全 文 ：Received 20 Oct. 2003 Accepted 26 Nov. 2003
Supported by the State Key Basic Research and Development Plan of China (G2000018604) and the Huo Ying Dong Education Foundation
(81014).
http://www.chineseplantscience.com
Inter-annual Changes in Eurasian Continent NDVI and Its Sensitivity to
the Large-scale Climate Variations in the Last 20 Years
GONG Dao-Yi, SHI Pei-Jun
(Key Laboratory of Environmental Change and Natural Disaster, Institute of Resources Science,
Beijing Normal University, Beijing 100875, China)
Abstract: The influence of climate change on the terrestrial vegetation health (condition) is one of the
most significant problems of global change study. The vegetation activity plays a key role in the global
carbon cycle. The authors investigated the relationship of the advanced very high resolution radiometer-
normalized difference vegetation index (AVHRR-NDVI) with the large-scale climate variations on the inter-
annual time scale during the period 1982-2000 for the growing seasons (April to October). A singular value
decomposition analysis was applied to the NDVI and surface air temperature data in the time-domain to
detect the most predominant modes coupling them. The first paired-modes explain 60.9%, 39.5% and 24.6%
of the squared covariance between NDVI and temperature in spring (April and May), summer (June and
August), and autumn (September to October), respectively, which implies that there is the highest NDVI
sensitivity to temperature in spring and the lowest in autumn. The spatial centers, as revealed by the
maximum or minimum vector values corresponding to the leading singular values, indicate the high
sensitive regions. Only considering the mode 1, the sensitive center for spring is located in western
Siberia and the neighbor eastern Europe with a sensitivity of about 0.308 0 NDVI/℃. For summer, there
are no predominantly sensitive centers, and on average for the relatively high center over 100º-120º E by
45º-60º N, the sensitivity is 0.248 0 NDVI/℃. For autumn, the center is located over the high latitudes of
eastern Asia (110º-140º E, 55º-65º N), and the sensitivity is 0.087 5 NDVI/℃. The coherent patters as
revealed by the singular decomposition analysis remain the same when coarser resolution NDVI data were
used, suggesting a robust and stable climate/vegetation relationship.
Key words: normalized difference vegetation index (NDVI); climate change; sensitivity; large-scale
Climate plays an important role in driving ecosystems
to change on both local and global scales. Global climate
change and its relationship with the ecological influence
and consequences, which involves the carbon (C) cycle
are among the most significant problems on the Earth
(Houghton et al., 2001; Walther et al., 2002). A numerous
researches, based on the ecological classification such as
the Holdridge life zone, studied the quantitative relation-
ship between the specific vegetation types and climate pa-
rameters such as temperature and precipitation (Li and Shi,
1999; Pan et al., 2003). On the other hand, many works
focused on the dynamic relation of vegetation and the cli-
matic driving factors by investigating the long-term varia-
tions in both vegetation condition and climate change (Fu
and Wen, 1999; Los et al., 2001; Kawabata et al., 2001;
Gong and Shi, 2003). The objective of the present study is,
by employing the long-term data, to investigate the large-
scale response of vegetation over the Eurasian continent
to the inter-annual time-scale fluctuations in climate during
the last two decades. As a variety of papers indicated that
the inter-annual variations in vegetation condition of most
northern forest are primarily influenced by temperature
(Zuzuki et al., 2000; 2001; Zhu et al., 2001), here we con-
sider only monthly temperature as the climate factor. In the
present study, what we are interested in includes whether
there are large-scale variations in the Eurasian continent
vegetation; if so, whether these variations are linked to the
similar spatial scale temperature fluctuation and, how and
where the strongest coupling between vegetation and tem-
perature exists, and to what extent they are coupled.
1 Data Preparation
Photosynthetic activity of vegetation can be inferred
using such satellite-derived vegetation index as the nor-
malized difference vegetation index (NDVI). In the present
study we utilized the widely employed advanced very high
resolution radiometer (AVHRR) Land Pathfinder produc-
tion of monthly maximum value composite. This data sets
are available through internet at the web site of http://
eosdata.gsfc.nasa.gov/. The original data are, in order to
Acta Botanica Sinica
植 物 学 报 2004, 46 (2): 186-193
GONG Dao-Yi et al.: Inter-annual Changes in Eurasian Continent NDVI and Its Sensitivity to the Large-scale Climate Variations
in the Last 20 Years 187
save space and facilitate transfer, archived in the form of
unsigned integer with the physical value range from 0 to
253. Zero means missing data. One denotes water, and two
indicates the blank region due to the map projection. Only
the values above three would be analyzed. The real geo-
physical values are obtained by rescaling the data by mi-
nus 128 and then multiplying by 0.008. Most of the real
geophysical NDVI vary between -0.2 and +0.7 (Agbu and
James, 1994).
Though some errors of the Pathfinder production have
still been found to exist due to incomplete atmospheric cor-
rection resulting from lack of appropriate water vapour and
aerosol data, the quality of this data set was significantly
improved by correcting for various factors such as intra-
sensor degradation (James and Kalluri, 1994). Changes in
satellites might induce inhomogeneities. The errors, if exist,
should be most manifest in the long-term trends. We com-
pared the trend of Eurasian mean NDVI with the result from
other authors based on the different data sets, and found
that there are no evident differences. For example, the trend
of mean spring (April to May) Pathfinder NDVI over Eurasia
is 10.6%/10 a for the period 1982-2000, and this value is
almost identical to the results of Zhu et al. (2000) (20.87%/
18 a for period 1982-1999, i.e. 11.6%/10 a), though they
take only vegetated pixel of 8 km resolution into account.
Thus we think the Pathfinder data is appropriate to investi-
gate the large-scale inter-annual variations, and do not ad-
just the possible errors.
Since the precision of the AVHRR visible channels de-
grades rapidly in twilight areas, most of data in high lati-
tude where the Sun is close to the horizon in winter are
missing. The surface snow cover, which exists from the late
autumn through the early spring, deteriorates the quality
of NDVI data, too. Therefore, in the present research we
consider only the growing seasons (from April to October).
The average of April to May denotes spring, June to Au-
gust stands for summer, and September to October means
autumn. Due to the satellite non-operation, the data for
period September to December in 1994 are not available. To
facilitate computation, the gap for September and October
of 1994 is filled using the long-term averages at each grid,
i.e., the missing data for September 1994 are assigned as
the September mean averaged over the 20 months (1981-
2001). The missing data for October1994 are assigned as
the October mean averaged over 19 months (1982-2000).
Then all data are presented in the form of anomalies from
the 1982-2000 means for each month and at each grid. This
kind of anomaly removes the annual cycle, and retains the
inter-annual fluctuations.
A land surface air temperature data set on a 5º longitude
× 5º latitude grid was utilized in the present study. The
observed station temperature data in the form of anomalies
from the 1961-1990 base-period were collected and inter-
polated to a regular set of grid boxes, finally resulting in
this data set (Jones, 1994; Jones et al., 2001). This data set
is used world-widely, and adopted by Intergovernment
Panel on Climate Change (IPCC) reports as basic scientific
information to assess the global temperature change
(Houghton et al., 2001). We used both NDVI and tempera-
ture data only for period 1982-2000 and for the domain of
0º-155º E by 20º-75º N.
2 Methods
The high spatial and temporal variability in NDVI arises
from a lot of factors, mainly due to local environmental
variables. Compared to these local factors, the large-scale
climate signals are usually very weak and hard to discern.
In order to suppress the high frequency noises, statistical
methods such as principal component analysis and spa-
tial-temporal average are often applied to NDVI data (Kogan,
2000). However, the averaging would smooth out the high
frequency noise as well as the climate signals
simultaneously. The principal component analysis can
surely detect the spatial-temporal features embedded in the
data set. But the disadvantage is that the detected leading
modes may be non-climate induced or even non-vegeta-
tion related in some circumstances.
In contrast to principal component analysis, the singu-
lar value decomposition (SVD) analysis is concerned with
the linear relationships between two different variables. This
technique is particularly suited for resolving problems such
as the coupling features between NDVI and temperature.
This method is often utilized in detecting the degree of
coupling between two geophysical variables. In the present
study, we applied the SVD technique to the covariance
matrix of NDVI and temperature. For more detail about this
technique and its application please refer to Wallace et al.
(1992) and Bretherton et al. (1992), and references wherein.
3 Results
We put temperature and NDVI data into two matrices
respectively. The two matrices have the same temporal
domains (1982-2000, 19 years) but different spatial dimen-
sions (due to the different spatial resolution and data
availabilities). After multiplying them we obtained their co-
variance matrix. We applied SVD analysis to this covari-
ance matrix, and finally got the results, including the singu-
lar values, paired singular vectors (paried-modes), and the
Acta Botanica Sinica 植物学报 Vol.46 No.2 2004188
corresponding expansion coefficients (time series).
As the singular values drop off monotonically with the
mode number, the importance of the paired-modes also
decreases steadily. Since the sum of the squares of the
singular values is equal to the total squared covariance
between all the elements of the temperature and NDVI, each
singular value indicates the relative importance of the cor-
responding spatial modes. A higher singular value means
the more importance of the associated paired-modes. So,
the ratio of each squared singular value to the total squared
covariance provides a quantitative measurement to its
significance. Usually the first several modes can explain a
large portion of the squared covariance. The last modes
probably make no sense, just noise or random disturbance.
Below we show the SVD analysis results and their mean-
ings for spring, summer and autumn separately.
3.1 Spatial feature for spring (April and May)
Table 1 presents the squared percent covariance ex-
plained by the first five paired-modes. Obviously, the pro-
portion of leading paired-modes in spring is the highest
among all seasons, 60.9%, almost 2.5 times of the value for
autumn. This implies that, in general, there is stronger con-
nection between NDVI and temperature, and also means
there is higher vegetation sensitivity to temperature in
spring than in summer and autumn. The related centers as
shown by the singular vectors are key regions where the
two variables display the tightest association or cause-
consequence relationship. Therefore, these centers dem-
onstrate the stronger NDVI sensitivity to temperature than
other regions. Almost entire mid- to high-latitude Eurasia
continent shows the same sign of temperature change, i.e.,
this temperature mode presents a continental scale variation,
and the whole region would be dominated by either posi-
tive temperature anomalies or the negative anomalies. The
corresponding time series show that there are upward
trends, implying that the warming temperature is prevailing
there. It should be noted that the temperature change is far
from uniform distribution, and the most important region
(center) is located in Siberia. Interestingly, for the first paired-
modes, NDVI shows a similarly spatial feature, i.e., in asso-
ciation with the positive-prevailing temperature mode, posi-
tive NDVI anomalies dominate over most of Eurasian
continent. Furthermore, the location of the center in NDVI
is identical to that of temperature (Fig.1). The above in-
phase relationship clearly reveals that the higher spring
temperature gives rise to active vegetation growth, leading
to a greener condition, and vice versa. Many studies re-
ported that during the last two decades there experienced
the significant warming over the Eurasian continent in win-
ter and spring (Houghton et al., 2001). Thus the first paired-
modes might be related to the global temperature warming
and the vegetation consequences. However, it should be
noted, as IPCC2001 report indicated, that the strongest
warming trend in spring occurred in the middle and eastern
portion of mid- to high Asian continent during the last two
decades, which is not the same region as our first modes
demonstrated. This suggests that the area experiencing the
strongest temperature variability is not necessarily the same
place where NDVI response has the strongest sensitivity.
The results also imply that NDVI sensitivity to temperature
is selectively significant, and the temperature-NDVI cou-
pling is dependent on the regions.
In addition, the second modes explains 13.2% of the
total amount. The spatial features are shown in Fig.2. Com-
pared to modes 1, the second paired-modes are of regional
scale. Europe and eastern Asia have the same temperature
signs, and the opposite signs appear in Northwest Asia.
This kind of pattern shows the similarity with some inher-
ent climate modes, such as the Eurasian pattern (Wallace
and Gutzler, 1981). It is interesting to note that the NDVI
also demonstrate the similar spatial features again. Thus,
the second paired-modes might present the temperature
influence of large-scale atmospheric circulation fluctuation
and the vegetation responses.
3.2 Spatial feature for summer (June to August)
Figure 3 shows the first paired-modes for summer. The
outstanding feature is that there are no evident coupling
centers, neither for temperature nor for NDVI. Of course, it
is not necessary to mean that there are no variation centers
for temperature itself, or for NDVI itself. If we perform the
spatial-temporal oriented analysis such as principal com-
ponent analysis or empirical orthogonal function analysis
for summer temperature anomalies over the same region,
and the leading mode does show apparently anomalous
centers. The results from Fig.3 just mean that there is no
region showing the strong coupling between summer tem-
perature and vegetation condition, which implies that there
Table 1 The squared percent covariance of NDVI and tempera-
ture explained by the five leading paired-modes for spring (Apr.
and May), summer (Jun. to Aug.) and autumn (Sept. and Oct.),
respectively
Spring Summer Autumn
Paired-modes 1 60.9% 39.5% 24.6%
Paired-modes 2 13.2% 19.0% 22.2%
Paired-modes 3 10.6% 12.3% 15.6%
Paired-modes 4 5.1% 6.8% 9.3%
Paired-modes 5 2.7% 5.8% 5.9%
Sum 92.5% 83.4% 77.6%
GONG Dao-Yi et al.: Inter-annual Changes in Eurasian Continent NDVI and Its Sensitivity to the Large-scale Climate Variations
in the Last 20 Years 189
is no high correlation existing in summer, and that NDVI
shows less dependence on and less sensitivity to summer
temperature. It is reasonable since summer is the warmest
season; heat limitation is not an essential factor. The NDVI
in northern forest is usually saturated during summer. If
the temperature is higher than normal, it would favor the
biological activity to less extent; if the temperature is lower
than normal, it would not impact the vegetation too much.
This case may be true for most of Eurasia. The second
paired-modes for summer is a little similar to the second
modes for spring, i.e., something like the Eurasian pattern.
However, the magnitude is much weaker than that in spring
(the figure is not shown).
3.3 Spatial feature for autumn (September and October)
Figure 4 exhibits the first paired-modes for autumn. Both
temperature and NDVI centers are located in eastern Asian
Acta Botanica Sinica 植物学报 Vol.46 No.2 2004190
continent of north of 50º N. The feature also indicates an
in-phase relationship, i.e., higher NDVI being associated
with the warmer temperature and lower NDVI with the colder
temperature. But the magnitude is much weaker than in
spring, even weaker than in summer, which suggests that
the NDVI sensitivity to temperature in autumn is the lowest
in the whole growing season. It may be partly due to the
facts that most annual vegetation conditions are generally
determined by the growth in spring and summer. That, if
true, implies that the autumn NDVI is primarily connected
to the spring-summer NDVI rather than the autumn
temperature, and compared to that the climate anomaly is
somewhat unimportant. Of course, the strong cold surge,
which usually comes forth first in autumn, might cause an
early stop of vegetation growth and early beginning of
winter conditions. The cold surges often form in the inte-
rior Asian continent, particularly in the middle to eastern
portion. It is noted that the high relation regions also cover
the middle to eastern Asian continent. We might conclude
that the paired-modes reveal the influence of the extremely
cold airflow. In addition, the strong cold surges can move
southward along the East Asia to the South China Sea.
Figure 4b also confirms this feature, since the coastal re-
gions in eastern China and southern Japan show the high
values of singular value vectors. The spatial pattern in the
second paired-mode in autumn is a little similar to the first
one, the explained squared covariance is also close to the
first paired-modes, thus the second modes might not be
independent from the first one.
Comparison of the results in Table 1 indicates that much
explained squared-covariance in spring is concentrated in
the first several paired-modes. For summer and autumn, it
is much scattered, suggesting that there might be less fac-
tors exerting significant influence on vegetation, and that
among these factors the temperature is the most important
one. However, for summer and autumn it is not the case. To
better understand the NDVI fluctuations in summer and
autumn, in addition to temperature, more other factors
should be taken into account.
3.4 Sensitivity analysis
The previous sections reveal that there are some key
regions where temperature and NDVI are highly correlated.
In order to quantitatively estimate the connection between
them over there, we analyzed the sensitivity of NDVI in this
section. According to Figs.1, 3 and 4, we chose three differ-
ent centers for spring, summer and autumn, respectively.
The mean NDVI and temperature averaged over these re-
gions were carried out, and an ordinary linear regression
analysis was applied to the time series. Finally we gained
the quantitative relationship between temperature and
NDVI for each region. Since the data samples used to cal-
culate the means would affect the variance, only the means
based on same spatial grids can be compared with each
other. Keeping this point in mind, we picked out NDVI cen-
ters with the same pixel numbers for spring, summer and
autumn. For spring, the rectangle is 60º-90º E by 55º-65º
N. Since there is no outstanding center for summer, we
chose a relatively high center of domain of 100º-120º E,
45º-60º N. For autumn the region is 110º-140º E, 55º-65º
N. All three regions selected contain the same NDVI pixel
number. The results are shown as follows:
Spring: NDVI = –0.015 + 0.308 T, r = 0.89, significant
level > 99%;
Summer: NDVI = 0.0068 9 + 0.248 T, r = 0.52, significant
level > 95%;
Autumn: NDVI = 0.009 2 + 0.087 5 T, r = 0.25，not
significant.
where T is temperature, and r is correlation coefficients.
Obviously, among all seasons and all key regions the high-
est sensitivity of NDVI to temperature occurs in spring,
0.308 NDVI/℃. In summer the sensitivity is lower than
spring. Although in autumn there is still positive correla-
tion between NDVI and temperature, the correlation is not
significant (Fig.5).
4 Discussion
It should be mentioned that the above analysis con-
cerns only the strong spatial-temporal coupling feature in-
volved in temperature-NDVI variations and would not be
expected to explain all of interannual variance in NDVI. For
example, Europe is not the center in the first paired-modes
for spring, but, some researches reported that there are
early-coming growth seasons, enhanced biological activ-
ity in spring during the recent years (Black et al., 2000;
Zhou et al., 2001), thus leading to strong upward NDVI
trends over there. As Fig.6 shows, during the last about 20
years, in addition to Europe, most of the Asian continent
also display the tendency to become greener. Some other
studies also reported that (Zhou et al., 2001). Mean NDVI
averaged over 40º-70º N, 0º-155º E shows a trend in order
of 0.021 2 per 10 a for April to October. This value is statis-
tically significant at the 99% confidence level. In all grow-
ing seasons, spring shows the strongest trend, too. The
trend for the same domain but for April and May is 0.026 3
NDVI per 10 a, also significant above the 99% confidence
level, which means that the trend in spring is 24% stronger
than the average of the entire growing season. In addition
to the linear trends, the year-to-year changes in April and
GONG Dao-Yi et al.: Inter-annual Changes in Eurasian Continent NDVI and Its Sensitivity to the Large-scale Climate Variations
in the Last 20 Years 191
May also show the high correlation with the average con-
dition for April to October as a whole. They correlate at a
value of 0.9 for the period 1982-2000, implying that about
80% of the interannual NDVI variation in growing seasons
is associated with the vegetation condition in spring.
Therefore, we should pay more attention to spring when
considering the response of ecosystem to climate changes.
Our analyses do not show any temperature-NDVI cou-
pling center in the leading SVD modes over China for all
seasons. This may be due to, compared to the other Eur-
asian continent regions, the lower NDVI sensitivity to tem-
perature over there. However, in both spring and whole
growing season there is a significant positive trend over
the eastern China, particularly in the northern plains. Tem-
perature in eastern China is increasing during the last 20
years, indicating there is general in-phase relationship be-
tween temperature and NDVI. However, for such monsoon
driving ecosystems as in East Asia the monsoon precipita-
tion plays an essential role in the vegetation condition and
its anomaly (Fu and Wen, 1999). In northern China, where
arid, semi-arid and semi-humid environments occupy most
of portion of areas, there is a positive correlation between
NDVI and rainfall. But observations show both annual and
growing season precipitation had not increased in the last
20 years，and contrary to this, it decreased indeed. The
inconsistency between the decreasing precipitation, warm-
ing temperature and greening vegetation may partly arise
from the advancement of agriculture activity in China. These
regions showing strong NDVI trend in eastern China (mostly
in northern plains) are part of the main advanced agricul-
tural regions. Vegetation here would benefit from the mod-
ern irrigation systems, which could evidently reduce the
impacts of natural rainfall deficit on plants, and also benefit
from a warming temperature.
In the present study, we do not use all of the 1º× 1º
NDVI data sets in the SVD analysis since, if so, the covari-
ance matrix is too large to perform the SVD on the personal
computer due to the limitation of computer memory. Instead,
we use the re-sampled 2º×2º resolution NDVI. In order to
check whether the changes in NDVI resolutions would
change the results or not, we perform the same process
using the remained 2º×2º data and an even coarser reso-
lution data set re-sampled at 3º×3º grids. We find that all
these have the similar results. For example, using the other
2º×2º resolution data for spring the first three modes ex-
plain 60.6%，13.3% and 10.8% of the total squared
covariance; for 3º×3º resolution the explained proportion
is 61.0%，12.7% and 10.8%, almost the same. In addition
to the explained covariance, the spatial patterns are alike,
too, which implies that the NDVI-temperature relationship
Fig.5. Scatter maps showing the relationships between normalized difference vegetation index (NDVI) and temperature (T) anomalies
over the most sensitive regions for spring (a), summer (b) and autumn (c), respectively. Both NDVI and temperature are anomalies.
Fig.6. Linear trends of normalized difference vegetation index (NDVI). a. Whole growing seasons (April to October). b. spring (April
and May) in period 1981-2000. Only strong trends with values ≥ 0.02 and ≤-0.02 NDVI per decade are shown.
Acta Botanica Sinica 植物学报 Vol.46 No.2 2004192
revealed by SVD is not dependent on the data resolution,
and that the features of strong large-scale temperature sig-
nals and the NDVI responses presented in the present study
are robust and stable.
References:
Agbu P A, James M E. 1994. The NOAA/NASA Pathfinder
AVHRR Land Data Set user’s manual. Goddard Distributed
Active Archive Center, NASA, Goddard Space Flight Center.
Technical Report. 1-105.
Black T A, Chen W J, Barr A G, Arain M A, Chen Z, Nesic Z,
Hogg E H, Neumann H H. 2000. Increased carbon sequestra-
tion by a boreal deciduous forest in years with a warm spring.
Geophys Res Lett, 27:1271-1274.
Bretherton C S, Smith C, Wallace J M. 1992. An intercomparison
of methods for finding coupled patterns in climate data. J
Climate, 5:541-560.
Fu C B, Wen G. 1999. Variation of ecosystems over East Asia in
association with seasonal, interannual and decadal monsoon
climate variability. Clim Change, 43:477-494.
Gong D Y, Shi P J. 2003. Northern hemispheric NDVI variations
associated with the large-scale climate indices in spring. Int J
Remote Sens, 24:2559-2566
Houghton J T, Ding Y H, Griggs D J, Noguer M, van der Linden
P J, Dai X, Maskell K, Johson C A. 2001. Climate change
2001: the scientific basis. Contribution of Working Group I to
the Third Assessment Report of the Intergovernmental Panel
on Climate Change. Cambridge, UK: Cambridge University
Press. 881.
James M E, Kalluri S N V. 1994. The Pathfinder AVHRR land
data set: an improved coarse-resolution data set for terrestrial
monitoring. Int J Remote Sens, 15:3347-3364.
Jones P D, Osborn T J, Briffa K R, Folland C K, Horton E B,
Alexander L V, Parker D E, Rayner N A. 2001. Adjusting for
sampling density in grid box land and ocean surface tempera-
ture time series. J Geophy Res, 106:3371-3380.
Jones P D. 1994. Hemispheric surface air temperature variations:
a reanalysis and an update to 1993. J Climate, 7:1794-1802.
Kawabata A, Ichii K, Yamaguchi Y. 2001. Global monitoring of
interannnual changes in vegetation activities using NDVI and
its relationships to temperature and precipitation. Int J Re-
mote Sen, 22:1377-1382.
Kogan F N. 2000. Satellite-observed sensitivity of world land
ecosystems to El Nino/La Nina. Rem Sen Environ, 74:445-
462.
Li X-B, Shi P-J. 1999. Research on regulation of NDVI change of
Chinese primary vegetation type based on NOAA/AVHRR
data. Acta Bot Sin, 41:314-324. (in Chinese)
Los S O, Collatze G J, Bounoua L, Sellers P J, Tucker C J. 2001.
Global interannual variations in sea surface temperature and
land surface vegetation, air temperature, and precipitation. J
Climate, 14:1535-1549.
Pan Y Z, Li X B, Gong P, He C Y, Shi P J, Pu R L. 2003. An
integrative classification of vegetation in China based on
NOAA/AVHRR and vegetation-climate indices of the
Holdridge life zone. Int J Remote Sens, 24:1009-1027.
Suzuki R, Nomaki T, Yasunari T. 2001. Spatial distribution and
its seasonality of satellite-derived vegetation index (NDVI)
and climate over Siberia. Int J Climato, 21:1321-1335.
Suzuki R, Tanka S, Yasunari T. 2000. Relationships between
meridional profiles of satellite-derived vegetation index (NDVI)
and climate over Siberia. Int J Climato, 20:955-967.
Wallace J M, Gutzler D S. 1981. Teleconnections in the
geopotential height field during the Northern Hemisphere
winter. Mon Wea Rev, 109:784-812.
Wallace J M, Smith C, Bretherton C S. 1992. Singular value de-
composition of wintertime sea surface temperature and 500-
mb height anomalies. J Climate, 5:561-576.
Walther G R, Post E, Convey P, Menzil A, Parmesan C, Beebee
T, Fromentin J M, Hoegh-Guldberg O, Bairlein F. 2002. Eco-
logical responses to recent climate change. Nature, 416:389-
395.
Zhou L M, Tucker C J, Kaufmann R K, Slayback D, Shabanov N
V, Myneni R. 2001. Variations in northern vegetation activity
inferred from satellite data of vegetation index during 1981 to
1999. J Geophy Res, 106:20069-20083.
(Managing editor: HAN Ya-Qin)