Hydrological Modeling of Highly Glacierized Basins (Andes, Alps, and Central Asia)

water

Article Hydrological Modeling of Highly Glacierized Basins (Andes, Alps, and Central Asia)

Nina Omani 1,*, Raghavan Srinivasan 2, Raghupathy Karthikeyan 3 and Patricia K. Smith 3

1 Former Graduate Research Assistant, Texas A&M University, College Station, TX 77843, USA 2 Spatial Sciences Laboratory, Department of Ecosystems Sciences and Management, Department of Biological and Agricultural Engineering, Texas A&M University, College Station, TX 77843, USA; [email protected] 3 Department of Biological and Agricultural Engineering, Texas A&M University, College Station, TX 77843, USA; [email protected] (R.K.); [email protected] (P.K.S.) * Correspondence: [email protected]; Tel.: +1-979-402-9391

Academic Editor: Richard Skefﬁngton Received: 4 December 2016; Accepted: 3 February 2017; Published: 10 February 2017

Abstract: The Soil and Water Assessment Tool (SWAT) was used to simulate five glacierized river basins that are global in coverage and vary in climate. The river basins included the Narayani (Nepal), Vakhsh (Central Asia), Rhone (Switzerland), Mendoza (Central Andes, Argentina), and Central Dry Andes (Chile), with a total area of 85,000 km2. A modified SWAT snow algorithm was applied in order to consider spatial variation of associated snowmelt/accumulation by elevation band across each subbasin. In previous studies, melt rates varied as a function of elevation because of an air temperature gradient while the snow parameters were constant throughout the entire basin. A major improvement of the new snow algorithm is the separation of the glaciers from seasonal snow based on their characteristics. Two SWAT snow algorithms were evaluated in simulation of monthly runoff from the glaciered watersheds: (1) the snow parameters are lumped (constant throughout the entire basin) and (2) the snow parameters are spatially variable based on elevation bands of a subbasin (modified snow algorithm). Applying the distributed SWAT snow algorithm improved the model performance in simulation of monthly runoff with snow-glacial regime, so that mean RSR decreased to 0.49 from 0.55 and NSE increased to 0.75 from 0.69. Improvement of model performance was negligible in simulations of monthly runoff from the basins with a monsoon runoff regime.

Keywords: SWAT; glacier; snow; modeling; watershed

1. Introduction The two basic snowmelt approaches generally used in hydrologic modeling are energy balance models and temperature-index models [1]. Temperature-index models are widely used in hydrologic studies because of the models’ performance and simplicity as well as the wide availability of temperature data [2–4]. In temperature-index based runoff models such as SWAT, melt rates only vary as a function of elevation, because of the air temperature gradient [5]. To overcome this drawback, a modiﬁed snow process was incorporated into SWAT in order to consider spatial variation of snowmelt/accumulation parameters by elevation band across subbasins. Previous studies using SWAT kept these parameters constant for an entire basin [6–10]. While this method was successful in simulation of snowmelt ﬂow, simulation of runoff from glaciered watersheds demands a distributed model for distinguishing seasonal snow from glacier. The new approach allows for the separation of seasonal snowmelt from glaciers’ melt based on the spatial variability of associated melt parameters.

Water 2017, 9, 111; doi:10.3390/w9020111 www.mdpi.com/journal/water Water 2017, 9, 111 2 of 19

The SWAT model has been applied worldwide, and its hydrologic components have been successfully tested in cases where streamflow was predominantly generated from rainfall events [11–15]. The model less frequently has been applied in mountainous watersheds, and a few recent studies have been conducted to test and improve SWAT’s snow hydrology component. Fontaine et al. [16] incorporated the elevation bands method with SWAT’s original snowmelt algorithm (temperature-index model), which improved the Nash–Sutcliffe coefficient of monthly runoff simulation from −0.70 to 0.86 in the 4999-km2 Rocky Mountain Basin in Wyoming. Debele and Srinivasan [4] incorporated a modified version of SNOW17 [17] into SWAT and compared its performance with the temperature-index model in three watersheds (ranging from 22.28 to 7106.82 km2), the results of which showed that the temperature-index model performed better than the SNOW17 model. Debele et al. [18] incorporated the distributed process-based energy budget SNOWEB in the pixel and elevation band scales into SWAT and compared its performance with the temperature-index model. In this method, it was assumed that solar radiation varies not only with latitude and altitude of subbasins, which is applied in the current version of SWAT, but also with land surface inclinations (aspect and slope). The temperature-index based snowmelt computation method had overall Nash-Sutcliffe model efficiency coefficient ranging from 0.49 to 0.73 in simulation of monthly streamflow while the energy budget based approach had Nash-Sutcliffe efficiency coefficient ranging from 0.33 to 0.59. Zhang et al. [8] applied SNOW17 in SWAT at the pixel scale. The SWAT model with temperature-index plus elevation bands performed as well as the SWAT model with SNOW17. Rahman et al. [19] applied the modified temperature-index method [5] to simulate the runoff from fully and partially glaciated subbasins of the Rhone River Basin. The applied method showed improvement in simulated streamflow from the main outlet of the basin even though the snowmelt parameters were lumped across the basin. Distributed, process-based energy budget models have been tested in SWAT, but we focused on another component of frequently used glacier/snowmelt models. A major difference between SWAT melt processes and the melt routine of hydroglacial models arises from the associated melt parameter distribution (melt factor). In previous studies using SWAT, associated snowmelt parameters were considered to be uniform within each subbasin, but in the hydroglacial models, melt factors are spatially variable in pixel or band scales. A common approach is to assign two different melt factors to ice and snow. This enables the user to treat seasonal snow and glaciers differently; melt factors are generally reported to be higher for glaciers and lower for snow [20]. The applied approach in this study separates seasonal snow from glaciers by setting the accumulation/melt parameters, including maximum and minimum melt factors, melt lag factor, melt temperature and snow fall temperature for the elevation bands of each subbasin. The details of the proposed approach are available at Omani et al. [21] for three benchmark glaciers in Asia and Europe. The objective of this study was to extend the applied method to macro-scale river basins that are global in coverage and vary in climatic condition. Two snow process algorithms (lumped and distributed) were examined for their ability to simulate flow in five river basins that provide global assessment and feature contrasts in climatic conditions.

2. Materials and Methods

2.1. Study Watersheds This study focuses on areas where climate variability has had a strong impact on highly glaciated areas. Five river basins with a total area of 85,000 km2, in the northern and southern hemispheres, for which sufﬁcient hydrologic information is available were selected for this study. These river basins have different spatial scales and climatic situations, from extreme maritime to extreme continental climates. They include: Vakhsh in Tajikistan, Narayani in Nepal, Mendoza in Argentina, ﬁve individual glaciated watersheds in the central dry Andes of Chile, and Upper Rhone in Switzerland (Figure1). Water 2017, 9, 111 3 of 19

Climate, hydrologic regime, soil, and land cover information for these watersheds can be found elsewhereWater 2017, 9, [ 22111]. 3 of 20

Figure 1. Location of ﬁvefive river basins: (1) Upper Rhone, (2) Vakhsh, (3) Narayani, (4) Mendoza, and (5) central Chile.

2.2. Hydrologic Data 2.2. Hydrologic Data For this study, a 90‐m resolution Global SRTM Digital Elevation Model contributed by For this study, a 90-m resolution Global SRTM Digital Elevation Model contributed by USGS/ USGS/NASA was used to describe topographic conditions such as slope and slope length, to create NASA was used to describe topographic conditions such as slope and slope length, to create ﬂow flow direction and accumulated flow, and to delineate watershed boundaries and channel networks direction and accumulated ﬂow, and to delineate watershed boundaries and channel networks [23]. [23]. Land use information was adopted from USGS Global Land‐Cover Characteristics (GLCC). Land use information was adopted from USGS Global Land-Cover Characteristics (GLCC). GLCC was GLCC was developed using satellite data collected from 1992 to 1993 with 1 km spatial resolution for developed using satellite data collected from 1992 to 1993 with 1 km spatial resolution for the entire the entire globe [24]. Land use information can be found in Omani et al. (2016b). The soil properties globe [24]. Land use information can be found in Omani et al. (2016b). The soil properties dataset with dataset with 5 arc‐minutes resolution (Ver. 1.2) was taken from ISRIC‐WRS [25]. The major soil type 5 arc-minutes resolution (Ver. 1.2) was taken from ISRIC-WRS [25]. The major soil type in the upper in the upper areas of the river basins is Lithosols, which are very shallow and occur mainly on steep areas of the river basins is Lithosols, which are very shallow and occur mainly on steep slopes, often slopes, often with exposed rock debris. Weather data were obtained from the Climate Forecast with exposed rock debris. Weather data were obtained from the Climate Forecast System Reanalysis System Reanalysis (CFSR) global meteorological dataset [26]. Global CFSR data are available going (CFSR) global meteorological dataset [26]. Global CFSR data are available going back to 1979 at a back to 1979 at a 38‐km resolution. Daily discharge records for model calibration were collected from 38-km resolution. Daily discharge records for model calibration were collected from local hydrologic local hydrologic administrators and online databases (Table 1). administrators and online databases (Table1).

2.3. Model Simulation This sectionsection focuses focuses on on glacier glacier modelling modelling in SWAT. in SWAT. More informationMore information about SWAT about snow SWAT processes snow andprocesses theoretical and theoretical details of details the applied of the approachapplied approach in this study in this can study be foundcan be infound Omani in Omani et al. [ 21et ].al. [21].

2.3.1. Subbasin Delineation and HRU DefinitionDefinition The watershedwatershed boundary boundary of theof the river river basins basins was determinedwas determined using the using SWAT the interface SWAT automaticinterface delineationautomatic delineation procedure (Figureprocedure2). Based (Figure on 2). the Based thresholds on the selected, thresholds there selected, were a total there of were 675 subbasins a total of in675 the subbasins five river in basins. the five SWAT river basins. further SWAT divides further each subbasin divides each into smallersubbasin areas into calledsmaller Hydrological areas called ResponseHydrological Units Response (HRUs). Units There (HRUs). were a totalThere of were 9878 a HRUs total of in 9878 the five HRUs river in basins. the five river basins.

Water 2017, 9, 111 4 of 19

Table 1. All available ﬂow gauge stations with drainage areas for each river basin.

Basin and Tributaries Flow Gauge Station Latitude Longitude Area (km2) Elevation (m) Calibration Period Validation Period Narayani a Burhi Gandaki Arughat 28.04 84.82 4270 485 1981–1985 - Marsyandi Bimal Nagar 27.95 84.43 3850 354 1981–1992 - Kali Gandaki Setibeni 28.01 83.60 6630 410 1981–1987 1989–1993 Seti Phoolbari 27.71 84.43 31,100 830 1981–1985 1991–1993 Narayani Gandak Devghat 28.23 84.00 582 180 1981–1984 - Vakhsh a Muksu Davsear 39.13 71.57 6550 2220 1981–1985 - Obikhingou Tavildara 38.70 70.52 5390 1616 1981–1985 - Vakhsh Komsomolabad 38.87 69.98 29,500 1258 1981–1987 1988–1989 Rhone b Rhone Gletsch 46.56 8.36 39 1761 1993–2007 2008–2010, 1982–1992 Goneri Oberwald 46.53 8.36 40 1385 1993–2007 2008–2010, 1982–1992 Massa Blattenbei Naters 46.39 8.01 195 1446 1993–2007 2008–2010, 1982–1992 Lonza Blatten 46.42 7.82 78 1520 1993–2007 2008–2010, 1982–1992 Central Chile c Juncal En Juncal −32.85 −70.17 - 1800 1993–2007 2008–2010, 1982–1992 Blanco En Rio Blanco −32.90 −70.28 - 1420 1993–2007 2008–2010, 1982–1992 Aconcagua En Rio Blanco −32.90 −70.30 - 1420 1993–2007 2008–2010, 1982–1992 Colorado En Colorado −32.85 −70.40 - 1062 1993–2007 2008–2010, 1982–1992 Aconcagua En Chacabuquito −32.85 −70.50 - 950 1993–2007 2008–2010, 1982–1992 Estero Yerba Loca Antes J. S. F. −33.33 −70.35 - 1350 1993–2007 2008–2010, 1982–1992 Olivares Antes Junta Rio Colorado −33.48 −70.13 - 1500 1993–2007 2008–2010, 1982–1992 Colorado Antes Junta Rio Olivares −33.48 −70.13 - 1500 1993–2007 2008–2010, 1982–1992 Colorado Antes Junta Rio Maipo −33.58 −70.37 - 890 1993–2007 2008–2010, 1982–1992 Maipo En el Manzano −33.58 −70.37 - 850 1993–2007 2008–2010, 1982–1992 Maipo En las Hualtatas −33.97 −70.13 - 1820 1993–2007 2008–2010, 1982–1992 Volcan En Queltehues −33.80 −70.20 - 1365 1993–2007 2008–2010, 1982–1992 Maipo En San Alfonso −33.73 −70.30 - 1092 1993–2007 2008–2010, 1982–1992 Pangal En Pangal −34.23 −70.32 - 1500 1993–2007 2008–2010, 1982–1992 En Pte. Termas de Cachapoal −34.25 −70.57 - 700 1993–2007 2008–2010, 1982–1992 Cauquenes Cachapoal Aguas Abajo −34.33 −70.37 - 1127 1993–2007 2008–2010, 1982–1992 Tinguiririca Bajo los Briones −34.72 −70.82 - 560 1993–2007 2008–2010, 1982–1992 Claro En el Valle −34.68 −70.87 - 476 1993–2007 2008–2010, 1982–1992 Mendoza d Colorado Punta de Vacas −32.83 −69.70 - - 1993–2007 2008–2010, 1982–1992 Cuevas Punta de Vacas −32.87 −69.77 - - 1993–2007 2008–2010, 1982–1992 Mendoza Guido −32.92 −69.24 - - 1993–2007 2008–2010, 1982–1992 Tupungato Punta de Vacas −32.88 −69.77 - - 1993–2007 2008–2010, 1982–1992 Vacas Punta de Vacas −32.85 −69.76 - - 1993–2007 2008–2010, 1982–1992 Notes: a Source: The Global Runoff Data Centre (GRDC); b Source: Switzerland Federal Ofﬁce for the Environment (FOEN); c Source: National Water Information System of Argentina; d Source: Chilean Dirección General de Aguas (DGA). Water 2017, 9, 111 5 of 19 Water 2017, 9, 111 6 of 20

(a) (b)

(e)

Figure 2. Subbasin delineation, glacier outlines, and locations of dams and flow gauge stations in (a) Figure 2. Subbasin delineation, glacier outlines, and locations of dams and ﬂow gauge stations in Rhone River Basin; (b) Narayani River Basin; (c) Chilean River Basins; (d) Mendoza River Basin; and (a) Rhone River Basin; (b) Narayani River Basin; (c) Chilean River Basins; (d) Mendoza River Basin; (e) Vakhsh River Basin. and (e) Vakhsh River Basin. 2.3.2. Initial Glacier Storage 2.3.2. Initial Glacier Storage The equilibrium line altitude (ELA) is a theoretical line where climatic mass balance is zero on Theaverage equilibrium over a number line of altitude years and (ELA) determines is a theoretical the boundary line of where the accumulation climatic mass zone balance and ablation is zero on averagezone over on a a glacier. number The of ELA years of anda glacier determines is sometimes the boundary denoted ELA of the0 (the accumulation balanced‐budget zone ELA). and The ablation zone on a glacier. The ELA of a glacier is sometimes denoted ELA0 (the balanced-budget ELA). The ELA is determined by climate and the aspect of the glacier. It is not inﬂuenced by glacier dynamics, Water 2017, 9, 111 7 of 20

ELAWater 2017is determined, 9, 111 by climate and the aspect of the glacier. It is not influenced by glacier dynamics,6 of 19 extent and hypsometry, and thus reveals a largely unfiltered climatic signal [27]. Therefore, it can be a representation of the lowest boundary of climatic glacierization [28]. extent and hypsometry, and thus reveals a largely unﬁltered climatic signal [27]. Therefore, it can be a In this study, it was assumed that the ELA0 represented the glacier boundary. The physical representation of the lowest boundary of climatic glacierization [28]. boundaries of glaciers were then corrected based on glacier inventory data, GLIMS glacier outlines In this study, it was assumed that the ELA0 represented the glacier boundary. The physical and MODIS products for the glacier‐free areas at elevations higher than the ELA0, and the areas boundaries of glaciers were then corrected based on glacier inventory data, GLIMS glacier outlines and climatically suited for glacier formation at altitudes lower than ELA0. Elevation bands below the MODIS products for the glacier-free areas at elevations higher than the ELA0, and the areas climatically ELA0 were considered to be seasonal snow cover regardless of glacial tongues extending into lower suited for glacier formation at altitudes lower than ELA . Elevation bands below the ELA were elevations. Debris‐covered areas of glaciers, accumulated0 wind‐blown snow, and small isolated0 considered to be seasonal snow cover regardless of glacial tongues extending into lower elevations. glaciers with an area of 0.1 km2 or less were ignored when estimating the total glaciated area of the Debris-covered areas of glaciers, accumulated wind-blown snow, and small isolated glaciers with an river basins. ELA2 0 values were obtained from the literature and are available in Table 2. area of 0.1 km or less were ignored when estimating the total glaciated area of the river basins. ELA0 values were obtained from the literature and are available in Table2. Table 2. The calculated area of glaciers using GLIMS, MODIS, and modeled glacier area in this study.

Table 2. The calculated area of glaciers using GLIMS,Modeled MODIS, and modeled glacier area in this study. Total Area GLIMS% MODIS% Area% ELA0 (m.s.l.) Source (km2) Total Area (ThisModeled Study) Area% 2 GLIMS% MODIS% ELA0 (m.s.l.) Source Narayani 31,698(km ) 9.20 7.32 (This11.0 Study) 5000 in east to 5600 in west [29,30] Narayani 31,698 9.20 7.32 11.0 5000 in east to 5600 in west [29,30WGMS] Vakhsh 28,907 9.66 13.1 12.3 4300 in north to 4500 in south WGMSdatabase [31] Vakhsh 28,907 9.66 13.1 12.3 4300 in north to 4500 in south Upper database [31] 4513 13.61 13.36 14.2 2800 in north to 3100 in south [28] RhoneUpper 4513 13.61 13.36 14.2 2800 in north to 3100 in south [28] MendozaRhone 7092 4.34 4.25 4800 at 33° S to over 5000 at 32° S [32] CentralMendoza 7092 4.34 4.25 48005000 at 33at◦ 30°S to S, over and 5000 drops at 32 to◦ S[32] 14,342 4.35 4.37 [33–36] ChileCentral 4300–44005000 at around 30◦ S, and 32.5° drops S to to 33.0° S 14,342 4.35 4.37 [33–36] Chile 4300–4400 around 32.5◦ S to 33.0◦ S Glacier covered areas were extracted from the Global Land Ice Measurements from Space (GLIMS)Glacier [37,38] covered dataset areas and were the extractedWorld Glacier from Inventory the Global (WGI). Land IceIf the Measurements GLIMS dataset from did Space not completely(GLIMS) [37 cover,38] datasetthe entire and study the Worldarea it Glacierwas complimented Inventory (WGI). with glaciated If the GLIMS area by dataset the Moderate did not Resolutioncompletely Imaging cover the Spectroradiometer entire study area (MODIS) it was complimented maximum snow with extent glaciated product area (MOD10A2) by the Moderate with eightResolution‐day composites Imaging Spectroradiometer at 500‐m resolution. (MODIS) The minimum maximum snow snow cover extent area product at the end (MOD10A2) of the melting with season,eight-day from composites February at 2002 500-m to resolution.2010, was considered The minimum to be snow the glacier/permanent cover area at the end snow of the cover melting area (Figureseason, 3). from The February percentage 2002 of tothe 2010, glacier was covered considered area of to each be the river glacier/permanent basin by each model, snow MODIS cover areaand GLIMS,(Figure3 is). Thepresented percentage in Table of the 2. It glacier is assumed covered that area GLIMS of each glacier river outlines basin by represent each model, the glacial MODIS extent and atGLIMS, the end is presentedof the reference in Table period2. It is as assumed a starting that point GLIMS for glacierthe future outlines simulations represent of the glacial glacial extent. extent The at meanthe end ice of thickness the reference in each period elevation as a starting band pointwas calculated for the future by simulationsaveraging the of glacialmeasured extent. ice Thethickness mean valuesice thickness from the in eachWGI. elevation When no banddata were was calculatedavailable for by high averaging elevations, the measured ice thickness ice thicknesswas assumed values to befrom 1000 the m WGI. water When equivalent no data (m.w.e) were (SNOEB available = for999,999 high mm elevations, H2O). Figures ice thickness A1 and was A2 in assumed Appendix to beA show1000 m thewater measurement equivalent locations (m.w.e) (SNOEB of glacier = 999,999depth and mm area H2O). from Figures the WGI A1 anddata A2 inventory in Appendix for theA show river basins.the measurement locations of glacier depth and area from the WGI data inventory for the river basins.

Upper Rhone 100 2000 %

2001 80 2002 srea 60 2003 2004

cover 40 2005 20 2006 Snow 0 2007

8 2008 ‐ 72 40 ‐ ‐ 1 360 328 296 264 232 200 168 136 104 ‐ ‐ ‐ ‐ ‐ ‐ ‐ ‐ ‐ 2009 65 33

97 2010 353 321 289 257 225 193 161 129 Day Average

(a)

Figure 3. Cont.

Water 2017, 9, 111 7 of 19 Water 2017, 9, 111 8 of 20

Vakhsh Narayani 100 100 % %

80 80 area area

60 60 40 40 cover cover

20 20 Snow 0 Snow 0 8 8 ‐ 40 72 ‐ 40 72 ‐ ‐ 1 360 328 296 264 232 200 168 136 104 ‐ ‐ 1 360 328 296 264 232 200 168 136 104 ‐ ‐ ‐ ‐ ‐ ‐ ‐ ‐ ‐ ‐ ‐ ‐ ‐ ‐ ‐ ‐ ‐ ‐ 33 65 33 65 97 97 353 321 289 257 225 193 161 129 353 321 289 257 225 193 161 129 Day Day (b) (c) Central Chile 100 Mendoza

100 %

% 80 80 area

area 60 60

cover 40

40 cover

20 20 Snow

Snow 0 0 8 8 ‐ ‐ 40 72 40 72 ‐ ‐ 1 ‐ ‐ 1 360 328 296 264 232 200 168 136 104 360 328 296 264 232 200 168 136 104 ‐ ‐ ‐ ‐ ‐ ‐ ‐ ‐ ‐ ‐ ‐ ‐ ‐ ‐ ‐ ‐ ‐ ‐ 33 65 33 65 97 97 353 321 289 257 225 193 161 129 353 321 289 257 225 193 161 129 Day Day (d) (e)

Figure 3. Eight‐day snow cover area variations extracted from MODIS products (MOD10A2) from Figure 3. Eight-day snow cover area variations extracted from MODIS products (MOD10A2) from 2000 2000 to 2010 are represented as a percentage of the total area of river basins: (a) Upper Rhone; (b) to 2010 are represented as a percentage of the total area of river basins: (a) Upper Rhone; (b) Vakhsh; Vakhsh; (c) Narayani; (d) Central Chile; and (e) Mendoza. (c) Narayani; (d) Central Chile; and (e) Mendoza.

Subbasins over 2000 m in altitude were divided into 10 elevation bands with 100‐m to 200‐m intervals,Subbasins depending over 2000on the m elevation in altitude range were of divided the subbasin. into 10 Smaller elevation elevation bands band with 100-mintervals to enable 200-m intervals,SWAT to model depending the glacier on the boundaries elevation range more of accurately. the subbasin. It was Smaller assumed elevation that the band glacier intervals boundary enable in SWATa subbasin to model matched the glacier the lowest boundaries altitude more of the accurately. elevation Itband was if assumed more than that 50% the of glacier the elevation boundary band in a subbasinarea was covered matched by the a glacier. lowest altitude of the elevation band if more than 50% of the elevation band area was covered by a glacier. 2.3.3. Model Calibration and Validation 2.3.3. Model Calibration and Validation The model was calibrated using monthly streamflow data at the gauge stations listed in Table 1, The model was calibrated using monthly streamflow data at the gauge stations listed in Table1, focusing on glaciated subbasins. The non‐glaciated and snow‐free gauged watersheds located at focusing on glaciated subbasins. The non-glaciated and snow-free gauged watersheds located at lowlands and glaciated subbasins with insufficient flow data, as well as dammed subbasins, were lowlands and glaciated subbasins with insufficient flow data, as well as dammed subbasins, were ignored in the model calibration. The model was validated using the monthly streamflow from the ignored in the model calibration. The model was validated using the monthly streamflow from the Rhone, Mendoza, and Chile for two periods: recent years from 2008 to 2010, and early years from Rhone, Mendoza, and Chile for two periods: recent years from 2008 to 2010, and early years from 1982 to 1992. This was to confirm that the model works well for different periods. For the Vakhsh 1982 to 1992. This was to confirm that the model works well for different periods. For the Vakhsh and and Nepal River Basins, model validation periods were variable based on data availability. Nepal River Basins, model validation periods were variable based on data availability. Monthly calibration was performed using the SWAT Calibration and Uncertainty Program Monthly calibration was performed using the SWAT Calibration and Uncertainty Program (SWAT‐CUP) [39]. SWAT‐CUP is a generic interface to provide a link between any automatic (SWAT-CUP) [39]. SWAT-CUP is a generic interface to provide a link between any automatic calibration/uncertainty or sensitivity program and SWAT. Sequential Uncertainty Fitting Version 2 calibration/uncertainty or sensitivity program and SWAT. Sequential Uncertainty Fitting Version (SUFI2) incorporated in SWAT‐CUP allows for calibration and validation of the model at multiple 2 (SUFI2) incorporated in SWAT-CUP allows for calibration and validation of the model at multiple gauging stations and multiple observed datasets simultaneously. SUFI2 calculates a weighted gauging stations and multiple observed datasets simultaneously. SUFI2 calculates a weighted multi‐component objective function (weighted summation of the square errors) based on simulated multi-component objective function (weighted summation of the square errors) based on simulated variables and observed time series of all gauged watersheds and then minimizes it by searching the Latin Hyperbolic [40] generated parameters for the best solution. In order to perform automatic calibration by SUFI2, the initial parameter values and ranges were determined. A list of all snowfall/melt parameters and associated groundwater parameters and

Water 2017, 9, 111 8 of 19 variables and observed time series of all gauged watersheds and then minimizes it by searching the Latin Hyperbolic [40] generated parameters for the best solution. In order to perform automatic calibration by SUFI2, the initial parameter values and ranges were determined. A list of all snowfall/melt parameters and associated groundwater parameters and their ranges [23] are presented in Omani et al. [22]. Since most of the watersheds studied are glacier dominated and above the tree lines, the land use and soils in the upper catchment had little or no significant impact on hydrology such as evapotranspiration (ET). So, none of the land use or soil parameters were found to be sensitive in the model calibration process, which confirmed that the roles of these parameters are not important to the water budget. The groundwater delay (GW-DELAY) and baseflow recession constant (ALPHA-BF) were initially set for simulation of low flow during the winter months. The parameter values then were optimized during the automatic model calibration. The baseflow recession constant is directly proportional to groundwater flow response to changes in recharge. The parameters were adjusted for the elevation bands above the ELA0 as an accumulation zone or glacier and below the ELA0 as an ablation zone with seasonal snow. The melt parameters for the elevation bands were adjusted in order to match the observed and simulated average monthly flow curves and then the parameters were optimized by automatic model calibration using SUFI2. Briefly, model calibration consisted of two main steps. First, parameters were automatically calibrated for an entire basin so that the snow parameters were uniform throughout the entire basin. In the next step, the results were improved by calibrating the model for snow parameters at the elevation band-subbasin scale. Model performance was tested in some of the smaller subbasins first and then extended into the larger subbasins to test the hypothesis that the hydrologic significance of meltwater may be negligible at the macro scale despite the presence of large glaciers in the headwaters area [41,42]. Calibration and validation were evaluated using the root mean square error-observations standard deviation ratio (RSR), the Nash–Sutcliffe Coefficient of Efficiency (NSE) [43] and the Percent Bias (PBIAS). RSR is a ratio of the RMSE and standard deviation of measured data. It ranges from the optimal value of 0 to a large positive value [44]. NSE indicates how well the plot of observed versus simulated data fits the 1:1 line. PBIAS measures the average tendency of the simulated data to be larger or smaller than their observed counterparts [45]. The optimal value of PBIAS is 0.0, with low values indicating accurate model simulation in term of magnitude.

3. Results and Discussion Tables3–7 show an average of adjusted values and range of the calibration parameters throughout the river basins for lumped and modiﬁed snow algorithms. The symbol “s” in the tables stands for glacier-free elevation bands or snowy elevation bands and “g” stands for glaciered elevation bands over the ELA0 altitudes. The calibration parameter values are averaged for the elevation bands and subbasins and the parameter range shows minimum and maximum parameter values based on elevation band-subbasin scale throughout the entire river basin. Mean SMFMX (Maximum melt factor) throughout all river basins is between 2.80 and 5.38, and 3.30 to 7.71, for snow and glacier, respectively. TIMP (Snow temperature lag factor) values are lower in the high elevation bands where the glaciers exist and range between 0.010 to 1.000 with average values between 0.025 for high elevation bands and 0.740 for low elevation bands. It can be seen from Table3 that the average values for SMFMX and SMFMN (Minimum melt factor) for the Narayani River Basin are larger than the obtained values for Rhone River Basin (Table4) as also was indicated by Kayastha et al. [46]. Water 2017, 9, 111 9 of 19

Table 3. Range and mean values of calibration parameters in subbasin-elevation band scale, Narayani River Basin.

Modiﬁed Model Parameter Name Lump Model Average Range SFTMP 0.9 2.09 2,4 SMTMP 0.6 0.83 0,2 SMFMX s a 7.33 5.38 4,7 SMFMX g b 7.33 7.71 5,9 SMFMN s 6.22 4.29 3,6 SMFMN g 6.22 5.59 4,7 TIMP s 0.42 0.67 0.27,0.87 TIMP g 0.42 0.12 0.01,0.5 TLAPS −6.5 −6.5 −6.5 PLAPS −100 −100 0,−200 ALPHA_BF 0.053 0.053 0.048,1 Notes: a s: elevation bands below 5000 m; b g: glaciered elevation bands over 5000 m.

Table 4. Range and mean values of calibration parameters in subbasin-elevation band scale, Rhone River Basin.

Modiﬁed Model Parameter Name Lump Model Average Range SFTMP s a 0.4 1.0 −1,2 SFTMP g b 0.4 2.0 1,3 SMTMP s 1.7 0.6 0,2 SMTMP g 1.7 1.5 0.5,3.0 SMFMX s 2.8 3.54 2,4 SMFMX g 2.8 5.50 4,6 SMFMN s 2.4 3.00 1.5,4.0 SMFMN g 2.4 3.45 1.5, 5.0 TIMP s 0.96 0.57 0.50,0.70 TIMP g 0.96 0.025 0.01,0.05 TLAPS −6.7 −6.7 −5.0,−7.0 PLAPS 267 267 0,400 ALPHA_BF 0.036 0.036 0.010,0.048 Notes: a s: elevation bands below 2900 m; b g: glaciered elevation bands over 2900 m.

Table 5. Range and mean values of calibration parameters in subbasin-elevation band scale, Vakhsh River Basin

Modiﬁed Model Parameter Name Lump Model Average Range SFTMP 2.24 1.72 1,3 SMTMP s a 4.97 1.41 0,2 SMTMP g b 4.97 1.67 0.5,2.0 SMFMX s 6.96 4.27 2,6 SMFMX g 6.96 5.77 4,8 SMFMN s 2.72 3.68 2,6 SMFMN g 2.72 4.77 3,7 TIMP s 0.14 0.65 0.5,1.0 TIMP g 0.14 0.06 0.03,0.12 TLAPS −7.70 −7.70 −7.5,−7.8 PLAPS 312.0 312.0 135,500 ALPHA_BF 0.01 0.01 0.004,0.020 GW_DELAY 40.7 40.7 20,60 Notes: a s: elevation bands below 4300 m; b g: glaciered elevation bands over 4300 m. Water 2017, 9, 111 10 of 19

Table 6. Range and mean values of calibration parameters in subbasin-elevation band scale, Mendoza River Basin.

Modiﬁed Model Parameter Name Lump Model Average Range SFTMP 4.5 2.2 2.0,3.4 SMTMP 1.7 3 2,4 SMFMN s a 1.9 2.3 2,4 SMFMN g b 1.9 3.1 2,4 SMFMX s 2.0 2.8 2.0,3.5 SMFMX g 2.0 3.3 2.0,4.5 TIMP s 1.0 0.33 0.1,1 TIMP g 1.0 0.15 0.01,0.3 TLAPS −7.2 −7.2 −6.9,−7.3 PLAPS 80 80 −300,365 ALPHA_BF 0.008 0.008 0.006,0.01 Notes: a s: elevation bands below 4700 m; b g: glaciered elevation bands over 4700 m.

Table 7. Range and mean values of calibration parameters in subbasin-elevation band scale, Chile River Basin.

Modiﬁed Model Parameter Name Average Range SFTMP l a 0.77 −4.6,3.0 SFTMP s b 1.01 −4.6,3.0 SFTMP g c 1.14 −4.6,3.0 SMTMP l 0.42 −3.0,4.0 SMTMP s 0.90 −1.0,4.0 SMTMP g 1.66 0.5,5.0 SMFMN s 4.06 2.00,6.25 SMFMN g 4.93 2.0,7.0 SMFMX s 4.64 2.00,6.25 SMFMX g 5.38 3.0,8.0 TIMP s 0.74 0.09,1.00 TIMP g 0.50 0.01,1.00 Notes: a l: elevation bands below 3300 m; b s: elevation bands between 3300 and 4300 m; c g: glaciered elevation bands over 4300 m.

Figure4 shows how the simulated monthly flow and mean monthly flow fit the observed mean monthly flow cycle using the modified algorithm for some of the selected reaches. This degree of accuracy in simulation of the seasonal pattern of monthly flow is only achievable by adjusting melt parameters based on elevation bands. As demonstrated in Figure5, SMFMX changes in lower elevation bands had more influence on the rising limb of the flow curve, whereas the descending limb was more sensitive to SMFMX changes in the higher elevation bands (glaciered areas). This indicates that the late spring flow is under the influence of snowmelt at lower elevations and that late summer flow is controlled by glacier melt at higher elevations. This can also be investigated by analyzing the melting lag time at the elevation bands. Seasonal melting from October 1998 to October 1999 from each of the elevation bands of a typical subbasin with a wide range of elevations is presented in Figure6a. Melting for glacier-free elevation bands with seasonal snow cover are presented in Figure6b for high elevation bands with permanent snow cover or glacier. In Figure6a, the snowpack in elevation bands 1, 2, 3, 4, and 5 has completely vanished by mid-August, whereas in Figure6b the glacier ablation reaches its peak in August and continues through October. The melt was decreased at lower altitudes by decreasing the SMTMP (snow melt base temperature) and SFTMP (snowfall temperature) at lower elevation bands. Water 2017, 9, 111 11 of 19 Water 2017, 9, 111 12 of 20

100 Reach 23 ‐Rhone 50 Reach 23_Rhone 45 80 40 35 60 30 (cms) 25 40

(cms) 20

Flow 15 20 Flow 10 5 0 0 05 04 00 01 03 96 97 98 99 94 93 07 06 02 04 95 93 ‐ ‐ ‐ ‐ ‐ ‐ ‐ ‐ ‐ ‐ ‐ ‐ ‐ ‐ ‐ ‐ ‐ Jul p Jan Jun Oct Apr Sep Feb Dec Aug Nov Mar May Jul Jan Jan Jun Oct Oct Apr Se Feb Sep Dec Dec Aug Nov Nov Mar May Observed Lump model Modified model

16 Reach 2‐Rhone 10 Reach 2‐Rhone 14 9 12 8 10 7

(cms) 6 8 5 (cms) 6 Flow 4 4 Flow 3 2 2 0 1 93 94 95 96 97 98 99 00 01 02 03 04 05 06 07

‐ ‐ ‐ ‐ ‐ ‐ ‐ ‐ ‐ ‐ ‐ ‐ ‐ ‐ ‐ 0 Jul Jan Jan Jan Jan Jan Jan Jan Jan Jan Jan Jan Jan Jan Jan Jan Jan Jun Oct Apr Feb Sep Dec Aug Nov Mar May

600 Reach 133‐Vakhsh 450 Reach 133‐Vakhsh 400 500 350 400 300 250 (cms) 300 200 (cms)

200 Flow 150 100 Flow 100 50 0 0 85 84 83 82 81 85 84 85 83 84 82 83 81 82 81 Jul Jan ‐ ‐ ‐ ‐ Jun ‐ ‐ ‐ ‐ ‐ ‐ ‐ ‐ ‐ ‐ ‐ Oct Apr Sep Feb Dec Aug Nov Mar May Jan Jan Jan Jan Jan Sep Sep Sep Sep Sep May May May May May Observed Lump model Modified model 14 Reach79‐Mendoza 30 Reach79‐Mendoza 12 25 10 20 8 (cms)

15 6 (cms)

10 Flow 4 Flow 5 2 0 0 Jul 04 03 99 98 94 93 07 06 05 02 03 01 00 97 98 96 95 93 Jan Jun Oct Apr Sep Feb ‐ ‐ ‐ ‐ ‐ ‐ ‐ ‐ ‐ ‐ ‐ ‐ ‐ ‐ ‐ ‐ ‐ ‐ Dec Aug Nov Mar May Jul Jul Jul Jan Jan Jan Sep Sep Sep Nov Nov Nov Mar Mar Mar May May May Observed Lump model Modified model

Figure 4. Cont.

Water 2017, 9, 111 12 of 19 Water 2017,, 9,, 111 13 of 20 Water 2017, 9, 111 13 of 20

120 Reach84‐‐Mendoza 120 Reach84‐Mendoza 70 70 Reach84‐‐Mendoza 100 70 Reach84‐Mendoza 100 60 80 60 80 50 80 50 (cms) (cms) 60 40 (cms)

(cms) 60 40

(cms) (cms) 60 40 30 Flow

Flow 30

Flow 40 40 Flow 30 Flow 40 Flow 20 20 20 20 10 10 0 0 0 0 03 04 98 99 93 94 06 07 05 01 02 03 00 96 97 98 95 93 0 Jul ‐ ‐ ‐ ‐ ‐ ‐ ‐ ‐ ‐ ‐ ‐ ‐ ‐ ‐ ‐ ‐ ‐ ‐ 03 04 98 99 93 94 06 07 05 01 02 03 00 96 97 98 95 93 Jan Jun Oct Apr Sep Feb Jul Dec ‐ ‐ ‐ ‐ ‐ ‐ ‐ ‐ ‐ ‐ ‐ ‐ ‐ ‐ ‐ ‐ ‐ ‐ Aug Nov Mar Jan May Jun Oct 03 04 98 99 93 94 06 07 05 01 02 03 00 96 97 98 95 93 Apr Sep Feb Dec Aug Nov Mar Jul ‐ ‐ ‐ ‐ ‐ ‐ Jul Jul Jul ‐ ‐ ‐ ‐ ‐ ‐ ‐ ‐ ‐ ‐ ‐ ‐ May Jan Jan Jan Jan Jun Oct Apr Sep Feb Sep Sep Sep Dec Jul Jul Jul Aug Nov Nov Nov Nov Mar Mar Mar Mar May Jan Jan Jan May May May Sep Sep Sep Nov Nov Nov Mar Mar Mar Jul Jul Jul May May May Jan Jan Jan Sep Sep Sep Nov Nov Nov Mar Mar Mar May May May

9 Reach90‐Mendoza 9 Reach90‐Mendoza 8 Reach90‐Mendoza 87 7 6 Reach90‐Mendoza 7 6 Reach90‐Mendoza 6 Reach90‐Mendoza 5 (cms) 65

(cms) 5 5

(cms) 5 4 4 4

Flow 4 (cms)

4 Flow 3 (cms) 3 3

Flow 3 (cms) 32 3

2 Flow 2 21 Flow 2 1 Flow 2 1 1 0 1 0 1 06 07 05 01 02 03 03 04 00 96 97 98 98 99 95 93 93 94 0 ‐ ‐ ‐ ‐ ‐ ‐ ‐ ‐ ‐ ‐ ‐ ‐ ‐ ‐ ‐ ‐ ‐ ‐ 06 07 05 01 02 03 03 04 00 96 97 98 98 99 95 93 93 94 0 ‐ ‐ ‐ ‐ ‐ ‐ ‐ ‐ ‐ ‐ ‐ ‐ ‐ ‐ ‐ ‐ ‐ ‐ 06 07 05 01 02 03 03 04 00 96 97 98 98 99 95 93 93 94 0 Jul Jul Jul Jul ‐ ‐ ‐ ‐ ‐ ‐ ‐ ‐ ‐ ‐ ‐ ‐ ‐ ‐ ‐ ‐ ‐ ‐ Jan Jun Jan Jan Jan Oct Apr Feb Sep Jul Dec Sep Sep Sep Jul Jul Jul Aug Nov Mar Nov Nov Nov Mar Mar Mar Jan May Jun Jan Jan Jan May May May Oct Apr Feb Sep Dec Sep Sep Sep Aug Nov Mar Nov Nov Nov Jul Mar Mar Mar Jul Jul Jul May May May May Jan Jun Jan Jan Jan Oct Apr Feb Sep Dec Sep Sep Sep Aug Nov Mar Nov Nov Nov Mar Mar Mar May May May May

FigureFigure 4. Observed4. Observed and and simulated simulated monthly monthly runoff runoff and and mean mean monthly monthly runoff runoff using using SWAT SWAT with different with Figure 4. Observed and simulated monthly runoff and mean monthly runoff using SWAT with different melt algorithms for the calibration period at selected reached. meltdifferent algorithms melt algorithms for the calibration for the calibration period at period selected at reached.selected reached. 60 60 Glacier free bands 60 Glaciated bands 60 Glacier free bands 60 Glaciated bands 50 GlacierBelow free2900 bands m.s.l. GlaciatedOver 2900 bands m.s.l. 50 Below 2900 m.s.l. 50 Over 2900 m.s.l. 50 Below 2900 m.s.l. Over(Rhone 2900‐Reach m.s.l. 23) (Reach(Reach 23‐‐Rhone) 50 (Rhone‐Reach 23) 40 (Reach 23‐Rhone) 40 (Rhone‐Reach 23) 40 40 30 30 (cms)

30 30 (cms) (cms)

(cms) 30 30

(cms)

(cms) 20 20 20 Flow Flow 20 20Flow Flow

Flow 10 Flow 10 10 10 10 0 0 0 0 0 Jul Jul Jan Jun Oct Apr Sep Feb Jul Dec Jan Aug Jun Nov Oct Apr Mar Feb Sep Jul Jan Dec May Aug Jun Nov Oct Apr Mar Sep Feb Dec Jan Aug May Jun Nov Oct Apr Mar Jul Feb Sep Dec May Aug Nov Mar Jul Jan Jun May Oct Apr Sep Feb Dec Jan Aug Jun Nov Oct Apr Mar Feb Sep Dec May Aug Nov Mar May Figure 5. Variability of monthly flow from Reach 23 (65% glaciered area) to SMFMX changes FigureFigure 5. Variability5. Variability of monthlyof monthly ﬂow flow from from Reach Reach 23 23 (65% (65% glaciered glaciered area) area) to SMFMXto SMFMX changes changes (2 to Figure 5. Variability−1 −1 of monthly flow from Reach 23 (65% glaciered area) to SMFMX changes (2 to 8− mm/d1 −1−1∙∙c−1) in the upper and lower elevation bands. 8 mm/d(2 to 8 mm/d·c −)1∙c in−1) the in the upper upper and and lower lower elevation elevation bands. bands.

800 Glacier‐free bands Glaciered bands 800 Glacier‐free bands 700 Glaciered bands 700 GlacierBelow‐ 2900free bands m.s.l. 700 EB6 GlacieredOver 2900 bands m.s.l. 700 Below 2900 m.s.l. Over 2900 m.s.l.

700 700 Below 2900 m.s.l. EB6EB7 Over 2900 m.s.l.

EB7 600 EB1 500 EB7EB8

EB1 (mm

600 500 EB8 EB1EB2 (mm 600 500 EB8EB9

EB2 (mm 500 EB2EB3 EB9 w.e.)

500 EB3 300 EB9EB10 w.e.) 500 EB3EB4 300 EB10 w.e.)

EB4 300 EB10

400 balance

400 EB4EB5 balance (mm

EB5 (mm

400 balance 100

EB5 100 (mm 300

300 100w.e.) mass w.e.)

melt 300 mass

w.e.) melt

mass

200 ‐100 Jul

Jan melt Jun Oct Oct Apr Feb Sep

‐100 Jul Dec Dec

200 Aug Nov Nov Mar Jan May Jun Oct Oct Apr Feb Sep Dec Dec Aug Nov Nov Mar

‐100 Jul

200 May Jan Jun Oct Oct Apr Feb Sep Dec Dec Aug Nov Nov Snow Mar

100 May specific Snow

100 ‐300

specific 1998 1999 Snow 100 ‐300 1998 1999 specific 0 ‐300 1998 1999 0 Net 0 Net ‐500 99 99 99 99 99 99 99 99 98 98 99 99 99 98 Net ‐ ‐ ‐ ‐ ‐ ‐ ‐ ‐ ‐ ‐ ‐ ‐ ‐ ‐ ‐500 99 99 99 99 99 99 99 99 98 98 99 99 99 98 ‐ ‐ ‐ ‐ ‐ ‐ ‐ ‐ ‐ ‐ ‐ ‐ ‐ ‐ ‐500 99 99 99 99 99 99 99 99 98 98 99 99 99 98 Jul ‐ ‐ ‐ ‐ ‐ ‐ ‐ ‐ ‐ ‐ ‐ ‐ ‐ ‐ Jan Jun Oct Oct Apr Sep Feb Jul Dec Aug Nov Nov Mar Jan May Jun Oct Oct Apr Sep Feb ‐700 Dec Aug Nov Nov Mar Jul May ‐700 Jan Jun Oct Oct Apr Sep Feb Dec Aug Nov Nov Mar May ‐700 (a) (b) (a) (b) Figure 6. Seasonal melting from the elevation bands from October 1998 to October 1999: (a) glacier Figure 6. Seasonal melting from the elevation bands from October 1998 to October 1999: (a) glacier Figurefree elevation 6. Seasonal bands melting with seasonal from the snow elevation cover; bands (b) high from elevation October bands 1998 with to Octoberpermanent 1999: snow (a cover) glacier free elevation bands with seasonal snow cover; (b) high elevation bands with permanent snow cover freeor elevation glacier. bands with seasonal snow cover; (b) high elevation bands with permanent snow cover or glacier. or glacier.

Water 2017, 9, 111 13 of 19

Calibration and validation results for all river basins are given in Tables8 and9. In the watersheds with highly glaciered drainage areas (Reaches 2 and 23) in the Rhone River Basin, PBIAS was improved when the modified snow algorithm was used instead of lumping all of the parameters for entire watershed area Both RSR and NSE also were improved using the modified model, although the improvement is more significant in simulation of magnitude of flow. There is negligible improvement in the simulated flow from watersheds with small areas of glacier (Reach 4) when using the modified snow algorithm in comparison to the lumped snow algorithm.

Table 8. Model performance by both lumped and modiﬁed snow algorithms for calibration period.

Calibration RSR NSE PBIAS Performance Glacier% Reach# MODIS 1 2 1 2 1 2 2 (GLIMS) 2 41.00 (46.15) 0.49 0.41 0.76 0.83 −24.7 −13.2 Very good 4 0.00 (5.1) 0.51 0.51 0.75 0.74 +2.7 +5.5 Very good Rhone 14 26.46 (26.97) 0.39 0.30 0.85 0.91 −7.3 −1.5 Very good 23 62.34 (65.10) 0.38 0.23 0.86 0.95 +25.9 +2.2 Very good 96 0.44 0.44 0.81 0.80 −6.6 −8.0 Very good 122 14.93 (10.91) 0.48 0.47 0.77 0.78 +10.5 +19.8 Satisfactory Narayani 123 9.77 (9.38) 0.39 0.40 0.85 0.84 +2.1 +5.3 Very good 133 16.55 (13.16) Good 159 9.20 0.56 0.57 0.69 0.68 +39.5 +34.9 Unsatisfactory 72 31.00 (31.00) 0.51 0.44 0.74 0.80 +18.5 +2.8 Very good Vakhsh 133 11.10 0.50 0.35 0.75 0.87 +31.2 +9.1 Very good 109 9.66 0.57 0.53 0.67 0.72 −3.8 −7.4 Good 90 8.75 0.82 0.64 0.32 0.59 −15.8 +3.1 Satisfactory 84 15.00 0.55 0.48 0.70 0.77 +22.5 +5.0 Very good Mendoza 82 3.21 0.69 0.65 0.52 0.58 +7.2 +7.1 Satisfactory 79 6.00 0.85 0.69 0.28 0.50 −17.9 +5.5 Satisfactory 86 4.34 0.59 0.66 0.65 0.56 −2.9 −14.3 Satisfactory 6 0.17 0.58 0.67 −1.3 Good 66 9.68 0.70 0.47 +8.1 Unsatisfactory 5 5.70 0.62 0.61 +33 Unsatisfactory 7 15.47 0.62 0.62 +10.9 Satisfactory 9 12.73 0.76 0.41 −32.4 Unsatisfactory 14 16.00 0.52 0.73 +10.3 Satisfactory 37 27.47 0.76 0.42 +4.3 Unsatisfactory 38 14.63 0.84 0.29 −42.7 Unsatisfactory Chile 39 15.70 0.82 0.33 +38 Unsatisfactory 40 10.64 0.68 0.53 +13.7 Satisfactory 55 9.03 0.70 0.50 +18.4 Satisfactory 59 16.89 0.67 0.55 +36.5 Satisfactory 76 15.91 0.82 0.33 +8.8 Unsatisfactory 77 12.28 0.77 0.41 −6.0 Unsatisfactory 86 12.28 0.58 0.66 +6.7 Good 100 16.35 0.68 0.53 +3.8 Satisfactory 108 0.00 0.67 0.56 +34.2 Unsatisfactory Notes: 1 Lump model; 2 Distributed model. Water 2017, 9, 111 14 of 19

Table 9. Model performance by modiﬁed snow algorithms for validation period.

Reach# RSR NSE PBIAS Performance Reach# RSR NSE PBIAS Performance 2 0.44 0.81 −21.6 Satisfactory 6 0.81 0.34 +50.0 Unsatisfactory 4 0.62 0.61 −9.4 Good 66 0.89 0.21 −18.7 Unsatisfactory Rhone 14 0.42 0.82 −16.1 Satisfactory 5 0.79 0.38 +41.0 Unsatisfactory 23 0.33 0.89 −13.1 Good 7 0.88 0.22 +4.7 Unsatisfactory 96 9 0.83 0.31 +40.2 Unsatisfactory 122 14 0.89 0.21 +51.4 Unsatisfactory Narayani 123 0.47 0.78 +22.5 Satisfactory 37 0.94 0.11 +46.3 Unsatisfactory 133 0.55 0.70 +18.0 38 0.92 0.15 +44.0 Unsatisfactory 159 0.55 0.70 +12.9 GoodChile 39 1.00 0.00 +41.2 Unsatisfactory 72 40 0.71 0.50 +11.0 Satisfactory Vakhsh 133 55 0.75 0.44 −0.2 Unsatisfactory 109 0.52 0.73 −8.15 Good 59 0.59 0.65 +31.0 Unsatisfactory 90 1.11 −0.24 +5.8 Unsatisfactory 76 1.07 −0.14 −8.3 Unsatisfactory 84 0.54 0.71 +6.2 Good 77 Mendoza 82 0.64 0.59 +12.5 Satisfactory 86 0.66 0.57 +2.6 Satisfactory 79 0.73 0.46 −2.1 Unsatisfactory 100 0.77 0.41 −1.8 Unsatisfactory 86 0.87 0.24 −21.7 Unsatisfactory 108 0.77 0.41 +39 Unsatisfactory

For Narayani River Basin, PBIAS and NSE values in Table8 reveal similar model performance by both lumped and modified snow algorithm methods, possibly as a result of the diminishing influence of glacier/snowmelt on runoff resulting from coincident summer monsoon precipitation in the Narayani River Basin. This monsoonal influence leads to accurate prediction of seasonal flow by rainfall–runoff models without explicitly considering the snow hydrology process. In Table8, positive PBIAS in simulation of flow magnitude from all gauge stations (except Reach 96) means that the model has underpredicted the volume of flow. This systematic bias might be due to an underestimation of the NCEP reanalysis spring/summer precipitation (May–September) in the Narayani River basin of 20%. The modified snow algorithm performed very well in simulation of monthly flow from the glaciered areas of the Vakhsh River Basin with RSR smaller than 0.5, PBIAS smaller than 10 percent and NSE greater than 0.70. Major difficulties in simulation of monthly flow from the Vakhsh River Basin were a lack of data for calibration and the low quality of the climate data, especially in the northern half of the river basin. The short calibration period of 2.5 years for Reach 72 was not long enough to capture the long-term variability of the monthly flow. The simulated mean monthly flow volume and cycle from the glaciated areas (drainage area to Reaches 72 and 133) by the modified snow algorithm during the summer months is in better agreement with the observed data in comparison to the mean monthly flow simulated using the lumped model for Reach 133 (Figure4). At the main outlet (Reach 109), only the monthly flow variation shows improvement. Consequently, for an entire river basin it cannot be concluded that the modified model has better performance than the lumped model. Comparison between observed and simulated monthly flow from the main outlet of the basin during the validation period indicated that there was good agreement, verified by RSR, PBIAS, and NSE of 0.52, −8.15, and 0.73, respectively. Among the gauged watersheds of the Mendoza River Basin, the drainage area of Reach 84 contains the largest area covered by glaciers. Subbasin 90 is a single gauged small subbasin, so it is a good example of the streamflow response to melt parameter distribution. There was a considerable improvement in the model performance when using the modified model instead of the lumped model in Reaches 79, 84, and 90. Figure4 shows that the peaks simulated by the modified model match the observed peak flows. This level of accuracy is only achievable by adjusting the melt parameters for seasonal snow and permanent snow for each elevation band since the lumped model was not able to capture the peaks by calibrating the model for many different sets of melt parameters. Although the results from the modified model show significant improvement in highly glaciated areas and negligible improvement in less glaciated areas, the simulated monthly flow at the main outlet (Reach 86) does not show any improvement with the modified model. It can be seen in Table8 that the model performance in simulation of downstream flow (e.g., Outlet 86 in the Mendoza River basin) declines when applying Water 2017, 9, 111 15 of 19 the modified snow algorithm. The new approach enabled calibration of the model versus the flow from upland catchments and optimization of the simulation results by adjusting the parameters separately for subbasin/elevation bands, while in the lump method the parameters were adjusted in order to get the reasonable result and were not necessarily the best at the gauged streams. It seems that the parameter value optimization for upland basins negatively affects the downstream flow in comparison to the lump parameter adjustment method. In other words, factors other than snowmelt could affect the downstream flow, and further model setup and parameter adjustment (groundwater, plant growth and soil moisture parameters, etc.) are necessary for lowlands and flat areas, which were out of the scope of this study. Another reason for declining the accuracy of downstream flow simulation could be the use of inaccurate reanalysis precipitation data. While snowfall/melt is the dominant hydrologic process in dry highlands (highland areas of Mendoza are very dry), flow is less affected by precipitation events, and snow parameter adjustment improves the flow simulation considerably. In lowlands, where the rainfall-runoff model is dominant, the flow is directly under the influence of precipitation events and any problem in input precipitation data could directly decline the accuracy of downstream flow simulation. So, when the results for snowy basins show improvement, flow in lowland areas may not necessarily improve. Comparisons between observed and simulated monthly streamflow during the validation period indicate range of model performances from good to unsatisfactory. Comparisons between observed and simulated monthly flow from Chilean river basins during the calibration period indicate poor to satisfactory simulation. The RSR and NSE values are generally lower than those values obtained in simulation of monthly flows from the other river basins in this study. Winter precipitation (October to March) inter-annual variability in the Andes is linked to ENSO (El Niño–Southern Oscillation) events and consequently reveals a more complex response of streamflow [47], which may explain poor model performance in simulation of seasonal flow variability in the Mendoza River Basin and the Chilean river basins in the dry central Andes.

4. Conclusions Treating the glaciered and un-glaciered areas in the watersheds separately improved SWAT model performance significantly in simulation of volume and variation of runoff in glaciered areas. Spatial and temporal variations of melt rates mainly depend on the spatial and temporal variations of melt factors in hydro-glacial models. While temporal variations of melt factors have been considered in the SWAT model in the past, there has been no consideration of spatial variations in melt factors and lag time factor, which are directly influenced by surface type (snow and ice). In this study, these spatial variations were specifically taken into account. Model performance using the snow algorithms also depends on the climate of a river basin. Significance of melt water may be negligible in watersheds where the melt season coincides with monsoon precipitation, resulting in no significant difference in the simulation results from the two melt algorithms when monsoons were a factor. For the river basins in the central Andes, applying the modified distributed snow algorithm considerably improved the model performance in simulation of runoff volume in comparison with the lumped snow algorithm, while variation of runoff did not show any improvement. This may be due to high inter-annual and annual variability of flow in these regions, which are more dominated by rainfall-runoff relationships than snowmelt. We can conclude that considering the spatial variations of associated melt parameters significantly improves the SWAT performance in simulation of runoff volume and its seasonal variation in highly glaciated river basins. While distributed process-based energy budget models have been tested in the SWAT model, no studies have been done to incorporate enhanced temperature-index models into SWAT. Incorporation of an enhanced temperature-index model into the modified snow algorithm of SWAT should be an objective of future studies on enhancing the SWAT snow hydrologic process. The modified snow algorithm did not enhance SWAT model performance in simulation of the streamflow at the main outlet of the river basins. Based on this, one might be tempted to draw the conclusion that at the macro scale the lumped model is preferable to the distributed model. This, however, requires further Water 2017, 9, 111 16 of 19 investigation into modeling such hydrologic processes as glacier surface mass balance and transient snow line altitude. Therefore, it is suggested that the model not only be calibrated using runoff but also be calibrated against the individual snow hydrology components.

Author Contributions: For research articles with several authors, a short paragraph specifying their individual contributions must be provided. The following statements should be used “Nina Omani and Raghavan Srinivasan conceived and designed this study and prepared the initial manuscript; Raghavan Srinivasan, Raghupathy Karthikeyan and Patricia K. Smith supervised reporting and manuscript preparation. All the authors finalized the research article together.” Conflicts of Interest: The authors declare no conflict of interest.

AppendixWater A 2017, 9, 111 17 of 20 Water 2017, 9, 111 17 of 20 Appendix A Appendix A

(a) (b)

Figure A1. Glacier snow(a) depth and area across the (a) Rhone and (b) Narayani(b) river basins extracted Figure A1. Glacier snow depth and area across the (a) Rhone and (b) Narayani river basins extracted Figurefrom the A1. GLIMS Glacier glacier snow depthoutlines. and Green area acrossdots show the ( alocations) Rhone andof glaciers (b) Narayani with no river depth basins information. extracted from the GLIMS glacier outlines. Green dots show locations of glaciers with no depth information. from the GLIMS glacier outlines. Green dots show locations of glaciers with no depth information.

(a) (b) (a) (b)

Figure A2. Cont.

Water 2017, 9, 111 17 of 19 Water 2017, 9, 111 18 of 20

(c)

Figure A2. Glacier snow depth and area across the (a) Mendoza, (b) central Chile, and (c) Vakhsh Figure A2.riverGlacier basins. snowMinimum depth snow and cover area area across from the MOD10A2 (a) Mendoza, at the ( bend) central of the Chile,melting and season (c) Vakhshfrom river basins.February Minimum 2002 snow to 2010 cover was area considered from MOD10A2 as glacier or at permanent the end ofsnow the cover melting for Mendoza, season from central February 2002 toChile, 2010 wasand part considered of the Vakhsh as glacier basin. Green or permanent dots show snowlocations cover of glaciers for Mendoza, with no depth central information. Chile, and part of the Vakhsh basin. Green dots show locations of glaciers with no depth information. References

References1. Anderson, E.A. National Weather Service River Forecast System‐Snow Accumulation and Ablation Model; NOAA Technical Memorandum NWS‐HYDRO‐17; National Oceanographic and Atmospheric Administration 1. Anderson,Silver E.A. SpringsNational Md: Washington, Weather Service DC, USA, River 1973. Forecast System-Snow Accumulation and Ablation Model; NOAA Technical2. Moor, Memorandum R.D. Application NWS-HYDRO-17; of a Conceptual Streamflow National OceanographicModel in a Glacierized and Drainage Atmospheric Basin. Administration J. Hydrol. 1993, Silver Springs150 Md:, 151–168. Washington, DC, USA, 1973. 3. Schaper, J.; Seidel, K.; Martinec, J. Precision Snow Cover and Glacier Mapping for Runoff Modelling in a 2. Moor, R.D. Application of a Conceptual Streamﬂow Model in a Glacierized Drainage Basin. J. Hydrol. 1993, High Altitude Alpine Basin. In Remote Sensing and Hydrology, Proceedings of the Santa Fe Symposium 2000; 150, 151–168.International [CrossRef Association] of Hydrological Sciences: Wallingford, CT, USA, 2001; pp. 105–111. 3. Schaper,4. Debele, J.; Seidel, B.; Srinivasan, K.; Martinec, R. Comparison J. Precision of the Performances Snow Cover of Temperature and Glacier‐Index Mapping and Process for‐Based Runoff Energy Modelling Budget in a High AltitudeSnowmelt Alpine Estimation Basin. Approaches In Remote in the SensingSWAT Model and; Hydrology,Texas A&M ProceedingsUniversity Spatial of the Sciences Santa Fe Laboratory: Symposium 2000; InternationalCollege Association Station, TX, USA, of Hydrological 2005; p. 40. Sciences: Wallingford, CT, USA, 2000; pp. 105–111.

4. Debele,5. Hock, B.; Srinivasan, R. Temperature R. Comparison Index Melt Modelling of the Performances in Mountain of Areas. Temperature-Index J. Hydrol. 2003, and 282, Process-Based 104–115. Energy Budget 6. Wang, X.; Melesse, A.M. Evaluation of the SWAT Model’s Snowmelt Hydrology in a Northwestern Snowmelt Estimation Approaches in the SWAT Model; Texas A&M University Spatial Sciences Laboratory: Minnesota Watershed. Trans. ASAE 2005, 48, 1359–1376. College7. Ahl, Station, R.S.; Woods, TX, USA, S.W.; 2005; Zuuring, p. 40. H.R. Hydrologic Calibration and Validation of SWAT in a Snow‐Dominated 5. Hock, R.Rocky Temperature Mountain Watershed Index Melt Montana Modelling U.S.A. in J. MountainAm. Water Resour. Areas. Assoc.J. Hydrol. 2008, 442003, 1411–1430., 282, 104–115. [CrossRef] 6. Wang,8. X.;Zhang, Melesse, X.S.; Srinivasan, A.M. Evaluation R.; Debele, of theB.; Hao, SWAT F.H. Model’s Runoff Simulation Snowmelt of Hydrology the Headwaters in a of Northwestern the Yellow River Minnesota Watershed.UsingTrans. the SWAT ASAE Model2005 with, 48 ,Three 1359–1376. Snowmelt [CrossRef Algorithms.] J. Am. Water Resour. Assoc. 2008, 44, 48–61.

7. Ahl,9. R.S.;Stehr, Woods, A.; Debels, S.W.; P.; Zuuring, Arumi, J.L.; H.R. Romero, Hydrologic F.; Alcayaga, Calibration H. Combining and Validation the Soil and of Water SWAT Assessment in a Snow-Dominated Tool (SWAT) and MODIS Imagery to Estimate Monthly Flows in a Data‐Scarce Chilean Andean Basin. Hydrol. Rocky Mountain Watershed Montana U.S.A. J. Am. Water Resour. Assoc. 2008, 44, 1411–1430. [CrossRef] Sci. J. 2009, 54, 1053–1067. 8. Zhang,10. Pradhanang, X.S.; Srinivasan, S.M.; Anandhi, R.; Debele, A.; Mukundan, B.; Hao, F.H. R.; RunoffZion, M.S.; Simulation Pierson, D.C.; of the Schneiderman, Headwaters E.M.; of theMatonse, Yellow River Using theA.; SWATFeri, A. ModelApplication with of Three SWAT Snowmelt Model to Algorithms.Assess SnowpackJ. Am. Development Water Resour. and Streamflow Assoc. 2008 in the, 44 , 48–61. [CrossRefCannonsville] Watershed New York USA. Hydrol. Process. 2011, 25, 3268–3277. 9. Stehr,11. A.;Srinivasan, Debels, R.; P.; Ramanarayanan, Arumi, J.L.; Romero, T.S.; Arnold, F.; Alcayaga,J.G.; Bednarz, H. S.T. Combining Large Area the Hydrologic Soil and Modeling Water Assessmentand Assessment: Part II. Model Application. J. Am. Water Resour. Assoc. 1998, 34, 91–101. Tool (SWAT) and MODIS Imagery to Estimate Monthly Flows in a Data-Scarce Chilean Andean Basin. 12. Arnold, J.G.; Muttiah, R.S.; Srinivasan, R.; Allen, P.M. Regional Estimation of Base Flow and Groundwater Hydrol. Sci. J. 2009, 54, 1053–1067. [CrossRef] Recharge in the Upper Mississippi Basin. J. Hydrol. 2000, 227, 21–40. 10. Pradhanang,13. Arnold, S.M.; J.G.; Anandhi,Srinivasan, A.; R.; Mukundan,Muttiah, R.S.; R.; Allen, Zion, P.M.; M.S.; Walker, Pierson, C. Continental D.C.; Schneiderman, Scale Simulation E.M.; of Matonse, the A.; Feri, A.Hydrologic Application Balance. of SWAT J. Am. Model Water Resour. to Assess Assoc. Snowpack 1999, 35, 1037–1052. Development and Streamﬂow in the Cannonsville Watershed14. Arnold, New J.G.; York Srinivasan, USA. Hydrol. R.; Muttiah, Process. R.S.;2011 ,Williams,25, 3268–3277. J.R. Large [CrossRef Area ]Hydrologic Modeling and 11. Srinivasan,Assessment: R.; Ramanarayanan, Part I. Model Development. T.S.; Arnold, J. Am. J.G.; Water Bednarz, Resour. Assoc. S.T. 1998 Large, 34, 73–89. Area Hydrologic Modeling and 15. Di Luzio, M.; Srinivasan, R.; Arnold, J.G. Integration of Watershed Tools and SWAT Model into BASINS. J. Assessment: Part II. Model Application. J. Am. Water Resour. Assoc. 1998, 34, 91–101. [CrossRef] Am. Water Resour. Assoc. 2002, 38, 1127–1141. 12. Arnold, J.G.; Muttiah, R.S.; Srinivasan, R.; Allen, P.M. Regional Estimation of Base Flow and Groundwater Recharge in the Upper Mississippi Basin. J. Hydrol. 2000, 227, 21–40. [CrossRef] 13. Arnold, J.G.; Srinivasan, R.; Muttiah, R.S.; Allen, P.M.; Walker, C. Continental Scale Simulation of the Hydrologic Balance. J. Am. Water Resour. Assoc. 1999, 35, 1037–1052. [CrossRef] 14. Arnold, J.G.; Srinivasan, R.; Muttiah, R.S.; Williams, J.R. Large Area Hydrologic Modeling and Assessment: Part I. Model Development. J. Am. Water Resour. Assoc. 1998, 34, 73–89. [CrossRef] Water 2017, 9, 111 18 of 19

15. Di Luzio, M.; Srinivasan, R.; Arnold, J.G. Integration of Watershed Tools and SWAT Model into BASINS. J. Am. Water Resour. Assoc. 2002, 38, 1127–1141. [CrossRef] 16. Fontaine, T.A.; Cruickshank, T.S.; Arnold, J.G.; Hotchkiss, R.H. Development of a Snowfall-Snowmelt Routine for Mountainous Terrain for the Soil Water Assessment Tool (SWAT). J. Hydrol. 2002, 262, 209–223. [CrossRef] 17. Anderson, E.A. A Point Energy and Mass Balance Model of a Snow Cover; NOAA Technical Report NWS 19; National Oceanic and Atmospheric Administration Silver Springs Md: Washington, DC, USA, 1976; 150p. 18. Debele, B.; Srinivasan, R.; Gosain, A.K. Comparison of Process-Based and Temperature-Index Snowmelt Modeling in SWAT. Water Resour. Manag. 2010, 24, 1065–1088. [CrossRef] 19. Rahman, K.; Maringanti, C.; Beniston, M.; Widmer, F.; Abbaspour, K.; Lehmann, A. Streamflow Modeling in a Highly Managed Mountainous Glacier Watershed using SWAT: The Upper Rhone River Watershed Case in Switzerland. Water Resour. Manag. 2013, 27, 323–339. [CrossRef] 20. Braithwaite, R.J. Temperature and Precipitation Climate at the Equilibrium-Line Altitude of Glaciers Expressed by the Degree-Day Factor for Melting Snow. J. Glaciol. 2008, 54, 437–444. [CrossRef] 21. Omani, N.; Srinivasan, R.; Smith, P.K.; Karthikeyan, R. Glacier mass balance simulation using SWAT distributed snow algorithm. Hydrol. Sci. J. 2016.[CrossRef] 22. Omani, N.; Srinivasan, R.; Karthikeyan, R.; Venkatta, K.; Smith, P.K. Impacts of climate change on the glacier melt runoff from five river basins. Trans. ASABE 2016, 59.[CrossRef] 23. Neitsch, S.L.; Arnold, J.G.; Kiniry, J.R.; Williams, J.R. Soil and Water Assessment Tool Theoretical Documentation Version 2009; Texas Water Ressources Institute Technical Report 406; Texas A&M University System: College Station, TX, USA, 2011. 24. Brown, J.F.; Loveland, T.R.; Merchant, J.W.; Reed, B.C.; Ohlen, D.O. Using Multisource Data in Global Land Cover Characteristics: Concepts Requirements and Methods. Photogramm. Eng. Remote Sens. 1993, 59, 977–987. 25. Batjes, N.H. ISRIC-WISE Derived Soil Properties on a 5 by 5 Arc-Minutes Grid (Version 1.2); Report 2012/01; ISRIC—World Soil Information Wageningen: Wageningen, The Netherlands, 2012. 26. Saha, S. The NCEP Climate Forecast System Reanalysis. Bull. Am. Meteorol. Soc. 2010, 91, 1015–1057. [CrossRef] 27. Huss, M.; Farinotti, D.; Bauder, A.; Funk, M. Modelling Runoff from Highly Glacierized Alpine Drainage Basins in a Changing Climate. Hydrol. Process. 2008, 22, 3888–3902. [CrossRef] 28. Zemp, M.; Hoelzle, M.; Haeberli, W. Distributed Modeling of the Regional Climatic Equilibrium Line Altitude of Glaciers in the European Alps. Glob. Plant. Chang. 2007, 56, 83–100. [CrossRef] 29. Williams, V.S. Present and Former Equilibrium-Line Altitudes near Mount Everest, Nepal and Tibet. Arct. Alp. Res. 1983, 15, 201–211. [CrossRef] 30. Savoskul, O.S.; Smakhtin, V. Glacier Systems and Seasonal Snow Cover in Six Major Asian River Basins: Hydrological Role under Changing Climate; International Water Management Institute (IWMI): Colombo, Sri Lanka, 2013; 53p. 31. Zemp, M.; Frey, H.; Gärtner-Roer, I.; Nussbaumer, S.U.; Hoelzle, M.; Paul, F.; Haeberli, W.; World Glacier Monitoring Service (WGMS). Fluctuations of Glaciers 2005–2010 (Vol. X); ICSU(WDS)/IUGG(IACS)/UNEP/ UNESCO/WMO; World Glacier Monitoring Service: Zurich, Switzerland, 2012. 32. Brenning, A. Geomorphological, Hydrological, and Climatic Significance of Rock Glaciers in the Andes of Central Chile (33–35 S). Permafr. Periglac. Process. 2005, 16, 231–240. [CrossRef] 33. Hastenrath, S. On the Pleistocene Snow-Line Depression in the Arid Regions of the South American Andes. J. Glaciol. 1971, 10, 225–267. [CrossRef] 34. Kull, C.; Grosjean, M.; Veit, H. Modeling Modern and Late Pleistocene Glacio-Climatological Conditions in the North Chilean Andes. Clim. Chang. 2002, 52, 359–381. [CrossRef] 35. Lliboutry, L. Rock Glaciers in the Dry Andes. In Proceedings of the International Symposium ‘Glacier Mass Balance, Fluctuations and Runoff’, Alma-Ata, Kazakhstan, 30 September–5 October 1985. 36. Lliboutry, L. Glaciers of the dry Andes. In Satellite Image Atlas of Glaciers of the World: South America; Williams, R.S., Jr., Ferrigno, J.G., Eds.; (USGS Professional Paper 1386-I); US Geological Survey: Reston, VA, USA, 1999; pp. I119–I147. Water 2017, 9, 111 19 of 19

37. Raup, B.H.; Kieffer, H.H.; Hare, T.M.; Kargel, J.S. Generation of Data Acquisition Requests for the ASTER Satellite Instrument for Monitoring a Globally Distributed Target: Glaciers. IEEE Trans. Geosci. Remote Sens. 2000, 38, 1105–1112. [CrossRef] 38. Armstrong, R.; Raup, B.; Khalsa, S.J.S.; Barry, R.; Kargel, J.; Helm, C.; Kieffer, H. GLIMS Glacier Database. U.S. National Snow and Ice Data Center Boulder Colorado USA: Digital Media. 2010. Available online: http://www.glims.org (accessed on 23 February 2012). 39. Abbaspour, K.C.; Yang, J.; Maximov, I.; Siber, R.; Bogner, K.; Mieleitner, J.; Zobrist, J.; Srinivasan, R. Spatially-Distributed Modelling of Hydrology and Water Quality in the Prealpine/Alpine Thur Watershed Using SWAT. J. Hydrol. 2007, 333, 413–430. [CrossRef] 40. McKay, M.D.; Beckman, R.J.; Conover, W.J. Comparison of Three Methods for Selecting Values of Input Variables in the Analysis of Output from a Computer Code. Technometrics 1979, 21, 239–245. [CrossRef] 41. Rees, H.G.; Collins, D.N. Regional Differences in Response of Flow in Glacier-fed Himalayan Rivers to Climatic Warming. Hydrol. Process. 2006, 20, 2157–2169. [CrossRef] 42. Immerzeel, W.W. Historical Trends and Future Predictions of Climate Variability in the Brahmaputra Basin. Int. J. Clim. 2008, 28, 243–254. [CrossRef] 43. Nash, J.E.; Sutcliffe, J.V. River Flow Forecasting Through Conceptual Models Part I A Discussion of Principles. J. Hydrol. 1970, 10, 282–290. [CrossRef] 44. Moriasi, D.N.; Arnold, J.G.; Van Liew, M.W.; Bingner, R.L.; Harmel, R.D.; Veith, T.L. Model Evaluation Guidelines for Systematic Quantification of Accuracy in Watershed Simulations. Trans. ASABE 2007, 50, 885–900. [CrossRef] 45. Gupta, H.V.; Sorooshian, S.; Yapo, P.O. Status of Automatic Calibration for Hydrologic Models: Comparison with Multilevel Expert Calibration. J. Hydraul. Eng. 1999, 4, 135–143. [CrossRef] 46. Kayastha, R.B.; Ageta, Y.; Fujita, K. Use of Positive Degree-Day Methods for Calculating Snow and Ice Melting and Discharge in Glacierized Basins in the Langtang Valley Central Nepal. In Climate and Hydrology in Mountain Areas; De Jong, C., Collins, D., Ranzi, R., Eds.; Wiley: Chichester, UK, 2005; pp. 5–14. 47. Waylen, P.R.; Caviedes, C.N. Annual and Seasonal Streamflow Fluctuations of Precipitation and Streamflow in the Aconcagua River Basin Chile. J. Hydrol. 1990, 120, 79–102. [CrossRef]

© 2017 by the authors; licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).