蒙古高原NDVI的空间格局及分异_英文_

蒙古高原NDVI的空间格局及分异_英文_ J. Geogr. Sci. (2009) 19: 403-415 DOI: 10.1007/s11442-009-0403-7 ? 2009 Science in China Press Springer-Verlag NDVI spatial pattern and its differentiation on the Mongolian Plateau 1,2*1,3111,2ZHANG Xueyan, HU Yunfeng, ZHUANG Dafang, QI Yongqing, MA Xin 1. Institute of Geographic Sciences and Natural Resources Research, CAS, Beijing 100101, China; 2. Graduate University of Chinese Academy of Sciences, Beijing 100049, China; 3. Department of Urban Planning & Environment, Royal Institute of Technology, Stockholm 10044, Sweden Abstract: GIMMS NDVI database and geo-statistics were used to depict the spatial distribu- tion and temporal stability of NDVI on the Mongolian Plateau. The results demonstrated that: (1) Regions of interest with high NDVI indices were distributed primarily in forested moun- tainous regions of the east and the north, areas with low NDVI indices were primarily distrib- uted in the Gobi desert regions of the west and the southwest, and areas with moderate NDVI values were mainly distributed in a middle steppe strap from northwest to southeast. (2) The maximum NDVI values maintained for the past 22 years showed little variation. The average NDVI variance coefficient for the 22-year period was 15.2%. (3) NDVI distribution and vege- tation cover showed spatial autocorrelations on a global scale. NDVI patterns from the vegetation cover also demonstrated anisotropy; a higher positive spatial correlation was in- dicated in a NW-SE direction, which suggested that vegetation cover in a NW-SE direction maintained increased integrity, and vegetation assemblage was mainly distributed in the same specific direction. (4) The NDVI spatial distribution was mainly controlled by structural factors, 88.7% of the total spatial variation was influenced by structural and 11.3% by random factors. And the global autocorrelation distance was 1178 km, and the average vegetation patch length (NW-SE) to width (NE-SW) ratio was approximately 2.4:1.0. Keywords: GIMMS NDVI; spatial pattern; spatial differentiation; spatial statistics; Mongolian Plateau 1 Introduction Land-use and cover change (LUCC) is an important branch of global change research, and spatial pattern and differentiation of terrestrial vegetation is one of the fundamental interests of LUCC. An understanding of the structure, function and relevant processes of regional ecosystems through quantitative, spatial, temporal and multi-scale perspectives is vital in assessing landscape dynamics, and during the past 15 years, many studies have implemented Received: 2008-10-13 Accepted: 2009-02-10 Foundation: National Natural Science Foundation of China, No.40701150; National Science and Technology Support Project, No.07H70163AR; National Key Project of Scientific and Technical Supporting Programs, No.2006BAC08B03, No.2008BAC34B06 Author: Zhang Xueyan (1979?), Ph.D Candidate, specialized in remote sensing and GIS application research. E-mail: xyzhang@lreis.ac.cn *Corresponding author: Hu Yunfeng, E-mail: huyf@ lreis.ac.cn www.scichina.com www.springerlink.com these approaches (Zeng and Levy, 1995, Liu and Buheaosier, 2000, Wu and Hobbs, 2002, Liu et al., 2003, Weng, 2003, Aragao et al., 2005). Furthermore, scientists have recognized the value of evaluating spatial ecosystem patterns and temporal processes using data ar- chives. Rapid developments in earth observation technology and decades of remotely sensed data have enabled scientists to implement these studies more effectively (Zhou et al., 2001, Weng, 2002, Liu et al., 2003, Fang et al., 2004, Gong and Shi, 2004, Hu et al., 2008). As a land-locked highland of the Eurasian continent, the Mongolian Plateau plays an im- portant role in the meteorological and ecological systems of North China, and perhaps even the entire East Asian continent. Many studies have referred to the Mongolian Plateau as the major dust source in North China and East Asia, where dust storms are regarded as one of the most severe environmental problems impacting these regions (Ye et al., 2000, Husar et al., 2001, Natsagdorj et al., 2003, Zhang et al., 2003). Therefore, the establishment of re- gional sustainability research, particularly in relationship to terrestrial vegetation cover and its dynamic evolutionary processes is important for the Mongolian Plateau. Unfortunately, most previous studies have been limited to the Inner Mongolia region of China due to the practical difficulties of cross-national research (Zhuang et al., 1999, Hu et al., 2003, Liu et al., 2004, Hu et al., 2005). Few studies have been conducted that include the Mongolian Plateau, with little data on the physical geography, resources and environment of the entire region (Yang et al., 2004, Dill et al., 2006, Onda et al., 2007, Hu et al., 2008, Liu et al., 2008). Classical statistical approaches have served as effective tools to address ecological and LUCC research, resulting in several foundational studies (Qi and Wu, 1996, Cain et al., 1997, Perry et al., 2002, Overmars et al., 2003). However, the most important premise in classical statistics applications, i.e. the data independency hypothesis, has conflicted with the nature of geographical data. Therefore, a method to bridge classical statistics and geo-applications has not been feasible, and any results and conclusions drawn from such attempts are suspect. Consequently, a new branch of statistics, i.e. geo-statistics has been developed based on regionalized variables and spatial distribution theory that abandons the classical independency hypothesis and provides a logical approach to study the rules em- bedded in spatial data. The purpose of this study was to present a NDVI (Normalized Differential Vegetation In- dex) spatial pattern and its spatial differentiation rules for the Mongolian Plateau. The main questions posed here include: (1) Is an explicit NDVI distribution pattern indicated for the Mongolian Plateau and if so, what defines the pattern? (2) What factors control spatial pat- terns and differentiation of NDVI on the Mongolian Plateau and what dictates spatial dif- ferentiation? A derived mean annual-maximum NDVI dataset was first composed using the MVC method and validated by other data sources. Subsequently, NDVI temporal stability was assessed using 22 years of archived data to assess the feasibility of the derived dataset as representative for further study. Thirdly, the NDVI spatial pattern was generated, and the relationship between NDVI and vegetation cover was discussed. Finally, three spatial statis- tical methods were selected to explore NDVI spatial differentiation on the Mongolian Pla- teau. All together, spatial autocorrelation, factors resulting in spatial differentiation, and di- rection effects were always concentrated upon, and the geographical natures of statistical indices were demonstrated by combining with other knowledge. 405 ZHANG Xueyan et al.: NDVI spatial pattern and its differentiation on the Mongolian Plateau 2 Study region, data and methods 2.1 Study region The Mongolian Plateau is a highland of the Eurasian continent lying between 87?40′-122?15′N and 37?46′-53?08′E. It includes the entire country of Mongolia, a portion of the Russian Siberian region, and part of North China, including the Inner Mongolia and Xinjiang Uygur autonomous regions. In the present study, all of Mongolia and Inner Mon- golia Autonomous Region of China were evaluated. The Mongolian Plateau is a land-locked highland comprised of spacious plains and high mountains that shape the geomorphology of the region (Figure 1). The Greater Hinggan Mountains in the east, the Sayan and Hentiy Mountains in the north, the Altay Mountains in the west, and the Yinshan Mountains in the south define the perimeter of the plateau. In ad- dition to the mountainous regions, spacious high plains with typical steppes and deserts are characteristic components of the area. The Mongolian Plateau typically ranges between 1000–1500 m in elevation. The highest region occurs in the Altay Mountains with altitudes as high as 3000–3500 m, and the lowest region is the 700 m Hulun Buir plateau at the base of the Greater Hinggan Mountains. The Mongolian Plateau experiences a typical continental climate with low annual pre- cipitation, frequent drought, and windy episodes during the winter and spring seasons. The average temperature is ?26? in January and 17? in July. The annual average precipitationin most regions is less than 200 mm, while it may reach 400 mm or higher in eastern, north- eastern, or northern mountainous areas. These climatic conditions is a main cause of steppe vegetation, including meadow steppe, typical steppe and desert steppe. Some forest commu- nities are distributed in the eastern and northern mountains, and deserts are located in large areas of the Gobi and western and southwestern regions. Figure 2 depicts the Mongolian Plateau terrestrial vegetation cover according to the GLCF (Global Land Cover Facility) (Figure 2) (GLCF, 2008). 2.2 GIMMS NDVI NDVI can be used to coarsely evaluate the ecological status and processes operating in a region, including measurements of vegetation growth, degree of cover, and biomass, among others. Therefore, it is widely used in biogeochemical models to calculate rates of photo- synthesis, land-surface evaportranspiration, and the absorption and release of energy by the land surface. GIMMS NDVI is processed by the Global Inventory Monitoring and Modeling Studies (GIMMS) at the National Aeronautics and Space Administration (NASA). The da- tabase was launched in 1986, and now provides 22-year monthly records about terrestrial vegetation cover. New database features include reduced NDVI variation resulting from view geometry, volcanic aerosols, and other effects not directly related to actual vegetation (Tucker et al., 2005). The GIMMS NDVI dataset is considered the 3rd generation NDVI database and superior to any other existing NDVI database, such as the PAL NDVI (Path- finder AVHRR Land NDVI). The original GIMMS NDVI database had an 8-km spatial resolution and a 15-day tempo- ral resolution. The dataset is processed using the Maximum Value Composites (MVC) Figure 1 Terrain map of the Mongolian Plateau Figure 2 Land cover map of the Mongolian Plateau Annotation: E-N: Evergreen Needle-leaved; D-N: Deciduous Needle-leaved; D-B: Deciduous Broad-leaved method, which means the pixel value of each dataset represents the maximum NDVI during the last 15 days. This methodology dramatically reduces the adverse impacts caused by clouds, aerosols, view geometry, solar altitude, as well as other limiting factors. The algo- rithm of MVC approach can be depicted as the following function: NDVI= Max{ndvi} i i,j where, i=1, 2 …24; j=1, 2…15; NDVIis the ith composite value in a year; and ndviis the i i,j jth day value in the ith half-month period. 2.3 Data pre-processing and validation To improve computing space and time efficiency and decrease seasonal data undulation and 407 ZHANG Xueyan et al.: NDVI spatial pattern and its differentiation on the Mongolian Plateau long-term data redundancy, long-term NDVI spatial-temporal pattern studies are often con- ducted on an inter-annual scale. However simply using annual mean NDVI values without other appropriate pre-processing is considered a weak approach, as it may underestimate real values. Therefore, several specific rectification algorithms have been developed, including the BISE method (Best Index Slope Extraction), S-G method (Savitzky-Golay), Fourier transform method, and the wavelet method (Viovy et al., 1992, Azzali and Menenti, 2000, Li and Kafatos, 2000, Roerink et al., 2000, Lanfredi et al., 2003, Chen et al., 2004). In the present study, MVC was combined with BISE and S-G methods and applied to the half-month GIMMS NDVI database to generate reasonable annual NDVI datasets. The MVC method and corresponding annual NDVI products were finally selected as representa- tive due to the following characteristics: (1) the influence of clouds, atmosphere and solar altitude were further lowered and therefore did not underestimate real NDVI value; (2) a practical sense of agriculture, husbandry, and dust storm research identified the periods to best estimate vegetation cover; and (3) the simplest algorithm and better computing effi- ciency was available among all NDVI pre-processing programs. GIMMS NDVI databases are derived for global studies, therefore this study necessitated an initial validation to ascertain its utility for a regional study. MODIS (Moderate Resolution Imaging Spectroradiometer) NDVI was selected to validate the derived mean-annual-maximum GIMMS NDVI. Because a higher NDVI variance is typical of mountainous regions, the Greater Hinggan Mountains of the Mongolian Plateau was selected as the case study to conduct a cross validation between 8 km GIMMS NDVI and 1 km MODIS NDVI. Compared with 1 km MODIS NDVI data from 2002, the derived mean-annual-maximum GIMMS NDVI data demonstrated a general mean error less than 11.1%, and the number of pixels with an error of less than 10% accounted for 69.1% of the pixels. Therefore, the derived GIMMS NDVI dataset was reasonable and chosen for further study (Figures 3a?3d). Figure 3 Cross validation between GIMMS NDVI and MODIS NDVI 2.4 Methods NDVI distribution patterns and spatial differentiation can be investigated by a series of models derived from geo-statistics theory and relevant methodology. The models posed here include a temporal stability analysis based on the coefficient of variance index, spatial auto- correlation analysis based on Moran’s I index, semi-variation analysis, and fractal dimension analysis. All models can be implemented on a global extent, and also along different geo- graphic directions. 2.4.1 Temporal stability analysis The temporal dynamics of NDVI on the Mongolian Plateau during 1982 – 2003 were evalu- ated using a time series analysis to assess temporal stability. The NDVI coefficient of varia- tion (CV) was calculated according to the following formula: σ CV = x where CV is the coefficient of variation, σ is the standard square deviation, and x is the NDVI mean value from 1982–2003 (22-year period). Generally, a larger CV indicates a dis- crete distribution or violent fluctuations, while a smaller CV indicates a close distribution or stability. 2.4.2 Spatial autocorrelation analysis Several methods are available to measure spatial autocorrelation. Here, Moran’s I index was selected. Moran's I coefficient was represented by the following formula: n n nw( x? x )( x ? x )ij i j ?? i =1 j =1 I = n n n 2w ( x ? x ) i ij ?? ? i =1 j =1 i =1 where xand xare values of the variable x in two positions, x is the mean value of vari- i j able x, wis the neighboring weight, and n is the total number of variable x. Moran’s I coef- ij ficient ranges from –1 to 1. Generally, a higher positive Moran’s I coefficient represents a tendency toward aggregation, and similar vegetation cover is inclined to form larger assem- blages in space; while lower negative Moran’s I coefficient indicates a tendency toward fragmentation, and similar vegetation cover is inclined to scatter as smaller assemblages in space. 2.4.3 Semi-variance analysis Semi-variance analysis was applied to present integrated information regarding spatial variation and distribution. Semi-variance analysis was derived as follows: N (h)1 2 Z ( x ) ? Z ( x + h) γ (h) = []i i ? 2 N (h) i =1 where x is a spatial variant, h is the distance between xand x, and Z(x) and Z(x+ h) are i i+hii values in position xand x, respectively. In the diagram of semi-variance function, there i i+h are three key parameters: semi-variation sill (C+C), nugget variance (C), and variance 00range (a). The ratio of nugget variance (C) to sill (C+C) is used to estimate of the variance 000 409 ZHANG Xueyan et al.: NDVI spatial pattern and its differentiation on the Mongolian Plateau nature. A higher ratio indicates more random variance components (i.e. the variance is largely influenced by microclimate and local relief), while a lower ratio indicates more structural components (i.e. the variance is mainly induced by regional climatic conditions and regional geological structure). Furthermore, the parameter of variance range (a) is used 0to estimate assemblage’s size with convinced spatial autocorrelation. 2.4.4 Fractal dimension analysis Fractal theory determines distribution characteristics under different measuring scales. The fractal dimension (D) is decided by the variance function γ(h) and the separation distance h: (4?2D)2γ(h) = h A higher fractal dimension (D) is also used to characterize the variance. The higher ratio indicates more random components, while a lower ratio suggests more structural factors. The fractal dimension, semi-variation sill (C+C), and nugget variance (C) possess a certain 00 relationship with each other. When the ratio of semi-variation sill (C+C) to nugget variance 0 (C) is higher, the fractal dimension is greater. Both of these measures depict the same char- 0 acteristics from different forms of variance. In this study, the NDVI pre-handling works were implemented by AML (Arc/Info Macro Language) programming in the Arc/Info Workstation platform; the spatial differentiation +modeling works were supported by GS7.0 package; and the interpretation and analysis of spatial distribution patterns and spatial differentiation were carried out in the ArcMap envi- ronment. 3 Results and analysis 3.1 Distribution pattern of NDVI and LUCC on the Mongolian Plateau The mean value from the annual-maximum-NDVI during 1982–2003 showed clear spatial differentiation for the Mongolian Plateau (Figure 4). The following trends were observed: pixels with high NDVI values were distributed in the mountainous forest regions of the east and north, pixels with low NDVI values were distributed in the Gobi desert regions of the west and southwest, and pixels with moderate NDVI values were distributed in a middle steppe strap from the northwest to the southeast. The NDVI distribution patterns were closely linked with vegetation cover. According to the GLCF vegetation cover map (Figure 2), the detailed relationship between NDVI and LUCC was consistent with our previous research and was depicted as follows (Hu et al., 2008). Due to high precipitation and low evaporation, the eastern (Greater Hinggan Mountains and Hulun Buir high plain) and northern (Sayan Mountains, Hentiy Mountains, and Hangay Mountains) Mongolian Plateau was characterized as forest, forest steppe and meadow steppe. Correspondingly, the NDVI in these regions (>0.6) was higher than that in other regions. In the western Mongolian Plateau (Altay Mountains and Gobi Desert), the main terrestrial land cover types were desert steppe, steppe desert, and Gobi desert, supported by low NDVI in- dices (<0.4). In the transitional regions (from the southeast including Xinlin Gol plateau and Ulan Qab plateau to the northwest including Dornogovi, Dundgovi, Ovorhangay, Arhangay, and Dzavhan), typical steppe zone or agriculture–pasture ecotones were depicted, and the NDVI in these transitional areas was moderate (between 0.4–0.6). 3.2 Temporal stability analysis Over the past 22 years, the average NDVI variance coefficient for the entire Mongolian Pla- teau was 15.2%. The number of pixels with a C.V. less than 15% accounted for 53% of the pixels, and the pixels with C.V. less than 30% accounted for 97% of the data. Moreover, the temporal variance was low in the high NDVI regions (i.e. the Greater Hinggan Mountains, Hentiy Mountains, Sayan Mountains, and Hangay Mountains regions with forest, forest steppe, and meadow steppe cover) or high in low NDVI regions (i.e. Gobi–Altay Mountains and Desert regions with desert and Gobi cover), and high in typical steppe regions with moderate NDVI values (i.e. in the transitional zone between the east and west). Figure 5 demonstrated the detail pattern of coefficient of variance for the annual-maximum NDVI Figure 4 NDVI mean-annual-maximum spatial patterns for the Mongolian Plateau during 1982–2003 Figure 5 Coefficient of variance for the annual-maximum NDVI spatial pattern during 1982–2003 411 ZHANG Xueyan et al.: NDVI spatial pattern and its differentiation on the Mongolian Plateau during 1982–2003. Temporal analysis indicated a low variance for the annual-maximum NDVI during the past 22 years, suggesting vegetative cover stability. Therefore, the mean-annual-maximum NDVI dataset is likely representative of the Mongolian Plateau vegetation cover, and is suitable for further spatial differentiation analysis. 3.3 Spatial differentiation analysis Spatial autocorrelation analysis 3.3.1 Results of the Moran’s I coefficient at the global scale showed a positive value (0.21) (Fig- ure 6a and Table 1), indicating a strong spatial autocorrelation for NDVI and the terrestrial vegetation cover distribution in the Mongolian Plateau. The pixels with similar NDVI values tended to coalesce, and the terrestrial vegetative cover of the Mongolian Plateau was intact with negligible fragmentation at this scale. These results were consistent with the distribu- tion patterns observed for terrestrial vegetation cover. According to the land cover map (Figure 2), the vegetation cover of the Mongolian Plateau can be divided into large vegeta- tion assemblages with clear boundaries, including forest assemblages in the mountainous regions, waste land assemblages in the Gobi desert regions, and large areas of typical steppe assemblages on the plateau. Further analysis along different geographic directions also showed that the NDVI pattern and vegetation cover has clear anisotropy (Figures 6b–6e and Table 1). A higher positive spatial correlation was revealed in the NW–SE direction (Moran’s I coefficient = 0.41), fol- lowed by the W–E direction, while a negative spatial correlation was indicated in the N–S and NE–SW directions. The higher positive spatial correlation for the NW–SW direction indicated that vegetation cover in this orientation maintained better integrity and vegetation assemblages. Such results are reasonable when compared with the land cover map. Most terrestrial vegetation cover for the Mongolian Plateau was distributed as strips or assem- blages along a NW–SW orientation. From E–W the vegetation followed a water/moisture availability gradient progressing from forest, meadow steppe, typical steppe, to Gobi and desert. 3.3.2 Semi-variance analysis The semi-variance analysis generated a variance range (a) of approximately 1178 km on a 0 global scale (Figure 6f and Table 1), which supported a vegetation autocorrelation with a correlation radius equal to 1178 km. Furthermore, the ratio of nugget (C) to sill (C+C) was 00small (0.113), which indicated that random factors have less impact on NDVI spatial distri- bution patterns, while structural factors have the greatest influence. The proportion of the spatial variance induced by random factors accounted for 11.3% of the total with structural factors comprising 88.7% of the variance. Structural factors included broad climatic regimes, regional topography, and geological structure, while random factors are likely influenced by micro-climate, local relief, and human activities. Furthermore, detailed calculations along different geographic directions were developed (Figures 6g–6i and Table 1). The ratios of nugget (C) to sill (C+C) were less than 2%, 00 suggesting that the spatial variance induced by random factors was weaker and structural factors prevailed in all different geographic directions. In addition, variance range (a) 0analysis was the largest (3183 km) along a NW–SE direction. These results demonstrated Figure 6 Moran’s I coefficient, semi-variance and fractal dimension model Table 1 Semi-variance function index and Moran’s I coefficient 32 3Extent or direction a(km) C/sill Fitting model Moran’s I coefficient ) rSill (×10C(×10) 0 00 Global model 0.21 8 72 1178 0.113 0.998 Gaussian N–S 2 100 1259 0.018 0.860 Spherical model –0.0042 NE–SW Spherical model 2 111 1314 0.016 0.882 –0.038 W–E 0 119 2916 0.001 0.865 Exponential model 0.30 NW–SE 0 135 3181 0.001 0.864 Exponential model 0.41 413 ZHANG Xueyan et al.: NDVI spatial pattern and its differentiation on the Mongolian Plateau that mainly impacted by structural factors, the terrestrial vegetation cover on the Mongolian Plateau is distributed in a NW–SE direction, and the average length (NW–SE) to width (NE–SW) ratio for the vegetation assemblages was approximately 2.4:1.0. 3.3.3 Fractal dimension analysis Fractal dimension analysis generated a NDVI global fractal dimension value of 1.573 for the Mongolian Plateau and the fractal dimension along different directions ranged from 1.474 to 1.723 (Figures 6k–6o). The fractal dimension along a NE–SW direction was small, which determined that the vegetation cover along this direction was fragmented and the spatial variance was mainly caused by random factors. The fractal dimension in a NW–SE direction was large, indicating that the vegetation cover along this direction was intact and the spatial variance resulted from structural factors. The analyses based on semi-variance and fractal dimension were congruent in terms of NDVI spatial distributions, and explained by the same physical and geographical influences. 4 Conclusions and discussion 4.1 Conclusions The results of this study provided a depiction of NDVI spatial distribution and temporal sta- bility on the Mongolian Plateau based on GIMMS NDVI data analysis. Furthermore, a de- tailed analysis of spatial variability derived from Moran’s I coefficient, semi-variance func- tion and fractal dimension was developed to explore NDVI spatial differentiation. The re- sults were as follows: (1) The derived mean-annual-maximum GIMMS NDVI dataset exhibited less error, which was indicated by validation analysis from 1 km MODIS NDVI data in Greater Hing- gan Mountains of the Mongolian Plateau in 2002. Temporal analysis also verified that the maximum NDVI values during the past 22 years showed little variation i.e. values remained stable. Both analyses supported the derived mean-annual-maximum GIMMS NDVI dataset as robust for further study. (2) Based on these derived datasets, the basic NDVI distribution patterns were investi- gated in detail, and a close relationship between NDVI and terrestrial vegetative cover was demonstrated. Regions of interest with high NDVI values were distributed primarily in for- ested mountainous regions of the east and north, areas with low NDVI indices were primar- ily distributed in the Gobi desert regions of the west and southwest, and areas with moderate NDVI indices were mainly distributed in a middle steppe strap from northwest to southeast. The maximum NDVI values maintained for the past 22 years showed little variation. The average NDVI variance coefficient for the 22-year period was 15.2%. Furthermore, the temporal variance was low in both the high (forests) and low (deserts) index regions, while it was high in steppe regions that demonstrated moderate NDVI values. (3) NDVI distribution and vegetation cover showed spatial autocorrelations on a global scale. Pixels with similar NDVI values showed a tendency to coalesce, and vegetation cover was generally intact for the Mongolian Plateau. NDVI patterns from the relevant vegetation cover also demonstrated anisotropy; a higher positive spatial correlation was indicated in a NW–SE direction, which suggested that vegetation cover distributed in a NW–SE direction maintained increased integrity, i.e. vegetation assemblages were mainly distributed in this specific direction. (4) Based on semi-variance and fractal dimension analysis, the factors controlling spatial variance on the Mongolian Plateau were investigated. Although simultaneously affected by both structural and random factors, the NDVI spatial distribution was mainly controlled by structural factors; 88.7% of the total spatial variation was influenced by structural and 11.3% by random factors. (5) Spatial autocorrelation analysis from both semi-variance and fractal dimension analy- ses showed anisotropy in NDVI distribution on the Mongolian Plateau. Further analysis in- dicated a global autocorrelation distance of 1178 km, and an average length (NW–SE) to width (NE–SW) ratio for the vegetation assemblages was approximately 2.4:1.0. 4.2 Limitations and future research The results of this research provided evidence that the derived mean-annual-maximum NDVI can serve as a reliable representative dataset for future studies, and spatial statistics can characterize the physical attributes of spatial variants. However, the following limita- tions were noted: (1) Although the dataset derived from this study was determined acceptable, it is impor- tant to stress that the purpose of this study was to elucidate long-term spatial distribution and differentiation patterns, and not transient patterns on an annual basis. (2) Spatial statistics and spatial analysis were both characterized, including spatial scale and temporal scale effects. However, an increase in detailed distribution and change pattern detection could be obtained if datasets with higher spatial and/or temporal resolution are applied in the future, although the basic patterns and principles would remain largely unal- tered. References Aragao L, Shimabukuro Y, Santo F et al., 2005. Landscape pattern and spatial variability of leaf area index in eastern Amazonia. Forest Ecology and Management, 211(3): 240–256. Azzali S, Menenti M, 2000. Mapping vegetation–soil–climate complexes in southern Africa using temporal Fou- rier analysis of NOAA–AVHRR NDVI data. International Journal of Remote Sensing, 21(5): 973–996. Cain D H, Riitters K, Orvis K, 1997. A multi-scale analysis of landscape statistics. Landsc. Ecol., 12(4): 199–212. Chen J, Jonsson P, Tamura M et al., 2004. A simple method for reconstructing a high-quality NDVI time-series data set based on the Savitzky–Golay filter. Remote Sensing of Environment, 91(3/4):332–344. Dill HG, Khishigsuren S, Majigsuren Y et al., 2006. Geomorphological studies along a transect from the taiga to the desert in central Mongolia: Evolution of landforms in the mid-latitude continental interior as a function of climate and vegetation. J. Asian Earth Sci., 27(2): 241–264. Fang J Y, Piao S L, He J S et al., 2004. Increasing terrestrial vegetation activity in China, 1982–1999. Science in China (Series C), 47(3): 229–240. (in Chinese) GLCF, 2008. GLCF: AVHRR global land cover classification. Gong D, Shi P, 2004. Inter-annual changes in Eurasian continent NDVI and its sensitivity to the large-scale cli- mate variations in the last 20 years. Acta Botanica Sinica, 46(2): 186–193. (in Chinese) Hu Y F, Ban Y F, Zhang Q et al., 2008. Spatial-temporal pattern of gimms NDVI and its dynamics in Mongolian Plateau. In: Earth Observation and Remote Sensing Applications, 2008, Beijing: IEEE, 1–6. Hu Y F, Liu J Y, Zhuang D F et al., 2003. Study on the relationship between land-use dynamics and wind erosion dynamics in Inner Mongolia during late 1990s. Progress in Geography, 22(6): 541–550. (in Chinese) Hu Y F, Liu J Y, Zhuang D F et al., 2005. Distribution characteristics of (cs)-c-137 in wind-eroded soil profile and 415 ZHANG Xueyan et al.: NDVI spatial pattern and its differentiation on the Mongolian Plateau its use in estimating wind erosion modulus. Chinese Science Bulletin, 50(11): 1155–1159. (in Chinese) Husar R B, Tratt D M, Schichtel B A et al., 2001. Asian dust events of April 1998. Journal of Geophysical Re- search-Atmospheres, 106(D16): 18317–18330. Lanfredi M, Lasaponara R, Simoniello T et al., 2003. Multiresolution spatial characterization of land degradation phenomena in southern Italy from 1985 to 1999 using NOAA–AVHRR NDVI data. Geophysical Research Letters, 30(2): 1069–1081. Li Z T, Kafatos M, 2000. Interannual variability of vegetation in the United States and its relation to El Nino/Southern Oscillation. Remote Sensing of Environment, 71(3): 239–247. Liu J Y, Buheaosier, 2000. Study on spatial-temporal feature of modern land-use change in China: Using remote sensing techniques. Quaternary Sciences, 20(3): 229–239. (in Chinese) Liu J Y, Liu M L, Zhuang D F et al., 2003. Study on spatial pattern of land-use change in China during 1995–2000. Science in China (Series D), 46(4): 373–384. (in Chinese) Liu J Y, Qi Y Q, Shi H D et al., 2008. Estimation of wind erosion rates by using Cs-137 tracing technique: A case study in Tariat–Xilin Gol transect, Mongolian Plateau. Chinese Science Bulletin, 53(5): 751–758. (in Chinese) Liu J Y, Wang S Q, Chen J M et al., 2004. Storages of soil organic carbon and nitrogen and land use changes in China: 1990–2000. Acta Geographica Sinica, 59(4): 483–496. (in Chinese) Natsagdorj L, Jugder D, Chung Y S, 2003. Analysis of dust storms observed in Mongolia during 1937–1999. Atmos. Environ., 37(9/10): 1401–1411. Onda Y, Kato H, Tanaka Y et al., 2007. Analysis of runoff generation and soil erosion processes by using envi- ronmental radionuclides in semiarid areas of Mongolia. J. Hydrol., 333(1): 124–132. Overmars K P, de Koning G H J, Veldkamp A, 2003. Spatial autocorrelation in multi-scale land use models. Eco- logical Modeling, 164(2/3): 257–270. Perry J N, Liebhold A M, Rosenberg M S et al., 2002. Illustrations and guidelines for selecting statistical methods for quantifying spatial pattern in ecological data. Ecography, 25 (5): 578–600. Qi Y, Wu J G, 1996. Effects of changing spatial resolution on the results of landscape pattern analysis using spatial autocorrelation indices. Landsc. Ecol., 11(1): 39–49. Roerink G J, Menenti M, Verhoef W, 2000. Reconstructing cloudfree NDVI composites using Fourier analysis of time series. International Journal of Remote Sensing, 21(9): 1911–1917. Tucker C J, Pinzon J E, Brown M E et al., 2005. An extended AVHRR 8-km NDVI data set compatible with modis and spot vegetation NDVI data. International Journal of Remote Sensing, 26(20): 4485–4498. Viovy N, Arino O, Belward A S, 1992. The best index slope extraction (BISE): A method for reducing noise in NDVI time-series. International Journal of Remote Sensing, 13(8): 1585–1590. Weng Q H, 2002. Land use change analysis in the Zhujiang Delta of China using satellite remote sensing, GIS and stochastic modeling. J. Environ. Manage., 64(3): 273–284. Weng Q H, 2003. Fractal analysis of satellite-detected urban heat island effect. Photogram. Eng. Remote Sens., 69(5): 555–566. Wu J G, Hobbs R, 2002. Key issues and research priorities in landscape ecology: An idiosyncratic synthesis. Landsc. Ecol., 17(4): 355–365. Yang X P, Rost K T, Lehmkuhl F et al., 2004. The evolution of dry lands in northern China and in the Republic of Mongolia since the last glacial maximum. Quatern Int., 118/119: 69–85. Ye D, Chou J, Liu J et al., 2000. Causes of sand-stormy weather in northern China and contral measures. Acta Geographica Sinica, 55(5): 513–521. (in Chinese) Zeng L X, Levy G, 1995. Space and time aliasing structure in monthly mean polar-orbiting satellite data. Journal of Geophysical Research-Atmospheres, 100(D3): 5133–5142. Zhang X Y, Gong S L, Zhao T L et al., 2003. Sources of Asian dust and role of climate change versus desertifica- tion in Asian dust emission. Geophys Res. Lett., 30(24): 2272. (in Chinese) Zhou L M, Tucker C J, Kaufmann R K et al., 2001. Variations in northern vegetation activity inferred from satel- lite data of vegetation index during 1981 to 1999. Journal of Geophysical Research-Atmospheres, 106(D17): 20069–20083. Zhuang D F, Liu J Y, Liu M L, 1999. Research activities on land-use/cover change in the past ten years in China using space technology. Chinese Geographical Science, 9(4): 330–334. (in Chinese)