Assessment of the Timing of Daily Peak Streamflow during the MeltSeason in a Snow-Dominated Watershed

XING CHEN, MUKESH KUMAR, AND RUI WANG

Nicholas School of the Environment, Duke University, Durham, North Carolina

ADAM WINSTRALaAND DANNY MARKS

Northwest Watershed Research Center, Agricultural Research Service, USDA, Boise, Idaho

(Manuscript received 19 August 2015, in final form 23 February 2016)

ABSTRACT

Previous studies have shown that gauge-observed daily streamflow peak times (DPTs) during spring

snowmelt can exhibit distinct temporal shifts through the season. These shifts have been attributed to three

processes: 1) melt flux translation through the snowpack or percolation, 2) surface and subsurface flow ofmelt

from the base of snowpacks to streams, and 3) translation of water flux in the streams to stream gauging

stations. The goal of this study is to evaluate and quantify how these processes affect observed DPTs vari-

ations at the Reynolds Mountain East (RME) research catchment in southwest Idaho, United States. To

accomplish this goal, DPTs were simulated for the RME catchment over a period of 25 water years using a

modified snowmelt model, iSnobal, and a hydrology model, the Penn State Integrated Hydrologic Model

(PIHM). The influence of each controlling process was then evaluated by simulating the DPT with and

without the process under consideration. Both intra- and interseasonal variability in DPTs were evaluated.

Results indicate that the magnitude of DPTs is dominantly influenced by subsurface flow, whereas the

temporal shifts within a season are primarily controlled by percolation through snow. In addition to the three

processes previously identified in the literature, processes governing the snowpack ripening time are iden-

tified as additionally influencing DPT variability. Results also indicate that the relative dominance of each

control varies through the melt season and between wet and dry years. The results could be used for sup-

porting DPTs prediction efforts and for prioritization of observables for DPT determination.

1. Introduction

Snowmelt is the major source of groundwater re-

charge and streamflow inmountainous areas throughout

the western United States. More than 50%–70% of the

western U.S. water supply originates in snow-fed upland

watersheds (Barnett et al. 2005; Carroll et al. 2006;

DeWalle and Rango 2008). Accurate quantification of

the variability of streamflow from these watersheds can

be crucial for both stream ecology and sustainable

management of water supply resources.

Streamflow variability in the western United States

exhibits spatial and temporal dependencies. This study

examines the temporal dependencies, in particular,

those that occur at subdaily time scales (Lowry et al.

2010; Lundquist and Cayan 2002; Lundquist and

Dettinger 2005; Tobin et al. 2013). Subdaily streamflow

variations are important for fluvial processes and

aquatic ecology as well as reservoir operations aimed at

reducing the magnitude of high flows (Graf 1999; Poff

et al. 1997) and modulating downstream aquatic com-

munities (Bain et al. 1988) and water quality variables

such as stream temperature (Neumann et al. 2003) and

dissolved oxygen. The diurnal variations in streamflow

are mainly introduced by ice or snowmelt and evapo-

transpiration and have been the focus of many studies

(Caine 1992; Gribovszki et al. 2008, 2010; Johnson et al.

2013; Loheide 2008; Loheide and Lundquist 2009;

Lundquist and Dettinger 2005; Soylu et al. 2012). We

focus on assessing the intra- and interseasonal variations

a Current affiliation: WSL Institute for Snow and Avalanche

Research SLF, Davos Dorf, Switzerland.

Corresponding author address: Mukesh Kumar, Nicholas School

of the Environment, Duke University, 450 Research Dr., LSRC

A207A, Durham, NC 27708.

E-mail: mukesh.kumar@duke.edu

AUGUST 2016 CHEN ET AL . 2225

DOI: 10.1175/JHM-D-15-0152.1

� 2016 American Meteorological Society

in daily streamflow peak times (DPTs) during the melt

season and identifying the process dependencies of

these variations. DPT is defined as the hour when

streamflow is the maximum within a day, and melt sea-

son is considered to span from the first to last day that

shows a melt-affected diurnal signal (characterized by a

rapid increase followed by a gradual decrease in daily

hydrograph; Lundquist and Cayan 2002).

DPTs during the snowmelt season can demonstrate

varied shifts. As the melt season progresses, DPTs may

occur earlier in the diurnal cycle (Caine 1992; Jordan

1983a,b), later (Grover and Harrington 1943; Lundquist

and Cayan 2002), or have no significant temporal shift

(Lundquist et al. 2005). Earlier/later DPTs indicate

shorter/longer time intervals between peak melt pulse

and peak subdaily streamflow. These disparate shifts

have been attributed to a combination of three primary

controls: 1) the percolation of liquid water through the

snowpack (Ambach et al. 1981; Caine 1992; Colbeck and

Davidson 1973; Dunne et al. 1976; Jordan 1983a; Pfeffer

et al. 1990), 2) the translation ofmeltwater from the base

of snowpack to the river channel (Caine 1992; Dunne

and Black 1971; Flerchinger et al. 1992; Kobayashi 1986;

Maulé and Stein 1990), and 3) the translation of water

flux in the river channel to the stream gauging station

(Lundquist and Dettinger 2005). These three controls

affect the time difference between the peakmelt pulse at

the snowpack surface due to energy inputs and peak

subdaily streamflow at the stream gauging station; for

example, peak melt pulse typically occurs in the early

afternoon, while peak subdaily streamflow generally

occurs late at night or even the following morning

(Lundquist and Dettinger 2005).

The first control affects the DPTs during the melt

season by altering percolation time of liquid water

through the snowpack as the melt season progresses.

During spring melt when snow cover densities exhibit

low variability, liquid water translation times are gen-

erally longer through thicker snowpacks than thinner

ones. At the beginning of the melt season when snow-

pack is generally thicker, the translation time is longer.

Refreezing of meltwater may further contribute to slow

effective percolation velocity early in the melt season

(Colbeck and Davidson 1973; Colbeck 1975; Pfeffer

et al. 1990). In contrast, the percolation time is reduced

during the late melting season as the snowpack thins

(Caine 1992; Jordan 1983a,b; Kobayashi and Motoyama

1985). Percolation rates may also be affected by the

evolution of preferential flow paths in the snowpacks

during the melt season (Marsh and Woo 1985; Marsh

and Pomeroy 1996).

The second control on shifts in the timing of DPTs is

associated with the time difference between surface

water input from the base of the snow cover and themelt

response signal in the stream. In larger watersheds that

span over a wide range of elevations, streamflow re-

sponse time will increase during the melt season as the

snow line retreats to higher elevations. This results in a

shift of DPTs to later in the day (Caprio 1966; Lundquist

and Cayan 2002; Lundquist and Dettinger 2005;

Lundquist et al. 2004, 2005). In watersheds with frac-

tured basalt in the subsurface, DPTs have been reported

to vary nonmonotonically depending on the snow dis-

tribution and distance of residual snow cover from the

stream gauging station (Flerchinger et al. 1992). In some

large watersheds with a highly heterogeneous snow

cover distribution, the delaying effect of retreating snow

line may offset the effect of percolation time change

within the snowpack, resulting in an essentially un-

changed DPTs through the melt season (Lundquist

et al. 2005).

The third control on changes in DPTs during the melt

season is caused by differences in daily average flow

velocity in the stream channel. Flow velocity varies with

streamflow volume, with higher velocities around the

streamflow peak and lower velocities late in the melt

period (Lundquist and Dettinger 2005).

As the snow cover amount and properties vary inter-

seasonally, the effects of each aforementioned control

on DPTs is expected to change from one year to the

next. Using 5 years of data in Martinelli snowpatch

(0.08 km2 area), Caine (1992) suggested that the DPT

occurred later in the day during years with larger basin-

scale snow depth. Similar conclusions were also drawn

by Lundquist and Dettinger (2005) based on data from

Marble Fork watershed.

The primary purpose of this paper is to evaluate the

role of the process controls on both intra- and inter-

seasonal variations in DPTs. Lundquist et al. (2005)

assessed the impacts of percolation time through the

snowpack and the travel time of meltwater in the river

channel on DPT shifts while assuming that travel times

between the base of the snowpack and the stream were

negligible [see paragraph 19 of Lundquist et al. (2005)].

However, this assumption cannot be universally applied,

especially in watersheds where streamflow has a large

groundwater contribution. To explore the impact of

percolation of liquid water through the snow cover and

translation of water from the base of the snow cover to

the stream and to evaluate the role of these and any

additional processes that may determine DPT variation,

we couple a snowmelt model, ISNOBAL, with a hy-

drology model, Penn State Integrated Hydrologic

Model (PIHM), to simulate the DPTs. The intra- and

interseasonal variations in both observed and modeled

DPTs are identified, and the capability of the coupled

2226 JOURNAL OF HYDROMETEOROLOGY VOLUME 17

ISNOBAL–PIHM to estimate DPTs and its variations

during the melt season is evaluated. Through a series of

process-unmixing experiments, the coupled modeling

system is then used to isolate the role of individual

processes in determining DPTs. Aside from the three

previously reported controls, the physically based cou-

pled modeling framework allows us to also assess the

roles of additional processes on observed DPT varia-

tions. These experiments were used to answer the

following two questions: 1) How do DPTs vary intra-

seasonally and what processes are critical to de-

termine the magnitude of those variations? and

2) How doDPTs vary interseasonally and what are the

processes that control that variation?

This paper is organized as follows. Section 2 pres-

ents the study area, the modeling methodology, and

related calibration and validation details of the linked

model (ISNOBAL–PIHM). Section 3 describes the

design of process-unmixing experiments and the in-

formation obtained from these experiments. Section

4 describes the results from the process-unmixing

experiments and discusses process-dependent con-

trols on intra- and interseasonal variations of DPTs.

Section 5 summarizes the results and takeaways from

this study.

2. Setting and model details

a. Study area and relevant datasets

Reynolds Mountain East (RME), a snow-dominated

headwater catchment within the Reynolds Creek Ex-

perimental Watershed (RCEW; Marks 2001), was

selected for this study (Fig. 1). Elevations in RME

(0.38 km2) range from 2028 to 2137m above mean sea

level. Hillslope angles range from 08 to 21.48. Soil texturein the watershed ranges from loam to clay, with variably

fractured and altered basalt underneath. The watershed

is covered by dense willows and aspen near the riparian

zone and mixture of sparse Douglas fir, Vaseyana

sagebrush, scattered dry meadow, and aspen patches

farther away from the stream (Grant et al. 2004; Reba

et al. 2011a). Based on the 25-water-year (WY)1 dataset

presented in Reba et al. (2011a) (and discussed further

below), the monthly average temperature in the water-

shed ranges from 248C in December to 178C in July.

Over 70% of the precipitation occurs in the form of

snow during the winter months, with July–September

being typically very dry. WY precipitation at the snow

pillow site (filled circle in Fig. 1) varies from 584 to

1537mm with a WY average of 967mm. In contrast,

WY precipitation at the exposed site, which is only

around 350m away and is located in an exposed area

within a sagebrush community (filled triangle in

Fig. 1), varies from 454 to 1201mm with aWY average

of 779mm (Marks andWinstral 2001). The differences

in precipitation are due to complex snow accumula-

tion patterns produced by wind scour and drifting.

Heterogeneity in mass and energy inputs in this

FIG. 1. Land-cover distribution and observation sites in the RME watershed.

1 A water year runs from 1 Oct of the previous year through 30

Sep of the given WY and is used in regions dominated by winter

precipitation, like the western United States. In this paper, any

reference to years is to water years, with the year referring to the

calendar year of that spring’s snowmelt runoff.

AUGUST 2016 CHEN ET AL . 2227

watershed produce spatial–temporal differences in

melt production with exposed areas producing a

greater proportion of early spring melt and sheltered

regions contributing higher percentages in late spring

(Marks et al. 2002).

The aforementioned 25-WY (1984–2008) hydro-

climatic dataset (Reba et al. 2011a) was used to evaluate

DPT variations. Relevant datasets were hourly stream-

flow at the basin outlet (WYs 1984–2008), hourly

groundwater at three wells (WYs 2006–08), and hourly

soil moisture at one site (WYs 2005–08). Streamflow

from the RME watershed ranges from 125mm in the

driest WY to 1106mm in the wettest WY, with an av-

erage of 518mm for the 25-WY period. The runoff ratio

varies from 0.24 to 0.90, with an average runoff ratio of

0.61 for the 25-WY simulation period. The three years

containing fine temporal groundwater and soil moisture

data encompass a wide range of hydroclimatic condi-

tions, with 2006 and 2007 WYs being among the wettest

and driest five WYs, respectively, over the 25-WY

dataset (Reba et al. 2011a). WY 2008 was an average

year with precipitation of 856mm at the exposed site and

996mm at the sheltered site. All relevant topographic,

physiographic, and hydroclimatic datasets for WYs 1984

to 2008 are obtainable from the Northwest Watershed

Research Center, USDA (ftp://ftp.nwrc.ars.usda.gov/

public/RME_25yr_database). Detailed descriptions of

these datasets can be found in Hanson et al. (2001).

b. Modeling methodology

For this analysis, ISNOBAL was coupled to PIHM,

simulating snow accumulation, melt, and hydrologic

processes across the RME watershed for the 25-WY

simulation period. ISNOBAL is a grid-based two-layer

energy- and mass-balance snow model (Marks et al.

1999), forced with distributed fields of precipitation, air

temperature, net shortwave radiation, downwelling

thermal radiation, wind speed, and relative humidity.

ISNOBAL simulates snow states including snow tem-

perature, density, and snow water equivalent for each

layer at every time step. Snowmelt is initiated with ad-

ditional energy input once snow temperature in either of

the two layers reaches 08C. Melt water is retained in the

snowpack by capillary forces (surface tension) until

threshold saturation levels are reached. At this point,

surface tension can no longer retain any additional liq-

uid water against gravity and any additional energy in-

put results in meltwater percolating downward.

ISNOBAL delivers excess meltwater, or rain on bare

ground to the soil surface as surface water input (SWI)

for each hourly time step, assuming zero lag time.

ISNOBAL has been previously applied across sites with

varying snow and climate conditions in the United

States (Garen and Marks 2005; Link and Marks 1999;

Reba et al. 2011b; Winstral et al. 2009). In this work, a

lag function was incorporated in ISNOBAL to delay the

SWI fluxes using (Anderson 1976; Flerchinger 2000)

L5L

max

C2W1 1

, (1)

where L is the SWI delay time (h); C2 is an empirical

coefficient with the value 0.01m21;W is the depth of the

SWI flux (m); and Lmax is the maximum lag time (h),

calculated as

Lmax

5C1[12 exp(20:0253 d

s/r

s)], (2)

where C1 is the maximum allowable lag obtained from

best fit of experimental data (Anderson 1976),C15 10 h;

ds is snow cover depth (m); and rs is snow cover density

(kgm23).

Even though this ‘‘lag and route’’ approach calculates

the percolation time of excess liquid water without ac-

counting for the preferential flow paths (Marsh andWoo

1985; Marsh and Pomeroy 1996) and an explicit dis-

tinction of the snowpack into unsaturated and saturated

layers (Gray et al. 1985), it has been shown to be ef-

fective in simulating hourly outflow from the bottom of

snowpack (Barry et al. 1990). Equation (1) suggests that

for a ripe snowpack with a density of 480kgm23, depth

of 0.90m, and melt rate of 1.2mmh21, the liquid water

flux lag time at a point would be around 3.5 h. This

propagation time is similar to the one obtained using the

equations presented in Lundquist and Dettinger (2005),

which calculates a translation time of 3.7 h. Dunne et al.

(1976) reported the observed shift in lag time in the

range of 0.5–7.9 h from plots sampled at a range of as-

pects, vegetation cover, snow depth, and density. After

accounting for the delay in SWI flux translation through

snowpacks, the resulting adjusted SWI is input to PIHM

(Kumar 2009; Kumar et al. 2009; Qu and Duffy 2007).

PIHM is a physics-based distributed hydrologic

model, which employs a semidiscrete finite volume

formulation to simulate a range of processes, includ-

ing snowmelt, evapotranspiration [Penman–Monteith

equation (Allen et al. 1998)], interception [Rutter

model (Rutter et al. 1971)], overland flow [2D diffu-

sion wave equation (Gottardi and Venutelli 1993)],

unsaturated zone infiltration [1D approximation of

Richards’s equation (Richards 1931)], groundwater

flow (3D, Richards’s equation), and streamflow [1D

diffusive wave equation (Strelkoff 1970)]. It is to be

noted that one-way linking between ISNOBAL

and PIHM involved deactivating the temperature

index snowmelt utility in PIHM and replacing it with

2228 JOURNAL OF HYDROMETEOROLOGY VOLUME 17

the physically based snowmelt SWI output from

ISNOBAL. Additionally, simulated ground evapora-

tion in PIHM is shut off at locations shielded by accu-

mulated snow, but evapotranspiration from protruding

vegetation still occurs. Energy exchanges at snow-free

locations remain the same. Both ISNOBAL and

PIHM are set to run at an hourly time step; however,

the spatial resolution of ISNOBAL and PIHM are

different. ISNOBAL was run on a total of 3978 10m310m structured grids in the model domain, while

PIHM has 101 unstructured triangular grids in the

same domain. SWI for each triangle grid contains the

sum of SWI output from ISNOBAL structured grids,

which has the centroid within the triangle. PIHM has

been previously applied across watersheds with dif-

ferent scales and climate conditions (Chen et al 2015;

Kumar and Duffy 2015; Shi et al. 2014; Yu et al. 2014).

The linked framework has been previously used in

Kumar et al. (2013) and Wang et al. (2013) to model

relevant hydrologic states and fluxes in RME.

c. Model calibration and validation

1) MODEL CALIBRATION

Model calibration is performed only in PIHM.

Though ISNOBAL is uncalibrated, the terrain-based

wind redistribution parameters tuned to topographic

and vegetation structure of the RME catchment as

presented by Winstral et al. (2009) were used for the

ISNOBAL simulation. The ability of the model to sim-

ulate snow distribution and total melt and surface water

output has already been demonstrated in Winstral and

Marks (2002). Reba et al. (2011b) showed that for an

uncalibrated ISNOBAL simulation, the average Nash–

Sutcliffe model efficiency (NSE) over the 25-WY period

for predicted versus measured SWE at the snow pillow

was 0.90, with a range of values between 0.74 and 0.98.

Calibration of parameters for PIHM simulations in-

volved nudging of hydrogeological parameters uni-

formly across the model domain (Refsgaard 1997) to

match the baseflow magnitude, groundwater head dis-

tribution, and decay rate of the hydrograph during the

recession period. Calibration experiments were per-

formed for two periods: 1) a dry period with no appre-

ciable antecedent recharge (10–25 October 2005) and

2) a wet and cold period with peak streamflow response

(caused by rainfall event) and negligible evapotranspi-

ration (from 30 December 2005 to 30 January 2006).

Streamflow during the first calibration period was as-

sumed to be predominantly base flow and controlled by

subsurface properties. It was assumed that streamflow

during the second calibration period was controlled by

both surface and subsurface properties. This strategy

ensured that calibration of the coupled snow and hy-

drology model did not depend on the ISNOBAL SWI

output. Hydrogeological parameters that were adjusted

during the calibration procedure included soil drainage

parameters, such as van Genuchten coefficients (Van

Genuchten 1980), porosity, and conductivity of the top

soil and the underlying subsurface. More details about

the calibration methodology are presented in Kumar

et al. (2013). It is to be noted that parameter optimiza-

tion did not include matching the magnitude or timing

of streamflow peak response at either daily or seasonal

scales.

2) MODEL VALIDATION

(i) Streamflow, groundwater, and soil moisturevalidation

Results of streamflow validation for the simulation

period 1984–2008 (Fig. 2a) showed NSE of 0.86 and

coefficient of determination r 2 of 0.93 for hourly data,

0.88 and 0.94 for daily data, 0.86 and 0.96 for monthly

data, and 0.86 and 0.96 for yearly data. The simulated

and observed annual runoff means during WYs 1984–

2008 were 528 and 468mm, respectively.

Groundwater table dynamics validation for the period

WY 2006–08 is presented in Fig. 2b. Though the ob-

served groundwater responses at wells 1 and 2 were

unusually steep, for example, varying by as much as

7.5m in a matter of 3 h, the model was able to capture

the general shape and structure of these dynamics across

all years where groundwater height data were available

[see Kumar et al. (2013) for details]. Figure 2b shows

that the model captures the slower dynamics equally

well. More detailed characterization of the subsurface,

finer spatial data and model grid resolution, and an au-

tomated calibration strategy may possibly help resolve

the discrepancies. It is possible that the simple repre-

sentation of macroporous flow behavior in the sub-

surface is not adequate for highly transient subsurface

flow systems.

Comparisons between observed soil moisture at

multiple locations has already been shown in Fig. 2 of

Kumar et al. (2013), where the simulated top soil satu-

ration magnitude and timing reasonably matches the

observation. Even though hourly soil moisture data exist

for the snow pillow site fromWYs 2005 to 2008, out of 67

melt days during this period only 6 exhibited a peak

signal. On these 6 days, average error between peak

hour in the observed soil moisture data and the modeled

lagged SWI flux was within an hour. On the rest of the

61 days, soil is either completely saturated during the

day or it saturates well before the melt peak is reached.

Despite using a nonlocalized calibration approach,

AUGUST 2016 CHEN ET AL . 2229

model predictions for streamflow, groundwater table,

and soil moisture variations can be considered adequate

and underscore the potential of the coupled ISNOBAL–

PIHM system for simulating hydrologic responses and

understanding the role of process controls on DPTs.

(ii) DPT validation

In this study, melt season DPTs were calculated from

simulated streamflow and compared to observed DPTs

for WYs 1984–2008. Days without a melt-affected (di-

urnal) signal were excluded from the analysis (i.e., the

hydrograph showed monotonic variations within a day).

Figure 3 shows the typical streamflow diurnal signals

during the melt season and corresponding DPTs. Per-

formance of the linked model was evaluated based on

the following two metrics: 1) presence/absence of an

observed diurnal signal in simulation results and 2) DPT

simulation.

A. PRESENCE /ABSENCE OF AN OBSERVED DIURNAL SIGNAL

IN SIMULATION RESULTS

For the 25-WY simulation period, a total of 811 and

595 days showed melt-affected signal in observed and

simulated streamflow, respectively, with 519 days (64%

of observed melt-affected days) showing melt-affected

signal in both observed and simulated streamflow. For

the wettest 5 years, which is defined as the 5 years with

the largest precipitation amount before themelt-out date

at the snowpillow site, a total of 195 and 169 days showed

melt-affected signal in observed and simulated stream-

flow, respectively, with 150 days (77% of observed melt-

affected days) showing melt-affected signal in both

observed and simulated streamflow data. In contrast, for

the driest 5 years, which is defined as the 5 years with the

smallest precipitation amount before the melt-out date

at the snow pillow site, a total of 124 and 57 days showed

FIG. 2. Modeled and observed (a) streamflow for WYs 1984–2008 and (b) groundwater

depth for WYs 2006–08 [light gray vertical lines in (b) demarcate period from 1 to 13 Apr in

WY 2006].

2230 JOURNAL OF HYDROMETEOROLOGY VOLUME 17

melt-affected signal in observed and simulated stream-

flow, respectively, with 56 days (45% of observed melt-

affected days) showing melt-affected signal in both

observed and simulated streamflow. The results indicate

that the model could capture the days with diurnal signal

much more accurately in wet years than in dry years.

The observed disparity in accuracy between wet and

dry years is largely due to the effectiveness of the model

during the early melt period. Isolating the first 5 days of

the season withmelt-affected diurnal signals in observed

streamflow, modeled streamflow similarly exhibits a di-

urnal signal on 88%of the days in the wettest 5 years and

FIG. 4. Comparison of simulated and observed streamflow during the melt season for (a) a very wet, warm snow

season (WY 1996) and (b) a dry, cool snow season (WY 1999). See Reba et al. (2011b) for details on conditions

during WYs 1996 and 1999.

FIG. 3. Streamflow time series for WY 2006 during melt season. Also illustrated is the DPT

hour (after noon) for every day. Times in black indicate peak hour for days with distinct diurnal

signal. Times in gray boxes indicate peak hour for days with multiple peaks. All times are with

respect to the local noon. Note that negative value means time before noon, for example, 22

means 1000 local time (LT).

AUGUST 2016 CHEN ET AL . 2231

only 52% of the days in the driest 5 years. The inability

of the model to capture melt-affected streamflow in

early spring, especially during dry years, is highlighted in

Fig. 4. The streamflow simulation is fairly accurate

throughout the melt season for a very wet, warm snow

season (WY 1996, Fig. 4a). In contrast, Fig. 4b shows that,

for a dry, cool snow season (WY 1999), the model under-

estimates streamflow early in the melt season and over-

estimates streamflow later in the melt season. One possible

reason could be that PIHM did not simulate soil temper-

ature, and hence the consequent reduction in infiltration

capacity under frozen/near-frozen soil conditions was not

accounted for (Burt and Williams 1976; Flerchinger et al.

2006; Horiguchi and Miller 1983; Watanabe and Flury

2008). As a result, instead of surface runoff during the cool

WY 1999 snow season, the model overestimated recharge

to the subsurface, thus missing the melt-affected stream-

flow in early spring. These conclusions are consistent with

reported cold snow and soil temperatures during the WY

1999 snow season (see Reba et al. 2011b).

B. DPT MAGNITUDE

The DPTs were compared for the days when both ob-

served and simulated data showed a melt-affected signal.

Since both observed and modeled daily hydrographs may

often consist of almost flat and slightly noisy high-flow

periods lasting few hours (,4h) with very small hour-to-

hour variations, comparison of modeled and observed ab-

solute DPTs can sometimes be muddled by noise (Fig. 5).

Hence, comparisons were also performed against ‘‘mod-

eled Q1% peak hour,’’ which is defined as the closest hour

to the observed DPTs in which the modeled streamflow is

within the top 1% of the modeled daily streamflow range.

An identical observed DPT and modeled Q1% peak hour

indicates that the top 1% of modeled daily streamflow

occurred within the same hour as the observed daily peak.

In contrast, a larger difference between modeled DPT and

modeled Q1% DPT indicates a longer period of relatively

flat (and possibly noisy) hydrograph near the peak.

DPTs in all the figures presented here are with respect

to the local noon, that is, 9 indicates 9 h after noon,

which is 2100 local time (LT). DPT simulation results for

the 24 WYs are shown in Fig. 6. WY 1992 is not shown

here since themodel could not capture the diurnal signal

exhibited on any of the 5 days with an observed diurnal

signal during this extremely dry year. The r 2 value be-

tween modeled and observed DPTs for the 519 days

with diurnal signal in both simulated and observed

FIG. 5. Schematic of observed and modeled daily streamflow hydrographs with (a) identical observed and

modeled daily peak hour (here observed peak hour5modeled absolute peak hour5modeledQ1% peak hour) and

(b) different observed and modeled peak hour with very small variations near the top 1% peak discharge (here

observed peak hour 5 modeled absolute peak hour 2 1 5 modeled Q1% peak hour).

2232 JOURNAL OF HYDROMETEOROLOGY VOLUME 17

FIG. 6. Observed and modeled daily peak time for 24 WYs ranging from WYs 1984 to 2008 (WY 1992 was excluded). The x axis

represents WY day, and the y axis indicates the daily peak time hour with respect to local noon. Note that negative value means time

before noon, for example, 24 means 0800 LT.

AUGUST 2016 CHEN ET AL . 2233

streamflows was 0.58, while r 2 between Q1% DPTs and

observed DPTs was 0.74. However, 130 of these days

had multimodal peaks because of rain or snow events

that interfered with the diurnal variation in energy

forcing to the snowpack. For days with unimodal peak,

r2 between modeled and observed DPTs was 0.61 and

between Q1% DPTs and observed DPTs was 0.80.

Figure 6 shows that both modeled and Q1% DPTs ap-

pear to follow the seasonal variations exhibited by the

observed data. The mean and variance of observed,

modeled, andQ1%DPTswere (5.8, 7.4), (5.6, 9.3), and (6.0,

7.7) h, respectively. Further analyses indicated that, of the

number of days considered in Fig. 6, years with more than

80% of days with j(observed DPT 2 modeled absolute

DPT)j and j(observed DPT2 modeledQ1% DPT)j beingsmaller than or equal to 1h are 6 and 14 years, respectively.

For the 24 years under consideration, more than 59% of

days had a difference between observed and modeled

DPTs of less than 1h, and more than 78% of days had this

difference in the range less than 2h. The corresponding

numbers for the difference between Q1% modeled and

observed DPTs were 76% and 88%, respectively.

Presented results indicate that the simulated hour of

peak discharge, especially the hour of the Q1% peak,

closely matched the hour of observed daily streamflow

peak during the snowmelt season, especially in wetter

years. However, ISNOBAL–PIHM was not as success-

ful in capturing diurnal variations in streamflow, espe-

cially early in the melt season during drier years.

3. Experiment design and details

a. Daily peak time and melt season duration

The initial day of the melt period, identified by the

melt-induced signal in streamflow, does not always

mean that snowmelt begins on this given day. For ex-

ample, analysis of the simulated daily SWI volume forWY

2006 indicates 12 days of melt before a melt-affected

streamflow signal was registered. Furthermore, it also

does not mean that the SWI volume on this day is neces-

sarily larger than that on the previous days. For example,

on the previous and following 5 days of initiation of the

melt period (13 April, day 195 of WY 2006), simulated

melt amount only differed marginally (Fig. 7). Simulation

results suggest that the melt-affected diurnal signal in this

year was expressed only after groundwater table was

shallow enough for it to substantially contribute to

streamflow. Before this, most of the melt recharges the

subsurface moisture deficit and groundwater. This expla-

nation is supported by the observed groundwater table and

streamflow data for WY 2006 (see Figs. 2, 3) where the

melt-affected signal in streamflow starts on 13 April (WY

day 195), only after the observed groundwater levels in

wells 1 and 2 have become shallow enough after un-

dergoing gradual increase from 1 to 13 April (this time

interval is highlighted by two circular dots in well 1 and 2

groundwater series in Fig. 2) in response to melt recharge.

Groundwater control on diurnal streamflow response is

further evident from diurnal streamflow fluctuations that

aremuch smaller (both in terms of volume and amplitude)

than melt (see Fig. 7). It is clear in Fig. 7 that before WY

day 195, the magnitude of SWI is comparable to that after

day 200; however, SWI at the beginning of the melt season

(before WY day 195) does not lead to substantial

streamflow as it does later in the melt season (e.g., after

WY day 200). Also, the observed streamflow magnitude

on the first melt-affected day is small and increases in

subsequent days (concomitantly with the increase in

groundwater height) even though a similar increasing

trend inmelt and SWI does not exist. This indicates a large

FIG. 7. Modeled daily total melt and streamflow during melt period in WY 2006.

2234 JOURNAL OF HYDROMETEOROLOGY VOLUME 17

groundwater contribution to streamflow response

(Bengtsson 1982). In fact, simulation results suggest that

the groundwater contribution to streamflow is close to

100% during the melt season in the RME watershed. This

indicates that DPTs analyzed in this watershed are domi-

nantly controlled by subsurface flow processes, including

infiltration of SWI into the upper unsaturated zone and its

translation to groundwater.

b. Isolating the relative role of individual processcontrols on DPTs through process-unmixingexperiments

The variation of DPTs within a melt season has been

previously attributed to three processes (see section 1 for

detailed discussion).We hypothesize that the time for snow

cover ripening may also contribute to intraseasonal varia-

tions in DPTs. Therefore, in the ensuing analyses, we

consider the role of the following four processes on

DPT variations: 1) net incoming energy (turbulent

heat 1 radiation 1 advection) to the snowpack and

snowpack properties (depth1 temperature1moisture

content), which determines the snowpack ripening time; 2)

percolation rate of liquid water flux through the ripened

snowpack; 3) translation of meltwater output from the

bottom of the snowpack (SWI) to the stream channel

through surface and/or subsurface pathways; and 4) trans-

lation ofwater in the river channel to the streamflowgauge.

Notably, many other factors, such as snowpack and its

distribution and watershed properties, may also affect the

peak time, but they do so by indirectly influencing the

aforementioned four processes. The scope of this paper is

restricted to assessing the influence of the four basic pro-

cesses on DPTs.

To isolate the influence of each process on the timing

of daily peak streamflow, DPTs are evaluated with and

without the process under consideration. DPTs from dif-

ferent processes and relations between them are listed in

Table 1. It is to be noted that the peak time delay (P122P1)

caused by translationwithin a snowpack is not equivalent to

the average translation time through the snowpack. Instead,

(P122 P1) represents the time difference between instances

that receive the largest water flux in the two simulations and

hence is a function of bothmelt flux amount and the time of

translation. For example, between two cases with identical

spatial distribution of percolation times through the snow-

pack in awatershed, ifmelt contributions from thinner snow

(with relatively shorter percolation times) are larger than

that from deeper snow, then (P12 2 P1) will be smaller.

Conversely, if melt contributions are mostly from deeper

snowpacks through which the translation time is relatively

longer, (P12 2 P1) can be expected to be larger. Similarly,

(P123 2 P12) expresses the contribution from surface/

subsurface flow processes in delaying the peak time.

4. Results and discussion

The results are presented thematically following the

two questions outlined in section 1.

a. How do DPTs vary intraseasonally and what arethe key processes that determine its magnitude andvariations?

1) INTRASEASONAL DPT VARIATIONS

Within each melt season, DPTs generally exhibited a

shift to earlier in the diurnal cycle as the melt season

TABLE 1. Description of processes considered and process-unmixing approach.

Processes considered Streamflow DPT Process-unmixing approach

1 Net incoming energy to the snowpack

and snowpack characteristics that

determines the time it takes for

snowpack to ripen

P1 DPT of total melt flux over the watershed

from ISNOBALwithout accounting for

translation process through snow, i.e.,

by setting L 5 0 in Eq. (1)

P1 DPT from snowpack ripening

process

2 Percolation rate of liquid water flux

through the ripened snowpack

P12 DPT of liquid water flux (by accounting

for translation process in the snowpack)

over the watershed, i.e., by calculating

L in Eq. (1) based on snow density, melt

flux amount, and snow depth at each

grid cell

P12 2 P1 DPT delay from liquid water

flux percolation through

snowpack

3 Translation of melt flux output from

the bottom of the snowpack to the

stream channel through surface

and/or subsurface pathways

P123 DPT of streamflow simulated by linked

ISNOBALand PIHM, wherein the flow

translation process in the river channel

was discounted by setting Manning’s

roughness coefficient in the river to be

infinitesimally small

P123 2 P12 DPT delay from liquid water

flux translation through

surface/subsurface

pathways

4 Translation of water in the river

channel to the streamflow gauge

P1234 DPT from the linked ISNOBAL and

PIHM detailed in sections 2b and 2c

P1234 2 P123 DPT delay fromwater traveling

in river channel

AUGUST 2016 CHEN ET AL . 2235

progressed, that is, DPT on the last day of the melt

season was earlier than that on the start day. The change

in DPTs from the start to end of the melt season across

the 25-WY period was from 10 to 1 h in the observed

data, from 12 to 1h in the ISNOBAL–PIHM results, and

from 11 to 1h inQ1%modeled results. The shift of DPTs

to earlier in the day was often obscured by abrupt non-

monotonic day-to-day variations. For example, in WY

1993 (Fig. 6), the observed DPTs shifted from hour 12 to

3 (2400–1500 LT) between WY days 207 and 215,

abruptly shifting to hour 11 (2300 LT) on WY day 217

(identified by an arrow in Fig. 6) and then gradually

shifted earlier again to hour 4 (1600 LT) byWY day 232.

During the simulation period, abrupt shifts ($2 h) in

DPTs were observed on 40 days in the observation data,

with a maximum shift equal to 8 h. Comparatively,

abrupt shifts occurred on 43 days in the modeled

streamflow and on 36 days in Q1% modeled results,

with a maximum shift time of 5 and 6h, respectively.

2) INFLUENCE OF PHYSICAL CONTROLS ON

INTRASEASONAL PEAK TIME AND ITS

VARIATIONS

Using the strategy detailed in section 3b, the time of

peak streamflow (e.g., P1, P12, and P123) was obtained

after discounting the effect of individual process con-

trols. Results (Fig. 8) suggest that discounting the role of

individual processes affects both the timing of daily

simulated peak streamflow and its variation during the

season. The daily peak time was earlier, as expected, as

increasing number of process controls were discounted,

that is, P1 , P12 , P123 , P1234. In addition, the dif-

ference between P1, P12,P123, andP1234 was much larger

in the beginning of the season than at the end for each

year. However, the relative contribution of each process

control was observed to vary from day to day. In the next

section, we evaluate the role of each process control on

the intraseasonal timing and variation of DPTs.

(i) Role of net incoming energy and evolvingsnowpack properties

Increasing energy inputs and concomitant changes in

snowpack properties are expected to result in decreasing

DPTs. This is because as net energy input increases,

snow cover ripening may occur earlier during the day. In

addition, changing snowpack properties such as in-

creasing snowpack temperature and liquid water con-

tent and decreasing depth would require less energy to

ripen the snowpack, leading to earlier melt. The average

P1 difference from the start to end of the snow season

during the simulation period was 1.1 h. While in some

years P1 at the start of the season was up to 10h later

than that at the end, other years showed no differences.

Since a shift in P1 is a compound effect of both in-

creasing incoming energy and changing snowpack

properties, two additional ISNOBAL experiments were

performed to assess influences of the two factors.

To evaluate the influence of increasing net energy input

on the DPTs, ISNOBAL snow states for each day within

themelt seasonwere fixed to conditions on the first day of

melt.With fixed snow states and varying net energy input,

the average P1 difference from start to end of the season

during the simulation period was around 0.3h. Similarly,

to evaluate the influence of changing snowpack proper-

ties on the DPTs, ISNOBAL incoming energy for each

day within the melt season was fixed to conditions on the

first day of melt. With fixed incoming energy and varying

snowpack properties, the averageP1 difference from start

to end of the season during the simulation period was

around 0.7h. The results from the two additional exper-

iments indicate that evolving snowpack properties play a

more important role than the incoming energy in themelt

timing. In reality, the variation of P1 (shown in Fig. 8)

during the melt season is not monotonic, as it is de-

pendent on net energy fluxes and snowpack properties,

which exhibit strong variability. Sudden decreases in

temperature or radiation, or new snow events, which in-

crease albedo and snow depth, can lower the net energy

inputs and modulate temperature of the snowpack. In

fact, all 95 days showing a delayed streamflow peak time

in P1 had either large decreases in temperature or radi-

ation relative to the prior day. Of these 95 days, 55%

showed a concomitant increase in P1234. In summary, an

increasing trend in radiation and temperature, with con-

comitant increase in snowpack temperature andmoisture

content and decrease in snowpack depth during the melt

season, generally leads to a decreasing trend in P1.

However, the isolated impact of net energy input is only

marginal. Sudden decreases in radiation or temperature

(e.g., due to cloudiness) and precipitation eventsmay lead

to an abrupt P1 increase. Changes in P1234 could mirror

changes in P1. Evaluation of the isolated impact of cloud

and/or precipitation on P1234 is reserved for future study.

(ii) Role of snow percolation process

Delay in the DPTs (i.e., the time difference in DPTs

if a process’ contribution is considered or not) caused by

the process of snow percolation (P12 2 P1) is also ex-

pected to decrease as the melt season progresses be-

cause of decreasing snow depth and increasing liquid

water content. This is evident in Figs. 8 and 9, where the

difference between P12 and P1 is generally larger in the

beginning than the end of the melt season for all years.

The average (P12 2 P1) at the start and end of the melt

season over all years is 2.7 and 0.2 h, respectively.

However, the delay from translation of meltwater

2236 JOURNAL OF HYDROMETEOROLOGY VOLUME 17

FIG. 8. Results of peak hour variation after discounting different processes for 24WYs:P1234 is the original modeled peak hour;P1 is the

peak hour of melt flux when translation time through snowpack is not accounted for, P12 is the peak hour from translated liquid water flux,

and P123 is based on the original model but assuming translation in the river does not happen. Detailed explanations of the four exper-

iments are presented in section 3a.

AUGUST 2016 CHEN ET AL . 2237

through the snowpack showed a nonmonotonic decrease

and could vary significantly from day to day. One ex-

ample of such day-to-day variance could be seen onWY

days 196 and 197 in WY 2006 (Fig. 9). On WY day 196,

P1-forced DPT was determined by an intense rain event

occurring late in the day at 1800 LT. This intense rain

event resulted not only in accelerated melt, but trans-

lation of rainwater through the snowpack, leading to a

reduction in snow translation delay (P12 2 P1) of 3 h

[based on Eq. (1)]. The following day (WY day 197),

where melt was forced primarily by diurnal energy fluxes,

the snow translation delay was 6h. The difference in the

time of meltwater translation through the snow between

WY days 196 and 197 was mainly caused by the intense

rain event on WY day 196 that increased the percolation

flux. Our analysis shows that a decreasing trend in snow

depth during the melt season generally leads to a de-

creasing trend in (P12 2 P1). However, timing of rain-on-

snow events during the day may interfere with this trend

and can lead to an abrupt increase in (P12 2 P1).

(iii) Role of surface/subsurface flow

The streamflow in RME watershed is controlled pri-

marily by subsurface processes (see section 3a). As the

melt season progresses, soil moisture storages become

increasingly saturated, water-table height increases, and

the contributions of subsurface processes to discharge

increase. Since the subsurface hydrologic conductivity in

the watershed decreases with depth, net increase in

water-table height during the melt season leads to a

consequent increase in effective hydraulic conductivity

of the watershed (Wildenschild and Jensen 1999; Yeh

and Harvey 1990; Quinton and Marsh 1998; Quinton

and Gray 2003). Though the groundwater table during

the melt period generally reduced after the peak melt, it

still remained higher than its level at the beginning of

melt season. In fact, the average difference in ground-

water table between the beginning and end of the melt

season across the simulation years was 2.1m. As a result,

the delay in timing of peak streamflow caused by sub-

surface flow (P123 2 P12) is expected to decrease. This is

evident in Fig. 8, where the difference between P123 and

P12 is generally larger in the beginning than the end of

the melt season. Average (P123 2 P12) at the start and

end of melt season over all years was 4.0 and 1.6 h, re-

spectively. However, (P123 2 P12) does not decrease

monotonically during the melt season. A closer look at

the melt flux peak and corresponding streamflow sug-

gested that mild increases in (P1232 P12) on consecutive

days were often observed, mostly because of a flat hy-

drograph around the peak (see section 4a). However, on

certain days, (P123 2 P12) exhibited jumps as large as

5 h (an abrupt increase during WY 1984 is highlighted

in Fig. 10). Interestingly, the abrupt jump identified in

WY 1984 occurred on days with very small differences

in melt and groundwater distributions. Further investi-

gation of WY 1984 melt flux and streamflow onWY day

236 revealed that both the original melt flux and trans-

lation of meltwater through the snow were trimodal on

this day, with the middle peak being the largest. How-

ever, daily streamflow showed only two peaks due to dry

antecedent groundwater conditions early in the day.

Because of the multiple melt recharge pulses during the

day, the difference in the hour of peak streamflow for

FIG. 9. Peak time and delay caused by different factors in WY 2006 (P1 is peak hour from melt

flux when translation time through snowpack is not accounted for, P12 2 P1 is peak hour delay

caused by snow translation process, P123 2 P12 is peak hour delay caused by ground translation

process, and P1234 2 P123 is peak hour delay caused by in channel translation process).

2238 JOURNAL OF HYDROMETEOROLOGY VOLUME 17

days with multimodal melt flux could not be compared

to days with a single peak. Variations in (P123 2 P12)

were also affected by dynamic changes in subsurface

moisture and its distribution, which influenced in-

filtration capacity and drainage pathways. Our analysis

shows an increasing trend in effective hydraulic con-

ductivity due to a rising groundwater table during the

melt season, which generally leads to a decreasing trend

in (P123 2 P12). However, days with a multimodal melt

flux can exhibit abrupt jumps in (P123 2 P12) because of

dynamic changes in losses and subsurface storage.

(iv) Role of stream channel translation

Impacts of stream channel translation onDPTs during

melt season were generally within an hour because of

the small size of the RME catchment (0.4 km2) and the

limited stream channel (,500m). This is illustrated by

Fig. 8, which shows that P123 and P1234 overlap each

other on most of the days during the melt season. Re-

solving DPTs at finer time steps might reveal contribu-

tions from this process to be on the order of minutes for

days with overlapping P123 and P1234; however, to

maintain consistency in the analyses given that the

temporal resolution of the observation data is an hour,

only hourly difference in P123 and P1234 is reported here.

(v) Relative role of each process in delaying the timeof peak streamflow

Among the above four processes, P1 controls the

timing of melt flux while the other three processes

contribute to the delay in the translation of meltwater

from the snow through the soil and groundwater to the

stream channel and finally to the stream gauge. The

peak streamflow timing delays attributable to each

process are taken as differences in simulated peak flows

before and after including each process in the simula-

tions. Average delays due to percolation of flow through

the snowpack (P2), flow through the subsurface (P3),

and flow in the stream channel (P4) were 1.7, 2.3, and

0.2 h, respectively. The average delay values indicate

that subsurface translation played the dominant role in

delaying the peak times, with snow percolation being the

secondmost important factor. The influence of stream

translation process on the streamflow DPTs was negli-

gible compared to the other two factors, which is con-

sistent with the findings in Lundquist et al. (2005).

As noted in the previous sections, the delay effect

caused by the percolation and subsurface processes was

generally larger in the beginning of the melt season than

the end. To identify the relative contribution from each

process at different times during the melt season, the melt

period was evenly divided into three periods: 1) the early

melt period (deep snowpack 1 deep groundwater table),

2) the rapid melt period (rapidly reducing snowpack 1rapidly increasing groundwater table), and 3) the late

melt period (thin snowpack1 shallow groundwater table).

Average timing delay for DPTs during the three intervals

as caused by each process was calculated and compared.

Results indicate that the relative importance of indi-

vidual processes change during the melt season.

FIG. 10. (a) Peak time and delay caused by ground translation process (P123 2 P12) in WY

1984. (b) Original melt, melt after translation through snowpack, and streamflow time series

for WY 1984.

AUGUST 2016 CHEN ET AL . 2239

During the early melt period in the wettest 5 years,

average delay caused by snow translation and sub-

surface flow was 3.4 and 2.3 h, respectively. During this

period, translation through snow was the dominant

control in determining DPTs. In the rapid melt period,

delays caused by translation through snow and sub-

surface processes were 1.3 and 2.2 h, respectively. Dur-

ing this period, subsurface processes became the

dominant factor. In the late melt period, translations

through snow and the subsurface contributed evenly to

the delay in DPTs, with delay time being 1.2 h for both.

For the driest 5 years, subsurface processes were the

dominant factor for delay throughout the melt season.

Average delay caused by translation through the sub-

surface for the early melt period, rapid melt period, and

late melt period were 7.8, 4.7, and 2.7 h, respectively,

while corresponding delays caused by snow translation

were only 1.3, 0.7, and 0.6 h, respectively. The contri-

bution to DPT from translation through snow was gen-

erally larger in the early melt period, while the contribution

from the subsurface process was much more prominent

in the later periods when most of the snow had ablated.

The importance or controlling influence of a process on

changes to DPT during particular times of the melt sea-

son does not always mean that it is also the main de-

terminant on seasonal DPT variations, especially the

seasonal decreasing trend. Our analysis indicates that

while subsurface flow processes were more dominant in

determining the actual streamflow peak time, the de-

creasing trend in DPT was controlled primarily by the

snow translation process. Variations in net energy input,

on the other hand, mainly affected day-to-day variations

in DPT during the melt season.

b. How do DPTs vary interseasonally and what arethe processes that control this variation?

1) INTERSEASONAL DPT VARIATIONS

Seasonal average DPTs (for days exhibiting melt-

affected signal) showed significant variations among

years, ranging from 4.0 to 7.1 h after local noon in the

observed data, 3.2 to 7.6 h in the modeled absolute

DPTs, and 4.4 to 7.4 h in Q1% modeled results. These

variations are likely the result of interaction among the

controlling processes identified in section 3b.

It is important to recognize that the differences in

seasonal average DPT among individual years are also

influenced by the variations in the number and timing of

melt days between different years, which in turn are

controlled by the peak SWE, soil temperature, abrupt

changes in meteorological conditions, and water-table

depth. Initiation of the melt season varied by as many as

28 days (earliest start WY day 5 183, latest start WY

day5 211) in the observed data and 40 days in modeled

streamflow results (earliest start WY day 5 184, latest

start WY day 5 224). Similarly, the number of days

with a diurnal melt signal ranged from 13 to 54 days in

observed data and 6 to 38 days in modeled DPT results.

Depending on the timing of melt-affected days during

the melt season, the process controls that determined

DPTs such as energy input to the snow, its depth, and

melt recharge into the soil may vary, thus resulting in

interseasonal variations in DPTs.

Interseasonal variations also occurred in seasonal

shifts in DPTs. Seasonal DPT shift, defined as the dif-

ference in daily peak time between the first third and the

last third of the melt season, averaged 2.4 h (range of

1–5 h) in the observed data, 3.6 h (range of 1–9.8 h) in the

modeled data, and 3.1 h (range of 0.8–6.3 h) in Q1%

modeled results, among the years of study. The average

peak hour difference between the first and last days of

the melt season for the three streamflow series was

found to be 4.1, 5.5, and 4.1 h, respectively. This differ-

ence ranged from 10 (early season) to 1 h (late season) in

observed data, 12 to 1 h in modeled results, and 11 to 1h

in Q1% modeled results.

2) INFLUENCE OF PHYSICAL CONTROLS ON

INTERSEASONAL PEAK TIME AND ITS

VARIATIONS

Analyses of the relative role of each process control

on average DPT delay for the entire simulation period

[see section 4a(2)(v)] demonstrated that subsurface

translation played the most dominant role on the aver-

age DPT, followed by translation through snow.

However, a closer look revealed that delay times caused

by snow translation, subsurface flow, and river trans-

lation processes at a seasonal scale varied by as much as

0.5–3.2, 0.9–6.4, and 0.1–0.9 h, respectively. The wide

range in delay time caused by each individual process

raises the possibility of different processes dominating in

different years. Further analyses revealed that the snow

translation process was dominant in delaying DPTs in

wetter years while subsurface flowwas dominant in drier

years. For example, average delay times caused by snow

translation, subsurface flow, and stream channel trans-

lation time were 2.8, 1.9, and 0.3 h, respectively, in the

wettest fiveWYs during the simulation period, while the

corresponding delay times in the driest five WYs were

1.2, 5.8, and 0.4 h, respectively. The delay in DPTs from

snow translation process was more prominent in early

melting period. The average delay times due to snow

translation, subsurface flow, and stream channel trans-

lation in the five wettest years during the first third of the

melt season were 4.1, 2.4, and 0.4 h, respectively. The

delay from subsurface flow in dry years tended to be

2240 JOURNAL OF HYDROMETEOROLOGY VOLUME 17

prominent throughout the melt period. For example,

average delays due to subsurface flow processes in the

five driest years were 6.5 and 4.6 h during the first and

last third of the melt period, while the delays due to

snow translation for the corresponding periods were

only 2.0 and 0.9 h. Overall, the average translation time

through snow showed a high correlation with snowfall

amount (r5 0.80). This was not surprising, as years with

greater snowfall amounts tended to generate snowpacks

with greater depths, leading to longer melt translation

times. In addition, more snowfall resulted in a higher

groundwater table near the end of melt season, which in

turn reduced the delay attributable to subsurface flow

processes. As a result, variation of average delay time

caused by subsurface flow processes had a negative

correlation with annual snowfall (r520.46). This weak

correlation revealed that other confounding factors such

as the multimodal behavior of melt flux and change in

net storage and losses [as discussed in section 4a(2)(iii)]

may also contribute to variations in delay time between

years. Since increased snowfall can lead to a larger DPT

delay from the snow translation process but a smaller

DPT delay from subsurface flow, these two translation

processes may compensate each other. This results in

seasonally moderate average DPT delays in wetter

years. The correlation between seasonally averaged

DPT and snowfall amount is only 0.19, indicating that

the average DPT delay may not show a clear relation-

ship to snow depth andmelt recharge volume. However,

the shift in DPT during each melt season (i.e., the dif-

ference in DPT between beginning and end of the melt

season) was still larger in years with high snowfall

amounts because of both larger decreases in snow

translation and subsurface flow time between the begin-

ning and end of each melt season. Average DPT shift for

each year, calculated as the difference in the last third and

first third of themelt season, had a correlation of 0.42with

the snowfall amount. Our analysis indicates that wet

years tended to have larger DPT shift within the melt

season. While the process of translation of meltwater

through snow was a more dominant control on DPTs in

wetter years than in drier years, the influence was more

important in the beginning of themelt period. In contrast,

subsurface flow processes were a more important control

in drier years and maintained this influence on DPTs

throughout the melt season in these years.

5. Summary and conclusions

This study explored the day-to-day intra- and inter-

seasonal variation in the timing of peak daily streamflow

(DPT) in a snow-dominated watershed. We quantita-

tively evaluated the role of hydrologic process controls

on DPT and its variations. Results indicate that the

physics-based integrated model, which consisted of a

snowmelt model coupled to a hydrologic model, was

able to reasonably capture both DPT and its seasonal

variations, with the exception of the first few days in the

melt season in drier years. For days when the model

could capture the melt-affected signal shown in ob-

served data, 80% of variation in observed DPT could be

captured by the model. A process-unmixing approach

was then used to evaluate the relative influence of pro-

cess controls in determining DPT and its variation dur-

ing the melt season. Results show that subsurface flow

was the most dominant influence on DPT delays in this

catchment. The influence of meltwater translation

through the snowpack was a close second. Translation

time in the river channel had negligible contributions to

both peak hour and its variations because of the small

spatial area assessed in this study. Though subsurface

flow was assumed to have minimal effect on DPT in

previous studies (e.g., Lundquist et al. 2005), the results

presented here indicate that for basins with groundwater

flow as the dominant control in the streamflow diurnal

signal (such as RME), the effect of subsurface could not

be neglected.

The relative influence of process controls on DPT

varies during the melt season. Intraseasonally, delay in

peak time from translation of meltwater through snow

was generally larger early in the melt period while sub-

surface flow played a much more prominent role later in

the melt period when most of the snow had ablated. The

average shift in DPT to earlier in the day in the later part

of themelt season wasmostly contributed by the process

of meltwater translation through the snowpack, with

contributions from subsurface flow being secondary.

Interseasonally, translation of meltwater through snow

is the most dominant process in delaying DPT in wet

years while subsurface flow plays a more dominant role

in drier years. Overall, the contribution of meltwater

translation through snow and the subsurface on DPT

delay show a high positive correlation and a moderate

negative correlation with seasonal snow amount, re-

spectively. Average DPT delay due to translation

through snow is larger in wet years because of larger

snow depth. At the same time, the contribution of

translation through subsurface on DPT delay is smaller

in wet years because of increased effective subsurface

conductivity. While of only marginal influence in this

watershed, translation time through stream channels can

be expected to be more important in larger watersheds

with longer river reaches. Our results identify the dominant

process controls on DPTs at both intra- and interseasonal

scales and could be used to prioritize measurement strate-

gies for determination andmonitoring the timing of peak

AUGUST 2016 CHEN ET AL . 2241

daily streamflow. The results suggest that one should be

cautious while analyzing DPT delays, as they are the

product of interactions between several processes con-

trolled by the complex physiography and variable weather

conditions typically encountered in mountain regions.

The physically based coupled modeling system pre-

sented here is ideally suited for process-unmixing ex-

periments to isolate the role of individual processes on

daily peak time. The methodology can be used in other

watersheds and in conjunction with other process-

explicit models. The presented analyses also highlight

that diurnal streamflow data may carry important in-

formation regarding the hydrologic partitioning in the

watershed and can further aid in model diagnosis and

validation. For example, overall increase in observed

daily streamflow magnitude (before the seasonal peak)

even as the daily melt peak varied contrapositively (see

section 3a) indicates that groundwater contribution to

streamflow during the melt season was significant in the

RME watershed. Similarly, the inability of the model to

capture early diurnal signals in dry, cool years [see sec-

tion 2c(2)(ii)(A)] while it overestimated the ground-

water recharge indicates errors in simulated runoff

during early melt season and highlight the need to in-

corporate the effects of changes in hydraulic conduc-

tivity vis-à-vis soil temperature.

It is important to be aware that the magnitude of DPT

delays and the relative contribution of each process

presented in this paper are specific to this watershed and

may vary elsewhere. The relative controls on meltwater

translation through snowpack, the subsurface, and in the

stream may vary between watersheds and can be influ-

enced by a range of factors, including hydrologic parti-

tioning of melt between surface and subsurface flow,

changes in streambed hydraulic conductivity triggered

by temperature variations, dynamics of watershed hy-

drologic connectivity and preferential pathways in the

snowpack, and anthropogenic activities (Gribovszki et al

2006; Lundquist and Cayan 2002; Marks et al 1998;

McNamara et al 2005; Morgenschweis 1995). The

modeling experiment conducted here does not account

for uncertainty. Further studies focused on providing de-

tailed maps of subsurface properties, evolution of pref-

erential pathways in snowpack, and relative contributions

of surface and subsurface discharge that could be used to

refine our results and provide an estimate of, and ulti-

mately help reduce, the uncertainty.

Acknowledgments. Datasets used in the paper are

publicly available from the Northwest Watershed Re-

search Center, USDA anonymous ftp site (ftp://ftp.nwrc.

ars.usda.gov/public/RME_25yr_database). The data and

analysis presented in this paper were funded in part by

NSF CAREER award (EAR 1454983) and USDA-ARS

CRIS Snow and Hydrologic Processes in the Intermoun-

tain West (5362-13610-008-00D). We thank Jeff Dozier

and Jessica Lundquist for providing constructive com-

ments that greatly improved this manuscript.

REFERENCES

Allen, R. G., L. S. Pereira, D. Raes, and M. Smith, 1998: Crop

evapotranspiration: Guidelines for computing crop water

requirements. FAO Irrigation and Drainage Paper 56, 300

pp. [Available online at www.fao.org/docrep/X0490E/

X0490E00.htm.]

Ambach, W., M. Blumthaler, and P. Kirchlechner, 1981: Applica-

tion of the gravity flow theory to the percolation of melt water

through firn. J. Glaciol., 27 (95), 67–75.

Anderson, E. A., 1976: A point energy andmass balancemodel of a

snow cover. NOAA Tech. Rep. NWS 19, 150 pp. [Available

online at http://amazon.nws.noaa.gov/articles/HRL_Pubs_PDF_

May12_2009/HRL_PUBS_51-100/81_A_POINT_ENERGY_

AND_MASS.pdf.]

Bain, M. B., J. T. Finn, and H. E. Booke, 1988: Streamflow regu-

lation and fish community structure. Ecology, 69, 382–392,

doi:10.2307/1940436.

Barnett, T. P., J. C. Adam, and D. P. Lettenmaier, 2005: Po-

tential impacts of a warming climate on water availability in

snow-dominated regions. Nature, 438, 303–309, doi:10.1038/

nature04141.

Barry, R., M. Prévost, J. Stein, and A. P. Plamondon, 1990: Ap-

plication of a snow cover energy and mass balance model in a

balsam fir forest. Water Resour. Res., 26, 1079–1092,

doi:10.1029/WR026i005p01079.

Bengtsson, L., 1982: Groundwater and meltwater in the snowmelt

induced runoff. Hydrol. Sci. J., 27, 147–158, doi:10.1080/

02626668209491097.

Burt, T. P., and P. J. Williams, 1976: Hydraulic conductivity in

frozen soils. Earth Surf. Processes, 1, 349–360, doi:10.1002/

esp.3290010404.

Caine, N., 1992: Modulation of the diurnal streamflow response by

the seasonal snowcover of an alpine basin. J. Hydrol., 137,

245–260, doi:10.1016/0022-1694(92)90059-5.

Caprio, F. M., 1966: Pattern of plant development in the western

United States. Montana Agricultural Experiment Station

Bulletin 607, 42 pp.

Carroll, T., D. Cline, C. Olheiser, A. Rost, A. Nilsson, G. Fall,

C. Bovitz, and L. Li, 2006: NOAA’s national snow analyses.

Proc. 74th Annual Western Snow Conf., Las Cruces, NM,

Western Snow Conference, 31–43. [Available online at http://

www.westernsnowconference.org/node/837.]

Chen, X., M. Kumar, and B. L. McGlynn, 2015: Variations in

streamflow response to large hurricane-season storms in a

southeastern U.S. watershed. J. Hydrometeor., 16, 55–69,

doi:10.1175/JHM-D-14-0044.1.

Colbeck, S. C., 1975: A theory for water flow through a layered

snowpack. Water Resour. Res., 11, 261–266, doi:10.1029/

WR011i002p00261.

——, and G. Davidson, 1973: Water percolation through homo-

geneous snow. IAHS Publ., 107, 242–257.

DeWalle, D. R., and A. Rango, 2008: Principles of Snow Hydrol-

ogy. Cambridge University Press, 428 pp.

Dunne, T., and R. D. Black, 1971: Runoff processes during snowmelt.

Water Resour. Res., 7, 1160–1172, doi:10.1029/WR007i005p01160.

2242 JOURNAL OF HYDROMETEOROLOGY VOLUME 17

——,A.G. Price, and S. C. Colbeck, 1976: The generation of runoff

from subarctic snowpacks. Water Resour. Res., 12, 677–685,

doi:10.1029/WR012i004p00677.

Flerchinger, G. N., 2000: The simultaneous heat and water (SHAW)

model: Technical documentation. Tech. Rep. NWRC 2000-09, 40

pp. [Available online at http://www.ars.usda.gov/SP2UserFiles/

Place/20520000/ShawDocumentation.pdf.]

——, K. R. Cooley, and D. R. Ralston, 1992: Groundwater re-

sponse to snowmelt in a mountainous watershed. J. Hydrol.,

133, 293–311, doi:10.1016/0022-1694(92)90260-3.

——,M. S. Seyfried, and S. P. Hardegree, 2006: Using soil freezing

characteristics to model multi-season soil water dynamics.

Vadose Zone J., 5, 1143–1153, doi:10.2136/vzj2006.0025.

Garen, D. C., and D. Marks, 2005: Spatially distributed energy

balance snowmelt modelling in a mountainous river basin:

Estimation of meteorological inputs and verification of model

results. J. Hydrol., 315, 126–153, doi:10.1016/j.jhydrol.2005.03.026.

Gottardi,G., andM.Venutelli, 1993:A control-volumefinite-element

model for two-dimensional overland flow. Adv. Water Resour.,

16, 277–284, doi:10.1016/0309-1708(93)90019-C.

Graf, W. L., 1999: Dam nation: A geographic census of American

dams and their large-scale hydrologic impacts. Water Resour.

Res., 35, 1305–1311, doi:10.1029/1999WR900016.

Grant, L., M. Seyfried, and J. McNamara, 2004: Spatial variation

and temporal stability of soil water in a snow-dominated,

mountain catchment. Hydrol. Processes, 18, 3493–3511,

doi:10.1002/hyp.5798.

Gray, D. M., P. G. Landine, and R. J. Granger, 1985: Simulating

infiltration into frozen prairie soils in streamflowmodels.Can.

J. Earth Sci., 22, 464–472.Gribovszki, Z., P. Kalicz, and M. Kucsara, 2006: Streamflow

characteristics of two forested catchments in Sopron Hills.

Acta Silvatica et Lignaria Hung., 2, 81–92.

——, ——, J. Szilágyi, and M. Kucsara, 2008: Riparian zone

evapotranspiration estimation from diurnal groundwater level

fluctuations. J.Hydrol., 349, 6–17, doi:10.1016/j.jhydrol.2007.10.049.

——, J. Szilágyi, and P. Kalicz, 2010: Diurnal fluctuations in shal-

low groundwater levels and streamflow rates and their

interpretation—A review. J. Hydrol., 385, 371–383,

doi:10.1016/j.jhydrol.2010.02.001.

Grover, N. C., and A. W. Harrington, 1943: Stream Flow: Mea-

surements, Records and Their Uses. Wiley, 363 pp.

Hanson, C. L., D. Marks, and S. S. Van Vactor, 2001: Long-term

climate database, Reynolds Creek Experimental Watershed,

Idaho, United States. Water Resour. Res., 37, 2839–2841,

doi:10.1029/2001WR000417.

Horiguchi, K., and R. D. Miller, 1983: Hydraulic conductivity

functions of frozen materials. Proc. 4th Int. Conf. on Perma-

frost, Fairbanks, AK, University of Alaska Fairbanks, 504–508.

Johnson, B., B. Malama, W. Barrash, and A. N. Flores, 2013:

Recognizing and modeling variable drawdown due to evapo-

transpiration in a semiarid riparian zone considering local

differences in vegetation and distance from a river source.

Water Resour. Res., 49, 1030–1039, doi:10.1002/wrcr.20122.

Jordan, P., 1983a:Meltwatermovement in a deep snowpack: 1. Field

observations. Water Resour. Res., 19, 971–978, doi:10.1029/

WR019i004p00971.

——, 1983b: Meltwater movement in a deep snowpack: 2. Simu-

lation model. Water Resour. Res., 19, 979–985, doi:10.1029/

WR019i004p00979.

Kobayashi, D., 1986: Separation of a snowmelt hydrograph by

stream conductance. J. Hydrol., 84, 157–165, doi:10.1016/

0022-1694(86)90049-1.

——, and H. Motoyama, 1985: Effect of snow cover on time lag of

runoff from a watershed. Ann. Glaciol., 6, 123–125.

Kumar, M., 2009: Toward a hydrologic modeling system. Ph.D.

thesis, The Pennsylvania State University, 274 pp.

——, and C. J. Duffy, 2015: Exploring the role of domain parti-

tioning on efficiency of parallel distributed hydrologic model

simulations. J. Hydrogeol. Hydrol. Eng., 4 (1), doi:10.4172/

2325-9647.1000119.

——, ——, and K. M. Salvage, 2009: A second-order accurate, fi-

nite volume–based, integrated hydrologic modeling (FIHM)

framework for simulation of surface and subsurface flow.

Vadose Zone J., 8, 873–890, doi:10.2136/vzj2009.0014.

——, D. Marks, J. Dozier, M. Reba, and A. Winstral, 2013: Eval-

uation of distributed hydrologic impacts of temperature-index

and energy-based snowmodels.Adv.Water Resour., 56, 77–89,

doi:10.1016/j.advwatres.2013.03.006.

Link, T., and D. Marks, 1999: Distributed simulation of snowcover

mass- and energy-balance in the boreal forest. Hydrol. Pro-

cesses, 13, 2439–2452, doi:10.1002/(SICI)1099-1085(199910)13:14/

15,2439::AID-HYP866.3.0.CO;2-1.

Loheide, S. P., II, 2008: A method for estimating subdaily evapo-

transpiration of shallow groundwater using diurnal water table

fluctuations. Ecohydrology, 1, 59–66, doi:10.1002/eco.7.

——, and J. D. Lundquist, 2009: Snowmelt-induced diel fluxes

through the hyporheic zone. Water Resour. Res., 45, W07404,

doi:10.1029/2008WR007329.

Lowry, C. S., J. S. Deems, I. I. Loheide, P. Steven, and J. D.

Lundquist, 2010: Linking snowmelt-derived fluxes and

groundwater flow in a high elevation meadow system, Sierra

Nevada Mountains, California. Hydrol. Processes, 24, 2821–

2833, doi:10.1002/hyp.7714.

Lundquist, J. D., and D. R. Cayan, 2002: Seasonal and spatial

patterns in diurnal cycles in streamflow in the western

United States. J. Hydrometeor., 3, 591–603, doi:10.1175/

1525-7541(2002)003,0591:SASPID.2.0.CO;2.

——, and M. D. Dettinger, 2005: How snowpack heterogeneity

affects diurnal streamflow timing. Water Resour. Res., 41,

W05007, doi:10.1029/2004WR003649.

——, D. R. Cayan, and M. D. Dettinger, 2004: Spring onset

in the Sierra Nevada: When is snowmelt independent of

elevation? J. Hydrometeor., 5, 327–342, doi:10.1175/

1525-7541(2004)005,0327:SOITSN.2.0.CO;2.

——, M. D. Dettinger, and D. R. Cayan, 2005: Snow-fed

streamflow timing at different basin scales: Case study of

the Tuolumne River above Hetch Hetchy, Yosemite,

California. Water Resour. Res., 41, W07005, doi:10.1029/

2004WR003933.

Marks, D., 2001: Introduction to special section: Reynolds Creek

Experimental Watershed. Water Resour. Res., 37, 2817,

doi:10.1029/2001WR000941.

——, and A. Winstral, 2001: Comparison of snow deposition, the

snow cover energy balance, and snowmelt at two sites in a

semiarid mountain basin. J. Hydrometeor., 2, 213–227,

doi:10.1175/1525-7541(2001)002,0213:COSDTS.2.0.CO;2.

——, J. Kimball, D. Tingey, and T. Link, 1998: The sensitivity

of snowmelt processes to climate conditions and forest

cover during rain-on-snow: A case study of the 1996 Pa-

cific Northwest flood. Hydrol. Processes, 12, 1569–1587,

doi:10.1002/(SICI)1099-1085(199808/09)12:10/11,1569::

AID-HYP682.3.0.CO;2-L.

——, J. Domingo, D. Susong, T. Link, and D. Garen, 1999:

A spatially distributed energy balance snowmelt model

for application in mountain basins. Hydrol. Processes, 13,

AUGUST 2016 CHEN ET AL . 2243

1935–1959, doi:10.1002/(SICI)1099-1085(199909)13:12/13,1935::

AID-HYP868.3.0.CO;2-C.

——, A.Winstral, andM. Seyfried, 2002: Simulation of terrain and

forest shelter effects on patterns of snow deposition, snowmelt

and runoff over a semi-arid mountain catchment. Hydrol.

Processes, 16, 3605–3626, doi:10.1002/hyp.1237.

Marsh, P., and J. W. Pomeroy, 1996: Meltwater fluxes at

an arctic forest-tundra site. Hydrol. Processes, 10, 1383–1400, doi:10.1002/(SICI)1099-1085(199610)10:10,1383::

AID-HYP468.3.0.CO;2-W.

——, and M. K. Woo, 1985: Meltwater movement in natural het-

erogeneous snow covers. Water Resour. Res., 21, 1710–1716,doi:10.1029/WR021i011p01710.

Maulé, C., and J. Stein, 1990: Hydrologic flow path definition and

partitioning of spring meltwater.Water Resour. Res., 26, 2959–2970, doi:10.1029/WR026i012p02959.

McNamara, J. P., D. Chandler, M. Seyfried, and S. Achet, 2005:

Soil moisture states, lateral flow, and streamflow generation

in a semi-arid, snowmelt-driven catchment.Hydrol. Processes,

19, 4023–4038, doi:10.1002/hyp.5869.

Morgenschweis, G., 1995: Kurzzeitige Vorhersage der Wasser-

entnahme aus einem Flussgebiet. Vortragsmanusskript zur 8

Wissenschaft Tagung Hydrologie und wasserwirtschaft zum

Thema Verfügbarkeit von Wasser, Bochum, Germany, Ruhr-

Universität Bochum, 16 pp.

Neumann, D. W., B. Rajagopalan, and E. A. Zagona, 2003:

Regression model for daily maximum stream temper-

ature. J. Environ. Eng., 129, 667–674, doi:10.1061/

(ASCE)0733-9372(2003)129:7(667).

Pfeffer, W. T., T. H. Illangasekare, andM. F. Meier, 1990: Analysis

andmodeling of melt-water refreezing in dry snow. J. Glaciol.,

36 (123), 238–246.

Poff, N. L., J. D. Allan, M. B. Bain, J. R. Karr, K. L. Prestegaard,

B. D. Richter, R. E. Sparks, and J. C. Stromberg, 1997: The

natural flow regime. BioScience, 47, 769–784, doi:10.2307/

1313099.

Qu, Y., and C. J. Duffy, 2007: A semidiscrete finite volume for-

mulation for multiprocess watershed simulation. Water Re-

sour. Res., 43, W08419, doi:10.1029/2006WR005752.

Quinton, W. L., and P. Marsh, 1998: The influence of mineral

earth hummocks on subsurface drainage in the continuous

permafrost zone. Permafrost Periglacial Processes, 9, 213–

228, doi:10.1002/(SICI)1099-1530(199807/09)9:3,213::AID-

PPP285.3.0.CO;2-E.

——, andD.M. Gray, 2003: Subsurface drainage from organic soils

in permafrost terrain: The major factors to be represented in a

runoff model. Proc. of the Eighth Int. Conf. on Permafrost,

Davos, Switzerland, International Permafrost Association,

917–922. [Available online at http://research.iarc.uaf.edu/

NICOP/DVD/ICOP%202003%20Permafrost/Pdf/Chapter_

161.pdf.]

Reba, M. L., D. Marks, M. Seyfried, A. Winstral, M. Kumar,

and G. Flerchinger, 2011a: A long-term data set for hy-

drologic modeling in a snow-dominated mountain catch-

ment. Water Resour. Res., 47, W07702, doi:10.1029/

2010WR010030.

——, ——, A. Winstral, T. Link, and M. Kumar, 2011b: Sensitivity

of the snowcover in a mountain basin to variations in climate.

Hydrol. Processes, 25, 3312–3321, doi:10.1002/hyp.8155.

Refsgaard, J. C., 1997: Parameterisation, calibration and validation

of distributed hydrological models. J. Hydrol., 198, 69–97,

doi:10.1016/S0022-1694(96)03329-X.

Richards, L. A., 1931: Capillary conduction of liquids through

porous mediums. J. Appl. Phys., 1, 318–333, doi:10.1063/

1.1745010.

Rutter, A. J., K. A. Kershaw, P. C. Robins, and A. J. Morton, 1971: A

predictive model of rainfall interception in forests, 1. Derivation

of the model from observations in a plantation of Corsican pine.

Agric. Meteor., 9, 367–384, doi:10.1016/0002-1571(71)90034-3.

Shi, Y., K. J. Davis, F. Zhang, and C. J. Duffy, 2014: Evaluation of

the parameter sensitivities of a coupled land surface hydro-

logic model at a critical zone observatory. J. Hydrometeor., 15,

279–299, doi:10.1175/JHM-D-12-0177.1.

Soylu, M. E., J. D. Lenters, E. Istanbulluoglu, and S. P. Loheide,

2012: On evapotranspiration and shallow groundwater fluc-

tuations: A Fourier-based improvement to theWhite method.

Water Resour. Res., 48, W06506, doi:10.1029/2011WR010964.

Strelkoff, T., 1970: Numerical solution of Saint-Venant equations.

J. Hydraul. Div., 96 (1), 223–252.

Tobin, C., B. Schaefli, L. Nicótina, S. Simoni, G. Barrenetxea,

R. Smith, M. Parlange, and A. Rinaldo, 2013: Improving the

degree-day method for sub-daily melt simulations with

physically-based diurnal variations. Adv. Water Resour., 55,

149–164, doi:10.1016/j.advwatres.2012.08.008.

Van Genuchten, M. T., 1980: A closed-form equation for

predicting the hydraulic conductivity of unsaturated soils.

Soil Sci. Soc. Amer. J., 44, 892–898, doi:10.2136/

sssaj1980.03615995004400050002x.

Wang, R., M. Kumar, andD.Marks, 2013: Anomalous trend in soil

evaporation in a semi-arid, snow-dominated watershed. Adv.

Water Resour., 57, 32–40, doi:10.1016/j.advwatres.2013.03.004.

Watanabe, K., and M. Flury, 2008: Capillary bundle model of hy-

draulic conductivity for frozen soil. Water Resour. Res., 44,W12402, doi:10.1029/2008WR007012.

Wildenschild, D., and K. H. Jensen, 1999: Laboratory in-

vestigations of effective flow behavior in unsaturated hetero-

geneous sands. Water Resour. Res., 35, 17–27, doi:10.1029/98WR01958.

Winstral, A., and D. Marks, 2002: Simulating wind fields and snow

redistribution using terrain-based parameters to model snow

accumulation and melt over a semi-arid mountain catchment.

Hydrol. Processes, 16, 3585–3603, doi:10.1002/hyp.1238.

——, ——, and R. Gurney, 2009: An efficient method for distrib-

uting wind speeds over heterogeneous terrain. Hydrol. Pro-

cesses, 23, 2526–2535, doi:10.1002/hyp.7141.Yeh, T. C. J., and D. J. Harvey, 1990: Effective unsaturated hy-

draulic conductivity of layered sands. Water Resour. Res., 26,

1271–1279, doi:10.1029/WR026i006p01271.

Yu, X., C. Duffy, D. C. Baldwin, and H. Lin, 2014: The role of

macropores and multi-resolution soil survey datasets for dis-

tributed surface–subsurface flow modeling. J. Hydrol., 516,

97–106, doi:10.1016/j.jhydrol.2014.02.055.

2244 JOURNAL OF HYDROMETEOROLOGY VOLUME 17