Climate Dynamics

, Volume 32, Issue 7, pp 1173–1186

Moisture variability across China and Mongolia: 1951–2005

Authors

    • Tree-Ring Laboratory, Lamont-Doherty Earth ObservatoryColumbia University
    • Center for Arid Environment and Paleoclimate Research, MOE Key Laboratory of West China’s Environmental SystemLanzhou University
  • Edward R. Cook
    • Tree-Ring Laboratory, Lamont-Doherty Earth ObservatoryColumbia University
  • Rosanne D’arrigo
    • Tree-Ring Laboratory, Lamont-Doherty Earth ObservatoryColumbia University
  • Fahu Chen
    • Center for Arid Environment and Paleoclimate Research, MOE Key Laboratory of West China’s Environmental SystemLanzhou University
  • Xiaohua Gou
    • Center for Arid Environment and Paleoclimate Research, MOE Key Laboratory of West China’s Environmental SystemLanzhou University
Article

DOI: 10.1007/s00382-008-0436-0

Cite this article as:
Li, J., Cook, E.R., D’arrigo, R. et al. Clim Dyn (2009) 32: 1173. doi:10.1007/s00382-008-0436-0

Abstract

Moisture variability across China and Mongolia (hereafter, CM) during 1951–2005 was investigated using the recently developed monthly Palmer Drought Severity Index (PDSI) dataset. In total there are 206 PDSI grid points across CM, based on a 2.5° × 2.5° gridding system. For CM as a whole a significant decreasing trend in mean moisture availability was observed during 1951–2005, but with strong decadal (17.1-year) and interannual (5.0-year, 3.2-year, 2.4–2.8 year) variations. The areas affected by moderate and severe moisture deficit over CM have increased significantly since the mid-1950s. In contrast, there is a significant decreasing trend for areas affected by moderate wetness since the mid-1950s, and no significant trend was found for the areas affected by severe wetness. Ten moisture-related spatial patterns were objectively defined for CM using rotated Empirical Orthogonal Function (REOF) analysis. These patterns are related to distinct geographical areas and are associated with distinct temporal variations. Four of these patterns, in Northeast China (NE), North China (NC), Central China (CC), and East China (EC), generally demonstrate a significant decreasing trend in moisture availability. Two patterns located in western areas of Northwest China (NW) and the Tibetan Plateau (TP) show a significant moisture increase, while four patterns in Mongolia (MN), far western China (FW), South China (SC), and Southwest China (SW) do not have significant moisture trends during 1951–2005. Based on REOF results we propose that CM should be divided into ten coherent moisture divisions. Moisture variations within each division are generally coherent, but may show either similar or contrasting covariability with adjacent divisions.

1 Introduction

China and Mongolia (CM) are located in the eastern part of the Eurasian continent, a region characterized by complex topography and climate (Domrös and Peng 1988). Due to the varying effects of multiple climate forcings, moisture availability within CM can vary dramatically. The most dramatic moisture contrast is found between the regions of southeastern China (annual rainfall over 1500 mm; controlled by the monsoonal climate) and northwestern China and southern Mongolia (annual rainfall less than 200 mm; controlled by the continental climate) (Yatagai 2003; Ye et al. 2004).

Despite the great diversity in mean moisture conditions, a common feature across CM is that drought and flood extremes can impact any region. Such extreme climate events typically occur over a relatively large area, enforcing strong effects on regional human activities and ecosystems. A few well-known such events include the 1998 flood in eastern China (Zong and Chen 2000), the 2001 persistent, severe drought in northern China (People’s Daily, 31 May 2001), and a multi-year drought across northern China and Mongolia in the late-1920s (Davi et al. 2006; Liang et al. 2006: Li et al. 2007). Therefore, there is a need to investigate spatial and temporal patterns of moisture variability across the entire CM. Such investigations will aid water resource management for different regions, and will lead to a better understanding of the underlying climate forcings that impact CM.

A number of studies have examined moisture variability (particularly precipitation) over CM. However, most of these studies focused either on specific regions (e.g., Wang and Li 1990; Liang et al. 1995; Wang et al. 2003; Xin et al. 2006), or on large areas identified geographically (e.g., Wang and Zhou 2005; Zhai et al. 2005; Zou et al. 2005) or by temperature derived boundaries (e.g., Qian and Zhu 2001). Exceptions include the study of Qian et al. (2003), who analyzed spatial and temporal dryness/wetness variations for eastern China based on historical documents derived dryness/wetness index (DWI), and Bordi et al. (2004), who analyzed the spatio-temporal variability of dry/wet periods in eastern China by using the standardized precipitation index (SPI). To our knowledge no studies have described the spatial patterns and coherency of moisture variations across either all of China or the entire CM. One reason for the scarcity of such studies is that until recently an appropriate moisture parameter, with homogeneous data distribution across the entire area and one that allows direct comparison of spatial and temporal variations, did not exist. Fortunately, this difficulty has partially been alleviated with the recent development of large-scale datasets of the PDSI (Dai et al. 2004; Zou et al. 2005).

The PDSI was first designed by Palmer (1965) with the intent of measuring regional moisture availability by incorporating antecedent precipitation, soil moisture demand, and supply into a primitive hydrological accounting system. The PDSI is an index of meteorological drought, and expresses moisture conditions ranging from extremely dry to extremely wet levels, generally within a range of −6 to 6 (Heim 2002). As the PDSI is standardized to enable direct time and space comparisons of moisture variability for diverse climates, it has become a standard for measuring meteorological drought across the United States in the years since its development (Alley 1984; Heim 2002), and is now being utilized as a drought metric for much of the globe (Briffa et al.1994; Lloyd-Hughes and Saunders 2002; Ntale and Gan 2003; Dai et al.2004; Zou et al.2005; van der Schrier et al. 2006a, b).

It is worth a noting that a drought index called China-Z Index (CZI; Wu et al. 2001) has been successfully used as an operational tool to monitor real-time droughts and floods over China. Similar to the SPI, the CZI considers only precipitation and assumes the precipitation data follow the Pearson Type III distribution (Wu et al. 2001). It has been shown that the performance of the CZI and the SPI is virtually identical in many regions (Wu et al. 2001; Morid et al. 2006). However, it has also been shown that such probability-based indexes may be biased on reporting the severity of droughts in arid regions (Hayes et al. 1999; Wu et al. 2007). Considering a large portion of CM is under arid or semi-arid climate, the PDSI is therefore preferably used in the current study.

The PDSI, as well as the newly designed “self-calibrating” PDSI (Wells et al. 2004), has been employed to investigate spatial and temporal variations of moisture conditions across North America and Europe with much success (Karl and Koscielny 1982; Briffa et al.1994; Cook et al.1999; van der Schrier et al. 2006a, b). However, no such work has been done for eastern Eurasia. In this paper, we investigate spatial and temporal moisture variations across CM using the recently developed PDSI dataset of Dai et al. (2004; hereafter, DAI PDSI). We first test the validity of the DAI PDSI for describing moisture conditions over CM. We then assess changes in mean moisture conditions as well as the relative representation of moderate and severe dry/wet areas. We define ten spatial moisture patterns for the entire CM using the REOF analysis, and describe the temporal changes of moisture availability associated with each of these regional patterns. Finally, we propose to divide CM into ten coherent moisture divisions based on REOF results.

The current paper is organized as follows. Sect. 2 describe the DAI PDSI dataset and the analytical techniques that were employed in this study. Sect. 3 describe the validation of the DAI PDSI dataset. Sect. 4 describe observed changes in mean moisture conditions as well as the relative coverage of moderate and severe dry/wet areas. In Sect. 5 we present the ten defined spatial moisture patterns and their associated temporal variations, and define the ten moisture divisions over CM. The study is summarized in Sect. 6.

2 Data and methods

2.1 The DAI PDSI dataset

The monthly DAI PDSI dataset used herein is an update of the earlier version of Dai et al. (1998). The DAI PDSI dataset has global coverage based on a 2.5° × 2.5° gridding system, and, for version 3 (updated in November 2006), spans from 1870 to 2005. Calculation of this dataset involved the use of two precipitation datasets (pre-1948 data from Dai et al. (1997) and data for 1948–2005 from the National Centers for Environmental Prediction (NCEP) Climate Prediction Center (Chen et al. 2002)), the Climate Research Unit (CRU) surface air temperature dataset (Jones and Moberg 2003), and a water-holding capacity dataset by Webb et al. (1993). Details on the development of the DAI PDSI dataset are found in Dai et al. (2004).

There are a total of 206 PDSI grid points across CM, all of which were used in this study (Fig. 1). Since most of the meteorological stations in China were not established until the 1950s, we only used the monthly DAI PDSI for the period from 1951 to 2005 (a total 660 months). It should be noted, however, despite the use of monthly PDSI in the current study, the PDSI is imprecise in its treatment of all precipitation as rainfall, not accurately accounting for frozen soils and snow (Alley 1984; Dai et al. 2004). There are two considerations regarding the use of monthly PDSI for the current study. First, there is a large portion of CM where it rarely snows, such as south and southeast China. Second, the PDSI is an accumulating index that has a fairly long memory (Alley 1984; Dai et al. 2004). As such, climate conditions in prior winter and spring will affect summer PDSI, and summer climate conditions will affect fall and winter PDSI as well. Such high autocorrelations among individual months make the results virtually identical when examining the PDSI trends in winter and summer months (Dai et al. 2004). Considering our main purpose here is to identify spatial patterns and temporal changes of moisture over CM, we deem it appropriate to use the monthly PDSI for this study. Therefore, the above PDSI data (206 grids times 660 months) provided the basis for the current study.
https://static-content.springer.com/image/art%3A10.1007%2Fs00382-008-0436-0/MediaObjects/382_2008_436_Fig1_HTML.gif
Fig. 1

Map of China and Mongolia showing the locations of the 206 PDSI grid points used in this study. The PDSI grid points were developed by Dai et al. (2004) for global coverage based on a 2.5° × 2.5° gridding system

2.2 Analytical techniques

We used REOF method to objectively define the most significant regional patterns of moisture variability across CM. REOF analysis has been widely used in meteorology and climatology to define spatial and temporal variability of a large set of variables (Richman 1986; Briffa et al.1994; Cook et al.1999; Qian et al. 2003). This technique is one of the best ways to define the most significant regional patterns (rather than for pure data reduction), as it can efficiently recover the underlying “simple structure” (Thurstone 1947; Kaiser 1958) and thus facilitate the search for physical interpretations.

The rotation method used in this study is the correlation-based Varimax rotation (Kaiser 1958). We selected the correlation-based matrix (rather than the covariance-based matrix) for rotation since we are mainly interested in defining spatial covariability, rather than in isolating patterns in regions of exceptionally high variability. Varimax is generally considered the most accurate analytic orthogonal rotation, and is the only orthogonal rotation method that has been used widely (Richman 1986). Varimax attempts to simplify the structure of the patterns by making each factor to have a small number of large loadings and a large number of the loadings towards zero. After Varimax rotation, each original variable tends to be associated with one (or a small number of) factor and each factor represents only a small number of variables. For comparison, we also performed Promax rotation (Hendrickson and White 1964), which is considered to be one of the best oblique rotation methods for recovering underlying true simple structure (Richman 1986). Additional information on REOF analysis can be found in the review papers of Richman (1986) and Hannachi et al. (2007).

The significance of trends was evaluated using the Mann–Kendall method (Mann 1945; Kendall 1975). Spectral analyses were performed using the Multi-Taper Method (MTM; Mann and Lees 1996) in order to examine changes in mean moisture conditions in the frequency domain.

3 Evaluation of the DAI PDSI

We first conducted statistical analyses to evaluate the ability of the DAI PDSI dataset to describe moisture conditions across CM. As shown in Fig. 2a, overall the PDSI values have a normal distribution, although slightly skewed toward negative values (skewness = −0.15). The mean and median PDSI values are −0.43 and −0.52, respectively, which are within or close to the range of the defined “normal” moisture status (PDSI = 0.0 ± 0.5; Palmer 1965). Most of the PDSI values are within the range of ±6.0, although a slightly higher proportion of extreme values (3.37%; defined as |PDSI| ≥ 6.0) are reported.
https://static-content.springer.com/image/art%3A10.1007%2Fs00382-008-0436-0/MediaObjects/382_2008_436_Fig2_HTML.gif
Fig. 2

Plots showing the frequency of monthly PDSI values over (a) the entire range of PDSI values and (b) the major PDSI categories. All the PDSIs across CM during 1951–2005 were used in this study

Sorting of the PDSI into major PDSI categories (Wells et al. 2004) reveals a weakly bimodal distribution, with one peak representing mildly dry regimes and the other representing mildly wet regimes (Fig. 2b). The frequency of PDSI in the mildly dry regime (24.02%) is higher than for mildly wet regime (17.86%), and both show a higher frequency than the near normal regime (14.16%). Overall, the frequencies of near normal and mild conditions account for 56.05%, which is quite reasonable. However, the frequency of reported extremely dry regimes (PDSI ≤ −4.0) is 7.84%, even higher than that of reported severely dry regimes (−4.0 < PDSI ≤ −3.0, 6.93%), hardly corresponds to the classification of “extreme” drought. Extreme wet conditions (PDSI ≥ 4.0) have also been slightly over-reported (4.57%).

The problem of over-reporting of extreme and/or severe moisture conditions in the DAI PDSI was examined further. As shown in Fig. 3a, c, over-reported extreme and severe dry regimes are not represented in the traditionally arid and semi-arid regions (i.e., northwestern China and southern Mongolia), and instead are mainly found in North and Northeast China, where a significant drying trend has occurred since the 1950s (Wang and Zhou 2005; Zou et al.2005). Similarly, over-reported extreme and severe wet spells are not found in the traditionally wet regions (i.e., eastern and southern China), and instead are mainly found over the Tibetan Plateau and the western areas of Northwest China (Fig. 3b, d), where significant positive trends in moisture have occurred in the last several decades (Liu et al. 1998; Bräuning and Mantwill 2004; Li et al. 2006).
https://static-content.springer.com/image/art%3A10.1007%2Fs00382-008-0436-0/MediaObjects/382_2008_436_Fig3_HTML.gif
Fig. 3

The frequency at which the PDSI reported (a) an extreme dry spell (percentage of months in 1951–2005 with PDSI ≤ −4.0), (b) an extreme wet spell (percentage of months in 1951–2005 with PDSI ≥ 4.0), (c) a severe or extreme dry spell (percentage of months in 1951–2005 with PDSI ≤ −3.0), and (d) a severe or extreme wet spell (percentage of months in 1951–2005 with PDSI ≥ 3.0)

It has been shown that the behavior of the PDSI at various locations is inconsistent, resulting in the report of more severe PDSI in some regions than others (Alley 1984; Guttman et al. 1992). Such a drawback makes spatial comparison of the PDSI over diverse climatological regions difficult, and stimulates the design of “self-calibrating” PDSI (Wells et al. 2004; van der Schrier et al. 2006a, b). However, the problem of the PDSI on over-reporting extreme values over certain regions may reveal another property of the PDSI that emerges when describing moisture conditions for areas with persistent positive or negative trends in moisture (Fig. 3). Since the PDSI was designed to measure the cumulative departure (relative to local mean conditions) in moisture supply and demand at the surface (Dai et al. 2004), persistent moisture increase (decrease) will lead to the occurrence of some clusters of extreme positive (negative) PDSI values. Therefore, such behavior in over-reporting extreme values may also be a manifestation of the cumulative effects of persistent moisture excess/deficit over the areas shown in Fig. 3. Consequently, the problem of over-reporting extreme values in the DAI PDSI may not be as serious as first indicated.

4 Trends in mean moisture conditions and the relative coverage of dry/wet areas

Monthly PDSI values were averaged over the entire CM for 1951–2005 in order to investigate recent changes in mean moisture conditions. As shown in Fig. 4a, mean moisture availability across CM shows a persistent decreasing trend during 1951–2005 (statistically significant at the 0.01 level), with overlapping pronounced decadal and interannual variations. This decreasing trend is most significant from the early 1990s to 2001, with 2001 being the driest year in the entire interval. Regarding decadal-scale variability, above normal conditions were mainly found in the late half of the 1950s, mid-1970s, and around 1990, with below normal conditions in the mid-1960s, late-1970s to early-1980s, and early twenty first century.
https://static-content.springer.com/image/art%3A10.1007%2Fs00382-008-0436-0/MediaObjects/382_2008_436_Fig4_HTML.gif
Fig. 4

a Time series of monthly PDSI values averaged over the entire CM for 1951–2005, and b its MTM spectral density. In (a) the horizontal line indicates the mean value of the PDSIs and in (b) the bold line indicates the null hypothesis, and the dash, dash-dot and dotted lines indicate the 90, 95 and 99% significance levels, respectively

To further investigate the interannual and decadal variations in the DAI PDSI, spectral properties of the record of mean moisture conditions were examined using the MTM method. As shown in Fig. 4b, there is statistically significant spectral activity at greater than 28.4-year, which is a manifestation of the long-term moisture trend. Also indicated is a significant 17.1-year peak and several significant interannual peaks at 5.0, 3.2, and 2.4–2.8 years. These interannual and decadal cycles may be intrinsic to the Asian monsoon system, the El Niño-Southern Oscillation (ENSO; Allan et al. 1996) or other aspects of tropical to higher latitude atmosphere-ocean variability (Feng and Hu 2004; Chan and Zhou 2005; Shen et al. 2006; Sung et al. 2006).

It is important to obtain additional information about long-term trends in regions influenced by different moisture patterns. To this end, the spatial extents of moderate and/or severe moisture conditions during the 1951–2005 periods were investigated. As in Briffa et al. (1994) and van der Schrier et al. (2006b), moderate moisture conditions are defined as |PDSI| > 2.0, and severe moisture conditions as |PDSI| > 4.0. As shown in Fig. 5a, despite a decline in the early 1950s, the areas affected by moderate and severe moisture deficit have increased during 1951–2005 (both trends significant at the 0.01 level). On average, about 20% of the area of CM was affected by moderate drought in the 1950s. This number increased to about 30% in the 1980s and to about 50% in the late-1990s to early twenty first century. Similarly, areas affected by severe drought increased from about 5% in the 1950s to about 10% in the 1980s, and then to above 20% in the late-1990s to early twenty first century. Most significant increases in moderate and severe moisture deficit were found during 1990–2001, and in 2001 (the driest year in the 1951–2005 period) up to 65% (50) of CM was affected by moderate (extreme) drought.
https://static-content.springer.com/image/art%3A10.1007%2Fs00382-008-0436-0/MediaObjects/382_2008_436_Fig5_HTML.gif
Fig. 5

Indices of changing areas of moderate (gray line; defined by |PDSI| > 2.0) and severe (black line; defined by |PDSI| > 4.0) (a) drought, (b) moisture excess, and (c) either drought or moisture excess. These areas are shown as percentages of the area of the entire CM

The areas affected by moderate wetness are generally less than that affected by moderate drought (Fig. 5b), except for the late half of the 1950s when about 30% of CM was affected by moderate wetness. The latter half of the 1950s is also the period with the most areas affected by severe wetness (~10%). Since the mid-1950s there is a decreasing trend for areas affected by moderate wetness (significant at the 0.01 level), although no significant trend was found for areas affected by severe wetness. For 1993, the wettest year during the 1951–2005 period, up to 46% (27) of CM was affected by moderate (extreme) moisture excess.

Figure 5c shows the overall percentage of areas affected by moderate and severe moisture (either dry or wet). Areas affected by moderate moisture deficit or excess increased in the early half of the 1950s, decreased from the mid-1950s to 1970, and then increased significantly after 1970. Areas affected by severe moisture deficit or excess were rather stationary at about 10% during 1951–1969, but started to increase dramatically after 1970, especially for 1990–2001. Since the Asian Monsoon system is the dominant controller of moisture changes over CM, these strong long-term secular variations of moderate and severe moisture conditions indicate that the strength of the monsoon and its spatial variations have changed greatly during the last several decades (Yu et al. 2004; Cheng et al. 2005; Zhai et al. 2005; Li et al. 2006).

5 Regional patterns of moisture variability

5.1 Effects of rotation

REOF analysis was used to objectively define the most significant regional patterns of moisture variability across CM. In this study the “variables” refer to the 206 PDSI grid points and the “observations” are the monthly PDSI values for each grid point during 1951–2005.

Determining the number of EOF modes to be rotated is an important issue in REOF analysis, as it directly affects the resulting spatial patterns and temporal variations, facilitating or misleading the search of physical interpretations. We first performed a rigorous “red noise” version of the Rule N test (Preisendorfer 1988), which is based on the Monte Carlo procedure. With the use of orthogonal Varimax rotation, the first eight EOFs (or factors) were found to be significant at the 95% confidence level. However, rotation of eight EOFs blended some discrete patterns together, resulting in the bias of “underfactoring” (i.e., selecting too few factors for retention; Richman 1981). This result is consistent with the finding that the Rule N test tends to underestimate the number of significant EOF modes (Preisendorfer 1988). Further evaluation indicated that the rotation of more than ten factors increased the retention of noise, resulting in the appearance of some patterns with no clear spatial distributions (i.e., the problem of “overfactoring”; Richman 1981). Therefore, only ten EOFs were rotated for this study, producing ten clearly defined spatial patterns (Fig. 6).
https://static-content.springer.com/image/art%3A10.1007%2Fs00382-008-0436-0/MediaObjects/382_2008_436_Fig6_HTML.gif
Fig. 6

The ten factors for CM identified through REOF analysis. Note that the percentage variance accounted for by each factor is labeled in each map

For comparison, the oblique Promax rotation was also applied to the first ten EOFs. The ten loading patterns achieved from the two types of rotations are virtually identical (figures not shown). Similar results were reported in other studies relating to large-scale drought patterns and spatial patterns of tree growth anomalies (Karl and Koscielny 1982; Meko et al. 1993; Cook et al.1999). Therefore, the orthogonal Varimax rotation appears to be an efficient way to recover regional patterns of moisture variability across CM.

Table 1 shows the percentage variance explained by each of the first ten factors before and after the Varimax rotation. Together, the first ten factors account for 58.28% of the total variance, with the remaining variance discarded as noise. The ten REOF loading patterns are shown in Fig. 6, and their associated scores in Fig. 7. We also calculated the correlations between the ten series of scores and the time series of mean moisture conditions across CM. These results are shown in Table 2.
Table 1

The variance explained by the first ten factors before and after the Varimax rotation

Component

Unrotated EOF

Rotated EOF

% of Variance

Cumulative %

% of Variance

Cumulative %

1

16.46

16.46

10.13

10.13

2

8.37

24.83

7.10

17.23

3

6.23

31.06

6.37

23.60

4

6.05

37.12

6.02

29.62

5

4.75

41.87

5.88

35.49

6

4.42

46.29

5.40

40.89

7

3.36

49.64

5.23

46.12

8

3.17

52.81

4.82

50.94

9

2.83

55.64

3.68

54.62

10

2.64

58.28

3.67

58.28

https://static-content.springer.com/image/art%3A10.1007%2Fs00382-008-0436-0/MediaObjects/382_2008_436_Fig7_HTML.gif
Fig. 7

The ten normalized REOF scores corresponding to the ten factor patterns shown in Fig. 6

Table 2

Correlation matrix of the ten score series with the time series of mean moisture availability across CM

 

NE

NC

MN

NW

FW

SC

CC

TP

EC

SW

Mean

0.58**

0.41**

0.46**

0.07

0.33*

0.14

0.28*

−0.01

0.21

0.10

*Significant at p < 0.05

**Significant at p < 0.01

5.2 Regional patterns and temporal variations

We now describe the ten most significant regional patterns of moisture variability across CM, defined in terms of the factor maps and the associated scores from the orthogonal Varimax rotation (Figs. 6, 7).

REOF #1 essentially represents the coherent moisture variability in the northeastern part of CM (NE pattern). This pattern explains the largest percentage of the total PDSI variance (10.13%), and covers most of Northeast China and easternmost Mongolia. The scores for this pattern show a moisture increase in the early 1950s, followed by a decreasing trend from the late 1950s to the late 1970s. The moisture level recovered somewhat in the 1980s, but decreased dramatically beginning in the early 1990s. The overall drying trend during 1951–2005 is statistically significant at the 0.01 level. The wettest period in this region occurred during the latter half of the 1950s, and the driest period is found in the early twenty first century. As shown in Table 2, the series of scores for this pattern is best correlated with the time series of mean moisture variability across CM (r = 0.58), suggesting that moisture condition over the NE is a reasonably good indicator of mean moisture condition across the entire CM.

REOF #2 explains the second largest component of the total PDSI variance (7.10%) and highlights common moisture variability in North China (NC pattern). This pattern is heavily loaded over North China, especially in the middle and lower reaches of the Yellow River. The time series for this pattern basically shows above-normal moisture conditions during 1950s–1970s, and below-normal moisture conditions during 1980s to the early twenty first century. Overall there is a significant drying trend for this region during 1951–2005 (at the 0.01 level). As shown in Table 2, the time series for this pattern is also highly correlated with the record of mean moisture variability across CM (r = 0.41).

REOF #3 mainly represents coherent moisture variability over most of Mongolia (MN pattern). This pattern explains 6.37% of the total variance, with the largest loading in western and central Mongolia. The scores associated with this pattern indicate no significant moisture trend during the study period, but demonstrate strong decadal and interannual variations. The wettest period in this region occurred in the mid-1990s, and the driest period is found in the early twenty first century. As shown in Table 2, scores associated with this pattern are also highly correlated with changes in mean moisture conditions across the entire CM (r = 0.46).

REOF #4 represents coherent moisture variability over the western areas of Northwest China (NW pattern), where significant moisture increase since the early twentieth century has been identified (Li et al. 2006). This pattern explains 6.02% of the total variance, with its largest loading in the central Tien Shan area. The scores associated with this pattern indicate that there is a significant increasing trend in moisture during the study period (at the 0.01 level), especially since the mid-1970s. The relatively wet period around 1960 is consistent with the results of Li et al. (2006).

REOF #5 is heavily loaded over far western China (FW pattern). This factor explains 5.88% of the total variance and represents moisture variability in the western part of the Tarim basin and its surrounding areas. The scores suggest that moisture availability in this region fluctuated about its long-term mean during the study period, with no persistent moisture trend. In general there are no extended excursions towards very wet or dry conditions in this region. The wettest and driest period are both found in the 1980s.

REOF #6 explains 5.40% of the total variance and represents the common moisture variability over South China (SC pattern). The scores for this pattern do not indicate any persistent moisture trend during the study period, although moisture availability increased significantly from the late-1980s to 2003. Scores for this pattern also show strong multi-annual variations, with relatively wet periods in the early 1950s, early 1960s, mid-1970s, early 1980s, and late-1990s to 2003, and relatively dry periods in the late 1950s, mid-1960s, late 1970s and late 1980s. The wettest (driest) year during the study period is 1983 (2003), respectively.

REOF #7 describes moisture variability in Central China (CC pattern), particularly in the Sichuan basin and extending into the northeastern Tibetan Plateau. Variance explained by this pattern is 5.23%. The associated scores demonstrate that moisture conditions in this region are highly variable during 1951–2005, with a significant moisture increase during the 1950s–1980s and a dramatic decrease in the 1990s. Overall, moisture availability in this region decreased significantly (at the 0.01 level) during the study period. The driest period in this region is from 1995 to 2003.

REOF #8 explains 4.82% of the total variance and represents coherent moisture variability over the Tibetan Plateau (TP pattern). This pattern is most heavily loaded over the southern and central TP. The scores indicate that moisture availability over the TP has increased significantly during the study period (at the 0.01 level), although this trend was interrupted by a severe drought in the mid-1990s. Moisture increase over the TP, reported previously (e.g., Liu et al. 1998; Bräuning and Mantwill 2004), is consistent with a number of modeling and other studies (Mitchell et al. 1995; Johns et al. 1997; Meehl and Arblaster 2003; D’Arrigo et al. 2006).

REOF #9 is most heavily loaded over East China (EC pattern), particularly between the middle and lower reaches of the Yangtze and Huaihe Rivers. This loading pattern explains 3.68% of the total PDSI variance. The scores show a significant decreasing trend in moisture availability during the study period (at the 0.01 level). The moisture decrease identified here does not conflict with previous findings demonstrating an increase in precipitation in East China, since this previously observed increase was mostly found in the Yangtze River valley for the summer season (Zhai et al. 2005).

Finally, REOF #10 describes coherent moisture variability centered over Southwest China (SW pattern). This pattern has an explained variance of 3.67%. The scores for this pattern do not indicate any persistent moisture trend during the whole study period, although moisture has increased significantly since the early 1980s. The relatively wet periods in this region are found in the mid-1950s, early 1990s and around 2000, and the relatively dry periods are found in the late 1960s, early 1980s, and mid-1990s.

5.3 Moisture divisions over China and Mongolia

Many studies have investigated moisture variability across China using either geographically defined (e.g., Wang and Zhou 2005; Zhai et al. 2005; Zou et al. 2005) or temperature derived (e.g., Qian and Zhu 2001) boundaries. However, difficulties arise when applying these defined boundaries to the study of regional moisture variations over CM. First, areas within a particular geographical division may belong to different climate regimes, with different moisture-related trends. For example, the Sichuan basin is often included with other areas of Southwest China when studying moisture changes for this region (Zhai et al. 2005; Zou et al. 2005). However, as we have demonstrated, these two regions actually have contrasting trends and thus cannot be treated as one coherent division. The eastern areas of Northwest China are another example. This is a region where several climate regimes intersect, and it therefore cannot be treated as one coherent moisture division. Second, there are some neighboring areas that have similar trends in moisture, but which were previously included in separate moisture divisions, such as the defined Central China and East China in Qian and Zhu (2001). Finally, some regions, such as the FW pattern identified in the present study, have a distinct pattern of moisture variability and should be considered separately from the others.

The above observations indicate clearly that a more objective method is needed to define moisture divisions over CM. Since the ten factors identified by REOF analysis are located in different geographic areas and have distinct moisture-related patterns, we propose that moisture divisions over CM should be defined by the ten loading patterns that we have identified in this study (Fig. 8). Figure 8a show the core areas of the ten coherent moisture divisions, defined by REOF scores over 0.6. These core areas are well separated and cover regions of varying spatial extent. Areas covered by each of the ten divisions are generally defined by REOF scores over 0.4 (Fig. 8b). However, due to the opposite signs of moisture trends in neighboring divisions, some areas cannot be clearly ascribed to a particular division. These areas include eastern Northwest China, the southwestern and eastern Tibetan Plateau, and part of Southwest China. Thus, we did not ascribe these areas to any particular moisture division. A higher-resolution PDSI dataset is planned that will help clarify the boundaries of these areas in future analyses.
https://static-content.springer.com/image/art%3A10.1007%2Fs00382-008-0436-0/MediaObjects/382_2008_436_Fig8_HTML.gif
Fig. 8

The ten coherent moisture divisions across CM identified through REOF analysis. Shown in (a) are the core areas of the ten moisture divisions (defined by scores over 0.6), and in (b) are the areas covered by each of the ten moisture divisions (defined by scores over 0.4)

As expected, moisture changes in each region identified by the REOF method may demonstrate either similar or different trends to those identified by traditional ways, depending on the significance of discrepancy in boundary areas. For instance, the identified moisture changes by different division strategies are consistent in northeast China, north China, and south China, where the boundary areas identified by these strategies are similar (Qian and Zhu 2001; Zou et al. 2005). But the identified moisture changes are quite different in southwest China, northwest China, and the Tibetan Plateau, where large difference in boundary areas exists (Qian and Zhu 2001; Zou et al. 2005). In particular, an increasing trend in moisture on the Tibetan Plateau was identified by the REOF method, while no obvious trend was found by the geographically defined strategy (Fig. 9).
https://static-content.springer.com/image/art%3A10.1007%2Fs00382-008-0436-0/MediaObjects/382_2008_436_Fig9_HTML.gif
Fig. 9

Moisture trends on the Tibetan Plateau identified by the REOF method (solid line) and by the geographically defined boundary (dashed line; Zou et al. 2005). Note that the moisture trend of Zou et al. (2005) is based on changes of percentage areas in drought conditions (PDSI < −1.0)

Based on the above analyses, we propose that the REOF-based division strategy may provide better information on the identification of boundary areas as well as temporal moisture changes than using traditional ways, especially over complex regions such as CM. Such improvements on characterizing moisture trends and divisions will help understanding regional moisture variability as well as the underlying physical forcings that impact CM.

6 Concluding remarks

We have used the DAI PDSI dataset to investigate moisture variability across CM from 1951 to 2005. Our analyses indicate that the DAI PDSI performs reasonably well when used to describe moisture conditions across CM. However, we have identified some over-reporting of extreme moisture excess/deficit in our study, which could be either due to the failure of the PDSI scaling or due to a manifestation of the cumulative effects of persistent moisture excess/deficit over some areas. This over-reporting problem of the PDSI needs to be further investigated in the future studies.

As we have shown, moisture availability across CM has been highly variable during 1951–2005. Over the entire CM, mean moisture availability has decreased significantly over this period, but with considerable decadal and interannual variations (Fig. 4). Areas affected by moderate and severe moisture deficit have increased significantly since the mid-1950s. In contrast, a significant negative trend has been found for areas affected by moderate wetness since the mid-1950s, with no significant trends found for areas affected by severe wetness. As a result, the relative area affected by moderate moisture deficit or excess increased in the early half of the 1950s, decreased from the mid-1950s to 1970, and then increased significantly after 1970. The relative area affected by severe moisture deficit or excess were rather stationary during 1951–1969, but began to increase dramatically after 1970, especially for 1990–2001. Since most of CM is under the influence of the Asian Monsoon system, the above strong secular variations of moderate and/or severe moisture conditions across CM provide evidence that the Asian Monsoon system may have changed dramatically during the last several decades.

The ten most significant regional patterns of moisture variability were objectively defined for CM using REOF analysis. Among these, the NE, NC, CC, and EC patterns generally demonstrate decreasing trends in moisture availability, the NW and TP patterns show strong increases in moisture, and the other four patterns (MN, FW, SC, and SW) do not show significant moisture-related trends during 1951–2005. Based on these results, we propose that there are ten coherent moisture divisions over CM. As discussed in Sect. 5, moisture variations within each of these divisions are reasonably coherent; but can have either similar or contrasting covariability with neighboring divisions.

Our findings should prove useful for understanding moisture variations across CM, and for improving strategies for water resource management in different CM regions. However, many important questions still remain with regard to moisture variations over this complex region, including the need to evaluate the stability of these moisture divisions at different time scales and the underlying climate forcings influencing moisture variability in each division. Future work on these topics will further improve our understanding of moisture variability across CM as well as the entire region impacted by the Asian Monsoon system.

Acknowledgments

This research was funded by the National Science Foundation (Grant OCE 04-02474), the NSFC Innovation Team Project (No. 40721061), and the Program of Introducing Talents of Discipline to Universities (B06026) from China’s Ministry of Education. The authors gratefully acknowledge Mr. Paul J. Krusic for technical assistance. This is Lamont–Doherty Earth Observatory Contribution (No. 7175).

Copyright information

© Springer-Verlag 2008