HYDROLOGICALPROCESSES Hydrol.Process.(2013) PublishedonlineinWileyOnlineLibrary (wileyonlinelibrary.com)DOI:10.1002/hyp.10005 Application and evaluation of a snowmelt runoff model in the Tamor River basin, Eastern Himalaya using a Markov Chain Monte Carlo (MCMC) data assimilation approach Prajjwal K. Panday,1* Christopher A. Williams,2 Karen E. Frey2 and Molly E. Brown3 1WoodsHoleResearchCenter,149WoodsHoleRoad,Falmouth,MA,02540,USA 2 GraduateSchoolofGeography,ClarkUniversity,950MainStreet,Worcester,MA,01610,USA 3 BiosphericSciences,NASAGoddardSpaceFlightCenter,Greenbelt,MD,20771,USA Abstract: Previousstudieshavedrawnattentiontosubstantialhydrologicalchangestakingplaceinmountainouswatershedswherehydrologyis dominatedbycryosphericprocesses.Modellingisanimportanttoolforunderstandingthesechangesbutisparticularlychallengingin mountainousterrainowingtoscarcityofgroundobservationsanduncertaintyofmodelparametersacrossspaceandtime.Thisstudy utilizesaMarkovChainMonteCarlodataassimilationapproachtoexamineandevaluatetheperformanceofaconceptual,degree-day snowmeltrunoffmodelappliedintheTamorRiverbasinintheeasternNepaleseHimalaya.Thesnowmeltrunoffmodeliscalibrated usingdailystreamflowfrom2002to2006withfairlyhighaccuracy(averageNash–Sutcliffemetric~0.84,annualvolumebias<3%). TheMarkovChainMonteCarloapproachconstrainstheparameterstowhichthemodelismostsensitive(e.g.lapserateandrecession coefficient) and maximizes model fit and performance. Model simulated streamflow using an interpolated precipitation data set decreases the fractional contribution from rainfall compared with simulations using observed station precipitation. The average snowmeltcontributiontototalrunoffintheTamorRiverbasinforthe2002–2006periodisestimatedtobe29.7±2.9%(whichincludes 4.2±0.9% from snowfall that promptly melts), whereas 70.3±2.6% is attributed to contributions from rainfall. On average, the elevation zone inthe 4000–5500m range contributes the mosttobasinrunoff,averaging 56.9±3.6% of all snowmelt input and 28.9±1.1% of all rainfall input to runoff. Model simulated streamflow using an interpolated precipitation dataset decreases the fractional contribution from rainfall versus snowmelt compared with simulations using observed station precipitation. Model experimentsindicatethatthehydrographitselfdoesnotconstrainestimatesofsnowmeltversusrainfallcontributionstototaloutflow butthatthisderivesfromthedegree-daymeltingmodel.Lastly,wedemonstratethatthedataassimilationapproachisusefulfor quantifyingandreducinguncertaintyrelatedtomodelparametersandthusprovidesuncertaintyboundsonsnowmeltandrainfall contributionsinsuchmountainouswatersheds.Copyright©2013JohnWiley&Sons,Ltd. Supportinginformationmay be foundintheonline versionofthis article. KEYWORDS snowmelt; hydrology;SRM; APHRODITE; MODISsnow; Himalaya;MCMC;data assimilation Received 11January2013; Accepted 6August 2013 INTRODUCTION tropospherictemperaturesfromthelongestavailablerecord of microwave satellite measurements reveals widespread Growingevidenceindicatesthatthelow-latitudemountain- pre-monsoonal and mean annual warming over the HKH ous cryosphere is undergoing change owing to recent region at a rate of 0.21±0.08°C/decade (Gautam et al., warming(IPCC,2007;Yaoetal.,2012).Geographicareas 2010). Regional climate projections by the Intergovern- wherethewatercycleisdominatedbysnowmelthydrology mental Panel on Climate Change (IPCC, 2007) indicate areexpectedtobemoresusceptibletoclimatechange,asit CentralAsiamaywarmbyamediantemperatureof3.7°C affectstheseasonalityofrunoff(Barnettetal.,2005;Adam by the end of the 21st century, with warming over high etal.,2009).TheHinduKush-Himalaya(HKH),regarded altitudesacrosstheHKH(Pandayetal.,inreview).Thereis asthe‘watertowers’ofAsia,isonesuchcriticalregionasit an increasingly large body of evidence of a glacio- nourishes several of the major Asian watersheds through hydrological response along the east–west transect of the perennialriverssuchastheGanges,IndusandBrahmaputra HKH corresponding to the these climatic changes, with (Messerli et al., 2004). Most recent research analyzing glaciers in the Eastern Himalaya exhibiting retreat and negative mass balance and glaciers in the Karakoram and northwestern Himalaya exhibiting a positive mass balance *Correspondenceto:PrajjwalK.Panday,WoodsHoleResearchCenter, over the last few decades (Bolch et al., 2012; Yao et al., 149WoodsHoleRoad,Falmouth,MA,02540,USA. E-mail:[email protected] 2012).Suchclimate-drivenresponsesofmountainousriver Copyright©2013JohnWiley&Sons,Ltd. P.K.PANDAYETAL. hydrology may pose significant challenges for this region. observationstoconstrainmodelswiththegoalofproviding Despiteitsregionalimportanceandvulnerabilitytoclimate better understanding of the underlying processes change, there is uncertainty associated with the physical (McLaughlin et al., 2005). The Bayesian approach in processes that control runoff generation at high altitudes hydrologicalapplicationsallowstosystematicallyincorpo- (Bolch et al., 2012; Pellicciotti et al., 2012) and subse- rate prior information on model parameters in the data quently with the rates and magnitude of climate change assimilationframework(Zobitzetal.,2011)andtocompute impactsonsnowcoverandsnowmelthydrology. probabilitydistributionsofmodelledoutputsandcharacter- Water resource management and the evaluation of ize uncertaintyinthe model,usuallythroughMonteCarlo impacts of climatic change require quantification of simulation (Mishra, 2009). This provides a means of streamflow variability. Hydrological models provide a objectively and quantitatively diagnosing the parameters framework to investigate these relationships and subse- andprocessestowhichamodelismostsensitiveandfrom quently use a scenario approach to evaluate projections of whichonecanlearn.Fewstudieshaveemphasizedtheneed hydrologicalvariables(Maureretal.,2009).Theassessment for parameterization in the Himalayan region and have of hydrological impacts of climate change is particularly performed extensive calibration (Konz et al., 2007; challenginginmountainousenvironmentsastheyconstitute Immerzeel et al., 2011; Pellicciotti et al., 2012) and complex hydrological systems owing to extreme heteroge- validation with additional data sets such as glacier mass neity in vegetation, soils, topography and spatially and balancesandremotelysensedsnowcover.Pellicciottietal. temporallyvaryingsnowcoverandsnowmeltpatterns(Gurtz (2012) also emphasized that an assessment of the internal et al., 2005; Panday et al., 2011). Continued efforts in consistencyofthemodelisimportantowingtotheproblem hydrologicalmodellingofsnowandglaciermeltintheHKH ofmultiplicityofparametersetsorequifinality. regionmostlythroughconceptuallylumped,semi-distributed In addition to parameter uncertainty, sensitivity of (ratherthanphysicallybasedmodels)haveprovidedbaseline hydrological models to input data such as precipitation patternsofspatiotemporalpatternsofprecipitationandrunoff also needs to be evaluated particularly in complex at very large scales (Bookhagen and Burbank, 2010; topography and high elevations. Hydrological models in Immerzeel et al., 2010). However, poor accessibility and mountainousregionswithscarceobservationsareusually scarcityofadequatenetworkofhydrometeorologicalstations forced with precipitation data located at lower altitudes, at high altitude regions still remain a major impediment to which can introduce added uncertainty. Satellite-derived runoff modelling. As a result, only a few studies have precipitation estimates and interpolated rain-gauge obser- exploredsnowmeltrunoffmodelsatthecatchmentorriver- vations have been used in other studies as inputs to a basinscaleinthisregion(Braunetal.,1993;Shresthaetal., snowmelt runoff model (SRM) instead of gauged precip- 2004;Akhtaretal.,2008;Tahiretal.,2011). itation. Bookhagen and Burbank (2010) used improved Parameterizationisusuallychallengingowingtoscarcity Tropical Rainfall Measuring Mission rainfall estimates to of data in mountainous regions such as the HKH, and a characterize the spatiotemporal distribution of rainfall, soundunderstandingofthesnowmelthydrologyneedstobe snowfall and evapotranspiration across the Himalayan emphasized prior to assessment of water resources in a region, whereas Tahir et al. (2011) successfully used changing climate. Several catchment-scale and basin-scale interpolateddailyprecipitationdatatosimulatesnowmelt modelling studies have addressed the glacio-hydrological hydrologyintheHunzaRiverbasin. processes in the Himalayan region (Konz et al., 2007; A thorough assessment of hydrologic processes, Akhtaretal.,2008;ButtandBilal,2011;Immerzeeletal., patternsandtrendsrequiresminimization ofuncertainties 2011;Tahiretal.,2011).Amajorityofthesestudieshave in modelling and available input data. The objective of adopted parameter values from the literature without thisstudyistoexaminespatialandtemporalvariationsin rigorouscalibrationandwithoutquantifyinghowparameter snowmelt contributions to runoff in the Hindu Kush- uncertaintyinfluencesmodelledresults.Transferofparam- HimalayanregionandTamorRiverbasinbyusingadata etersacrossspaceandtimeinhydrologicalmodellingcanbe assimilation approach that both combines available input problematic,particularlyinconceptualmodelsofsnowmelt data with a degree-day based SRM and estimates model whenparameterseitherdonothaveaphysicalmeaningor parameters by using a Markov Chain Monte Carlo are not easily measurable (Sun et al., 2012). In order to (MCMC) technique. Additionally, we investigate and addresstheissuesofparameterizationandmodeluncertain- evaluate the performances of the SRM when driven by ty, data assimilation methods are typically applied in observed precipitation as well as interpolated gridded hydrology to integrate different datasets, combine model precipitation data. We demonstrate how a data assimila- outputs with observations, update hydrological models, tion approach can be coupled with a SRM to investigate quantifymodeluncertaintiesandsensitivityand/orimprove the issues of model parameter sensitivity and uncertainty modelaccuracy(BevenandFreer,2001;Clarketal.,2006). and improve performance in a data-scarce basin in the Data assimilation can be viewed as a method to use Eastern Himalaya. Copyright©2013JohnWiley&Sons,Ltd. Hydrol.Process.(2013) APPLICATIONANDEVALUATIONOFASNOWMELTRUNOFFMODELINTHEHIMALAYA METHODS Study area This geographic domain of this study is the upstream portionoftheTamorRiverbasinintheEasternHimalaya of Nepal (Figure 1). The Tamor basin covers an area of 4274km2, with an elevation range of 422 to 8505m (Figure 2). The Tamor River, which is one of the major tributaries of the Sapta Koshi River, flows between high peaks such as Kanchenjunga (8586m) and Everest (8848m) on gorges reaching depths of 2000m. Precip- itation for the region is highly seasonal and concentrated duringthesummermonsoonmonths(JunetoSeptember) (Figure 3). The upper reaches of the basin are snow covered and glaciated, contributing melt water to streamflow. Figure2. Area-elevationcurvefortheTamorRiverbasin Hydrometeorological data acquired from the Consultative Group on International Hydrometeorological measurements for the Tamor AgriculturalResearch–ConsortiumforSpatialInformation. River basin in the Eastern Himalaya were acquired from Thebasinwasdividedintofourzones,andthezonalmean the Department of Hydrology and Meteorology, Nepal. hypsometricelevation(Figure2)wasusedastheelevation Daily temperature data at Taplejung station (27.35°N, towhichbasestationtemperatureswereextrapolatedforthe 87.67°E at 1732m elevation) and daily precipitation data calculationofzonaldegreedays(TableI). measured at Lungthung station (27.55°N, 87.78°E at 1780m elevation) in the Taplejung district were used for ModerateResolutionImagingSpectroradiometer(MODIS) the hydrological modelling (Figure 1). Discharge data measured daily from 2002–2006 for the Tamor River at snow cover Majhitarstation(27.15°N,87.71°Eat533melevation)were Satellite-derived snow cover is the most efficient and usedforcalibration.TheTamorbasinwasdelineatedatthe readily available input for snowmelt runoff modelling. Majhitar station using the 90m digital elevation data Snow cover was derived using the MODIS/Terra Snow Figure1. LocationandtopographyoftheTamorRiverbasinintheeasternNepaleseHimalaya Copyright©2013JohnWiley&Sons,Ltd. Hydrol.Process.(2013) P.K.PANDAYETAL. AsianPrecipitation–HighlyResolvedObservational DataIntegrationtowardsEvaluation(APHRODITE) precipitationdata Spatiallyrepresentativegriddedprecipitationdataarealso used to drive a snowmelt runoff model, as valley stations may not be representative for basin precipitation. The APHRODITEofthewaterresourcesprojecthascompileda daily gridded precipitation dataset covering 1951–2007 periodbyanalyzingrain-gaugeobservationdataacrossAsia (Yatagaietal.,2009).TheAPHRODITEdatausedinthis research (APHRO_MA_V1003R1) have a 0.25×0.25° gridded spatial resolution, which was interpolated from meteorological stations throughout this region. Whereas satellitedataconsistentlyunderestimateoroverestimatethe amountofprecipitation,particularlyinthemonsoonseason Figure 3. Average monthly streamflow (m3/s) and total monthly (Duncan and Biggs, 2012), APHRODITE data have been precipitation(mm)from1996to2006atMajhitarstation shown to provide the best precipitation estimates in most regionsexceptathighelevationswhereaccuracyofthisdata setislimited(Andermannetal.,2011).Weusedthebasin- TableI. Elevationzone,elevation range,mean elevationand averaged APHRODITE precipitation as input to the SRM zonalarea forthe four zonesof theTamor River basin assumingittobeconstantacrosselevationzones.Themodel Elevation Meanelevation Zonal wasrunusingAPHRODITEprecipitationdatatodetermine Elevationzone range(m) ofzone(m) area(km2) thereliabilityofthedatasetforsnowmeltrunoffmodelling intheTamorRiverbasin. 1 533–2500m 1685.9 1355.5 2 2500–4000m 3208.1 1111.3 3 4000–5500m 4764.3 1297.9 Snowmelt runoff modelling 4 >5500 5988.8 509.2 Modelling of stream discharge was based on the SRM Totalbasin — — 4273.8 developed by Martinec et al. (2008). SRM is a conceptual, deterministic, degree-day hydrological model that simulates and forecasts daily runoff resulting from snowmelt and Cover8-DayL3Global500mGrid(MOD10A2;Halletal., precipitationfromhydrometeorologicalinputdata.Themodel 2006)andupdatedeveryeightdaysusingasnowdepletion was designed for streamflow modelling in basins where curve.TheMODISsnowcoverisbasedonasnowmapping algorithmthatusesat-satellitereflectanceinMODISbands snowmelt is a major contributor, and more recently, it has 4(0.545–0.565μm)and6(1.628–1.652μm)tocomputethe beenappliedforevaluationofimpactsofclimatechangeon seasonal snow cover and runoff (Immerzeel et al., 2010). Normalized Snow Difference Index (Hall et al., 2002). A SRM has been applied to basins ranging from 0.76 to pixel will be mapped as snow if the Normalized Snow Difference Index≥0.4 and MODIS band 2 (0.841– 917444km2andacrosselevationrangesfrom0to8840min 0.876μm) >11%; however, if the MODIS band 4 over 100 basins in 29 different countries (Martinec et al., reflectanceis<10%,thepixelwillnotbemappedassnow 2008). This model was chosen in this study to assess snowmelthydrologyowingtoitsminimaldatarequirements even if the other criteria are met. The MODIS product andtransferabilitytoavarietyofgeographicregions.Onthe distributedthroughtheNationalSnowandIceDataCentre basisoftheinputofdailytemperature,precipitationandsnow- was acquired for the scene h25v06, covering the study coveredareavalues,SRMcomputesthedailystreamflowas region from 2002 to 2006. The MODIS images were reprojected to Universal Transverse Mercator zone 45 projectionsystem,andtheTamorRiverbasinareawasthen Qnþ1 ¼∑½CsznaznðTnþΔTznÞSzn (1) extracted toassessthe fraction ofsnowcover for different altitudinal zones. Average snow cover was estimated by þCrzn PznÞAzð10000=86400Þð1(cid:2)knþ1Þ interpolating linearly between the previous and next available cloud free images for days with cloud cover þQn knþ1; following Tahir et al. (2011). The conventional depletion knþ1 ¼xQyn; (2) curveswerederivedforeachelevationzoneusingthe8-day snowcoverdata,whichprovidetheinputtotheSRMasthe where Q (m3/s) is discharge at day n+1, C is the runoff decimalportionofsnow-coveredareaforeachday. coefficienttosnowmelt(Cs )andtorain(Cr )foreachzone n n Copyright©2013JohnWiley&Sons,Ltd. Hydrol.Process.(2013) APPLICATIONANDEVALUATIONOFASNOWMELTRUNOFFMODELINTHEHIMALAYA (subscript z), a (cm/°C/day) is the degree-day factor, toaccountingforlossesviarunoffcoefficients,introducing T+ΔT(°C)arethedegreedays,P(cm)istheprecipitation animplicitassumptionthatfractionalcontributionstorunoff perday,A(km2)istheareaofthebasin,Sisthefractional areproportionaltotherelativeabundanceofeachinput. snowcoverandkisthedischargerecessioncoefficient.The Equation(1)isappliedtoeachzoneofthebasin,andtotal daily average streamflow on day n+1 is computed by discharge iscomputedas sumofallzonal discharges.The addition of snowmelt and precipitation that contributes to SRM structure as applied inthis study assumes a timelag runoff with the discharge on the preceding day. The between the daily temperature cycle and the resulting snowmelt of the preceding day is a product of the degree- dischargecycleof18h.Theuseofthistimelagissupported day factor (a), zonal degree days (T+ΔT) and the snow- bytheexpectedrelationbetweenlagtimeandbasinsizeas coveredareapercentage(S).Thesnowmeltrunoffcoefficient shown in a SRM intercomparison report by the World (Cs) and the zonal area (A) are further multiplied with the MeteorologicalOrganization(1986).Althoughthetimelag previousproducttocomputethepercentagethatcontributes parameter can be adjusted to improve synchronization of to runoff. Similarly, the product of the rainfall runoff simulated and observed peaks of average daily flows, coefficient (Cr) and the zonal area (A) determines the calibration of the recession coefficients provides a similar precipitationcontributiontorunoff.Therecessioncoefficient effect (Martinec et al., 2008). The parameters for runoff (k)isalsoanimportantparameterasitdescribesthestructure coefficients, lapse rate, critical temperature, recession of the falling limb of the hydrograph and correspondingly coefficient and degree-day factors reflect characteristics at determinestheproportionofdailymeltwaterproductionthat abasinscaleandaredeterminedinthisstudywithanMCMC appears immediately in the runoff. This is obtained with dataassimilationbymaximizingthefitbetweentheobserved analysis of historical discharge data based on Equation 2, and modelled streamflows. The modelled streamflow is whereparametersxandyarederivedbyanalyzingdischarge assessedforaccuracyusingtheNash–Sutcliffe(NS)statistic onagivendayasafunctionofthevalueonthepreviousday. and the volume difference (D ), where volume difference v A critical temperature is used to determine whether a demonstrates bias. (Supplementary information, A) precipitationeventisrainorsnowoverdifferentpartsofthe (Martinecetal.,2008). basin. When precipitation is determined to be snow, its contributiontorunoffdependsonwhetheritfallsonasnow- Model calibration using MCMC optimization covered or previously snow-free area. Snowfall on snow- AdataassimilationtechniqueknownasaMCMCroutine covered areas is assumed to become part of the seasonal (Zobitz et al., 2011) was used to explore the parameter snowpack, its depth is not tracked further, and its spacesintheSRM.TheapplicabilityofBayesianmethods contributiontorunoffisdeterminedbythesnowdepletion inhydrological studieshasincreased owing totheiruse in curve. Daily snowmelt depth from snow-covered areas is parameter estimations and uncertainty assessments (Smith determined on the basis of degree days and a factor that and Marshall, 2008). The MCMC algorithm is one of convertsthistomassrelease,qualitativelyrepresentingthe severaldataassimilationtechniquesandusestheMetropolis maximumpotentialmeltgiventheenergyinput.Incontrast, Hastings algorithm (Metropolis et al., 1953), which snow falling as precipitation over snow-free areas is iteratively adjusts the model parameters to yield the best accumulated until the next day warm enough to produce matchbetweentheobservedandmodelledstreamflow.The melting at which time it is depleted on the basis of the output provides a set of accepted parameter values and degree-days approach. During the winter half year information about the posterior distributions of the (October–March), the model cancels any existing storage parameters consistent with the prior information, model of preceding temporal snowfalls when snow cover is structureandobservations.AdditionaldetailsoftheMCMC maximum such that snow on previously snow-free areas algorithmusedforoptimizationaredescribedmorefullyin becomes part of the seasonal cover. When precipitation is Supplementary information, B and in Zobitz et al. (2011) consideredtoberainduringthecoldseason,rainfallrunoff andBraswelletal.(2005). contributestototalrunoffonlyfromthesnow-freearea,and Basedontheliteratureandrangeofparametervaluesused therainfalldepthisreducedbytheratioofsnow-freeareato in other snowmelt runoff models in the HKH region zonalarea.Atalaterstageduringtheseason,whichissetto (Immerzeeletal.,2010; ButtandBilal,2011; Tahiret al., April 1, rain falling on snow-covered areas is allowed to 2011), we first specified prior ranges for the parameters. fully contribute to runoff such that rain from entire zonal TableIIshowstheparameterswiththepriorrangesthatwere area is accounted. The SRM approach estimates fractional assigned seasonally or annually across different elevation contributions to total discharge from (1) daily snowmelt zones. Seasonally varying parameters were optimized for from the snowpack for each elevation, (2) the melt of the snowmelt and monsoon period (June–August) and the precipitation delivered as snow but to areas described as snow accumulation and dry period (September–May). snow-free at the time of snowfall and (3) zonal rainfall. Snowmelt and rainfall runoff coefficients were varied Contributionsofeachtototalriverflowareestimatedprior seasonallyandbyelevationsinformedbyTahiretal.(2011) Copyright©2013JohnWiley&Sons,Ltd. Hydrol.Process.(2013) P.K.PANDAYETAL. TableII. Initial parameter rangesusedfor MarkovChainMonte Carlodata assimilation Parameterranges across eachelevation zone Parameter Zone1 Zone2 Zone 3 Zone 4 (533–2500m) (2500–4000m) (4000–5500m) (>5500m) Snowrunoff coefficient June–August 0.3–0.8 0.3–0.8 0.3–0.8 0.3–0.8 September–May 0.3–0.8 0.3–0.8 0.3–0.8 0.3–0.8 Rainfall runoffcoefficient June–August 0.3–0.8 0.3–0.8 0.3–0.8 0.3–0.8 Sep–May 0.3–0.8 0.3–0.8 0.3–0.8 0.3–0.8 Degree-day factor(cm/°C/day) Annual 0.3–0.6 0.3–0.6 0.5–0.9 0.5–0.9 Lapserate (°C/100m) Annual 0.5–0.7 Critical temperature(°C) Annual 0.0–2.0 xcoefficient Annual 0.9–1.4 ycoefficient Annual 0.0–0.25 SnowmeltandmonsoonalmonthsareJune–August,andSeptember–Mayisthesnowaccumulationanddryperiod.Lapserate,criticaltemperatureandx andycoefficientsarenotvariedacrosselevationzones. andwereassignedequalpriorparameterranges.Although Model sensitivity and uncertainty degree-day factors tend to vary with different seasons, In addition to model calibration, we also carried out climate and surfaces, under similar conditions they are several experimental runs. The model was allowed to expectedtoincreasewithincreasingelevation(Hock,2003). calibrate via MCMC by setting the snowmelt runoff Similar to previous studies (Butt and Bilal, 2011; Tahir coefficients to zero such that snowmelt contributions were etal.,2011),whichutilizeadegree-dayfactorthatincreases neglectedinthesimulationofthehydrograph.Similarly,the with elevation, we also increase the parameter range rainfall runoff coefficients were also set to zero such that graduallyacrosselevationzones.Theelevationzonegreater rainfallcontributionswereexcludedinthefinalsimulationof than5500m,whichareglacierandicecoveredareassumed the hydrograph. In addition to these experiments, we tohaveahigherdegree-dayfactorasthevaluestendtobe conducted additional sensitivity and uncertainty analyses considerably higher for ice compared with snow (Hock, viaensemblestreamflowsimulations(1000runs)toidentify 2003). Parameters such as lapse rate, critical temperature parametersthatcontrolmodelperformanceanduncertainties and x and y coefficients were calibrated annually. Other in modelled streamflow and total input contributions. studies using SRM have used a similar variation of Parameter sensitivity and uncertainty in streamflow in the parameters.Althoughmostparameterscanpotentiallyvary time history were analyzed as 5% and 95% confidence across elevations and seasons, increasing parameter sets intervals of the ensemble runs. First, we generated an increase potential ranges of parameter values and thereby ensembleofmodelruns(‘bestensemble’hereafter)fromthe increase uncertainty (Seibert, 1997). For this study, we posteriordistributionofacceptedparametersetsfromMCMC implementedtheMCMCtechniqueusingthreechains,each that represent the most likely set of parameters. Second,the ofwhichconsistedof10000iterations.Thebestparameter sensitivity to parameters was tested by perturbing the valuesfromthesechainswereusedinthefinalsetof10000 distribution of a parameter by some amount and computing iterations, which is used to determine the parameter thecorrespondingchangesinmodeloutputwhilepreserving distributions. The range of parameter values was deliber- thedistributionofotherparameters.Thisisperformedinthis atelysettobelargeatfirsttoexamineifrealisticparameter study by randomly drawing from the altered distribution of values were selected by the MCMC technique. If the each parameter (e.g. ±10% from the mean value of a posterior parameter distributions were less plausible, the parameter from the posterior distribution) to generate parameter ranges were logically constrained by literature ensemble runs. This allows quantifying the effects of each valuestobemorerealistic.Thesnowmeltandrainfallrunoff parameter or a combination of parameters on model output coefficients were constrained in the ranges specified in and on total input contributions to streamflow by providing Table II based on Tahir et al. (2011) and Butt and Bilal uncertaintyboundsontheoutputstatistics.Third,anensemble (2011),degree-dayfactorrangeswerespecifiedonthebasis ofparametersetswasgeneratedbyrandomlysamplingfrom ofHock(2003),andpriorrangesforlapserateandcritical the cumulative distribution curve constructed from the temperaturewerespecifiedfromTahiretal.(2011)andButt posteriordistributionofeachparameter.Theensemblemodel andBilal(2011).TheSRMwasrunviaMCMCwithprior simulation uses the full range of parameter variations and parameterrangesasspecifiedinTableIIforthehydrological combinationsandtherebyprovidesoveralluncertaintiesinthe years2002to2006. choiceofmodelparametersgeneratedfromMCMC. Copyright©2013JohnWiley&Sons,Ltd. Hydrol.Process.(2013) APPLICATIONANDEVALUATIONOFASNOWMELTRUNOFFMODELINTHEHIMALAYA RESULTS maximum snow cover of ~40% with summer zonal snow coveragefallingto~12%.Elevationsupto4000mhadon Summary of meteorological parameters average~9%annualmaximumsnowcover. Long term annual temperatures (1970–2006) at the Taplejungstationaveraged16.3°C,whereasaveragewinter Model parameterization and summer temperatures were 10°C and 21.3°C, respectively. Mann–Kendall tests indicate significant Whenthepriorrangesoftheparametersweredeliberately (p<0.05) positive trends in both annual maximum and set to be large, the model calibrated well with a good fit average temperatures for the Taplejung station during this between the observed and modelled streamflow; however, period. Average annual precipitation at the Lungthung thisalsoresultedinseveralmodelswithsimilarperformance station was ~2484mm for the 1996–2006 period. On and physically implausible coefficients. Logically cons- average,approximately76%oftheprecipitationisreceived training the prior parameter ranges provided models with during the summermonsoon.Themonthly streamflowfor betterfitandparametersthatwerephysicallyplausiblewhen theTamorRiverbasinattheMajhitarstationalongwiththe varied across seasons and elevations. The posterior totalmonthlyprecipitationatLungthungstationindicatesa distributions of parameters after the MCMC optimization fingerprint of monsoonal precipitation in the Eastern forallparametersforthe2002–2003hydrologicalyearsare Himalaya from June to September (Figure 3). Average showninFigure5.Whenthemodelwasoptimizedusingthe streamflow measured at the Majhitar station averaged MCMC approach, some of the most sensitive parameters 256.2m3/s annually during the same period, whereas the (suchasxandycoefficients,criticaltemperatureandlapse monsoon months had an average streamflow of 573m3/s. rate) exhibited a narrow, unimodal distribution with their Snow cover depletion curves for the Tamor River basin ranges significantly reduced from the initial parameter indicatethatthehighestelevationzone(above5500m)had ranges. The x and y coefficients required to compute the a maximum ~85% snow-covered area for the 2002–2006 recessioncoefficientwerethemostsensitiveparameters,as period, dropping to ~73% during summer (Figure 4). The they determinedthe structureof hydrograph recessionand elevation zone between 4000 and 5550m had an annual correspondingproportionofdailymeltwaterthatappeared (a) (b) (c) Figure4. SnowcoverdepletioncurvesfromtheModerateResolutionImagingSpectroradiometer8-daymaximumsnowcover500-mresolutionproductinthe TamorRiverbasinfrom2002to2006in(a)zone2,elevationrangefrom2500–4000m,(b)zone3,elevationrangefrom4000–5500mand(c)zone4,elevation rangegreaterthan5500m. Copyright©2013JohnWiley&Sons,Ltd. Hydrol.Process.(2013) P.K.PANDAYETAL. Figure5. Posteriorparameterdistributionsforthe2002–2003hydrologicalyearsashistograms immediately in the runoff. This shows that the MCMC that contributes to runoff and are expected not to vary as approach performs well in constraining the most sensitive much across types of precipitation. Some of the observed parameters of the model. Once a narrow posterior differencesinthesecoefficientswithinasinglezonecould distribution is established for the sensitive parameters bearesultofthetrade-offamongparametersafterthemost during the MCMC approach, changing other parameters sensitiveparametershavebeenconstrainedwell.Inadequate does not result in a significant increase in model precipitation data from higher elevations zones in the performance.Therefore,someoftheremainingparameters mountainous regions further hindered our ability to suchassnowmeltandrainfallrunoffcoefficientsexhibited constrain the runoff coefficients. Parameter estimates tend broadposteriordistributions,andlargestandarddeviations to be stable across years, though Tcrit showed large where good simulations of the observed streamflow were variation, but also had a large spread. Degree-day factor obtained over broad ranges (Figure 5 and Table III). The estimatestendedtobethehighestforelevationsgreaterthan posteriordistributionsofrunoffcoefficientsforbothrainfall 4000m (zones 3 and4), which had the highestfraction of andsnowmeltarecomparableacrosstypesofprecipitation snow-coveredarea.Thepeaksinposteriordistributionsfor and all simulated years. These coefficients represent the the degree-day factor correspondingly rise from zone 1 to fraction of liquid water, from either snowmelt or rainfall, zone3elevationranges. Copyright©2013JohnWiley&Sons,Ltd. Hydrol.Process.(2013) APPLICATIONANDEVALUATIONOFASNOWMELTRUNOFFMODELINTHEHIMALAYA Table III. Meanandstandard deviation ofparameters from theposterior distribution Meanandstandard deviation ofparameteracross zones Parameter Zone 1 Zone2 Zone3 Zone 4 Year (533–2500m) (2500–4000m) (4000–5500m) (>5500m) Cs June–August 2002 0.55±0.15 0.59±0.13 0.72±0.07 0.56±0.13 2003 0.46±0.13 0.44±0.11 0.47±0.12 0.52±0.15 2004 0.54±0.14 0.54±0.14 0.68±0.11 0.55±0.14 2005 0.55±0.14 0.55±0.15 0.69±0.11 0.55±0.15 September–May 2002 0.58±0.14 0.61±0.11 0.59±0.13 0.54±0.15 2003 0.60±0.14 0.56±0.13 0.71±0.07 0.56±0.15 2004 0.70±0.09 0.71±0.07 0.67±0.11 0.54±0.14 2005 0.59±0.14 0.43±0.11 0.40±0.0/8 0.54±0.14 Cr June–August 2002 0.73±0.05 0.74±0.05 0.69±0.09 0.55±0.15 2003 0.67±0.11 0.56±0.13 0.54±0.15 0.54±0.14 2004 0.71±0.05 0.68±0.09 0.59±0.14 0.55±0.14 2005 0.76±0.04 0.74±0.08 0.76±0.05 0.56±0.14 September–May 2002 0.69±0.07 0.68±0.09 0.65±0.10 0.55±0.14 2003 0.37±0.07 0.43±0.09 0.40±0.08 0.54±0.14 2004 0.71±0.07 0.70±0.08 0.54±0.14 0.56±0.14 2005 0.69±0.07 0.70±0.08 0.72±0.07 0.51±0.13 a Annual 2002 0.48±0.08 0.50±0.0 0.61±0.08 0.68±0.102 2003 0.42±0.08 0.41±0.08 0.79±0.0 0.69±0.11 2004 0.53±0.05 0.53±0.06 0.81±0.05 0.70±0.`2 2005 0.45±0.09 0.44±0.09 0.80±0.08 0.71±0.12 LR Annual 2002 0.64±0.01 2003 0.51±0.01 2004 0.68±0.01 2005 0.54±0.01 Tcrit Annual 2002 1.26±0.28 2003 1.16±0.49 2004 1.83±0.13 2005 0.94±0.62 x coefficient Annual 2002 1.34±0.03 2003 1.26±0.01 2004 1.29±0.02 2005 1.11±0.01 y coefficient Annual 2002 0.08±0.01 2003 0.06±0.01 2004 0.07±0.01 2005 0.03±0.01 Cs,snowmeltrunoffcoefficient;Cr,rainfallrunoffcoeficient;a,degreedayfactor;LR,lapserate. The SRM is usually calibrated manually where known months. Across all simulated years, the observed and ranges ofparameter values aremaintained while changing modelled streamflow show a linear association with some otherparametersbytrialanderror.Thistypeofcalibration biasduringhighdischargeperiods[Figure6(b,d,fandh)]. can be inefficient in exploring the parameter space and ThemodelsimulatedstreamflowshaveahigherNSstatistic resulted in an average NS statistic of 0.64 and volume in the years where the summer peak discharge is lower differenceof10%acrossallsimulatedyearsforthisstudy. relativetootheryears(TableIVandFigure6). ModelcalibrationusingMCMCoptimizationhadagreater overall improvement with NS statistic of ~0.84 averaged Model sensitivity and uncertainty acrossallsimulatedyears.Figure6(a,c,eandg)showsthe Theresultsoftheexperimentrunsindicatethatwiththe modelledstreamflowagainstobservedstreamflowusingthe snowmelt runoff coefficients (Cs) set to zero, the model MCMC approach for four hydrological years (utilizing optimized as well as when Cs values were also included observedprecipitationattheLungthungstation).TheSRM through the MCMC technique. The rainfall runoff is able to capture the daily streamflow in the annual coefficients (Cr) are parameterized to higher values, and hydrologyofthebasin;however,themodelunderestimates themodelisabletosimulatethehydrographbyoptimizingx high discharge events, particularly during the monsoonal andycoefficientswithmodelperformanceequivalenttothat Copyright©2013JohnWiley&Sons,Ltd. Hydrol.Process.(2013) P.K.PANDAYETAL. Figure6. TimeseriesandscatterplotsofobservedandmodelledstreamflowusingobservedprecipitationatLungthungstationfor2002–2003(a,b), 2003–2004(c,d),2004–2005(e,f)and2005–2006(g,h) withthebestestimatesofparametersasinTableIV.Thefirst accepted by the MCMC approach. Because the most ensembleparametersets(bestensemble)randomlyselected sensitive parameters have narrow, unimodal distributions, fromtheposteriordistributionoutputfromMCMCdidnot similarmodelfitandperformanceisachievedwithinabroad exhibit any significant variations in model fit and perfor- distribution of other parameters. However, when the mance. This is to be expected as the parameter sets are distribution of the most sensitive parameters was altered Copyright©2013JohnWiley&Sons,Ltd. Hydrol.Process.(2013)
Description: