105.59
P-16Langtang River before LangshishaTrishuli river4111.06911−15.20−112.25
P-17ALangtang River after LangshishaTrishuli river4096.66302−15.19−113.24
P-17BSalbhachum glacier, side valleyTributary3994.66257−13.43−102.26
P-18Yala glacier near Kyanjin GompaTributary3868.17176−13.55−102.15
P-20Small clean glacier above LangtangTributary3548.45418−11.82−86.86
P-21Side valley, after LangtangTributary3357.95101−13.56−97.37
P-22At bridge before Baub villageTrishuli river1993.05848−11.98−86.97
Table 4. Water Samples Collected Along the Hinku River and Tributaries, Dudh Kosi Basin
CodeSite DescriptionSampleElevation (m)δ18O (‰)δD (‰)
M01AHinku riverMain river3370.26−14.99−108.47
M01BHinku riverMain river3370.26−14.96−108.50
M02Kote villageMain river3514.36−15.11−109.26
M03Sabai glacier outletTributary4256.20−13.87−98.43
M04Above Tangnag villageMain river4263.58−16.60−121.50
M05After mixingMain river4429.10−16.52−121.68
M06Dig glacier outlet before mixingTributary4430.77−16.00−115.66
M07Debris cover glacier outletTributary4584.43−16.72−122.73
M08AMera glacier at DigTributary4620.31−16.32−118.54
M08BMera glacier at DigTributary4620.31−16.19−118.58
M09Snow above base camp MeraSnow5306.25−16.93−124.84
M11Snow above high camp MeraSnow5775.03−23.18−172.15
M13Snow on Mera glacierSnow6160.76−21.13−158.24
M14Snow below summitSnow6383.30−22.65−168.47
M16Spring (represents groundwater)Spring4300.00−12.75−89.75

[22] Samples collected for geochemical analysis in Trishuli basins were filtered through a 47 mm Costar Nucleopore 0.7 µm membrane filter. All samples of snow, ice, and spring water were shipped to the Institute of Arctic and Alpine Research (INSTAAR) at University of Colorado-Boulder and analyzed at the Kiowa Wet Chemistry Laboratory. In this study, we focus only on hydrogen and oxygen isotopes (δD and δ18O). Isotopic analyses of 18O were conducted using the CO2–H20 equilibration technique at the Stable Isotope Laboratory at the Institute of Arctic and Alpine Research in Boulder, CO. The 18O values are expressed in conventional delta (δ) notation in units of per million (‰) relative to Vienna Standard Mean Ocean Water (V-SMOW), with a precision of ±0.05‰ (equation (7)):

display math(7)

[23] The δD is calculated in a similar fashion, with a precision of ±0.5‰. The new and old water components are estimated using δ18O [e.g., Liu et al., 2004] by (equation (8)):

display math(8)

with the water balance constraint that (equation (9)):

display math(9)

where Q is volume flow rate, C is δ18O content, and the subscripts indicate the water source. Here “new” water is water released from the glacial outlet streams to the catchment and “old” water is water already stored in the catchment before the release of the new water from glacial outlet streams.

[24] We analyzed spatial patterns of δ18O samples in relation to proximity to glaciers and elevation of the headwaters using the SRTM DEM and grid-based functions in GIS. Hydrologic functions were used to determine the upstream contributing area of each sampling site as well as to extract the median and maximum elevation of the headwaters. A regression analysis was conducted on to δ18O values with sample elevations and headwaters elevation to evaluate spatial patterns in δ18O.

4. Results

4.1. Precipitation and Runoff Patterns in Langtang

[25] Discharge and precipitation patterns at Kyangjin station in the Langtang Khola watershed are shown in Figure 4. Daily discharge measurements at this station show a clear monsoon signal, with a sharp increase in runoff starting in June and a maximum in late July/early August. The maximum discharge in July to August (summer monsoon) is due to increased glacier melt resulting from high rainfall, high solar radiation and high air temperatures [Thayyen and Gergan, 2010]. The hydrograph has only one peak, confirming that snow and ice melt happen concomitantly in this monsoon-dominated area of the Himalaya. This is in contrast with other areas such as the Yarkant River which flows north from the Karakoram Mountains, where two peaks of discharge from snow and ice are clearly visible on the hydrograph [Li and Williams, 2008], indicative of a snow-dominant alpine regime in the spring and glacier melt in the late summer [Thayyen and Gergan, 2010].

Figure 4.

Precipitation and runoff patterns in Langtang Khola Wastershed: (a) daily discharge patterns at Kyangjing station, averaged monthly for the length of record; (b) monthly precipitation patterns at Kyangjing meteorological station, averaged for the length of record. Data are from the Department of Hydrology and Meteorology (DHM), Nepal.

[26] The monthly precipitation graph (Figure 4b) shows maximum precipitation during the months of June to August. For the available period, we calculated that 73% of the annual precipitation occurred during the monsoon, which is consistent with trends noted by Ageta and Higuchi [1984] and Shairawa et al. [1992], based on independent field measurements in the same basin. The average annual precipitation at Kyangjin was 646.5 mm for this period, in close agreement with values reported in a previous study by Kayastha et al. [2005]. Discharge reached its maximum in the first half of July (4 mm/day) and then decreased to about 0.6 mm/day during the winter months. The low values of winter runoff are generally correlated with mean temperatures below 0°C [Motoyama et al., 1987] and are considered base flow values. Such discharge patterns are characteristic of a monsoon-dominated system, in contrast with the western areas of the Himalaya, which experience less summer monsoon precipitation [Bookhagen and Burbank, 2006]. There is a seasonal dependency of precipitation with altitude in this area: precipitation peaks at lower elevations and then decreases with altitude up to about 4000 m during the monsoon period due to reduced water vapor and increases with altitude during the winter due to uplift of moist air from synoptic (western) disturbances [Seko, 1987]. Interestingly, however, annual precipitation amounts at Kyangjin (541 mm) are comparable to those measured at the outlet of Langtang glacier at 5300 m (556 mm). Shairawa et al. [1992] showed that Kyangjin station receives less precipitation than similar elevations in the valley and elsewhere due to its location at the bottom of a narrow valley. Monsoon winds bring precipitation on the southern slopes of the valley, but their movement northward is blocked by the east-west topographic barrier. As a result, the valley bottom receives less precipitation than south-facing upper ridges of Langtang valley.

[27] Similar to precipitation, specific discharge in these basins increases as elevation decreases (Table 5). Long-term mean annual specific discharge at the high-elevation hydrologic station of Kyangjin was 0.6 m/yr, which is considerably lower than specific discharge at the stations downstream: Betrawati (2.1 m/yr) and Trishuli (2.2 m/yr). Higher amounts of specific discharge at lower elevations are induced by higher amounts of precipitation at low elevations, driven by monsoon rains in this area. This is in contrast to most midlatitude mountain environments where discharge at low altitudes is driven mainly by snow and ice melt at high elevations, and specific discharge generally increases with elevation [Williams et al., 2011].

Table 5. Characteristics of the Basins and Subbasins Used for the Ablation Modela
BasinAverage Basin Elevation (m)Basin Area (km2)Average Annual Measured Discharge (km3)Glacierized Area (% Basin)Ice-Melt Contribution (%)
Clean IceDebris-Covered IceTotalClean IceDebris-Covered IceTotal
  1. a

    Also shown is the contribution of ice melt to streamflow runoff in each basin for these two years, discussed later in the text (section 'Ice Ablation Model Results').

Langtang Khola5170352.
Dudh Kosi37363711.

4.2. Glacier Area and ELA

[28] The total percent glacierized areas (clean and debris-covered ice) for each of the nested watersheds in Trishuli basin (Langtang Khola, Betrawati, and Trishuli) was 43.5%, 18.6%, and 11.4%, respectively (Table 5). Debris-covered ice covered 7.8%, 3.0%, and 1.9% of the basin areas, respectively. The ablation area of glaciers in the Trishuli basin was 330 km2 (56% of the total glacierized area), of which 70.7 km2 was situated in Langtang Khola watershed (Table 6). Our estimates of the glacier areas are in close agreement with Immerzeel et al. [2012], who calculated 46% glacierized area for the Langtang Khola watershed. In the Dudh Kosi basin, the percent glacierized areas were 34.7% (Hinku watershed) and 13.9% (entire Dudh Kosi). Of this, debris-covered glaciers covered 5.6% and 3.7% of the basin areas, respectively (Table 5). The ice ablation area covered 261.5 km2 or 51% of the entire glacierized area in Dudh Kosi, with 27.7 km2 situated in the Hinku watershed (Table 6).

Table 6. Glacier Characteristics Per Basin, Derived From SRTM DEM and ASTER-Based Glacier Outlines
BasinGlacier Elevation (m)Glacier Area (km2)% Total Area
Min ZMed ZMax ZAccumulationAblationTotalAccumulationAblation
Langtang Khola39975552718382.670.7153.35347
Dudh Kosi434955408132248.1261.5509.64951

[29] Glacier elevations in the Trishuli basin ranged from 3608 m at the glacier termini to 7345 m at headwaters, with a median of 5445 m (Table 6) and 4349–8132 m with a median of 5540 m in the Dudh Kosi basin (Table 6). Glacier termini elevations were ∼740 m higher in Dudh Kosi basin compared to the Trishuli basin. The histogram of glacier elevations (Figures 5a and 5b) shows that most glacier area is situated in the elevation band of 5400–5500 m for Trishuli and 5500–5600 m for Dudh Kosi basins, respectively (black lines on Figure 5).

Figure 5.

(a) Glacier hypsometry for the Trishuli basin and subbasin; (b) glacier hypsometry for the Dudh Kosi basin and subbasin, derived from glacier outlines and the SRTM DEM. Light gray bars represent hypsometry for the large basins (Trishuli at Betrawati and Dudh Kosi at Rabuwa Bazar, respectively); dark gray bars represent the hypsometry for the small nested watersheds (Langtang Khola and Hinku, respectively). The elevation band corresponding to the basin-wide average ELA derived from the broadband albedo is shown in black.

[30] The elevations of the ELAs in the Trishuli basin ranged from 4869 to 6231 m, with a mean of 5468 m and a standard deviation of 251 m (Figure 6). In the Dudh Kosi basin, the ELA elevations ranged from 5172 to 6047 m, with a mean of 5568 m and a standard deviation of 198 m. The distribution of ELAs shows a spatial trend, with values increasing from southwest to northeast (Figure 6). The trend is represented by a tilted surface, which dips at a rate of 5 m per 1 km change in northing, oriented toward the NE (22.6 degrees for Trishuli and 51.3 degrees for Dudh Kosi). These findings reflect the orographic forcing of monsoon air masses over the Himalaya, which causes a reduction in the moisture content of the masses as they ascend over the topographic barrier. Benn and Owen [2005] showed that the drier climate on the northern slopes results in higher ELAs (6000–6200 m) compared to ELAs in the southern slopes (4600–5600 m). Our results are in agreement with these findings.

Figure 6.

Broadband albedo derived from ASTER bands 1, 3, 5, 6, 7, and 9 for a subset of the Langtang region. The ELAs for individual glaciers, digitized based on albedo values for snow, ice, and debris covered ice are also shown [Paterson, 1994].

[31] ELA values for Trishuli are consistent with field-based measurements on Yala glacier in the same area (ELAYala ∼ 5300–5400 m) [Fujita et al., 1998; Konz et al., 2006; Morinaga et al., 1987]. ELA values for Dudh Kosi are within the range of yearly ELA values measured on two glaciers in Khumbu and reported in Wagnon et al. (submitted manuscript, 2013) (ELAMera = 5615 m for 2007–2012; ELAPokhalde = 5625 m for 2009–2010). Similar values for modern ELAs were reported in Benn and Owen [2000] for the Langtang valley in Trishuli basin (ELALangtang = 5320 m), and the south side of the Everest area (ELAEverest = 5200 m). We consider these values to be representative of present-day ELA in these areas (ELATrishuli = 5468 m, and ELADudh Kosi = 5568 m).

4.3. Ice Ablation Model Results

[32] Estimates of glacier melt and its contribution to measured discharge at various points along Trishuli and Koshi rivers are presented in Table 5 and Figure 7. Long-term mean annual measured discharge values at Betrawati (Trishuli basin) and Rabuwa Bazar (Dudh Kosi basin) were 7.3 and 6.2 m3, respectively). Long-term mean annual discharge values at the smaller watersheds (Kyangjin in Langtang Khola and P1 in Hinku) were 0.2 and 0.3 m3, respectively (Table 5). The ice ablation model yielded a significant contribution of glacier ice melt to river flow in Langtang Khola watershed (∼58.3% total glacier ice melt contribution at Kyangjin hydrologic station). The total ice melt contribution decreased with elevation to about 9.5% of streamflow at Betrawati (600 m), ∼50 km from the glacier termini, and 4.5% of streamflow at Trishuli (150 m), ∼75 km downstream from the glacier termini. The contribution of debris-covered glaciers alone to measured streamflow was 17.7% of streamflow at Kyangjin and decreased downstream to 0.5% of discharge at Trishuli (Table 5).

Figure 7.

Contribution of glacier runoff to annual streamflow at three points along Trishuli in the Narayani river basin: Trishuli (estimated discharge), Betrawati, Kyangjin (measured discharge), and the percent glacierized area in each basin.

[33] In the Dudh Kosi basin, the total contribution of glacier ice melt was 21.1% of the streamflow estimated at P1 (∼3824 m), about 7 km downstream from the outlet of glaciers, and decreased to 7.4% of the measured discharge at Rabuwa Bazar hydrologic station at 460 m, about 50 km from the glacier termini. The contribution of debris covered ice to measured streamflow decreased from 4.1% of discharge at P1 to 3.0% at Rabuwa Bazaar (Table 5).

[34] Sensitivity tests performed on the ELA (±200 m) for clean ice showed that the ablation model was highly sensitive to the choice of ELA (Table 7). For example, at Kyangjin station in Langtang Khola watershed, an increase in ELA by 200 m while maintaining the same ablation gradient at 0.8 m/100 m caused a 45% increase in glacier ice melt contribution to runoff. The clean ice melt estimated under this scenario was 85.3% of the measured discharge at Kyangjin (Table 7). Lowering the ELA by 200 m caused the glacier contribution to decrease by 23% of the measured discharge compared to normal ELA. In the Hinku basin, raising the ELA by 200 m doubled the ice contribution to increase compared to a normal scenario, and lowering the ELA by 200 m caused a 10% decrease in the ice melt contribution.

Table 7. Sensitivity of the Clean Ice Melt Model to ELA (± 200 m) and Ice Ablation Gradient for the Three Basins in Trishuli and Dudh Kosi (δb/δz = 0.6 – 1.3 m/100m)
ParameterScenarioLangtang Khola (Kyangjin)Trishuli (Betrawati)Trishuli
Clean Ice Runoff (km3)Clean Ice Contribution (%)Clean Ice Runoff (km3)Clean Ice Contribution (%)Clean Ice Contribution (%)
ELAELA = 5268 m0.0417.
ELA = 5468 m0.0940.
ELA = 5668 m0.1985.
δb/δzδb/δz = 0.6m/100 m0.0930.
δb/δz = 0.8m/100 m0.1440.
δb/δz = 1.3m/100 m0.1565.
ParameterScenarioHinkuDudh Kosi
Clean Ice Runoff (km3)Clean Ice Contribution (%)Clean Ice Runoff (km3)Clean Ice Contribution (%)
ELAELA = 5368 m0.
 ELA = 5568 m0.0417.10.34.4
 ELA = 5768 m0.0934.60.59.0
δb/δzδb/δz = 0.6m/100 m0.0312.80.23.8
 δb/δz = 0.8m/100 m0.0417.10.34.4
 δb/δz = 1.3m/100 m0.0727.80.47.1

[35] Sensitivity analysis performed on the nonlinear debris-covered ice ablation gradient (0.4 m/100 m to 1.3 m/100 m) for the two basins is presented in Table 8. A larger ablation gradient corresponds to a thin, patchy debris cover, where ablation rates are close to those of clean ice; a smaller ablation gradient corresponds to thick debris cover, which reduces the melt rates of the ice underneath the debris [Foster et al., 2012; Mihalcea et al., 2008b]. At Kyangjin, the contribution of debris-covered glaciers was 7.6% of the streamflow runoff assuming a smaller ablation gradient (0.4 m/100 m) (Table 8). A larger ablation gradient (1.3 m/100 m) increased the contribution of the debris-covered ice to 30.2% of the measured discharge at Kyangjin.

Table 8. Sensitivity of the Ablation Melt Model for Debris-Covered Ice to the Choice of the Ice Ablation Gradient for the Three Basins in Trishuli and Dudh Kosi (δb/δz = 0.4 – 1.3 m/100 m)
ParameterScenarioLangtang Khola (Kyangjin)Trishuli (Betrawati)Trishuli
Debris-Covered Ice Runoff (km3)Debris-Covered Ice Contribution (%)Debris-Covered Ice Runoff (km3)Debris-Covered Ice Contribution (%)Debris-Covered Ice Runoff (km3)Debris-Covered Ice Contribution (%)
δb/δzδb/δz = 0.4m/100 m0.
δb/δz = 0.6m/100 m0.0417.
δb/δz = 1.3m/100 m0.0730.
ParameterScenarioHinkuDudh Kosi
Debris-Covered Ice Runoff (km3)Debris-Covered Ice Contribution (%)Debris-Covered Ice Runoff (km3)Debris-Covered Ice Contribution (%)
δb/δzδb/δz = 0.4m/100 m0.0072.70.122.0
 δb/δz = 0.6m/100 m0.
 δb/δz = 1.3m/100 m0.0311.00.396.4

[36] Similarly, in the small Hinku basin, the contribution of debris-covered ice to discharge at P1 ranged from 2.7 to 11% of discharge when applying these two gradients, respectively.

4.4. Stable Isotopes and Mixing Model Results

[37] We examined trends in oxygen isotopes with elevation for 22 water samples collected in the Trishuli basin and 16 water samples in the Dudh Kosi basin. The δ18O values for the Langtang region ranged from −11.64‰ at the lowest elevation sampling site along the main stem of the river to −15.20‰ on the main stem near the tongue of the main Langtang glacier (Table 1, Figure 8). The distribution of δ18O with elevation (Figure 8a) shows decreasing trends in δ18O with elevation for both Trishuli and Dudh Kosi samples. The rate of decrease in δ18O for river samples only is −0.6‰ per 1000 m elevation for the Langtang and −1.1‰ per 1000 m elevation for the Dudh Kosi. In contrast, the rate of decrease in δ18O for snow samples in Dudh Kosi is four times larger than the rate for river samples (−4.4‰ per 1000 m). The correlation between elevation and δ18O for river samples is significant at 95% confidence interval at both sites (Pearson's r = −0.6). The combined δ18O values for river samples display a spread of about ∼3‰ in the 4000–5000 m elevation range.

Figure 8.

(a) δ18O trends with elevaton for sampling sites in Langang and Dudh Kosi. Open circles denote snow samples on Mera glacier; filled symbols represent surface water sampled. There is a decreasing trend in δ18O with elevation at both sites; (b) δ18O and δD relationship for stream waters in the Langtang and Dudh Kosi study areas. Isotopics compositions of the water follow closely the Global Meteoric Water Line (GMWL), with little different between the two basins.

[38] There was a difference in δ18O values of main glaciers compared to side glaciers. The outlet of the west Langtang Glacier at an elevation of 4660 m had δ18O values of −13.16‰, while the outlet of the main Langtang glacier at an elevation of 4464 m was about 2‰ more depleted in δ18O, with a value of −15.15‰. The δ18O values of samples collected several kilometers below the outlets of smaller glaciers along the side of the main drainage were consistently 1–2‰ more enriched than the outlet of the main Langtang glacier. The δ18O values along the main stem (Langtang Khola) were consistently more enriched with distance downstream. Similar to the Langtang drainage, the δ18O values at Mera became progressively more depleted with increasing elevation, ranging from −12.75‰ at a spring that we believe represents groundwater to −16.32‰ from the outlet of the Mera Glacier near Dig (Table 4; Figure 8a).

[39] It is worth noting that snow collected below the equilibrium line at an elevation of about 5300 m in Dudh Kosi had a δ18O value of −16.93‰, similar to that of glacier outflow collected at the tongue of the glacier near Dig. In contrast, values of δ18O collected above the equilibrium ranged from −21.13 to −23.18‰, which is much more depleted than the glacier outflow. These values suggest the possibility that snow and glacier ice located below the ELA may have overlapping isotopic values, potentially making it difficult to separate snow- and ice-melt contributions below the ELA to surface waters using only stable water isotopes. However, the much more depleted snow values above the ELA are consistent with little contribution or no contribution of snowmelt above the ELA to surface flows, an assumption in our ice ablation model.

[40] We also investigated the relationship between hydrogen and oxygen isotopes in the two basins (δD−δ18O) (Figure 8b). Generally, δD–δ18O in precipitation and surface waters are related according to the relationship δD = 8 δ18O + 10, defined as the Global Meteoric Water Line (GMWL) [Craig, 1961]. The δD–δ18O relationship for all water samples plots near GMWL, with only slight difference between the two basins (Figure 8b), yielding a relationship of (equation 10):

display math(10)

[41] These results suggest very little loss of snow and ice to sublimation or evaporation, similar to other alpine regions [Williams et al., 2006].

[42] The δD–δ18O values of stream water fall along a well-developed mixing line bounded on the depleted end by meltwaters from glacial outflow (“new water”) and on the enriched end by samples collected from springs (“old water”). Because the isotopic waters fall on a well-defined mixing line, two sources of water are sufficient to form stream water [Williams et al., 2009]. Here we assume that water from springs represents groundwater. For the Trishuli basin, we parameterized groundwater from a spring (P5, known as the Landslide spring to trekkers) with a value of −13.30‰. Glacier outflow (some combination of snow and ice melt, along with potential storage of monsoon rain) was estimated from the Langtang River before Langshisha (P16). Mixing models show that the glacier contribution to streamflow was 70% at the gauging station of Kyangjin (3800 m), and 37% at Lama Hotel (1448 m). In the Dudh Kosi basin, the contribution of glacier melt was 95% close to the Mera glacier outlet (4585 m) and 53% at 3370 m (Figure 9). The glacier component decreased rapidly with decreasing altitude in both basins, as the contribution of groundwater increased (73% groundwater contribution within 30 km of the terminus of Langtang glacier and 46% within 10 km of the terminus of Mera glacier). These values are higher than those obtained using the ice ablation model (∼58.3% ice melt contribution at Kyangjin in Trishuli, and 21.2% ice melt within 7 km of the terminus of Mera glacier) and likely reflect conditions during the synoptic survey that were post-monsoon in timing. The consistent decrease in δ18O isotopic values downriver in both Trishuli and Dudh Kosi are consistent with a decrease in the contribution of glacial meltwater to discharge at lower elevation. However, since we don't have independent values of groundwater isotopic content or in rain at lower elevations, we cannot run our mixing models at lower elevations.

Figure 9.

The contribution of melting snow and ice versus groundwater to streamflow in two watersheds: (a) Langtang Khola and (b) Hinku watershed. Overall, the contribution of snow and ice to streamflow decreases to less than 50% within 10–30 km of glacier termini.

5. Discussion

5.1. The Role of Ice Melt in Streamflow

[43] The contribution of the melting ice to annual discharge depends largely on the altitude of the basin and the percent of glacierized area in a basin. We found the contribution of ice melt in a small high-altitude subbasin (Langtang Khola, 43.5% glacierized) to be about five times that of ice melt over the larger Betrawati basin, which has less glacierized area relative to its size (11% glacierized). Similarly, the contribution of glaciers in the Hinku basin (34.7% glacierized) was about three times more than the contribution of glaciers of the entire Dudh Kosi basin (13.9% glacierized).

[44] The total contribution of glacier ice melt to streamflow runoff was proportional to the percent of glacierized area in each basin (Table 5 and Figure 7). Furthermore, the contribution of glacier ice melt decreased rapidly as watershed area increases, similar to results from other studies [Prasch et al., 2012].

[45] Our estimated glacier ice melt contribution to discharge for Langtang Khola watershed (58.3%) is comparable to that of Yamada and Motoyama [1988], who reported 54% ice melt contribution to the measured annual runoff in this watershed for 1985/1986. The difference in the glacier ice contribution may be in part due to the fact that our analysis included the smaller Lirung Khola watershed, while their study did not, and/or to the different periods of measurements. Our calculation of the percentage contribution of glacial melt to annual discharge for the Langtang Khola catchment is reasonably similar to that calculated by Immerzeel et al. [2012] using a much more sophisticated approach. They used a high-resolution, combined cryospheric/hydrological model that explicitly simulated glacier evolution and all major hydrological processes to evaluate source waters for discharge. For their base year of 2005, they report that annual discharge for the Langtang Khola was composed of 45% glacial ice melt, 6% snowmelt, 17% groundwater, and 28% rainfall [Immerzeel et al., 2012]. Our calculation of a contribution of 58.3% glacial meltwater to annual discharge for the Langtang Khola is larger than the 45% calculated for the same basin by Immerzeel et al. [2012], but within the boundaries of our sensitivity analysis (Tables 7 and 8) and provides an independent check that our ice ablation approach appears to work reasonably well. Prasch et al. [2012] also found an ice-melt contribution of 50% to total runoff in a central Himalaya basin, based on a regional climate model and process-oriented and hydrologic modeling.

[46] For Dudh Kosi, our calculated volume of glacier ice melt (0.5 km3 mean annual ice-melt volume for the period 1988–2006) is reasonably close to that calculated by Andermann et al. [2012] of 0.6 km3 using 30 years of precipitation and discharge data from the same stations in Nepal. In our model, debris-covered glacier runoff contributed a smaller percent to measured discharge at high elevations in both basins, compared to clean ice. This is consistent with the presence of thick debris on the glacier tongues, as observed in the field, which may reduce sub-debris glacier ice melt over parts of the glacier [Kayastha et al., 2000]. This may also partly due to the smaller area covered by these glacier tongues compared to clean ice.

[47] Changes in the ELA cannot be decoupled from potential changes in the ice ablation gradient. An increase in ELA under a higher precipitation scenario might imply a steepening of the mass balance gradient, as already noted worldwide [Dyurgerov and Dwyer, 2000]. Varying the clean ice ablation gradient to 0.6 m/100 m and 1.3 m/100 m while keeping the ELA constant induced −10 to +20% difference in the percent of ice melt contribution to annual discharge at Kyangjin (30.4 and 65.9%, respectively) in our study. Similarly, in the Hinku basin, varying the clean ice ablation gradient in the same way induced a change of −5% to +10% in the percent ice melt contribution to discharge. In a potential warming scenario, a steepening of the ice ablation gradient concomitant with a rise in ELA is possible and would result in a higher contribution to glacier runoff than shown here. For example, a rise in the ELA (+26 m) has been documented in this area in the last decades [Kayastha and Harrison, 2008; Kayastha et al., 2005] and was potentially related to increases in air temperature [Shrestha et al., 1999]. However, ELA changes would have to be on the order of a few hundred meters in order to significantly affect the glacier melt contribution to streamflow, as determined from our sensitivity analysis.

[48] In contrast to our isotope results, Maurya et al. [2011] report that glacial ice melt accounted for about 32% of annual discharge for the Ganges river at Rishikesh (372 m) in the Indian Himalaya, west of our study area. This is about six times greater than the contribution we calculated for the low-elevation Trishuli and Dudh Kosi catchments. They employed a three-component mixing model using the values of δ18O and electrical conductivity (EC) of the river water, and its constituents, to estimate the contribution of surface runoff, glacial melt, and groundwater to discharge. They report a δ18O value of −15.3‰ for their glacial meltwater end-member, which is similar to our values for the Langtang and Dudh Kosi basins. Groundwater was parameterized with isotopic and EC values collected from the river during baseflow, and the resulting 15% groundwater contribution appears reasonable. They parameterized the isotopic content of the surface runoff component with values of snowfall in winter and rainfall during the summer monsoon, based on a review of published values. They report an average δ18O of rainwater of −10.1‰, with values of snow ranging from −4 to −14‰, resulting in a value of −10.1‰ for the isotopic value of the surface runoff component. The δ18O values of precipitation (both snow and rain) are poorly constrained in the eastern Himalaya. If the isotopic values of precipitation in their region are more enriched that those used in the mixing model, the surface runoff contribution to annual discharge would increase and the glacial melt contribution would decrease. A potential reason that the ice-melt contribution to annual discharge calculated by Maurya et al. [2011] might be over-estimated is that the isotopic values of precipitation are more enriched than the value of −10.1‰ they used. More information on the time—and elevation—varying isotopic content of precipitation is needed to adequately parameterize hydrologic mixing models for the Himalaya.

[49] Both the ice ablation model and the mixing models indicate a decrease in the contribution of melting ice to streamflow with increasing distance from the glacierized areas, in agreement with large-scale studies in other parts of the Himalaya [Immerzeel et al., 2009, 2010]. Past studies showed that in the central/eastern area of the Himalaya, snow melt and monsoon rainfall constitutes a significant component of the water budget [Bookhagen and Burbank, 2010]. Seasonal snowmelt and rain from the nonglacierized areas in basins such as Dudh Kosi, with outlets at lower elevations, may be much more important to total annual streamflow than the melting ice occupying a smaller percent of the basin. Any proportional increase in glacier ice melt contribution to total streamflow may be due to a reduction in the contribution of nonglacierized areas by snowmelt and rain, rather than an increase in glacier ice melt only [Thayyen and Gergan, 2010].

5.2. Uncertainty and Limitations

[50] The methodology developed in this study allows using multitemporal imagery to assess the impact of changes in glacier area at the same time as a change in ELA and/or ice ablation gradient and capture the temporal changes in ice-melt contribution. Remote-sensing techniques show promise for understanding the contribution of glacial melt to streamflow runoff over large areas in the Himalaya where only limited runoff measurements exist. Challenges associated with estimating remote-sensing glacier area, ELA, and δb/δz on a yearly basis are addressed below. The accuracy of the ice-melt estimate depends on several factors: (1) the quality of the remote-sensing data used for glacier delineation (instrument gain settings, seasonal snow or clouds, time of year of the image acquisition, and geolocation errors); (2) uncertainty associated with glacier mapping (shadows, seasonal snow on glaciers, highly reflective water bodies, and debris-covered areas); (3) uncertainty in the estimates of ELA associated with the satellite-derived albedo and the DEM; (4) uncertainty in the ice ablation gradient due to lack of mass balance measurements on a glacier-by-glacier basis. A detailed description of these challenges, along with some recommendations is provided in Racoviteanu et al. [2008, 2009]. Here we summarize the main error sources as they pertain to either the ice ablation model and/or the isotope analysis.

[51] 1. In this study, we assumed a nominal uncertainty in glacier outlines of 1 ASTER pixel size (±15 m). Based on validation with high-resolution IKONOS imagery (4 m), we estimate the errors in glacier delineation to be ±15 m. The IKONOS scene was acquired at 7-day difference from the ASTER scenes, but the lack of snow in the IKONOS scene justifies its use for validation of the ASTER-derived outlines. For the Hinku watershed, due to the discrepancy in the year of acquisition of the two types of imagery, IKONOS scenes were only used to adjust pixels inside the glaciers (shadow areas, clouds) and not at the boundaries of the glaciers, which might have changed in the three-year period. Errors in the glacier delineation on shadow areas, small water bodies, and snow are considered minimal after validation with IKONOS data.

[52] 2. There is more uncertainty in estimating glacier ice melt for the debris-covered glacier tongues because of limited field-based measurements on these debris-covered glacier tongues. Observed ice melting rates under the debris cover depend on the thickness of the supraglacial debris and are highly variable [Brock et al., 2010; Foster et al., 2012; Kayastha et al., 2000; Lambrecht et al., 2011; Mihalcea et al., 2008b; Sakai, 2002; Takeuchi et al., 2000; Tangborn and Rana, 2000]. Furthermore, ice melt rates may be enhanced by the presence of ice cliff walls and supraglacial lakes [Zhang et al., 2011]. At a “critical thickness” of a few centimeters, sub-debris ice ablation rates are equal to those of clean ice [Brock et al., 2010; Foster et al., 2012; Zhang et al., 2011]. Below this critical thickness, ice-melt rates accelerate due to the low albedo of the debris [Singh et al., 2000; Takeuchi et al., 2000], and above this critical thickness, ablation rates of the ice underneath decrease due to the low thermal conductivity of debris. The accelerated ice-melt rates under a thin debris cover corresponding to the “rising limb” of ablation in Østrem [1969], however, are localized, occur at small temporal scales and generally under wet debris cover [Nakawo and Young, 1982; Nicholson and Benn, 2006], and do not necessarily apply over large areas. Since we do not have sufficient field observations of debris thickness, in our model we do not directly account for patchy areas of thin debris cover where ice melt may be accelerated. We also do not account for potential contribution of supra-glacial ice melt at the ice walls and supra-glacial lakes [Sakai, 2002; Sakai et al., 1998]. While we may underestimate the contribution of ice melt under debris cover at the upper parts of the debris cover tongues, we do provide a sensitivity analysis with a large debris-covered ice-ablation gradient (1.3 m/100 m) to address the potential impact of a thin debris cover on sub-debris glacier ice melt.

[53] 3. Errors in ELA estimates arise due to the vertical accuracy of the SRTM DEM (31 ± 10 m). The choice of the ELA is key, as it closely relates to glacier mass balance on a yearly basis, and as shown in section 'Stable Isotopes and Mixing Model Results', the ablation model is extremely sensitive to the choice of ELA. We have shown that a choice of either a steeper mass balance gradient and/or a higher ELA would significantly increase the glacier runoff contribution to measured discharge (Table 7). ELA values are ideally estimated from field measurements, corresponding to the altitude where the glacier mass balance (bn) is zero. However, field-based ELA measurements exist only from a few glaciers in the Himalaya [Karma et al., 2003; Konz et al., 2006; Kulkarni et al., 2004; Wagnon et al., 2007]. Some values are compiled in Benn and Owen [2000] on the basis of various studies, but the field measurements are not available in digital form. The basin-wide remote-sensing ELA value obtained for Langtang (∼5500 m) is in agreement with field-based ELA measurements in the area [Fujita et al., 1998] as well as values obtained using other methods for estimating the ELA [Kayastha and Harrison, 2008] (see section 'Glacier Area and ELA'). While we are using ELA values based on images from one year only, we conducted a sensitivity analysis allowing us to simulate various climate scenarios and to extend the estimates over the whole period of discharge measurements.

[54] 4. There are significant uncertainties associated with the discharge measurements in the Himalaya. Results of the ablation model are highly sensitive to measured discharge values, especially for small basins. To minimize uncertainties, we used long-term average discharge measurements for Trishuli and Dudh Kosi, thus neglecting the inter-annual variability in discharge.

[55] 5. Uncertainties in estimates of snow/ice melt based on isotope analysis are due to (a) the choice of the sampling sites, (b) parameterization of end members, and (c) limitations of synoptic surveys. End members were selected based on protocols described in Liu et al. [2004]; however, there is uncertainty associated with this, since we only have two components for the mixing model. We do not have monsoon rain samples to further constrain our mixing model. Isotopic sampling at a weekly to monthly frequency would improve the temporal resolution of the mixing models.

[56] 6. For both the ablation model and isotope methods, there is uncertainty in separating the snow and ice melt components of streamflow in this area of the Himalaya. The simultaneous accumulation and ablation in this monsoon-dominated area makes it challenging to distinguish between the snowmelt and ice melt components of streamflow using either the ice ablation gradient or the isotope methods. In the case of the former, we applied the ablation gradient exclusively to the area below the ELA, which by definition includes only ice, and it was delineated from images acquired at the end of the ablation season (October/November). Here we are not concerned with the annual accumulation occurring above the ELA. Estimating the snow and ice melt due to the glacier-wide mass balance using an ablation and accumulation gradient is beyond the scope of this paper.

[57] 7. Another limitation of our approach is that it does not address the seasonality in the production of glacial meltwater. For example, Li and Williams [2008] have shown that for the northwestern Himalayas glacial melt is an important component of river flow in late August. Moreover, melting of glacial ice plays an important role in maintaining water security during times of drought or similar climate extremes. For example, in the European Alps during the drought year of 2003 glacial melt contributions to August discharge of the Danube River were about three times greater than the 100 year average [Huss, 2011]. Addressing this seasonality in glacial melt is also beyond the scope of this paper.

6. Summary and Outlook

[58] Concerns over the current retreat of Himalayan glaciers and its implications for water supplies are widespread. Addressing these concerns requires a clarification of the interaction between stream flow, glacier mass balance, and seasonal snowmelt in different climate regimes across the Himalaya. An accurate quantitative assessment of the sources of river flow is crucial to effectively manage regional water resources now or in the future. Any attempt to generalize the response of Himalayan glacierized basins to climate forcing is challenged by the complex hydrology and the different sources of moisture across this mountain range. Furthermore, potential water shortage in a basin depends on a combination of factors, related to both cryosphere and socio-economics, such as the magnitude of cryosphere changes, the timing of the discharge, water demand for consumption and irrigation, population density, and the capacity for adaptation, among others.

[59] This study focused on assessing the relative contributions of ice melt to streamflow in the eastern, monsoon-influenced part of Himalaya. In this study, we complement and improve an existing ice ablation model by using updated remote sensing derived glacier data and separate ice ablation gradients for clean ice and debris-covered ice to estimate melt. The ice ablation model computes the ice melt volume over clean and debris-covered ice, by using an ice ablation gradient for each 100 m elevation band. We used satellite remote-sensing data (ASTER, IKONOS-2 and SRTM) combined with field-based discharge measurements for estimating the contribution of glacier ice to streamflow. Field-based stable isotope measurements allowed quantifying the contribution of groundwater and glacier ice/snow, as tools for validating the remote-sensing techniques. Using a combination of ablation models and isotope studies as two independent methods has the potential to reduce the uncertainty in estimates of the contribution of snow and glacier ice melt to streamflow in the Himalaya. The ice-melt estimates obtained from the two methods are in close agreement and show that at low altitudes, groundwater dominates river flow within only tens of kilometers of the glacier outlet. Our results are in agreement with recent studies [Bookhagen and Burbank, 2010; Immerzeel et al., 2009, 2010; Kaser et al., 2010], which showed only a modest contribution of ice melt to streamflow in monsoon-dominated climates of Brahmaputra and Ganges basins. This is in contrast with more significant ice-melt contributions in basins situated in arid climates (Indus basin). Furthermore, we found that runoff from debris-covered glaciers constitutes only a small percent of the measured streamflow, due to the small coverage of these glacier tongues relative to basin area.

[60] Our results do not support the widespread concerns that glacier retreat will induce severe water scarcity at lowlands across the Himalaya, as stated previously [IPCC, 2007]. We conclude that in the monsoon-dominated eastern Himalaya, in large basins with a low percentage of glacier cover, precipitation (as summer monsoon rain and winter snowfall) may be more important than glacier melt in contributing to streamflow at lower elevations, where the majority of the population lives. Lowland areas may face water shortages in the future, but these may be due to a combination of a high water demand for irrigation or consumption rather than climate-induced glacier changes alone.


[61] This research was supported by a NASA Earth System Sciences (ESS) fellowship (NNX06AF66H), a National Science Foundation doctoral dissertation improvement grant (NSF DDRI award B.C. 0728075), a CIRES research fellowship, and a graduate fellowship from CU-Boulder. Fieldwork funds were provided by the NSF-funded Niwot Ridge Long-Term Ecological Research program and an NSF OISE grant. A. Racoviteanu's post-doctoral research is supported by the Centre National d'Études Spatiales (CNES), France. Additional funding came from the World Bank project South Asia Sustainable Development Network (SASDN), the NASA-funded HIMALA project NNH08ZDA001N, and the USAID Cooperative Agreement AID-OAA-A-11–00045. We also thank Jonathan Taylor at California State University for facilitating access to high-resolution imagery through the NASA appropriations grant NNA07CN68G, and the NASA-funded Global Land Ice Measurements from Space (GLIMS) project for access to ASTER imagery at no cost. We thank Don Alford for the initial development of the methodology used in this study; N. Harshadeep, World Bank, for providing comments on previous reports; J.P. Lama at Guide for All Seasons trekking agency in Nepal for providing logistical support; Ang Kipa Sherpa, Debendra Karki, and Rory Cowie for help in the field and IRD France/GLACIOCLIM project for logistical support at Mera base camp. We acknowledge Patrick Wagnon and Yves Arnaud at IRD France for their constructive feedback on this paper.