sh

sk

nds

Received 25 June 2013Received in revised form 4 February 2014Accepted 22 March 2014Available online 1 April 2014This manuscript was handled byKonstantine P. Georgakakos, Editor-in-Chief,with the assistance of EmmanouilN. Anagnostou, Associate Editor

Circulation Models (GCMs) under three emission scenarios (B1, A1B and A2). The low resolution future

2011). In many parts of the world, climate change will most likelybe expressed through changes in freshwater availability. Hydrolog-ical systems are anticipated to experience not only changes in theaverage availability of water, but also in extreme events such asoods and droughts. Changes in precipitation are expected tosignicantly affect cryospheric processes and the hydrology ofheadwater catchments in the Himalayas (Immerzeel et al., 2012).

asing grees output r

relatively coarse in resolution and is generally coninsufcient for representing local variability necessary for cchange impact studies. Translating projections of changeson a global scale to the regional scale is important for all waterbased activities. These activities include irrigation, hydropowerdevelopment, and the reduction of risks related to oods anddroughts. A number of methods have been used to addressthese scale differences; these methods range from the simpleinterpolation of climate model results to dynamic or statisticaldownscaling methods (Bates et al., 2008).

Corresponding author. Tel.: +66 25245790.E-mail address: msbabel@ait.asia (M.S. Babel).

Journal of Hydrology 513 (2014) 422434

Contents lists availab

H

elsmelting of snow and ice, increasing evaporation and atmosphericwater vapour, and changes in soil moisture and runoff (Xu et al.,

regarding climate change in response to incregases (Mearns et al., 2003). However, the GCMhttp://dx.doi.org/10.1016/j.jhydrol.2014.03.0470022-1694/ 2014 Elsevier B.V. All rights reserved.nhouseemainssideredlimatemadeGlobal climate change is expected to have serious implicationsfor the Earths environment. Water is the one natural resource thatis expected to be most severely affected by climate change(Minville et al., 2008). Global warming is associated with changesin a number of components of hydrological systems such asprecipitations patterns, its intensity and extremes, the widespread

environmental change, which also affects important environmen-tal services that these regions provide, such as water supply tolowlands (Buytaert et al., 2010).

To prepare adaptation strategies for changing climatic condi-tions, decision makers require quantitative projections on regionalto local scales, depending on their purpose. Over several decades ofdevelopment, GCMs have consistently provided robust informationKeywords:Climate changeGCMKoshi basinUncertaintyIndices

1. Introductionprecipitation data obtained from the GCMs was downscaled using the statistical downscaling modelLARS-WG. The data was downscaled for 48 stations located in the six physiographic regions in the Koshibasin. The precipitation projections for three future periods, i.e. 2020s, 2055s and 2090s, are presentedusing empirical Probability Density Functions (PDFs) for each physiographic region. The differencesbetween the mean values of individual GCM projections and the mean value of the multi-model forthe three scenarios allow for the estimation of uncertainty in the projections. We also analyzed theprecipitation of the baseline and future periods using six indices that are recommended by the ExpertTeam on Climate Change Detection, Monitoring and Indices (ETCCDMI). Results indicate that not all GCMsagree on weather changes in precipitation will be positive or negative. A majority of the GCMs and theaverage values of all the GCMs for each scenario, indicate a positive change in summer, autumn andannual precipitation but a negative change in spring precipitation. Differences in the GCM projectionsexist for all the three future periods and the differences increase with time. The estimated uncertaintyis higher for scenario A1B compared to B1 and A2. Differences among scenarios are small during the2020s, which become signicant during the 2055s and 2090s. The length of the wet spell is expectedto increase, whereas the length of the dry spell is expected to decrease in all three future periods. Thereis a large scatter in the values of the indices: number of days with precipitation above 20 mm, 1-daymaximum precipitation, 5-day maximum precipitation, and amount of precipitation on the days withprecipitation above 95th percentile, both in direction and magnitude of the change.

2014 Elsevier B.V. All rights reserved.

Mountainous regions are fragile and are easily affected byArticle history: We analyzed precipitation projections for the Koshi river basin in Nepal using outputs from 10 GeneralAnalysis of future precipitation in the Ko

Anshul Agarwal a, Mukand S. Babel a,, Shreedhar MaaAsian Institute of Technology, P.O. Box 4, Klong Luang, Pathumthani 12120, ThailandbUNESCO-IHE Institute for Water Education, P.O. Box 3015, 2601 DA Delft, The Netherla

a r t i c l e i n f o s u m m a r y

Journal of

journal homepage: www.i river basin, Nepal

ey b

le at ScienceDirect

ydrology

evier .com/locate / jhydrol

by dividing the basin into six physiographic regions based on ele-

mean annual temperature is 20 C in the Hills and 16 C in the

HydThe Long Ashton Research Station Weather Generator (LARS-WG) is a stochastic weather generator developed by Semenovand Barrow (1997) for statistical downscaling. Several studies(such as Hashmi et al., 2011) have compared the performance ofLARS-WG with other statistical downscaling techniques and haveconcluded that LARS-WG can be adopted with condence forclimate change studies. Hashmi et al. (2009) analyzed theperformance of LARS-WG for the Auckland (in New Zealand) andconcluded that it prove to be an efcient tool for simulatingpresent climate and projecting its future states in terms of complexstatistics by using the information provided by a GCM. LARS-WGhas been applied in climate change impact studies in manyresearches, such as in the Saguenay watershed in northern Qubec,Canada (Dibike and Coulibaly, 2005); in Montreal, Canada(Nguyen, 2005) and different locations in Europe (Semenov andStratonovitch, 2010). The description of the latest version ofLARS-WG, called LARS-WG 5 and its capabilities is given inSemenov and Stratonovitch (2010). LARS-WG 5 incorporatesclimate projections from 15 GCMs used in the IPCC-AR4.LARS-WG 5 was used in the present study to downscale daily pre-cipitation data in the Koshi river basin, which is located in theHimalayan region. To the best of our knowledge, LARS-WG hasnot been applied to the Himalayan region before.

Substantial uncertainty remains regarding the precise impact ofclimate change on water resources (Kingston and Taylor, 2010).The estimates of uncertainty present plausible future climateswhich would help in investigating the potential consequences ofanthropogenic climate change. Such estimates are valuable for pol-icy makers and planners (Stott and Kettleborough, 2002). Theseestimates are also of fundamental importance for preparingapproaches to adaptation and mitigation (Deser et al., 2012).

Uncertainty in climate projections comes mainly from GCMs,SRES scenarios, downscaling methods and the innate internal var-iability of climate (Hawkins and Sutton, 2010; Hu et al., 2012). Inthis study the difference between different GCM outputs are con-sidered to be GCM uncertainty. This uncertainty arises because ofdifferences in the numerical and physical formulations of GCMs(e.g. spatial resolution, vertical layers, the representation of clouds,the convection process, the boundary layer, etc.). GCMs may yielddifferent responses to the same external conditions and this resultsin differences in their output (Fowler et al., 2007). The differencesin GCM outputs also come from the inadequate representation ofland surface and its features like vegetation and soil characteristics.Besides, GCMs cannot fully represent the effects of an enhancedlevel of CO2, or parameters like atmospheric chemistry, interactivebiogeochemistry, aerosols, dynamic vegetation, ice sheets, etc. Thedifference in the mean values of all GCMs for each of the threescenario (here B1, A1B and A2) is considered to be scenario uncer-tainty, in this study. Scenario uncertainty arises because of differ-ent assumptions of external factors like GHG emissions, whichinuence a climate system. In addition, there is internal variability,the natural variability of the climate system that occurs due to thebasic atmospheric system itself (Deser et al., 2012). Uncertaintiesalso arise from the incorporation of climate model results intohydrological models mainly because of different spatial scales.Countries like Nepal, which are dependent on rainfall and snow-fed rivers, may face greater problems as far as climate change isconcerned because uncertainty in precipitation (magnitude, timingand frequency) increases.

Many parts of Nepal are currently experiencing changes in pre-cipitation patterns (Shrestha et al., 2000; Baidya et al., 2008;Bartlett et al., 2010). Earlier assessments of the impact of climatechange on water resources in the Koshi basin were based on one

A. Agarwal et al. / Journal ofor two climate models, for example, Gosain et al. (2010) used Had-RM2 and HadRM3 and WWF (2009) used HadCM3. On the otherhand the present study considered outputs from ten GCMs forMountains. In general, the temperature decreases from South toNorth. Precipitation in the Koshi basin increases from the Low RiverValleys to the Mountains and then decreases in regions of higherelevation like the High Mountains and Himalayas. Maximum pre-cipitation is observed in the Mountains while minimum precipita-tion is observed in the Himalayas. The majority of the population(almost 70%) in the basin is dependent on rainfed agriculture forits livelihood (Dixit et al., 2009). The Hills and the Terai Plains con-tribute the maximum area to agriculture in the basin. Water stressduring the dry season and frequent oods during the monsoons arethe biggest challenges in the Koshi basin (NCVST, 2009).

2.2. Observed precipitation datavation differences. Furthermore, we analyzed the baseline andfuture periods precipitation data to study the changes that mayoccur in the various indices for precipitation extremes. The mainobjective of this study was to assess precipitation projections forthe Koshi river basin in Nepal and to analyze uncertainty in theseprojections by taking into account the differences between differ-ent GCM projections as well as between different future emissionscenarios. This study focus on analysis for three periods over the21st century: an early-century period of 20112030 (2020s), amid-century period of 20462065 (2055s), and a late-century per-iod of 20802099 (2090s).

2. Study area and data

2.1. Study area

This study was conducted in the Koshi river basin, in Nepal. TheKoshi ows through China, Nepal and India and is one of the largesttributaries of the Ganges. The Koshi River, along with its tributaries,drains a total area of 69,300 km2 up to its conuence with the Gan-ges in India (WWF, 2009; Gosain et al., 2010). In Nepal, Koshi is thelargest river basin, covers 18 districts (administrative boundaries inNepal) and nearly 30,000 km2 of land from the Himalayas to theagricultural lowlands of the Terai Plains. It consists of seven majorsub-basins (Sun Koshi, Indrawati, Dudh Koshi, Tama Koshi, Likhu,Arun and Tamor), all originating from the Himalayas. The basin arealies within latitudes 26510 and 29790N, and longitudes 85240 and88570E. The altitude of the basin ranges from 65 mamsl (metersabove mean sea level) in the Terai Plains to over 8000 mamsl inthe High Himalayas (Dixit et al., 2009). Given the signicantaltitudinal variation in the basin, it was divided into six distinctphysiographic regions: the Terai Plains (

Table 1Precipitation stations in the Koshi basin (Source: DHM, Nepal).

Station ID Lat_N Long_E Alt_m Station name

Terai Plains1319 26.48 87.27 72 Biratnagar1223 26.55 86.73 91 Rajbiraj

n stations, and APHRODITE data grid nodes (grey squares) along with respective grid IDs.

Hydrology 513 (2014) 422434Hydrology and Meteorology (DHM), Nepal. The precipitation dataof 35 gauging stations (Table 1) located in four physiographicregions (Mountains, Hills, Low River Valleys and Terai Plains)was used in this study. The locations of the stations are shown inFig. 1. Based on the available precipitation data it can be derivedthat the annual average precipitation is 2000 mm, 1670 mm,

Fig. 1. Location of the Koshi river basin in Nepal and its elevation ranges, precipitatio

424 A. Agarwal et al. / Journal of1200 mm and 1865 mm in the Mountains, Hills, Low River Valleysand Terai Plains respectively. More than 75% of the precipitation isconcentrated in the four months of the summer season (JuneSep-tember). Spring (MarchMay) receives about 15% of the annualprecipitation, while the combined precipitation during the monthsof winter (DecemberFebruary) and autumn (October and Novem-ber) is less than 10% of the total annual precipitation.

2.3. Gridded precipitation data

Observed precipitation data is not available for the Himalayasand the High Mountains regions of the Koshi basin. For theseregions, different sources of gridded precipitation data {TropicalRainfall Measuring Mission (Product: Daily TRMM 3B42 (V7)),Climatic Research Unit (Product: CRU TS3.21) and AsianPrecipitation Highly Resolved Observational Data IntegrationTowards Evaluation of Water Resources (APHRODITE) (Product:APHRO_MA_V1003R1)} were analyzed. The gridded precipitationdata products were judged based on its availability, spatial resolu-tion and comparison with the observed data. The data product fromAPHRODITE, which is available for Asia for the period of 19512007,was found to be most suitable. The APHRODITE data set is based onobserved precipitation data in conjunction with other precompileddatasets (Yatagai et al., 2009, 2012). APHRODITE used the observeddata from the maximum possible gauges available over Nepal togenerate the gridded precipitation. The use of best possibleavailable observation data has resulted in the generation of one ofthe most reliable daily precipitation products over Nepal and theHimalayan region (Duncan and Biggs, 2012). Andermann et al.(2011) have qualitatively assessed the reliability of the APHRODITEdaily precipitation data for Nepal.

1212 26.73 86.93 100 Phatepur1316 26.82 87.17 183 Chatara1320 26.70 87.27 200 Tarhara1311 26.82 87.28 444 Dharan

Low River Valleys1309 26.93 87.15 143 Tribeni1322 26.97 87.17 158 Machuwa Ghat1305 27.13 87.28 410 Leguwa Ghat1210 27.13 86.43 497 Kurule Ghat1028 27.57 85.75 633 Pachuwar Ghat

Hills1420 27.35 87.60 763 Dovan1036 27.68 85.63 865 Panchkal1115 27.45 85.82 1098 Nepal thok1325 27.37 87.15 1190 Dingla1307 26.98 87.35 1210 Dhankuta1027 27.78 85.90 1220 Bahrabise1211 27.03 86.83 1295 Khotang Bazar1303 27.28 87.33 1329 Chainpur (east)1107 27.28 85.97 1463 Sindhuli Gadhi

Mountains1024 27.62 85.55 1552 Dhulikhel1207 27.48 86.42 1576 Mane Bhanjyang1222 27.22 86.80 1623 Diktel1314 27.13 87.55 1633 Terhathum1206 27.32 86.50 1720 Okhaldhunga1405 27.35 87.67 1732 Taplejung1403 27.55 87.78 1780 Lungthung1406 27.20 87.93 1830 Memeng Jagat1102 27.67 86.05 1940 Charikot1203 27.43 86.57 1982 Pakarnas1006 27.87 85.87 2000 Gumthang1103 27.63 86.23 2003 Jiri1043 27.70 85.52 2163 Nagarkot1317 27.77 87.42 2590 Chepuwa1202 27.70 86.72 2619 Chaurikhark

scenarios are in most cases valid for other SRES scenarios as well(IPCC, 2007).

3. Methods

3.1. Downscaling

LARS-WG was used to downscale the low resolution precipita-tion projections obtained from the GCMs for all the 48 stationsconsidered in this study. First, the performance of LARS-WG wasanalyzed for each precipitation station using the baseline periods

Table 2Grids of APHRODITE in the High Mountains and the Himalayas regions in the Koshibasin.

High Mountains(27004000 m_amsl)

Himalayas(>4000 m_amsl)

GridID

Minelev

Maxelev

Meanelev

GridID

Minelev

Maxelev

Meanelev

27 1174 5186 2836 38 1266 7149 446028 1068 6494 3326 39 2681 6920 496029 1113 6424 3811 40 3374 8806 545130 545 6247 3108 44 3769 6795 536332 1457 5756 3955 47 1741 6940 4281

A. Agarwal et al. / Journal of Hydrology 513 (2014) 422434 4252.4. Future precipitation data

Ten GCMs that are included in IPCC AR4 were considered forthis study. These GCMs cover different resolutions (varying from1.4 1.4 to 4.0 5.0), come from different climate centreslocated around the world, and have vintage beyond the year2000. The data for these GCMs for selected SRES scenarios areavailable through LARS-WG5. The GCMs used in the study arelisted in Table 3.

Three emission scenarios B1, A1B and A2 were considered inthis study. These scenarios represent low, medium and high emis-sions of GHGes with respect to prescribed concentrations relativeto SRES. B1 generally represents the lowest GHG emissions duringthe 21st century (Maurer et al., 2009). A1B represents a balancedThe APHRODITE data product (APHRO_MA_V1003R1), whichhas the spatial resolution of 0.25 0.25 available at a daily scale,was used in this study for the Himalayas and High Mountains. Thegrids of the APHRODITE data over the Koshi basin in Nepal areshown in Fig. 1. Table 2 presents the elevation range for gridswhich fall under the Himalayas and High Mountains. The APHRO-DITE dataset was veried for the Koshi basin using observed sta-tion data for those grids for which the observation station datawas available. With reasonably good agreement between the twodatasets, we considered the APHRODITE datasets satisfactory touse in this study for the regions where observed data were notavailable. The annual average precipitation based on the APHRO-DITE data is 800 mm in the Himalayas and 1500 mm in the HighMountains.

33 1387 6260 384637 983 5882 307646 1605 6071 3875scenario which lies between B1 and A2. While A2 does not repre-sent the highest GHG emissions as per the SRES scenarios, it never-theless represents the highest emission scenario for which futuresimulations are available from most GCMs. Qualitative conclusionsrelated to future climate projections derived from these three

Table 3The global climate models (GCMs) used in this study.

No. Model Research centre Resolut

Atmosp

1 ECHAM 5 MPI, Germany 1.87 2 MRI-CGCM2.3 MRI, Japan 2.8 3 HadCM3 Hadley Centre UK 2.5 4 CGCM 3.1 CCCMA Canada 2.8 2.8 1.4 1.0 A1B 20055 MK3 CSIRO, Australia 1.9 6 CNCM3 CNRM, France 1.9 7 IPCM4 IPSL, France 2.5 8 GFCM21 GFDL, USA 2.0 9 CCSM3 NCAR, USA 1.4

10 INCM3 INM, Russia 4.0 1.9 1.9 1.9 B1, A1B 20011.9 2.0 2.0 A1B, A2 2004

3.75 2.0 2.0 B1, A1B, A2 20052.5 1.0 1.0 B1, A1B, A2 2005(19712000) dataset. LARS-WG uses Semi Empirical Distribution(SED) to approximate probability distributions of the dry and wetseries and of daily precipitation amounts in each month(Semenov and Stratonovitch, 2010). SED here is a discrete cumula-tive probability density function which uses 23 intervals in the cur-rent version (LARS-WG 5.5). Each interval contains a number ofevents based on the observed daily precipitation time series. Thesize of the intervals is exible and is decided based on the numberof events after each interval. Precipitation occurrence is modeledas alternate wet and dry series, wherein a wet day is dened as aday with precipitation > 0.0 mm. The length of each series is cho-sen randomly from the wet or dry semi-empirical distribution forthe month in which the series starts. For a wet day, the precipita-tion value is generated from the semi-empirical precipitation dis-tribution for the particular month, independent of the length ofthe wet series or the amount of precipitation on previous days.The calibration of the model is based on the comparison of the sta-tistical properties of the synthetic time series with those of theobserved data. The statistical properties include monthly mean,standard deviation, 95 percentile of monthly precipitation,monthly maximum precipitation, t-test and f-test (Khan et al.,2006; Hashmi et al., 2011).

For the models validation, a synthetic weather time series wasgenerated using the parameter le derived during the modelcalibration step. In this study, a 30-year long daily time series foreach precipitation station was generated. These time series werethen analyzed for mean, standard deviation, and maximumprecipitation and also to determine if there were any statisticallysignicant differences (at a 5% level) between the observed andsimulated data. To generate future climate scenarios for a station,D-change factors regarding a number of parameters (as calculatedduring the model calibration) were calculated on a monthly scalefrom the differences between the data of future periods and thebaseline period, as indicated by the outputs from the GCMs.These D-change factors were then applied as relative changes toLARS-WG parameters for the baseline precipitation of each stationin order to generate daily time series for future periods. A detaileddescription of the different steps involved in LARS-WG fordownscaling climate change projections is given in Semenov andStratonovitch (2010).

ion Scenarios Vintage

heric Ocean

1.87 1.5 1.5 B1, A1B, A2 20052.8 2.5 2.0 B1, A1B 2003

3.75 1.25 1.25 B1, A1B, A2 20001.4 1.0 1.0 B1, A1B, A2 20055.0 2.5 2.0 B1, A1B, A2 2004

Team on Climate Change Detection, Monitoring and Indices

The later may account for some gaps in the time series. The analy-

HydThe precipitation data for the future periods (and also the base-line period) were generated (downscaled) using calibrated and val-idated LARS-WG for all the GCMs for the available SRES scenarios(Table 3). For each future period 100 years of data were generated.It is important to note that these 100 years do not represent con-secutive years, but ensembles of precipitation time series withthe same statistics from which they were generated. The down-scaled precipitation at stations located in each physiographicregion was averaged to represent the precipitation over thatregion. The statistical signicance (at a 5% condence level) of eachGCM output was evaluated to analyze whether the projected pre-cipitation in future periods was signicantly different from thebaselines period precipitation. Each GCM output for a certain SRESscenario was considered to be independent. The annual precipita-tion data from hundred ensembles generated by LARS-WG (down-scaled) from the given GCM output under A1B scenario ispresented using Probability Distribution Functions (PDFs). Thespread of individual GCMs PDFs represents the natural variabilityof precipitation, while the inter-model difference represents theGCMs uncertainty. A combined PDF was constructed from thedownscaled time series of all the GCMs available for a given sce-nario with the assumption that the projections from each GCMwere equally likely to be true. The spread of these combined PDFsrepresents inter-annual variability, and includes GCM uncertainty.The differences between the median values of these combinedPDFs represent the scenarios uncertainty.

3.2. Uncertainty due to GCMs and SRES scenarios

To adequately illustrate the uncertainty in precipitation projec-tions in the various physiographic regions of the Koshi basin, aplausible range was quantied. This range was estimated fromthe changes in the mean value of precipitation for each future per-iod (an average of 20 years) from the baseline period. The rangewas estimated for seasonal and annual precipitation. The seasonalresults were based on four clearly distinguished seasons in Nepal:winter (DJF), spring (MAM), summer (JJAS), and autumn (ON)(Shrestha et al., 1999). Several probability distribution functionswere tted to the GCM projections for each scenario and theirgoodness-of-t was evaluated using the KolmogorovSmirnov test.This test was used to check if a sample came from a hypothesizedcontinuous distribution. The KolmogorovSmirnov statistic (D) isbased on the largest vertical difference between the theoreticaland the empirical Cumulative Distribution Function (CDF). TheKolmogorovSmirnov statistic (D) is:

D max16i6n

Fxi i 1n ;in Fxi

The hypothesis regarding the distributional form was analyzedat a 5% signicance level. CDFs were computed using the best tdistribution for seasonal and annual precipitation changes infuture periods (relative to the mean of 19712000) projected bythe GCMs under B1, A1B and A2 scenarios. Among various distribu-tion functions analyzed, the Weibull three parameter, Loglogisticthree parameter, Lon-normal three parameter and Beta were foundto be the best t for most cases. The 5 and 95 percentile intervalsfrom the CDFs are presented as the uncertainty range.

3.3. Precipitation indices

Precipitation data for the baseline period and future periods isrepresented by certain indices to determine the changes in

426 A. Agarwal et al. / Journal ofextreme precipitation events during different future periods in allsix physiographic regions of the Koshi basin. The indices werechosen so as to represent a variety of rainfall characteristics andsis of the seasonal results indicates a higher bias value during themonsoon season (JJAS) compared to other seasons in all the regionsof the Koshi basin.

4.2. Precipitation projections

In this section, the projection of precipitation is presented(Fig. 4) using empirical Probability Density Functions (PDFs) forthe ten GCMs under the A1B scenario and for three SRES scenarios.The probability that precipitation will be below a certain value isequal to the area under the curve to the left of the given value.The total area under each PDF is equal to 1. In general, PDFs indi-cate inter-annual variability (as individual GCM PDFs) as well asprobable uncertainty (PDFs of different GCMs) of average annualprecipitation projections. The changes in precipitation projectedby most GCMs under A1B scenario for all three future periodswas statistically signicant (at a 5% condence level), with someexceptions during the 2020s in the Mountains (CGMR, MPEH), Hills(ETCCDMI). The software RClimDex, which is developed byETCCDMI and is available at http://cccma.seos.uvic.ca/ETCCDMI/index.shtml, was used for calculating these indices.

4. Results and discussion

4.1. Calibration and validation of LARS-WG

The observed and simulated mean monthly precipitation, stan-dard deviation and monthly maximum precipitation at one stationlocated in the Low River Valleys and at another in the Mountainsregion of the Koshi basin are shown in Fig. 2. The results indicatethat the model produces the precipitation parameters quite well.The statistical results were satisfactory for all the 48 stationslocated in the six physiographic regions of the Koshi basin andthese results match their use in previous precipitation-relatedstudies as well (for example, Hashmi et al., 2011; Dibike andCoulibaly, 2005).

The models validation results in terms of mean, standard devi-ation, 95 percentile precipitation, and monthly maximum precipi-tation at each station were also found satisfactory. The averagedmonthly time series for the baseline period for the stations locatedin each physiographic region of the Koshi basin is shown in Fig. 3.The models performance in the Himalayas and High Mountainswas better compared to its performance in the other four regions.This may be because of the continuous time series data used for theHimalayas and High Mountains region, obtained from APHRODITE,as compared to the observed station data used for other regions.rainfalls extreme behavior. We selected six indices: the length ofwet spells i.e., the maximum number of consecutive days with pre-cipitation above 1 mm (CWD); the length of dry spells i.e., the max-imum number of consecutive days with precipitation less than1 mm (CDD); number of days with heavy precipitation here, calcu-lated as the total number of days with precipitation above 20 mm(R20); the most intense rainfall event in 1 day (RX1 day); the mostintense rainfall event in 5 consecutive days (RX5 day); and thetotal amount of precipitation on those days that have precipitationabove 95th percentile, i.e. precipitation on very wet days (R95p).Similar indices have also been used in previous studies (Vincentet al., 2011; Baidya et al., 2008; Kruger, 2006) and are recom-mended by the World Meteorological Organization, the projecton Climate Variability and Predictability (CLIVAR) and the Expert

rology 513 (2014) 422434(CSMK, GFCM) and Low River Valleys (CGMR, GFCM and MPEH).For the 2055s and 2090s, all projections show a signicant change(at a 5% condence level) in all the regions. Similarly, the projected

d in

Hydchanges for B1 and A2 scenarios were statistically signicant, witha few exceptions during the 2020s.

A majority of the GCMs projected an increase in precipitationduring all the three future periods, with the exception of IPCMand CSMK (Fig. 4). The results from 25 ensembles (projections from10 GCMs and 3 scenarios) considered in this study indicate that formean annual precipitation change, 17 ensembles show an increasewhile 8 show a decrease in precipitation. More precisely, precipita-tion projections from IPCM and CSMK show negative changes dur-ing all the three future periods under all three scenarios whileMPEH shows negative changes only for the 2055s and 2090s. Theaverage value gleaned from the 10 GCMs under A1B scenario, aswell as median values for PDFs of all the three scenarios, indicatesan increase in precipitation. This shows that the chances ofincrease in precipitation are more likely in the Koshi basin. A phys-

Fig. 2. Comparison of observed and LARS-WG simulated data at one station locatebaseline period (19712000).

A. Agarwal et al. / Journal ofical law (the ClausiusClapeyron relation) determines that thewater-holding capacity of the atmosphere increases by about 7%for every 1 C rise in temperature. The increase in water vapourcontent of the atmosphere resulting from expected increase intemperature is one the main reasons which leads to the changein precipitation. Climate model simulations and empirical evi-dences conrm that the warmer climates, owing to increasedwater vapour will lead to more intense precipitation events(IPCC, 2007).

In the early-century period, different GCM projections showclose agreement and are assumed to have the same inter-annualvariability as they do in the baseline period. This is indicated bythe overlap between the PDFs for the 2020s and the baseline per-iod, as shown in Fig. 4. Inter-annual variability increases during themid and late-century periods, as shown by the more at PDFs inFig. 4. The overlaps between the baseline period PDFs and futureperiod PDFs also indicate that the amount of precipitation in dryyears may remain the same, but precipitation will signicantlyincrease during the wet years, a projection made by most GCMs.The difference between individual GCM PDFs increases with thetime period in all the regions of the Koshi basin. Also, the PDFsfor the three scenarios are closer to each other during the early-century period but show increased differences during the midand late-century periods. Inter-annual variability is larger thanthe differences between GCMs or scenario PDFs during the early-century period. On the other hand, inter-annual variability is smal-ler than, the differences between GCMs or scenario PDFs during themid and late-century periods. The differences between scenarioPDFs also increases with the time period.

Fig. 4 indicates that among the 10 GCMs considered here, theprojections from three GCMs lie at extreme ends (that is, the lowerend (IPCM) and the upper end (MIHR and INCM)) for most cases.The resolution of the Ocean model for these three GCMs is lowerthan the other seven GCMs (Table 3). This may be one of the rea-sons for these GCMs to behave differently than the majority. Thehigher resolution of the Ocean model in GCMs led to a better sim-ulation of the sea surface temperature, sea ice extent, surface heat,momentum ux and the ocean heat transfer process, all of whichhave considerable impact on climate processes (CCSP, 2008). Also,two of these models, MIHR and INCM require ux adjustments toprevent large climate drifts in their simulations. These adjustmentsare not required in other GCMs considered in this study. Other key

the Low River Valleys and one station in the Mountains of the Koshi basin for the

rology 513 (2014) 422434 427features which differentiate GCMs are: compatibility between theatmospheric models and ocean models heat and water budget,prognostic variables that are considered for cloud characterization,etc. (CCSP, 2008; Randall et al., 2007). The differences in variousSRES scenario projections mainly arise because of the varyingassumptions related to economic, social and environmental modes,that form the basis of these scenarios. The differences among thescenarios during the early-century period are relatively small butthe differences increase with the time period (Stott andKettleborough, 2002; Knutti et al., 2003).

The ndings here are consistent with the projected meanannual precipitation for Nepal, as reported by NCVST (2009). Nei-ther nding shows a clear trend with reference to increase anddecrease in precipitation. Studies analyzing future precipitationbased on GCM projections agree that in the future periods, precip-itation may become more variable with the increase in the numberof extreme events (Sun et al., 2012; Bates et al., 2008; NSF, 2002).

4.3. Uncertainty in annual and seasonal precipitation

As discussed in the previous section, PDFs show inter-annualvariability in precipitation projections for different future periods.Differences in projections also exist because of different GCMs andvarious emission scenarios. This section aims to outline the rangeof uncertainty arising from differences in projections from differ-ent GCMs under the three emission scenarios. The best t CDFsshowing the changes in seasonal and annual precipitation during

Hyd428 A. Agarwal et al. / Journal ofthe three future periods of the 2020s, 2055s and 2090s (relative tothe 19712000 average) were analyzed. The uncertainty range wasestimated within the central 595% value of CDFs. The CDFs forseasonal precipitation change projections for the 2055s are shownin Fig. 5. The uncertainty range for all three future periods in all theregions of the Koshi basin is presented in Fig. 6.

The changes in precipitation are not uniform; rather they rangefrom negative to positive for all the three future periods withregard to seasonal as well as annual precipitation. No clear patternin precipitation changes is evident during any season, as can begleaned from Figs. 5 and 6. This may be due to the complexity ininterpreting precipitation projections, since different GCMs oftendo not agree on whether precipitation will increase or decreaseat a specic location and agree even less on the magnitude of that

Fig. 3. Observed and LARS-WG simulated precipitation in the six physiogrrology 513 (2014) 422434change (Girvetz et al., 2009). GCMs corresponding to the A1Bscenario show higher uncertainty than those corresponding to B1and A2.

Uncertainty increases in all the seasons in the mid and late-century periods. Fig. 6 indicates that the highest range of uncer-tainty is projected for the Mountains region, while the lowestrange is projected for the Himalayas. Winter precipitation, whichcontributes around 3% of the annual precipitation, is expected tochange by a small amount and has the least uncertainty, followedby autumn precipitation. Spring precipitation is expected tochange signicantly and shows higher uncertainty compared towinter and autumn precipitation. The summer season accountsfor maximum precipitation in a year in all the regions and hasthe highest uncertainty (changes in mm indicated by the large

aphic regions of the Koshi basin for the baseline period (19712000).

HydA. Agarwal et al. / Journal ofrange of 5 and 95 percentile bounds, Fig. 6) compared to the otherthree seasons during all the three future periods. During spring,there is a high probability that the change may be negative, whilesummer is highly likely to witness a positive change. The summermonsoon is expected to become more intense and also more vari-able. This may be because multiple variables inuencing climatecould potentially trigger abrupt transitions, leading to either adrier monsoon with signicantly less precipitation than currentlevels, or a more wet monsoon with much greater rainfall intensity(Zickfeld et al., 2005). Summer precipitation projections for Nepalin studies by NCVST (2009) and Bartlett et al. (2010) also showan increase towards the end of the 21st century.

Fig. 4. Probability Density Functions (PDFs) for annual precipitation data for the ten GCavailable from all the GCMs for each of the scenarios. Note: x is 100 for GCM PDFs andrology 513 (2014) 422434 429It is interesting to note (from Fig. 4) that in all the regions of theKoshi basin, during three future periods, a few GCMs (IPCM, MIHRand INCM) behave completely different from the rest of the GCMs.Such GCMs that behave differently are responsible for a very largeuncertainty range. Considering these GCMs as outliers, the uncer-tainty range may be reduced by almost 4050% for annual andseasonal precipitation change. For example, the uncertainty rangefor annual precipitation change in the Mountains region in 2055swill reduce in the span of 200 to 400 mm (instead of 600 to800 mm) if we exclude the projections from three outlier GCMs(IPCM, MIHR and INCM). As expected, the uncertainty range growsmonotonically with time in all the regions of the Koshi basin. This

Ms under the A1B scenario and for scenarios B1, A1B and A2 using the projections800, 1000 and 700 for B1, A1B and A2 PDF respectively.

Hyd430 A. Agarwal et al. / Journal ofmay be due to uncertainties in climate sensitivity and the carboncycle (Knutti et al., 2003). A similar behavior of uncertainty withrespect to GCMs, GHGes, and time horizon was also reported byresearchers such as Minville et al. (2008) and Chen at al. (2011).The uncertainty range projected in this study covers a considerablerange and the changes are physically plausible. The sampling pro-cess that led to the selection of these 25 ensembles from a hypo-thetical population of such samples took into consideration allthe extreme possibilities in future periods. The possibilities werebased on various SRES scenario assumptions as well as on differ-ences in GCM structures. The core utility of the projected uncer-tainty range is that it may provide possible and plausible ideasabout changes that may occur in precipitation in future periods,and this will aid in developing planning and coping tools and adap-tation strategies.

Fig. 5. Cumulative Distribution Functions (CDFs) for seasonal precipitation change projescenarios.rology 513 (2014) 4224344.4. Precipitation indices

To analyze the changes in precipitation characteristics duringthe future periods, three GCMs were selected from the 10 GCMsunder the A1B scenario (Fig. 4). These GCMs were chosen basedon the values of maximum decrease (low), as close as possible tothe mean of the ten GCMs (avg), and maximum increase (high)in the median value of their mean annual precipitation. It isassumed that the indices from the two GCMs at the extreme endscover the range of uncertainty in projections while the one closestto the mean represents the mean value of the 10 GCMs projec-tions. The selected GCMs and the values of their indices are pre-sented in Table 4.

The length of the wet spell (CWD) during the baseline period ishighest in the Mountains, followed by the High Mountains,

ctions during the 2055s (relative to the 19712000 average) under B1, A1B and A2

HydA. Agarwal et al. / Journal ofHimalayas, Hills, Terai Plains and nally lowest in the Low RiverValleys, as indicated in Table 4. CWD is expected to increase duringall three future periods in all the regions of the Koshi basin, as indi-cated by the GCM predictions. On the other hand, the length of thedry spell (CDD) is expected to decrease during all three future peri-ods, although large differences exist among the GCMs regardingthe magnitude of this change. During the baseline period, themaximum number of days with heavy precipitation (R20) occurredin the Terai Plains, followed by the Mountains, Hills, Low RiverValleys, High Mountains and Himalayas. According to all GCM pro-jections, R20 is expected to decrease during the 2020s in all theregions of the Koshi basin, with the exception of the projectionsof the GCM CGMR in the Terai Plains and MIHR in the Mountains.Large differences exist in GCM projections for the 2055s and 2090sas GCMs do not agree on the direction or the magnitude of change.The maximum difference is expected in the Mountains, with R20varying from 0 to 60 days during the 2055s.

Fig. 6. Range of change in seasonal and annual mean precipitation, as projected under B1and upper ends of the bar show the 5 percentile and 95 percentile intervals of the uncerology 513 (2014) 422434 431During the baseline period, the value of the index of the mostintense 1 dayprecipitation (RX1 day) decreaseswith increase in ele-vation,with thehighest valueestimated in the LowRiverValleys andthe lowest in the Himalayas. RX1 day is expected to follow a similarregional pattern in all the three future periods. It tends to decreaseduring the 2020s in all the regions as indicated by all the GCMs,but there is no consensus among the GCMs for the periods of the2055s and 2090s as far as the direction of change of RX1 day is con-sidered. GCMpredictions vary in themagnitude of change for all thethree future periods in all the regions. During the baseline period,the index of the most intense accumulated 5 day precipitation(RX5 day) had the highest value in the Hills, followed by the LowRiver Valleys, Terai Plains, Mountains, High Himalayas and Himala-yas. During the future periods, the regional pattern of RX5 day isexpected to change with the highest value projected for the TeraiPlains instead of theHills. TheRX5 day value is projected todecreaseduring the 2020s in all the regions, but the GCMs do not agree on the

, A1B and A2 scenarios in the six physiographic regions of the Koshi basin. The lowerrtainty range, while the dot represents the 50 percentile value.

coner th

2

Low Avg High L

I140825

I13

HydHimalayasIPCM MPEH MIHR

CWD (Days) 86 110 122 128CDD (Days) 55 42 38 34R20 mm (Days) 1 0 0 0RX1 day (mm) 19.9 11.5 14 16.1RX5 day (mm) 56.1 35.6 42.9 50.7R95P (mm) 115 74.2 99.8 119.4

High MountainsIPCM MPEH MIHR

CWD (Days) 120 162 171 171CDD (Days) 48 35 34 26Table 4Precipitation indices for the six physiographic regions of the Koshi basin, by taking intothe average of the 10 GCMs (Avg) and maximum increase (High) in precipitation und

Baseline 2020s

432 A. Agarwal et al. / Journal ofdirection of change during the 2055s and 2090s. The maximumvalue of the total amount of precipitation on very wet days (R95p)during the baseline period was in the Terai Plains, followed by theHills,Mountains, LowRiver Valleys, HighMountains andHimalayas.R95p is projected to decrease during the 2020s according to all theGCMs, while differences exist in the direction of change during the2055s and 2090s. A large difference exists among the GCMs regard-ing the magnitude of change of R95p as well in all the three futureperiods in all the regions of the Koshi basin.

5. Conclusions

This study analyzed precipitation projections in the six physio-graphic regions of the Koshi basin in Nepal: the Terai Plains, LowRiver Valleys, Hills, Mountains, High Mountains and Himalayas.For future precipitation projections, three SRES scenarios (B1,A1B and A2) were selected, each of which assumes a distinctly

R20 mm (Days) 8 0 1 3 0RX1 day (mm) 32.3 15.3 20 23.1 1RX5 day (mm) 97.5 57.2 73.1 84.1 4R95P (mm) 232 138.3 181.6 209.3 9

MountainsIPCM GFCM MIHR I

CWD (Days) 131 166 167 173 1CDD (Days) 43 19 20 18 2R20 mm (Days) 20 3 19 23 0RX1 day (mm) 39.9 24.5 32.6 33 1RX5 day (mm) 134.6 84.6 111.7 115.2 5R95P (mm) 311.1 223.4 298.7 327.2 1

HillsIPCM GFCM MIHR I

CWD (Days) 68 90 105 115 8CDD (Days) 52 24 26 21 2R20 mm (Days) 17 3 12 15 1RX1 day (mm) 57.4 27.5 36 37.5 2RX5 day (mm) 147 75.3 99.4 109.9 5R95P (mm) 334.7 207.5 272.7 297.4 1

Low River ValleysIPCM GFCM MIHR I

CWD (Days) 20 24 29 30 2CDD (Days) 60 39 35 34 3R20 mm (Days) 14 5 11 13 3RX1 day (mm) 69.1 33.8 44.2 48.9 2RX5 day (mm) 140.7 70 94.7 104.2 6R95P (mm) 290.3 189.5 254.4 283.1 1

Terai PlainsIPCM CGMR INCM I

CWD (Days) 63 54 63 69 4CDD (Days) 20 36 31 31 3R20 mm (Days) 24 15 25 30 1RX1 day (mm) 55.2 44.4 55.5 65.5 3RX5 day (mm) 140.1 111.3 141.3 168.3 9R95P (mm) 356.9 285.2 364.8 418.2 2sideration data from the three GCMs that represent maximum decrease (Low), close toe A1B scenario.

055s 2090s

ow Avg High Low Avg High

PCM NCCCM INCM IPCM HadCM MIHR01 123 125 95 125 1331 35 38 49 30 46

0 4 0 0 1.9 17.1 25.9 8.1 16.8 19.32.5 48.6 88.3 25.2 48.9 63.52.4 113.3 184.7 45.3 117.7 147.2

PCM NCCCM INCM IPCM HadCM INCM57 169 162 150 166 1634 27 35 38 20 39

rology 513 (2014) 422434different direction of future developments. We used multipleGCM projections for each of the three scenarios. A statistical down-scaling model (LARS-WG) was used to downscale the projectionsfrom the GCMs under the three SRES scenarios. Three future peri-ods of the 21st century were taken into consideration: an early-century period from 2011 to 2030 (2020s), a mid-century periodfrom 2046 to 2065 (2055s), and a late-century period from 2080to 2099 (2090s). The range of uncertainty was derived using the595% value of the best t distribution function for changes inprecipitation for the future periods (with respect to the baselineperiod), as projected by different GCMs for each of the threescenarios. To determine the changes in extreme precipitationevents, the data for the baseline period and the future periodswas represented by six indices.

The calibration and validation results for the baseline periodsprecipitation suggest that the downscaling model LARS-WGperformed well for all the stations in the Koshi basin and thus canbe used for downscaling data regarding future precipitation. A

1 42 0 3 281.6 20.7 40.4 11.2 22.9 33.80.5 76.2 152.6 41.4 83.6 126.85.6 197.6 364.2 94 216.9 296.5

PCM CGMR INCM IPCM NCCCM INCM65 174 164 153 178 1740 21 22 22 18 24

27 60 0 11 557.1 36.2 65.8 18.1 29.1 51.16.6 126 229.6 62 98.5 181.460.3 345.2 586.9 155.1 282.2 466

PCM GFCM INCM IPCM GFCM INCM5 109 110 74 108 1143 24 25 27 25 30

18 42 1 18 341.2 40.3 71.1 20.8 41.2 59.85.4 114.1 215.8 56 116.4 167.555.2 331.4 540.6 144.3 316.2 416.9

PCM GFCM INCM IPCM GFCM INCM0 32 41 19 33 408 37 38 42 37 41

14 25 2 14 229.1 49.8 81.2 26.9 50.1 70.41 106.3 180.5 57.3 105.5 155.358.8 283.2 442.1 141.2 279.1 388

PCM NCCCM INCM IPCM GFCM INCM9 64 84 47 70 845 30 33 63 36 380 28.9 47 8 30 448.7 59.4 102 35.9 64.7 91.48.4 147 265.3 91.3 163.4 236.251.3 395.4 621.2 223.5 404.7 562.8

Hydmajority of the GCMs project an increase in precipitation for all thethree future periods. Themedian value of the PDFs for the three sce-narios also indicates an increase in precipitation in all the regions ofthe Koshi basin. The amount of precipitation in dry years mayremain the same but the amount is expected to increase signi-cantly during the wet years in the future periods, as per the projec-tions of most GCMs. Precipitation is also expected to become morevariable in future periods. The difference between the individualGCM PDFs increases with the time period. Inter-annual variabilityin precipitation amounts is larger than the differences betweenthe GCMs or scenario PDFs during the early-century period. Duringthe mid and late-century periods, differences between GCM PDFsare higher compared to the inter-annual variability. These resultsimply that precipitation projections are subject to considerableuncertainty. GCMs show different signs of change in seasonal aswell as annual precipitation. The range of uncertainty in the pro-jected change is maximum in the Mountains and minimum in theHimalayas. There is high probability that precipitation duringspring may reduce in future periods and increase in the other sea-sons. The same is true for annual precipitation as well. The highestrange is projected under the A1B scenario for the 2020s and 2055sand under the A2 scenario for the 2090s. Uncertainty grows mono-tonically with time in all the regions of the Koshi basin.

The indices representing the characteristics of precipitationindicate that the length of the wet spell is expected to increase dur-ing all the three future periods in all the regions of the Koshi basin.The length of the dry spell is expected to decrease during all thethree future periods, although large differences exist in the magni-tude of change. The maximum number of days with heavy precip-itation is expected to decrease during the 2020s while noticeabledifferences exist in GCM projections for the 2055s and 2090s forthe same. This is because the GCMs do not agree on the directionor the magnitude of change. The precipitation indices, most intense1 day precipitation (RX1 day), most intense accumulated 5 daysprecipitation (RX5 day) and the maximum value of the totalamount of precipitation on very wet days (R95P) tend to decreaseduring the 2020s in all the regions. For the 2055s and 2090s, thereis no consensus among the GCMs about the direction of change ofthese indices. GCM predictions vary in the magnitude of change forall the three future periods in all the regions.

These results demonstrate that while future precipitation pro-jections over the Koshi basin are highly uncertain, the use ofmulti-model ensembles may provide expected range of future cli-mate for climate change impact assessments on a regional scale.

Acknowledgements

The authors acknowledge the nancial support provided in partby the UNESCO-IHE, Delft, Netherlands through the UNESCO-IHEPartnership Research Fund (UPaRF), and in part by the Asian Insti-tute of Technology, Thailand to conduct this research. The authorswould also like to thank the Department of Hydrology and Meteo-rology, Nepal for providing meteorological data and the Depart-ment of Irrigation, Nepal for supporting the acquisition of the data.

References

Andermann, C., Bonnet, S., Gloaguen, R., 2011. Evaluation of precipitation data setsalong the Himalayan front. Geochem. Geophys. Geosyst. 12 (7). http://dx.doi.org/10.1029/2011GC003513.

Baidya, S.K., Shrestha, M.L., Sheikh, M.M., 2008. Trends in daily climatic extremes oftemperature and precipitation in Nepal. J. Hydrol. Met. 5 (1), 3851.

Bartlett, R., Bharati, L., Pant, D., Hosterman, H., McCornick, P., 2010. Climate changeimpacts and adaptation in Nepal. In: IWMI Working Paper 139, InternationalWater Management Institute, Colombo, Sri Lanka, p. 35. Doi: 10.5337/2010.227.

A. Agarwal et al. / Journal ofBates, B.C., Kundzewicz, Z.W., Wu, S., Palutikof, J.P. (Eds.), 2008. Climate change andwater. In: Technical Paper of the Intergovernmental Panel on Climate Change.IPCC Secretariat, Geneva, p. 210.Buytaert, W., Vuille, M., Dewulf, A., Urrutia, R., Karmalkar, A., Celleri, R., 2010.Uncertainties in climate change projection and regional downscaling in thetropical Andes: implications for water resources management. Hydrol. EarthSyst. Sci. 14, 12471258.

CCSP, 2008. Climate Models: An Assessment of Strengths and Limitations. In: AReport by the U.S. Climate Change Science Program and the Subcommittee onGlobal Change Research. Department of Energy, Ofce of Biological andEnvironmental Research, Washington, D.C., USA, p. 124.

Chen, J., Brissette, F.P., Leconte, R., 2011. Uncertainty of downscaling method inquantifying the impact of climate change on hydrology. J. Hydrol. 401, 190202.

Deser, C., Phillips, A., Bourdette, V., Teng, H., 2012. Uncertainty in climate changeprojections: the role of internal variability. Clim. Dyn. 38, 527546.

Dibike, Y.B., Coulibaly, P., 2005. Hydrologic impact of climate change in theSaguenay watershed: comparison of downscaling methods and hydrologicmodels. J. Hydrol. 307, 145163.

Dixit, A., Upadhya, M., Dixit, K., Pokhrel, A., Rai, D.R., 2009. Living With Water Stressin the Hills of the Koshi Basin, Nepal. ICIMOD, Lalitpur, Nepal.

Duncan, J.M.A., Biggs, E.M., 2012. Assessing the accuracy and applied use ofsatellite-derived precipitation estimates over Nepal. Appl. Geogr. 34, 626638.

Fowler, H.J., Blenkinsop, S., Tebaldi, C., 2007. Linking climate change modeling toimpact studies: recent advances in downscaling techniques of hydrologicalmodeling. Int. J. Climatol. 27, 15471578.

Girvetz, E.H., Zganjar, C., Raber, G.T., Maurer, E.P., Kareiva, P., et al., 2009. Appliedclimate-change analysis: the climate wizard tool. PLoS ONE 4 (12), e8320.http://dx.doi.org/10.1371/journal.pone.0008320.

Gosain, A.K., Shrestha, A.B., Rao, S., 2010. Modelling Climate Change Impact on theHydrology of the Eastern Himalayas. TR 4, ICIMOD, Lalitpur, Nepal.

Hashmi, M.Z., Shamseldin, A.Y., Melville, B.W., 2009. Downscaling of future rainfallextreme events: a weather generator based approach. In: 18th World IMACS/MODSIM Congress, Cairns, Australia, 1317 July, 2009. .

Hashmi, M.Z., Shamseldin, A.Y., Melville, B.W., 2011. Comparison of SDSM andLARS-WG for Simulation and downscaling of extreme precipitation events in awatershed. Stoch. Environ. Res. Risk Assess. 25, 475484.

Hawkins, E., Sutton, R., 2010. The potential to narrow uncertainty in projections ofregional precipitation change. Clim. Dyn. 37, 407418.

Hu, Y., Maskey, S., Uhlenbrook, S., 2012. Downscaling daily precipitation over theYellow River source region in China: a comparison of three statisticaldownscaling methods. Theor. Appl. Climatol., Doi: 10.1007/s00704-012-0745-4.

Immerzeel, W.W., Beek, P.H.V., Konz, M., Shrestha, A.B., Bierkens, M.F.P., 2012.Hydrological response to climate change in a glacierized catchment in theHimalayas. Clim. Change 110, 721736.

IPCC, 2007. Climate Change 2007: The Physical Science Basis. In: Solomon, S., Qin,D., Manning, M., Chen, Z., Marquis, M., Averyt, K.B., Tignor, M., Miller, H.L. (Eds.),Contribution of WG I to AR4 of IPCC, Cambridge University Press, ISBN 978-0-521-88009-1 (pb: 978-0-521-70596-7).

Khan, M.S., Coulibaly, P., Dibike, Y., 2006. Uncertainty analysis of statisticaldownscaling methods. J. Hydrol. 319, 357382.

Kingston, D.G., Taylor, R.G., 2010. Sources of uncertainty in climate change impactson river discharge and groundwater in a headwater catchment of the Upper NileBasin, Uganda. Hydrol. Earth Syst. Sci. 14, 12971308.

Knutti, R., Stocker, T.F., Joos, F., Plattner, G.K., 2003. Probabilistic climate changeprojections using neural networks. Clim. Dyn. 21, 257272.

Kruger, A.C., 2006. Observed trends in daily precipitation indices in South Africa:19102004. Int. J. Climatol. 26, 22752285.

Maurer, E.P., Adam, J.C., Wood, A.W., 2009. Climate model based consensus on thehydrologic impacts of climate change to the Rio Lempa basin of CentralAmerica. Hydrol. Earth Syst. Sci. 13, 183194.

Mearns, L.O., Giorgi, F., Whetton, P., Pabon, D., Hulme, M., Lal, M., 2003. Guidelinesfor Use of Climate Scenarios Developed from Regional Climate ModelExperiments. Available from the IPCC, Task Group on Scenarios for ClimateImpact Assessment.

Minville, M., Brissette, F., Leconte, R., 2008. Uncertainty of the impact of climatechange on the hydrology of a nordic watershed. J. Hydrol. 358, 7083.

National Science Foundation, 2002. Increase in rainfall variability related to globalclimate change; impacts on ecosystems are greater than previously anticipated.Science Daily. From , (retrieved 13.01.14).

NCVST (Nepal Climate Vulnerability Study Team), 2009. Vulnerability through theeyes of the vulnerable: climate change induced uncertainties and Nepalsdevelopment predicaments. Institute for Social and Environmental Transition(ISET), Kathmandu, Nepal.

Nguyen, V.T.V., 2005. Downscaling methods for evaluating the impacts of climatechange and variability on hydrological regime at basin scale. Role of watersciences in transboundary river basin management, Thailand, 2005.

Randall, D.A., Wood, R.A., Bony, S., Colman, R., Fichefet, T., Fyfe, J., et al., 2007.Cilmate models and their evaluation. In: Climate Change 2007: The PhysicalScience Basis. Contribution of Working Group I to the ARIV of IPCC. CambridgeUniversity Press, Cambridge, United Kingdom and New York, NY, USA.

Semenov, M.A., Barrow, E.M., 1997. Use of stochastic weather generator in thedevelopment of climate change scenarios. Clim. Change 35, 397414.

Semenov, M.A., Stratonovitch, P., 2010. The use of multi-model ensembles fromglobal climate models for impact assessments of climate change. Clim. Res. 41,

rology 513 (2014) 422434 433114.Shrestha, A.B., Aryal, R., 2011. Climate change in Nepal and its impact on Himalayan

glaciers. Reg. Environ. Change 11, 6577.

Shrestha, A.B., Wake, C.P., Mayewski, P.A., Dibb, J.E., 1999. Maximum temperaturetrends in the Himalaya and its vicinity: an analysis based on temperaturerecords from Nepal for the period 197194. J. Clim. 12, 27752786.

Shrestha, A.B., Wake, C.P., Dibb, J.E., Mayewski, P.A., 2000. Precipitation uctuationsin the Nepal Himalaya and its vicinity and relationship with some large scaleclimatological parameters. Int. J. Climatol. 20, 317327.

Stott, P.A., Kettleborough, J.A., 2002. Origins and estimates of uncertainty inpredictions of twenty rst century temperature rise. Nature 416, 723726.

Sun, F., Roderick, M.L., Farquhar, G.D., 2012. Changes in the variability of global landprecipitation. Geophys. Res. Lett. 39, L19402. http://dx.doi.org/10.1029/2012GL053369.

Vincent, L.A. et al., 2011. Observed trends in indices of daily and extremetemperature and precipitation for the countries of the western Indian Ocean,19612008. J. Geophys. Res. 116, D10108. http://dx.doi.org/10.1029/2010JD015303.

WWF, 2009. Climate Change Impact on Discharge at Koshi River Basin usingRegional Climate and Hydrological Models. In: Final Report on a CollaborativeProject of WWF-Nepal/DHM.

Xu, H., Taylor, R.G., Xu, Y., 2011. Quantifying uncertainty in the impacts of climatechange on river discharge in sub-catchments of the Yangtze and Yellow RiverBasins, China. Hydrol. Earth Syst. Sci. 15, 333344.

Yatagai, A., Arakawa, O., Kamiguchi, K., Kawamoto, H., Nodzu, M.I., Hamada, A.,2009. A 44-year daily gridded precipitation dataset for Asia based on a densenetwork of rain gauges. Sci. Online Lett. Atmos. 5, 137140.

Yatagai, A., Kamiguchi, K., Arakawa, O., Hamada, A., Yasutomi, N., Kitoh, A., 2012.APHRODITE: constructing a long-term daily gridded precipitation dataset forAsia based on a dense network of rain gauges. Bull. Am. Meteorol. Soc. 93,14011415.

Zickfeld, K., Knopf, B., Petoukhov, V., Schellnhuber, H.J., 2005. Is the Indian summermonsoon stable against global change? Geophys. Res. Lett. 32, L15707.

434 A. Agarwal et al. / Journal of Hydrology 513 (2014) 422434

Analysis of future precipitation in the Koshi river basin, Nepal1 Introduction2 Study area and data2.1 Study area2.2 Observed precipitation data2.3 Gridded precipitation data2.4 Future precipitation data

3 Methods3.1 Downscaling3.2 Uncertainty due to GCMs and SRES scenarios3.3 Precipitation indices

4 Results and discussion4.1 Calibration and validation of LARS-WG4.2 Precipitation projections4.3 Uncertainty in annual and seasonal precipitation4.4 Precipitation indices

5 ConclusionsAcknowledgementsReferences