Noah land surface model modifications to improve snowpackprediction in the Colorado Rocky Mountains

Michael Barlage,1 Fei Chen,1 Mukul Tewari,1 Kyoko Ikeda,1 David Gochis,1

Jimy Dudhia,1 Roy Rasmussen,1 Ben Livneh,2 Mike Ek,3 and Ken Mitchell3

Received 12 November 2009; revised 4 August 2010; accepted 17 August 2010; published 16 November 2010.

[1] Simulated snowpack by the Noah land surface model (LSM) shows an early depletiondue to excessive sublimation and too early onset of snowmelt. Tomitigate these deficiencies,five model modifications are tested to improve snowpack performance: (1) time‐varyingsnow albedo, (2) solar radiation adjustment for terrain slope and orientation, (3) reducing thesurface exchange coefficient for stable boundary layers, (4) increase of fresh snow albedo,and (5) adjusting surface roughness length when snow is present. The Noah LSM is executedfrom 1 November 2007 to 1 August 2008 for the headwater region in the Colorado RockyMountains with complex terrain, and its results are evaluated against 1 km Snow DataAssimilation System (SNODAS) output and individual Natural Resources ConservationService Snowpack Telemetry (SNOTEL) sites. The most effective way to improvemagnitude and timing of seasonal maximum snowwater equivalent (SWE) is the introductionof the time‐varying albedo formulation and the increase in fresh snow albedo. Minorimprovement is obtained by reducing nighttime sublimation through adjusting the stableboundary layer surface exchange coefficient. Modifying the surface roughness length oversnow surfaces and adding a terrain slope and orientation adjustment for radiation has littleeffect on average SWE conditions for the entire modeling domain, though it can have asignificant effect in certain regions. The net effect of all changes is to improve the magnitudeand timing of seasonal maximum SWE, but the snow period end is now somewhat toolong. Adding the terrain slope and orientation effects does have an effect on local surfaceenergy flux components depending on the cell slope and orientation.

Citation: Barlage, M., F. Chen, M. Tewari, K. Ikeda, D. Gochis, J. Dudhia, R. Rasmussen, B. Livneh, M. Ek, and K. Mitchell(2010), Noah land surface model modifications to improve snowpack prediction in the Colorado Rocky Mountains, J. Geophys.Res., 115, D22101, doi:10.1029/2009JD013470.

1. Introduction

[2] The snowpack evolution in the Rocky Mountains ofthe Western United States directly affects the hydrologicresources of the millions of people who rely on the ColoradoRiver. According to Edwards and Redmond [2005], on aver-age, 85% of the streamflow in the Colorado River comes fromsnowmelt in high mountain regions in the upper portion ofthe basin. However, the seasonal evolution of snowpack inthe Colorado River basin (i.e., accumulation and ablation)exhibits significant interannual variability. Furthermore, sev-eral portions of the western U.S. are presently exhibitingsignificant decreases in peak snowpack accumulation, earlierdates of peak runoff, longer frost‐free periods and lowersummertime streamflow values, presumably in response to

long‐term climate change [e.g., Easterling, 2002;Milly et al.,2005; Hamlet et al., 2007; Mote, 2006; Barnett et al., 2008;Hidalgo et al., 2009]. Although intermodel differences aresubstantial, the consensus of climate models participating inthe Fourth Assessment Report of the IntergovernmentalPanel on Climate Change [2007] climate model projectionssuggest that much of the western U.S. will experienceincreased warm season aridity due to higher temperaturesand evaporative demand [Hoerling et al., 2007; Seageret al., 2007; Hoerling and Eischeid, 2008]. Taking intoaccount rapidly increasing demand for Colorado River waterfrom fast‐growing population centers in the southwest,improving predictive capabilities of snowpack processesin models is imperative for comprehensive water resourcemanagement.[3] To assess the impact of future climate change on

snowpack in the headwater region of the Colorado Rockies, ahigh‐resolution (2 km) regional climate model, based on theWeather Research and Forecast (WRF) model coupled tothe Noah land surface model (LSM), was recently appliedfor multiyear snow season simulations [Ikeda et al., 2010].While the WRF/Noah model simulated winter precipitation

1National Center for Atmospheric Research, Boulder, Colorado, USA.2Department of Civil and Environmental Engineering, University of

Washington, Seattle, Washington, USA.3National Centers for Environmental Prediction, Camp Springs,

Maryland, USA.

Copyright 2010 by the American Geophysical Union.0148‐0227/10/2009JD013470

JOURNAL OF GEOPHYSICAL RESEARCH, VOL. 115, D22101, doi:10.1029/2009JD013470, 2010

D22101 1 of 15

verified well against Natural Resources Conservation ServiceSnowpack Telemetry (SNOTEL) observations, the snowpackcharacteristics modeled by the Noah LSM exhibited a tooearly snow period end, and too low and too early seasonalmaximum snow water equivalent. Past studies have found anegative bias in the Noah representation of total snow waterequivalent (SWE) and snow cover extent, and in the timing ofspring snow depletion. Sheffield et al. [2003] found that overthe United States, Noah tended to underestimate snow coverextent by about 22% in 12 km simulations. For mountainousregions, the under prediction was as much as 65%. AlthoughNoah usually under predicted throughout the season, thelargest under prediction occurred in spring. Noah also tendedto have a shorter annual snow period (later start and earlierend) than observations. Livneh et al. [2009] also showed thatcompared to satellite and ground‐based observations Noahunderestimates SWE and snow cover extent throughout theUnited States. This low bias in snow cover and SWE seems acommon weakness in other land surface models as well. Forinstance, the four LSMs in NLDAS (Noah, MOSAIC, VIC,and SAC) were found to under predict seasonal maximumSWE during a 3 year period [Pan et al., 2003]. In their pan‐Arctic hydrology study, Slater et al. [2007] found that Noahhad similar performance to other models (VIC, CHASM,CLM, and ECMWF) in simulated seasonal SWEmean squareerror. Even the best models in SWE bias had problems withsnowpack timing; being too high in the early snow season andtoo low in melt phase. Slater et al. [2007] also found that theNoah prediction of the mean end day of seasonal snowpackhad a slight high bias, but performed as well or better than theother LSMs in their study.[4] The community Noah LSM has been widely used in

weather and regional climate models [Chen et al., 1996;Chenand Dudhia, 2001; Ek et al., 2003; Leung et al., 2005, 2006;Jiang et al., 2008] and is currently used in the predictionmodels of the National Centers for Environmental Predictionand in land data assimilation systems such as the NorthAmerica Land Data Assimilation System [Mitchell et al.,2004] and the Land Information System [Peters‐Lidardet al., 2007]. Therefore, the goal of this paper is to addressthe early snowmelt problem in Noah over the mountainousColorado Rocky regions by incrementally adding modifica-tions to the Noah land surface model. In general, models usedfor snowpack prediction vary substantially from complexmodel such as SNTHERM [Jordan, 1991], which considersmultiple snow layers, solar radiation penetration, liquid waterretention and snow grain growth, to simple single‐layermodels with minimal parameters [e.g., Essery and Etchevers,2004]. Noah can be considered on the low‐complexity side ofthe snow model spectrum, and a community effort to includeamultilayer snowpackmodel within Noah is underway (G. Y.Niu et al., The community Noah land surface model withmultiphysics options, submitted to Journal of GeophysicalResearch, 2010). However, this study will only focus onimproving the publically available Noah V3.0, whose inher-ent limitations include its single snow layer, lack of snowwater retention, and snowmelt based on residual energy fromthe surface energy balance. Noah does consider patchy frac-tional snow cover, snowpack heat flux, and spatially varyingsnow albedo. One important purpose of this study is to under-stand the extent to which a simple vegetation‐snow blendedmodel can be improved to represent seasonal evolution of

snowpack, because this type of snow model is commonlyused in weather and regional climate models. We also seek tounderstand the physical processes and model parameters thatsignificantly control partitioning of incoming energy at thesnow surface through surface‐atmospheric exchange. Modelresults are compared to available observation‐based dataassimilation output and ground‐based observations.[5] This paper is organized as follows: section 2 describes

the modeling setup and observations used for comparison;section 3 presents the results of the control simulation andidentifies the deficiencies addressed in the study; section 4describes each of the model modifications; section 5 com-pares the incremental model modification results to obser-vations; and the conclusions are presented in section 6.

2. Model Setup and Observations

2.1. Noah Land Surface Model

[6] Surface processes are simulated using the Noah landsurface model [Chen et al., 1996;Chen and Dudhia, 2001; Eket al., 2003], which is based on a diurnally dependent Penmanpotential evaporation approach, a multilayer soil model, anda primitive canopy model. Noah has been extended to includea modestly complex canopy resistance and frozen groundphysics. When snow is present, Noah considers a blendedsnow‐vegetation‐soil layer and simulates the snow accumu-lation, sublimation, melting, and heat exchange at snow‐atmosphere and snow‐soil interfaces, based on a simple snowparameterization of Koren et al. [1999]. Noah does not haveindependent snow layers, nor does Noah have a canopy snowinterception component.[7] Albedo for a grid containing snow is calculated using a

linear weighting between the nonsnow background albedo onthe snow‐free fraction and a blended snow‐vegetation albedoon the snow‐covered fraction. Snow‐covered fraction (fs) isdetermined as a function of SWE:

fs ¼ 1� exp �2:6SWE

SWEmax

� �� SWE

SWEmaxexp �2:6ð Þ

� �; ð1Þ

when SWE is below the vegetation type‐dependent SWEmax

(40 cm for grass/crop/shrub and 80 cm for forests). WhenSWE is greater than SWEmax, fs equals 1. The fraction of thegrid covered by snow is parameterized differently betweenland surface models [e.g., Sheffield et al., 2003] and can bea very important determinant of model performance. Differ-ent fs formulations are not considered this study becauseequation (1) was found to alleviate near‐surface temperaturebiases in coupled versions of Noah [Ek et al., 2003].[8] The Noah LSM has been implemented and tested

extensively in both offline modes and land data assimilationsystems [e.g., Chen et al., 1996; Chen and Mitchell, 1999;Mitchell et al., 2004; Wood et al., 1998; Boone et al., 2004;Chen, 2005], and coupled modes such as the MM5 andWeather Research and Forecasting (WRF) [Skamarock et al.,2008] model and several National Centers for EnvironmentalPrediction (NCEP) weather and climate models [Chen et al.,1997; Chen and Dudhia, 2001; Yucel et al., 1998; Marshallet al., 2003; Ek et al., 2003]. Noah can be executed in off-line or coupled mode and provides fluxes of energy andwater, including snow sublimation and snowmelt, from theland surface, while alsomaintaining stores of water, including

BARLAGE ET AL.: NOAH SNOWPACK PREDICTION D22101D22101

2 of 15

snowpack, and energy in four or more soil layers. Noah alsorequires several input parameter data sets that are based onin situ or satellite observations.

2.2. High‐Resolution Land Data Assimilation System

[9] For this study, Noah model simulations are conductedusing the High Resolution Land Data Assimilation System(HRLDAS) [Chen et al., 2007]. The HRLDAS was devel-oped to provide consistent land‐surface input fields for WRF(or any mesoscale or global model) that are on the samemodel grid and have had time to develop their equilibriumclimatology, a process that takes up to several years and thuscannot be reasonably handled within the much more com-putationally expensive WRF framework [Chen et al., 2007].To run the HRLDAS, land surface properties and atmosphericforcing data are required. The HRLDAS is flexible enough touse a wide variety of satellite, radar, model, and in situ inputsand can also be used, as it is in this study, simply as a driverfor Noah.[10] The simulation domain has also been used for coupled

WRF/Noah regional climate simulation for the ColoradoHeadwater region [Ikeda et al., 2010], which is approxi-mately 1200 km × 1000 km and centered on Colorado (shownin Figure 1). The central and western portions of the domainare dominated by mountainous terrain while the easternportion slopes gently toward the east. All simulations arecompleted at a 2 km horizontal resolution on a grid with 601 ×501 cells. Themodel integrates forward in time using a 1 hourtime step. Most atmospheric forcing data used to drive theHRLDAS are taken from a WRF simulation over the samedomain (see Ikeda et al. [2010] for a description and valida-tion of the WRF simulation). For this study, the HRLDAS is

modified to use model level output directly from 2 km WRFsimulations. Since the model forcing fields are taken fromWRF before any modifications are done to the Noah model,there may be biases in the surface fields due to poorly mod-eled snowpack. Using the first model level (∼20 m) insteadof the 2 m diagnostic variables should alleviate the poten-tial feedback. Model‐generated output being used as forcingare the lowest model layer temperature, humidity, and windsalong with pressure, downward longwave radiation, andprecipitation at the surface. Precipitation partitioning is doneusing the WRF model temperature (< = 0°C implies snow)and Noah model skin temperature (< = 0°C implies surfacefreezing of liquid temperature).[11] All of the above variables are difficult to obtain over

such a large area and at a high spatial resolution. However,it is possible to get accurate observations of solar radiationthrough GOES satellite measurements [Pinker and Laszlo,1992]. Since the HRLDAS can readily use these observa-tions when available, downward solar forcing is taken fromthe GOES surface downward flux historical database (avail-able at http://www.atmos.umd.edu/∼srb). These data areavailable hourly and at a horizontal resolution of 0.5°, whichis then interpolated to the HRLDAS 2 km grid.[12] Model state variables, including snow water equiv-

alent and snow depth, are initialized using the WRF ini-tialization on 1 November 2008, except for the SNOTELlocations, where snow states are initialized with observedSNOTEL fields.[13] As a validation of the WRF precipitation, Figure 2

shows a scatterplot of total accumulated WRF precipitationversus accumulated SNOTEL precipitation for the 9 monthsimulation period. Although there is spread in the data, noobvious bias or dependence on terrain height is displayedexcept for a low model bias for SNOTEL locations receivinggreater than 1000 mm of precipitation, which may be due tosub‐2‐km terrain effects [see Ikeda et al., 2010].

2.3. SNOTEL Observation Sites

[14] SNOTEL sites are an extensive, automated system tocollect snowpack and related climatic data in the WesternUnited States. A typical site has a pressure sensing snowpillow, storage precipitation gage, and an air temperaturesensor and these measurements have been found to comparefavorably to snow course measurements [Serreze et al.,1999]. SNOTEL data have been used as validation data inprevious Noah studies [e.g., Livneh et al., 2009; Pan et al.,2003]. Snow water equivalent (SWE) observations at 112SNOTEL sites, mostly in Colorado, are used to verify theNoah model output (see Figure 1). SNOTEL precipitationis also used to justify the use of WRF model output andas forcing data to test model sensitivity. To compare modeloutput to SNOTEL observations, the four nearest WRF gridpoints to the observation location are averaged using a dis-tance weighting method. Results are found to be insensitiveto different averaging techniques. A majority of the locationshave average terrain differences compared to WRF that areless than 100 m and nearly 80% are less than 200 m. Forthe 448 WRF grid points used to compare with SNOTEL,land cover is dominated by evergreen needleleaf forest (41%),savannah (37%), and mixed forest (13%).

Figure 1. Terrain height (m) of the WRF domain includingmarkers indicating location of SNOTEL sites. The KILNSNOTEL site (39.3°N, −106.6°E; 2926 m) is circled.

BARLAGE ET AL.: NOAH SNOWPACK PREDICTION D22101D22101

3 of 15

2.4. SNODAS Output

[15] There are no large‐scale observations of SWE. TheNational Operational Hydrologic Remote Sensing Centerproduces a gridded SWE product output from their SnowData Assimilation System (SNODAS) [Barrett, 2003].Although not an observational product, SNODAS doesintegrate a physically based, spatially distributed energy andmass balance model with observed snow data from satelliteand airborne platforms, and ground stations. SNODAS out-put has high spatial (30 arcsec or ∼1 km) and temporal (1 h)resolutions over the United States. In this study, SNODASoutput is interpolated to the HRLDAS grid using bilinearinterpolation and is used to compare against domain‐widesimulations.

3. Control Simulation

[16] The control simulation consists of the HRLDAS as thedriver for the standard release Noah model version 3.0. Allsimulations are done on the 2 km domain shown in Figure 1.The simulation period is 1 November 2007 to 1 August 2008,the same period for evaluating coupled WRF/Noah sim-ulations, and is executed at a 1 hour time step. Simulationdescription labels consisting of two‐letter codes indicatedifferent model changes or forcing data. In the control sim-ulation, GOES solar (GS) andWRFmodel level (ML) forcingare used. Therefore, the control simulation has the code nameGSML. The two‐letter codes for all simulations are summa-rized in Table 1.

[17] Three metrics are used for model performance evalu-ation: seasonal maximum SWE, date of seasonal maximumSWE, and date of last significant SWE. Similar statisticshave been used in other studies using Noah [Pan et al., 2003;Slater et al., 2007]. Using these three metrics, a majority ofimportant characteristics for snowpack hydrology, includingthe shape and timing of the SWE curve, can be determined.Seasonal maximum SWE and its date of occurrence aresimply found by searching the model SWE time series duringthe 9 month simulation. Date of last significant SWE is foundby searching from the simulation end until a time whengreater than 5 mm SWE exists at the grid point.[18] As discussed previously, the control simulation pro-

duces too little peak SWE, the peak is too early in the snowseason and the snowmelts too fast in spring. These results aresimilar to those found by other studies using the Noah model[Livneh et al., 2009; Pan et al., 2003; Sheffield et al., 2003;Niu et al., submitted manuscript, 2010]. Tables 2–4 sum-

Table 1. Two‐Letter Codes Used for Simulation Labels

Code Definition

GS use GOES solar forcingML use WRF model level forcingLV Livneh et al. [2009] time‐decaying snow albedoTA terrain adjustment for grid point slope and aspectCH use WRF stability adjustment85 set maximum albedo to 0.85ZE use zo based on exposed vegetationOP use SNOTEL observed precipitation

Figure 2. Scatterplot of total accumulated WRF/HRLDAS precipitation (mm) versus accumulatedSNOTEL precipitation (mm) for 1 November 2007 to 1 August 2008. Each SNOTEL location is associatedwith the four nearest WRF grid points. Data are binned by SNOTEL terrain: black (0–2500 m), red (2500–3000 m), blue (3000–3500 m), and green (3500+ m).

BARLAGE ET AL.: NOAH SNOWPACK PREDICTION D22101D22101

4 of 15

marize the results of the full HRLDAS domain and theSNODAS output. Table 2 shows that the control simulationpredicts maximum SWE 60% lower than SNODAS for low‐elevation locations (less than 2000 m) and varies between45% and 28% lower for elevations between 2000 and 3500m.Table 3 shows that the control simulation peak SWE is about20–30 days too early compared to SNODAS for grids pointslower than 3500 m. For the same elevations, Table 4 showsthe disappearance of the snowpack comes about 15–30 daystoo early. Although a relatively small fraction of the domain,very high elevation grid points above 3500 m compare rea-sonably well for all three metrics.[19] Tables 5–7 show the same comparison as above but

with the 112 SNOTEL observation locations. Model resultsuse the four closest grid points to each SNOTEL location.As above, maximum SWE is about 35–40% too low for theSNOTEL locations between 2500 and 3500 m (Table 5).There are only 3 SNOTEL sites below 2500 m and 3 sitesabove 3500 m so statistics for these locations are not robust.SWE peaks about 40–50 days too early (Table 6) and dis-appears about 12–14 days too early (Table 7).

4. Model Modifications

[20] To address some of the deficiencies in the controlsimulation, several modifications are applied to the Noahmodel. Each modification is described below. The changesare added sequentially meaning each modification is added toa model containing all the modifications listed above it. Alsonote that because it is not practical to show detailed resultsfrom all SNOTEL sites, at times detailed results are presentedfrom a representative site referred to as “KILN” that is locatedin central Colorado (see Figure 1).

4.1. Livneh et al. [2009] Time‐Varying Albedo

[21] One of the most important purposes of using a landsurface model is the correct partitioning of incident radiation

into radiative, sensible heat and latent heat fluxes. Netradiation drives the terrestrial surface energy budget. Surfacealbedo, the ratio of reflected to incident solar radiation, isan important surface characteristic that determines energytransfer at the surface. One of the fastest modifications tosurface albedo is snow cover. Because the albedo of snow‐covered land compared to snow‐free land is so extreme,correctly predicting the presence of snow can greatly affectthe near‐surface energy balance. Snow cover can change thealbedo of grassland by a factor of 3–4 and forested regions bya factor of 2–3 [Betts and Ball, 1997; Thomas and Rowntree,1992; Jin et al., 2002]. In modeling studies, these changeshave been shown to have large effects on the surface energybudget and near‐surface temperature [Thomas and Rowntree,1992; Viterbo and Betts, 1999]. Modeling studies have alsoshown that snow‐covered albedo changes can greatly affectthe timing of spring snowmelt and subsequent streamflowpeaks [Thomas and Rowntree, 1992]. Therefore, albedo, andmore generally, the surface energy budget are the main areaswhere model improvement is sought.[22] The Noah model uses a linear blending of background

vegetation/soil albedo and spatially dependent snow‐coveredalbedo [Robinson and Kukla, 1985] based on snow coverfraction (see equation (1)). If snow cover does not change,albedo does not change. Livneh et al. [2009] incorporated analbedo decay scheme adapted from the DHSVM [Wigmostaet al., 1994] and VIC [Cherkauer et al., 2003] models,which are based on the observed albedo decay rates initiallymade by the U.S. Army Corps of Engineers [1956]. Thescheme imposes an exponential decay in albedo (asnow) as thetime between snow events increases:

�snow ¼ �maxABt; ð2Þ

where A = 0.94 (0.82) and B = 0.58 (0.46) during the accu-mulation (ablation) phase and t is the time in days sincethe last snowfall. The albedo decay occurs at different rates

Table 2. Domain‐Averaged Maximum SWE for Each HRLDAS Simulation and SNODAS Outputa

0–1000 m 1000–1500 m 1500–2000 m 2000–2500 m 2500–3000 m 3000–3500 m 3500–4500 m

Number of grid points 35918 79639 90941 63694 20463 8523 1923Control 10 11 16 66 190 355 481+ LV 10 12 17 73 213 394 517+ TA 10 12 17 73 213 394 517+ CH 11 14 23 87 225 407 538+ 85 12 16 30 111 280 478 613+ ZE 12 16 29 106 274 470 604SNODAS 22 26 45 119 306 491 498D precipitation 23 −76 −149 −116 −78 18 76

aSWE is measured in millimeters. Data are binned based on WRF terrain. Also included are differences between WRF precipitation forcing and SNODASprecipitation.

Table 3. Domain‐Averaged Julian Date of Maximum SWE for Each HRLDAS Simulation and SNODAS Outputa

0–1000 m 1000–1500 m 1500–2000 m 2000–2500 m 2500–3000 m 3000–3500 m 3500–4500 m

Control −6 −4 3 27 45 69 92+ LV −5 −4 3 29 52 78 100+ TA −5 −4 3 29 52 78 100+ CH −9 −6 7 33 54 78 98+ 85 −7 −1 14 43 69 91 111+ ZE −7 −3 12 41 67 89 109SNODAS 25 32 25 48 77 99 103

aData are binned based on WRF terrain. Negative numbers imply days into December.

BARLAGE ET AL.: NOAH SNOWPACK PREDICTION D22101D22101

5 of 15

depending on snow surface temperature, decaying slowerduring the accumulation phase and faster during ablation.The new snow albedo (amax) is taken to be some adjust-able fraction (currently 0.5) of the difference between theRobinson and Kukla value and ground observations of freshsnow (∼0.85) [Wiscombe and Warren, 1980].[23] Figure 3 shows the albedo for the month of November

at the KILN SNOTEL site using both the control simulationalbedo formulation and that of Livneh et al. [2009]. TheRobinson and Kukla [1985] value for snow‐covered albedo atthis site is 0.52. Even with substantial snow cover, the controlsimulation albedo rarely gets above 0.5. The Livneh et al.formulation shows the decay of albedo with time and also thesharp increase in albedo when fresh snow falls. The down-ward spikes in the early part of the month occur when thesnowpack is assumed to have liquid water at the surface.

4.2. Adjustment for Terrain Slope and Aspect

[24] The slope and orientation of the surface can makesubstantial differences to the amount of incident shortwaveradiation through modification of the incident solar zenithangle [Whiteman, 2000]. Specifically, for the northernhemisphere, the directional orientation (aspect) of the landsurface determines if incident radiation will increase (southfacing), decrease (north facing), or have a diurnal adjustment(east and west facing). Figure 4 shows the grid point aspectvalues binned to the cardinal directions. There are nearlyequal number of north and south facing grids, but due to theexpansive, eastward sloping eastern portion of the domain,there are nearly twice as many east facing grids as west. Inaddition to terrain aspect, terrain slope also plays an importantrole in modifying the incident solar zenith angle. Figure 5shows the terrain slope values calculated from the 2 km ter-rain field generated by WRF. At this resolution, althoughmost mountainous regions have maximum slopes in the 4–6 degree range, there are several locations with slope valuesgreater than 10 degrees. For instance, the Grand Canyon can

clearly be seen in the southwest section of the domain. Forthis study, the prescribed GOES downward solar radiationis adjusted based on estimation of the local incident solarzenith angle using local latitude, slope and aspect valuesderived from the 2 km grid along with time of day and day ofyear values. This formulation has been used recently by Zängl[2004, 2009] and is based on the original formulation ofGarnier and Ohmura [1968]. The methodology also parti-tions incident shortwave radiation into direct and diffusecomponents in which, for a given daylight time and location,the diffuse component is a constant fraction of the top ofatmosphere radiation values while the direct component isaltered based on the local incident solar zenith angle.

4.3. Stability Adjustment for Stable Boundary Layers

[25] New in WRF version 3.0 is an adjustment to the sur-face exchange coefficient when the atmospheric boundarylayer is stable. The adjustment effectively decouples thesurface from the atmosphere by increasing the Zilitinkevichcoefficient, which then decreases the exchange coefficient[see Chen et al., 1997]. If a surface inversion exists, theZilitinkevich coefficient is increased by a factor of three. Anadditional increase is applied as a quadratic function of terrainheight (h) for elevations greater than 580 m (9.0 × 10−6 h2).The effect of the adjustment can easily be seen in Figure 6,which shows the surface exchange coefficient for the first2 days of November over the KILN SNOTEL location bothwith and without the stability adjustment. During the stablenighttime periods, surface exchange is decreased by morethan 80%, substantially reducing snow sublimation and down-ward sensible heat flux from the atmosphere.

4.4. Sensitivity to Maximum Snow Albedo Parameter

[26] The Livneh et al. [2009] albedo scheme describedin section 4.1 uses a factor of 0.5 to weight the Robinson andKukla [1985] maximum snow albedo and the albedo for freshsnow (0.85), and resulted in too low albedo for new snow

Table 4. Domain‐Averaged Last Julian Date of 5 mm SWE for Each HRLDAS Simulation and SNODAS Outputa

0–1000 m 1000–1500 m 1500–2000 m 2000–2500 m 2500–3000 m 3000–3500 m 3500–4500 m

Control 33 32 35 75 115 143 167+ LV 33 32 35 75 117 148 171+ TA 33 32 35 75 117 148 171+ CH 33 34 42 79 119 148 171+ 85 35 39 50 90 133 167 191+ ZE 34 37 47 87 131 165 189SNODAS 63 75 70 98 132 158 165

aData are binned based on WRF terrain.

Table 5. Average Maximum SWE for Each HRLDAS Simulation at SNOTEL Locations and SNOTEL Sitesa

2000–2500 m 2500–3000 m 3000–3500 m 3500–4500 m Correlation

Number of sites 3 47 59 3Control 262 320 445 444 0.82+ LV 287 368 494 482 0.82+ TA 285 369 495 482 0.82+ CH 284 379 502 505 0.80+ 85 330 465 576 575 0.79+ ZE 328 462 573 572 0.80+ OP 333 476 563 544 0.92SNOTEL 499 545 678 540

aSWE is measured in millimeters. Data are binned based on SNOTEL terrain.

BARLAGE ET AL.: NOAH SNOWPACK PREDICTION D22101D22101

6 of 15

cover. This modification eliminates the Robinson and Kuklaalbedo and sets all fresh snow to have an albedo (amax inequation (2)) of 0.85. The albedo continues to decay in timefrom this maximum fresh snow value using equation (2). Thered curve shown in Figure 3 already has this 0.85 modifica-tion. This modification works in conjunction with the Livnehet al. albedo modification. Setting albedo to an unvaryingconstant of 0.85 is inappropriate because of processes such assnow aging and snow falling off canopy.

4.5. Roughness Length Adjustment for VegetationWith Snow

[27] Currently, the Noah model uses a background vege-tation roughness length that is independent of the presence ofsnow. A modification is added so that only the exposed partof the vegetation is used in the roughness length calculation.To calculate the new roughness length, the canopy heightis first approximated by multiplying the background rough-ness length by 7 [see Arya, 1988]. Next, model snow depthis subtracted from the canopy height to approximate theexposed vegetation. This exposed canopy height is thendivided by 7 to get the new exposed roughness length. Theexposed roughness length is applied only over the snow‐covered fraction of the grid.

5. Results

[28] Results are given in two categories: (1) the entiredomain compared with SNODAS output and (2) model gridpoints associated with the SNOTEL observation sites, whichare usually located at high elevations.

5.1. Domain‐Wide Effects

[29] The domain‐wide effects of each modification aresummarized in Tables 2–4 for maximum SWE, date ofmaximum SWE and date of last 5 mmSWE, respectively. Formaximum SWE shown in Table 2, the individual modifica-

tions have little effect for grid points lower than 1500 m. Thismay be because most of these points are in the “marginal” or“transient” snow zones in the south and southeast portions ofthe domain and have much more dependence on seasonaltiming of precipitation forcing. For higher‐elevation loca-tions, it is clear that the major improvements are made byemploying the Livneh et al. [2009] albedo scheme (LV),setting fresh snow albedo to 0.85(85), and using stabilityadjustment (CH). The net effect of all modifications in thethree terrain bins between 2000 and 3500 m is to reduce theunder prediction compared to SNODAS from 45%, 38%, and28% to 11%, 10%, and 4%, respectively.[30] The timing of maximum SWE in Table 3 shows a

similar improvement for high‐elevation grid points with littlechange in the lower elevations. Again, the Livneh et al. [2009]albedo scheme and fresh snow albedo change appear to havethe greatest positive impact increasing the maximum SWEdate by 2–9 and 10–15 days, respectively, for high elevations.[31] For timing of snow disappearance, Table 4 shows

similar results to the timing of maximum SWE. For low‐elevation grid points, the snow still disappears 30 days earlierthan SNODAS, while for grid points higher than 3000 m thesnow is not melting early enough.[32] The terrain adjustment has very little effect on any of

the three domain‐averagedmetrics since the net orientation of

Table 6. Average Julian Date of Maximum SWE for EachHRLDAS Simulation at SNOTEL Locations and SNOTEL Sitesa

2000–2500 m 2500–3000 m 3000–3500 m 3500–4500 m

Control 64 61 78 96+ LV 65 74 85 102+ TA 65 75 85 102+ CH 63 74 84 101+ 85 69 88 93 112+ ZE 69 87 92 110+ OP 69 92 101 118SNOTEL 107 110 101 112

aData are binned based on SNOTEL terrain.

Table 7. Average Last Julian Date of 5 mm SWE for Each HRLDAS Simulation at SNOTEL Locations and SNOTEL Sitesa

2000–2500 m 2500–3000 m 3000–3500 m 3500–4500 m Correlation

Control 119 133 150 167 0.56+ LV 121 137 155 172 0.60+ TA 121 137 155 172 0.60+ CH 121 139 155 172 0.63+ 85 135 156 175 191 0.70+ ZE 134 156 174 190 0.70+ OP 140 159 175 192 0.75SNOTEL 140 147 162 140

aData are binned based on SNOTEL terrain.

Figure 3. Albedo for the month of November 2007 at theKILN SNOTEL site using the Livneh et al. [2009] albedoformulation (red) and the control simulation (black).

BARLAGE ET AL.: NOAH SNOWPACK PREDICTION D22101D22101

7 of 15

the high sloping terrain is almost neutral. The roughnesslength adjustment has little effect or makes some of themetrics worse because decreasing the roughness lengtheffectively reduces the surface fluxes. The extra energyretained at the surface then goes into melting snow; there-fore even though sublimation is reduced, net SWE impactis negative. Since the stability adjustment predominantlyfunctions during the night when there is no solar input, sub-limation is reduced. Periods of excessive downward night-time sensible heat flux are also reduced and the snowpackis increased.[33] Figure 7 shows the simulation success in snow cover

comparison to SNODAS for all points and for terrain binsusing all model modifications (GSMLLVTACH85ZE). Theblue lines show the percentage of points when both HRLDASand SNODAS predict no snow and the green lines showwhenboth predict snow. The black lines indicate the fraction whereHRLDAS predicts snow and SNODAS does not and thered lines show when SNODAS predicts snow and HRLDASdoes not. The sum of the black and red lines being zero wouldmean perfect correlation between the models. The black linebeing equal to the red line means the snow‐covered area isthe same, but is misplaced. This is not unexpected consider-ing HRLDAS and SNODAS have difference precipitationforcing.[34] For lower elevations (<2000 m), HRLDAS tends

to overpredict snow extent in the early part of the snow sea-son, peaking in January (black line). For higher elevations(>2500 m) where most SNOTEL sites are located, HRLDASstarts with too much snow. From January to March all high‐elevation locations are snow covered and near the end of the

season HRLDAS keeps snow too long as was seen in Table 4.Since most of the time the black line is higher than thered line, this implies the HRLDAS snow is, in general, tooextensive compared to the SNODAS snow. Except for thelower elevation, midwinter peak in the HRLDAS snow rel-ative to SNODAS, agreement between the models is usuallygreater than 80%.[35] See the auxiliary material for a similar snow cover

comparison with the Moderate Resolution Imaging Spectro-

Figure 4. Terrain aspect binned into cardinal directions:north (−45°–45°), east (45°–135°), south (135°–225°), andwest (225°–315°). Aspect is calculated from the 2 km WRFterrain field.

Figure 5. Terrain slope (degrees) derived from 2 km WRFterrain field.

Figure 6. Exchange coefficient (CHX) using the new WRFstability adjustment for stable boundary layers (red) and theoriginal formulation (black).

BARLAGE ET AL.: NOAH SNOWPACK PREDICTION D22101D22101

8 of 15

Figure 7

BARLAGE ET AL.: NOAH SNOWPACK PREDICTION D22101D22101

9 of 15

radiometer (MODIS) snow cover product.1 The results ofthe comparison with MODIS snow cover are very similar.

5.2. Comparison With SNOTEL Observations

[36] The control simulation shows the same pattern ofunder prediction of the three metrics for the 112 SNOTELsites as in domain‐wide SNODAS comparison (Tables 5–7).An additional simulation is completed over 448 WRF‐SNOTEL locations (the 4 closest WRF points to the 112SNOTEL sites) using the SNOTEL observed precipitation(OP). For statistical robustness, only results for the 47 sitesbetween 2500 and 3000 m and 59 sites between 3000 and3500 m are discussed. For each of these statistics, averageof the four closest WRF points are used to compare to theobservation site. Using the five model changes, maxi-mum SWE simulation shown in Table 5 improves for thelower‐elevation sites from an under prediction of 41% to anunder prediction of 15% using WRF precipitation forcing(GSMLLVTACH85ZE) and 13% using SNOTEL forcing

(GSOPMLLVTACH85ZE). Likewise, the higher‐elevationsites improve maximum SWE from an under prediction of34% to under predictions of 15% and 17% depending on theprecipitation forcing. Regardless of the model modificationadded, correlation between model results and SNOTELobservations remains at about 0.8 (see Table 5), but the useof SNOTEL precipitation increases the correlation to 0.92.[37] Figure 8 shows a scatterplot of seasonal maximum

SWE for the control simulation and simulation with all fivemodel modifications. Figure 8 shows that the modified modelreduces the low bias in SWE. However, the spread in eachsimulation is very similar, therefore, not affecting the cor-relation as seen in Table 5. The largest disagreement isat locations with SNOTEL maximum SWE greater than900 mm, where HRLDAS greatly underestimates maximumSWE. This is not surprising considering the precipitationscatterplot in Figure 2 shows precipitation under predictionwhen SNOTEL precipitation is greater than 1000 mm.[38] The date of peak SWE improves by 26 and 14 days

for low‐ and high‐elevation sites, respectively (Table 6).The model simulations are improved even more by theuse of observed precipitation. However, even though the

Figure 8. Seasonal maximum SWE (mm) scatterplot for HRLDAS compared to SNOTEL at all SNOTELstations. Each HRLDAS point is an average of the four closest points to the SNOTEL station. The controlsimulation is in black, and the simulation with all model changes is in red.

Figure 7. Time series of HRLDAS snow cover with all model modifications compared to SNODAS snow cover, where snowcover is defined as grid points with greater than 1 mm SWE. Blue lines show locations where both HRLDAS and SNODASindicate no snow cover, green lines show locations where both models indicated snow cover, black lines show locations whereHRLDAS indicates snow and SNODAS does not, and red lines show locations where SNODAS indicates snow and HRLDASdoes not. (a) Data for the entire domain. Separation of model grid points based on terrain heights of (b) less than 1000 m,(c) 1000–1500 m, (d) 1500–2000 m, (e) 2000–2500 m, and (f) greater than 2500 m.

1Auxiliary materials are available in the HTML. doi:10.1029/2009JD013470.

BARLAGE ET AL.: NOAH SNOWPACK PREDICTION D22101D22101

10 of 15

high‐elevation sites get the timing of peak correct, the lower‐elevation sites are still early by 18 days. Timing of the lastdate of 5 mm SWE go from under predictions of 12 days toover predictions of about 12 days (Table 7). Although cor-relation improves incrementally with model changes, themajor impact occurs with maximum albedo, which results insnow remaining for too long.[39] The seasonal SWE time series for each of the model

simulations averaged over all SNOTEL sites are shown inFigure 9. The control simulation clearly has too early and toolow peak SWE and melts the snow too early in spring. TheLivneh et al. [2009] albedo formulation shows incrementalimprovement in all three metrics (note in Figure 9 that theblack dashed line overlies the blue dashed line). Terrain andstability adjustments (TA and CH) do not have a large overallnet effect. The largest incremental improvement in peak SWEamount and timing occurs with the adjustment of maximumalbedo (85), although this change keeps snow on the groundtoo late in the spring. Finally, the roughness length (zo) for-mulation (ZE) has very little effect on average.[40] Although in practice the 85 simulation should be

coupled to the LV simulation (see section 4.4), a test simu-lation where the snow albedo is permanently set at 0.85produces a snowpack that remains for about a month longerthan observations (SNOTEL). However, the simulation doesnot produce a large difference in timing of maximum SWE(<1 day difference) and magnitude of maximum SWE (∼20–30 mm difference).[41] To analyze why the SWE curve changes between the

control simulation and simulation with all modifications,

snow sublimation and snowmelt are added to the SWE curvein Figure 10. Figure 10 clearly shows that the model changessuppress sublimation throughout the simulation while keep-ing the snowmelt curves similar through April and May untilthey diverge in lateMay. In the control simulation, about 45%of the snow sublimates, while about 15% sublimates in thefully modified simulation.[42] Small differences in precipitation can make large dif-

ferences in SWE especially in the late spring melt season.Figure 11 shows the SWE time series for the KILN SNOTELsite using the fully modified model with both WRF andSNOTEL precipitation forcing. Total accumulated precipi-tation for both WRF and SNOTEL are also shown. At thislocation, SNOTEL precipitation is higher by between 10 and70 mm during the simulation. For most of the simulation,SWE differences are comparable to precipitation differences.However, during the spring melt period SWE differencescan be as much as 170 mm immediately before the springsnowpack collapse. Also shown in this plot are the four pointsclosest to the SNOTEL site used to calculate the average.Even at 2 km resolution there is a substantial amount ofspatial variability, which must be considered when addres-sing any direct comparison between SNOTEL site observa-tions and model grid point output. This spatial variability isalso one of the reasons the results presented focus on theSNOTEL average and not individual SNOTEL locations.

5.3. Terrain Slope and Orientation Effect

[43] The adjustment for terrain slope and orientation hasvery little effect on the domain‐averaged and site‐averaged

Figure 9. Time series of SWE for the 9 month simulation starting 1 November 2007 averaged over allSNOTEL locations. The control simulation is the dashed red line. The other lines are the accumulated addi-tion of model changes: Livneh et al. [2009] albedo formulation (dashed blue line), surface terrain effect(dashed black line), WRF stability formulation (solid red line), snow albedo set to 0.85 (solid blue line), andzo formulation (solid black line). Black dots are the SNOTEL SWE observations.

BARLAGE ET AL.: NOAH SNOWPACK PREDICTION D22101D22101

11 of 15

Figure 10. Time series of SWE (solid lines), snow sublimation (long‐dashed lines), and snowmelt (short‐dashed lines) for the 9 month simulation starting 1 November 2007 averaged over all SNOTEL locations.Control simulation is in red, and simulation with all model changes is in blue. Black dots are the SNOTELSWE observations.

Figure 11. Time series of SWE at KILN using all model changes with precipitation forcing from SNOTEL(thick blue line) and WRF (thick red line). Their difference is shown by the short‐dashed blue line. SWE atthe four WRF grid points used for the average is shown by thin red and blue lines. SNOTEL and WRF pre-cipitation forcing and their difference are shown by green solid, long‐dashed, and short‐dashed lines,respectively. The black line shows the KILN SWE observations.

BARLAGE ET AL.: NOAH SNOWPACK PREDICTION D22101D22101

12 of 15

results shown above. However, Figure 12 shows that for agrid point near the KILN SNOTEL site with a slope of 6.7°and west facing orientation of 254° from 0° North, monthlyaveraged diurnal cycle incident shortwave differences canreach about −70 W/m2 in the morning and 80 W/m2 in theafternoon. These differences result in latent and sensibleheat decreases of about 10W/m2 in the morning and increasesof 15 and 30 W/m2, respectively, in the afternoon. Theremaining energy goes into heating the surface and meltingsnow.[44] Even though the net terrain effect is small, Table 8

summarizes the effect of terrain slope and orientation onmaximum SWE. East and west facing slopes have little effectregardless of slope magnitude. North and south facing slopeshave complimentary effects with north facing slopes increas-ing maximum SWE and south facing slopes decreasingmaximum SWE. The magnitude of the increase or decreasedepends directly on the magnitude of the slope.

6. Conclusions

[45] Modifications are made to the Noah land surfacemodel to try to improve snowpack simulation in the ColoradoRocky Mountains. The most effective way to improve mag-nitude and timing seasonal maximumSWEwas to introduce atime‐varying albedo formulation and increase the fresh snowalbedo. Minor improvement was obtained by introducinga stable boundary layer adjustment to surface exchange.Adjusting the surface roughness length when snow is presentand adding a terrain slope and orientation adjustment hadlittle effect on average SWE simulation. The end date of snow

cover is too early in the season with the control simulation,but becomes too late with the increase of maximum albedo.[46] Although albedo is allowed to decrease with time, the

LV simulation still uses a higher albedo for snow than thecontrol simulation, and using the maximum snow equal to0.85 (labeled as the 85 simulation) further increases the freshsnow albedo. Therefore, the increase in snow resulting fromthese two changes is due to the decrease of available energy atthe surface. The improvement from the CH simulationdecreased downward sensible heat flux (which brings warmerair aloft to the snow surface) and hence reduced sublimationduring stable boundary layer conditions.[47] Adding the terrain slope and orientation effects into

Noah does not have a net effect on average simulated SWE.However, the addition does have an effect on individual gridcells. North and south facing grid will receive less and moresolar radiation, respectively. East and west facing slope willchange the diurnal distribution of incoming radiation. Thechanges are magnified with increasing slope and may beimportant in boundary layer development and convectiveprecipitation triggering when coupled to an atmosphericmodel.[48] Note that small differences in precipitation input for

Noah made quite substantial differences in the accumulationand ablation of snow, particularly in the late spring snowmeltseason. There is also an interesting interplay of sublimationand melting. For instance, aggressively reducing the sur-face exchange coefficient during stable regime will slightlydecrease snow sublimation, but significantly increase snow-melt, leading to an early snow disappearance. Although notshown, the runoff predicted by the model has a very similartiming and magnitude to the snowmelt curve shown inFigure 10. This means that for water resources prediction, themodel now predicts a much larger spring/summer runoffmagnitude with the timing unchanged at the beginning of themelt period but extend for longer into the summer.[49] The modifications suggested in this work use Noah

(V3.0) as a reference and have in some form been introducedinto subsequent versions of Noah. Dynamic albedo and a newexchange coefficient formulation have been added in Noah(V3.1). The roughness length and slope‐aspect modificationsalong with a user‐defined option to adjust maximum snowalbedo are included in Noah (V3.2).[50] The treatment of snow in the current version of Noah

(V3.0) tested in this study is not complex. The lack of over-lying canopy and liquid water retention in the snow matrixmake it difficult to reproduce the natural evolution of the

Figure 12. April mean diurnal incident shortwave radiationdifference between simulations with and without the surfaceterrain adjustment (black) at a grid point near KILN with aslope of 6.7° and orientation of 254°E (west facing). Sensibleand latent heat flux differences are shown by red and blue,respectively.

Table 8. Difference in Maximum SWE Between Simulation WithTerrain Orientation Adjustment (GSMLLVTA) and WithoutAdjustment (GSMLLV) Binned by Slope of Terrain and TerrainOrientationa

North East South West

0°–2° +1 0 0 02°–4° +3 0 −4 04°–6° +7 +1 −7 06°–8° +12 +1 −12 −18°–10° +15 +2 −16 −110°+ +13 0 −19 −2

aSWE is measured in millimeters.

BARLAGE ET AL.: NOAH SNOWPACK PREDICTION D22101D22101

13 of 15

snowpack. This study shows that improvement can be maderelative to observations in the Colorado Rocky Mountainswithout drastically changing the existing Noah model. Com-parison to observations in other environments where, forexample, snowpack liquid water retention is more important(e.g., the Pacific Northwest) may show that more detailed pro-cess representations are necessary. To completely capture thephysical representation of natural snow phenomena like canopyintercepted snow sublimation and springtime collapse of thesnow, more complex processes will need to be incorporated.

[51] Acknowledgments. This work was supported by the NationalScience Foundation Project ARC‐0733058 and by the National Center forAtmospheric ResearchWater Systems program. Greg Fall and Carrie Olheiserfrom NOAA/NOHRSC assisted with the SNODAS output. We thank threeanonymous reviewers for their helpful comments on an earlier version ofthe manuscript.

ReferencesArya, S. P. (1988), Introduction to Micrometeorology, 303 pp., Academic,San Diego, Calif.

Barnett, T. P., et al. (2008), Human‐induced changes in the hydrology ofthe western United States, Science, 319, 1080–1083.

Barrett, A. (2003), National Operational Hydrologic Remote SensingCenter Snow Data Assimilation System (SNODAS) products at NSIDC,NSIDC Spec. Rep. 11, 19 pp., Natl. Snow and Ice Data Cent., Boulder,Colo.

Betts, A. K., and J. H. Ball (1997), Albedo over the boreal forest, J. Geophys.Res., 102, 28,901–28,909, doi:10.1029/96JD03876.

Boone, A., et al. (2004), The Rhône‐Aggregation Land Surface Schemeintercomparison project: An overview, J. Clim., 17, 187–208,doi:10.1175/1520-0442(2004)017<0187:TRLSSI>2.0.CO;2.

Chen, F. (2005), Variability in global land surface energy budgets during1987–1988 simulated by an offline land surface model, Clim. Dyn., 24,667–684, doi:10.1007/s00382-004-0439-4.

Chen, F., and J. Dudhia (2001), Coupling an advanced land surface‐hydrology model with the Penn State–NCAR MM5 modeling system.Part I: Model implementation and sensitivity, Mon. Weather Rev., 129,569–585, doi:10.1175/1520-0493(2001)129<0569:CAALSH>2.0.CO;2.

Chen, F., and K. Mitchell (1999), Using GEWEX/ISLSCP forcing data tosimulate global soil moisture fields and hydrological cycle for 1987–1988, J. Meteorol. Soc. Jpn., 77, 1–16.

Chen, F., K.Mitchell, J. Schaake, Y. Xue, H. Pan, V. Koren, Y. Duan,M. Ek,and A. Betts (1996), Modeling of land‐surface evaporation by fourschemes and comparison with FIFE observations, J. Geophys. Res., 101,7251–7268, doi:10.1029/95JD02165.

Chen, F., Z. Janjic, and K. Mitchell (1997), Impact of atmospheric surface‐layer parameterizations in the new land‐surface scheme of the NCEPmesoscale Eta model, Boundary Layer Meteorol., 85, 391–421,doi:10.1023/A:1000531001463.

Chen, F., et al. (2007), Evaluation of the characteristics of the NCAR high‐resolution land data assimilation system, J. Appl. Meteorol. Climatol., 46,694–713, doi:10.1175/JAM2463.1.

Cherkauer, K. A., L. C. Bowling, and D. P. Lettenmaier (2003), VariableInfiltration Capacity (VIC) cold land process model updates, GlobalPlanet. Change, 38, 151–159, doi:10.1016/S0921-8181(03)00025-0.

U.S. Army Corps of Engineers (1956), Summary report of the snow inves-tigations: Snow hydrology, North Pac. Div., Portland, Oreg.

Easterling, D. R. (2002), Recent changes in frost days and the frost‐freeseason in the United States, Bull. Am. Meteorol. Soc., 83, 1327–1332.

Edwards, L. M., and K. T. Redmond (2005) Climate factors on ColoradoRiver basin water supply, paper presented at Association of CaliforniaWater Agencies and Colorado River Water Users Association Confer-ences, Las Vegas, Nev.

Ek, M. B., K. E. Mitchell, Y. Lin, E. Rogers, P. Grunmann, V. Koren,G. Gayno, and J. D. Tarpley (2003), Implementation of Noah land surfacemodel advances in the National Centers for Environmental Predictionoperational mesoscale Eta model, J. Geophys. Res., 108(D22), 8851,doi:10.1029/2002JD003296.

Essery, R., and P. Etchevers (2004), Parameter sensitivity in simulations ofsnowmelt, J. Geophys. Res., 109, D20111, doi:10.1029/2004JD005036.

Garnier, B. J., and A. Ohmura (1968), A method of calculating the directshortwave radiation income of slopes, J. Appl. Meteorol., 7, 796–800,doi:10.1175/1520-0450(1968)007<0796:AMOCTD>2.0.CO;2.

Hamlet, A. F., P. W. Mote, M. P. Clark, and D. P. Lettenmaier (2007),Twentieth‐century trends in runoff, evapotranspiration, and soil moisturein the western United States, J. Clim., 20, 1468–1486, doi:10.1175/JCLI4051.1.

Hidalgo, H. G., et al. (2009), Detection and attribution of streamflowtiming changes to climate change in the western United States, J. Clim.,22, 3838–3855, doi:10.1175/2009JCLI2470.1.

Hoerling, M., and J. Eischeid (2008), Past peak water in the southwest,Southwest Hydrol., 35, 18–19.

Hoerling, M., J. Eischeid, X. W. Qian, and T. Y. Xu (2007), Explaining therecord U.S. warmth of 2006, Geophys. Res. Lett., 34, L17704,doi:10.1029/2007GL030643.

Ikeda, K., et al. (2010), Simulation of seasonal snowfall over Colorado,Atmos. Res., 97, 462–477, doi:10.1016/j.atmosres.2010.04.010.

Intergovernmental Panel on Climate Change (2007), Climate Change 2007:The Physical Science Basis: Working Group I Contribution to the FourthAssessment Report of the IPCC, edited by S. Solomon et al., CambridgeUniv. Press, New York. (Available at http://ipcc‐wg1.ucar.edu/index.html.)

Jiang, X., C. Wiedinmyer, F. Chen, Z.‐L. Yang, and J. C.‐F. Lo (2008),Predicted impacts of climate and land use change on surface ozone inthe Houston, Texas, area, J. Geophys. Res., 113, D20312, doi:10.1029/2008JD009820.

Jin, Y., C. B. Schaaf, F. Gao, X. Li, A. H. Strahler, X. Zeng, and R. E.Dickinson (2002), How does snow impact the albedo of vegetated landsurfaces as analyzed with MODIS data?, Geophys. Res. Lett., 29(10),1374, doi:10.1029/2001GL014132.

Jordan, R. (1991), A one‐dimensional temperature model for a snow cover:Technical documentation for SNTHERM.89, Spec. Rep. 91‐16, 58 pp.,U.S. Army Corps of Eng., Hanover, N. H.

Koren, V., J. Schaake, K. Mitchell, Q.‐Y. Duan, F. Chen, and J. M. Baker(1999), A parameterization of snowpack and frozen ground intended forNCEP weather and climate models, J. Geophys. Res., 104, 19,569–19,585, doi:10.1029/1999JD900232.

Leung, L. R., J. Done, J. Dudhia, T. Henderson, M. Vertenstein, and B. Kuo(2005), Preliminary results ofWRF for regional climate simulations, paperpresented at Workshop on Research Needs and Directions of RegionalClimate Modeling Using WRF and CCSM, Natl. Cent. for Atmos. Res.,Boulder, Colo. (Available at http://box.mmm.ucar.edu/events/rcm05/presentations/leung_rcm_workshop.pdf.)

Leung, L. R., Y.‐H. Kuo, and J. Tribbia (2006), Research needs and direc-tions of regional climate modeling using WRF and CCSM, Bull. Am.Meteorol. Soc., 87, 1747–1751, doi:10.1175/BAMS-87-12-1747.

Livneh, B., Y. Xia, K. E. Mitchell, M. B. Ek, and D. P. Lettenmaier (2009),Noah LSM snow model diagnostics and enhancements, J. Hydrometeorol.,11, 721–738.

Marshall, C. H., K. C. Crawford, K. E. Mitchell, and D. J. Stensrud (2003),The impact of the land‐surface physics in the operational NCEP Etamodel on simulating the diurnal cycle: Evaluation and testing usingOklahoma Mesonet data, Weather Forecasting, 18(5), 748–768,doi:10.1175/1520-0434(2003)018<0748:TIOTLS>2.0.CO;2.

Milly, P. C. D., K. A. Dunne, and A. V. Vecchia (2005), Global pattern oftrends in streamflow and water availability in a changing climate, Nature,438(7066), 347–350, doi:10.1038/nature04312.

Mitchell, K. E., et al. (2004), The multi‐institutional North American LandData Assimilation System (NLDAS): Utilizing multiple GCIP productsand partners in a continental distributed hydrological modeling system,J. Geophys. Res., 109, D07S90, doi:10.1029/2003JD003823.

Mote, P. W. (2006), Climate‐driven variability and trends in mountainsnowpack in western North America, J. Clim., 19, 6209–6220,doi:10.1175/JCLI3971.1.

Pan, M., et al. (2003), Snow process modeling in the North American LandData Assimilation System (NLDAS): 2. Evaluation of model simulatedsnow water equivalent, J. Geophys. Res., 108(D22), 8850, doi:10.1029/2003JD003994.

Peters‐Lidard, C. D., et al. (2007), High‐performance Earth system model-ing with NASA/GSFC’s Land Information System, Innovations Syst.Software Eng., 3(3), 157–165, doi:10.1007/s11334-007-0028-x.

Pinker, R. T., and I. Laszlo (1992), Modeling surface solar irradiance forsatellite applications on a global scale, J. Appl. Meteorol., 31, 194–211,doi:10.1175/1520-0450(1992)031<0194:MSSIFS>2.0.CO;2.

Robinson, D. A., and G. Kukla (1985), Maximum surface albedo ofseasonally snow‐covered lands in the Northern Hemisphere, J. Clim.Appl. Meteorol., 24, 402–411, doi:10.1175/1520-0450(1985)024<0402:MSAOSS>2.0.CO;2.

Seager, R., et al. (2007), Model projections of an imminent transition to amore arid climate in southwestern North America, Science, 316, 1181–1184, doi:10.1126/science.1139601.

Serreze, M. C., M. P. Clark, R. L. Armstrong, D. A. McGinnis, and R. S.Pulwarty (1999), Characteristics of the western United States snowpack

BARLAGE ET AL.: NOAH SNOWPACK PREDICTION D22101D22101

14 of 15

from snowpack telemetry (SNOTEL) data, Water Resour. Res., 35(7),2145–2160, doi:10.1029/1999WR900090.

Sheffield, J., et al. (2003), Snow process modeling in the North AmericanLand Data Assimilation System (NLDAS): 1. Evaluation of model sim-ulated snow cover extent, J. Geophys. Res. , 108(D22), 8849,doi:10.1029/2002JD003274.

Skamarock, W. C., et al. (2008), A description of the Advanced ResearchWRF Version 3, Tech Note 475+STR, Natl. Cent. for Atmos. Res.,Boulder, Colo.

Slater, A. G., T. J. Bohn, J. L. McCreight, M. C. Serreze, and D. P.Lettenmaier (2007), A multimodel simulation of pan‐Arctic hydrology,J. Geophys. Res., 112, G04S45, doi:10.1029/2006JG000303.

Thomas, G., and P. R. Rowntree (1992), The boreal forests and climate,Q. J. R. Meteorol. Soc., 118, 469–497, doi:10.1002/qj.49711850505.

Viterbo, P., and A. K. Betts (1999), Impact on ECMWF forecastsof changes to the albedo of the boreal forests in the presence of snow,J. Geophys. Res., 104, 27,803–27,810, doi:10.1029/1998JD200076.

Whiteman, C. D. (2000),Mountain Meteorology: Fundamentals and Appli-cations, 355 pp., Oxford Univ. Press, New York.

Wigmosta, M. S., D. P. Lettenmaier, and L. W. Vail (1994), A distributedhydrology‐vegetation model for complex terrain, Water Resour. Res.,30(6), 1665–1679, doi:10.1029/94WR00436.

Wiscombe, W. J., and S. G. Warren (1980), A model for the spectral albedoof snow. I: Pure snow, J. Atmos. Sci., 37, 2712–2733, doi:10.1175/1520-0469(1980)037<2712:AMFTSA>2.0.CO;2.

Wood, E., et al. (1998), The Project for Intercomparison of Land‐SurfaceParameterization Schemes (PILPS) Phase‐2(c) Red‐Arkansas RiverExperiment: 1. Experiment description and summary intercomparisons,Global Planet. Change, 19, 115–135, doi:10.1016/S0921-8181(98)00044-7.

Yucel, I., W. J. Shuttleworth, J. Washburne, and F. Chen (1998), Evaluat-ing NCEP Eta model derived data against observations, Mon. WeatherRev., 126, 1977–1991, doi:10.1175/1520-0493(1998)126<1977:ENEMDD>2.0.CO;2.

Zängl, G. (2004), The sensitivity of simulated orographic precipitation tomodel components other than cloud microphysics, Q. J. R. Meteorol.Soc., 130, 1857–1875, doi:10.1256/qj.03.119.

Zängl, G. (2009), The impact of weak synoptic forcing on the valley‐windcirculation in the Alpine Inn Valley, Meteorol. Atmos. Phys., 105(1–2),37–53, doi:10.1007/s00703-009-0030-y.

M. Barlage, F. Chen, J. Dudhia, D. Gochis, K. Ikeda, R. Rasmussen, andM. Tewari, National Center for Atmospheric Research, PO Box 3000,Boulder, CO 80305, USA. (barlage@ucar.edu)M. Ek and K. Mitchell, National Centers for Environmental Prediction,

5200 Auth Rd., Camp Springs, MD 20746, USA.B. Livneh, Department of Civil and Environmental Engineering,

University of Washington, Seattle, WA 98195, USA.

BARLAGE ET AL.: NOAH SNOWPACK PREDICTION D22101D22101

15 of 15