Abstract

Many studies have developed snow process understanding by exploring the impact of snow model complexity on simulation performance. This paper revisits this topic using several recently developed land surface models, including the Simplified Simple Biosphere Model (SSiB); Noah; Variable Infiltration Capacity (VIC); Community Land Model, version 3 (CLM3); Snow Thermal Model (SNTHERM); and new field measurements from the Cold Land Processes Field Experiment (CLPX). Offline snow cover simulations using these five snow models with different physical complexity are performed for the Rabbit Ears Buffalo Pass (RB), Fraser Experimental Forest headquarters (FHQ), and Fraser Alpine (FA) sites between 20 September 2002 and 1 October 2003. These models simulate the snow accumulation and snowpack ablation with varying skill when forced with the same meteorological observations, initial conditions, and similar soil and vegetation parameters.

All five models capture the basic features of snow cover dynamics but show remarkable discrepancy in depicting snow accumulation and ablation, which could result from uncertain model physics and/or biased forcing. The simulated snow depth in SSiB during the snow accumulation period is consistent with the more complicated CLM3 and SNTHERM; however, early runoff is noted, owing to neglected water retention within the snowpack. Noah is consistent with SSiB in simulating snow accumulation and ablation at RB and FA, but at FHQ, Noah underestimates snow depth and snow water equivalent (SWE) as a result of a higher net shortwave radiation at the surface, resulting from the use of a small predefined maximum snow albedo. VIC and SNTHERM are in good agreement with each other, and they realistically reproduce snow density and net radiation. CLM3 is consistent with VIC and SNTHERM during snow accumulation, but it shows early snow disappearance at FHQ and FA. It is also noted that VIC, CLM3, and SNTHERM are unable to capture the observed runoff timing, even though the water storage and refreezing effects are included in their physics.

A set of sensitivity experiments suggest that Noah’s snow simulation is improved with a higher maximum albedo and that VIC exhibits little improvement with a larger fresh snow albedo. There are remarkable differences in the vegetation impact on snow simulation for each snow model. In the presence of forest cover, SSiB shows a substantial increase in snow depth and SWE, Noah and VIC show a slight change though VIC experiences a later onset of snowmelt, and CLM3 has a reduction in its snow depth. Finally, we observe that a refined precipitation dataset significantly improves snow simulation, emphasizing the importance of accurate meteorological forcing for land surface modeling.

1. Introduction

Snow cover plays an important role in climate because of its high albedo, low thermal conductivity, roughness length, and ability to store water. The accurate parameterization of snow cover feedbacks within the climate system is essential for accurate prediction using general circulation models (GCMs; Yeh et al. 1983; Marshall et al. 1994). During the past couple of decades, many efforts have been dedicated to the development of snow model parameterization schemes (Dickinson et al. 1986; Sellers et al. 1986). The performance of seven snow models coupled with GCMs was investigated by Foster et al. (1996), who showed the models to be consistent in the simulation of snow water equivalent (SWE) but varied in the magnitude and temporal distribution within transitional seasons.

How do snow parameterizations in land surface models (LSMs) contribute to snow model performance differences? Do complicated snow modeling schemes with multilayer, finer vertical resolution and detailed interlayer snow processes tend to produce better results than relatively simple snow schemes? Jin et al. (1999) investigated the relationship between model performance and snow physics using three snow models of different complexity. Their results indicate that the complex snow model gives the best match with the observations, though it was computationally more expensive. Schlosser et al. (2000) and Slater et al. (2001) investigated snow physical processes with 21 land surface schemes. They show that the models as a group can capture the general pattern of accumulation and ablation on an interannual basis but with considerable variability. Boone and Etchevers (2001) suggest that surface energy parameterizations partly contribute to the large differences in total snow water equivalent. Luo et al. (2003) point out that more sophisticated snow models tend to simulate snow more accurately, especially snow ablation. In contrast, Etchevers et al. (2004) suggest that complicated models do not necessarily produce better simulations.

Given the spread of opinions in the literature, this study aims to address the question of required snow model complexity by inspecting how five snow models with different complexity perform against new observations available from the Cold Land Processes Field Experiment (CLPX) performed in Colorado. These five models include i) the Center for Ocean–Land–Atmosphere (COLA) Simplified Simple Biosphere Model (SSiB; Xue et al. 1991), which represents mass and energy transfer between land surface and atmosphere using a resistance formulation; ii) the National Centers for Environment Prediction (NCEP) Noah land surface model (Noah; Ek et al. 2003), which predicts snowpack based on energy–mass balance; iii) the Variable Infiltration Capacity (VIC) Macroscale Hydrologic Model (Liang et al. 1994), which solves the full energy and water balance on a gridcell basis; iv) the Community Land Model, version 3 (CLM3; Dai et al. 2003), which solves snow cover variability with a nested grid; and v) the one-dimensional U.S. Army Cold Regions Research and Engineering Laboratory Snow Thermal Model (SNTHERM; Jordan 1991), which simulates multiphase water and energy transfer processes in snow layers based on mixture theory.

There are several distinct features of this work that set it apart from previous intercomparison studies. Namely, CLPX provides a new benchmark in high-quality meteorological and snowpack measurements covering a wide area but of fine resolution, which is advantageous for monitoring spatial and temporal variations in the growth and melting of the snowpack. These new observations provide a good test bed for evaluating snow scheme parameterizations and enable a better understanding of snow processes and physics. This paper aims to explore the variations of snow accumulation and ablation of each model arising from the model’s parameterization of surface energy and hydrology. In addition, we attempt to investigate whether the snow models show consistent performance at topographically varying locations within one regional study area. Finally, several sensitivity studies are performed to investigate the impact of albedo, forest cover, and precipitation forcing on snow simulation.

An introduction to the snow models and their physics are briefly summarized in section 2. The CLPX measurement and model simulation setup are described in section 3. The model control simulation results and discussion are presented in section 4. The sensitivity tests are shown in section 5. Finally, a summary is given in section 6.

2. Snow schemes of land surface models

a. SSiB model

SSiB is a biophysically based land surface model developed for providing fluxes of radiation, momentum, sensible heat, and latent heat to general circulation models and regional models. It is a simplified version of the Simple Biosphere Model (SiB; Sellers et al. 1986), with reduced physical parameters and improved computational efficiency. The mass and energy transfer between land surface and atmosphere is represented using a resistance formulation,

 
formula
 
formula

where LH and H are latent heat and sensible heat fluxes, respectively; ρ is the air density; Cp is the heat capacity; λ is latent heat of vaporization Tm, qm; and Ta, qa are temperature and specific humidity at the reference height and canopy air space, respectively; r is aerodynamic resistance, which depends on vegetation height, surface roughness length, and wind speed. The surface and deep soil temperature is solved based on the force–restore method (Deardorff 1978).

SSiB includes a relatively simple snow submodel. Snow is accumulated on the ground and canopy when the surface air temperature is below freezing. The downward solar radiation is attenuated through the canopy with multiple scattering through the canopy and snowpack, but radiative attenuation within the snow layer is ignored because the snowpack is modeled as one bulk layer. Snow albedo is controlled by the spectral and angular distribution of solar radiation incident on the surface (i.e., direct or diffuse, and infrared or visible), surface type (soil or vegetation), the solar zenith angle, and snow cover. Snow starts to melt or the melting water and canopy water freeze if canopy/ground temperature is above or below freezing point, respectively.

b. Noah model

The Noah LSM was originally developed in the 1980s (Mahrt and Pan 1984). It has been implemented in operational weather and climate models as a result of its moderate complexity and computational efficiency. The community 2.7.1 version of the one-dimensional column of Noah is employed for this study. Noah simulates land surface temperature, snow depth, snow water equivalent, canopy water content, the components of the surface energy balance and the surface water balance, and the evolution of soil temperature and soil moisture. The snow parameterization is based on the energy and mass balance of snowpack with snow compaction and subgrid variability components. Snow albedo is formulated as a composite of a snow-covered and snow-free surface,

 
formula

where α, α0, and αs are the actual, snow-free, and maximum snow surface albedo, respectively, and σs is the snow cover fraction. The upper bound of snow albedo is set to the prescribed maximum albedo (0.44) under snow conditions. The shading effect of vegetation on albedo is also taken into account. The vegetation density [small versus large leaf area index (LAI)], and coverage (sparse versus dense) are considered in Noah. Moreover, snow interception, drip, and melt on canopy surfaces are neglected in the model. Snow surface roughness is not modified in the presence of vegetation.

The model explicitly solves liquid water retention and percolation within the snowpack. It also accounts for changes in the hydraulic and thermal properties of snow due to meltwater refreezing. The affect of frozen ground on reducing infiltration reduction is included based on probabilistic averaging of spatially variable ice content of the soil profile (Koren et al. 1999).

c. VIC model

The VIC model (Liang et al. 1994) is a macroscale hydrologic model, which solves the full energy and water balance on a gridcell basis. For our study, version 4.0.5 was used with three vertical soil layers. In general, VIC simulates the physical processes of moisture state changes and the related partitioning of energy fluxes. VIC employs two snow layers of variable thickness. The thin upper snow layer is used to solve the surface energy balance, whereas the lower layer is used to simulate deeper snowpacks. VIC runs full energy physics, which means that snowpack thermodynamic processes are coupled into the energy transfer processes of the entire model (Cherkauer and Lettenmaier 1999). The shortwave radiation is attenuated through canopy by a vegetation-dependent parameter defined as

 
formula

The aerodynamic resistance controls the interaction between the atmosphere and the vegetation surface, using a similar expression with Eq. (1). The canopy intercepts both solid and liquid precipitation, which in turn influences the snowpack mass balance. Snow albedo is expressed as (Wigmosta et al. 1994)

 
formula
 
formula

where N is the age of the snow surface in days since the last snow storm, and αf is the predefined fresh snow albedo (0.85). The melt period is defined as the period after 1 March, when the cold content of the snowpack is greater than zero. The temperatures of the soil column, soil surface, and snowpack layers are solved from heat transfer/balance equations for the entire system including soil, snowpack, vegetation, and air together with the corresponding water balance equations. To compute snow cover extent, VIC uses both subgrid vegetation tiling and elevation banding and assumes that any snow present fully covers the tile. The snowpack depth depends on SWE and the snowpack density.

d. CLM3

CLM3 focuses on the biogeophysics of the land surface and includes vegetation dynamics and river routing modules (Anderson 1976; Jordan 1991; Dai and Zeng 1997). The model has 10 soil layers and 5 snow layers, depending on the snow depth. Snow can also be present in the model without any snow layer if the snow depth is less than 0.01 m. CLM3 applies a two-stream approximation for radiative transfer calculation, the attenuation of solar radiation by canopy is also taken into account. Snow albedo is based on Dickinson et al. (1993) 

 
formula
 
formula

where αdiff,0 is the fresh snow albedo, C is an empirical constant, Fage is a transformed snow age used to give the fractional reduction of snow albedo as a result of snow aging for solar zenith angle less than 60°, μ is solar zenith angle, and the function f (μ) is a factor between 0 and 1 that is needed to increase the snow albedo as a result of the solar zenith angle exceeding 60°.

The model calculates surface vertical kinematic fluxes of momentum, sensible and latent heat based on the Monin–Obukhov similarity theory. For instance, latent and sensible heat fluxes employ an aerodynamic resistance-based formulation similar to Eq. (1). For the vegetated surface, the turbulent heat flux is the sum of the fluxes from the vegetation and the ground. CLM3 uses the principle of energy conservation in the form of the continuity equation to calculate soil and snow temperatures for all layers and uses a discretized version of Darcy’s law for vertical downward flow of water within soil layers (Oleson et al. 2004). The vegetation effect is accounted for in snow accumulation, melt and interception through fall and drip by canopy. CLM3 calculates the water transfer between snow layers, infiltration, surface runoff, subsurface drainage, and redistribution within the soil column, whereas the water vapor transport within the snowpack is neglected.

e. SNTHERM

The public version of the single-column SNTHERM model has been upgraded since its first release in 1989 (Cline 1997; Jordan et al. 1999). SNTHERM employs a mixture theory formulation to calculate the mass balance for the phases of water, dry air, and dry soil conditions. Mass conservation is maintained by liquid water and diffusive water vapor fluxes, sublimation, and phase changes relating to melt processes. SNTHERM also considers grain growth, densification and settlement, and energy exchange between the bottom layer of the snowpack and the top soil layer. The heat balance equations use specific enthalpy to represent the heat content within the snowpack layers and the transfer of heat through vapor diffusion, heat conduction, and downward shortwave radiation. The albedo in SNTHERM is based on Marks’s (1988) 

 
formula

where fvis is the fraction of visible to total incident radiation, αvis and αnir are clear-sky albedo for visible and near-infrared radiation, respectively, which are given by

 
formula
 
formula

where r is the optically equivalent grain radius, fdir is the fraction of direct to total incident radiation, and θz is the solar zenith angle. The solar radiation is exponentially attenuated within the snowpack. The model can adjust for variations in the snow physics, and the snow layers adapt to those changes, making the amount of layers variable (Jordan 1991).

3. Study sites, measurement, and model setup

Meteorological measurements were collected at 10 sites within the CLPX small regional study area (SRSA), located in north-central Colorado (39.5°–41°N, 105°–107.5°W) between 20 September 2002 and 1 October 2003 (Elder and Goodbody 2004). Specifically, we focus on three CLPX sites for our primary study areas: Fraser Alpine (FA), Fraser Experimental Forest headquarters (FHQ), and Rabbit Ears Buffalo Pass (RB). Each site is characterized by its typical physiographic attributes. The elevation at FA (39°51′N, 105°52′W), the highest CLPX site, is 3557 m, and it is covered with alpine tundra, generally north-facing with moderate relief. FHQ (39°54′N, 105°53′W) is characterized with lower elevation (2760 m) and subalpine forest coverage. RB (40°32′N, 106°41′W, 3144-m elevation) is located in dense coniferous forest, interspersed with open meadows and featuring low rolling topography.

Corresponding to their distinctive topographical features, the three chosen CLPX sites exhibit large variation in seasonal variation of wind speed, specific humidity, air temperature, and radiation (Fig. 1). It is evident that air temperature, specific humidity, and incoming solar radiation exhibit a seasonal cycle with the minimum in winter and the maximum in summer. Downward longwave radiation shows small seasonal changes over the entire observation period. Wind at FA and RB is stronger in winter and spring and weaker in summer. The average wind speeds for RB, FA, and FHQ are 3.1, 5.5 and 0.6 m s−1. FHQ has a nearly constant wind speed that is much lower than that of RB and FA. The air is drier and colder at FA, which is understandable because the altitude at FA is higher than RB and FHQ. The air temperature is below the freezing point in winter and spring at the three sites. FHQ has the highest air temperature, and the maximum air temperature difference between FHQ and FA is 5°C in March. The incoming shortwave radiation for FA is the largest, but the air temperature at FA is also the lowest. RB and FHQ show similar patterns in receiving the shortwave radiation from winter to spring. The downward atmospheric longwave radiation is largest at RB and smallest at FA because of its lower air temperature.

Fig. 1.

Monthly atmospheric forcing at RB (open circle), FA (cross), and FHQ (closed circle) from October 2002 to September 2003. (a) Wind speed (m s−1), (b) specific humidity (g kg−1), (c) air temperature (°C), and (d) incoming shortwave (solid) and downward longwave radiation (dash; W m−2).

Fig. 1.

Monthly atmospheric forcing at RB (open circle), FA (cross), and FHQ (closed circle) from October 2002 to September 2003. (a) Wind speed (m s−1), (b) specific humidity (g kg−1), (c) air temperature (°C), and (d) incoming shortwave (solid) and downward longwave radiation (dash; W m−2).

Meteorological towers in FA and RB are located close to the center of each intensive study area (ISA). The FHQ meteorological tower is located close to the Local Scale Observation Site (LSOS) within the Fraser mesocell study areas (MSAs). The CLPX meteorological observations include air temperature, relative humidity, radiation, wind speed and direction, solar and longwave radiations, soil temperature, and soil moisture as well as snow depth and snow surface temperature with a 10-min temporal resolution (Table 1). Air temperature and humidity were measured at 2 m above the ground surface for Fraser and 4 m for Rabbit Ears, respectively, whereas wind speed is obtained at 10 m. Snow pit data were also collected periodically at or near each of the meteorological towers, including snow depth, SWE, and profiles of snow density and snow temperature with 10 cm of vertical distance.

Table 1.

List of instruments used in CLPX meteorological measurement.

List of instruments used in CLPX meteorological measurement.
List of instruments used in CLPX meteorological measurement.

Driven by the same observed meteorological forcing, the five snow models were run without being coupled to any atmospheric model component. They also use the same initial conditions and have as similar soil and vegetation parameters as possible (Table 2). If there was no snow on the ground or the canopy at the start of the simulation periods, then initialization should have little influence on snow simulations. The observed 10-min meteorological variables were averaged to 1-h intervals to reduce instrument noise and to conform to standard model input requirements. Because precipitation is not provided by CLPX, we used the hourly 1/8° merged gauge–radar precipitation product available from the North American Land Data Assimilation System project (NLDAS; Cosgrove et al. 2003). The integration periods for the five models are from 23 September 2002 to 27 September 2003 for RB, from 15 November 2002 to 29 September 2003 for FHQ, and from 19 December 2002 to 27 September 2003 for FA. The model time step is 1 h for Noah and VIC and 30 min for the other three models.

Table 2.

Summary of key parameters for model initialization.

Summary of key parameters for model initialization.
Summary of key parameters for model initialization.

4. Results and discussion

a. RB

The snow depth, SWE, snow density, snow albedo, and snow temperature fields from the five models are compared with acoustic depth sounder snow depth (obs) and snow pit measurements (SPs), which are shown in Fig. 2. The snow starts to accumulate in October and gradually builds up a deep snowpack, with the maximum snow depth reaching 3.5 m. There are eight small snowmelt events followed by major snowmelt in mid-May, until the snow finally disappears on 25 June. All five models have the ability to accumulate snow on the ground during the winter snowfall and ablate it completely before mid-June. It is evident that the models underestimate snow depth and SWE (Figs. 2a and 2b), which could be partially attributed to errors in the atmospheric forcing, such as the NLDAS precipitation. SSiB and Noah show significant snow melting in mid-March, and the simulated snowpack completely ablates by the end of May, almost one month prior to that of the observation. VIC, CLM3, and SNTHERM realistically capture the timing of snowmelt but they all produce an earlier disappearance of snow. The simulated snow density is calculated as the ratio of SWE and snow depth (Fig. 2c). As the accumulation season progresses and the snowpack becomes deeper, VIC and SNTHERM realistically reproduce the snow density increase. The snow density in SSiB is a constant value of ∼200 kg m−3. Noah and CLM3 give smaller snow density values, but the CLM3 snow density approaches a maximum of 502 kg m−3 near the time of snowpack ablation (Fig. 2c). The observed “effective” albedo is calculated from the ratio of the ground measured downward and upward shortwave radiation (Fig. 2d). SSiB and Noah have lower than observed snow albedo and hence absorb more incoming solar radiation (Fig. 3a), producing a warmer snow surface, which leads to a faster snowmelt. In contrast, high snow albedo values in VIC, CLM3, and SNTHERM cause less solar radiation absorption, resulting in less energy for snowmelt. It is not surprising to observe that the snow accumulation is sustained longer in these models. The warmer SSiB and Noah snow surfaces emit larger outgoing longwave radiation in early winter, which ultimately cools the surface. During spring, the net longwave radiation from each model is in good agreement with the observations (Fig. 3b). The stronger shortwave radiations in SSiB and Noah provide an available energy source for large turbulent heat fluxes during late spring (Fig. 3c). The transfer of sensible heat from the atmosphere to the snow surface in VIC is supplied to melt the snowpack. These simulations reveal that snow sublimation is much greater than evaporation and plant transpiration during the snow accumulation period. Therefore, variation in wintertime latent heat is closely related to the change in snow sublimation. It can be seen that latent heat fluxes vary considerably (Fig. 3d), although each model produces very similar snow depth and SWE in the early snow season. Because of the lack of ground measurements, it is hard to verify whether snow sublimation or snowmelt dominates snow ablation during the snow season. Strong latent heat fluxes in Noah indicate that snow sublimation accounts for earlier snow ablation. Meanwhile, the latent heat flux in VIC and SNTHERM are marginal, which implies that snowmelt instead of sublimation plays a major role in snow ablation. On the contrary, snow sublimation and snowmelt (values not shown) are equivalently important in SSiB and CLM3 during winter, but snowmelt dominates snow removal in late spring, which directly contributes to large peak runoff in both models (Fig. 3e).

Fig. 2.

Daily (a) snow depth (m), (b) SWE (mm), (c) snow density (kg m−3), (d) snow albedo, and (e) snow temperature (°C) from SSiB (solid), Noah (long dash), VIC (short dash), CLM3 (dot), and SNTHERM (dot dash) compared with observed snow depth (obs, solid gray line) from acoustic depth sounder and snow pit SWE (SP, solid circle) at RB from 2 Oct 2002 to 30 Jun 2003.

Fig. 2.

Daily (a) snow depth (m), (b) SWE (mm), (c) snow density (kg m−3), (d) snow albedo, and (e) snow temperature (°C) from SSiB (solid), Noah (long dash), VIC (short dash), CLM3 (dot), and SNTHERM (dot dash) compared with observed snow depth (obs, solid gray line) from acoustic depth sounder and snow pit SWE (SP, solid circle) at RB from 2 Oct 2002 to 30 Jun 2003.

Fig. 3.

Monthly mean of (a) net shortwave (W m−2), (b) net longwave (W m−2), (c) sensible heat (W m−2), (d) latent heat (W m−2), and (e) runoff (mm) from SSiB (solid), Noah (long dash), VIC (short dash), CLM3 (dot), and SNTHERM (dot dash) compared with observation (solid line with open circle) at RB from October 2002 to August 2003.

Fig. 3.

Monthly mean of (a) net shortwave (W m−2), (b) net longwave (W m−2), (c) sensible heat (W m−2), (d) latent heat (W m−2), and (e) runoff (mm) from SSiB (solid), Noah (long dash), VIC (short dash), CLM3 (dot), and SNTHERM (dot dash) compared with observation (solid line with open circle) at RB from October 2002 to August 2003.

In this paper, the simulated runoff refers to the sum of the surface and subsurface runoff components. Substantial runoff in the SSiB emerges in February, which is one month prior to the increase of the observed soil moisture increase on 4 March 2003. This is because SSiB ignores liquid water retention capacity; thus, the meltwater immediately drains out of the snowpack once the snow melts. Noah, VIC, CLM3, and SNTHERM exhibit more delayed runoff than SSiB, primarily as a result of the inclusion of water storage and refreezing effect within the snowpack. However, only SNTHERM reproduces the runoff timing, and Noah, VIC, and CLM3 show disparity with the observations. The results suggest that the peak runoff in Noah, CLM3, and SNTHERM appear to be consistent among these models in May, but the magnitude varies drastically. The uncertainty in runoff simulation urges that it is essential to understand the snowmelt process and determine the drainage of meltwater that is restricted by the storage capacity. In addition to snowmelt, the models show disparity in the partitioning of meltwater into runoff and infiltration components. Noah and VIC indicate a substantial infiltration of meltwater, which recharges the soil moisture during snow ablation.

b. FHQ

Snow had already accumulated on the ground at FHQ when measurements began on 15 November 2002. There was only one major accumulation and melt cycle that occurred in mid-March that was not long enough to melt the entire snowpack (Fig. 4a). The five models are initialized with the same measured snow depth on 15 November, but they show significant discrepancy in snow depth and SWE. SSiB is close to the observations during the early snow accumulation, but it overestimates snow depth and SWE in early spring, which ultimately delays the onset of melting (Fig. 4b). In Noah, the snow depth and SWE are substantially underestimated. VIC and CLM3 agree well with the observations during the early snow season; however, they both underestimate snow depth and SWE in midwinter and exhibit early snowmelt. SNTHERM realistically reproduces snow accumulation and ablation. It is noted that the snow density in Noah is close to the observations, although the snow depth and SWE are poorly simulated (Fig. 4c). In addition to Noah, CLM3 also approaches the observed snow density variation, but VIC and SNTHERM apparently overestimate snow density, which is likely a result of predicted excessive overburden within the snowpack, contributing to higher SWE values. There are considerable differences in snow temperature among the snow models, which also reflects the variety in the simulated net longwave radiation.

Fig. 4.

Same as Fig. 2 but at FHQ from 15 Nov 2002 to 31 May 2003.

Fig. 4.

Same as Fig. 2 but at FHQ from 15 Nov 2002 to 31 May 2003.

SSiB and CLM3 exhibit comparable net radiation, but they show substantial discrepancy in the partitioning of the net radiations into turbulent heat fluxes (Fig. 5). SSiB provides a large sensible heat flux to the atmosphere, which eventually cools the snow surface and prevents the snowpack from melting. CLM3 produces a large latent heat flux to balance strong snowpack sublimation. As found at RB, SSiB and CLM3 snow sublimation and snowmelt (values not shown) are comparable during the snow accumulation phase, whereas snowmelt is dominant in late spring. In Noah, a large portion of shortwave radiation and a strong transfer of sensible heat from the atmosphere together warm the surface and melt the snow, leading to persistent snow ablation and a lack of snow accumulation on the ground. In addition, Noah exhibits large latent heat fluxes but small runoff, highlighting the potential role of sublimation during the snow ablation phase. The smaller net radiation in VIC is a result of its larger snow albedo; this additional surface available energy is primarily supplied to melt snow. It is not surprising to see that VIC generates near zero turbulent heat fluxes, hence marginal snow sublimation. Similar to RB, a large portion of meltwater in VIC infiltrates into the soil and is released into the atmosphere when the surface warms, whereas the melting water mostly converts to runoff in SSiB, CLM3, and SNTHERM. The initial significant runoff in these models is earlier than the observed soil moisture increase, which occurred on 9 March 2003. Similar to RB, the peak runoff in Noah, CLM3, and SNTHERM almost occurs around the same time, whereas SNTHERM experiences a larger period of snowmelt.

Fig. 5.

Same as Fig. 3 but at FHQ from November 2002 to June 2003.

Fig. 5.

Same as Fig. 3 but at FHQ from November 2002 to June 2003.

c. FA

Snow measurements were available from 19 December 2002 at FA. The snowpack is shown to be quite shallow in winter and early spring, and the maximum snow depth is less than 0.2 m (Fig. 6a). Until mid-March, deep snow accumulates on the ground because of the occurrences of strong precipitation events. The several snowpack removal events occurring in midwinter may not result from direct snowmelt because the temperature is very low. They are likely caused by high wind–induced snowdrift during snowfalls. As a matter of fact, snow pit–measured snow depth and SWE exhibit the spatial variability resulting from wind blowing (not shown here). The snow variability caused by snowdrift complicates snow simulation and presents a challenge for model physics. This point is further confirmed by the considerable discrepancy in the simulated snow depth and SWE from each model. SSiB, Noah, and CLM3 consistently overestimate the snow depth and show an early disappearance of snowpack. The SWE in Noah and CLM3 is close to the observations. VIC and SNTHERM realistically capture the duration of snow cover, but they overestimate the depth of snowpack and SWE. At the same time, they are in good agreement with the measured snow density (Fig. 6c). Despite the differences in snow accumulation and ablation, the models are consistent with the observed snow temperature.

Fig. 6.

Same as Fig. 2 but at FA from 19 Dec 2002 to 31 May 2003.

Fig. 6.

Same as Fig. 2 but at FA from 19 Dec 2002 to 31 May 2003.

SSiB and Noah receive more net shortwave radiation than the other models as a result of smaller snow albedo (Fig. 7a); therefore, they provide more energy to warm the snow surface. It is evident that all the models agree well with the observed net longwave radiation during the early snow season (Fig. 7b). There are considerable increases of latent heat transfer (Fig. 7d) caused by the enhancement of the water vapor pressure gradient between the snow and the underlying drier and colder air, combined with the prevailing high wind at FA. The total snow sublimation in Noah reaches 285 mm during the snow season, which is much larger than snowmelt, further reinforcing the significance of sublimation. In contrast to RB and FHQ, VIC has larger latent heat fluxes as a result of the increase in snow sublimation caused by the higher winds at FA. The high latent heat flux in SSiB and CLM3 correspond to large snow sublimation, indicating that snowmelt is secondary in importance in ablating snow during winter and early spring. SSiB and CLM3 experience their initial peak runoff in December (Fig. 7e), when thin snow is on the ground and only a small amount of meltwater is available. Therefore, it is the liquid precipitation that contributes to the early peak runoff. Because CLM3 has a higher threshold temperature (∼2.5°C) than SSiB (∼0°C), it is not surprising to notice that the peak runoff in CLM3 is larger than that in SSiB. There is a second peak runoff appearing in SSiB and CLM3 in April, which is mainly attributed to a large amount of snowmelt. In SNTHERM, latent heat is comparably larger at FA than at RB and FHQ, but it is still smaller than other models. The peak runoff in March from SNTHERM corresponds to a sharp decrease of SWE caused by a large amount of snowmelt. At FA, much of the meltwater in Noah and VIC percolates through a highly permeable sandy soil, which reduces the total amount of runoff compared with the other sites.

Fig. 7.

Same as Fig. 3 but at FA from December 2002 to August 2003.

Fig. 7.

Same as Fig. 3 but at FA from December 2002 to August 2003.

The significant discrepancy at FA suggests that several issues might influence the performance of snow simulations: i) thin snow (mean snow depth ∼10 cm), ii) lack of blowing snow processes, iii) snow memory, and iv) uncertainties in the forcing.

5. Sensitivity experiments

The previous section focuses the impact of model complexity on snow simulation performance. The results indicate that snow models are able to accumulate snow on the ground during snowfall and ablate it through snow sublimation and snowmelt, but they show disparities with the measurements. The simple snow model, such as SSiB, exhibits an inability to store water within the snowpack, owing to the lack of water retention. Noah overestimates surface temperature resulting from a low albedo. Although simple models may not be able to represent the complicated snow physics, it is possible that simple models can be tuned to agree with observations, but the tunable parameters could be case or region dependent. In addition to the simple models, the more complicated models, such as CLM3 and SNTHERM, also exhibit a deficiency in snow simulation, such as runoff timing. Regardless of the uncertainty in snow physics, external factors also account for the model performance, such as the atmospheric forcings. In fact, models show inconsistencies in their snow fields for each CLPX site, which can also be attributed to different atmospheric conditions. For example, snow sublimation in VIC is marginal in RB and FHQ, but it dominates in the snow ablation phase at FA. This site-dependent model performance further reveals the differences in model physics. Therefore, we examine the impact of albedo, vegetation, and precipitation on model performance in this section.

a. Albedo

This subsection examines the effect of albedo on snowpack simulation by changing the predefined maximum albedo in Noah and fresh snow albedo in VIC; albedo formulations in SSiB, CLM3, and SNTHERM complicate the procedure by artificially manipulating snow albedo. The maximum albedo for the RB area is 0.47 according to NCEP’s annual maximum snow albedo climatology dataset (Robinson and Kukla 1985). VIC originally uses 0.85 for fresh snow albedo and includes age effect for snow albedo calculations. The “effective” albedo is roughly 0.79 based on the calculation using CLPX upward and downward shortwave radiation. We set the maximum albedo for Noah and fresh snow albedo for VIC, with the arbitrary values of 0.5, 0.7, and 0.9 (Figs. 8b and 8d). The maximum snow depth and SWE increase with the enlarged snow albedo values for both models (Figs. 8a and 8c). The depth of snowpack in Noah considerably increases with the enhancement of albedo, with the maximum difference between peak snow depths to be about 1 m. When albedo equals 0.7, snow vanishes coincidently with the observations. Increasing the snow albedo results in a lower amount of received solar radiation, thus reducing the heating for the snowpack and prolonging the duration of snow cover. This leads to a slower snowmelt process, especially toward the end of the snow season. VIC displays a slight change in snow depth during winter and early spring, indicating the trivial impact of albedo on early snow accumulation. Beginning from late spring, substantial solar radiation reaches the snow surface, as shown in Fig. 3a. The snow surface with the smaller albedo receives more solar radiation, which in turn results in an earlier onset of snow ablation. Noah appears to be quite sensitive to the maximum albedo change, whereas VIC exhibits little sensitivity to predefined fresh snow albedo during the early accumulation period. Therefore, it is important to tune the Noah albedo to fit the observations, hence enhancing model performance. However, the tuned snow albedo depends on specific snow events or site characteristics. For instance, the observed average snow albedo is 0.79 at RB, 0.59 at FHQ, and 0.44 at FA. It is clear that the uncertainty of snow albedo impairs the accuracy of snow model. Therefore, it is crucial to improve the parameterization of snow albedo in future model development.

Fig. 8.

Comparison of snow depth (m) and snow albedo with observations (solid gray line) when maximum albedo and fresh snow albedo are set to 0.5 (long dash), 0.7 (dot), and 0.9 (dot dash), respectively, for (a), (b) Noah and (c), (d) VIC at RB from 3 Oct 2002 to 31 Jul 2003.

Fig. 8.

Comparison of snow depth (m) and snow albedo with observations (solid gray line) when maximum albedo and fresh snow albedo are set to 0.5 (long dash), 0.7 (dot), and 0.9 (dot dash), respectively, for (a), (b) Noah and (c), (d) VIC at RB from 3 Oct 2002 to 31 Jul 2003.

b. Forest cover

The RB site consists of forest (72.2%), rangeland (14.3%), and pasture (8%). The simulations discussed in the previous subsection are performed in a forest clearing where the RB tower instrumentation is installed. Considering the presence of forest cover within ∼50 m of the RB measurement location, we examine the impact of the heterogeneity of vegetation on snow accumulation and snowmelt. To evaluate the effect of vegetation on snow simulation, we conduct experiments with forest cover for the RB location.

The surface heterogeneity is not included in this version of SNTHERM, so we only use the remaining snow models to assess the changes in snow depth, SWE (Fig. 9), latent heat, sensible heat, and runoff (Fig. 10). Compared with Fig. 2a, SSiB simulates a deep snowpack and approaches the observed snow depth in this experiment. The net radiation in SSiB supplies the large upward transport of sensible heat instead of melting the snow, which is not surprising to note a large negative peak runoff in April. Corresponding to the substantially decreased latent heat flux in SSiB, snow sublimation is reduced during late spring. Therefore, the snow accumulation period is prolonged, which in turn results in a deep snowpack. Noah produces only a trivial change in snow depth, energy fluxes, and runoff. As described in section 2b, Noah ignores vegetation type and does not take into account snow interception, drip, and melt by the canopy. Therefore, Noah exhibits marginal spatial variability of snow depth and SWE on the forest floor. Snow accumulation in VIC is similar over grassland and forest, but the snowpack over the forest melts more slowly during the melt season. This is primarily caused by less radiation received in the forest scenario because shortwave radiation is attenuated by the forest canopy. In the presence of a canopy, there are two offsetting radiative fluxes: attenuation of shortwave transmission and enhancement of longwave irradiance. VIC agrees with Link and Marks (1999) that shortwave reduction plays a dominant role. Sicart et al. (2004) showed that longwave radiation increases compensate for shortwave reduction with increasing canopy coverage. CLM3 shows decreased snow depth and SWE, accompanied by a considerable increase of latent heat and decrease of runoff in March. This reduced net precipitation under forest canopy is attributed to the large snow interception and sublimation by forest canopy, which is consistent with Hardy et al. (1997) and Gelfan et al. (2004). The canopy influences mass and energy exchange during snow accumulation and snowmelt. The snow models exhibit large discrepancies in addressing the impact of canopy on radiative flux and snow interception. It is important to accurately portray the role of vegetation in simulating snow processes for future model development.

Fig. 9.

Comparison of daily (a) snow depth (m) and (b) SWE (mm) for SSiB (solid line), Noah (long dash), VIC (short dash), and CLM3 (dot) for forest vegetation type with obs snow depth (solid gray line) at RB from 3 Oct 2002 to 5 Jul 2003.

Fig. 9.

Comparison of daily (a) snow depth (m) and (b) SWE (mm) for SSiB (solid line), Noah (long dash), VIC (short dash), and CLM3 (dot) for forest vegetation type with obs snow depth (solid gray line) at RB from 3 Oct 2002 to 5 Jul 2003.

Fig. 10.

The differences of (a) sensible heat (W m−2), (b) latent heat (W m−2), and (c) runoff (mm) between forest cover and grassland from SSiB (solid line), Noah (long dash), VIC (short dash) and CLM3 (dot) at RB during October 2002 to August 2003.

Fig. 10.

The differences of (a) sensible heat (W m−2), (b) latent heat (W m−2), and (c) runoff (mm) between forest cover and grassland from SSiB (solid line), Noah (long dash), VIC (short dash) and CLM3 (dot) at RB during October 2002 to August 2003.

c. Improved precipitation

As mentioned in section 2, precipitation is not directly measured at the three CLPX sites, so NLDAS precipitation is used to force the snow models in our previous experiments. NLDAS precipitation is based on radar observations that are suspicious in mountain terrain and gauges that are sparse in this region, which could lead to inadequate precipitation estimates. We attempt to reconstruct a precipitation dataset (called PSD) based on snow depth measurements; the hourly snow depth difference is converted to precipitation using a fresh snow density derived from CLPX snow pit measurements. If the difference is less than zero, NLDAS precipitation is blended in. However, both monthly averaged NLDAS and PSD are smaller than the snowpack telemetry (SNOTEL) precipitation (the daily SNOTEL site that is about 500 m away from RB) during winter and spring (Fig. 11). The maximum difference between NLDAS and the SNOTEL gauge measurement is 136 mm in March, which is a significant snow accumulation period. We were motivated to use daily SNOTEL measurements while also accounting for the hourly variation of precipitation for model forcing. Two different sets of precipitation datasets based on SNOTEL precipitation are reconstructed to test the impact of improved precipitation on simulated snow depth and SWE: one precipitation dataset is based on NLDAS hourly variation, but it is scaled to match the monthly averages of the SNOTEL precipitation observations (NLDAS–obs). The other precipitation dataset is calculated from the snow depth increases (PSD) measured at RB and also scaled to the observed SNOTEL monthly precipitation (PSD–obs).

Fig. 11.

Monthly precipitation (mm) from (left) daily gauge measurement, (middle) NLDAS, and (right) PSD at RB during October 2002 to August 2003.

Fig. 11.

Monthly precipitation (mm) from (left) daily gauge measurement, (middle) NLDAS, and (right) PSD at RB during October 2002 to August 2003.

It is obvious that the models realistically reproduce snow accumulation and snow ablation with NLDAS–obs (Figs. 12a and 12b) compared with Fig. 2. SSiB captures snow accumulation but shows early snow disappearance. VIC and SNTHERM show consistent snowmelt and closely approach the observed SWE. CLM3 underestimates snow depth in March and snowmelt starts earlier, with ablation occurring later than the observed. The profiles of snow density and snow temperature in CLM3 and SNTHERM come close to the snow pit measurement when forced with improved precipitation observations (Fig. 13). CLM3 captures the vertical variation of snow density and snow temperature but with some underestimation (Fig. 13b). Because SNTHERM can simulate deep snow with hundreds of vertical layers, it gives a more realistic and detailed vertical variation compared with CLM3. However, SNTHERM generates a colder snow surface and warmer snow bottom, which leads to a large amount of snowmelt and results in larger snow density at the bottom snow layer.

Fig. 12.

Comparison of daily snow depth (m) and SWE (mm) using (a), (b) NLDAS–obs and (c), (d) PSD–obs for SSiB (solid line), Noah (long dash), VIC (short dash), CLM3 (dot) and SNTHERM (dot dash) with observed snow depth (solid gray line) and SP SWE (solid circle) at RB from 3 October 2002 to 9 Jul 2003.

Fig. 12.

Comparison of daily snow depth (m) and SWE (mm) using (a), (b) NLDAS–obs and (c), (d) PSD–obs for SSiB (solid line), Noah (long dash), VIC (short dash), CLM3 (dot) and SNTHERM (dot dash) with observed snow depth (solid gray line) and SP SWE (solid circle) at RB from 3 October 2002 to 9 Jul 2003.

Fig. 13.

Profiles of snow density (kg/m3) and snow temperature (°C) at RB for (a), (b) CLM3 and (c), (d) SNTHERM using NLDAS–obs with (e), (f) snow pit measurement on 17 Dec 2002 (solid line), 16 Jan 2003 (dash) and 29 Mar 2003 (dot dash).

Fig. 13.

Profiles of snow density (kg/m3) and snow temperature (°C) at RB for (a), (b) CLM3 and (c), (d) SNTHERM using NLDAS–obs with (e), (f) snow pit measurement on 17 Dec 2002 (solid line), 16 Jan 2003 (dash) and 29 Mar 2003 (dot dash).

All five models perform better when using the PSD–obs precipitation (Figs. 12c and 12d) than when using NLDAS precipitation. VIC and SNTHERM realistically capture the snow accumulation, the timing of melting, and the disappearance of snow. This study further confirms that the precipitation could be reconstructed using snow depth measurements.

In this study, the two modified precipitation datasets substantially improve model performance. Mote et al. (2003) also corrected the underestimation of snow depth and SWE by SNTHERM using the calibrated precipitation. Therefore, the model accuracy heavily depends on the forcing data.

6. Conclusions

The offline simulations using the recently developed models SSiB, Noah, VIC, CLM3, and SNTHERM are compared with CLPX field measurements to explore the impact of snow model complexity on snow cover simulation and to evaluate snow model performance at different geographical locations. The models display variations in simulating snow evolving processes when forced with the same meteorological observations, initial conditions, and similar soil and vegetation parameters.

All five models have the ability to capture basic snowpack features but with significant disagreement with the observations, which could result from uncertainty in model parameterizations or errors in atmospheric forcing. The degraded snow simulation using a simple model is largely caused by the inherent deficiency in representing internal snow physics. For instance, SSiB agrees well with other snow models in simulating snow depth at RB but underestimates SWE by assuming a constant snow density. In Noah, a maximum albedo is defined in the calculation of snow albedo, which is proved to be too small in the sensitivity test. Consequently, the net shortwave radiation is overestimated and more energy is supplied to ablate snow, resulting in an underestimation of snow depth and SWE, especially at RB and FHQ. Overall, the multilayer snow models exhibit better performance but sometimes also have performance issues. For example, the simpler parameterization in VIC agrees well with the more complicated SNTHERM in simulating snow depth, SWE, and snow density, and it even performs better than the complex CLM3, especially at FHQ and FA. However, the runoff timing in VIC, CLM3, and SNTHERM is not realistically captured, although the water storage and refreezing effects are included. This implies that some uncertainty is associated with snow melting parameterization, which needs to be improved in future model studies.

Generally speaking, snow temperature and net longwave radiation in SSiB are simulated well compared with observations, and the simulated snow depth during the snow accumulation period is consistent with the more complicated CLM3 and SNTHERM models. Noah is consistent with SSiB in simulating snow accumulation and ablation at RB and FA except at FHQ, where Noah apparently underestimates snow depth and SWE. VIC and SNTHERM are in good agreement with each other and realistically reproduce snow density and net radiation. CLM3 is consistent with VIC and SNTHERM during snow accumulation, but it shows early snow disappearance at FHQ and FA. Overall, the models exhibit consistent performance at RB with deep snow and FHQ with intermediate depth of snowpack. However, they display poor simulation at FA with thin snow. This may be caused by neglecting blowing snow physics, which accounts for most of the spatial variability of the snow distribution at FA.

The sensitivity experiments indicate that Noah’s snow simulation improves as its maximum predefined albedo increases toward observed values, but VIC shows little sensitivity to the fresh snow albedo. It is also found that the impact of vegetation on snow simulation is highly variable, depending on the specific snow model treatment of vegetation. SSiB substantially increases snow depth and SWE, Noah and VIC show slight changes, with VIC delaying snowmelt, and CLM3 considerably reduces its snowpack depth. The discrepancy of each model in portraying vegetation effects highlights the need to realistically address the vegetation impact on snow mass and energy exchange for future model improvement. It is demonstrated that all the models improve snow simulation using gauge-based precipitation forcing. It is important to thoroughly understand snow accumulation and ablation processes, precisely represent key snow physics in model parameterization, and obtain accurate atmospheric forcing to improve snow model performance.

Acknowledgments

The authors thank the CLPX measurement team, Dr. Kelly Elder, and the U.S. Forest Service for their provision of the observations used in this study. Xia Feng thanks Dr. ZhiChang Guo for his generous help with the SSiB model. The authors are also grateful to the three anonymous reviewers and to the guest editor for their constructive comments on an earlier version of this manuscript.

REFERENCES

REFERENCES
Anderson
,
E. A.
,
1976
:
A point energy and mass balance model of a snow cover. NOAA Tech. Rep. NWS 19, Office of Hydrology, National Weather Service, 150 pp
.
Boone
,
A.
, and
P.
Etchevers
,
2001
:
An intercomparison of three snow schemes of varying complexity coupled to the same land surface model: Local scale evaluation at an alpine site.
J. Hydrometeor.
,
2
,
374
394
.
Cherkauer
,
K. A.
, and
D. P.
Lettenmaier
,
1999
:
Hydrologic effects of frozen soils in the upper Mississippi River basin.
J. Geophys. Res.
,
104
,
19599
19610
.
Cline
,
D. W.
,
1997
:
Snow surface energy exchanges and snowmelt at a continental, midlatitude alpine site.
Water Resour. Res.
,
33
,
689
701
.
Cosgrove
,
B. A.
, and
Coauthors
,
2003
:
Real-time and retrospective forcing in the North American Land Data Assimilation System (NLDAS) project.
J. Geophys. Res.
,
108
.
D22
.
8842, doi:10.1029/2002JD003118
.
Dai
,
Y.
, and
Q.
Zeng
,
1997
:
A land surface model (IAP94) for climate studies. Part I: Formulation and validation in off-line experiments.
Adv. Atmos. Sci.
,
14
,
443
460
.
Dai
,
Y.
, and
Coauthors
,
2003
:
The Common Land Model.
Bull. Amer. Meteor. Soc.
,
84
,
1013
1023
.
Deardorff
,
J. W.
,
1978
:
Efficient prediction of ground surface temperature and moisture, with inclusion of a layer of vegetation.
J. Geophys. Res.
,
83
,
1889
1903
.
Dickinson
,
R. E.
, and
Coauthors
,
1986
:
Biosphere–Atmosphere Transfer Scheme (BATS) for the NCAR Community Climate Model. NCAR Tech. Note NCAR/TN-275+STR, 69 pp
.
Dickinson
,
R. E.
,
A.
Henderson-Sellers
, and
P. J.
Kennedy
,
1993
:
Biosphere–Atmosphere Transfer Scheme (BATS) version 1e as coupled to the NCAR Community Climate Model. NCAR Tech. Note NCAR/TN-387+STR, 72 pp
.
Ek
,
M. B.
, and
Coauthors
,
2003
:
Implementation of Noah land surface model advances in the National Centers for Environmental Prediction operational mesoscale Eta model.
J. Geophys. Res.
,
108
.
D22
.
8851, doi:10.1029/2002JD003296
.
Elder
,
K.
, and
A.
Goodbody
,
2004
:
CLPX-Ground: ISA main meteorological data. National Snow and Ice Data Center, Boulder, CO, digital media. [Available online at http://nsidc.org/data/docs/daac/nsidc0172_clpx_mainmet/.]
.
Etchevers
,
P.
, and
Coauthors
,
2004
:
Validation of the energy budgets of an alpine snowpack simulated by several snow models (SnowMIP project).
Ann. Glaciol.
,
38
,
150
158
.
Foster
,
J.
, and
Coauthors
,
1996
:
Snow cover and snow mass intercomparisons of general circulation models and remotely sensed datasets.
J. Climate
,
9
,
409
426
.
Gelfan
,
A. N.
,
J. W.
Pomeroy
, and
L. S.
Kuchment
,
2004
:
Modeling forest cover influences on snow accumulation, sublimation, and melt.
J. Hydrometeor
,
5
,
785
803
.
Hardy
,
J. P.
,
R. E.
Davis
,
R.
Jordan
,
X.
Li
,
C. E.
Woodcock
,
W.
Ni
, and
J. C.
McKenzie
,
1997
:
Snow ablation modeling at the stand scale in a boreal jack pine forest.
J. Geophys. Res.
,
102
,
29397
29405
.
Jin
,
J.
,
X.
Gao
,
Z-L.
Yang
,
R. C.
Bales
,
S.
Sorooshian
, and
R. E.
Dickinson
,
1999
:
Comparative analyses of physically based snowmelt models for climate simulations.
J. Climate
,
12
,
2643
2657
.
Jordan
,
R. E.
,
1991
:
A one-dimensional temperature model for a snow cover: Technical documentation for SNTHERM.89. CRREL Special Rep. 91-16, Cold Regions Research and Engineering Laboratory, 61 pp
.
Jordan
,
R. E.
,
E. L.
Andreas
, and
A. P.
Makshtas
,
1999
:
Heat budget of snow-covered sea ice at North Pole 4.
J. Geophys. Res.
,
104
,
C4
.
7785
7806
.
Koren
,
V.
,
J.
Schaake
,
K.
Mitchell
,
Q.
Duan
,
F.
Chen
, and
J.
Baker
,
1999
:
A parameterization of snowpack and frozen ground intended for NCEP weather and climate models.
J. Geophys. Res.
,
104
,
19569
19585
.
Liang
,
X.
,
D. P.
Lettenmaier
,
E. F.
Wood
, and
S. J.
Burges
,
1994
:
A simple hydrologically based model of land surface water and energy fluxes for GSMs.
J. Geophys. Res.
,
99
,
14415
14428
.
Link
,
T.
, and
D.
Marks
,
1999
:
Point simulation of seasonal snow cover dynamics beneath boreal forest canopies.
J. Geophys. Res.
,
104
,
27841
27857
.
Luo
,
L.
, and
Coauthors
,
2003
:
Effects of frozen soil on soil temperature, spring infiltration, and runoff: Results from the PILPS 2(d) experiment at Valdai, Russia.
J. Hydrometeor.
,
4
,
334
351
.
Mahrt
,
L.
, and
H. L.
Pan
,
1984
:
A two-layer model of soil hydrology.
Bound.-Layer Meteor.
,
29
,
1
20
.
Marks
,
D.
,
1988
:
Climate, energy exchange, and snowmelt in Emerald Lake Watershed, Sierra Nevada. Ph.D. dissertation, University of California, Santa Barbara, 158 pp
.
Marshall
,
S.
,
J. O.
Roads
, and
G.
Glatzmaier
,
1994
:
Snow hydrology in a general circulation model.
J. Climate
,
7
,
1251
1269
.
Mote
,
T. L.
,
A. J.
Grundstein
,
D. J.
Leathers
, and
D. A.
Robinson
,
2003
:
A comparison of modeled, remotely sensed, and measured snow water equivalent in the northern Great Plains.
Water Resour. Res.
,
39
.
1209, doi:10.1029/2002WR001782
.
Oleson
,
K. W.
, and
Coauthors
,
2004
:
Technical description of the Community Land Model (CLM). NCAR Tech. Note NCAR/TN-461+STR, 174 pp
.
Robinson
,
D. A.
, and
G.
Kukla
,
1985
:
Maximum surface albedo of seasonally snow-covered lands in the Northern Hemisphere.
J. Climate Appl. Meteor.
,
24
,
402
411
.
Schlosser
,
C. A.
, and
Coauthors
,
2000
:
Simulations of a boreal grassland hydrology at Valdai, Russia: PILPS phase 2(d).
Mon. Wea. Rev.
,
128
,
301
321
.
Sellers
,
P. J.
,
Y.
Mintz
,
Y. C.
Sud
, and
A.
Delcher
,
1986
:
A simple biosphere model (SiB) for use within general circulation models.
J. Atmos. Sci.
,
43
,
505
531
.
Sicart
,
J. E.
,
J. W.
Pomeroy
,
R. L. H.
Essery
,
J.
Hardy
,
T.
Link
, and
D.
Marks
,
2004
:
A sensitivity study of daytime net radiation during snowmelt to forest canopy and atmospheric conditions.
J. Hydrometeor.
,
5
,
774
784
.
Slater
,
A. G.
, and
Coauthors
,
2001
:
The representation of snow in land surface schemes results from PILPS 2(d).
J. Hydrometeor
,
2
,
7
23
.
Wigmosta
,
M. S.
,
L. W.
Vail
, and
D. P.
Lettenmaier
,
1994
:
A distributed hydrology-vegetation model for complex terrain.
Water Resour. Res.
,
30
,
1665
1679
.
Xue
,
Y.
,
P. J.
Sellers
,
J. L.
Kinter
III
, and
J.
Shukla
,
1991
:
A simplified biosphere model for global climate studies.
J. Climate
,
4
,
345
364
.
Yeh
,
T-C.
,
R. T.
Wetherald
, and
S.
Manabe
,
1983
:
A model study of the short-term climatic and hydrologic effects of sudden snowcover removal.
Mon. Wea. Rev.
,
111
,
1013
1024
.

Footnotes

Corresponding author address: Xia Feng, COLA, IGES, 4041 Powder Mill Rd., Ste. 302, Calverton, MD 20705-3106. Email: xfeng@gmu.edu