Recent trends in changes of vegetation over East Asia coupled with temperature and rainfall variations

Authors


Abstract

[1] In this study, we investigated whether long-term normalized difference vegetation index (NDVI) data show climate change signals after the mid-1990s which are inferred from other studies on changing trends in precipitation and dust frequencies. In doing so, mean NDVI data for the growing seasons (April–October) from 1982 to 2006 were used for examining the spatiotemporal variations in the vegetation over East Asia, in conjunction with precipitation and temperature data. Results indicate that there was a prominent change in the trend of NDVI around the mid-1990s: a pronounced positive trend over most of the East Asian domain before the mid-1990s (1982–1996) and a reverse (or weakened) trend after the mid-1990s (1997–2006). The reverse trend is evident over the higher-latitude regions north of 50°N and the eastern Mongolian border area. The EOF and SVD analysis suggest that the dominant warming trend until the mid-1990s led to the increased NDVI over the high-latitude regions. However, after the mid-1990s, the reverse NDVI trend found primarily in the east of Lake Baikal and the arid and semiarid regions south of 50°N seems to be closely linked to local precipitation changes occurred abruptly in the mid-1990s. However, precipitation influences on the reverse NDVI changes are not clear over the high-latitude regions north of 50°N.

1. Introduction

[2] It has been reported that the East Asian region including China, Mongolia, and South Siberia have experienced a significant increase in surface air temperature, in response to the global warming trend [Cruz et al., 2007]. In line with such climate change, the precipitation pattern has also undergone a substantial change during the past several decades, showing more extreme climate phenomena during recent years [Fang et al., 2005; Su and Wang, 2007; Yao et al., 2008]. These climate fluctuations should exert profound impacts on vegetation activity and its variability, because the growth or decay of biomass is closely associated with meteorological parameters like temperature and/or precipitation [Nemani et al., 2002].

[3] A number of studies have quantified ecological responses to recent climate fluctuations at various spatial scales, from regional to global, using the satellite-based normalized difference vegetation index (NDVI), which has been available since the early 1980s [Myneni et al., 1998; Kawabata et al., 2001; Tucker et al., 2001; Zhou et al., 2003; Gong and Ho, 2003; Piao et al., 2004]. One of the common findings from these studies is that vegetation activity has increased over the two-decade period until the late 1990s in the northern middle and high latitudes of East Asia primarily due to the lengthened growing season (i.e., earlier green-up and delayed dormancy associated with warming). These studies also suggested that variations in NDVI over the forest area in the 40°–70°N latitude region are mainly associated with the temperature increase which took place from the early 1980s to the late 1990s. Consistent with the increased vegetation, the frequency of strong dust storm events in northern China was noticeably reduced during the period from the early 1980s to the mid-1990s [Qian et al., 2002; Zhao et al., 2004; Wang et al., 2008].

[4] However, Zou et al. [2005], with more extended data up to year 2003, reported that dryness has prevailed after the late 1990s in most of northern China (except the western part of northwest China, north of 35°N and west of 90°E). With deteriorated drought conditions after the late 1990s, the frequency of dust storms tended to increase after the late 1990s in northern China [Li and Zhai, 2003; Wang et al., 2005; Xu et al., 2006].

[5] Summarizing previous studies, there appear to be changes in the climate trend around the mid-1990s, which separate the last two or three decades into two periods, before and after the mid-1990s, in terms of the pattern of precipitation and/or dust storm frequency. If these changes are recognized, concomitant signals can be found in the long term NDVI data, because the ecosystem is directly influenced by changes in meteorological parameters such as precipitation, and because the vegetation changes can have a direct influence on dust emission.

[6] Thus, the natural questions to be asked are: (1) How do vegetation activities differ between the periods before and after the mid-1990s in East Asia? (2) What are the changes like in the regional climate conditions (e.g., precipitation or temperature) after the mid-1990s? (3) How are meteorological parameters linked with the vegetation changes in East Asia? To answer some of those questions, we examine the NDVI variations over East Asia including more recent years (here 1982–2006), in conjunction with precipitation and temperature data. We also investigate how the temperature and precipitation changes are coupled with vegetation activities throughout the analysis period. This study will help us understand how recent climate-driven vegetation changes have an impact on phenomena like the dust storm frequency, desertification, and surface radiative forcing over East Asia because they are closely related to surface vegetation changes.

2. Data Set Used and Analysis Methods

[7] Analysis is carried out on data over the study area of (35°–55°N, 90°–130°E), which covers arid and semi-arid regions of East Asia including northern China, Mongolia, and southern part of Siberia–see Figure 1 for the analysis domain. This area is interesting because it is an area susceptible to desertification due to natural or man-made climate changes and land-use changes. Examining climate variations and associated vegetation changes, we use NDVI, Global Precipitation Climatology Project (GPCP) precipitation, and surface temperature data for the 25-year period (1982–2006) from National Centers for Environmental Prediction and National Center for Atmospheric Research (NCEP-NCAR) reanalysis.

Figure 1.

Geographical distribution of the 25-year (1982–2006) mean growing season NDVI over East Asia.

[8] NDVI estimates are from AVHRR measurements of NOAA satellites (NOAA 7, 9, 11, 14, 16 and 17). The NDVI is defined as (NIR − VIS)/(NIR + VIS), where VIS and NIR are reflected solar radiation over visible and near infrared spectra, respectively, and NDVI products have been widely used as an indicator of ecosystem productivity [Tucker, 1979; Myneni et al., 1995; Herrmann et al., 2005]. Bimonthly NDVI data with an 8 km spatial resolution for the 25-year (1982–2006) period are obtained from the Global Inventory Monitoring and Modeling Studies (GIMMS) of the National Aeronautics and Space Administration (NASA). The GIMMS NDVI has been corrected to remove some non-vegetation effects such as sensor degradation, intersatellite difference and volcanic aerosol effects-detailed corrections are given by Kaufmann et al. [2000], Zhou et al. [2001], and Tucker et al. [2005]. In this study, monthly mean NDVI data are produced from 15-day interval data by applying a maximum value decomposition method, which selects the highest NDVI for each pixel during a given month, and then data are binned into a 0.1° × 0.1° grid.

[9] Monthly mean precipitation and near surface air temperature data are also employed to derive their coupled relationships with the vegetation variability over East Asia. As a proxy for the rainfall, we use the GPCP precipitation estimates in a 2.5° grid resolution [Adler et al., 2003]. The monthly near surface air temperature data in a 2.5° grid format are obtained from the NCEP-NCAR reanalysis [Kalnay et al., 1996].

[10] In order to investigate the spatially coherent patterns of temporal variations of the vegetation and climate variables over East Asia, the Empirical Orthogonal Function (EOF) analysis is conducted for the NDVI, precipitation and temperature data. The EOF analysis is considered to be an effective way of analyzing temporally varying spatial patterns of any field of interest over a specified area [Bjornsson and Venegas, 1997]. Then, in order to isolate linearly coupled modes between near surface climate variables (e.g., precipitation and temperature) and vegetation activity over East Asia, the Singular Value Decomposition (SVD) method is employed, which can identify well-coupled modes of spatiotemporal variability between two geophysical variables in various spatial scales [Bretherton et al., 1992; Wallace et al., 1992].

3. Results

[11] In this study, we focus on the growing season, which is defined as a 7 month period from April to October [e.g., Fang et al., 2005; Piao et al., 2006]. Throughout the paper, we present results based on data analysis for the growing season.

3.1. The 25-Year Mean NDVI Climatology

[12] We first examine the geographical distribution of 25-year (1982–2006) mean NDVI in the analysis domain–Figure 1. The spatial distributions of the NDVI indicate a variety of biomass over this East Asian analysis domain. The Gobi desert and the nearby area show NDVI values smaller than 0.1. Surrounding vegetated areas show NDVI up to 0.7. NDVIs are shown in the range of 0.2–0.5 over Mongolia and Inner Mongolia region, 0.3–0.7 in northeast China east of 120°E, and 0.5–0.7 in the high latitude regions north of 50°N. Since the vegetation areas are generally represented by NDVI values larger than 0.1 [Zhou et al., 2001], we mainly focus on the area showing mean NDVI larger than 0.1 on a monthly time scale.

[13] In addition to the 25-year mean NDVI distribution, it is also of interest to examine whether meaningful NDVI trends are found during the 25-year period in response to the global warming trend. The overall 25-year mean trend of the growing-season NDVI is derived from its time series at each grid point. Here, anomaly trends are used and their geographical distributions and magnitudes are given in Figure 2.

Figure 2.

Geographical distribution of the 25-year mean trend of growing season mean NDVI. The mean trend given in percent is obtained using the NDVI departure from 25-year mean value at the given point. Contour interval is 0.1, and zero line is omitted. Solid and dashed lines represent positive and negative values, respectively, and negative areas are shaded.

[14] Despite the expectation that the greenness may be increased due to the increased growing period, the general patterns seem to be regionally dependent. It can be seen that NDVI increased significantly during the 25-year period in the semi-arid region over northeastern China (110°–125°E and 35°–45°N). In general, locally varying trends are found over the area from east Mongolia to Manchuria. On the other hand, high latitude regions north of 50°N, where vegetation greenness has been increased over the two decadal period until the late 1990s due to the lengthened growing season [Tucker et al., 2001; Zhou et al., 2001; Slayback et al., 2003], show a relatively weak increasing trend. Some areas, such as southeast of Lake Baikal, show an overall decreasing trend. Relatively weak trends or locally varying trends appear to be somewhat inconsistent with the expected trend from the temperature increase and its associated lengthened growing season over mid and high latitudes in the Northern Hemisphere shown in previous studies [Zhou et al., 2003; Piao et al., 2003, 2004; Fang et al., 2005]. At this point we may conjecture that it is because of different NDVI trends between the periods ‘before the mid-1990s’ and ‘after the mid-1990s’, as shown in other studies about prevailed drought conditions or increased dust storm frequencies. In other words, recent changes in climate trends over East Asia might obscure the long-term vegetation trend.

3.2. EOF Analyses of NDVI, Precipitation, and Temperature

[15] We attempt to identify temporal patterns of vegetation changes and relate them to climate variables by separating the total growing season mean NDVI, precipitation, and surface temperature anomalies into independent modes by conducting EOF analysis. In this analysis NDVI/precipitation data over the arid and semi-arid regions are included. We may concern about the inclusion of NDVI/precipitation data over the desert and semi-arid region in the analysis because less accurate NDVI/precipitation data over those regions may introduce biases in other areas. However, its influences appear to be insignificant (not shown).

[16] Figure 3 shows the eigenvectors and associated principle components (PCs) for the first three leading EOF modes of NDVI, which explain 44% of the total variance. The first EOF mode representing about 21% of the total variance shows positive values over most regions except for the mid to western part of Mongolia and the eastern part of Inner Mongolia. It is noted that significant variations of the growing season vegetation are found north of 50°N with a maximum region around the 110°–125°E area, implying that the vegetation varied more significantly in the high latitude regions. The time series of the first PC shows a large interannual variability; however, there appears to be an increasing trend until 1997 and a decreasing trend afterwards.

Figure 3.

(a, c, and e) First three leading eigenvectors and (b, d, and f) associated principal components obtained from EOF analysis of the growing season mean NDVI anomalies over East Asia in the period of 1982–2006. Contour interval is 0.003, and zero line is omitted. Solid and dashed lines represent positive and negative values, respectively, and negative areas are shaded.

[17] The second EOF mode, explaining about 12.6% of the total variance, appears to be related to regionally dominating vegetation changes; i.e., weak positive areas scattered largely over the west of 105°E in Mongolia and east China, and negative areas scattered in the mid-east of Mongolia and east areas of Lake Baikal. Relatively strong variations in NDVI are seen over the region of 35°–45°N and 110°–120°E. The corresponding time series shows a gradual increase for the past 25 years with a stronger NDVI variation over the last few years. Considering the spatial and temporal patterns, the second mode indicates that vegetation has steadily decreased in the area from mid-east Mongolia to the east of Lake Baikal and Manchuria while steadily increased in the eastern part of China.

[18] The third EOF mode, explaining 10.3% of the total variance, shows a distinct feature over the eastern border area of Mongolia. The corresponding time series shows a generally increasing trend until the mid-1990s and then a generally decreasing trend, which is somewhat similar to the trend noted in the first EOF.

[19] Although there are significant interannual fluctuations, the three modes together reflect a substantial change of vegetation patterns since the mid-1990s; EOF modes 1 and 3 show a trend change around the mid-1990s. This result suggests that there may have been a noticeable change of local climate around the mid-1990s.

[20] In order to examine whether NDVI changes noted in the EOF analysis are related to other climate variables, EOF analyses of precipitation and temperature are conducted. Figure 4 shows spatial patterns of the first three leading EOF modes of precipitation during the growing season and their corresponding PCs. The combined total explained variance by first three leading modes accounts for about 59%.

Figure 4.

Same as Figure 3 but for precipitation anomalies. Contour interval is 1.0.

[21] The first EOF mode, explaining about 31% of the total variance, shows a spatial pattern of rainfall climatology associated with the Asian summer monsoon. Significant variations are noted over east of 110°E with a maximum region over east China and Korea where the monsoon front and related precipitation prevail during the summer. This wet region is contrasted with the dry region in the west. It is also noted that a relatively strong rain belt appears to extend along the border between Mongolia and Russia. The corresponding PC time series shows a strong interannual variability. But there seems to be varying trends separated in the late 1990s; a negative slope turned into a positive slope after late 1990s. In particular, 1998 and 2003 years showing local peaks in the PC were recorded as flood years related to the variation of East Asia summer monsoon [Zou et al., 2005]. Combining the spatial pattern with PC of mode 1 suggests a general precipitation decrease before the late 1990s and a recovery of precipitation after the late 1990s, seemingly associated with interannual variation of monsoon activities.

[22] The second EOF mode of the precipitation, explaining about 15.7% of the total variance, shows an extended anomaly band from the east of Lake Baikal to Manchuria, with an anomaly centered over the northeast border of Mongolia. The associated PC time series of mode 2 shows a sudden shift of trend; general positive anomalies before 1997 in contrast to general negative anomalies after 1997, reflecting an apparent decrease in precipitation over the area from east of Lake Baikal to Manchuria through the northeast border of Mongolia. These changes appear to be consistent with the decrease in vegetation over those regions after late 1990s.

[23] The third EOF mode accounting for 11.8% of the total variance seems to indicate subregional dipole-like precipitation changes between north Manchuria and North Korea. The associated PC time series shows a strong interannual variation although there seems to be a weak decreasing trend after mid-1990s.

[24] EOF analysis results for temperature are given in Figure 5. It is shown that the first three leading modes of the growing season mean temperature anomalies explain about 81% of the total variance. The first EOF mode explaining about 62% of the total variance indicates that temperature variability during the growing season is largely associated with the first mode, and other modes appear less important. The spatial pattern of the first EOF mode is largely characterized by the same positive sign over the whole region, with the stronger temperature variations occurred over Mongolia north of 45°N. The PC time series shows a distinct increasing trend throughout the analysis period, however, an abrupt change in anomaly occurred around 1997 is evident. Although this mode generally indicates the recent global warming trend, the abrupt change appears to be closely related to the EOF mode 2 of precipitation, which also shows strong anomaly centers situated north of 45°N and an abrupt change in anomaly around 1997.

Figure 5.

Same as Figure 3 but for temperature anomalies. Contour interval is 0.1, and zero line is omitted.

[25] The remaining second and third modes seem to separate the signal according to the constraints of spatial and temporal orthogonality; e.g., a north-south pattern in the second mode and an east-west pattern in the third mode. Moreover, compared to the dominant first mode, the second and third modes seem less significant although there is an indication of somewhat increasing temperature trend after 1997 in the mode 3.

[26] In summary, the EOF analyses of the precipitation and temperature indicate that there was a sudden increase in temperature over the most of the analysis domain which seems to be related to the sudden decrease in precipitation around 1997. Since the vegetation is greatly influenced by those climate variables it is natural to examine further the NDVI variations before 1997 and after 1997. Also expected is the possible cause and effect linked between NDVI and precipitation/temperature. In this purpose we adopt the SVD analysis for the further examination.

3.3. Trend Changes of NDVI Around 1997

[27] Because of the quite distinct changes in the sign or the trend before and after 1997 found in the EOF analysis of precipitation and temperature, we separate the analysis period into two parts, i.e., ‘before the mid-1990s’ (1982–1996) and ‘after the mid-1990s’ (1997–2006) to explore how the vegetation activities vary from one period to another. To examine how the trends of NDVIs vary between the two divided periods over East Asia, the growing season mean trends of NDVI are calculated separately using a linear regression method. Obtained regression slopes are provided in Figure 6.

Figure 6.

Mean trends of growing season mean NDVI anomalies for (a) the period before the mid-1990s (1982–1996) and (b) the period after the mid-1990s (1997–2006).

[28] The significance test shows that the slopes with a 90% significance level and higher are sparse because of a smaller number of data at a given grid point after the separation. Nonetheless it is clear that the trends are significantly different between the two periods. Before the mid-1990s (Figure 6a), NDVI during the growing season appears to increase in most areas over East Asia, consistent with findings from other studies [Kawabata et al., 2001; Slayback et al., 2003; Piao et al., 2006]. In contrast, a distinct decline in vegetation greenness is noted after the mid-1990s (Figure 6b) over the high latitude regions north of 50°N and over the Mongolian border east of 110°E. Much of northeast China and Korea exhibits a negative or weakly negative trend after the mid-1990s.

3.4. NDVI Changes Related to Precipitation/Temperature

[29] The above trend analysis revealed the stressed vegetation since the mid-1990s specifically in the high latitude regions north of 50°N and some parts of semi-arid regions, in contrast with the increased vegetation before the mid-1990s. In this section, we further examine how interannual variations of the vegetation are linked to changes in precipitation and temperature using SVD analysis of two covariance matrix fields (here NDVI versus precipitation, and NDVI versus temperature) for the whole period. The SVD technique provides means for separating coupled modes of variability between two fields and identifying their relationship [Wallace et al., 1992]. In the SVD analysis, each pair of spatial patterns describes a fraction of the squared covariance (i.e., the ratio of each squared singular value to the total squared covariance) between two variables, while the temporal correlation between pairs of PC (expansion coefficients) time series indicates the degree of coupling between two variables [Bretherton et al., 1992].

[30] Figure 7 shows spatial structures and time series of the first three leading paired modes of NDVI-precipitation, which together explain about 75.9% of the total squared covariance. The first mode explains 42.7% of the total squared covariance between two fields, with a correlation of 0.79 between the two PC time series. The NDVI field of mode 1 is characterized by negative anomalies in the high latitude regions north of 50°N and positive anomalies in the semi-arid regions south of 50°N (Figure 7a). The corresponding precipitation anomaly pattern characterized by positive signs over most of the domain (Figure 7b) with local maxima over Manchuria and east China seems to represent the precipitation variabilities associated with the Asian summer monsoon, similar to the spatial pattern noted in EOF mode 1 of precipitation (Figure 4a). High latitude regions north of 50°N show somewhat out-of-phase relationship between NDVI and precipitation. However, the weak squared covariance of precipitation over high latitudes suggests that the impact of precipitation changes on the vegetation growth may be minor. On the other hand, the in-phase relationship with stronger covariance shown in the water-limited semi-arid regions indicates that more precipitation during the growing season yielded more active vegetation growth. The time series for mode 1 shows significant interannual fluctuations with a 3–4 year period (Figure 7c), quite similar to the time series of precipitation EOF mode 1. Also found is a well-known climate features of reduced precipitation in 1997 and in the 1999–2002 period, corresponding to severe droughts in East Asia.

Figure 7.

Spatial patterns and time series of expansion coefficients for first three leading modes between NDVI and precipitation anomaly fields for the period of 1982 to 2006. The time amplitudes are normalized by respective standard deviation.

[31] The major spatial feature of the mode 2, explaining about 20.1% of covariability, is nearly in-phase joint pattern particularly in the area extending from the eastern border of Mongolia to the east of Lake Baikal. Again, the SVD spatial pattern of precipitation is very similar to the EOF mode 2 of precipitation. Time series is also consistent with one found in EOF mode 2 of precipitation, i.e., different trends separated in mid-1990s. This sharp increase reflects that the decreased precipitation after the mid-1990s contributed to the reduced vegetation particularly over east Mongolia and eastern parts of Lake Baikal. Overall, the in-phase relationship in SVD mode 2 strongly suggests that vegetation changes over those regions are positively coupled with precipitation changes.

[32] The third mode explaining 13.1% of covariability shows positive NDVI anomalies over most of the domain while precipitation anomalies exhibit an east-west separation in the south of 50°N. Although the positive trend before the mid-1990s and weak negative trend afterwards may carry physical meaning, the locally scattered spatial patterns in both variables suggest that the mode 2 may not be physically significant.

[33] Figure 8 displays covariability patterns and time series of the first two leading paired modes between NDVI and temperature fields for the period of 1982–2006. About 85.7% of the total variance is explained by two modes, implying a strong linkage between NDVI and temperature. The mode 1 explains about 78.8% of the total squared covariance with a 0.84 correlation coefficient, which implies a strong covariability between two fields throughout the analysis period. Dominant positive NDVI anomalies are found over the north of 50°N with a predominantly positive temperature anomaly. This in-phase relationship found in mode 1, particularly over the high latitude regions north of 50°N, suggests that either the higher temperature during the growing season is giving rise to more active vegetation growth or vice versa. It is also noted that the spatial pattern of temperature is very similar to one found in EOF mode 1. The corresponding time series (Figure 8c) show a gradual increase in magnitude with an abrupt change in temperature anomaly occurred around 1997, which is similar to one found in EOF mode 1 of temperature. The spatial and temporal structure supports a notion that mode 1 reflects a warming trend and its associated increase in vegetation greenness. Such a notion is particularly true in the high latitudes north of about 50°N.

Figure 8.

Same as Figure 7 but for the first two leading modes between NDVI and temperature anomaly fields.

[34] On the other hand, the mode 2 of NDVI-temperature fields explains about 6.9% of total covariance. The vegetation pattern again shows a separation between the semi-arid region and its surroundings, as in mode 1 of the vegetation-precipitation pair (Figure 7a). In the temperature field, a largely east-west pattern of separation is found. However, considering that explained covariance is less than 7%, the mode may not be considered significant.

[35] Overall SVD results suggest that increased vegetation greenness in the high latitude regions before the mid-1990s (depicted in Figure 8a) is largely due to the general increase of temperature (in mode 1 of Figure 8). These results confirm the findings from earlier studies that greening over the northern mid to high latitudes between the early 1980s and the late 1990s can mainly be attributed to the global warming [Gong and Ho, 2003; Slayback et al., 2003; Zhou et al., 2003]. However, it is also noted that the increasing became much weaker (or reversed) after the mid-1990s, as inferred from the time series of SVD mode 1 of NDVI-temperature fields (in Figure 8c). On the other hand, the weakened (or reversed) trends of vegetation greening found after the mid-1990s over the eastern parts of Lake Baikal and eastern Mongolian border appear to be closely associated with the local precipitation variability–see the mode 2 of Figure 7. The results of SVD analysis support the notion that locally reduced precipitation suppressed the vegetation growth by aggravating the surface water availability despite above normal temperature conditions.

4. Summary and Discussion

[36] Recognizing the different climate patterns after the mid-1990s inferred from other studies on the trends of change in precipitation and/or dust frequencies, it was our objective to examine whether concomitant signals could be found in the long term NDVI data, in response to such climate variations after the mid-1990s. We further intended to understand how climatically important meteorological parameters are linked to the vegetation changes in East Asia, if there exist changes in climate trends after the mid-1990s. In pursuing these objectives, we examined the spatiotemporal variations in the vegetation over East Asia using the growing-season (April–October) mean NDVI data for the period of 1982–2006. In particular, we examined how changes of climate parameters (e.g., precipitation and temperature) are interrelated with the vegetation changes using the SVD analysis.

[37] It is noted that there have been general trends of change in NDVI around the mid-1990s, separating the locally dependent 25-year mean (1982–2006) trend into a predominantly positive trend over most of the East Asian domain before the mid-1990s (1982–1996), and a weakened (or reversed) trend after the mid-1990s (1997–2006). The latter reversed or weak signal was particularly dominant in the high latitudes north of 50°N. Because of contrasting trends, the overall 25-year mean trend of NDVI seems to be obscured and appears locally dependent, in contrast to the predominantly increasing vegetation trend over most of East Asia found in other literatures. The predominantly positive NDVI trend continued until the mid-1990s and then a reversed (or weakened) NDVI trend appeared after the mid-1990s. This trend change appears to be linked to sudden changes in temperature and precipitation as shown EOF analysis.

[38] The possible influences of precipitation and temperature changes on such NDVI variations were further explored using the SVD technique. It is difficult to pin down the cause of increased vegetation during the former period because of interrelated climate variables. However, the high percentage of captured covariance between NDVI and temperature and the clear increasing trends found in the time series strongly suggest that the evident warming trend noted until the mid-1990s would be the most important factor for inducing the positive NDVI trend, in particular over the high latitude regions. In addition, the decreased local precipitation before the mid-1990s in the humid regions north of 50°N seemed to contribute to the enhanced vegetation over those regions.

[39] However, the positive NDVI trends appeared to be disturbed after the mid-1990s in spite of a prevailing temperature increase. The area north of 50°N showed an increased vegetation trend before the mid-1990s, after which the trend was reversed, but the causes to such changes are not clear from the SVD analysis. Although overall out-of-phase relationships between NDVI and precipitation are found over the high latitudes north of 50°N in the SVD modes (mode 1 and 3), the role of precipitation in the trend change is not clear.

[40] On the other hand, in areas such as the east of Lake Baikal and east Mongolia, local precipitation changes seemed to play an important role in the vegetation changes around the mid-1990s, since the second SVD mode of NDVI/precipitation fields depicts a reversed trend after the mid-1990s. With above-normal temperature conditions, the obviously reduced precipitation after the mid-1990s seemed to contribute to reduced vegetation over those regions.

[41] In this study we tried to link vegetation changes over East Asia with variations of natural climate variables. But, the ecosystems over the ecologically sensitive regions such as arid and semi-arid areas are influenced not only by climate changes but also by intensive human activities such as wildfires, overgrazing, illegal deforestation, and land use conversion. Forest and grassland fires can be one typical example. In China, the occurrence of forest fires has been dramatically decreased since 1980s and has remained relatively low and stable afterwards [Yan et al., 2006; Lü et al., 2006] while grassland fires as a minor source of biomass burning in China have also substantially decreased after mid-1990s [Yan et al., 2006], suggesting a general recovery of vegetation in China after those periods. In addition, steadily increased irrigations and improved agricultural production over the Inner Mongolia from mid-1980s and 1990s might have contributed to the NDVI increase in those farming areas [Xin et al., 2008].

[42] By contrast, in Russia, extreme forest fires and burned areas have been increased in recent years with severest seasons in 2002 and 2003, including the land degradation over the east of Lake Baikal after the mid-1990s [Sukhinin et al., 2004]. Mongolia has also experienced a continuous increase in forest fires since 1990s with extreme fire hazards in 1996 and 1997 [Goldammer et al., 2004]. Majority of burned areas are located in the central and eastern parts of the forest/grassland area, leading to a conjecture that forest fires may be one of the causes of reduced vegetation after the mid-1990s in those regions. Since those periods and locations are coincident with geographical locations and time showing negative anomalies of precipitation with positive temperature anomalies, we speculate that general changes in climate conditions have helped fires occur easier and spread wider. Such speculation leads to an indirect effect of climate change on the land cover changes through human activities. Although it is very difficult to prove how those fires are linked to climate changes and thus beyond the scope of this study, we bring attention to interpret current conclusions because of the possible man-made influences on vegetation changes.

Acknowledgments

[43] The authors would like to thank three anonymous reviewers for their constructive and valuable comments, which led to an improved version of the manuscript. This work was supported by the Korea Meteorological Administration Research under NMRI's Satellite Application Program and by the BK21 Program of the Korean Government.

Ancillary