1 JUNE 2011
LIN AND WU
2801
Contribution of the Autumn Tibetan Plateau Snow Cover to Seasonal
Prediction of North American Winter Temperature
HAI LIN AND ZHIWEI WU
Meteorological Research Division, Environment Canada, Dorval, Quebec, Canada
(Manuscript received 11 June 2010, in final form 14 December 2010)
ABSTRACT
Predicting surface air temperature (T ) is a major task of North American (NA) winter seasonal prediction.
It has been recognized that variations of the NA winter T ’s can be associated with El Niño–Southern Oscillation (ENSO) and the North Atlantic Oscillation (NAO). This study presents observed evidence that
variability in snow cover over the Tibetan Plateau (TP) and its adjacent areas in prior autumn (September–
November) is significantly correlated with the first principal component (PC1) of the NA winter T ’s, which
features a meridional seesaw pattern over the NA continent. The autumn TP snow cover anomaly can persist
into the following winter through a positive feedback between snow cover and the atmosphere. A positive TP
snow cover anomaly may induce a negative sea level pressure and geopotential height anomaly over the eastern
North Pacific, a positive geopotential height anomaly over Canada, and a negative anomaly over the southeastern United States—a structure very similar to the positive phase of the Pacific–North America (PNA)
pattern. This pattern usually favors the occurrence of a warm–north, cold–south winter over the NA continent.
When a negative snow cover anomaly occurs, the situation tends to be opposite. Since the autumn TP snow
cover shows a weak correlation with ENSO, it provides a new predictability source for NA winter T ’s.
Based on the above results, an empirical model is constructed to predict PC1 using a combination of autumn
TP snow cover and other sea surface temperature anomalies related to ENSO and the NAO. Hindcasts and
real forecasts are performed for the 1972–2003 and 2004–09 periods, respectively. Both show a promising
prediction skill. As far as PC1 is concerned, the empirical model hindcast performs better than the ensemble
mean of four dynamical models from the Canadian Meteorological Centre. Particularly, the real forecast of
the empirical model exhibits a better performance in predicting the extreme phases of PC1—that is, the
extremely warm winter over Canada in 2009/10—should the model include the autumn TP snow cover impacts. Since all these predictors can be readily monitored in real time, this empirical model provides a realtime forecast tool for NA winter climate.
1. Introduction
Large-scale seasonal anomalies of near-surface air
temperature (T ) in North American (NA) winters have
a significant impact on societal and economical activities.
In the past several decades, both extreme cold and warm
events have become more severe and frequent. For instance, the 2009/10 winter was noteworthy for its anomalously below-normal T’s across much of the United States,
whereas in Canada, it was the warmest and shortest winter
within the past several decades. A useful seasonal forecast
of T’s is thus of great value.
Corresponding author address: Dr. Hai Lin, MRD, ASTD, Environment Canada, 2121 Trans-Canada Highway, Dorval QC H9P
1J3, Canada.
E-mail: [email protected]
DOI: 10.1175/2010JCLI3889.1
It is known that the potential seasonal predictability
of the atmosphere mostly arises from coupled mechanisms that involve low boundary forcing (e.g., Charney
and Shukla 1981), because the atmosphere, on its own,
lacks the mechanisms to generate predictable variations
beyond synoptic time scales (Lorenz 1963). El Niño–
Southern Oscillation (ENSO) is a primary predictability
source for interannual variations of NA winter T ’s (e.g.,
Ropelewski and Halpert 1986; Hurrell 1996; Shabbar
and Khandekar 1996). The influence of ENSO extends
to North America through atmospheric teleconnections
related to tropical diabatic forcing (e.g., Horel and
Wallace 1981). On the other hand, as one of the most
important patterns of atmospheric variability, the North
Atlantic Oscillation (NAO) accounts for 31% of the
hemispheric T ’s interannual variance over the past 60
winters (Hurrell 1996). The NAO is a large-scale seesaw
2802
JOURNAL OF CLIMATE
in atmospheric mass between the subtropical high and
the polar low (e.g., Wallace and Gutzler 1981; Barnston
and Livezey 1987; Hurrell 1995; Li and Wang 2003; Wu
et al. 2009). Coherent with the NAO, the North Atlantic
tripole sea surface temperature anomaly (SSTA) might
be another potential predictability source for NA winter
T ’s (e.g., Rodwell et al. 1999).
As ENSO and the NAO account for only a portion of
climate variability across the globe (Ting et al. 1996;
Hurrell 1996; Wu and Li 2008), other lower boundary
forcing mechanisms need to be investigated to further
improve seasonal predictions (Wu et al. 2011). Of these,
one of the most important is likely to be snow cover, as
an anomaly in snow cover is associated with changes in
solar radiation absorption and surface heat fluxes.
The effect of Eurasian snow cover on global climate
has long been noticed (e.g., Gong et al. 2002, 2003, 2004;
Fletcher et al. 2009; Sobolowski et al. 2010). A possible
relation between the variability of the Indian summer
monsoon and Eurasian snow cover was reported (Bamzai
and Shukla 1999). The snow cover anomalies over the
Tibetan Plateau (TP) were found to have a close connection with Asian climate, especially the East Asian
summer climate (e.g., Wu and Kirtman 2007). The winter
Arctic Oscillation (AO) was reported to be modulated by
the variability of Eurasian snow cover through troposphere–
stratosphere interactions (Gong et al. 2007). On the other
hand, Clark and Serreze (2000) suggested that the impact
of snow cover anomalies over East Asia (1058–1508E)
may be limited to short-to-medium-range weather forecasting applications, as they observed a lack of persistence in snow cover anomalies.
The Tibetan Plateau represents the highest elevated
land on earth. The snow cover over the TP undergoes
a strong annual cycle and interannual variability, which
leads to anomalies in atmospheric diabatic heating in
the middle troposphere. In this paper, we attempt to
investigate whether and how the TP snow cover anomaly influences the NA winter climate, and to what extent it contributes to the seasonal prediction of NA
winter T ’s.
Section 2 introduces the datasets, model, and methodology used in this study. Section 3 identifies the leading
modes of NA winter T’s using an empirical orthogonal
function (EOF) analysis. The first principal component
(PC1) of the first EOF (EOF1) is found to be significantly
correlated with the interannual variability of the autumn
TP snow cover anomaly. Section 4 presents the large-scale
three-dimensional circulation features associated with the
three leading modes. In section 5 the impact of autumn TP
snow cover anomaly on the Northern Hemisphere circulation and NA temperature is discussed. In section 6, an
empirical model is constructed to predict the PC1 of NA
VOLUME 24
winter T’s based on the autumn TP snow cover and other
predictors related to ENSO and the NAO. A seasonal
hindcast is performed for the 1972–2003 period, and the
result is compared with the multimodel ensemble (MME)
forecast of four dynamical models from the Canadian
Meteorological Centre. A real forecast experiment is conducted for the 2004–09 period. The last section summarizes our major findings and discusses some outstanding
issues.
2. Data, model, and methodology
The main datasets employed in this study include 1)
the 40-yr European Centre for Medium-Range Weather
Forecasts (ECMWF) reanalysis (ERA-40; Uppala et al.
2005) data and the National Centers for Environmental
Prediction–National Center for Atmospheric Research
(NCEP–NCAR) Global Reanalysis 1 (NCEP-1; Kalnay
et al. 1996) data; 2) the Met Office Hadley Centre’s sea
surface temperature (SST) datasets gridded at 1.08 3 1.08
resolution (Rayner et al. 2003); and 3) monthly Northern
Hemisphere snow cover data (in unit of percent) gridded
at 2.08 3 2.08 resolution, calculated with weekly snow
cover data and programs from the National Oceanic and
Atmospheric Administration’s (NOAA) Climate Prediction Center (www.cpc.ncep.noaa.gov/data/snow/).
To get a longer time length covering the period from
December 1957 through February 2010, the ERA-40
and NCEP-1 data are combined. Since the NCEP-1 data
may have systematic errors in the period before 1980
(Wu et al. 2005), we use the ERA-40 data for the period
1957–2001 and extend the data from December 2002 to
February 2010 using NCEP-1 data (Wang et al. 2010). To
maintain temporal homogeneity, the 2002–10 NCEP-1
data were adjusted by removing the climatological difference between the ERA-40 and NCEP-1 data. In this
study, autumn refers to September–November (SON), and
winter is defined as December–February (DJF).
The output of the MME hindcast conducted with four
dynamical models from the Canadian Meteorological
Centre under the second phase of the Historical Forecasting Project (HFP2) is used in this study. The four
models are the second- and third-generation atmospheric GCMs (GCM2 and GCM3, respectively) of the
Canadian Centre for Climate Modelling and Analysis
(CCCma) (Boer et al. 1984; McFarlane et al. 1992),
a reduced-resolution version of the global Spectral à
Éléments Finis (SEF) model of the Recherche en prévision numérique (RPN) (Ritchie 1991), and the Global
Environmental Multiscale-Climate Version (GEM-CLIM)
model of the RPN (Côté et al. 1998). The MME hindcast
covers 24 winters from 1969/70 through 2002/03. An ensemble of 10 parallel integrations of 4-month duration
1 JUNE 2011
LIN AND WU
2803
FIG. 1. (a) Spatial pattern, (d) corresponding PC, and (g) spectrum of PC of EOF1 mode of the winter (DJF) mean 2-m T over North
America (108–708N, 1508–408W). (b),(e),(h) and (c),(f),(i) As in (a),(d), and (g), but for the second and the third modes. The numbers in
the brackets indicate fractional variance of the EOF modes. The red circles indicate the extremely anomalous 2009/10 winters. The red line
in (d) represents the prior autumn snow cover anomaly averaged over Tibetan Plateau and the adjacent areas.
was conducted using each model starting from the beginning of each month. The initial atmospheric conditions
were at 12-h intervals preceding the start of the forecasts,
taken from the NCEP–NCAR reanalysis. Global SST was
predicted using the persistence of the anomaly of the
preceding month; that is, the SST anomaly from the previous month was added to the climate of the forecast
period. Sea ice (ICE) extents were initialized with the
analysis and relaxed to climatology over the first 15 days
of integration. The experiment setup of the hindcast was
described in Lin et al. (2008) and Jia et al. (2010). In this
study the MME hindcast for winter seasons starting from
1 December is used to compare with the statistical model,
as introduced in section 6.
To identify the dominant modes, we perform an EOF
analysis with winter (DJF) T ’s over the NA domain
(108–708N, 1508–408W). The EOF analysis is carried out
by constructing an area-weighted covariance matrix.
3. Distinct leading modes of NA winter T ’s
variability
To better understand the relationship between the
Tibetan Plateau snow cover and NA winter T ’s, first we
derive the leading modes of NA winter T ’s. Figure 1
presents the spatial patterns and the corresponding PCs
of the three leading modes that jointly account for 63%
of the total variances of the interannual variability of
NA wintertime seasonal mean T ’s. The first mode explains around 30% of the total variance and is statistically distinguished from the rest of the eigenvectors in
terms of the criterion defined by North et al. (1982).
The most prominent characteristic of the EOF1 mode is
a meridional dipole structure with anomaly centers of
opposite sign over the northern and southern NA continent (Fig. 1a), at approximately 608 and 308N, respectively. A positive (negative) PC1 winter is characterized
2804
JOURNAL OF CLIMATE
VOLUME 24
(TPSCI) (red curve). A remarkable feature of PC1 is that
it experiences a strongly coherent interannual variability
with TPSCI. The correlation coefficient between them
reaches 0.57, exceeding the 99% confidence level based
on the Student’s t test. It is noteworthy that the extremely
anomalous 2009/10 winter had its T’s strongly projected to
a positive EOF1, with PC1 reaching the highest value
within the past 53 yr (shown by a red circle in Fig. 1d).
Meanwhile, the autumn TP snow cover anomaly also
reached its highest value since 1972. This, along with the
apparent coherence between TPSCI and PC1, indicates
that the autumn TP snow cover may be linked to the origin of the EOF1 mode. We will come back to this point in
section 5.
The second mode shows a monosign pattern over most
of NA, except for the northeastern flank of the continent,
with the maximum loading located in the central NA
continent and Davis Strait (Fig. 1b). A positive (negative)
phase of EOF2 corresponds to a cold (warm) NA continent. The spectrum of PC2 exhibits significant periodicities of 3–4 and 7 yr (Fig. 1h). It is interesting that the
EOF2 mode (not shown fully) resembles an NAO-like
anomaly pattern over the North Atlantic, namely, a tripole T’s anomaly pattern (e.g., Wu et al. 2009). The third
mode features a zonal dipole pattern with an opposite sign
between the eastern and western NA continent (Fig. 1c).
A strong (or weak) third mode corresponds to a cold–west,
warm–east (warm–west, cold–east) T distribution. PC3
basically bears three significant spectrum peaks: 2, 3.3,
and 25 yr (Fig. 1i).
The distinct spatial–temporal structures of the three
leading modes imply they may have different physical origins and predictability sources. In the next section, we will
analyze the large-scale circulation anomalies and predictability sources associated with the three leading modes.
4. Dynamic structures and predictability sources of
the leading modes
a. Planetary-scale circulation anomalies
FIG. 2. SLP (contours in units of hPa), 2-m T ’s (color shadings,
units of 8C), and 925-hPa winds (vectors, units of m s21) for (a)
climatological winter mean, and anomalies regressed with reference to (b) PC1, (c) PC2, and (d) PC3.
by a warm–north, cold–south (cold–north, warm–south) T
distribution over the NA continent. PC1 is dominated by
interannual variability, as is evidenced in Fig. 1d. Its power
spectrum peaks at a periodicity around 3.3 years (Fig. 1g).
Embedded in Fig. 1d is the autumn snow cover anomaly
averaged over the TP and the adjacent region (258–508N,
908–1058E), which is defined as a TP snow cover index
Figure 2 shows the anomalous surface circulation regressed to three PCs along with the climatology. For
EOF1, there are two negative sea level pressure (SLP)
anomaly centers (dashed contours in Fig. 2b). The SLP
anomaly distribution is reminiscent of a positive phase of
the Pacific–North America (PNA) pattern. Indeed, the
correlation coefficient between PC1 and the DJF PNA
index reaches 0.72, well passing the 99% confidence level.
The PNA index is obtained online (www.cpc.ncep.noaa.
gov); the index was calculated based on the rotated principal component analysis of Barnston and Livezey (1987).
A large area of negative SLP anomalies is located over
the northeastern Pacific, with a major trough extending
1 JUNE 2011
LIN AND WU
FIG. 3. A 500-hPa geopotential height H (contours in units of
10 gpm) and winds (vectors, units of m s21) for (a) climatological
winter mean, and anomalies regressed with reference to (b) PC1,
(c) PC2, and (d) PC3.
2805
southeastward along the west coast of NA. Compared
with the climatology (Fig. 2a), this pattern reflects a deeperthan-normal and eastward-shifted Aleutian low pressure
system. The northerly surface wind anomalies prevail in
the central North Pacific (NP), accompanied by increased
advection of cold air from the north, which decreases SST
over the central North Pacific. The southerly surface
winds along the west coast of the NA continent transport
from the south warmer and moister air into this region,
leading to increased T’s in Alaska and the west coast of
NA. The other negative SLP center, which is relatively
minor, is located in the midwestern Atlantic, which implies weakening of the subtropical high in this region.
Cold anomalies are found in the southeastern United
States, extending eastward toward the midwestern Atlantic. The northerly surface winds over the eastern and
southeastern United States transport cooler and drier air
toward the Gulf of Mexico and the western Atlantic,
which decreases SST over that region.
A pronounced feature of EOF2 is the NAO-like structure in SLP and 925-hPa winds over the North Atlantic
(Fig. 2c). The correlation coefficient between PC2 and the
DJF NAO index is 20.76, well exceeding the 99% confidence level. A strong positive SLP center associated with
anticyclonic surface wind dominates the Greenland and
Iceland region, whereas a negative SLP center associated
with cyclonic surface wind appears over the midlatitude
North Atlantic. This pattern is similar to the negative
phase of the NAO, with a weak subtropical high and
a weak Icelandic low. A warm anomaly center is located
over Greenland. The NA continent is basically controlled
by positive SLP and cold T anomalies. The SLP and T
anomaly centers tilt southeastward from western Canada
to the southeastern United States. The pattern accompanying EOF3 is a wave train with two positive SLP centers
over the northeastern Pacific and the eastern NA continent
and a negative SLP center over the western NA continent.
Figure 3 compares the anomalous large-scale circulations at midtroposphere that are associated with the three
leading modes along with the climatology. For the first
mode, there is a negative height anomaly center over the
North Pacific, a positive center in central Canada, and
a negative center over the southeastern United States and
southwestern Atlantic region—a distribution similar to
the PNA pattern. The strong low-pressure anomaly center
is located over the northeastern Pacific above the cold-air
mass, which is consistent with the hypsometric equation
(Fig. 3b). Because the climatological ridge over the northeastern Pacific and the west coast of the NA continent
tilts southeastward from Alaska to the Rocky Mountains
(Fig. 3a), the anomalous high center over central Canada
implies an eastward shift of the high ridge toward the
central continent, which decreases cold-air activities over
2806
JOURNAL OF CLIMATE
TABLE 1. Correlation coefficients between PCs and some major
autumn potential predictability sources. The bold italic correlation
coefficients exceed the 95% confidence level based on the Student’s t test, and the bold italic values in the combined skill column
represent the best performance.
PCs
Niño-3.4 index
NAO index
TP snow
cover index
PC1
PC2
PC3
0.42
20.12
0.19
20.07
20.2
20.16
0.57
0.34
20.1
the NA continental region north of 408N and increases T’s
in the mid–high latitudes. Another low-pressure anomaly
center is seen over the midwestern North Atlantic. Since
the climatological NA trough tilts southwestward along the
east coast of the NA continent, the anomalous low over
the midwestern Atlantic indicates a southward extension
of the NA trough, which favors more cold-air activities
over the eastern United States and the midwestern North
Atlantic. The 500-hPa circulation anomalies have centers
that in general match those in the lower troposphere
(Fig. 2b), indicating a quasi-barotropic vertical structure
of the circulation patterns. The 500-hPa circulation patterns associated with the second and the third modes also
resemble those at the surface. For the second mode,
pronounced NAO-like anomalies prevail over the North
Atlantic sector with a meridional seesaw in geopotential
height (H) between the high and middle latitudes. For the
third mode, a notable wave train propagates from the
northeastern Pacific to the eastern NA continent.
Overall, the circulation anomalies associated with the
three leading modes may be viewed as departures from
the climatological norm. The first mode tends to shift the
Aleutian low eastward (Fig. 2). In the midtroposphere,
the first mode represents an eastward shift of the high
ridge over the western NA continent with a positive
anomaly center over central Canada, and a southward
extension of the NA trough with negative anomalies
over the southeastern United States and the western
Atlantic (Fig. 3). EOF1 is closely related to the PNA
pattern. The second mode primarily exhibits NAO-like
circulation anomalies, whereas the third mode bears a
west–east-oriented wave train pattern. Thus, the three
leading temperature modes are closely related to largescale circulation variability.
VOLUME 24
autumn Niño-3.4 index, NAO index, and the TP snow
cover index (Table 1). PC1 has a significant correlation
not only with the Niño-3.4 SST but also with the TP snow
cover anomalies in SON. Interestingly, the correlation
with the TP snow cover is even higher than with the Niño3.4 index. The second mode also shows a link with the
autumn TP snow cover, whereas the third mode does not
exhibit any notable connection with the above major potential predictability sources. Thus, the first mode is likely
to be the most predictable mode, and this study will focus
on seasonal prediction of the first mode.
An important feature that has not been noticed before
is that the TP snow cover might provide a new predictability source for seasonal mean NA winter T’s besides ENSO. The SON TP snow cover has only a weak
correlation with the SON Niño-3.4 SST, with a correlation coefficient of 0.22 for the 1972–2008 period. Clark
and Serreze (2000) found that the East Asian snow cover
may significantly affect the circulation over the downstream North Pacific on the synoptic time scale. However,
according to their lag correlation statistics, the East Asian
snow cover anomalies cannot persist beyond about a week.
Therefore, they proposed that the East Asian snow cover
was limited for short-to-medium-range weather forecasting
applications other than for problems on longer time scales.
In the present work, we found that the autumn snow cover
anomaly over the TP has a good persistence, and that it is
closely connected to the first two leading modes of the NA
surface air temperature—the first mode in particular. We
will discuss how the autumn TP snow cover influences the
NA winter T’s in the next section.
Although the NAO is correlated with the simultaneous NA winter T ’s (see section 1; Hurrell 1996), the
autumn NAO does not show any considerable impact on
the NA winter T ’s because of a lack of persistence (see
Table 1). Thus, the autumn NAO, by itself, may not
serve as a good predictor for the NA winter T ’s. The
NAO in fall, however, may extend its influences through
some kind of lower boundary forcing, such as the tripole
SST anomalies in the North Atlantic (e.g., Wu et al.
2009). Indeed, the autumn tripole SST anomalies in the
North Atlantic do show a strong correlation with the first
leading mode in the following winter (see section 6).
Thus, the autumn tripole SST related to the NAO is also
used as a predictor in this study.
b. Predictability sources for the three leading modes
In this paper, we select the autumn ENSO, NAO, and
the TP snow cover as three major potential predictability
sources for wintertime NA temperature. To examine how
the three leading modes of NA winter T’s are related to
these predictability sources, we calculate the correlation
coefficients between the three leading PCs and the prior
5. Impacts of autumn TP snow cover
The first mode of the NA winter T’s has a significant
positive correlation with the preceding autumn snow cover
anomalies over the TP and the adjacent areas (Fig. 4);
12.4% of the entire figure area exceeds the 95% significance level. A Monte Carlo test is performed to test
1 JUNE 2011
LIN AND WU
2807
FIG. 5. The lead–lag correlation coefficients between PC1 and
the TPSCI from JJA(0) to FMA(1). The dotted line represents the
95% confidence level based on a Student’s t test. The vertical line
indicates D(0)JF(1) where the simultaneous correlations between
PC1 and TPSCI are shown.
FIG. 4. Correlation coefficient map between PC1 and prior autumn snow cover over the TP and adjacent areas. The interval of
contours is 0.1. The shaded regions represent correlation coefficients exceeding the 95% confidence level. The snow cover
anomalies averaged in the black box is defined as a Tibetan Plateau
snow cover index. The altitudes of areas within bold curves are
higher than 1500 m.
the field significance of Fig. 4 (Livezey and Chen 1983).
To do so, the PC1 time series is replaced with a series of
38 numbers randomly selected from a normal distribution that has the same autocorrelation as PC1. This experiment is repeated 1000 times. The case of a greater
than 12.4% area with correlations statistically significant
at the 95% level happens only 5 times. Therefore, it is
highly unlikely (at the 99% level) that the results in Fig.
4 were a chance occurrence. An excessive TP snow cover
in autumn is likely an antecedent anomalous boundary
condition for the occurrence of a high PC1 winter over
the NA continent. Likewise, a reduced TP snow cover in
autumn tends to lead a low PC1 winter in NA.
To describe the interannual variability of TP snow
cover, a TPSCI is defined with the snow cover anomalies
averaged over TP and the adjacent region (258–508N, 908–
1058E) (red box in Fig. 4). To see whether the snow cover
anomaly over this elevated land area has a persistent
linkage with the first leading mode, Fig. 5 shows the lead–
lag correlations between PC1 and the TPSCI time series
from June to August [JJA(0)] to February to April
[FMA(1)]. Here ‘‘0’’ denotes the simultaneous year and
‘‘1’’ denotes the following year. JJA(0) leads the simultaneous winter December–February [D(0)JF(1)] by
6 months. The correlation coefficients between TPSCI
and PC1 become significant from August to October
[ASO(0)] and persist throughout the following winter
D(0)JF(1). Thus, the TP snow cover has a strong and
persistent connection with the NA T ’s leading mode,
which makes it of seasonal prediction value.
The autocorrelation of TPSCI between SON and DJF is
0.46, which passes the 99% confidence level. How can the
TP snow cover anomalies sustain from autumn through
winter? To shed light on its possible mechanism, we take
the autumn TPSCI as a reference and compute the lead–
lag correlation with 200-hPa geopotential height fields.
The results are shown in Fig. 6. The absolute values of
the correlation coefficients continuously increase from a
1-month lead through to a 3-month lag. In the 1-month
lead (Fig. 6a), before the TP snow cover becomes excessive (reduced), negative (positive) H anomalies emerge
over the TP. It indicates that the atmospheric anomalies are responsible for the underneath TP snow cover
anomalies, which lead to excessive (reduced) snow cover
anomalies. From a 0-month lead through to a 3-month
lag (Figs. 6b–e), low-pressure anomalies are enhanced
(weakened) over the TP and the adjacent areas, which
indicates that the TP snow cover anomalies have a significant positive feedback to the atmosphere. Thus, the persistence of the autumn snow cover anomalies is likely due
to a positive feedback between the atmospheric circulation
and the snow cover underneath. The possible physical
mechanisms can be interpreted as follows. An excessive
(reduced) snow cover usually would favor low (high) H
anomalies because of reduced (increased) solar radiation
absorption at the surface and suppressed (enhanced) sensible heat and radiative fluxes from the ground. On the
other hand, the low (high) H anomalies and cold (warm)
upper-tropospheric temperature anomalies induced by
the snow cover anomaly would produce increased (decreased) upper-tropospheric potential vorticity (Hoskins
et al. 1985; Shaman and Tziperman 2005). Anomalous
surface convergence (divergence) as well as anomalous
vertical motions would follow that would lead to increased
2808
JOURNAL OF CLIMATE
VOLUME 24
FIG. 6. The lead–lag correlation patterns between 200-hPa H (H200) and the autumn TPSCI. The TPSCI leads the H200 by (a) 21, (b) 0,
(c) 1, (d) 2, and (e) 3 months. The dark (light) shaded areas denote significant positive (negative) correlation exceeding the 95% confidence level. Note that 21, 0, 1, 2, and 3 months correspond to ASO, SON, October–December (OND), November–January (NDJ), and
DJF, respectively.
(decreased) cloudiness and increased (decreased) precipitation in the region. This would maintain the excessive
(reduced) snow cover anomalies. Such positive feedback
may sustain such an air–snow cover anomaly pattern. In
addition, the atmospheric anomalies tend to propagate
eastward from a 0-month lead to a 3-month lag (Figs. 6b–e).
In a 3-month lag, a large area of negative correlation is
formed in the northeast Asian and northwestern Pacific
region near the westerly jet. This pattern is reminiscent of
the remote response to an anomalous TP snow cover as
revealed in Wang et al. (2008), who completed a numerical
experiment with the ECHAM4 model and found that in
summer, the TP warming can induce two distinct Rossby
wave trains: one in the tropics that propagates along the
low-level monsoon westerly and the other in the extratropics along the upper-level westerly jet.
To understand the impact of autumn TP snow cover
anomalies on the downstream large-scale circulation and
the NA T’s in winter, Fig. 7 presents regressions of DJF
atmospheric variables with respect to SON TPSCI. For the
500-hPa geopotential height field (Fig. 7a), negative anomaly centers are found over the eastern North Pacific and the
southeastern United States, with positive height anomalies
over Canada. This PNA-like pattern is very similar to that
associated with PC1 (Fig. 3b), confirming that the autumn
TP snow cover anomaly leads to an atmospheric circulation anomaly that is associated with the leading mode of
NA T’s. In the North Atlantic, the 500-hPa geopotential
height anomaly pattern appears to have a negative NAO
structure. It is well known that the tropical diabatic heating
anomaly of an ENSO event in the equatorial Pacific can
excite an extratropical Rossby wave train that takes the
form of the PNA (e.g., Wallace and Gutzler 1981). How
a snow cover anomaly around TP also generates a PNAlike circulation anomaly is of great interest. One hypothesis is that the PNA is a preferred mode of atmospheric
variability that can be generated by an appropriate external forcing that can be either a tropical diabatic
heating anomaly of an ENSO event or an extratropical diabatic heating anomaly of TP snow cover. Another
1 JUNE 2011
2809
LIN AND WU
FIG. 8. Correlation between autumn TPSCI and DJF synoptic
eddy activities, which is defined as the variance of daily meridional
wind perturbations at 300 hPa every winter season. The dark
(light) shaded areas denote significant positive (negative) correlation exceeding the 95% confidence level.
FIG. 7. (a),(b) As in Figs. 3b and 2b, respectively, but for regressed
to the autumn TPSCI.
Lin and Derome 1997). This southward shift of the westerly jet modulates the DJF synoptic eddy activities over
the North Pacific (Fig. 8). Here the synoptic eddy activity is
defined as the variance of bandpass-filtered daily mean
meridional winds at 300 hPa, calculated for every winter
season, where a Lanczos filter with 201 weights is used to
retain a period of 2–8 days. Following a positive autumn
TP snow cover anomaly, there is a southward shift of
synoptic-scale eddy activity over the eastern North Pacific
in the following winter.
6. Seasonal prediction of NA winter T ’s
possible mechanism is the barotropic instability of the twodimensional winter mean flow as proposed in Simmons
et al. (1983), where a disturbance induced by an external
forcing can grow by extracting kinetic energy from the
mean flow in the eastern North Pacific and North American
region. A contribution from the feedback of synopticscale transient eddies can also help to maintain the PNAlike circulation anomaly (e.g., Held et al. 1989; Lin and
Derome 1997).
The regressions of DJF SLP, surface wind, and T’s with
respect to autumn TPSCI are shown in Fig. 7b. The SLP
anomaly has a similar distribution as a 500-hPa geopotential height, reflecting an equivalent barotropic vertical structure of the PNA-like circulation pattern. Over
NA T ’s display a north–warm, south–cold dipole with
two warm centers over Alaska and northeastern Canada.
In general, the distribution of surface anomalies is similar
to that related to PC1 (Fig. 2b), indicating that the main
feature of EOF1 is related to a TP snow cover anomaly.
Compared to Fig. 2b, Fig. 7b has a weaker warm anomaly
over central Canada. It is possible that over central
Canada, the PC1-related T anomaly is also contributed
by other processes, for example, an interaction between
local snow cover and the atmosphere.
The strong eastern North Pacific negative geopotential
anomaly center associated with the autumn TP snow cover
(Fig. 7a) will shift the local westerly jet southward, in a way
similar to that which accompanies a positive PNA (e.g.,
It has been generally accepted that ENSO is perhaps the
most important predictability source for seasonal prediction of the global climate, including NA T’s (Ropelewski
and Halpert 1986). Besides ENSO, the NAO may be another potential predictability source (e.g., Hurrell 1996;
Wu et al. 2009). The result in this study suggests that the
autumn TP snow cover could provide an additional physically based predictor for NA winter T’s.
Here we focus on the seasonal prediction of PC1, which
represents the variability of the dominant mode of NA
T’s. To assess the contribution of the autumn TP snow
cover to the seasonal prediction skill of PC1, an empirical
seasonal prediction model is developed using a linear
regression method for the period of 1972–2009, which is
expressed as
y 5 a0 1 a1 x1 1 a2 x2 1 a3 x3 1 a4 x4 ,
(1)
where x1 denotes the autumn TPSCI defined above, and
x2, x3, and x4 refer to the autumn normalized SSTAs
averaged in tropical Indian Ocean (TIO; 208S–208N,
508–858E), Pacific Ocean [(108S–108N, 1558E–808W) plus
(508–608N, 1708E–1408W)], and North Atlantic Ocean
[(08–308N, 708–208W) plus (558–658N, 658–208W) minus
(258–358N, 958–708W)], respectively, that are selected
according to correlations between the global autumn
SST and PC1 (boxes in Fig. 9). The SSTA in the Pacific
can be considered as an ENSO-related predictor, whereas
2810
JOURNAL OF CLIMATE
FIG. 9. Correlation between SON SSTA and DJF PC1. SSTAs in
the black boxes represent other predictors for PC1 besides the
autumn TP snow cover anomalies.
that in the North Atlantic can be regarded as the NAOrelated predictor, namely, a so-called tripole SST anomaly pattern, which was documented in previous studies
(e.g., Rodwell et al. 1999; Wu et al. 2009). There is evidence from previous studies that the Indian Ocean SSTA
may provide useful information for long-term Northern
Hemisphere extratropical circulation variability (e.g.,
Hoerling et al. 2004). Therefore, although it is not
independent of the Niño-3.4 index (their correlation coefficient being 0.54), we consider it not negligible, and
it is used as a complement to the other predictors. All
these predictors show a significant correlation with PC1
(Fig. 10). The correlation coefficient with the autumn TP
snow cover is calculated for the 1972–2009 period because of snow cover data limitation, while the rest are
calculated for the 1957–2009 period. If we use the same
period of 1972–2009, then the correlation coefficients
between PC1 and x2, x3, and x4 are 0.39, 0.36, and 0.56,
respectively. Thus, the relative importance of the predictors to the overall model is ordered by the TP snow cover
(0.57), the North Atlantic tripole SST anomaly (0.56), the
Indian Ocean SSTA, and the Pacific SSTA in terms of the
partial correlations in parentheses.
To test the predictive capability of the above empirical model, the cross-validation method is performed to
hindcast PC1 for the 1972–2003 period (Wu et al. 2009).
To warrant the robustness of the hindcast result, we
choose a leaving-eight-out strategy. The procedure is as
following: the cross-validation method systematically
deletes 8 yr from the period 1972–2003 (namely, years
left out in four blocks of 1972–79, 1980–87, 1988–95, and
1996–2003), derives a forecast model from the remaining years, and tests it on the deleted cases. Note that the
choice of ‘‘leaving eight out’’ is not random. Randomly
choosing 20%–30% of the data to be in the test data and
the remainder as the training set can prevent overfitting
or data wasting (e.g., Wu et al. 2009). For PC1, 25% of
the whole hindcast period (32 yr) equals 8 yr.
The cross-validated estimates of PC1 are shown in
Fig. 11. The correlation coefficient between the observations (black line in Fig. 11) and the cross-validated
estimate of the empirical model (red line in Fig. 11)
VOLUME 24
FIG. 10. Time series of PC1 and its autumn SST predictors in the
North Pacific (NP), tropical Indian Ocean (TIO,) and North Atlantic
(NA). The correlation coefficients are indicated in the brackets.
reaches 0.55 and the statistical significance test P value
equals 0.0012, exceeding the 99.9% confidence level. If
we established the prediction model with the TP snow
cover excluded and did the same hindcast (green line
in Fig. 11), then the correlation coefficient between
the observations and the 32-yr cross-validated estimate
would drop to 0.43, and the P value increases to 0.0135.
This indicates that the autumn TP snow cover does
significantly improve the seasonal prediction skill of the
empirical model. For comparison, we also presented the
observations and the MME hindcast (1969–2002) of
the four dynamic models (GEM-CLIM, GCM2, GCM3,
and SEF) from the Canadian Meteorological Centre. The
MME hindcast of DJF seasonal mean 2-m air temperature initialized from 1 December of each year is projected
onto the observed first mode. The hindcast of the new
empirical model and that of the MME (blue line in Fig. 11)
are both in general agreement with the observations. The
correlation coefficient between the observations and the
34-yr MME hindcast is 0.32. The root-mean-square errors
for the empirical model and the MME are 0.59 and 0.70.,
respectively. Therefore, the empirical model has a better
prediction skill than the MME as far as PC1 is concerned.
To examine the real forecast skill of the empirical
model, we use it to predict PC1 for 2004–09. The prediction equation is established based on training data of
FIG. 11. Comparison of the hindcast PC1 made by the empirical
model (with and without TP snow cover) and by the MME mean of
the four dynamical models (GEM-CLIM, GCM2, GCM3, and SEF).
1 JUNE 2011
LIN AND WU
2811
FIG. 12. The first leading mode of North American winter T’s and the corresponding PC1 derived from the
four-model ensemble hindcast for the period 1969–2002.
all the years prior to 2004. It is worth mentioning that the
extremely abnormal NA 2009/10 winter is realistically
forecast (shown by green circle in Fig. 11). Since all these
predictors can be readily monitored in real time, this
empirical model provides a real-time forecast tool and
may operationally facilitate the seasonal prediction of
the NA winter climate.
7. Conclusions and discussion
The seasonal prediction of anomalous NA winter T ’s
is of great importance. Many previous studies have
shown that NA winter T variations are closely related
to ENSO and the NAO (e.g., Ropelewski and Halpert
1986; Hurrell 1996). This paper, for the first time, presents
observed evidence that the autumn TP snow cover anomalies can make a significant impact on the abnormal winter T ’s over North America. A positive TP snow cover
anomaly may induce a pattern similar to the positive
phase of the PNA. This usually favors the occurrence of
warm–north, cold–south winter over the NA continent.
When a negative TP snow cover anomaly occurs, the
situation tends to be opposite.
Since the autumn TP snow cover shows only a weak
linkage with ENSO and the tropical Indian Ocean SSTA
(their correlation coefficient being 0.2), it may serve as
a new predictability source for the dominant mode of NA
winter T ’s. Based on the above results, an empirical
model is established to predict PC1 of the NA winter T’s
using a combination of the ENSO-related predictors, the
NAO-related predictors, and the TP snow cover anomalies in prior autumn. A hindcast experiment is performed for the 1972–2003 period, which shows a better
prediction skill than the MME hindcast of the four
dynamical models. Since all these predictors can be
readily monitored in real time, this empirical model
provides a real-time forecast tool and may facilitate the
seasonal prediction of the NA winter climate.
In this study we focused on the seasonal prediction of
the leading mode of NA T ’s (EOF1), which explains
about 30% of the seasonal mean variance of surface
temperature variability over NA. As the EOF was done
over a large area covering the whole continent and some
adjacent water, the percentage of variance explained
refers to an average of the whole analysis area. For local
regions, such as the anomaly centers over Canada and
the southeastern United States, the variance explained
by this mode would be much larger. The contribution of
TP snow cover on NA temperature seasonal prediction
is also evidenced in the skill distribution of T ’s forecast
at each grid point. In Fig. 1d, it appears that there is an
upward trend in the time series of PC1 and TPSCI. The
trend, however, is not significant, as can be seen from the
spectrum analysis (Fig. 1g), which shows that PC1 is
dominated by interannual variability. To test how much
of the skill on the interannual time scale could be contributed by this trend, the hindcast experiment with the
empirical model is repeated after removing the linear
trend. The skill turns out to be very similar, and there is
no change to the conclusion.
Our results indicate that the contribution of the preceding TP snow cover anomalies should be correctly
taken into account in a dynamical model. In general, the
MME hindcast of four dynamical models used here has
a realistic performance in capturing the first mode of NA
winter T ’s (Fig. 12). The meridional seesaw pattern over
the NA continent is well reproduced, and the correlation
coefficient between observed and hindcast PC1 is 0.41,
2812
JOURNAL OF CLIMATE
exceeding the 95% confidence level. However, our empirical model performs even better than the MME of the
dynamical models. One possible reason is that the TP
snow cover influence may not have been adequately
represented in the models. Our results thus suggest a future direction for the dynamical model improvement.
The empirical seasonal prediction model in this study is
based on the seasonal mean time scale, and it is assumed
that PC1 is stable. However, many climate extreme
events occur on a subseasonal time scale. Does a similar
relationship between the TP snow cover and NA T’s exist
that can be used to predict the subseasonal events? If not,
what are the sources of predictability for the subseasonal
extreme events? If the leading mode changes with time,
then the predictors and relevant mechanisms may also
change, correspondingly. These are still open questions
for future investigations. In addition, the hypotheses
concerning the origins of the first leading mode call for
further numerical and theoretical studies, which will be
discussed in the future.
Acknowledgments. We thank Dr. Ruping Mo for his
helpful comments on an earlier version of this manuscript.
Zhiwei Wu is supported by the Sustainable Agriculture
Environment Systems (SAGES) research initiative of
Agriculture and Agri-Food Canada through the Natural
Sciences and Engineering Research Council of Canada
(NSERC) fellowship program.
REFERENCES
Bamzai, A. S., and J. Shukla, 1999: Relation between Eurasian
snow cover, snow depth, and the Indian summer monsoon: An
observational study. J. Climate, 12, 3117–3132.
Barnston, A. G., and R. E. Livezey, 1987: Classification, seasonality
and persistence of low-frequency atmospheric circulation
patterns. Mon. Wea. Rev., 115, 1083–1126.
Boer, G. J., N. A. McFarlane, R. Laprise, J. D. Henderson, and
J.-P. Blanchet, 1984: The Canadian Climate Centre spectral atmospheric general circulation model. Atmos.–Ocean, 22,
397–429.
Charney, J. G., and J. Shukla, 1981: Predictability of monsoons. Monsoon Dynamics, J. Lighthill and R. P. Pearce, Eds., Cambridge
University Press, 99–109.
Clark, M. P., and M. C. Serreze, 2000: Effects of variations in East
Asian snow cover on modulating atmospheric circulation over
the North Pacific Ocean. J. Climate, 13, 3700–3710.
Côté, J., S. Gravel, A. Méthot, A. Patoine, M. Roch, and
A. Staniforth, 1998: The operational CMC–MRB Global Environmental Multiscale (GEM) model. Part I: Design considerations and formulation. Mon. Wea. Rev., 126, 1373–1395.
Fletcher, C. G., S. C. Hardiman, P. J. Kushner, and J. Cohen, 2009:
The dynamical response to snow cover perturbations in a large
ensemble of atmospheric GCM integrations. J. Climate, 22,
1208–1222.
Gong, G., D. Entekhabi, and J. Cohen, 2002: A large-ensemble
model study of the wintertime AO–NAO and the role of interannual snow perturbations. J. Climate, 15, 3488–3499.
VOLUME 24
——, ——, and ——, 2003: Relative impacts of Siberian and North
American snow anomalies on the winter Arctic Oscillation.
Geophys. Res. Lett., 30, 1848, doi:10.1029/2003GL017749.
——, ——, ——, and D. Robinson, 2004: Sensitivity of atmospheric
response to modeled snow anomaly characteristics. J. Geophys. Res., 109, D06107, doi:10.1029/2003JD004160.
——, J. Cohen, D. Entekhabi, and Y. Ge, 2007: Hemispheric-scale
climate response to northern Eurasia land surface characteristics and snow anomalies. Global Planet. Change, 56, 359–370.
Held, I. M., S. W. Lyons, and S. Nigam, 1989: Transients and the
extratropical response to El Niño. J. Atmos. Sci., 46, 163–174.
Hoerling, M. P., J. W. Hurrell, T. Xu, G. T. Bates, and A. Phillips,
2004: Twentieth century North Atlantic climate change. Part
II: Understanding the effect of Indian Ocean warming. Climate Dyn., 23, 391–405.
Horel, J. D., and J. M. Wallace, 1981: Planetary-scale atmospheric
phenomena associated with the Southern Oscillation. Mon.
Wea. Rev., 109, 813–829.
Hoskins, B. J., M. E. McIntyre, and A. W. Robertson, 1985: On the
use and significance of isentropic potential vorticity maps.
Quart. J. Roy. Meteor. Soc., 111, 877–946.
Hurrell, J. W., 1995: Decadal trends in the North Atlantic Oscillation: Regional temperature and precipitation. Science, 269,
676–679.
——, 1996: Influence of variations in extratropical wintertime
teleconnections on Northern Hemisphere temperature. Geophys. Res. Lett., 23, 665–668.
Jia, X., H. Lin, and J. Derome, 2010: Improving seasonal forecast
skill of North American surface air temperature in fall using
a postprocessing method. Mon. Wea. Rev., 138, 1843–1857.
Kalnay, E., and Coauthors, 1996: The NCEP/NCAR 40-Year Reanalysis Project. Bull. Amer. Meteor. Soc., 77, 437–471.
Li, J. P., and J. X. L. Wang, 2003: A modified zonal index and its
physical sense. Geophys. Res. Lett., 30, 1632, doi:10.1029/
2003GL017441.
Lin, H., and J. Derome, 1997: On the modification of the high- and
low-frequency eddies associated with the PNA anomaly: An
observational study. Tellus, 49A, 87–99.
——, G. Brunet, and J. Derome, 2008: Seasonal forecasts of Canadian winter precipitation by postprocessing GCM integrations. Mon. Wea. Rev., 136, 769–783.
Livezey, R., and W. Y. Chen, 1983: Statistical field significance and
its determination by Monte Carlo techniques. Mon. Wea. Rev.,
111, 46–59.
Lorenz, E. N., 1963: Deterministic nonperiodic flow. J. Atmos. Sci.,
20, 130–141.
McFarlane, N. A., G. J. Boer, J.-P. Blanchet, and M. Lazare, 1992:
The Canadian Climate Centre second-generation general circulation model and its equilibrium climate. J. Climate, 5, 1013–1044.
North, G. R., T. L. Bell, R. F. Cahalan, and F. J. Moeng, 1982:
Sampling errors in the estimation of empirical orthogonal
functions. Mon. Wea. Rev., 110, 699–706.
Rayner, N. A., D. E. Parker, E. B. Horton, C. K. Folland,
L. V. Alexander, D. P. Rowell, E. C. Kent, and A. Kaplan,
2003: Global analyses of sea surface temperature, sea ice, and
night marine air temperature since the late nineteenth century. J. Geophys. Res., 108, 4407, doi:10.1029/2002JD002670.
Ritchie, H., 1991: Application of the semi-Lagrangian method to
a multilevel spectral primitive-equation model. Quart. J. Roy.
Meteor. Soc., 117, 91–106.
Rodwell, M. J., D. P. Rowell, and C. K. Folland, 1999: Oceanic
forcing of the wintertime North Atlantic Oscillation and European climate. Nature, 398, 320–323.
1 JUNE 2011
LIN AND WU
Ropelewski, C. F., and M. S. Halpert, 1986: North American precipitation and temperature patterns associated with the El Niño/
Southern Oscillation (ENSO). Mon. Wea. Rev., 114, 2352–2362.
Shabbar, A., and M. Khandekar, 1996: The impact of El Niño–
Southern Oscillation on the temperature field over Canada.
Atmos.–Ocean, 34, 401–416.
Shaman, J., and E. Tziperman, 2005: The effect of ENSO on Tibetan Plateau snow depth: A stationary wave teleconnection
mechanism and implications for the South Asian monsoons.
J. Climate, 18, 2067–2079.
Simmons, A. J., J. M. Wallace, and G. W. Branstator, 1983: Barotropic wave propagation and instability, and atmospheric teleconnection patterns. J. Atmos. Sci., 40, 1363–1392.
Sobolowski, S., G. Gong, and M. Ting, 2010: Modeled climate state
and dynamic responses to anomalous North American snow
cover. J. Climate, 23, 785–799.
Ting, M. F., M. P. Hoerling, T. Y. Xu, and A. Kumar, 1996: Northern
Hemisphere teleconnection patterns during extreme phases of
the zonal-mean circulation. J. Climate, 9, 2614–2633.
Uppala, S. M., and Coauthors, 2005: The ERA-40 Re-Analysis. Quart.
J. Roy. Meteor. Soc., 131, 2961–3012, doi:10.1256/qj.04.176.
Wallace, J. M., and D. S. Gutzler, 1981: Teleconnections in the
geopotential height field during the Northern Hemisphere
winter. Mon. Wea. Rev., 109, 784–812.
2813
Wang, B., Q. Bao, B. Hoskins, G. Wu, and Y. Liu, 2008: Tibetan
Plateau warming and precipitation change in East Asia. Geophys. Res. Lett., 35, L14702, doi:10.1029/2008GL034330.
——, Z. Wu, J. Liu, C.-P. Chang, J. Li, and T.-J. Zhou, 2010: Another
look at climate variations of the East Asian winter monsoon:
Northern and southern modes. J. Climate, 23, 1495–1512.
Wu, R., and B. P. Kirtman, 2007: Observed relationship of spring
and summer East Asian rainfall with winter and spring Eurasian snow. J. Climate, 20, 1285–1304.
——, J. L. Kinter III, and B. P. Kirtman, 2005: Discrepancy of interdecadal changes in the Asian region among the NCEP–NCAR
reanalysis, objective analyses, and observations. J. Climate, 18,
3048–3067.
Wu, Z., and J. Li, 2008: Prediction of the Asian–Australian monsoon
interannual variations with the grid-point atmospheric model of
IAP LASG (GAMIL). Adv. Atmos. Sci., 25, 387–394.
——, B. Wang, J. Li, and F.-F. Jin, 2009: An empirical seasonal
prediction model of the East Asian summer monsoon using
ENSO and NAO. J. Geophys. Res., 114, D18120, doi:10.1029/
2009JD011733.
——, J. Li, Z. Jiang, and J. He, 2011: Predictable climate dynamics
of abnormal East Asian winter monsoon: Once-in-a-century
snowstorms in 2007/2008 winter. Climate Dyn., doi:10.1007/
s00382-010-0938-4, in press.