Evaluating the Present Annual Water Budget of a Himalayan Headwater

PUBLICATIONS

Journal of Geophysical Research: Atmospheres

RESEARCH ARTICLE Evaluating the present annual water budget of a Himalayan 10.1002/2016JD026279 headwater river basin using a high-resolution Key Points: atmosphere-hydrology model • A significant disagreement in high-altitude precipitation estimates Lu Li1 , David J. Gochis2 , Stefan Sobolowski1 , and Michel D. S. Mesquita1 is found between gauge observations, TRMM, APHRODITE, and the WRF 1Uni Research Climate, Bjerknes Centre for Climate Research, Bergen, Norway, 2National Center for Atmospheric Research, • A large amount of precipitation in high mountainous areas in Beas Basin Boulder, Colorado, USA is occurring and is not properly accounted for in TRMM or APHRODITE • WRF-Hydro modeling system shows Abstract Understanding the present water budget in Himalayan Basins is a challenge due to poor in situ skill in capturing monthly discharge coverage, incomplete or unreliable records, and the limitations of coarse resolution gridded data set. In the variability and the daily discharge distribution study, a two-way coupled implementation of the Weather Research and Forecasting (WRF) Model and the WRF-Hydro hydrological modeling extension package (WRF/WRF-Hydro) was employed in its offline configuration, over a 10 year simulation period for a mountainous river basin in North India. A triple nest is employed, in which the innermost domain had 3 km for atmospheric model grids and 300 m for hydrological Correspondence to: L. Li, components. Two microphysical parameterization (MP) schemes are quantitatively evaluated to reveal how [email protected] differently MP influences orographic-related precipitation and how it impacts hydrological responses. The WRF-Hydro modeling system shows reasonable skill in capturing the spatial and temporal structure of fl Citation: high-resolution precipitation, and the resulting stream ow hydrographs exhibit a good correspondence Li, L., D. J. Gochis, S. Sobolowski, and with observation at monthly timescales, although the model tends to generally underestimate streamflow M. D. S. Mesquita (2017), Evaluating the amounts. The Thompson Scheme fits better to the observations in the study. More importantly, WRF shows present annual water budget of a “ ” Himalayan headwater river basin using that for high-altitude precipitation, a high bias is exhibited in winter precipitation from WRF, which is about a high-resolution atmosphere-hydrol- double to triple that as estimated from valley-sited rain gauges and remotely sensed precipitation estimates ogy model, J. Geophys. Res. Atmos., 122, from Tropical Rainfall Measuring Mission and Asian Precipitation - Highly-Resolved Observational Data 4786–4807, doi:10.1002/2016JD026279. Integration Towards Evaluation. Given the full annual cycle pattern and amount in high-altitude precipitation

Received 21 NOV 2016 and the statistical correspondence in discharge, it is concluded that the WRF-Hydro modeling system shows Accepted 1 APR 2017 potential for explicitly predicting potential changes in the atmospheric-hydrology cycle of ungauged or Accepted article online 7 APR 2017 poorly gauged basins. Published online 3 MAY 2017 Plain Language Summary Understanding the present water budget in Himalayan Basins is a challenge due to poor in situ coverage, incomplete or unreliable records, and the limitations of coarse resolution gridded data set. In a Himalayan headwater river basin, the Weather Research and Forecasting (WRF)-Hydro modeling system shows reasonable skill in capturing the precipitation and the resulting stream ﬂow hydrographs exhibit a good correspondence with observation at monthly timescales. More importantly, WRF shows that for high-altitude precipitation, a high “bias” is exhibited in winter precipitation from WRF, which is about double to triple that as estimated from valley-sited rain gauges and remotely sensed precipitation estimates from both Tropical Rainfall Measuring Mission and Asian Precipitation - Highly-Resolved Observational Data Integration Towards Evaluation. Given the full annual cycle pattern and amount in high-altitude precipitation and the statistical correspondence in discharge, it is concluded that the WRF-Hydro modeling system shows potential for explicitly predicting potential changes in the atmospheric-hydrology cycle of ungauged or poorly gauged basins.

1. Introduction The Himalayan Mountains are the source region of one of the world’s largest supplies of freshwater. More than 1.2 billion people in the surrounding regions are directly or indirectly reliant upon their resources [Kaser et al., 2010; Ménégoz et al., 2013]. Most of these areas are in developing countries with signiﬁcant rural, subsistence populations that are vulnerable to the hydrological impacts associated with changes in seasonal precipitation patterns associated with the Indian Summer Monsoon [Krishnamurthy et al., 2009; Christensen et al., 2013] and long-term storage associated with glacier growth/retreat [Yao et al., 2007].

©2017. American Geophysical Union. Despite this urgency, understanding even present-day hydroclimate variability in Himalayan Basins is a All Rights Reserved. challenge due to poor in situ coverage [Maussion et al., 2011], incomplete or unreliable records [Hewitt,

LI ET AL. EVALUATE WATER BUDGET IN HIMALAYAN BASIN 4786 Journal of Geophysical Research: Atmospheres 10.1002/2016JD026279

2005; Bolch et al., 2012; Hartmann and Andresky, 2013], and the limitations of coarse resolution dynamical models such as Global Circulation Models over complex terrain [Fyfe and Flato, 1999; Mass et al., 2002; Leung and Qian, 2003; Salathé et al., 2008]. A first step toward addressing these challenges has been the development of gridded precipitation estimates derived from a variety of sources such as satellite- based data, i.e., Tropical Rainfall Measuring Mission (TRMM) [Huffman et al., 2007; Yan et al., 2016; Adjei et al., 2016]; reanalysis data, i.e., ERA-Interim [Dee et al., 2011] and the WATCH Forcing Data Era-Interim [Weedon et al., 2014]; and rain gauge-based data, i.e., Climate Research Unit [Mitchell and Jones, 2005], Climate Prediction Center [Xie et al., 2010] and the Asian Precipitation - Highly-Resolved Observational Data Integration Towards Evaluation (APHRODITE) [Yatagai et al., 2012; Xu et al., 2016]. While the proliferation of remotely sensed, reanalysis and merged products has been a positive development for researchers, many studies find that the aforementioned data sets are often inconsistent with each other [e.g., Palazzi et al., 2013; Yatagai et al., 2012; Ménégoz et al., 2013]. Further, precipitation in the Himalayan region is strongly influenced by terrain, and regional patterns and amounts are not always captured by gridded precipitation data sets, which have grid spacing of 25 km at the high end. High-resolution dynamical simulations have shown promising in addressing some of the issues related to complex terrain. Numerical weather prediction models can provide reliable estimates of precipitation at high spatial and temporal resolutions [Zängl, 2007; Prein et al., 2013; Rasmussen et al., 2011; Collier et al., 2013; Ji and Kang, 2013a]. An added benefit of high-resolution dynamical approaches is that they not only provide data in great detail but also produce a full complement of variables, which allow for investigations of the physics that drive the local climate. This increases credibility and enables the simulation of complex local processes that might otherwise be overlooked [Arritt and Rummukainen, 2011; Pavelsky et al., 2012; Mayer et al., 2015].

Although Regional Climate Models (RCMs) outputs are able to describe the spatial variability of precipitation over the Himalayas, significant biases in precipitation still exist and a corollary uncertainty in the local- regional water budget remains unresolved. At relatively coarse resolution, RCMs (>20 km grid spacing) can represent atmospheric circulation features but they do not always provide adequate precipitation representation and exhibit large differences when compared to reanalyses, rain gauge, and satellite observations [Biskop et al., 2012; Dimri et al., 2013; Ménégoz et al., 2013; Tramblay et al., 2013; Ji and Kang, 2013b]. Conversely, high-resolution, convection permitting (<4 km grid spacing) RCMs have demonstrated reasonable skill in reproducing precipitation distribution and intensity patterns over complex terrain [e.g., Rasmussen et al., 2011, 2014; Collier et al., 2013]. Over the Himalayan region, in particular, Maussion et al. [2011] showed improved representation of precipitation in a high-resolution (2 km) dynamical simulation when compared to satellite products. Cossu and Hocke [2014] compared 13 WRF parameterization schemes over a bell-shaped mountain and found that the choice of microphysical scheme has important consequences for water phase component, hydrometeor distribution, and precipitation. At high resolution, and with selection of the proper parameterizations, RCMs have demonstrated the capability to capture the statistical features of orographic precipitation such as seasonality, relative intensity, and precipitation phase even if the timing of specific events is off due to internal model variability [Barstad and Caroletti, 2013; Rasmussen et al., 2011, 2014]. For example, an optimally performing WRF model setup, which employed the Thompson microphysical scheme [Thompson et al., 2008], was used to produce the High Asia Reanalysis (HAR) data set at 10 km resolution over the Tibetan Plateau [Maussion et al., 2014]. The resulting HAR precipitation data set validates well against in situ and satellite data and has high potential for glaciological and hydrological impact studies. But the authors emphasize the need for further verification and improvements. However, when it comes to hydrological applications, challenges persist despite the improvements in high- resolution modeling, due to the biases in RCM precipitation and temperature [Teutschbein and Seibert, 2010; Christensen et al., 2008; Graham et al., 2007]. Ideally, scientists would like to use RCM output data directly in hydrological applications. Even better would be to couple the atmospheric and hydrological modeling systems so that physical consistency, feedbacks, and exchanges are maintained throughout the processes. The advancements described in the previous paragraphs have brought us closer to these goals, but there are still challenges to overcome. Currently, many hydrological applications employ bias-corrected RCM output [i.e., Chen et al., 2011; Teutschbein and Seibert, 2012]. However, this results in physical inconsistencies [e.g., Immerzeel et al.,

LI ET AL. EVALUATE WATER BUDGET IN HIMALAYAN BASIN 4787 Journal of Geophysical Research: Atmospheres 10.1002/2016JD026279

Figure 1. The topography, stream network, and glacier cover of Beas river basin up to Pandoh Dam with seven rain gauges and Thalout discharge station.

2013] due to the fact that the corrected variables are not independent. For instance, although bias- corrected RCM precipitation data will improve the hydrological calibration results, it will no longer be consistent with modeled temperature, radiation, pressure, and winds thus reducing the utility of the full dynamical simulation. More fundamental perhaps is the fact that RCMs and hydrological models often use very different land surface model representations and spatial scales; there can be large differences in the water and energy budgets between modeling systems, which implies that mass and energy budgets are not conserved in uncoupled modeling systems. At time, these discrepancies can be as large as the potential climate change signals being investigated. A coupled implementation of Weather Research and Forecasting Model (WRF, v3.5.1) and the WRF-Hydro hydrological modeling extension package which run together in fully coupled (two way) or uncoupled (one way from atmosphere to land) modes can serve as hydrometeorological prediction system. In a fully coupled mode, the WRF-Hydro architecture provides a means to couple a hydrological model component to atmospheric models and other Earth System modeling architectures [Gochis et al., 2014]. The ofﬂine conﬁguration is preferable to the fully coupled mode for calibration and validation in this study. This is due in part to the very high resolution and 10 yearlong period of the atmosphere-hydrological simulations, the aim of which is to investigate the water budget. Furthermore, the differences between uncoupled and coupled modes in 10 years water budget are likely small compared Table 1. Area-Elevation Curve to other uncertainties and challenges in >Elevation (m) Area (%) this area, i.e., parameter sensitivity and 2000 88 data quality. Also, additional uncertainty 3000 62 exists due to the lack of a glacier module 4000 41 in the current version of WRF-Hydro, 4800 21 something which is currently underdeve- 6000 1 lopment. In this case, using fully coupled

LI ET AL. EVALUATE WATER BUDGET IN HIMALAYAN BASIN 4788 Journal of Geophysical Research: Atmospheres 10.1002/2016JD026279

Figure 2. WRF domains and model topography. (a) Three domains are indicated by black frames. (b) The inner domain and location of Beas river basin (more detail information can be seen in Figure 3).

mode may not able to capture sufficient/accurate feedback from land to atmosphere in such a glacier- fed basin. Therefore, we employ WRF-Hydro, in its offline configuration, over the topographically complex Beas river basin in India. Using the offline configuration allows for a clearer interpretation of the results, identification of uncertainties in the water budget (particularly in the precipitation forcings), and diagnosis of shortcomings and assessment of sensitivity to critical parameters in both the atmospheric (e.g., microphysics) and hydrological (e.g., infiltration, evapotranspiration, and surface/subsurface routing) components. The study area, data, and experimental approaches are intro- duced in the first three sections followed by the sections of results and discussion, including additional sensitivity tests to address known shortcomings. Despite the challenges the study represents an important step on the road toward a fully coupled atmospheric-hydrological modeling over the complex terrain of the Himalayas.

2. Study Area The Beas river basin (Figure 1) upstream of the Pandoh Dam has a drainage area of 5406 km2, of which 780 km2 (14%) is covered permanently by snow or ice [Kumar et al., 2007]. The mainstream Beas River is joined by five major tributaries: the Parvati, Tirthan, Sainj, Sabari Nala, and Bakhli Khad Rivers. Both the Parvati and Sainj Rivers are glacier fed (Figures 1 and 3). The Beas Basin is mountainous with 21% of the area above 4800 m above sea level (asl) (Table 1) and many peaks above 6400 m asl. This is important because little to no long-term data from weather stations exist at elevations over 4800 m asl [Winiger et al., 2005; Maussion et al., 2011; Böhner and Lucarini, 2015]. Even when available, reliable snowfall measurements are scarce as snowfall is generally underestimated by surface Table 2. Design of WRF Experiment rain gauges [Mair et al., 2013; Name Physical Schemes Hirabayashi et al., 2008; Judson and a Doesken, 2000]. These data are CU (1) Kain-Fritsch Scheme important for understanding the MP (3) Simple three-class Scheme MP (8) Thompson Scheme local-regional hydroclimate. Over LS (2) Unified Noah Land Surface Model the focus region of the present PBL (1) Yonsei Univeristy Scheme study, Kumar et al. [2007] estimated LW (1) RRTM Scheme that snow and glacier melt runoff SW (1) Dudhia Scheme contributes around 35% to the Boundary ECMWF ERA-Interim reanalysis data annual total flow in the Beas River a Only for the outer domain measured at the Pandoh Dam.

LI ET AL. EVALUATE WATER BUDGET IN HIMALAYAN BASIN 4789 Journal of Geophysical Research: Atmospheres 10.1002/2016JD026279

Table 3. WRF-Hydro Model Hydrological Routing Physics Processes Equations Physics Options ¼ β ; fi Subsurface routing qi;j T i;j tan i;jwi;j Surface ex ltration from saturated soil columns β < i;j 0 Lateral flow from saturated soil layers fl ¼ α β Surface overland ow qx xh Pounded water in excess of retention depth subject to routing over land flow ¼ ∂h Sfx Sox ∂x (an example for the 2-D diffusive wave overland flow x direction) Channel routing qffiffiffiffiffiffiffiffi 1-D diffusive wave ¼ ∂z ∂z fl Q sign ∂x K ∂x One-way: over ow into channel

= No subsurface losses ¼ Cm 2 3 K n AR No overbank ﬂow

αi zi Base flow Qbasei ¼ Ci e Empirical, exponential storage-discharge model Base flow combined with overland flow

The monthly mean temperature varies between 2 and 20°C [Kumar et al., 2007], and the mean annual precipitation is 1217 mm, and the mean annual runoff is 1195 mm, of which around 70% occurs during the summer monsoon season (July to September) [Kumar et al., 2007; Li et al., 2013a; Yin et al., 2016]. The land uses/covers in Beas Basin (up to Pandoh) mainly consist of forest (40.1%), ice and snow (38.7%), open land (11.2%), and open forest (8.6%). The high-hill and alpine zones are lying between 2000 and 3000 m above mean sea level (amsl), which are prone to soil erosion, and between 3000 and 4000 m amsl with acute erosion because of steep slopes, respectively [Pandey, 2002; Prasad and Roy, 2005; Kahn et al., 2012; Hegdahl et al., 2016]. In Beas Basin, 88% of the area is higher than 2000 m amsl, while 86% of the area has steep slope (> 25°). The study region and the three nested model domains (with spatial resolutions of 27 km, 9 km, and 3 km) are shown in Figure 2. The topography shown in Figure 2 was derived from 2 min grid of the ETOPO2v2 obtained from NOAA’s National Geophysical Data Center (http://www.ngdc.noaa.gov). The outer domain covers 3375 × 3375 km2 and includes parts of the Indian Ocean, the Bay of Bengal, and the entire Himalayan region. The extensive outer domain is selected to capture large-scale processes and to minimize model errors due to lateral boundary interactions with the complex topography [Warner, 2011; Heikkilä et al., 2011]. The medium and small domains cover an area of 900 × 900 km2 and 354 × 354 km2, respectively. Our analysis is focused on the small domain, which covers the Beas River.

3. Data Three gridded global data sets are employed for model validation and comparison. The Tropical Rainfall Measuring Mission 3B42 V7 (TRMM) data set is a daily satellite-based precipitation data set with 0.25° horizontal resolution [Huffman et al., 2007]. It covers the area between 50°S and 50°N. Additionally, a global reanalysis

Figure 3. The topography and stream channels with discharge stations of Beas river basin. (a) Noah land surface model grid in the inner domain (3 km), (b) the subgrid of WRF-Hydro routing processes (300 m), and (c) the stream channels/stream orders of Beas river basin.

LI ET AL. EVALUATE WATER BUDGET IN HIMALAYAN BASIN 4790 Journal of Geophysical Research: Atmospheres 10.1002/2016JD026279

product, the 0.25° ERA-Interim reanalysis data set [Simmons et al., 2007; Dee et al., 2011], and the APHRODITE V1101R2 daily precipitation (0.25°) [Yasutomi et al., 2011] are also employed. All data sets have been applied previously to hydrological analyses [Duethmann et al., 2012; Li et al., 2013a, 2013b; L. Li et al., 2015; Xu et al., 2016]. Seven rain gauge stations and two weather stations (1996–2005) obtained from Bhakra Beas Management Board (BBMB) in India were used to assess performance of the WRF model precipitation. In order to eliminate the regulation impacts from the Pandoh reservoir, the daily discharge of Thalout station, which is just upstream of Pandoh Dam and has a drainage area of approximately 5000 km2, was used for the WRF-Hydro discharge calibration and evaluation. Discharge data were obtained from BBMB in India. In our study the highest elevation rain gauge station is located below 2000 m amsl and the lack of a glacier module in the current version of WRF-Hydro presents challenges for the study area. In this case, we used a conceptual glacier-hydrological model in Beas river basin for supplemental water resource of the glacier wastage, which is added into WRF-Hydro discharge for comparison. More details of simulation experiments and uncertainty analysis will be explored in the results section.

4. Methodology 4.1. WRF Experiment Design The Advanced Research WRF (ARW) modeling system version 3.5.1 was set up with three nested domains (Figure 2). The topography is produced from the U.S. Geological Survey (USGS) with 30 arc sec resolution Digital Elevation Model (DEM) data, while land use (24 categories) and soil category (19 categories) are also derived from the USGS categories. The forcing at the lateral boundaries was obtained from the ERA-Interim reanalysis data set (spatial resolution of 0.75° and temporal resolution of 6-hourly) from the European Center for Medium Range Weather Forecasting [Dee et al., 2011]. The model was set up with 40 vertical levels in all domains, of which the lowest full model level was about 50 m (lowest half-sigma level at 25 m). The model top was at 50 hPa in all domains. In an investigation of the influence of microphysical schemes on precipitation, Cossu and Hocke [2014] show that most of these schemes (including WRF Single-moment 3-class scheme and Thompson), with the excep- tion of Kessler, have no extreme differences, though they differ in amount of water vapor and accumulated precipitation. The choice of microphysical scheme has important consequences for water phase component, hydrometeor distribution, and precipitation. From other studies of mountain precipitation of WRF, we find that the Thompson microphysical scheme performs well [Collier et al., 2013; Maussion et al., 2014; Rasmussen et al., 2011, 2014]. Therefore, considering the high-resolution and 10 yearlong simulation in our study, two experiments with two microphysics schemes were conducted to test the sensitivity of precipitation/runoff to these choices. All other model settings were kept identical. The first simulation, hereafter named WRF-MP3, employs a simple WRF Single-moment 3-class scheme [Hong et al., 2004]. It predicts only three categories of hydrometers: vapor, cloud water/ice, and rain/snow. For temperatures above freezing, it assumes cloud water and rain; for temperatures below freezing, it assumes cloud ice and snow. Though a simple scheme, it is very efficient in its computational cost [Skamarock et al., 2008], which makes it attractive to use for longer simulations. The second, hereafter named WRF-MP8, is the more complex New Thompson Scheme [Thompson et al., 2008], which computes cloud water, rain water, snow, graupel, and ice. It predicts ice and rain number concentrations, which makes it a double-moment ice scheme. Though it is computationally more expensive than the single moment MP3 scheme, MP8 has been shown to accurately reproduce precipitation over highly complex terrain [e.g., Rasmussen et al., 2011, 2014]. The Kain-Fritsch cumulus convection scheme was used over the outer domain, while the convection parameterization is turned off for the inner two domains. The Yonsei Univeristy scheme was used for the planetary boundary layer, the RRTM scheme for longwave radiation, and the Dudhia scheme for shortwave radiation. The model used adaptive time stepping for which the nominal, default time step was around 8 s over the innermost domain (Table 2). Sea surface temperatures were updated every 6 h from the ERA-Interim data set. The simulations were run from 1996 to 2005, with the first 2 years dedicated to spin-up and discarded. Calibration and validation were performed on the 1998–2005 time period. 4.2. WRF-Hydro System WRF-Hydro is a hydrological extension package to the WRF model, which, when run in fully coupled or uncoupled modes, can serve as an integrated hydrometeorological prediction system. In a fully coupled

LI ET AL. EVALUATE WATER BUDGET IN HIMALAYAN BASIN 4791 Journal of Geophysical Research: Atmospheres 10.1002/2016JD026279

Figure 4. Monthly precipitation (1982–2011) from TRMM 3B42, ERA-Interim, and gauge at (a) Manali (1926 m) and (b) Pandoh (899 m) in Beas river basin.

mode, the WRF-Hydro v2.0 architecture provides a means to couple a hydrological model component to atmospheric models and other Earth System modeling architectures [Gochis et al., 2014]. While different options exist in WRF-Hydro for representing land-surface column physics, we utilized the 1-D Noah land surface model (“Noah LSM”)[Mitchell et al., 2004; Ek et al., 2003] principally because the Noah LSM provides land surface model consistency when it is run in both fully coupled mode with WRF or in stand- alone, uncoupled mode. The Noah LSM calculates the vertical fluxes of energy, i.e., sensible and latent heat, net radiation, and moisture (including canopy interception, snowpack accumulation and ablation, infiltration, infiltration-excess, deep percolation, ponded water depth, and soil thermal and moisture states). On each land surface model time step the infiltration excess, ponded water depth, and soil moisture are disaggregated from the 1-D Noah LSM grid (3 km in the present case) and input to a high- resolution routing grid (300 m) by a time step weighted method [Gochis and Chen, 2003] and then passed to the subsurface and overland flow terrain-routing modules. Saturated subsurface overflow routing, surface overland flow routing, channel and lake routing, and base-flow modules are integrated in the WRF-Hydro system. Lakes and reservoirs were not considered in this study because their impact above the Pandoh Dam reservoir is considered to be minimal. Subsurface horizontal routing is calculated in sequence following the Noah LSM execution. The subsurface exfiltration from a saturated soil column experiencing continued moisture convergence is added to the infiltration excess from the land surface model. Overland flow routing is calculated next, using an unsteady, explicit, finite-difference, diffusive wave formulation similar to that of Julien et al. [1995]. The channel routing is calculated by 1-D variable time stepping diffusive wave formulation (Table 3). Both the overland flow and the gridded channel flow are time stepped according to Courant condition constraints that are associated with the 300 m routing grid resolution. Additionally, a conceptual base-flow parameterization is used in WRF-Hydro v2.0 to better attenuate deep soil drainage and release fluxes from the soil column into the gridded channel network. In this study, it is assumed that the groundwater basin has the same mask as the surface water basin, which

Table 4. Mean Annual Rainfall From TRMM, ERA-Interim and Gauge at Manali and Pandoh (Unit: mm/yr) Manali Pandoh

Period 1998–2006 1990–2005 1998–2006 1982–2005

Data TRMM Gauge ERA-Interim Gauge TRMM Gauge ERA-Interim Gauge Annual 832 1111 1503 1500 1168 1319 2360 1978 Bias 279 3 151 382

LI ET AL. EVALUATE WATER BUDGET IN HIMALAYAN BASIN 4792 Journal of Geophysical Research: Atmospheres 10.1002/2016JD026279

Figure 5. Mean annual precipitation (1998–2005) at 3 km resolution from WRF-MP3, WRF-MP8, TRMM, APHRODITE, and Gauge (dot) in Beas river basin.

means that the whole basin has a groundwater reservoir (“bucket”) with a conceptual depth and associated conceptual volume. The input and output variables, more detailed descriptions, and equations can be found in Gochis et al. [2014] and Yucel et al. [2015]. The WRF-Hydro modeling system has been tested in several different cases focusing on different hydrometeorological forecasting and simulation problems [e.g., Gochis et al., 2014; Yucel et al., 2015; Senatore et al., 2015; Arnault et al., 2016]. It simulated flood events with reasonable accuracy in both gauged and ungauged basins in the Black Sea region [Yucel et al., 2015]. Following multiparameter automated calibration, the WRF-Hydro system simulated a full annual cycle of the Crati river basin in southern Italy, achieving Nash-Sutcliffe efficiency of 0.8 [Senatore et al., 2015]. Use of the uncoupled WRF-Hydro modeling system (i.e., not coupled to the WRF atmospheric model) allows the user to assess the reliability, skill, and variability of streamflow predictions over a long time period. The reliability of the system over the Himalayan region at such timescales is currently unknown, and the current study will help fill this knowledge gap. In the present study WRF-Hydro was set up to run offline using the WRF atmospheric simulations as input (see previous subsection). The subgrid disaggregation/aggregation routing processes are executed on a 300 m resolution, which means that the integer divisor (AGGFACTR) is 10 in the study (Figure 3). The surface physiographic files are prepared by WRF and ARCGIS 10.0 [Gochis et al., 2014], including high-resolution terrain grids, which specify the topography, channel grid, flow direction, and stream order (for channel routing), which has five stream

Figure 6. Seasonal precipitation (1998–2005) of JAS (monsoon) and DJFM (winter) from WRF-MP3, WRF-MP8, TRMM, APHRODITE, and Gauge (dot) in Beas river basin.

LI ET AL. EVALUATE WATER BUDGET IN HIMALAYAN BASIN 4793 Journal of Geophysical Research: Atmospheres 10.1002/2016JD026279

Figure 7. Areal mean precipitation in Beas river basin from WRF-MP3, WRF-MP8, TRMM, APHRODITE, and gauge (1998–2005): (a) Monthly precipitation time series. (b) Monthly mean precipitation with 25% and 75% quantile values (error bar).

orders in the network of the study basin, lakes (optional), groundwater basin mask (optional), and the point of stream gauging stations which are necessary to route water across the landscape.

4.3. Glacier-Hydrological Modelling A conceptual glacier-hydrological model GSM-WAMOD [Xu, 2002; Widen-Nilsson et al., 2009; Gong et al., 2011; Li et al., 2013a, 2013b; L. Li et al., 2015] was applied to investigate how much glacier wastage may impact the ﬁnal headwater budget. The results of the simulation of glacier wastage are added in section 5.5.1. The daily gridded precipitation and temperature of WRF-MP8 are used to drive GSM-WAMOD for model calibration during 1998–2003 (spin-up from 1996 to 1997) and validation from 2004 to 2005. Calibration was performed by searching for an “optimal” parameter set at the Thalout discharge station (1996–2005), which was obtained from BBMB. In total 5000 parameter sets were obtained by Latin- Hypercube sampling [McKay et al., 1979] with prior uniform distribution from initial parameter-value ranges [Gong et al., 2011; Li et al., 2013b]. The 5000 runoff time series were obtained from the GSM-WASMOD runs with the previously deﬁned 5000 parameter-value sets. Model parameters were then calibrated by observed

Table 5. Seasonal and Annual Precipitation of Seven Rain Gauges (1998–2005) Precipitation (mm) Banjar Bhuntar Janjehali Larji Manali Pandoh Sainj

Gauge Elevation (m) 1000 1080 1784 995 1926 899 1384 Gauge Summer 469 345 651 506 580 878 469 Winter 283 309 300 283 225 146 283 Annual 1044 922 1326 1082 1176 1327 1044 WRF Elevation (m) 1502 1496 2105 1465 2849 1311 2075 WRF-MP8 Summer 494 376 695 720 195 714 494 Winter 179 259 239 198 631 159 179 Annual 773 709 1055 1031 979 960 773 WRF-MP3 Summer 175 107 292 366 66 269 175 Winter 135 230 250 153 709 125 135 Annual 379 392 631 593 923 438 379

LI ET AL. EVALUATE WATER BUDGET IN HIMALAYAN BASIN 4794 Journal of Geophysical Research: Atmospheres 10.1002/2016JD026279

Table 6. Spearman’s Rank and Pearson Correlation Coefﬁcients of WRF and Gauge Period Model Spearman’s Pearson

1998–2005 WRF-MP3 0.59 0.36 1998–2005 WRF-MP8 0.73 0.64 1998–2001 WRF-MP3 0.60 0.46 1998–2001 WRF-MP8 0.75 0.75 2002–2005 WRF-MP3 0.59 0.34 2002–2005 WRF-MP8 0.74 0.54

discharge (at Thalout station) using efﬁciency criteria, including Nash-Sutcliffe (NS) coefﬁcient [Nash and Sutcliffe, 1970], bias, and root-mean-square error (RMSE).

5. Results 5.1. Observation Intercomparison There are challenges in evaluating precipitation estimates in data sparse mountain regions and also uncertainty when comparing precipitation estimates with different spatial scales [Frei and Schär, 1998; Duethmann et al., 2013; Ji and Kang, 2015; Ji et al., 2015]. In our study basin, there are seven rainfall gauge stations with one highest gauge of 1926 m at Manali and one lowest gauge of 899 m at Pandoh. All of the rainfall gauges except Manali, which is located at the northern part of the basin, are clustered in the southwest corner of the basin. In this study, the Pandoh station, which has the longest record and lowest altitude, and Manali station, which has the highest altitude, are chosen for the intercomparison. This is in order to identify the baseline data quality of ERA-Interim, which provides the forcing data of WRF-Hydro modeling system. The TRMM 3B42 (0.25°) was also chosen to compare, and a nearest neighbor mapping approach was used to match precipitation product grid cells with station data. Monthly precipitation from TRMM 3B42, ERA-Interim, and gauge data at those two rainfall stations is compared in Figure 4. It shows that the ERA-Interim precipitation captures the variability at Manali and Pandoh stations quite well compared with TRMM and the in situ gauges but overestimates the magnitude in some months. The mean annual precipitation biases of TRMM at Manali and Pandoh compared to gauge measurements are 279 mm/yr and 151 mm/yr, respectively; ERA-Interim biases at Manali and Pandoh compared to gauge are 3 mm/yr and 382 mm/ yr, respectively (Table 4). From the comparison result, ERA-Interim performs fairly well as chosen for forcing data of WRF-Hydro modeling simulation. However, ERA-Interim, TRMM, and gauge data disagree in monthly precipitation at Manali.

5.2. Validation of High-Resolution Precipitation From WRF To evaluate the high-resolution precipitation from WRF, the seven rain gauges are used for validation. TRMM 3B42 (0.25°), which merges TRMM and other satellite estimates, is also used as complementary validation data set. Since we only have seven rain gauge stations in the study basin, especially lacking measurements at high altitudes (i.e., > 2000 m), APHRODITE is chosen as a complimentary “ground truth” data set, instead of the coarse gridded ERA-Interim. APHRODITE V1101R2 that gridded daily precipitation data set (0.25°) [Yatagai et al., 2012; Yasutomi et al., 2011] is constructed by interpolating rain gauge observations from meteorological and hydrological stations over Asia.

Table 7. Validation of High-Resolution Precipitation of WRF by Gauge and TRMM in Beas River Basin (1998–2005) Reference WRF Scores Manali Pandoh

Gauge WRF-MP3 Bias 42.70 61.90 Gauge WRF-MP8 Bias 33.74 23.67 Gauge WRF-MP3 RMSE 161.98 153.87 Gauge WRF-MP8 RMSE 146.34 105.08 TRMM WRF-MP3 Bias 17.75 43.03 TRMM WRF-MP8 Bias 8.79 4.81 TRMM WRF-MP3 RMSE 104.29 94.25 TRMM WRF-MP8 RMSE 88.23 91.97

LI ET AL. EVALUATE WATER BUDGET IN HIMALAYAN BASIN 4795 Journal of Geophysical Research: Atmospheres 10.1002/2016JD026279

Table 8. The Characteristics of Dry Spells From WRF and Gauge at Seven Gauge Stations (1998–2005) Index (Unit:Day) Banjar Bhuntar Janjehali Larji Manali Pandoh Sainj

The mean length of dry spell WRF-MP3 17 19 15 16 17 20 15 WRF-MP8 12 13 13 12 12 14 13 Gauge 7 7 8 7 8 7 8 The longest consecutive dry days WRF-MP3 149 149 102 149 100 149 148 WRF-MP8 136 145 118 118 132 118 136 Gauge 93 103 117 92 124 103 125

The mean annual precipitation from 1998 to 2005 from the two uncoupled WRF simulations (WRF-MP3 and WRF-MP8, 3 km), TRMM (0.25°), APHRODITE, and gauge rainfall are shown in Figure 5. From the figure, the orographic influence on precipitation is clear at the higher-spatial resolution of the 3 km WRF simulations. The precipitation from WRF-MP3 is about 400 mm/yr in the valleys and reaches over 1500 mm/yr on the mountains. A similar pattern and range can be seen in the WRF-MP8 (Thompson Scheme). The glaciated northeastern region exhibits high annual precipitation in the WRF simulations, but comparison with observations is complicated by the lack of gauge data in this region (see dots in Figure 5 for station locations). The coarse resolution of TRMM and APHRODITE results in patterns that show precipitation increasing toward the southwest part of the domain, corresponding with areas of generally lower elevation. Comparing the WRF model output and other precipitation estimate products to gauges in Figure 5 indicates that the WRF model estimates from WRF-MP8 are not very different than gauge values. For example, the Manali precipitation value of 1176 mm/yr is reasonably well captured by WRF, while both TRMM and APHRODITE significantly underestimate precipitation at Manali. In general, the annual precipitation from MP8 is somewhat larger than MP3, leaving MP3 with an apparent low bias in annual precipitation. Besides, the main difference of MP8 and MP3 precipitation is in valley area. Additional differences between WRF, TRMM, APHRODITE, and gauge data can be seen in Figures 6 and 7. Figure 6 shows the seasonal precipitation (mm) of WRF-MP3, WRF-MP8, TRMM, APHRODITE, and gauge data for the winter (December, January, February, and March (DJFM)) and monsoon seasons (July, August, and September (JAS)). Both WRF simulations capture the dry-season/wet-season patterns well, with precipitation largely falling over the mountains during winter and the valleys/lower altitudes in the summer. There is a rainfall peak over Pandoh in the summer monsoon period, while most precipitation is located over high-altitude areas in winter. The precipitation peaks in the WRF simulations move from higher to lower altitudes with the seasonal change from winter to summer. While there is reasonably good correspondence between gauge data and the WRF model simulations during the JAS summer monsoon period, differences occur during the winter months, especially at higher altitudes (i.e., Manali) (Table 5). One potential cause for this discre- pancy between the model and the observation could be the fact that winter precipitation in this region largely falls as snow and the precipitation gauges used in this study are not adequately designed to measure snowfall [c.f. Rasmussen et al., 2010]. WRF-simulated snowfall compared with snowfall measurements in the western U.S. has been shown to have significant skill [e.g., Rasmussen et al., 2011, 2014], but it is acknowl- edged that western U.S. and the Himalaya possess myriad differences in climate and physiographic conditions. It should also be noted that the lack of in situ measurements at high altitudes complicates evaluation of the high precipitation amounts shown in the WRF simulations in winter. Neither APHRODITE nor TRMM shows large amounts of wintertime precipitation at high altitudes, but both are compromised by lack

Figure 8. Monthly temperature from WRF-MP3, WRF-MP8, and observations at (a) Manali and (b) Pandoh in Beas river basin (1998–2005).

LI ET AL. EVALUATE WATER BUDGET IN HIMALAYAN BASIN 4796 Journal of Geophysical Research: Atmospheres 10.1002/2016JD026279

Table 9. Discharge Calibration and Validation of WRF-Hydro With MP8 (1998–2005, Spin-Up: 96–97, Unit: mm/yr)a Calibration (1998–2003) Validation (2004–2005)

Simulations RMSE BIAS NS RMSE BIAS NS

Default 178.3 0.65 0.022 197.9 0.68 0.11 Calibrated 69 0.54 0.32 81 0.56 0.27 Calibration with GlaW 61.3 0.42 0.58 78.5 0.46 0.46 Calibration with NoInf 57.7 0.20 0.63 70.3 0.27 0.57 Calibration with NoInf and GlaW 50.1 0.09 0.72 59.3 0.18 0.69 aNS stands for monthly Nash-Sutcliff efﬁcient, GlaW stands for glacier wastage, and NoInf stands for no inﬁltration.

of coverage, resolution, and skill (see discussion). Further, there is some support in the literature suggesting that the winter loading over the mountains in the WRF experiments is credible [Maussion et al., 2011]. More details are in the later discussions section. In Figure 7a, the areal mean monthly precipitation time series from the WRF experiments are compared with rain gauge, TRMM 3B42, and APHRODITE. The averaged values are used for estimating areal mean precipitation. The figure shows that WRF-MP8 overestimates the peak precipitation in the monsoon season, while WRF-MP3 underestimates it compared with the other three data sets. In general, WRF-MP8 appears to reproduce monthly variability more accurately than WRF-MP3 prior to 2002. After 2002 both WRF runs appear to overestimate precipitation peaks compared with the rain gauge data. The Spearman’s rank correlation coefficients and Pearson correlation coefficients between WRF and the Gauge rainfall data were calculated and are shown in Table 6. P values were also calculated and indicated that all correlations are significant at the 5% level. WRF-MP8 exhibits a stronger relationship to the gauge data than WRF-MP3 with coefficients of 0.73(0.64) versus 0.59(0.36) for the Spearman’s(Pearson’s) tests. The fact that the Spearman’s coefficients are all larger than their Pearson counterparts suggests that the relationship between the modeled and observed precipitation is not linear. Monthly mean precipitation totals are shown in Figure 7b, from which we can see that (i) in summer (JAS), the precipitation from WRF-MP8 is in closer agreement with gauge rainfall than WRF-MP3; (ii) in winter (DJFM), precipitation from both WRF-MP8 and WRF-MP3 is overestimated compared with observations, TRMM and APHRODITE; (iii) in the premonsoon season (April, May, and June), both WRF-MP8 and WRF-MP3 underestimate precipitation; and (iv) there is generally good agreement in the postmonsoon season between both the WRF simulations and gauge data. Whether or not the winter overestimation is due to poor sampling, lack of snowfall observations or a true positive bias in the WRF model simulations is unclear. Quantitative validation of WRF precipitation is shown in Table 7, which provides the bias and root-mean-square error (RMSE) of WRF compared with gauge and TRMM at Manali and Pandoh. Table 7 clearly shows that WRF-MP8 has closer agreement to both observed precipitation and the TRMM data set in terms of bias and RMSE. In summary, multiple lines of evidence point toward a more realistic representation of precipitation patterns and amounts in the WRF-MP8 simulations. Given the requirements of the hydrological modeling components of the WRF-Hydro system, this added precision justifies the additional computational cost.

5.3. Analysis of Dry Spells and Temperature Dry spell characteristics from WRF-MP3 and WRF-MP8 are analyzed using seven gauge data sets (1998–2005). The precipitation of the WRF grid, which is nearest to the rainfall gauge station, is chosen for comparison with gauge data. Precipitation threshold of 1 mm/d is used to deﬁne a dry day (dry spells) in the study [Isotta et al., 2014]. The mean length of dry spell and the longest length of dry spell (which is the maximum number of

Table 10. Mean Annual Water Balance of Beas River Basin up to Thalout by WRF-Hydro With MP8 (1998–2005; Spin-Up: 96–97; Unit: mm/yr) Simulations Precipitation Snowmelt Dischargea ET Soil Water Water Equivalent Snow Depth

Default 1255 480 595 488 -63 308 Calibrated 1255 578 906 264 50 15 Calibrated + NoInf 1255 556 1124 101 22 5 aObserved discharge is 1219.9 mm/yr, NoInf stands for no inﬁltration.

LI ET AL. EVALUATE WATER BUDGET IN HIMALAYAN BASIN 4797 Journal of Geophysical Research: Atmospheres 10.1002/2016JD026279

Figure 9. The calibrated monthly runoff (1998–2005) from WRF-Hydro routed runoff (WRF-MP3_rout and WRF_MP8_rout) compared with unrouted runoff from WRF-MP3, WRF-MP8, and the observation at Thalout.

consecutive days with precipitation less than 1 mm) are used for comparison [Sushama et al., 2014]. The results are shown in Table 8. From the table, we can see that (i) the mean length of dry spells is around 7–8 days (gauge), 12–14 days (WRF-MP8), and 15–20 days (WRF-MP3); (ii) the longest dry spells from WRF-MP3 and WRF-MP8 are generally longer than those from Gauge data; and (iii) in general, the characteristics of dry spells from WRF-MP8 are closer to gauge estimates than those from WRF-MP3. Furthermore, the monthly temperature (1998–2005) simulated from WRF-MP3, WRF-MP8, and observations at Manali and Pandoh are shown in Figure 8. A temperature lapse rate of 6.4°C/km [Azam et al., 2014b] is used for adjusting the monthly temperature from WRF based on the elevation difference between the grid of WRF and the gauge station. Results indicate reasonable performance of WRFs in capturing both the magnitude and seasonal variability of temperature, although an overestimation of the temperature during June–August is found in WRF simulations. Comparing WRF-MP3 and WRF-MP8, the monthly temperature from latter is closer to the observations.

5.4. Discharge Calibration Manual calibration [Boyle et al., 2000; Yucel et al., 2015] of WRF-Hydro was performed in this study into two steps. The first step selected the most sensitive parameters from a wide range of parameters, which are from three tables of Noah LSM parameters, including soil parameters (i.e., Saturation soil conductivity), vegetation parameters (i.e., optimum transpiration air temperature), and some general parameters (i.e., infiltration parameter in the surface runoff parameterization) and two tables of WRF-Hydro hydrological modules, including channel routing (i.e., Manning roughness coefficients) and groundwater bucket model (i.e., the groundwater bucket model exponent). Furthermore, two spatially distributed parameters were also considered into the calibration, i.e., the surface flow roughness scaling factor and the surface retention depth. Within the first step manual calibration, almost all available parameters are performed in the sensitivity tests. In the second step, eight more sensitive parameters were taken into further tuning in the stepwise calibration [Boyle et al., 2000; Yucel et al., 2015; Senatore et al., 2015], including infiltration factor (REFKDT) and soil evaporation exponent (FXEXP_DATA), number of soil layers reached by roots (NROOT), soil moisture reference value for plant

Figure 10. QQ plot of daily runoff data (i.e., 1998–2005 discharge from WRF-MP3, WRF-MP8, and Thalout station) versus gamma distribution.

LI ET AL. EVALUATE WATER BUDGET IN HIMALAYAN BASIN 4798 Journal of Geophysical Research: Atmospheres 10.1002/2016JD026279

transpiration (SMCREF), saturated hydraulic conductivity (DKSAT), retention depth (RETDEPRT), roughness parameter (OVROUGHRT), and manning routing (MANN). Since the manual calibration of the offline WRF-Hydro is computationally demanding, we calibrated WRF-Hydro based on WRF-MP8 output due to its improved performance compared to Figure 11. The distribution of daily runoff from WRF-MP3 and WRF-MP8 WRF-MP3 (see previous subsection). compared with observed discharge (gauge) (1998–2005). The water balance analysis discussed further below is also based on the WRF-Hydro simulation using WRF-MP8 as input. The calibration and validation results of WRF-Hydro with WRF-MP8, which are with/without glacier wastage and with/without infiltration, can be found in Table 9. The water balance of Beas river basin up to Thalout station is shown in Table 10, which indicates that calibration results in discharge increasing over the default parameter settings but that it is still underestimated compared with observed discharge. During 1998–2005, the mean annual snowmelt is about 577.8 mm/yr accounting for 46% of total precipitation 1255 mm/yr and the total evapotranspiration is about 263.6 mm/yr accounting for 21% of precipitation. Figure 9 shows the monthly discharge from the WRF-Hydro model during 1998–2005. The modeled hydrographs capture monthly variability well but at times significantly underestimate discharge compared with observations from the Thalout station. Comparing the routed runoff from WRF-Hydro and no-routed runoff from LSM, the hydrological routing processes corrected the error peak flow in premonsoon period (i.e., April–May) from LSM runoff. The routed discharge from WRF-Hydro (with MP8) captured the peak flows in some years, i.e., 2000 and 2003, which fits better to the observed discharge than WRF-Hydro (with MP3). Some of the most common and important probability distributions used in hydrology are the normal (fitting to the annual streamflow), lognormal (fitting to the annual streamflow), Gumbel (fitting to extreme values of hydrological variables), Weibull (fitting to extreme values of hydrological variables), and gamma. Daily streamflow data frequently follow gamma distribution [Prékopa and Szántai, 1978; Aksoy, 2000; Muneepeerakul et al., 2010]. Quantile-quantile plots of daily discharge from observation and WRF-Hydro driven by WRF-MP3 and WRF-MP8 output are shown in Figure 10. They indicate that the daily discharge (1998–2005) from observations and WRF-Hydro driven by WRF-MP8 fit a gamma distribution although the latter has heavier tails. Conversely, the daily discharge from WRF-Hydro driven by WRF-MP3 output does not fit the gamma distribution and is heavily skewed. From Figure 11, we can see that the PDF of WRF-Hydro peaks at <1 mm/d while the observed peak is around 1–2 mm/d suggesting that WRF-Hydro overestimates low flow occurrences. Conversely, larger flow events between 5 and 10 mm/d are relatively underestimated by WRF-Hydro compared to observations suggesting that, in addition to possessing a low flow bias, the model is overdampening runoff pulse events. This is partly due to the underestimation of medium intensity precipitation events, which can be inferred from Figure 12. For most of the gauges (except Bhuntar and Manali), the probabilities of precipitation (< 50 mm/d) from WRF-MP8 are lower than that the probabilities from observed rainfalls. More details can be found in Table 11, which provides the cumulative distributions of daily discharge from gauge, WRF-Hydro driven by WRF-MP3, and WRF-Hydro driven by WRF-MP8 for different discharge ranges.

5.5. Sensitivity Experiments 5.5.1. Infiltration Sensitivity The Beas Basin (up to Pandoh) largely consists of bare rock and steep slopes. The area of 0–5, 5–10, 10–15, 15–25, 25–35, 35–45, and >45° are 1.9%, 4.8%, 6.9%, 23.2%, 31%, 22.3%, and 10%, respectively. The area with steep slopes (> 25°) is over 86% of total basin. The infiltration in Beas river basin can be quite small because water falling on steep-sloped land runs off more quickly and infiltrates less than water falling on flat land [Ribolzi et al., 2011]. In order to investigate how much impact of infiltration process on headwater budget of Beas river basin is and analyze the sensitivity of parameter REFKDT, we compared the headwater

LI ET AL. EVALUATE WATER BUDGET IN HIMALAYAN BASIN 4799 Journal of Geophysical Research: Atmospheres 10.1002/2016JD026279

budget when inﬁltration parameter REFKDT is zero with the calibrated result. From Table 10, we can see that for REFKDT to be zero there is an increased discharge of 218 mm/yr and also largely decreased evapotranspiration (ET) from 264 mm/yr to 101 mm/yr. However, the discharge still underestimates compared with the observation, which might be due to water budget of glacier wastage that is missing. 5.5.2. Glacier Wastage 5.5.2.1. Calibration and Validation of GSM-WASMOD In the study basin, 780 km2 (14%) is covered permanently by snow or ice. However, there is lack of glacier module in the present version of WRF- Hydro model system in our study. This will certainly result in the underestimation of discharge. Numerous case studies over the Himalayan river basins estimate the contribution of snow/glacier melt to total runoff [Racoviteanu et al., 2013; Ahluwalia et al., 2013]. The Snowmelt Runoff Model or degree-day factor-based mass balance model, which requires reliable ground information and meteorological data, is widely used in hydrological simulations of Himalayan Basins [e.g., Prasad and Roy, 2005; Shrestha et al., 2012]. Over the focus region of the present study, the dedications of snow/glacier melting to total runoff are different from previous studies varying from 27.5% to 40% [Kumar et al., 2007; Li et al., 2013a; H. Li et al., 2015]. According to the study of Kääb et al. [2015] by ICESat satellite altimetry data, the per- centage of discharge equivalent from annual glacier wastage to river runoff Figure 12. The distributions of daily precipitation from WRF-MP3 and WRF- in Beas Basin over 2003–2008 is about MP8 compared with seven rain gauges (1998–2005). 5%. Furthermore, many studies investigated the glacier mass balance in Chhota Shigri Glacier, which is one of the main glaciers in Beas river basin (close to Bhuntar). Berthier et al. [2007] found that there was an agreement of negative glacier mass balance in Chhota Shigri Glacier, including 1to1.1 m/a w.e. in 1999–2004 for space-based and 1.13 m/a w.e. in 2002–2004 for ﬁeld-based estimation. Wagnon et al. [2007] also found that the glaciological mass balance of Chhota Shigri was 1.06 m/a w.e. in 2002–2003 and 1.2 m/a w.e. in 2003–2004. Recently, it is reported that Chhota Shigri Glacier has a negative mass balance of 4.8±1.8 m/a w.e. in 1999–2010 by Vincent et al. [2013].

LI ET AL. EVALUATE WATER BUDGET IN HIMALAYAN BASIN 4800 Journal of Geophysical Research: Atmospheres 10.1002/2016JD026279

Table 11. Cumulative Distribution of Daily Discharge From Gauge, WRF-MP3 and WRF-MP8 Discharge (mm/d) Gauge WRF-MP3 WRF-MP8

0–1 0.191 0.583 0.483 1–5 0.593 0.303 0.378 5–10 0.182 0.082 0.104 10–15 0.029 0.022 0.026 15–20 0.004 0.007 0.007 >20 0.001 0.003 0.002 Total 1 1 1

In order to investigate how much the glacier wastage may impact on the final headwater budget and related evaluation metrics, we added an experiment simulating glacier wastage using a conceptual glacier- hydrological model GSM-WAMOD [Li et al., 2013a, 2013b; L. Li et al., 2015]. In our study, the average glacier mass balance in Bear River basin is calibrated by the parameters of degree day factors [Li et al., 2013a; Azam et al., 2014b]. It results in 1.01 m/a w.e. of glacier mass balance, which deliver about 146 mm of the annual glacier wastage in Beas river basin during 1998–2005. The results show that 14% of total runoff is from glacier wastage in Beas river basin. The daily NS, bias, and RMSE from calibration are 0.70, 0.11, and 1.64, while those from validation are 0.66, 0.038, and 1.93, respectively. The monthly NS of calibration and validation is 0.79 and 0.75, respectively. 5.5.2.2. Experiment Results We added the glacier wastage into the WRF-Hydro discharge and recalculated the calibration and validation metrics (in Table 9). From Table 9, we can see that the water budget from glacier wastage largely improves the monthly Nash-Sutcliff efficacy from 0.32 to 0.58 and from 0.27 to 0.46 for calibration and validation periods, respectively. Besides, the bias and RMSE both reduced after adding glacier wastage for both calibration and validation periods. Figures 13 and 14 show the monthly (1998–2005) and daily (1998) hydrographs from no routed, routed, and the glacier wastage plus routed runoff of WRF/WRF-Hydro (with MP8) and observed discharge from Thalout station. The figures show that the glacier wastage plus routed runoff from WRF-Hydro (with MP8) fit the observed discharge better at either monthly or daily time steps. Especially, from the daily hydrograph of 1998, we can see that the routed runoff improved the overestimated flow in premonsoon period (March–June), while the added glacier wastage water budget further improved the routed runoff during summer monsoon period (July–September), although it still underestimates the runoff in general and mismatch a few peak flows.

6. Discussion With respect to the monsoon precipitation over the Himalayas, the space-time distribution of precipitation, the spatial variability of the diurnal cycle of convection, and the terrain are all linked from a process perspec- tive from the mesoscale to the synoptic scale [Barros et al., 2004]. At the synoptic scale, the summer monsoon and wintertime moisture transport by the westerly jet stream over Asia are the main sources of precipitation for the Indus Basin where the Beas river basin is located [Mölg et al., 2013; Immerzeel et al., 2015]. The intercomparison of high-resolution precipitation from WRF-MP3/MP8, gauge, TRMM, and APHRODITE shows that WRF-MP3 has trouble in reproducing the spatial pattern of summertime precipitation, while WRF-MP8 agrees

Figure 13. The WRF-Hydro monthly no-routed runoff (WRF-MP8), routed runoff (WRF-MP8_rout), routed runoff added with glacier wastage (WRF-MP8_rout + Glac), and the observed discharge of Thalout during 1998–2005.

LI ET AL. EVALUATE WATER BUDGET IN HIMALAYAN BASIN 4801 Journal of Geophysical Research: Atmospheres 10.1002/2016JD026279

Figure 14. The WRF-Hydro daily no-routed runoff (WRF-MP8) and routed runoff (WRF-MP8_rout), routed runoff added with glacier wastage (WRF-MP8_rout + Glac), and the observed discharge of Thalout in 1998.

well with observations and the gridded data sets. In winter, however, the patterns produced by both WRF-MP3 and WRF-MP8 are markedly different than those suggested by observations and gridded data products. The large disagreement between precipitation patterns in the model simulations and other data sources occurs over high altitudes of the Beas Basin. This disparity may partly arise due to the fact that there are no measurements in this study higher than 2000 m altitude and also due to the fact that the regional precipitation gauges used in this study were not designed to appropriately measure snowfall. Reliable snowfall measurements are scarce over the Himalayan region as snowfall is generally underestimated by surface rain gauges [Mair et al., 2013; Hirabayashi et al., 2008; Judson and Doesken, 2000]. The APHRODITE data, which are based on gauge measurements, also suffer from limited coverage and appear to underestimate precipitation in this region. Satellite remote sensing products such as the TRMM 3B42 V7 product have also been shown to have difficulty measuring snowfall; TRMM is mostly rainfall data and only partially includes snowfall since measuring snowfall with IR remote sensing technology is challenging [Huffman et al., 2007; Ji and Luo, 2013; Viste and Sorteberg, 2015]. These issues complicate the task of evaluating wintertime precipitation of the WRF simulations. More generally, high-altitude precipitation in the Himalaya region is poorly understood and has large uncertainty [Ragettli and Pellicciotti, 2012; Immerzeel et al., 2013, 2015; Viste and Sorteberg, 2015; Ji et al., 2015]. However, limitations related to snowfall measurements are likely not the entire story. The largest discrepancies between the model and observations are seen at the Manali station where WRF underestimates (overestimates) summer (winter) precipitation. Manali is quite a bit deeper into the mountains than the other stations so it is likely to be much more heavily influenced by the complex topography. The influence of this topography is clearly seen in Figures 5 and 6. Even though it is at a similar elevation to other gauges, the shortcomings in measuring snow will be more pronounced here given its location. There is also the fact that the Kain-Fritsch convective scheme is highly sensitive to the representation of low-level forcing fields [Stensrud, 2007], which might affect the underestimation in the summer over Manali, as these fields will be more challenging to reproduce here than over the other locations. In winter the opposite situation is seen and this could be due again to the complex topography in the area surrounding Manali. Too rapid orographic ascent (also a known issue with RCMs) will lead to excessive precipitation at the lower elevations on the mountain shoulders of and upper valleys. Investigating these speculations is beyond the scope of the present paper, but they provide motivation for further study to elucidate the dynamical and thermodynamical causes of these discrepancies. There is some evidence to support the winter precipitation patterns produced by our WRF simulations. Shrestha et al. [2012] reconstructed snowfall at a high elevation basin of the Nepal Himalaya by a snow depth measurement and corrected the precipitation data in order to simulate the snow depth at a 5035 m asl site, multiplying rain gauge data by a factor of 2.5. Winiger et al. [2005] found an increase of the annual precipitation with altitude, and the snowfall contribution exceeded 80% of the total annual precipitation at site up to 4500 m. In the study of Bookhagen and Burbank [2010], they found that two thirds of the precipitation in western Himalaya basins was associated with winter storms and snowmelt contributed 50% of the hydrological budget of western basins. Immerzeel et al. [2015] used glacier mass balance to inversely infer the

LI ET AL. EVALUATE WATER BUDGET IN HIMALAYAN BASIN 4802 Journal of Geophysical Research: Atmospheres 10.1002/2016JD026279

high-altitude precipitation over the upper Indus Basin where the Beas river basin is located. Their findings suggested that the high-altitude precipitation is much larger than previously thought. In our study, the spatial variability from WRF-MP3/MP8 precipitation is larger than other data sets. The highest (gauge or grid cell) precipitation amounts from gauge, TRMM, and APHRODITE in Beas (1998–2005) were 1326 mm/yr, 1136 mm/yr, and 1721 mm/yr, respectively. However, the highest precipitation in Beas from WRF-MP3 and WRF-MP8 reaches to 3251 mm/yr and 4273 mm/yr, respectively. These amounts are in closer agreement with over 3000 mm/yr estimated by Immerzeel et al. [2015] over the 2003–2007 period. The winter precipitation (DJFM) from WRF-MP3 and WRF-MP8 is 594 mm and 541 mm; while that from gauge, TRMM, and APHRODITE, respectively, are 282 mm, 237 mm, and 184 mm over the Beas river basin. Almost all of this precipitation falls at high altitudes (see Figure 6 and cf. Figure 2). There is no available gauge data over 2000 m amsl in the study, so we can only compare the results from WRF, TRMM, and APHRODITE. Over high elevation, glaciated areas (2358–6423 m asl) the average annual precipitation of WRF-MP3 and WRF-MP8 (2358–6423 m asl) is around 1444 and 1413 mm/yr, whereas TRMM and APHRODITE estimate 858 and 524 mm/yr. Further, the average winter precipitation from WRF-MP3 and WRF-MP8 over these areas is about 841 mm and 741 mm, which is almost 300–400% of that estimated by TRMM and APHRODITE (230 and 148 mm, respectively). The WRF-MP3/MP8-simulated high-altitude annual precipitation is comparable with the result from the Immerzeel et al. [2015], which estimates a corrected annual high-altitude (3751–4250 m asl) precipitation of ~1271 mm/yr in the upper Indus Basin, an increase of over 300% over the uncorrected case. It is further suggested that APHRODITE underestimates annual precipitation by as much as 200% over the upper Indus Basin [Immerzeel et al., 2015]. Although the present study lacks the observations to conclu- sively evaluate the accuracy of the modeled wintertime precipitation patterns and amounts, other studies have shown that WRF accurately reproduces winter precipitation over complex terrain, i.e., western U.S. [Rasmussen et al., 2011]. WRF appears very useful to estimate the regional and seasonal distribution of precipitation in the Himalaya basins even if it shows significant biases [Ménégoz et al., 2013]. Taken with the results from Immerzeel et al. [2015], the evidence suggests that the WRF-MP8 precipitation patterns and amounts are perhaps more representative of the full annual cycle of precipitation than the initial analysis would lead one to believe. The calibrated WRF-Hydro model exhibited reasonable performance in capturing the annual cycle of runoff at monthly timescales but systematically underestimates discharge in most years (Figure 9). There are some likely culprits for this systematic bias. One of the main reasons is that the use NOAH land surface model in this configuration of WRF-Hydro has a simplified snow submodel, which is recognized to underestimate snow water equivalent [Sultana et al., 2014]. It lacks the capability to predict the glacier mass balance, which reduces simulated ablation [Collier et al., 2013]. From the sensitivity experiment test by adding the glacier wastage from conceptual glacio-hydrological model (WASMOD), which calculated the glacier mass balance based on a degree-day factor method, not only the daily and monthly runoff better fitted to the observed discharge but also the monthly NS largely improved from 0.32 to 0.58 and 0.27 to 0.46 at calibration and validation periods, respectively. The negative glacier mass balance of 1.01 m/a w.e. during 1998–2005 in Beas river basin is comparable to other studies in Chhota Shigri Glacier [Wagnon et al., 2007; Azam et al., 2012; Pratap et al., 2016]. However, there is a high uncertainty for the mass balance in these field glaciers due to a lack of observational coverage of the accumulation area [Berthier et al., 2007; Azam et al., 2014a, 2014b]. There are also uncertainties due to the lack of an energy balance verification in GSM-WASMOD. Another experiment of infiltration parameter sensitivity given to us also provides insight into the extreme case of all the precipitation runoff as overland flow. The results of noninfiltration experiment show the increase of monthly NS and the decrease of monthly bias and RMSE, which inferred that the steep-sloped land in the upper Beas river basin could result in quite small infiltration. Furthermore, the sublimation from NOAH land surface model is highly overestimated, which is also a common problem [Mitchell et al., 2004; Wang et al., 2010]. This physics-based glacier module coupling will be available in the future WRF-Hydro model development but cannot be applied at present study.

7. Conclusions A hydrometeorological regional modeling system (WRF-Hydro) was implemented and analyzed over a 10 year simulation period for a mountainous, topographically complex basin over the Himalayan region of

LI ET AL. EVALUATE WATER BUDGET IN HIMALAYAN BASIN 4803 Journal of Geophysical Research: Atmospheres 10.1002/2016JD026279

northwest India. A triple nest of the WRF atmospheric model was used to provide precipitation forcing experiments over a small domain encompassing the Beas river basin with a 3 km WRF atmospheric and Noah land model grid spacing and a 300 m WRF-Hydro routing grid spacing. The high-resolution precipitation results from two microphysical parameterization (MP) schemes (i.e., single-moment 3-class scheme and the Thompson Scheme) were assessed and discussed. Although the in situ observation network is sparse and has a low elevation bias, the Thompson Scheme appears to give better precipitation estimates than the single-moment three-class scheme in both temporal and spatial variability compared with in situ measurements. A significant disagreement in high-altitude precipitation estimates is found between gauge observations, TRMM, APHRODITE, and the WRF model experiments. Although the explanation for these differences is hard to evaluate since there is no measurement over 2000 m amsl, our results show that the spatial distribution of annual precipitation in the Beas Basin from WRF varies much more than that from gauge, TRMM, or APHRODITE data set. The high-altitude precipitation between 2358 and 6423 m amsl from WRF is about twice or triple size as that from the gridded data, i.e., TRMM and APHRODITES. The precipitation from our WRF simulations confirms the results of the study from Immerzeel et al. [2015] and suggests that a large amount of precipitation in high-altitude mountainous areas in the Beas river basin is occurring and is not properly accounted for in the TRMM or APHRODITE products analyzed in this work. The WRF-Hydro modeling system shows skill in capturing monthly discharge variability and the daily discharge distribution. The simulated WRF-Hydro (with MP8) daily discharge agrees well with the gamma distribution from observed discharge at Thalout station although the probability of low flows (0–1 mm) is more than what is observed while the probability of larger flow events appeared to be underestimated from WRF-Hydro. Somehow, the daily discharge simulation is still a challenge, because of the uncertainty from long-term time period precipitation, and underestimates in glacier ablation, which was proved by the sensitivity experiment of adding glacier wastage water from a conceptual glacio-hydrological model. Also, the steep-sloped land of Beas river basin contributes to less infiltration than originally thought. But this needs further investigation since other meteorological variables such as temperature, humidity, wind, and solar radiation also need to be better validated in order to constrain the possibilities for the WRF-Hydro model bias. However, when forced with reasonable meteorological forcings, the WRF-Hydro model system shows the prediction skill in capturing the spatial and temporal structure of high-resolution precipitation and has potential capability of predicting stream flow hydrographs, especially in sparsely gauged basins, such as the Himalayas.

Acknowledgments References This study was joint funded by the Research Council of Norway (RCN) pro- Adjei, K. A. A., L. L. Ren, E. M. Appiah-Adjei, and S. H. Oddi (2016), Application of satellite-derived rainfall for hydrological modelling in the – ject 216576 (NORINDIA), project- data-scarce Black Volta trans-boundary basin, Hydrol. Res., 46(5), 777 791. JOINTINDNOR 203867, and project Ahluwalia, R. S., S. P. Rai, S. K. Jain, B. Kumar, and D. P. Dobhal (2013), Assessment of snowmelt runoff modelling and isotope analysis: A case – GLACINDIA 033 L164. We thank the study from the western Himalaya, India, Ann. Glaciol., 54(62), 299 304(6). Bhakra Beas Management Board for Aksoy, H. (2000), Use of gamma distribution in hydrological analysis, Turk. J. Eng. Environ. Sci., 24(6), 419–428. providing the data of the Beas Basin. Arnault, J., S. Wagner, T. Rummler, B. Fersch, J. Bliefernicht, S. Andresen, and H. Kunstmann (2016), Role of runoff-infiltration partitioning and Many thanks to Roy Rasmussen (NCAR) resolved overland flow on land-atmosphere feedbacks: A case-study with the WRF-Hydro coupled modeling system for West Africa, for his suggestions in early WRF domain J. Hydrometeorol., 17, 1489–1516, doi:10.1175/JHM-D-15-0089.1. design and figure improvements, Wei Arritt, R. W., and M. Rummukainen (2011), Challenges in regional-scale climate modeling, Bull. Am. Meteorol. Soc., 92(3), 365–368. Yu (NCAR) for assistance with the WRF- Azam, M. F., et al. (2012), From balance to imbalance: A shift in the dynamic behaviour of Chhota Shigri glacier, western Himalaya, India, Hydro model compilation, and Chong- J. Glaciol., 58(208), 315–324. yu Xu for his suggestions for improving Azam, M. F., P. Wagnon, C. Vincent, A. L. Ramanathan, V. Favier, A. Mandal, and J. G. Pottakkal (2014a), Processes governing the mass balance the manuscript. Also, we thank two of Chhota Shigri Glacier (western Himalaya, India) assessed by point-scale surface energy balance measurements, The Cryosphere, 8(6), anonymous reviewers for useful com- 2195–2217. ments. Part of simulation data are Azam, M. F., P. Wagnon, C. Vincent, A. Ramanathan, A. Linda, and V. B. Singh (2014b), Reconstruction of the annual mass balance of Chhota stored in Norwegian data infrastructure Shigri glacier, Western Himalaya, India, since 1969, Ann. Glaciol., 55(66), 69–80. Norstore, which are downloadable Barros, A. P., G. Kim, E. Williams, and S. W. Nesbitt (2004), Probing orographic controls in the Himalayas during the monsoon using satellite (http://ns9001k.norstore.uio.no). It is imagery, Nat. Hazards Earth Syst. Sci., 4,29–51, doi:10.5194/nhess-4-29-2004. also available from the authors upon Barstad, I., and G. N. Caroletti (2013), Orographic precipitation across an island in southern Norway: Model evaluation of time-step precipi- request ([email protected]). tation, Q. J. R. Meteorol. Soc., 139(675), 1555–1565. Berthier, E., Y. Arnaud, R. Kumar, S. Ahmad, P. Wagnon, and P. Chevallier (2007), Remote sensing estimates of glacier mass balances in the Himachal Pradesh (Western Himalaya, India), Remote Sens. Environ., 108(3), 327–338. Biskop, S., P. Krause, J. Helmschrot, M. Fink, and W. A. Flügel (2012), Assessment of data uncertainty and plausibility over the Nam Co Region, Tibet, Adv. Geosci., 31,57–65. Bookhagen, B., and D. W. Burbank (2010), Toward a complete Himalayan hydrological budget: Spatiotemporal distribution of snowmelt and rainfall and their impact on river discharge, J. Geophys. Res., 115, F03019, doi:10.1029/2009JF001426.

LI ET AL. EVALUATE WATER BUDGET IN HIMALAYAN BASIN 4804 Journal of Geophysical Research: Atmospheres 10.1002/2016JD026279

Böhner, J., and V. Lucarini (2015), Prevailing climatic trends and runoff response from Hindukush-Karakoram-Himalaya, upper Indus basin. arXiv preprint arXiv:1503.06708. Bolch, T., et al. (2012), The state and fate of Himalayan glaciers, Science, 336, 310–314. Boyle, D. P., H. V. Gupta, and S. Sorooshian (2000), Toward improved calibration of hydrologic models: Combining the strengths of manual and automatic methods, Water Resour. Res., 36(12), 3663–3674. Chen, J., F. P. Brissette, A. Poulin, and R. Leconte (2011), Overall uncertainty study of the hydrological impacts of climate change for a Canadian watershed, Water Resour. Res., 47, W12509, doi:10.1029/2011WR010602. Christensen, J. H., F. Boberg, O. B. Christensen, and P. Lucas-Picher (2008), On the need for bias correction of regional climate change projections of temperature and precipitation, Geophys. Res. Lett., 35, L20709, doi:10.1029/2008GL035694. Christensen, J. H., et al. (2013), Climate phenomena and their relevance for future regional climate change, in Climate Change 2013: The Physical Science Basis. Contribution of Working Group I to the Fifth Assessment Report of the Intergovernmental Panel on Climate Change, edited by T. F. Stocker et al., pp. 1229–1231, Cambridge Univ. Press, Cambridge, U. K., and New York. Collier, E., T. Mölg, F. Maussion, D. Scherer, C. Mayer, and A. B. G. Bush (2013), High-resolution interactive modelling of the mountain glacier– atmosphere interface: An application over the Karakoram, Cryosphere Discuss., 7(1), 103–144. Cossu, F., and K. Hocke (2014), Influence of microphysical schemes on atmospheric water in the weather research and forecasting model, Geosci. Model Dev., 7(1), 147–160. Dee, D. P., et al. (2011), The ERA-Interim reanalysis: Configuration and performance of the data assimilation system, Q. J. R. Meteorol. Soc., 137(656), 553–597. Dimri, A. P., T. Yasunari, A. Wiltshire, P. Kumar, C. Mathison, J. Ridley, and D. Jacob (2013), Application of regional climate models to the Indian winter monsoon over the western Himalayas, Sci. Total Environ., 468-469, S36–S47, doi:10.1016/j.scitotenv.2013.01.040. Duethmann, D., J. Zimmer, A. Gafurov, A. Güntner, B. Merz, and S. Vorogushyn (2012), Evaluation of areal precipitation estimates based on downscaled reanalysis and station data by hydrological modelling, Hydrol. Earth Syst. Sci. Discuss., 9, 10,719–10,773. Duethmann, D., J. Zimmer, A. Gafurov, A. Güntner, D. Kriegel, B. Merz, and S. Vorogushyn (2013), Evaluation of areal precipitation estimates based on downscaled reanalysis and station data by hydrological modelling, Hydrol. Earth Syst. Sci., 17(7), 2415–2434. Ek, M. B., K. E. Mitchell, Y. Lin, E. Rogers, P. Grunmann, V. Koren, G. Gayno, and J. D. Tarpley (2003), Implementation of Noah land surface model advances in the National Centers for Environmental Prediction operational mesoscale Eta model, J. Geophys. Res., 108(D22), 8851, doi:10.1029/2002JD003296. Frei, C., and C. Schär (1998), A precipitation climatology of the Alps from high-resolution rain-gauge observations, Int. J. Climatol., 18(8), 873–900. Fyfe, J. C., and G. M. Flato (1999), Enhanced climate change and its detection over the Rocky Mountains, J. Clim., 12, 230–243, doi:10.1175/ 1520-0442-12.1.230. Gochis, D. J., and F. Chen (2003), Hydrological enhancements to the Community Noah Land Surface Model: Technical Description. NCAR Science and Technical Note, TN454+STR, Sept., 2003. Gochis, D. J., W. Yu, and D.N. Yates (2014), The WRF-Hydro model technical description and user’s guide, version 2.0. NCAR technical document. 120 pages. Available at: WRF-Hydro 2.0 User Guide. Gong, L., S. Halldin, and C. Y. Xu (2011), Global scale river routing—An efficient time delay algorithm based on HydroSHEDS high resolution hydrography, Hydrol. Processes, 25(7), 1114–1128. Graham, L., J. Andréasson, and B. Carlsson (2007), Assessing climate change impacts on hydrology from an ensemble of regional climate models, model scales and linking methods—A case study on the Lule River basin, Clim. Change, 81,293–307, doi:10.1007/s10584-006-9215-2. Hartmann, H., and L. Andresky (2013), Flooding in the Indus River basin—A spatiotemporal analysis of precipitation records, Global Planet. Change, 107,25–35, doi:10.1016/j.gloplacha.2013.04.002. Hegdahl, T. J., L. M. Tallaksen, K. Engeland, J. F. Burkhart, and C.-Y. Xu (2016), Discharge sensitivity to snowmelt parameterization: A case study for Upper Beas basin in Himachal Pradesh, India, Hydrol. Res., 47(4), 683–700, doi:10.2166/nh.2016.047. Heikkilä, U., A. Sandvik, and A. Sorteberg (2011), Dynamical downscaling of ERA-40 in complex terrain using the WRF regional climate model, Clim. Dyn., 37(7–8), 1551–1564. Hewitt, K. (2005), The Karakoram anomaly? Glacier expansion and the ‘elevation effect’, Karakoram Himalaya, Mt. Res. Dev., 25, 332–340. Hirabayashi, Y., S. Kanae, K. Motoya, K. Masuda, and P. Döll (2008), A 59-year (1948-2006) global meteorological forcing data set for land surface models. Part II: Global snowfall estimation, Hydrol. Res. Lett., 2,65–69. Hong, S.-Y., J. Dudhia, and S.-H. Chen (2004), A revised approach to ice microphysical processes for the bulk parameterization of clouds and precipitation, Mon. Weather Rev., 132, 103–120. Huffman, G. J., R. F. Adler, D. T. Bolvin, G. Gu, E. J. Nelkin, K. P. Bowman, Y. Hong, E. F. Stocker, and D. B. Wolff (2007), The TRMM multi-satellite precipitation analysis: Quasi-global, multi-year, combined-sensor precipitation estimates at fine scale, J. Hydrometeorol., 8(1), 38–55. Immerzeel, W., F. Pellicciotti, and M. Bierkens (2013), Rising river flows throughout the twenty-first century in two Himalayan glacierized watersheds, Nat. Geosci., 6, 742–745, doi:10.1038/NGEO1896. Immerzeel, W. W., N. Wanders, A. F. Lutz, J. M. Shea, and M. F. P. Bierkens (2015), Reconciling high altitude precipitation in the upper Indus Basin with glacier mass balances and runoff, Hydrol. Earth Syst. Sci. Discuss., 12(5), 4755–4784. Isotta, F. A., et al. (2014), The climate of daily precipitation in the Alps: Development and analysis of a high-resolution grid dataset from pan- Alpine rain-gauge data, Int. J. Climatol., 34, 1657–1675, doi:10.1002/joc.3794. Ji, X., and Y. Luo (2013), The influence of precipitation and temperature input schemes on hydrological simulations of a snow and glacier melt dominated basin in Northwest China, Hydrol. Earth Syst. Sci. Discuss., 10(1), 807–853. Ji, Z. M., and S. C. Kang (2013a), Double-nested dynamical downscaling experiments over the Tibetan Plateau and their projection of climate change under two RCP scenarios, J. Atmos. Sci., 70(4), 1278–1290. Ji, Z. M., and S. C. Kang (2013b), Projection of snow cover changes over China under RCP scenarios, Clim. Dyn., 41, 589–600. Ji, Z. M., and S. C. Kang (2015), Evaluation of extreme climate events using a regional climate model for China, Int. J. Climatol., 35(6), 888–902. Ji, Z., S. Kang, Z. Cong, Q. Zhang, and T. Yao (2015), Simulation of carbonaceous aerosols over the Third Pole and adjacent regions: Distribution, transportation, deposition, and climatic effects, Clim. Dyn., 45(9–10), 2831–2846. Judson, A., and N. Doesken (2000), Density of freshly fallen snow in the central Rocky Mountains, Bull. Am. Meteorol. Soc., 81, 1577–1587, doi:10.1175/1520-0477(2000)081<1577:DOFFSI>2.3.CO;2. Julien, P., B. Saghafian, and F. Ogden (1995), Raster-based hydrological modeling of spatially-varied surface runoff, Water Resour. Bull., 31(3), 523–536. Kääb, A., D. Treichler, C. Nuth, and E. Berthier (2015), Brief communication: Contending estimates of 2003–2008 glacier mass balance over the Pamir–Karakoram–Himalaya, The Cryosphere, 9(2), 557–564.

LI ET AL. EVALUATE WATER BUDGET IN HIMALAYAN BASIN 4805 Journal of Geophysical Research: Atmospheres 10.1002/2016JD026279

Kahn, A. A., S. S. Randhawa, and D. C. Rana (2012), Training needs assessment of stakeholders in disaster management in the state of Himachal Pradesh. State Councile for Science Technology & Environment, Disaster Management Cell Govt of HP, National Insistue of Disaster Management Govt India. Kaser, G., M. Großhauser, and B. Marzeion (2010), Contribution potential of glaciers to water availability in different climate regimes, Proc. Natl. Acad. Sci. U.S.A., 107, 20,223–20,227, doi:10.1073/pnas.1008162107. Krishnamurthy, C. K. B., U. Lall, and H. H. Kwon (2009), Changing frequency and intensity of rainfall extremes over India from 1951 to 2003, J. Clim., 22, 4737–4746. Kumar, V., et al. (2007), Snow and glacier melt contribution in the Beas River at Pandoh Dam, Himachal Pradesh, India, Hydrol. Sci. J.-Des Sci. Hydrol., 52(2), 376–388. Leung, L. R., and Y. Qian (2003), The sensitivity of precipitation and snowpack simulations to model resolution via nesting in regions of complex terrain, J. Hydrometeorol., 4, 1025–1043, doi:10.1175/1525-7541. Li, H., S. Beldring, C.-Y. Xu, M. Huss, and K. Melvold (2015), Integrating a glacier retreat model into a hydrological model—Case studies on three glacierised catchments in Norway and Himalayan region, J. Hydrol., 527, 656–667, doi:10.1016/j.jhydrol.2015.05.017. Li, L., M. Engelhard, C. Y. Xu, S. J. Jain, and V. P. Singh (2013a), Comparison of satellite-based and reanalysed precipitation as input to glacio- hydrological modeling for Beas river basin, Northern India. Cold and Mountain Region Hydrological Systems Under Climate Change: Towards Improved Projections. IAHS Publ. 360. 45–52. Li, L., C. S. Ngongondo, C. Y. Xu, and L. Gong (2013b), Comparison of the global TRMM and WFD precipitation datasets in driving a large-scale hydrological model in Southern Africa, Hydrol. Res., 10, 2166. Li, L., I. Diallo, C.-Y. Xu, and F. Stordal (2015), Hydrological projections under climate change in the near future by RegCM4 in Southern Africa using a large-scale hydrological model, J. Hydrol., 528,1–16, doi:10.1016/j.jhydrol.2015.05.028. Mair, E., G. Bertoldi, G. Leitinger, S. D. Chiesa, G. Niedrist, and U. Tappeiner (2013), ESOLIP—Estimate of solid and liquid precipitation at sub- daily time resolution by combining snow height and rain gauge measurements, Hydrol. Earth Syst. Sci. Discuss., 10(7), 8683–8714. Mass, C. F., D. Ovens, K. Westrick, and B. A. Colle (2002), Does increasing horizontal resolution produce more skillful forecasts?, Bull. Am. Meteorol. Soc., 83, 407–430, doi:10.1175/1520-0477. Maussion, F., D. Scherer, R. Finkelnburg, J. Richters, W. Yang, and T. Yao (2011), WRF simulation of a precipitation event over the Tibetan Plateau, China—An assessment using remote sensing and ground observations, Hydrol. Earth Syst. Sci., 15, 1795–1817, doi:10.5194/hess- 15-1795-2011. Maussion, F., D. Scherer, T. Mölg, E. Collier, J. Curio, and R. Finkelnburg (2014), Precipitation seasonality and variability over the Tibetan Plateau as resolved by the high Asia reanalysis, J. Clim., 27(5), 1910–1927. Mayer, S., C. F. Maule, S. Sobolowski, O. B. Christensen, H. J. D. Sørup, M. A. Sunyer, K. Arnbjerg-Nielsen, and I. Barstad (2015), Identifying added value in high-resolution climate simulations over Scandinavia, Tellus A, 67, 24941. McKay, M. D., R. J. Beckman, and W. Conover (1979), A comparison of three methods for selecting values of input variables in the analysis of output from a computer code, Technometrics, 21(2), 239–245. Ménégoz, M., H. Gallée, and H. W. Jacobi (2013), Precipitation and snow cover in the Himalaya: From reanalysis to regional climate simulations, Hydrol. Earth Syst. Sci., 17(10), 3921–3936. Mitchell, K. E., et al. (2004), The multi-institution North American Land Data Assimilation System (NLDAS): Utilizing multiple GCIP products and partners in a continental distributed hydrological modeling system, J. Geophys. Res., 109, D07S90, doi:10.1029/2003JD003823. Mitchell, T. D., and P. D. Jones (2005), An improved method of constructing a database of monthly climate observations and associated high- resolution grids, Int. J. Climatol., 25, 693–712. Mölg, T., F. Maussion, and D. Scherer (2013), Mid-latitude westerlies as a driver of glacier variability in monsoonal High Asia, Nat. Clim. Change, 4,68–73, doi:10.1038/nclimate2055. Muneepeerakul, R., S. Azaele, G. Botter, A. Rinaldo, and I. Rodriguez-Iturbe (2010), Daily streamflow analysis based on a two-scaled gamma pulse model, Water Resour. Res., 46, W11546, doi:10.1029/2010WR009286. Nash, J. E., and J. V. Sutcliffe (1970), River flow forecasting through conceptual models—Part 1. A discussion of principles, J. Hydrol., 10(3), 282–290. Palazzi, E. V., J. Hardenberg, and A. Provenzale (2013), Precipitation in the Hindu-Kush Karakoram Himalaya: Observations and future scenarios, J. Geophys. Res. Atmos., 118,85–100, doi:10.1029/2012JD018697. Pandey, B. W. (2002), Geoenvironmental Hazards in Himalaya: Assessment and Mapping (the Upper Beas Basin), pp. 104–115, Mittal Publications. Pavelsky, T. M., S. Sobolowski, S. B. Kapnick, and J. B. Barnes (2012), Changes in orographic precipitation patterns caused by a shift from snow to rain, Geophys. Res. Lett., 39, L18706, doi:10.1029/2012GL052741. Prasad, V. H., and P. S. Roy (2005), Estimation of snowmelt runoff in Beas Basin, India, Geocarto Int., 20(2), 41–47. Pratap, B., D. P. Dobhal, R. Bhambri, M. Mehta, and V. C. Tewari (2016), Four decades of glacier mass balance observations in the Indian Himalaya, Reg. Environ. Change, 16(3), 643–658. Prein, A. F., G. Holland, R. M. Rasmussen, J. M. Done, K. Ikeda, M. Clark, and C. Liu (2013), Importance of regional climate model grid spacing for the simulation of heavy precipitation in the Colorado headwaters, J. Clim., 26, 4848–4857, doi:10.1175/JCLI-D-12-00727.1. Prékopa, A., and T. Szántai (1978), A new multivariate gamma distribution and its fitting to empirical streamflow data, Water Resour. Res., 14(1), 19–24. Racoviteanu, A. E., R. Armstrong, and M. W. Williams (2013), Evaluation of an ice ablation model to estimate the contribution of melting glacier ice to annual discharge in the Nepal Himalaya, Water Resour. Res., 49, 5117–5133, doi:10.1002/wrcr.20370. Ragettli, S., and F. Pellicciotti (2012), Calibration of a physically based, spatially distributed hydrological model in a glacierized basin: On the use of knowledge from glaciometeorological processes to constrain model parameters, Water Resour. Res., 48, W03509, doi:10.1029/ 2011WR010559. Rasmussen, R., et al. (2010), The NOAA/FAA/NCAR winter precipitation test bed: How well are we measuring snow?, Bull. Am. Meteorol. Soc., 93(6), 811–829. Rasmussen, R., et al. (2011), High-resolution coupled climate runoff simulations of seasonal snowfall over Colorado: A process study of current and warmer climate, J. Clim., 24(12), 3015–3048, doi:10.1175/2010JCLI3985.1. Rasmussen, R. M., et al. (2014), Climate change impacts on the water balance of the Colorado headwaters: High-resolution regional climate model simulations, J. Hydrometeorol., 15, 1091–1116, doi:10.1175/JHM-D-13-0118.1. Ribolzi, O., J. Patin, L. M. Bresson, K. O. Latsachack, E. Mouche, O. Sengtaheuanghoung, N. Silvera, J. P. Thiébaux, and C. Valentin (2011), Impact of slope gradient on soil surface features and infiltration on steep slopes in northern Laos, Geomorphology, 127(1–2), 53–63, doi:10.1016/ j.geomorph.2010.12.004.

LI ET AL. EVALUATE WATER BUDGET IN HIMALAYAN BASIN 4806 Journal of Geophysical Research: Atmospheres 10.1002/2016JD026279

Salathé, E. P., Jr., R. Steed, C. F. Mass, and P. H. Zahn (2008), A high-resolution climate model for the U.S. Pacific Northwest: Mesoscale feedbacks and local responses to climate change, J. Clim., 21, 5708–5726, doi:10.1175/2008JCLI2090.1. Senatore, A., G. Mendicino, D. J. Gochis, W. Yu, D. N. Yates, and H. Kunstmann (2015), Fully coupled atmosphere-hydrology simulations for the central Mediterranean: Impact of enhanced hydrological parameterization for short and long time scales, J. Adv. Model. Earth Syst., 7, 1693–1715, doi:10.1002/2015MS000510. Shrestha, M., L. Wang, T. Koike, Y. Xue, and Y. Hirabayashi (2012), Modeling the spatial distribution of snow cover in the Dudhkoshi Region of the Nepal Himalayas, J. Hydrometeorol., 13(204–222), doi:10.1175/JHM-D-10-05027.1. Simmons, A., S. Uppala, D. Dee, and S. Kobayashi (2007), ERA-Interim: New ECMWF reanalysis products from 1989 onwards, ECMWF Newsl., 110(110), 25–35. Skamarock, W. C., J. B. Klemp, J. Dudhia, D. O. Gill, D. M. Barker, M. G. Duda, X. Y. Huang, W. Wang, and J. G. Powers (2008), A description of the advanced research WRF version 3 (No. NCAR/TN-475+STR). National Center For Atmospheric Research Boulder Co Mesoscale and Microscale Meteorology Div. Stensrud, D. J. (2007), Parameterization Schemes: Keys to Understanding Numerical Weather Prediction Models, Cambridge Univ. Press, Cambridge, U. K. Sultana, R., K. L. Hsu, J. Li, and S. Sorooshian (2014), Evaluating the Utah Energy Balance (UEB) snow model in the Noah land-surface model, Hydrol. Earth Syst. Sci., 18(9), 3553–3570. Sushama, L., S. B. Said, M. N. Khaliq, D. N. Kumar, and R. Laprise (2014), Dry spell characteristics over India based on IMD and APHRODITE datasets, Clim. Dyn., 43(12), 3419–3437. Teutschbein, C., and J. Seibert (2010), Regional climate models for hydrological impact studies at the catchment scale: A review of recent modeling strategies, Geogr. Compass, 4(7), 834–860, doi:10.1111/j.17498198.2010.00357.x. Teutschbein, C., and J. Seibert (2012), Bias correction of regional climate model simulations for hydrological climate-change impact studies: Review and evaluation of different methods, J. Hydrol., 456,12–29. Thompson, G., P. R. Field, R. M. Rasmussen, and W. D. Hall (2008), Explicit forecasts of winter precipitation using an improved bulk microphysics scheme. Part II: Implementation of a new snow parameterization, Mon. Weather Rev., 136, 5095–5115. Tramblay, Y., D. Ruelland, S. Somot, R. Bouaicha, and E. Servat (2013), High-resolution MED-CORDEX regional climate model simulations for hydrological impact studies: A first evaluation of the ALADIN-climate model in Morocco, Hydrol. Earth Syst. Sci., 17, 3721–3739. Vincent, C., A. Ramanathan, P. Wagnon, D. P. Dobhal, A. Linda, E. Berthier, P. Sharma, Y. Arnaud, M. F. Azam, and J. Gardelle (2013), Balanced conditions or slight mass gain of glaciers in the Lahaul and Spiti region (northern India, Himalaya) during the nineties preceded recent mass loss, The Cryosphere, 7(2), 569–582. Viste, E., and A. Sorteberg (2015), Snowfall in the Himalayas: An uncertain future from a little-known past, Cryosphere Discuss., 9, 441–493. Wang, Z., X. Zeng, and M. Decker (2010), Improving snow processes in the Noah land model, J. Geophys. Res., 115, D20108, doi:10.1029/ 2009JD013761. Wagnon, P., et al. (2007), Four years of mass balance on Chhota Shigri Glacier, Himachal Pradesh, India, a new benchmark glacier in the western Himalaya, J. Glaciol., 53(183), 603–611. Warner, T. T. (2011), Numerical Weather and Climate Prediction,pp.96–113, Cambridge Univ. Press, New York. Weedon, G. P., G. Balsamo, N. Bellouin, S. Gomes, M. J. Best, and P. Viterbo (2014), The WFDEI meteorological forcing data set: WATCH Forcing Data methodology applied to ERA-Interim reanalysis data, Water Resour. Res., 50(9), 7505–7514, doi:10.1002/2014WR015638. Widen-Nilsson, E., L. Gong, S. Halldin, and C.-Y. Xu (2009), Model performance and parameter behavior for varying time aggregations and evaluation criteria in the WASMOD-M global water balance model, Water Resour. Res., 45, W05418, doi:10.1029/2007WR006695. Winiger, M. G. H. Y., M. Gumpert, and H. Yamout (2005), Karakorum–Hindukush–western Himalaya: Assessing high-altitude water resources, Hydrol. Processes, 19(12), 2329–2338. Xie, P., M. Chen, and W. Shi (2010), CPC unified gauge-based analysis of global daily precipitation, in Preprints, 24th Conf. on Hydrology, Am. Meteorol. Soc., vol. 2, Atlanta, Ga. Xu, C.-Y. (2002), WASMOD—The water and snow balance MODelling system, in Mathematical Models of Small Watershed Hydrology and Applications, edited by V. P. Singh and D. K. Frevert, pp. 555–590, Water Resources Publications, Chelsea, Mich. Xu, H. L., C.-Y. Xu, S. D. Chen, and H. Chen (2016), Similarity and difference of global reanalysis datasets (WFD and APHRODITE) in driving lumped and distributed hydrological models in a humid region of China, J. Hydrol., 542, 343–356. Yan, D. H., S. H. Liu, T. L. Qi, B. S. Weng, C. Z. Li, Y. J. Lu, and J. J. Liu (2016), Evaluation of TRMM precipitation and its application into distributed hydrological model in Naqu River Basin of the Tibetan plateau, Hydrol. Res., doi:10.2166/nh.2016.090. Yao, T., J. Pu, A. Lu, Y. Wang, and W. Yu (2007), Recent glacial retreat and its impact on hydrological processes on the Tibetan Plateau, China, and surrounding regions, Arct. Antarct. Alp. Res., 39(4), 642–650. Yasutomi, N., A. Hamada, and Y. Akiyo (2011), Development of a long-term daily gridded temperature dataset and its application to rain/snow discrimination of daily precipitation, Global Environ. Res., 15, 165–172. Yatagai, A., et al. (2012), PHRODITE: Constructing a long-term daily gridded precipitation dataset for Asia based on a dense network of rain gauges, Bull. Am. Meteorol. Soc., 93, 1401–1415. Yin, Y. X., C.-Y. Xu, H. S. Chen, L. Li, H. L. Xu, H. Li, and S. K. Jain (2016), Trend and concentration characteristics of precipitation and the related climatic teleconnections from 1982 to 2010 in the Beas river basin, India, Global Planet. Change, 145, 116–129. Yucel, I., A. Onen, K. K. Yilmaz, and D. J. Gochis (2015), Calibration and evaluation of a flood forecasting system: Utility of numerical weather prediction model, data assimilation and satellite-based rainfall, J. Hydrol., 523,49–66. Zängl, G. (2007), To what extent does increased model resolution improve simulated precipitation fields? A case study of two north-Alpine heavy-rainfall events, Meteorol. Z., 16(5), 571–580.

LI ET AL. EVALUATE WATER BUDGET IN HIMALAYAN BASIN 4807