Has spring snowpack declined in the Washington Cascades?

Abstract

Our best estimates of 1 April snow water equivalent (SWE) in the Cascade Mountains of Washington State indicate a substantial (roughly 15–35%) decline from mid-century to 2006, with larger declines at low elevations and smaller declines or increases at high elevations. This range of values includes estimates from observations and hydrologic modeling, reflects a range of starting points between about 1930 and 1970 and also reflects uncertainties about sampling. The most important sampling issue springs from the fact that half the 1 April SWE in the Cascades is found below about 1000 m, where sampling was poor before 1945. Separating the influences of temperature and precipitation on 1 April SWE in several ways, it is clear that long-term trends are dominated by trends in temperature, whereas variability in precipitation adds "noise" to the time series. Consideration of spatial and temporal patterns of change rules out natural variations like the Pacific Decadal Oscillation as the sole cause of the decline. Regional warming has clearly played a role, but it is not yet possible to quantify how much of that regional warming is related to greenhouse gas emissions.

Full text

HESSD
4, 2073–2110, 2007
Has spring snowpack
declined in the
Washington
Cascades?
P. Mote et al.
Title Page
Abstract Introduction
Conclusions References
Tables Figures
J I
J I
Back Close
Full Screen / Esc
Printer-friendly Version
Interactive Discussion
EGU
Hydrol. Earth Syst. Sci. Discuss., 4, 2073–2110, 2007
www.hydrol-earth-syst-sci-discuss.net/4/2073/2007/
© Author(s) 2007. This work is licensed
under a Creative Commons License.
Hydrology and
Earth System
Sciences
Discussions
Papers published in Hydrology and Earth System Sciences Discussions are under
open-access review for the journal Hydrology and Earth System Sciences
Has spring snowpack declined in the
Washington Cascades?
P. Mote1, A. Hamlet1,2 , and E. Salath´
e1
1Climate Impacts Group, University of Washington, Seattle, USA
2Department of Civil and Environmental Engineering, University of Washington, Seattle, USA
Received: 26 June 2007 – Accepted: 26 June 2007 – Published: 5 July 2007
Correspondence to: P. Mote (philip@atmos.washington.edu)
2073

HESSD
4, 2073–2110, 2007
Has spring snowpack
declined in the
Washington
Cascades?
P. Mote et al.
Title Page
Abstract Introduction
Conclusions References
Tables Figures
J I
J I
Back Close
Full Screen / Esc
Printer-friendly Version
Interactive Discussion
EGU
Abstract
Our best estimates of 1 April snow water equivalent (SWE) in the Cascade Mountains
of Washington State indicate a substantial (roughly 15–35%) decline from mid-century
to 2006, with larger declines at low elevations and smaller declines or increases at high
elevations. This range of values includes estimates from observations and hydrologic5
modeling, reflects a range of starting points between about 1930 and 1970 and also
reflects uncertainties about sampling. The most important sampling issue springs from
the fact that half the 1 April SWE in the Cascades is found below about 1000m, where
sampling was poor before 1945. Separating the influences of temperature and precip-
itation on 1 April SWE in several ways, it is clear that long-term trends are dominated10
by trends in temperature, whereas variability in precipitation adds “noise” to the time
series. Consideration of spatial and temporal patterns of change rules out natural vari-
ations like the Pacific Decadal Oscillation as the sole cause of the decline. Regional
warming has clearly played a role, but it is not yet possible to quantify how much of that
regional warming is related to greenhouse gas emissions.15
1 Introduction
Phase changes of water from ice to liquid are among the most visible results of a
warming climate, and include declines in summer Arctic sea ice, the Greenland ice
sheet, and most of the world’s glaciers (Lemke et al., 2007). These components of
the cryosphere have slow response times and take decades to centuries to come into20
equilibrium with a change in climate. By contrast, seasonal snow cover disappears
every year and therefore each year’s snow cover is the product of a single year’s climate
conditions. This makes changes happen more rapidly but also substantially decreases
the signal-to-noise ratio.
Northern hemisphere spring snow cover has declined about 8% over the period of25
record 1922–2005 (Lemke et al., 2007) and the declines have predominantly occurred
2074

HESSD
4, 2073–2110, 2007
Has spring snowpack
declined in the
Washington
Cascades?
P. Mote et al.
Title Page
Abstract Introduction
Conclusions References
Tables Figures
J I
J I
Back Close
Full Screen / Esc
Printer-friendly Version
Interactive Discussion
EGU
between the 0◦and 5◦C isotherms, underscoring the feedbacks between snow and
temperature near the freezing point. That is, small amounts of warming can provide
the impetus to melt snow, which increases absorption of solar radiation by the under-
lying surface and provides additional warming. This feedback is particularly strong in
spring in lower latitudes when solar radiation is rather strong; in winter solar radiation5
is considerably weaker, which helps explain why large-scale declines in December and
January are small to nonexistent.
For the mountainous regions of the western U.S., summers are usually dry and
snowmelt provides approximately 70% of annual streamflow. Observations of moun-
tain snowpack in the spring provide important predictions of summer streamflow for10
agriculture, hydropower, and flood control, among other needs. These mountain snow
observations also provide a unique climate record of changes in the mountains, and
demonstrate long-term declines at approximately 75% of sites since 1950 (Mote et al.,
2005) or 1960 (Mote, 2006). The enormous importance of snowmelt for western water
resources, and the sensitivity of snow accumulation and melt to air temperature, pro-15
vide strong motivations for better understanding the role that warming has played in
regional changes in snowpack.
Bales et al. (2006) estimated the fraction of annual precipitation falling between 0◦C
and –3◦C, what one might call warm snow. The Cascade and Olympic Mountains
of Washington and Oregon stand out as having the highest fraction of warm snow20
in the continental U.S. In this paper we examine changes in areally averaged 1 April
SWE, with three purposes in mind: First, to provide greater spatial specificity to the
results of Mote et al. (2005); second, to demonstrate the challenges in quantifying
such change; and third, to determine whether a human influence on snowpack can
be detected. This third point exploits the facts that a human influence on climate can25
be detected as a warming, at least on the scale of the western U.S. (Stott, 2003),
that the best estimates of anthropogenic change in precipitation for the Northwest are
for no measurable change (Salathe et al., 2007); and that the pattern of response to
warming is to lose more snow at low elevations than at high elevations (Mote et al.,
2075

HESSD
4, 2073–2110, 2007
Has spring snowpack
declined in the
Washington
Cascades?
P. Mote et al.
Title Page
Abstract Introduction
Conclusions References
Tables Figures
J I
J I
Back Close
Full Screen / Esc
Printer-friendly Version
Interactive Discussion
EGU
2005; Regonda et al., 2005).
2 Data and methods
Snow surveys have been conducted at over a thousand carefully surveyed and marked
locations, or “snow courses”, in western North America for decades. When new snow
courses were established, surveyors had several goals in siting them, for example to5
minimize the effects of blowing snow by siting the snow courses in natural clearings
on relatively level terrain, generally in bowls or benches. Since the purpose was water
supply forecasting, the measurements of SWE were most prevalent initially near the
date of peak SWE, 1 April. Over time the sites were typically visited more frequently,
monthly from 1 January to 1 June. Automated sites began to be used in the 1980s10
and in some cases replaced snow courses. For Washington’s Cascades (west of 120◦
longitude) and Olympics, 49 long term records are available (Fig. 1, Table 1). Spatial
coverage is uneven, and was especially uneven before 1938 when all snow courses
(and aerial markers) were at high elevation in the North Cascades.
The data are provided by the Natural Resources Conservation Service (NRCS) of15
the US Department of Agriculture (http://www.wcc.nrcs.usda.gov/snow/snowhist.html).
Climate data came from the Western Regional Climate Center (http://www.wrcc.dri.
edu/cgi-bin/divplot1 form.pl?4504). The North Pacific Index (NPI) was obtained from
http://www.cgd.ucar.edu/∼jhurrell/np.html. Detailed consideration of individual time se-
ries is beyond the scope of this paper, but users can plot maps of linear trends and also20
time series at individual sites at http://www.climate.washington.edu/trendanalysis/.
The smoothing performed in some of the figures uses the locally weighted regression
(“loess”) scheme (Cleveland, 1993) with partial reflection. Loess computes each value
for the smoothed time series by performing a least-squares linear fit on weighted adja-
cent data, where the analyst determines a suitable window to define adjacent and the25
weighting favors closer points. The smoothing parameter αis chosen so that the resid-
uals have very little low-frequency variability; in this case α=0.6. To handle variations
2076

HESSD
4, 2073–2110, 2007
Has spring snowpack
declined in the
Washington
Cascades?
P. Mote et al.
Title Page
Abstract Introduction
Conclusions References
Tables Figures
J I
J I
Back Close
Full Screen / Esc
Printer-friendly Version
Interactive Discussion
EGU
near the endpoints, a linear combination of two loess-smoothed versions is used: first,
the original time series, for which the end points are somewhat weakly constrained,
and second, the middle third of a time series formed by reflecting the original time se-
ries (x1,...xj) about its endpoints (xj...x1,x1,... xj, xj...x1) with the result that the first
derivative of the curve must be zero at the endpoints. This “partial reflection” approach5
seems to produce the best results at the endpoints.
For calculating statistical significance, a two-sided t-test (p<0.05) is used, and de-
grees of freedom are first estimated by checking for autocorrelation. In the time series
shown here, the lag-1 autocorrelation is negative, so the number of degrees of freedom
is simply the number of data points.10
3 Simulation with the VIC hydrology model
Physically based hydrologic models can be used to achieve some of the goals of this
study, by providing spatially uniform fields of estimation and also evaluating the climatic
factors behind the variability and trends in snowpack. In this study (as in Mote et al.,
2005, and Hamlet et al., 2005) we use the Variable Infiltration Capacity (VIC) hydrologic15
model (Liang et al., 1994; Cherkauer and Lettenmaier, 2003) implemented over the
western U.S. at 1/8th degree spatial resolution. As inputs to the VIC model, we used a
gridded dataset of daily precipitation and maximum and minimum temperature (Hamlet
and Lettenmaier 2005), at the same 1/8th degree resolution, from 1915 to 2003. Spatial
interpolation of daily data from the Cooperative weather (Coop) network, with terrain-20
dependent algorithms, provide the spatial coverage. The temporal characteristics of
the dataset are nudged toward the monthly values from the climate-quality dataset
from the Historical Climatology Network (HCN) (Karl et al., 1990). The methods used
to produce the driving data, and evaluation of the resulting hydrologic simulations using
observed streamflow records, are reported in more detail by Hamlet and Lettenmaier25
(2005). The VIC simulation was conducted at a daily time step for the water balance
and at a time step of one hour for the snow model. Additional details on the hydrologic
2077

HESSD
4, 2073–2110, 2007
Has spring snowpack
declined in the
Washington
Cascades?
P. Mote et al.
Title Page
Abstract Introduction
Conclusions References
Tables Figures
J I
J I
Back Close
Full Screen / Esc
Printer-friendly Version
Interactive Discussion
EGU
model and its implementation are reported by Hamlet et al. (2005).
Using VIC simulations of snow over the Western U.S., Mote et al. (2005) corrobo-
rated observed trends in snowpack derived from snow course observations from 1950–
1997. The model was shown to do a remarkably good job of reproducing topographic
gradients in snowpack trends over the West as a whole and particularly in the Cas-5
cades in the Pacific Northwest. Based on this evaluation of the model, the simulations
were extended to 1915–2003 and long term trends in snowpack for several different
periods from 1915–2003 were examined (Hamlet et al., 2005). That study also exam-
ined, in detail, the relative roles of temperature and precipitation trends on snowpack
trends in each period.10
We now examine the performance of the VIC model at the snow courses in Wash-
ington over the period 1950–1997. Note first that there are several reasons why the
SWE simulated by VIC and the SWE measured at a snow course might differ: terrain,
temporal offsets, interpolation of driving data, spurious trends in driving data, and land-
cover change at the snow course. These are discussed more by Mote et al. (2005).15
First, while a VIC grid cell includes 15 snow “bands” with different elevation and land
cover over a roughly 10×12 km area, a snow course reports the snowpack at a spe-
cific site in terrain that may or may not be representative of a larger area, though the
locations for snow courses are chosen with the goal of representing SWE over a signif-
icant portion of a watershed. Second, the actual date of a snow course measurement20
can be several days before 1 April, so a storm or melt event between the observation
date and the 1 April VIC date. Third, VIC uses weather data from stations that may be
50 km or more from the grid cell and may not represent local hydroclimatic conditions
despite the efforts of the terrain-dependent algorithm to do so. Fourth, the driving data
(especially precipitation) might contain spurious trends owing, for example, to growth25
of trees over a rain gauge; efforts are made in constructing the HCN dataset to find
and correct such sources of error, but some undoubtedly remain. Finally, changes in
land cover (e.g., incursion of forest into a previously open area) at the site of the snow
course may also introduce trends that are not shared over the 10×12km area of a VIC
2078

HESSD
4, 2073–2110, 2007
Has spring snowpack
declined in the
Washington
Cascades?
P. Mote et al.
Title Page
Abstract Introduction
Conclusions References
Tables Figures
J I
J I
Back Close
Full Screen / Esc
Printer-friendly Version
Interactive Discussion
EGU
grid cell. For all these reasons, and probably more, we would not expect a close match
between the mean VIC grid cell value and the mean snow course value nor between
the trends.
Despite these potential sources of differences between VIC SWE and observed
SWE, Table 2 and Fig. 2 show a strong correspondence at about three-quarters of5
the snow courses examined. Correlations are fairly high, with a median of 0.8, rms
differences are typically a small fraction of the mean, with a median of 18 cm, and
the median absolute difference in the trends is 16%. For about a quarter of the sites,
trends disagree by more than 26% as with Deer Park and Dock Butte aerial marker
(AM) shown here. From this analysis it is impossible to deduce the reasons for the dis-10
crepancies between VIC and observations at some sites. The strong correspondence
at the majority of sites, though, combined with the previously demonstrated success
of VIC at simulating streamflow in this region, provides a strong basis for using VIC to
produce spatially averaged estimates of total 1 April SWE.
Figure 3a shows the results of spatially averaging the VIC 1 April SWE over the Cas-15
cades domain for the 1916–2003 period of record. Note that a number of low-elevation
grid cells with no snow on 1 April are included in the domain for averaging, which pro-
duces a much lower average SWE than the observed time series we construct below.
The interannual variability of the time series shown in Fig. 3a is substantial, with a co-
efficient of variation of 0.35. Highest 1 April SWE occurred in 1956, followed by 1997,20
1950, and 1999; very low values occurred in 1941, 1977, and 1992. The smooth curve
shows the interdecadal variability, with relatively snowy periods centered around 1950,
1970, and 1999, and relatively snow-poor periods centered around 1992 and 1926. A
linear fit to the entire period of record gives a decline of 16%. We will explore below
the choice of starting date for reporting period-of-record trend.25
In order to elucidate the climatic factors behind these changes, we examine also
a simulation discussed by Hamlet et al. (2005, which see for details) in which the
interannual variability in precipitation is removed from the driving dataset. This way,
interannual variability in SWE is produced primarily from variations in temperature. The
2079

HESSD
4, 2073–2110, 2007
Has spring snowpack
declined in the
Washington
Cascades?
P. Mote et al.
Title Page
Abstract Introduction
Conclusions References
Tables Figures
J I
J I
Back Close
Full Screen / Esc
Printer-friendly Version
Interactive Discussion
EGU
result (Fig. 3b) has substantially less interannual variability, and slightly smaller trend
(13%). Trends since 1950 were larger, but again were dominated by temperature not
by precipitation.
4 Challenges in quantifying “the snowpack” of a mountain range
Using observations to estimate the total snow in a mountain range at a given time is5
difficult for several reasons, some of which were discussed above in connection with
the VIC simulation. Precipitation varies strongly with altitude, aspect, and slope, and
local winds interacting with terrain and vegetation cover can redistribute falling snow
leading to drifts, bare spots, and other heterogeneous features. Over the course of
the winter, melt events may remove snowpack as runoffor infiltration into the soil, and10
may also change or “ripen” the snowpack. Point measurements of snow depth or snow
water equivalent (SWE) are subject to all of these factors and more.
Sampling is another important consideration. Estimating the total SWE for a moun-
tain range depends very much on the choices one makes about which snow courses
to include. Horizontal coverage is one factor – note the patchiness of coverage in15
Fig. 1, with some river basins well-sampled and others unsampled. Ideally, an ob-
servationally derived estimate would balance spatial coverage with temporal longevity.
However, another crucial factor, even more than horizontal coverage, is vertical cov-
erage. As Fig. 4a shows, the number of long-term snow courses (those in existence
by 1960 and still providing data in 2006) changed rapidly during the 1940s, so that20
the choice of a cutoffyear (say 1945 or 1950) influences how many snow courses are
available for analysis and their mean elevation. In particular, the early courses had a
much higher mean elevation before about 1945, when the mean elevation stabilized
at about 1300 m. This fact is crucially important in selecting snow courses because
(a) according to our estimates from observations (below) and from VIC, roughly half25
the snowpack in the Cascades on 1 April lies at elevations below about 1000 m, so
it is important to adequately sample that half of the snowpack; and (b) low-elevation
2080

HESSD
4, 2073–2110, 2007
Has spring snowpack
declined in the
Washington
Cascades?
P. Mote et al.
Title Page
Abstract Introduction
Conclusions References
Tables Figures
J I
J I
Back Close
Full Screen / Esc
Printer-friendly Version
Interactive Discussion
EGU
snowpack is much more sensitive to temperature than high-elevation snowpack (Mote,
2003, Mote et al., 2005, Hamlet et al. 2005), hence any response to regional warming
is most likely to be detectable at lower elevations. Consequently, requiring that snow
courses be active before about 1945 leads to a considerable sampling bias. As we will
show below, there are ways of judiciously addressing this sampling bias to estimate5
variability before 1945.
Another consequence of the choice of starting date concerns the grand mean SWE
for the available snow courses. High elevation snow courses also have higher mean
SWE. Fig. 4b shows the mean SWE and elevation for each of the snow courses, and
shows how the lower elevations were poorly sampled before 1940 and much better10
sampled by 1945. The grand mean was lower in 1940 than in 1945.
As shown by Mote et al. (2005) and Mote (2006), the clearest signature of warming-
induced changes in snowpack is that trends become more positive with increasing
elevation. Figure 5 shows how consistently the trends depend on elevation, regardless
(to first order) of starting year, with lines drawn to indicate how the trends depend15
quantitatively on elevation. The slopes of the lines are 59±28%/1000 m for 1940-
present, 40±13%/1000 m for 1950-present, 37±13%/1000 m for 1960-present, and
41±13%/1000 m for 1970-present. Trends 1930-present (the plus symbols in the first
panel) are about 10% higher than 1940-present for those six snow courses or aerial
markers that date back to 1930, but the small number and especially the lack of sam-20
pling below 1400 m prevents us from drawing conclusions about trends vs. elevation
over the 1930-present period of record. The striking consistency in trends vs. elevation
from one decade to the next, even for 1940 with very poor sampling below 1400 m,
emphasizes the pervasive role of warming in affecting trends in SWE at lower eleva-
tions, as will be shown below. From this analysis it is clear that the vertical fingerprint25
of warming has been detected in Cascades snowpack.
2081

HESSD
4, 2073–2110, 2007
Has spring snowpack
declined in the
Washington
Cascades?
P. Mote et al.
Title Page
Abstract Introduction
Conclusions References
Tables Figures
J I
J I
Back Close
Full Screen / Esc
Printer-friendly Version
Interactive Discussion
EGU
5 Observationally based estimates of change over time
For each starting year from 1927 to 1960, we composed a separate time series of
total SWE for the Washington Cascades from available snow courses, with different
cutoffyears from 1935 to 1960 determining a different mix of snow courses. That is,
time series 1 goes from 1927 to 2006 and includes acceptable snow courses that span5
1927 to 2006; time series 2 goes from 1928 to 2006, etc. To be acceptable, a snow
course had to have some data on or before the cutoffyear, some data on or after 2003,
and be at least 80% complete. Missing data were filled by recursively comparing with
highly correlated time series as follows. For each snow course, use the best-correlated
other snow course along with the linear regression between them to fill missing years.10
If missing data remain, the next-best-correlated snow course is used, and so on.
The full set of regionally averaged time series is shown in Fig. 6. As each snow
course is added it changes the average SWE, as suggested in Fig. 4b, largely as a
function of elevation of the new snow course. Additions beyond about 1950 make
little difference, as suggested also by Fig. 4b. Smoothing the time series (Fig. 6b)15
clarifies the differences among the curves and also shows three interdecadal peaks in
SWE, centered on about 1952, 1973, and 1998, with low points in the 1980s and early
1990s, similar to the VIC results repeated from Fig. 3a. Despite the fact that 2006 was
a relatively good year for SWE, the very low years of 2001 and 2005 mean that the
smoothed time series all end in a dip, which lies below the previous dip in around 198020
dominated by the very snow-poor years of 1977 and 1981 and other below-average
years around that time. The ranking of snowiest or least snowy years, too, depends
on which curve is chosen; for the 1938 curve the snowiest years are 1973, 1976, and
1956 (quite different from VIC), whereas for the 1956 curve the snowiest years are
1956, 1973, and 1999. Clearly any results derived from these regionally averaged time25
series will depend on judicious choice of time series for further analysis.
We now explore the correspondence of linear fits between VIC and observations
for different starting points (Table 3). Each entry shows the trend from the starting
2082

HESSD
4, 2073–2110, 2007
Has spring snowpack
declined in the
Washington
Cascades?
P. Mote et al.
Title Page
Abstract Introduction
Conclusions References
Tables Figures
J I
J I
Back Close
Full Screen / Esc
Printer-friendly Version
Interactive Discussion
EGU
year to 2003 (the last year of the VIC simulation); the observational estimates are
for a changing mix of snow courses up to 1960. Most mid-century starting points
yield a trend around –25% for observations and VIC, with a range of about –15% to
–35%. Observed and VIC trends are within about 6% for starting points between 1945
and 1956; before 1945, the dearth of low-elevation snow courses affects the observed5
trends, as would be expected from Fig. 4. It is less clear why the trends diverge for
starting points beyond the late 1950s, perhaps land use change (encroachment of
forests into snow courses). Trends are statistically significant (i.e. the 95th percentil
estimate of the trend is negative) for observations for starting points between 1945
and 1954, and for VIC only between 1942 and 1950. The 5th percentile estimates10
are mostly in the –35% to –55% range for observed starting points between 1942 and
1974, and for all VIC starting points between 1916 (not shown in the table) and 1974.
Trends are generally positive, but not significant, for starting points after 1975.
With the goal of characterizing the variability over as long a time as possible without
too severely underemphasizing low elevations, we use the time series starting in 1944.15
From Table 3 it is clear that this is a fairly conservative estimate of the trends.
5.1 Regression analysis with temperature and precipitation
Complementing the analysis of the temperature-only VIC simulation presented in Sec-
tion 4, we use empirical relationships between 1 April SWE from each snow course and
reference time series for temperature and precipitation in November through March,20
roughly as in Mote (2006). For this analysis through 2006, US Historical Climate
Network (USHCN). USHCN data were not available for 2006, so instead we use the
Climate Division data for Washington Climate Division 4 (west slopes Cascades and
foothills). As a first step in understanding the linear relationships between SWE and the
climate variables, we show in Fig. 7 the correlations between 1 April SWE and Nov-25
Mar mean temperature and precipitation, for different periods of analysis. Note that,
consistent with Fig. 4 and the sensitivity of SWE at lower elevations to temperature,
the correlation of regionally averaged SWE with precipitation drops somewhat and the
2083

HESSD
4, 2073–2110, 2007
Has spring snowpack
declined in the
Washington
Cascades?
P. Mote et al.
Title Page
Abstract Introduction
Conclusions References
Tables Figures
J I
J I
Back Close
Full Screen / Esc
Printer-friendly Version
Interactive Discussion
EGU
correlation with temperature changes substantially from –0.3 to about –0.5. Next, the
climate data are used in multiple linear regression to estimate the regionally averaged
snowpack for the regionally averaged time series of 1 April SWE since 1944. Regres-
sion analysis against November–March precipitation and temperature is performed, as
in Mote (2006):5
SWE(t)=SWE0+aTT(t) +aPP(t) +ε(t)
where SWE(t) is the 1 April SWE in year t, SWE0is the mean SWE, aTand aPare
the regression coefficients for temperature and precipitation respectively, and T(t) and
P(t) are the values of temperature and precipitation in year t, and finally ε(t) is the
residual. We can then also define S(t)=SWE0+aTT(t) +aPP(t), that is, the time series10
generated directly from the climate data. Figure 8 shows SWE(t) and the resulting
S(t) for the time series starting in 1944. The correlation between the two is 0.88,
and the correspondence is obviously quite good in most years, though the trend in
S(t) is substantially less than observed for a number of possible reasons (mainly that
when variance is partitioned between a deterministic and a stochastic component, the15
regressed extreme values tend to be less extreme and the slope of the fitted trend is
consequently also reduced). There may also be some errors in the long-term trends
in the climate division data; this calculation could be redone using USHCN stations in
and near the Cascades, augmented for the most recent years using other data.
Using the regression approach permits a separation of the effects of temperature20
SWE0+aTT(t) and precipitation SWE0+aPP(t) (Fig. 9). The two contributions sum to
the –12 cm in Fig. 8, though the percentage changes do not sum since they are com-
puted relative to the value in 1944 for each linear fit, which is somewhat different. This
analysis shows that to first order the long-term change is a result of warming, with
precipitation adding noise that makes the trend detection more difficult, especially on25
shorter time periods (see Fig. 3 and Table 3). We return to this subject below, where
we examine the output of the simulation with the hydrology model. For the 1956–2006
time series (not shown), the observed decline is 27%, the trend in S(t) is –15%, the
trend in aPP is 0cm or 2% and the trend in aTT is –5 cm or –17%. Hence, even for
2084

HESSD
4, 2073–2110, 2007
Has spring snowpack
declined in the
Washington
Cascades?
P. Mote et al.
Title Page
Abstract Introduction
Conclusions References
Tables Figures
J I
J I
Back Close
Full Screen / Esc
Printer-friendly Version
Interactive Discussion
EGU
these fairly different starting points the results are similar: the trend is dominated by
aTT(t).
5.2 Further distinguishing roles of temperature and precipitation
The ratio of 1 April SWE over precipitation (Fig. 10) shows what fraction of precipitation
falling from November through March remains on the ground on 1 April, what we might5
call the storage efficiency; that fraction has declined by 28% for observations and by
33% for VIC. Note that because the precipitation measurements come from weather
stations, which are almost all located at low elevations where less precipitation falls,
the actual precipitation was multiplied by 1.5 to reflect high-elevation enhancement of
precipitation (this factor does not affect the computed trends). Storage efficiency was10
worst for 2005, when much of the winter precipitation fell in three warm wet storms in
what was otherwise not an exceptionally warm winter (Fig. 9).
We now show yet another way to examine the data with a view to determining
whether warming has played a role in the observed changes. Recognizing that the fun-
damental pattern of change in SWE in simulations of warming is a pronounced loss at15
lower elevations, we use the same set of snow courses from 1944 as in Figs. 6 through
9, and construct time series of average SWE at elevations above 1200 m (“high”) and
below 1200 m (“low”). The ratio of low to high elevation SWE is shown in Fig. 11, along
with a linear fit. The slope parameter is statistically significant.
5.3 Area-weighting with elevation20
In the foregoing analysis, time series at the various snow courses were combined by
simple linear averaging. The area in each elevation band, however, decreases with
elevation, so that even though low elevations have relatively little SWE per unit area
(Fig. 3), they hold a large total quantity of snow. In order to estimate the importance of
this effect, we construct elevation bands at 60m vertical intervals, based on National25
Geophysical Data Center 2-min etopo2 digital elevation for the Washington Cascades.
2085

HESSD
4, 2073–2110, 2007
Has spring snowpack
declined in the
Washington
Cascades?
P. Mote et al.
Title Page
Abstract Introduction
Conclusions References
Tables Figures
J I
J I
Back Close
Full Screen / Esc
Printer-friendly Version
Interactive Discussion
EGU
The hypsometric (cumulative area-altitude) relationship is shown in Fig. 12a along with
the altitudes of the elevation bands containing snow courses used in constructing the
1944–2006 time series shown in Figs. 6b–10. In Fig. 12b, the mean SWE from 1944
to 2006 at each snow course is multiplied by the area in that elevation band, and
the resulting profile is interpolated to the elevation bands in which there are no snow5
courses, then smoothed slightly. The median elevation is only 640 m, and half the SWE
is found at elevations below 1025 m, which is represented by only four snow courses
for the 1944–2006 time series (Table 1). Fully 90% of the snow on 1 April is found
below 1600 m.
Using the area-weighting represented in Fig. 12b, we construct another time series10
of regionally averaged 1 April SWE (Fig. 13). The strong weighting of the lower eleva-
tions, where observed declines have been larger, produces a much larger linear trend,
–35% over the 1944–2006 period, than in the unweighted time series (–24%, Fig. 8).
Furthermore, it brings the observed trend exactly into line with the VIC trend over the
1944–2003 period of the VIC simulation.15
6 Discussion and conclusions
When observations are weighted by the area of each elevation band, there is a re-
markable congruence between observed and modeled estimates of regionally aver-
aged SWE (Fig. 13). This is all the more remarkable because the point measurements
of the observations and the gridded values of the model are fundamentally different20
quantities, and furthermore the observations are potentially affected by a different set
of time-varying biases (e.g., land cover change). Both the observationally derived and
modeled estimates of total 1 April SWE have strengths and weaknesses. The model-
ing produces robust spatially distributed estimates of SWE and the resulting streamflow
agrees well with observations, as does the simulated snow for roughly three-quarters of25
the observed snow courses; on the other hand, VIC requires interpolated temperature
and precipitation which may not always accurately reflect the daily values experienced
2086

HESSD
4, 2073–2110, 2007
Has spring snowpack
declined in the
Washington
Cascades?
P. Mote et al.
Title Page
Abstract Introduction
Conclusions References
Tables Figures
J I
J I
Back Close
Full Screen / Esc
Printer-friendly Version
Interactive Discussion
EGU
in the remote mountains. The observed values are actual measurements but may be
influenced by site changes like canopy growth, and their spatial distribution is so sparse
that areally averaged estimates are sensitive to choices about which snow courses to
include. Such choices are especially important with respect to elevation, since (1) a
great deal of snow on 1 April lies at relatively low elevations that are represented by a5
very few snow courses and (2) long-term warming trends disproportionately affect low
elevation snow (Fig. 5).
The smooth curves in Figs. 3 and 6 emphasize that there is substantial variability in
1 April SWE on timescales longer than several years. During the period of adequate
instrumental coverage (1944–2006) three relatively snowy periods have occurred; in10
the observations, each of these had a slightly lower peak than the last. In the VIC
simulation, all three had roughly the same peak value. This difference is reflected also
in the larger negative linear trend in the observations than in the VIC simulation over
this time period and in the median difference between observed and VIC trends at
individual locations over the 1950-97 period of analysis (Table 2). Considerable further15
analysis will be required to elucidate the reasons for the differences in trends between
model and observations.
Within a decade or so of 1955, selection of starting date does not greatly influence
the linear declines computed from observations. The linear decline of 1 April SWE
with ending point 2006 and from starting points between 1916 and 1970, calculated20
either with VIC data or observations (the latter starting no earlier than 1944 to ensure
adequate sampling of lower elevations), is roughly –15% to –34% and mostly around
–25% (Table 3). Area-weighting the observations produces larger declines. For most
of the mid-century starting points the trend in observed regionally averaged 1 April
SWE is statistically significant. The drawback of using linear fits is that they obscure25
interesting and significant features like the local maximum in SWE in the late 1990s.
Fluctuations and trends in temperature and precipitation play distinct roles in the be-
havior of these time series. Rising temperatures have clearly dominated in determining
the regionally averaged trend in SWE over periods longer than about 30 years, as is
2087

HESSD
4, 2073–2110, 2007
Has spring snowpack
declined in the
Washington
Cascades?
P. Mote et al.
Title Page
Abstract Introduction
Conclusions References
Tables Figures
J I
J I
Back Close
Full Screen / Esc
Printer-friendly Version
Interactive Discussion
EGU
evident in (a) the VIC simulation with fixed precipitation (Fig. 3b); (b) the regression
analysis (Fig. 9); (c) the declining “storage efficiency” (Fig. 10); (d) the strong depen-
dence of trends on elevation (Fig. 5); and (e) the declining ratio of low elevation to
high elevation snowpack. Precipitation produces much of the interannual variability
but plays little role in trends except over short time periods (say, post-1970), and linear5
trends in precipitation are not statistically significant over any period in contrast to those
of temperature.
Further interpretation is required though. What is one to make of the positive (but
not significant) trends in SWE since the early 1970s? What roles have Pacific decadal
variability and human-induced global warming played in these changes?10
Before answering these questions, we first note that the Pacific Decadal Oscillation
(Mantua et al., 1997) was predominantly in one phase from 1945 to 1976 and in an-
other phase from 1977 to perhaps 1997, with ambiguous phase since then, and that
the period since 1997 (the past decade). We also note that with a declining storage
efficiency caused by rising temperatures, the extremely high seasonal precipitation in15
1997 and 1999, which helped make the most recent 10-year period the wettest period
in the instrumental record and produced the peak in the smooth curve in Figs. 3 and
6, only managed to raise 1 April SWE to about half the value of the ∼1950 peak in the
observed estimate, though they are closer in the VIC simulation.
We propose and test three hypotheses to answer these questions about interpre-20
tation; these hypotheses each have corrolaries about future variability of Cascades
SWE.
–Hypothesis 1: the variability and trends in SWE can be largely explained by Pacific
decadal variability. Table 3 can be interpreted as showing that the significant
trends found for starting points between 1945 and 1954 depend entirely on the25
1977 PDO regime shift.
–Corollary 1: Future variability in SWE is unpredictable but will vary around the
long-term mean.
2088

HESSD
4, 2073–2110, 2007
Has spring snowpack
declined in the
Washington
Cascades?
P. Mote et al.
Title Page
Abstract Introduction
Conclusions References
Tables Figures
J I
J I
Back Close
Full Screen / Esc
Printer-friendly Version
Interactive Discussion
EGU
–Hypothesis 2: Human-induced global warming in the Northwest includes large
increases in winter precipitation that have canceled the influence of warming dur-
ing the period when human influence on global climate emerged, i.e. since the
mid-1970s.
–Corollary 2: A human influence on Cascades SWE will be difficult to detect be-5
cause trends in precipitation will continue to cancel trends in temperature.
–Hypothesis 3: a long-term decline in 1 April SWE driven by warming is detectable,
but has been partly obscured in the past decade or so by random fluctuations of
precipitation.
–Corollary 3: future SWE will decline, especially at low elevations.10
Tools used to test these hypotheses include time series analysis, spatial analysis, and
modeling of future SWE with VIC.
1) The subject of natural variability and its connection to variations and trends in
western snowpack has been examined by Mote (2006), Stewart et al. (2005), and
Clark et al. (2001). Mote (2006) regressed individual time series of SWE in the West15
on the North Pacific Index, an atmospheric index sensitive to both ENSO and PDO
variability, and found that 10–60% of trend in SWE at snow courses in the Pacific
Northwest could be explained by the North Pacific Index (an indicator of the strength
of the Aleutian Low that responds both to the PDO and to ENSO), a larger fraction
than for the PDO index itself. Response to NPI is characterized by an out-of-phase20
relationship between the PNW and the Southwest, and the strong positive trends in
precipitation in the Southwest since 1950 (chiefly the result of a severe drought in
the 1950s and early 1960s) overwhelmed the influence of rising temperatures there,
producing positive trends in SWE at most sites.
In this hypothesis, the negative PDO phase that prevailed from 1945 to 1976 pro-25
duced the high values of SWE during that period, the positive phase of the PDO that
prevailed from 1977 to 1997 produced low values of SWE, and a negative PDO phase
2089

HESSD
4, 2073–2110, 2007
Has spring snowpack
declined in the
Washington
Cascades?
P. Mote et al.
Title Page
Abstract Introduction
Conclusions References
Tables Figures
J I
J I
Back Close
Full Screen / Esc
Printer-friendly Version
Interactive Discussion
EGU
beginning in 1999 produced high values of SWE observed since then. However, three
significant problems with this hypothesis emerge. First, the phase of the PDO is not
accurately represented by constant values that change sign every few decades, and
in particular the PDO index has not shown a clear preference for positive or negative
values in the past 10 years, nor a strong correspondence to PNW SWE. Second, even5
with the NPI or PDO regressed out of the time series of SWE, a substantial negative
trend in 1 April SWE remains and in fact becomes statistically significant. We cor-
relate and regress the 1944–2006 time series of 1 April SWE shown in Fig. 6 with
November–March NPI: the correlation is 0.53, so NPI plays a significant role in inter-
annual variability, but with NPI regressed out the trend changes only from –18.6% to10
–14.6% (Figure 14). The coefficient of variation is cut in half from 0.28 to 0.15 which is
why the trend becomes statistically significant. Third, if the period since 1997 or 1999
has been a negative phase PDO it does not explain the very low snow years of 2001
and 2005, nor the much lower storage efficiency for most years since 1997.
2) One could test the second hypothesis by testing for an emerging human influence15
on temperature and precipitation separately. A useful metric for emerging human influ-
ence is the logarithm of global CO2, a rough approximation of the radiative forcing and
hence the climatic influence of human activity. Regression of ln(CO2) on temperature
produces a fairly large coefficient, explaining much of the observed trend. However, for
precipitation the regression is close to zero. This hypothesis also is inconsistent with20
the twenty scenarios of Northwest climate produced under the auspices of the IPCC
and analyzed by Mote et al. (2005b), which suggest only modest changes (a few %)
in winter precipitation in response to rising greenhouse gases. This is consistent with
evidence in the literature that at least on the scale of the West a human influence on
air temperature can be detected (e.g., Stott, 2003) but also that neither globally nor25
regionally can a human influence on precipitation be detected (Gillett et al., 2004). In
simulations with the VIC hydrology model, even large increases in winter precipitation
were not sufficient to offset losses in the Cascades associated with warming (e.g.,
Hamlet and Lettenmaier, 1999).
2090

HESSD
4, 2073–2110, 2007
Has spring snowpack
declined in the
Washington
Cascades?
P. Mote et al.
Title Page
Abstract Introduction
Conclusions References
Tables Figures
J I
J I
Back Close
Full Screen / Esc
Printer-friendly Version
Interactive Discussion
EGU
3) The third hypothesis is supported by several lines of evidence. The time se-
ries analysis presented above suggests that precipitation behaves like random noise.
Exceptionally high values of winter precipitation in 1997 and 1999 made short-term
trends (early 1970s to present) near-zero or even positive, though none of these are
statistically significant. Indeed, the local maximum in the late 1990s was followed by5
exceptionally low values in 2001 and 2005 which produce a sharp dip in the smoothed
curves (Figs. 3b and 6). A second line of evidence concerns the spatial signal of warm-
ing, namely the elevation-dependent decline in snowpack, which clearly has emerged:
see Figs. 5 and 11. Modeling studies (e.g., Hamlet and Lettenmaier 1999, Hamlet et
al., 2005) clearly indicate that an elevation-dependent decline in snowpack is a robust10
response to regional warming under anthropogenic influence for both past and future
changes in temperature and precipitation combined. This is not to say that the ob-
served changes in SWE can be conclusively linked to rising greenhouse gases; such
attribution on the spatial scale of the Cascades is not yet possible. Furthermore, it
is not yet clear what role greenhouse gases may have played in changes in the NPI;15
some studies suggest that the observed changes in the NPI may themselves be re-
lated to anthropogenic climate change, in which case “removing” the NPI as was done
in testing Hypothesis 1 may remove both anthropogenic and natural factors.
None of these hypotheses is completely satisfactory, but it is clear that warming has
had a significant effect on Cascades snowpack especially at lower elevations where20
SWE has declined dramatically. These declines, while influenced by Pacific decadal
variability, are dominated by warming trends largely unrelated to Pacific climate vari-
ability and strongly congruent with trends expected from rising greenhouse gases.
Acknowledgements. Snow course measurements provide unique climate records of the moun-
tainous regions of the West, and we wish to thank heartily the dedicated snow surveyors of25
decades past and of today for making the measurements described herein, and NRCS for pro-
viding the data. We also thank R. Norheim for constructing the map in Fig. 1, and D. Hartmann,
R. Wood, H. Harrison, D. Lettenmaier, K. Redmond, and N. Mantua for helpful comments on
an earlier version of this manuscript. This publication was funded by the Joint Institute for the
Study of the Atmosphere and Ocean (JISAO) at the University of Washington under NOAA30
2091

HESSD
4, 2073–2110, 2007
Has spring snowpack
declined in the
Washington
Cascades?
P. Mote et al.
Title Page
Abstract Introduction
Conclusions References
Tables Figures
J I
J I
Back Close
Full Screen / Esc
Printer-friendly Version
Interactive Discussion
EGU
Cooperative Agreement No. NA17RJ1232, Contribution #1411.
References
Cherkauer, K. A. and Lettenmaier D. P.: Simulation of spatial variability in snow and frozen soil,
J. Geophys. Res., 108, 8858, doi:10.1029/2003JD003575, 2003.
Cleveland, W. S.: Visualizing Data, Hobart Press, Summit, N.J., 1993.5
Gillett, N. P., Weaver, A. J., Zwiers, F. W., and Wehner, M. F.: Detection of volcanic influence
on global precipitation, Geophys. Res. Lett., 31(12), L12217, doi:10.1029/2004GL020044,
2004b.
Hamlet, A. F. and Lettenmaier, D. P.: Production of temporally consistent gridded precipitation
and temperature fields for the continental U.S., J. Hydrometeorology, 6(3), 330–336, 2005.10
Hamlet, A. F., Mote, P. W., Clark, M. P., and Lettenmaier, D. P.: Effects of temperature and
precipitation variability on snowpack trends in the western U.S., J. Climate, 18, 4545–4561,
2005.
Karl, T. R., Williams Jr., C. N., Quinlan, F. T., and Boden, T. A.: United States Historical Cli-
matology Network (HCN) Serial Temperature and Precipitation Data, Environmental Science15
Division, Publication No. 3404, Carbon Dioxide Information and Analysis Center, Oak Ridge
National Laboratory, Oak Ridge, TN, 389 pp., 1990.
Lemke, P., Ren, J., Alley, R. B., Allison, I., Carrasco, J., Flato, G., Fujii, Y., Kaser, G., Mote,
P., Thomas, R. H., and Zhang, T., Observations: Changes in Snow, Ice and Frozen Ground.
Chapter 4 in: Climate Change 2007: The Physical Science Basis. Contribution of Work-20
ing Group I to the Fourth Assessment Report of the Intergovernmental Panel on Climate
Change, edited by: Solomon, S., Qin, D., Manning, M., Chen, Z., Marquis, M., Averyt, K. B.,
Tignor, M., and Miller, H. L., Cambridge University Press, Cambridge, United Kingdom and
New York, NY, USA, 2007.
Liang, X., Lettenmaier, D. P., Wood, E. F., and Burges, S. J.: A simple hydrologically based25
model of land surface water and energy fluxes for general circulation models, J. Geophys.
Res., 99(D7), 14 415–14 428, 1994.
Mote, P. W.: Climate-driven variability and trends in mountain snowpack in western North Amer-
ica, J. Climate, 19, 6209–6220, 2006.
2092

HESSD
4, 2073–2110, 2007
Has spring snowpack
declined in the
Washington
Cascades?
P. Mote et al.
Title Page
Abstract Introduction
Conclusions References
Tables Figures
J I
J I
Back Close
Full Screen / Esc
Printer-friendly Version
Interactive Discussion
EGU
Mote, P. W., Hamlet, A. F., Clark, M. P., and Lettenmaier, D. P.: Declining mountain snowpack
in western North America, Bull. Am. Meteorol. Soc., 86, 39–49, 2005.
Regonda S., Rajagopalan B., Clark M., and Pitlick J.: Seasonal cycle shifts in hydroclimatology
over the Western US, J. Climate, 18, 372–384, 2005.
Salath´
e, E. P., Mote, P. W., and Wiley, M. W.: Considerations for selecting downscal-5
ing methods for integrated assessments of climate change impacts, Intl. J. Climatol.,
doi:10.1002/joc.1540, 2007.
Stott, P. A.: Attribution of regional-scale temperature changes to anthropogenic and natural
causes, Geophys. Res. Lett., 30, 1728, doi:10.1029/2003GL017324
10
2093

HESSD
4, 2073–2110, 2007
Has spring snowpack
declined in the
Washington
Cascades?
P. Mote et al.
Title Page
Abstract Introduction
Conclusions References
Tables Figures
J I
J I
Back Close
Full Screen / Esc
Printer-friendly Version
Interactive Discussion
EGU
Table 1. Locations and starting year of snow courses used in this study, sorted by elevation.
“AM” refers to aerial markers, where snow depth is estimated from aerial photos of a snow
stake and SWE is inferred from snow depth.
name started longitude latitude elev (m)
MEADOW CABINS 1945 120.933 48.583 579
ROCKY CREEK 1959 121.800 48.683 640
BEAVER CREEK TRAIL 1944 121.200 48.833 670
S.FORK THUNDER CREEK 1959 121.667 48.600 670
DOMMERIE FLATS 1939 121.067 47.233 670
CITY CABIN 1948 121.517 47.317 728
THUNDER BASIN 1948 120.983 48.517 731
TUNNEL AVENUE 1941 121.350 47.317 746
TOATS COULEE 1959 119.733 48.850 867
AHTANUM RANGER STATION 1941 121.017 46.517 945
MT. GARDNER 1959 121.567 47.367 1006
FISH LAKE 1943 121.567 47.517 1027
SCHREIBERS MEADOW 1959 121.817 48.700 1036
FREEZEOUT CREEK TRAIL 1944 120.950 48.950 1067
MARTEN LAKE 1959 121.717 48.767 1097
OLALLIE MEADOWS 1945 121.450 47.383 1105
BEAVER PASS 1944 121.250 48.883 1121
DOCK BUTTE AM 1959 121.800 48.633 1158
STAMPEDE PASS 1943 121.333 47.283 1176
RUSTY CREEK 1943 119.867 48.533 1219
SATUS PASS 1957 120.683 45.983 1228
SASSE RIDGE 1944 121.050 47.367 1280
BLEWETT PASS NO. 2 1946 120.683 47.350 1301
UPPER WHEELER 1959 120.367 47.283 1341
SALMON MEADOWS 1938 119.833 48.667 1371
WATSON LAKES 1959 121.583 48.667 1371
HURRICANE 1949 123.533 47.967 1371
WHITE PASS (E. SIDE) 1953 121.383 46.633 1371
COX VALLEY 1959 123.483 47.967 1371
PARK CREEK RIDGE 1928 120.917 48.450 1402
BUMPING RIDGE 1953 121.333 46.817 1402
RAINY PASS 1930 120.717 48.567 1457
CHEWELAH 1958 117.583 48.283 1501
STEMILT SLIDE 1959 120.383 47.283 1524
EASY PASS AM 1959 121.433 48.867 1585
DEER PARK 1949 123.250 47.950 1585
BOYER MOUNTAIN 1946 117.433 48.200 1600
LITTLE MEADOWS AM 1927 120.900 48.200 1608
CAYUSE PASS 1940 121.533 46.867 1615
JASPER PASS AM 1959 121.400 48.783 1646
PARADISE PILLOW 1940 121.717 46.800 1676
MUTTON CREEK NO. 1 1938 119.867 48.667 1737
DEVILS PARK 1950 120.850 48.750 1798
LYMAN LAKE 1928 120.917 48.200 1798
GREEN LAKE 1941 121.167 46.550 1829
CORRAL PASS 1940 121.467 47.017 1829
MINERS RIDGE 1928 120.983 48.167 1890
HARTS PASS 1941 120.650 48.717 1981
CLOUDY PASS AM 1927 120.917 48.200 1981
2094

HESSD
4, 2073–2110, 2007
Has spring snowpack
declined in the
Washington
Cascades?
P. Mote et al.
Title Page
Abstract Introduction
Conclusions References
Tables Figures
J I
J I
Back Close
Full Screen / Esc
Printer-friendly Version
Interactive Discussion
EGU
Table 2. Comparison of VIC and observed 1 April SWE, 1950–1997.
snow course rms difference correlation observed trend VIC trend
Ahtanum R.S. 8 cm 0.71 –39% –54%
Beaver Creek Trail 29cm 0.64 –51% –7%
Beaver Pass 26 cm 0.60 –46% 0%
Big Boulder Creek 13 cm 0.84 –31% –34%
Blewett Pass No. 2 8 cm 0.87 –39% –43%
Boyer Mountain 34 cm 0.84 –3% –33%
Bumping Ridge Pillow 15cm 0.85 –14% –15%
Bunchgrass Meadow 13 cm 0.83 –20% –21%
Cayuse Pass 63 cm 0.78 –33% –10%
City Cabin 16 cm 0.89 –81% –65%
Corral Pass 16 cm 0.91 –25% –29%
Cox Valley 22cm 0.76 –30% –18%
Deer Park 16 cm 0.77 –64% –35%
Devils Park 19 cm 0.83 –13% –15%
Dock Butte AM 39cm 0.75 –35% –6%
Fish Lake 18 cm 0.86 –35% –44%
Freezout Creek Trail 28 cm 0.66 –45% –7%
Green Lake 16 cm 0.78 +2% +11%
Harts Pass 18 cm 0.78 –28% –2%
Hurricane Ridge 18 cm 0.80 –71% –35%
Lake Cle Elum 9 cm 0.89 –89% –65%
Lyman Lake 27 cm 0.80 –6% –11%
Marten Lake 44 cm 0.68 –5% +3%
Miners Ridge Pillow 23 cm 0.87 –11% –19%
Mount Gardner 17 cm 0.83 –56% –50%
Mutton Creek No. 1 10 cm 0.76 –30% –14%
Olallie Meadows 31cm 0.84 –58% –37%
Paradise 36 cm 0.84 –24% –10%
Park Creek Ridge 22 cm 0.80 –20% –20%
Rainy Pass 18 cm 0.72 –29% –5%
Rocky Creek 25 cm 0.78 –23% –15%
Rusty Creek 4 cm 0.82 –56% –25%
South Fork Thunder Creek 9 cm 0.77 –68% –8%
Salmon Meadows 6cm 0.80 –40% –14%
Schreibers Meadow 34 cm 0.76 –40% –5%
Stampede Pass Pillow 22 cm 0.84 –20% –28%
Stemilt Slide 8 cm 0.75 –17% –17%
Thunder Basin 42 cm 0.56 –36% –11%
Tunnel Avenue 14 cm 0.87 –56% –52%
Upper Wheeler 9 cm 0.64 –43% –17%
Watson Lakes 38 cm 0.69 –34% –5%
White Pass (East Side) 16 cm 0.92 –21% –15%
median 18 cm 0.80 median difference –15%
2095

HESSD
4, 2073–2110, 2007
Has spring snowpack
declined in the
Washington
Cascades?
P. Mote et al.
Title Page
Abstract Introduction
Conclusions References
Tables Figures
J I
J I
Back Close
Full Screen / Esc
Printer-friendly Version
Interactive Discussion
EGU
Table 3. Linear trend (%) from starting year to 2003 for observations and VIC, with 5% and
95% confidence levels of the trend.
starting yr observations VIC
5% mean 95% 5% mean 95%
1935 –15 5 27 –42 –20 1
1936 m m m –42 –19 3
1937 m m m –41 –18 5
1938 –15 1 18 –42 –19 4
1939 –18 0 17 –41 –17 7
1940 -25 -7 10 –42 –17 6
1941 –37 –9 8 –45 –21 2
1942 –39 –12 4 –49 –26 –3
1943 –44 –15 1 –51 –29 –7
1944 –47 –18 0 –50 –28 –5
1945 -53 –25 –7 –53 –32 –10
1946 –45 –27 –10 –56 –34 –13
1947 –44 –26 –8 –54 –32 –10
1948 –47 –29 –10 –56 –33 –11
1949 –48 –29 –11 –56 –33 –10
1950 –47 –28 –9 –53 –30 –6
1951 –46 –26 –6 –50 –25 0
1952 –45 –25 –5 –46 –21 4
1953 –45 –25 –4 –46 –19 6
1954 –45 –24 –3 –46 –19 7
1955 –43 –21 0 –45 –17 11
1956 –42 –20 1 –45 –16 12
1957 –37 –15 7 –34 –5 –23
1958 –37 –14 9 –35 –4 24
1959 –40 –17 6 –36 –6 26
1960 –40 –16 7 –36 –5 26
1961 –42 –19 4 –39 –8 23
1962 -43 –18 -6 –39 –7 24
1963 –45 –20 4 –41 –7 25
1964 –49 –26 –2 –45 –13 17
1965 —47 –23 1 –45 –11 21
1966 –47 –22 2 –44 –9 24
1967 –47 –21 3 –45 –10 25
1968 -46 –19 -7 –45 –8 28
1969 –49 –22 4 –50 –13 22
1970 –47 –19 8 –48 –10 27
1971 –50 –22 5 –53 –15 22
1972 –44 –14 15 –45 –5 35
1973 –35 –4 26 –37 5 49
1974 –39 –9 21 –43 –1 40
1975 –26 4 35 –30 14 60
2096

HESSD
4, 2073–2110, 2007
Has spring snowpack
declined in the
Washington
Cascades?
P. Mote et al.
Title Page
Abstract Introduction
Conclusions References
Tables Figures
J I
J I
Back Close
Full Screen / Esc
Printer-friendly Version
Interactive Discussion
EGU
Fig. 1. Map of Washington’s Cascades and Olympics with snow course locations indicated.
2097

HESSD
4, 2073–2110, 2007
Has spring snowpack
declined in the
Washington
Cascades?
P. Mote et al.
Title Page
Abstract Introduction
Conclusions References
Tables Figures
J I
J I
Back Close
Full Screen / Esc
Printer-friendly Version
Interactive Discussion
EGU
WA BUMPING LAKE o1051m v 71m
1950 1960 1970 1980 1990 2000
rmsd 9cm corr 0.87 obs tr -53% VIC tr -43%
0
20
40
60
80
100
WA BUMPING RIDGE PILLOW o1402m v -89m
1950 1960 1970 1980 1990 2000
rmsd 15cm corr 0.85 obs tr -14% VIC tr -15%
0
50
100
150
WA BUNCHGRASS MEADOW o1524m v -99m
1950 1960 1970 1980 1990 2000
rmsd 13cm corr 0.83 obs tr -20% VIC tr -21%
0
20
40
60
80
100
120
140
WA CAYUSE PASS o1615m v -16m
1950 1960 1970 1980 1990 2000
rmsd 63cm corr 0.78 obs tr -33% VIC tr -10%
0
100
200
300
400
WA CITY CABIN o 728m v 161m
1950 1960 1970 1980 1990 2000
rmsd 16cm corr 0.89 obs tr -81% VIC tr -65%
0
20
40
60
80
100
120
WA CORRAL PASS o1829m v -510m
1950 1960 1970 1980 1990 2000
rmsd 16cm corr 0.91 obs tr -25% VIC tr -29%
0
50
100
150
200
250
WA CORRAL PASS PILLOW o1829m v -461m
1950 1960 1970 1980 1990 2000
rmsd 16cm corr 0.81 obs tr -20% VIC tr -24%
0
50
100
150
WA COX VALLEY o1371m v -141m
1950 1960 1970 1980 1990 2000
rmsd 22cm corr 0.76 obs tr -30% VIC tr -18%
0
50
100
150
200
WA DEER PARK o1585m v -588m
1950 1960 1970 1980 1990 2000
rmsd 16cm corr 0.77 obs tr -64% VIC tr -35%
0
20
40
60
80
100
120
WA DEVILS PARK o1798m v -739m
1950 1960 1970 1980 1990 2000
rmsd 19cm corr 0.83 obs tr -13% VIC tr -15%
0
50
100
150
200
WA DOCK BUTTE AM o1158m v -73m
1950 1960 1970 1980 1990 2000
rmsd 39cm corr 0.75 obs tr -35% VIC tr -6%
0
50
100
150
200
250
300
WA FISH LAKE o1027m v 97m
1950 1960 1970 1980 1990 2000
rmsd 18cm corr 0.86 obs tr -35% VIC tr -44%
0
50
100
150
200
WA FISH LAKE PILLOW o1027m v 45m
1950 1960 1970 1980 1990 2000
rmsd 15cm corr 0.89 obs tr -29% VIC tr -33%
0
50
100
150
200
WA FREEZEOUT CR. TR. o1067m v -31m
1950 1960 1970 1980 1990 2000
rmsd 28cm corr 0.66 obs tr -45% VIC tr -7%
0
20
40
60
80
100
120
WA GREEN LAKE o1829m v -383m
1950 1960 1970 1980 1990 2000
rmsd 16cm corr 0.78 obs tr 2% VIC tr 11%
0
50
100
150
200
Figure 2. Observed (o) and VIC simulated (+) April 1 SWE at nine of the snow courses in Wash-
ington for 1950-97 (or shorter). Each frame includes statistics about the elevation of the snow
course (o1615m for Cayuse Pass), the difference in elevation between the VIC grid cell and the
snow course (positive means VIC higher), root mean square (rms) difference between simulated
and observed SWE, correlation, observed trend, and trend simulated by VIC. Data are given in
centimeters of SWE.
Fig. 2. Observed (o) and VIC simulated (+) 1 April SWE at nine of the snow courses in Wash-
ington for 1950–1997 (or shorter). Each frame includes statistics about the elevation of the
snow course (o1615m for Cayuse Pass), the difference in elevation between the VIC grid cell
and the snow course (positive means VIC higher), root mean square (rms) difference between
simulated and observed SWE, correlation, observed trend, and trend simulated by VIC. Data
are given in centimeters of SWE. 2098

HESSD
4, 2073–2110, 2007
Has spring snowpack
declined in the
Washington
Cascades?
P. Mote et al.
Title Page
Abstract Introduction
Conclusions References
Tables Figures
J I
J I
Back Close
Full Screen / Esc
Printer-friendly Version
Interactive Discussion
EGU
VIC swe
1910 1930 1950 1970 1990 2010
year
20
40
60
80
100
120
140
160
cm
1950-2003: -31%
1916-2003: -16%
VIC swe, fixed precipitation
1910 1930 1950 1970 1990 2010
year
20
40
60
80
100
120
140
cm
1950-2003: -20%
1916-2003: -13%
Fig. 3. 1 April SWE simulated by VIC, averaged over the domain of the Cascades and Olympics
for the full simulation (top) and for a temperature-only simulation (bottom), with linear fits and
loess smoothing for the time periods indicated.
2099

HESSD
4, 2073–2110, 2007
Has spring snowpack
declined in the
Washington
Cascades?
P. Mote et al.
Title Page
Abstract Introduction
Conclusions References
Tables Figures
J I
J I
Back Close
Full Screen / Esc
Printer-friendly Version
Interactive Discussion
EGU
Washington snow courses
1925 1935 1945 1955 1965
0
10
20
30
40
50
60
Number
1000
1250
1500
1750
2000
Elevation, meters
0 50 100 150 200 250 300
long-term mean SWE, cm
0
500
1000
1500
2000
Elevation, m
Fig. 4. Elevation, and hence temperature sensitivity, of the full set of snow courses for the
Washington Cascades and Olympics changes over time. Early snow courses substantially un-
dersampled middle and low elevations. (a) Number of snow courses reporting 1 April data
(dotted line) for each year; for each starting year, number of snow courses at least 80% com-
plete over the period of record (thick solid line), and their mean elevation (long dashed line,
right-hand axis). Note how the early snow courses were few and were at higher elevation. (b)
Scatterplot of mean 1 April SWE against elevation of all snow courses (open circles) and of
snow courses available in 1940 (solid circles). The mean value for each set is indicated by the
large circles; mean SWE changes by 26 cm and the mean elevation by 290 m.
2100

HESSD
4, 2073–2110, 2007
Has spring snowpack
declined in the
Washington
Cascades?
P. Mote et al.
Title Page
Abstract Introduction
Conclusions References
Tables Figures
J I
J I
Back Close
Full Screen / Esc
Printer-friendly Version
Interactive Discussion
EGU
Trends since 1940 (1930, +)
-120 -80 -40 0 40
Percent change
500
1000
1500
2000
Elevation, m
Trends since 1950
-120 -80 -40 0 40
Percent change
500
1000
1500
2000
Elevation, m
Trends since 1960
-120 -80 -40 0 40
Percent change
500
1000
1500
2000
Elevation, m
Trends since 1970
-120 -80 -40 0 40
Percent change
500
1000
1500
2000
Elevation, m
Fig. 5. Dependence of linear trends in 1 April SWE on elevation, for different starting years
through the end of the record (plus symbols in the first frame indicate trends from 1930). The
pronounced and consistent dependence of trends on elevation is an indication of the role of
warming trends (see Mote; 2003, 2006; Mote et al.; 2005; Hamlet et al., 2005; Regonda et al.,
2005).
2101

HESSD
4, 2073–2110, 2007
Has spring snowpack
declined in the
Washington
Cascades?
P. Mote et al.
Title Page
Abstract Introduction
Conclusions References
Tables Figures
J I
J I
Back Close
Full Screen / Esc
Printer-friendly Version
Interactive Discussion
EGU
Cascades SWE
1925 1940 1955 1970 1985 2000 2015
0
50
100
150
200
250
cm
Cascades SWE
1925 1940 1955 1970 1985 2000 2015
0
50
100
150
200
cm
Fig. 6. Twenty-six different time series of regionally averaged SWE for the Washington Cas-
cades are constructed by including all snow courses available by the date indicated. The bot-
tom panel shows the data, smoothed (see appendix for details). Thick dashed curve shows
VIC values from Fig. 3.
2102

HESSD
4, 2073–2110, 2007
Has spring snowpack
declined in the
Washington
Cascades?
P. Mote et al.
Title Page
Abstract Introduction
Conclusions References
Tables Figures
J I
J I
Back Close
Full Screen / Esc
Printer-friendly Version
Interactive Discussion
EGU
Correlations, Apr 1 SWE and NDJFM T and P
1930 1940 1950 1960 1970
Starting year
0.20
0.30
0.40
0.50
0.60
0.70
0.80
0.90
inches
- corr(SWE,T)
234567
Temperature, oC
0
50
100
150
SWE, cm
40 60 80 100 120 140 160
Precipitation, cm
0
50
100
150
SWE, cm
Fig. 7. Top panel: correlation between regionally averaged 1 April SWE and November–March
(top) precipitation and (bottom, negative) temperature for each starting year through 2006 (i.e.,
the 26 time series shown in Fig. 6 in which new snow courses are added through 1960; after
that the time window simply shrinks). Note that, consistent with Figs. 4–5 and the sensitivity
of SWE at lower elevations to temperature, the correlation of regionally averaged SWE with
precipitation drops somewhat and the correlation with temperature changes substantially from
–0.3 to about –0.5. Bottom two panels show scatterplots of 1 April SWE vs climate division 4
temperature (left) and precipitation (right), for the time series starting in 1944.
2103

HESSD
4, 2073–2110, 2007
Has spring snowpack
declined in the
Washington
Cascades?
P. Mote et al.
Title Page
Abstract Introduction
Conclusions References
Tables Figures
J I
J I
Back Close
Full Screen / Esc
Printer-friendly Version
Interactive Discussion
EGU
SWE, observed and reconstructed from T and P
1940 1950 1960 1970 1980 1990 2000 2010
20
40
60
80
100
120
140
160
obs:-24%
reg:-13%
Fig. 8. For the 1944–2006 time series of 1 April SWE, the observed time series (trend –24%,
–25 cm) and fitted time series S(t) (trend –13%, or –12 cm; diamond symbols).
2104

HESSD
4, 2073–2110, 2007
Has spring snowpack
declined in the
Washington
Cascades?
P. Mote et al.
Title Page
Abstract Introduction
Conclusions References
Tables Figures
J I
J I
Back Close
Full Screen / Esc
Printer-friendly Version
Interactive Discussion
EGU
Figure 9. Time series of (top) aPP(t), left axis and precipitation in cm, right axis;
(bottom) aTT(t), left axis and temperature (right axis, note inverted scale). Linear trends
for each quantity are both graphed and printed, and the loess curve is shown as well.
37
SWE and precipitation
1940 1950 1960 1970 1980 1990 2000 2010
20
40
60
80
100
120
140
SWE, cm
50
75
100
125
150
Precip, cm
P: 6cm, 5%
SWE: 5cm, 6%
SWE and temperature
1940 1950 1960 1970 1980 1990 2000 2010
40
60
80
100
120
SWE, cm
2
3
4
5
6
7
Temp, oC
T: 1.4oC
SWE: -17cm, -18%
Fig. 9. Time series of (top) aPP(t), left axis and precipitation in cm, right axis; (bottom) aTT(t),
left axis and temperature (right axis, note inverted scale). Linear trends for each quantity are
both graphed and printed, and the loess curve is shown as well.
2105

HESSD
4, 2073–2110, 2007
Has spring snowpack
declined in the
Washington
Cascades?
P. Mote et al.
Title Page
Abstract Introduction
Conclusions References
Tables Figures
J I
J I
Back Close
Full Screen / Esc
Printer-friendly Version
Interactive Discussion
EGU
SWE / NDJFM precip
1940 1950 1960 1970 1980 1990 2000 2010
year
0
20
40
60
80
%
-28%
-33%
Fig. 10. “Storage efficiency”, the ratio of 1 April SWE to November–March precipitation, with
linear fit, for (top) observations 1944–2006 and (bottom) VIC simulation 1944–2003. The aver-
age storage efficiency in VIC is lower because the domain includes a lot of low-elevation grid
points, but the percentage changes are similar.
2106

HESSD
4, 2073–2110, 2007
Has spring snowpack
declined in the
Washington
Cascades?
P. Mote et al.
Title Page
Abstract Introduction
Conclusions References
Tables Figures
J I
J I
Back Close
Full Screen / Esc
Printer-friendly Version
Interactive Discussion
EGU
Cascades SWE
1940 1950 1960 1970 1980 1990 2000 2010
0
20
40
60
80
inches
Low elev / high elev
1940 1950 1960 1970 1980 1990 2000 2010
0.0
0.2
0.4
0.6
0.8
Fig. 11. Ratio of 1 April SWE at low-elevation (<1200m) snow courses to SWE at high-elevation
(>1200 m) snow courses. With the linear fit, the ratio has declined from 0.42 to 0.33 from 1944
to 2006; note also that in 2005 and 1981 the low elevations had less than 15% as much SWE as
the high elevations. This metric is a good representation of the pattern expected from regional
warming. 2107

HESSD
4, 2073–2110, 2007
Has spring snowpack
declined in the
Washington
Cascades?
P. Mote et al.
Title Page
Abstract Introduction
Conclusions References
Tables Figures
J I
J I
Back Close
Full Screen / Esc
Printer-friendly Version
Interactive Discussion
EGU
Area-elevation for Washington Cascades
0 20 40 60 80 100
% of area
250
500
750
1000
1250
1500
1750
2000
Elevation, m
0 20 40 60 80 100
% of snow
250
500
750
1000
1250
1500
1750
2000
Elevation, m
Fig. 12. Cumulative area (left) and SWE (right) as a function of elevation in thousands of feet,
for the Washington Cascades. In the left panel the asterisks indicate the 60-m elevation bands
with at least one snow course. In the right panel the lines indicate the 50th and 90th percentiles.
2108

HESSD
4, 2073–2110, 2007
Has spring snowpack
declined in the
Washington
Cascades?
P. Mote et al.
Title Page
Abstract Introduction
Conclusions References
Tables Figures
J I
J I
Back Close
Full Screen / Esc
Printer-friendly Version
Interactive Discussion
EGU
Area-weighted Cascades SWE
1935 1950 1965 1980 1995 2010
0
50
100
150
cm
obs: -28%
VIC: -27%
Fig. 13. Regionally averaged 1 April SWE for observations (o) computed for the 1944–2006
snow courses using area-weighting and infilling of missing values with best-correlated time
series, and VIC (v). The VIC values have been scaled to the mean observed SWE. Linear fits
for observed (solid) and VIC (dashed) overlap.
2109

HESSD
4, 2073–2110, 2007
Has spring snowpack
declined in the
Washington
Cascades?
P. Mote et al.
Title Page
Abstract Introduction
Conclusions References
Tables Figures
J I
J I
Back Close
Full Screen / Esc
Printer-friendly Version
Interactive Discussion
EGU
Apr 1 SWE
1940 1950 1960 1970 1980 1990 2000 2010
20
40
60
80
100
120
140
160
cm
-18.6%
Apr 1 SWE minus NPI
1940 1950 1960 1970 1980 1990 2000 2010
50
60
70
80
90
100
110
120
cm
-14.6%
Fig. 14. Variability and trend in 1 April SWE (1944–2006 time series), top, and with NPI regres-
sion subtracted (bottom).
2110