Landscape Spatial Patterns in the Maowusu (Mu Us) Sandy Land, Northern China and Their Impact Factors

Home , Mu Us Desert, Taklamakan Desert, Thar Desert

Catena 145 (2016) 321–333

Contents lists available at ScienceDirect

Catena

journal homepage: www.elsevier.com/locate/catena

Landscape spatial patterns in the Maowusu (Mu Us) Sandy Land, northern China and their impact factors

Peng Liang a,b, Xiaoping Yang a,c,⁎ a Key Laboratory of Cenozoic Geology and Environment, Institute of Geology and Geophysics, Chinese Academy of Sciences, P.O. Box 9825, Beijing 100029,China b University of Chinese Academy of Sciences, Beijing 100049, China c Geographisches Institut, Universität Heidelberg, Im Neuenheimer Feld 348, 69120 Heidelberg, Germany article info abstract

Article history: In environmentally sensitive areas, particularly the arid or semi-arid regions, different factors such as relief and Received 9 September 2015 human activity, can greatly influence the landscape on a local scale. Using satellite imageries and field investiga- Received in revised form 11 June 2016 tions, we have studied the distribution of sand dunes in the Maowusu (Mu Us) Sandy Land in the southern Ordos Accepted 14 June 2016 Plateau, northern China. Key research questions are which factors are critically responsible for the spatial pattern Available online xxxx evident in the landscape and what is the relative importance of such factors? We used a new approach, i.e., the Geographical Detector to improve our spatial analysis. Using the combined methods of the Geographical Detec- Keywords: fi Landscape evolution tor, digital image processing, eld investigation and spatial and temporal analysis of the Normalized Difference Spatial analysis Vegetation Index (NDVI), we aimed to clarify the causal relations governing the different landform patterns with- Human impact in this Sandy Land and beyond. The primary factors include local relief, drainage, climatically controlled potential Desertification plant productivity, wind action and rock type as explanatory parameters x and the landscape properties as the Geographical Detector output y. Our results show the quantified links between the spatial distribution of various landscapes and their controlling factors. This study revealed that the climatic potential productivity, local relief and drainage are the key factors in shaping the landscape spatial pattern. Owing to different regional climates and land use histories, the responses to changes of external forcing between the western and eastern parts of the Maowusu Sandy Land are out-of-step. We conclude that the desertification of the eastern part of Maowusu Sandy Land has been mainly caused by human activities while the western part is governed by natural factors. © 2016 Elsevier B.V. All rights reserved.

1. Introduction Karnieli et al., 2014) and palaeoenvironmental responses to climate change over longer time scales (Stevens et al., 2013; Lancaster et al., Global landscapes show a high degree of spatial variation 2015). These studies emphasized temporal changes but paid little atten- (Slaymaker and Spencer, 1998; Blümel, 2013), reflecting the interplay tion to landscape spatial patterns or to different impact factors on a re- of diverse factors such as climate, topography and human intervention gional scale. But without a clear understanding of the spatial (Bürgi et al., 2004). Arid and semi-arid landscapes have been the focus heterogeneity or the geomorphological context within a study area, of considerable and varied research (Zhu et al., 1980; Wiggs et al., the representativeness of sedimentary sections is questionable (Huang 1995; Forman et al., 2001; Goudie, 2002; Eitel et al., 2005; Unkel et al., et al., 2009) and thus contradictory conclusions may be derived from 2007; Livingstone et al., 2010; Singhvi et al., 2010; Hugenholtz et al., various profiles within a small catchment. 2012; Yang et al., 2012, 2013; Williams, 2014, 2015, 2016). For example, Also, the desertification, which is believed to be caused by various Yang et al. (2012) considered precipitation, wind speed and geomor- factors, including climatic variations and human activities (Williams phic context as factors controlling the large-scale occurrence of deserts and Balling, 1996; Rasmussen et al., 2001), is a major environmental in northern China. In regard to desert dunes, Wiggs et al. (1995) sug- issue in drylands, and threatens socio-economic development and gested that surface activity mainly occurs on the crests and upper slopes human life (Yang et al., 2007; Williams, 2014). In the past thirty years, of the dunes due to difference in wind energy. the availability of remote sensing data has provided new opportunities Earlier research in the Maowusu Sandy Land of northern China has to investigate the desertification in drylands and stimulated a myriad of focused on two key questions, i.e., desertification or land cover change studies concerning drylands with respect to landscape change over time during historical times (Zhu and Liu, 1981; Runnström, 2003; Zhang (e.g., Runnström, 2003; Yang et al., 2007; Bayarsaikhan et al., 2009; et al., 2003; Deng et al., 2007; Mason et al., 2008; Huang et al., 2009; Karnieli et al., 2014; Li and Yang, 2014; Scuderi et al., 2015). Although some researchers attempted to use the landscape change detection ⁎ Corresponding author. methods to seek the predominant driving forces for the landscape E-mail addresses: [email protected], [email protected] (X. Yang). change, including desertification in drylands (e.g., Runnström, 2003;

Zhang et al., 2003), it is still problematic to evaluate the inter-relation- (Fig. 1a). Due to the reducing influence of the East Asian Summer Mon- ships between desertification and climate change, and human activities soon from southeast to northwest, the mean annual precipitation in our because of the difficulties and uncertainties of handling data of different study area decreases from 400 to 450 mm in the southeast to ca. qualities. 250 mm in the northwest (Zhu et al., 1980; Mason et al., 2008; Fig. Some studies on sand seas have offered insights by using satellite re- 1d), and is characterized with a high annual variability (Zhu et al., mote sensing data to aid the interpretation and classification of aeolian 1980; Karnieli et al., 2014). Summer (from June to September) rainfall dune landscapes (Beveridge et al., 2006; Livingstone et al., 2010)andto accounts for as much as 70% of the total annual precipitation (Domrös explore the dune-field patterns through the pattern analysis of dune- and Peng, 1988; Karnieli et al., 2014; Fig. 1d). Spring (from March to field parameters under different wind regimes or environmental condi- May) is characterized by moisture deficit in the study area, particularly tions (Ewing and Kocurek, 2010) from a spatial perspective. Unlike ac- in the western part (as illustrated by the climatic diagram of Walter and tive sand seas which are composed of various sand dunes, only sandy Lieth (1960) in Fig. 1d), and stronger winds compared with other patches or small dune fields occurred in the sandy land areas such as months (Liu et al., 2005). the Maowusu (Zhu et al., 1980). Some earlier studies have recognized the diversity of the landscapes on a regional scale (Deng et al., 2007; 3. Materials and methods Zuo et al., 2012; Karnieli et al., 2014). However, many questions still re- main unsolved such as the relationships between impact factors and 3.1. Datasets landscape spatial patterns, particularly on different spatial scales. By combining remote sensing, spatial analysis and geographical in- Taking vegetation cover, image quality and weather conditions into formation system (GIS) methods, especially a new approach, i.e., the account, we selected four sets of contiguous images from the Operation- Geographical Detector, this paper aims to demonstrate the landscape al Land Imager (OLI) images of Landsat 8 (data sources: http:// spatial patterns on various spatial scales and to reveal the relationships earthexplorer.usgs.gov/) to interpret the landscape thematic map (y). between the spatial distribution of various sandy landscapes and their The mean cloudiness of these images in our study domain is b6% triggering factors quantitatively and systematically. Furthermore, we (Supp.-Table 1). We used the satellite images captured in early August combine the NDVI spatiotemporal patterns to explore the relative 2013 (Supp.-Table 1), approximately consistent with the time of our roles of human activities and natural factors in desertification of the field investigation, to compile the landscape thematic map. In addition, Maowusu Sandy Land from a spatial perspective. the surface greenness of the study area reflected by NDVI reaches its maximum value in August (Mason et al., 2008; Karnieli et al., 2014). It 2. Regional setting is beneficial to identify and extract bare sand dune through the Landsat OLI images when the vegetation cover is most complete. Bearing this in Located in the transitional zone between agricultural farming and mind, we used the images of early August for maximum likelihood su- animal grazing in the southern Ordos Plateau, the Maowusu Sandy pervised classification in an effort to identify the landscape spatial pat- Land (Mu Us Desert) is bordered by the Kubuqi (Hobq) Desert in the terns. Ground information was collected during our field trip for the north, and the Yellow River in the east and west (Fig. 1a). The southern purpose of visual interpretation and classification accuracy assessment. boundary of the Maowusu is roughly consistent with the Great Wall We selected climatic potential productivity, wind speed, river net- built during the Chinese Ming Dynasty (A.D. 1368–1644), which also works, bedrock types and local relief as natural environment factors marks the boundary between sandy landscape and the Loess Plateau (x) for the causal relationship analysis. Climatic potential productivity (Deng et al., 2007). It is one of the twelve major dune fields in north and mean annual wind speed data were acquired from grid products China and belongs to ‘sandy lands’ (Yang et al., 2012; Fig. 1a). Our provided by the Data Sharing Union of Earth System Science (http:// study region is ~34,500 km2 in area, and its longitude and latitude www.geodata.cn). The spatial resolution of these two datasets is 1 km. range from 107.49°E to 110.20°E and 37.48°Nto39.56°N, respectively. The climatic potential productivity was defined as the result of interac- The Quaternary deposits (such as aeolian sand, alluvium and lacustrine tion between local solar radiation, temperature, and precipitation sediments), Neogene sandstones and Cretaceous glutenites crop out in (Huang, 1985; Huang et al., 1999), and it was estimated using: the Maowusu Sandy Land (Ma et al., 2004; Fig. 2). The topography of the study area is slightly undulating from south- Pc ¼ 21:9Q ðÞN=365 fwðÞ west to northeast, and the elevation increases gradually from 950 m in 2 the southeast to 1600 m in the northwest (Fig. 1b), resulting in the ma- Where Pc is the climatic potential productivity (kg/hm ·a); 21.9 is an jority of the rivers in this region flowing from northwest to southeast. experimental coefficient in China based on Huang (1985); Q = mean Geomorphic features of this sandy land include semi-fixed dunes, annual solar radiation, MJ/(m2 a); N is mean annual total days of frost- fixed dunes and well-vegetated inter-dune wetlands. Active dunes, free period; f (w) is a function of precipitation and temperature (if lakes and river flood plains are also common in this region (Zhu et al., P ≤ E0, f(w)=P/E0;ifP N E0, f(w)=1); P is the annual total precipitation 1980). Parabolic dunes, barchans and barchan chains reaching a maxi- and E0 is the annual total evaporation based on Penman Equation mal height of 20 to 30 m occur in the dune field (Yang et al., 2012). Riv- (Penman, 1948). This climatic potential productivity dataset is a kilome- ers and the diversity of the landscapes are the most remarkable features ter grid product that we calculated from the impact factors and is of this sandy land. Calligonum and Agriophyllum are the predominant expressed as kg/hm2 a. plants growing on the stable dunes and at the foot of semi-fixed We extracted the mean annual wind speed from the national mean dunes. Artemisia and shrubs, including Sanina, Tamarix, Hippophae and annual wind speed dataset of China from 1951 to 2000. This dataset is Salix, are found in the inter-dune lowlands. Also, farmland is widely dis- an interpolation product of the observed meteorological data from tributed in the riparian areas and inter-dune lowlands, with maize the 743 weather stations in China (1951–2000) based on the Wieringa dominant crop. The groundwater table of the study area is shallow, es- model (Wieringa, 1986). The unit of this grid dataset is 0.1 m/s. Since pecially in the inter-dune areas, often just 1–3 m below the ground the highest wind energy controlling the sand dune movement occurs al- (Ohte et al., 2003; Karnieli et al., 2014). most always in spring (from March to May) in the Maowusu Sandy Land The study area is classified as the Bsk (arid-steppe-cold) climate type (Liu et al., 2005), we selected the mean annual wind speed of these on the basis of Köppen–Geiger climate classification (Köppen, 1931; three months as our input data for analysis. Peel et al., 2007) and belongs to the semi-arid type of the Middle Tem- River networks, including perennial rivers and intermittent streams, perature Zone (Domrös and Peng, 1988). Based on the calculation of the were extracted from the images of Google Earth and validated by our aridity index (k) (Ren, 1999), the Maowusu is characterized by 2 b k b 5 field investigation. Bedrock types were acquired by digitizing the P. Liang, X. Yang / Catena 145 (2016) 321–333 323

Fig. 1. Overview of the study area. (a) Distribution of aridity index (modiﬁed from Liu, 2010) in northern China and location of the Maowusu (Mu Us) Sandy Land. (b) ASTER DEM map of the study area. (c) ETM+ image of the selected region (2002, band combination RGB = 7, 4, 1), original data source from Global Land Cover Facility (www.landcover.org). (d) Walter-Lieth climate diagrams (following Walter and Lieth, 1960) of Yanchi, Etuoke Qi and Yulin (underlined in panels a and b) weather stations. 324 P. Liang, X. Yang / Catena 145 (2016) 321–333

Fig. 2. Potential driving factors of landscape spatial patterns. (a) Climatic potential productivity (CPP). (b) Mean annual wind speed (MWS) from 1951 to 2000. (c) Buffer region of river systems. (d) Geological map. geological map of Shaanxi and Inner Mongolia from the Geological Atlas Wang and Hu, 2012). Based on our field investigation, the buffer region of China at the scale of 1:2,500,000 (Ma et al., 2004; Fig. 2d). Topograph- of rivers in terms of the Euclidean distance to river channel were carried ic data were acquired from the Advanced Spaceborne Thermal Emission out on the ArcGIS® for Desktop 10.3 platform with ‘buffer’ tool (ESRI, and Reflection Radiometer Digital Elevation Model product ASTER Glob- 2015; Fig. 2c). Since the landform change of the target region is so gen- al Digital Elevation Model (ASTER GDEM) version 2 downloaded from tle, the topographic layers have been classified into thirty levels rather GDEM official website (http://gdem.ersdac.jspacesystems.or.jp/), with than seven as other factors by using the quantile method to try to take a horizontal spatial resolution of 1 arc-second (approximately 30 m at small topographic features into account (Supp.-Table 8). the equator) (Tachikawa et al., 2011; Fig. 1b). All of the data have been projected to the WGS-1984 Universal Transverse Mercator zone 3.3. Land cover classification 49 N (Fig. 2). The GIMMS NDVI dataset (Tucker et al., 2004) from 1981 to 2006 In order to obtain the response variable (y), we applied the Landsat 8 over the interval July 1st to 15th and MODIS Vegetation Index Products OLI images for maximum likelihood (ML) supervised classification (MOD13C1) from 2005 to 2013 over the interval July 1st to 15th used in (Strahler, 1980; Lillesand et al., 2004; Campbell and Wynne, 2011; this study were obtained from the Global Land Cover Facility (www. Richards, 2013). The Google Earth images, field notes and photos were landcover.org) and MODIS website (http://modis.gsfc.nasa.gov/), re- combined to verify the classification. After many experiments, band 4 spectively. The former provides global NDVI values at 0.07272727° (8- (red), band 3 (green) and band 2 (blue) of the Landsat OLI images km) resolution, derived from Advanced Very High Resolution Radiome- were selected to overlay a natural color map (Supp.-Table 1). The ML su- ter (AVHRR) images. The latter is cloud-free NDVI product at 16-day pervised classification method provided by ERDAS IMAGINE® 9.2 pro- 0.05° (~5.6-km) resolution from NASA's Moderate Resolution Imaging gram was used to interpret the mosaic image and eliminated the Spectrometers (MODIS) on the Terra platforms (Tucker et al., 2005). fragments no more than the threshold with four pixels. We went through several process stages, including layers stack, histogram 3.2. Potential impact factors data preprocessing matching, mosaic, set projection and clip (Lillesand et al., 2004) before the supervised classification. Using a small set of basic training areas All of the continuous impact factors x except the topographic factor, which are primarily dependent on our visual interpretation according i.e., climatic potential productivity and wind condition, have been to the characters of ground objects and our field observations reclassified into seven levels (Fig. 2a and b) with the quantile method (Campbell and Wynne, 2011; Richards, 2013), we classified the land- (Cao et al., 2013; ESRI, 2015) provided by ArcGIS® for Desktop 10.3 be- scapes into five types, i.e., active dunes, semi-fixed dunes, shrub or crop- cause the GeogDetector needs discrete variables (Wang et al., 2010; lands, meadow or steppe, and water bodies, following the Land Cover P. Liang, X. Yang / Catena 145 (2016) 321–333 325

Classiﬁcation System (LCCS) of FAO/UNEP (Di Gregorio and Jansen, org/GeogDetector/). In this study, we put the Sand Cover Ratio calculat- 2000). By using this classiﬁed result, we compiled a landscape spatial ed from landscape thematic map (based on Eq. (1))and patterns map. abovementioned potential impact factors, i.e., climatic potential productivity, local relief, river buffer, wind speed and rock type as input data to 3.4. Sand Cover Ratio this model for our analysis.

For the sake of quantifying analysis conveniently, our study 3.6. Spatiotemporal analysis of NDVI binarized the landscape spatial patterns by applying the ‘reclassify’ package provided by ArcGIS® for Desktop 10.3 (ESRI, 2015). Through In order to enhance our understanding of the spatial patterns of this binarization processing, we classified the landscape into only two human activities, we analyzed spatiotemporal changes of the NDVI types, i.e., sand zone and vegetation zone, based on the degree of vege- from 1981 to 2013, pixel-by-pixel, using the Mann-Kendall test tation coverage (Fig. 5a). More specifically, its value equals zero when (Mann, 1945; Kendall, 1970). We used a similar navigation procedure the type of landscape are active dunes or semi-fixed dunes, and one oth- as in Mason et al. (2008) and Li and Yang (2014). We prolonged the erwise. And then, a square net at a length of 1 km was used to overlay time window from 2003 to 2013 following Mason et al. (2008).In the entire study area. Grid standard data were generated and then we order to fit the MODIS product's resolution to the GIMMS dataset (the calculated the Sand Cover Ratio (Fig. 5b) as follows: MODIS resolution is ~5.6 km while the GIMMS resolution is 8 km) as well as to compare these two dataset derived results expediently, the Sand dune area km2 MODIS NDVI data were resampled to 8 km × 8 km pixels using the Sand Cover Ratio ¼ 100% ð1Þ nearest neighbor approach with ArcGIS® for Desktop 10.3 software 2 1km (ESRI, 2015). The Mann–Kendall test was carried out with the ‘Kendall’ package in the R (RCoreTeam,2016). Where the denominator 1 (km2) is the area of the one net and the sand dune area (km2) is the area of sand dunes in each net. The details 4. Results of calculating the Sand Cover Ratio are illustrated in Fig. 5a. 4.1. Structure of landscape types 3.5. Modeling using Geographical Detector The landscape thematic map interpreted from Landsat 8 OLI The Geographical Detector (GeogDetector) was developed for ex- images is presented in Fig. 4 and the Sand Cover Ratio map is ploring the relationships between landscape spatial patterns and their presented in Fig. 5b. According to our field investigation and the impact factors (Wang et al., 2010; Wang and Hu, 2012). It is based on ERDAS IMAGINE® error matrix (Congalton, 1991), the classification spatial variance analysis, and its strength is in analyzing mixed data accuracy of remote sensing can be trusted (the overall accuracy is types such as categorical data (e.g., rock types) and continuous data 93.2%, Supp.-Tables 2, 3, 4). The area and the proportion of the land- (e.g., precipitation). The power of determinant (P) was defined as: scape types calculated from supervised classification are shown in X Table 1 and Fig. 4g. The result displayed that the landscapes of the m n Var Maowusu Sandy Land are dominated by meadow and sparse shrub ¼ − i¼1 Di Di ð Þ PD;R 1 2 whose proportion is approximately 70%. About 29.72% of the study nVar region is classified as active dunes or semi-fixed dunes. Although our results of the percentage of each landscape type, especially veg- PD, R indicates the power of determinant of D to R; n and Var are the total number of samples and the variance of Sand Cover Ratio of the etation coverage, are different from earlier studies (Zhu et al., 1980), the landscape types we identified are broadly in agreement with study area, respectively; nDi and VarDi are the number of samples and more recent studies (e.g., Deng et al., 2007; Mason et al., 2008; the variance of subregion Di; m is the number of subregions (e.g., in Fig. 3, for the impact factor C, m = 4). It's worth noting that when the Karnieli et al., 2014). subdivision of a factor perfectly explains the landscape spatial patterns, From a spatial distribution perspective, the number of active sand m dune patches increase gradually from southeast to northwest. The ma- ∑ n Var ð i¼1 Di Di Þ the second part nVar equals zero and PD, R = 1 (assuming jority of the active sand dune patches are located in the intermittent Var ≠ 0). This model consists of four parts, i.e., risk detector, factor detec- stream zones (Fig. 4). Moreover, these zones have higher altitudes tor, interaction detector and ecological detector (Wang et al., 2010; (Fig. 1b) and less precipitation (Fig. 2a). Every landscape is not regular Wang and Hu, 2012; GeogDetector Website, http://www.sssampling. in shape but occurs in many very small patches. The high Sand Cover Ratio (N70%) zone is mainly located in the southwest of Wushenzhao and west of Wushen Qi (Figs. 4 and 5b). On the contrary, the low Sand Cover Ratio (b10%) zone is mainly found in Jingbian County as well as in the northeast part of Yulin (Figs. 4 and 5b).

4.2. Triggering factors for the landscape spatial patterns disclosed by the GeogDetector

We used the factor detector to explore which geographic parameter is the most important impact factor for the landscape spatial patterns (Wang et al., 2010). The factor detector can quantify the impact of an assumed natural environment factor on the interpreted sand cover

spatial patterns. The factors are ranked by their inﬂuence (PD, R)on Sand Cover Ratio as follows:

ðÞ: % N ðÞ: % Fig. 3. Sand Cover Ratio layer R and the spatial pattern of suspect factors C and D and their Climatic potential productivity 26 4 local relief 22 5 N ðÞ: % N ðÞ: % N ðÞ: % : subdivision in the study region M. river buffer 20 1 wind condition 10 2 rock type 9 7 326 P. Liang, X. Yang / Catena 145 (2016) 321–333

Fig. 4. (a) Thematic map of the study area, compiled from Landsat 8 OLI images. The red line in the map is the demarcation line between intermittent streams' zone and perennial rivers' zone. Red circles (b–f) note ﬁve examples of iconic landscapes which are marked and presented in Fig. 4b–f. (g) The percentage of various landscape types in the study area (For interpretation of the references to color in this ﬁgure legend, the reader is referred to the web version of this article.).

According to the PD, R values shown above, we can see that the cli- Sand Cover Ratio pattern. The wind condition and rock type don't play matic potential productivity, local relief and river system (their PD, R any signiﬁcant role in forming the landscape spatial patterns. The eco- values N20%) are the ﬁrst three factors responsible for the observed logical detector's result revealed that the difference between the river

Fig. 5. (a) Binary map of landscape with square net and the example of Sand Cover Ratio calculation. (b) Sand Cover Ratio map derived from Fig. 5a. P. Liang, X. Yang / Catena 145 (2016) 321–333 327

Table 1 reaching its inflection point at level d whose wind speed ranges from Proportion of various landscape types in the Maowusu Sandy Land. 3.0 m/s to 3.1 m/s. Although there are some fluctuations in this relation- Active Semi-fixed Shrub & Meadow & Water ship, it seems not statistically significant (Supp.-Table 7). Type dunes dunes croplands steppe bodies

Area (km2) 4021.5 6213.1 10,644.7 13,410.2 150.8 4.3. Spatiotemporal pattern of NDVI from 1981 to 2013 Percentage 11.68 18.04 30.91 38.94 0.44 (%) Generally, NDVI exhibited a significant greening trend from 1981 to 2013 in the Maowusu Sandy Land (Fig. 7). It is consistent with an earlier investigation of NDVI and albedo between 1981 and 2010 (Karnieli et buffer and the wind condition is statistically significant (Table 2). The al., 2014). However, there are some differences on temporal and spatial results of factor detector and ecological detector, incontrovertibly, indi- scales. The NDVI increasing trend from 1981 to 2006 occurred only in cated that the climatic potential productivity plays the most important the eastern part of Maowusu Sandy Land. In contrast, the NDVI in the role in determining landscape spatial patterns because the PD, R value western part of Maowusu did not show any significant variation of climate potential productivity is the highest one and it has a statisti- trend, based on Mann–Kendall time series test (Fig. 7a). From 2005 to cally significant difference compared with other impact factors (Table 2013, significant increasing trends are found in the entire study area al- 2). though the eastern part has a more obvious tendency (Fig. 7b). In this study, we used the risk detector to detect which area has a higher Sand Cover Ratio to disclose the inner mechanism of forming 5. Interpretation and discussion the landscape spatial patterns. Except between level a and level b, level f and level g, the risk detector shows statistically significant differ- 5.1. Interpretation of results from the GeogDetector and possible landscape ences between all other climatic potential productivity levels (Supp.- forming mechanism Table 5). The climatic potential productivity strongly affects landscape spatial patterns, but a lower Sand Cover Ratio is not always associated The vegetation cover inferred from our work is greater than the with increased climatic potential productivity or precipitation, as value of 40–50% reported in Zhu et al. (1980) possibly due to different shown by the risk detector (Fig. 6). Generally, with the increase in cli- statistical methods or to the ecosystem rehabilitation effect since the matic potential productivity, the Sand Cover Ratio decreases significant- 1980s after the implementation of the state policy named “Three Norths ly first and then reaches the nadir (level e with the climatic potential Shelter Forest Program” (Zhang et al., 2003; Karnieli et al., 2014). The productivity of 3.98–4.18 kg/m2 a) at 18% and then increases to 23% at area of cropland also expanded during the past decades (Runnström, level f (Fig. 6). Although the Sand Cover Ratio increases from level a to 2003) and so adds to the increase in plant cover. This effect can over- level b, this trend is not statistically significant (Supp.-Table 5). The state the percentage of natural vegetation because the images' time apex of Sand Cover Ratio (36%) is within level b (Fig. 6) where the cli- window we selected coincides with the crop (mainly maize) growing matic potential productivity is at the second lowest level (Fig. 2a) and season (Supp.-Table 1) and it is not an easy task for us to separate culti- the mean annual precipitation is ca. 280 mm. It emerged that the south- vated crops from natural vegetation in the natural color map. This differ- east (level f and level g) has the highest climatic potential productivity ence reconfirms that there are crucial uncertainties while using the time (Fig. 2a) and a mean annual precipitation of ca. 400 mm but its Sand series analysis to detect the driving forces of desertification. Cover Ratio is not the lowest. This exception means that other factors From a spatial perspective, the landscape patterns represented by should have an additional impact on the landscapes' distribution on a the Sand Cover Ratio distribution in the Maowusu Sandy Land occur in regional scale. three class-levels which can be seen as the regular zonal changes. According to the factor detector results, local relief and river sys- First, the gradual increase in Sand Cover Ratio from southeast to north- tem also play important roles in landscape distribution, with PD, R west is similar to longitudinal zonality. Second, the repeated occurrence values of 22.5% and 20.1%, respectively. The difference in PD, R values of the strips of sand dunes from southwest to northeast is similar to lat- between local relief and river system is not statistically significant itudinal zonality although it has nothing to do with the changes of lati- (Table 2). The relationships between river networks and Sand tudes. Third, the patches of dunes and wetlands are azonal (Fig. 4). The Cover Ratio are shown in Fig. 6. It means that the average Sand smaller scale patterns like patches are nested within the larger scale Cover Ratio increased with the increase in the distance to river chan- patterns. The GeogDetector model results show that at different spatial nels (both present rivers and dry channels). Although the Sand Cover scales, landscape patterns respond to different driving factors. The cli- Ratio decreases from level b to c, and from level f to g, these changes matic potential productivity exhibits a similar spatial pattern as the are not statistically significant according to risk detector (Supp.- landscape changes from southeast to northwest, suggesting that climate Table6).ButtherelationshipbetweenthelocalreliefandtheSand controls the landscapes' distribution on a regional scale. Similarly, the Cover Ratio is not obvious based on the topography risk detector spatial pattern of river networks and local relief results in the striped (Supp.-Fig. 1; Supp.-Table 8). landscapes from southwest to northeast. But how to explain the large According to the risk detector's results about wind conditions, with number of small patches of bare sand which are usually surrounded the increase in the mean annual wind speed (from level a to level g in by vegetation or wetlands? We believe that many of the small patches Fig. 6), the Sand Cover Ratio first increases and then decreases after of bare sand were mainly caused by human activities. This opinion is supported by the satellite data and field investigation (Fig. 8). According to our field survey, many small patches of grass or shrub occur in the Table 2 fi Statistical significance of power of determinants between factors according to the ecolog- inter-dune areas of larger dune elds, while other bare sand patches ical detector. surrounded by meadows could be caused by overgrazing (Zhu, 1979). The different land cover types separated by the fence-line are ubiqui- Stat. Sig. Diff. Climate Local relief River buffer Wind condition Rock type tous in the Maowusu Sandy Land (Fig. 8c and d), similar to the cases re- Climate ported from other sandy environments in the world (e.g., Tsoar and Local relief Y River buffer Y N Karnieli, 1996; Yizhaq et al., 2007; Mason et al., 2008; Li and Yang, Wind condition Y Y Y 2014). It is worth mentioning that the human impact on landscape at Rock type Y Y Y N a fine scale hase a threshold. A five-year grazing experiment in Inner Y means the difference between two factors statistically significant with the confidence Mongolia conducted by Zhao et al. (2005) showed that continuous ≥95%, and N means not. heavy grazing could result in small bare spots on the ground, which 328 P. Liang, X. Yang / Catena 145 (2016) 321–333

Fig. 6. Results of risk detector. Different factors, i.e., climatic potential productivity, river system and wind, refer to their original definitions for level a to level g (see Fig. 2 or Supp.-Table 5, Supp.-Table 6, Supp.-Table 7). merge into larger bare land later in the rangeland. Inconsistent changes Kumar, 2001; Kim and Mohanty, 2016). In the inter-dune areas of the in the Sand Cover Ratio and the climatic potential productivity trends Maowusu, the groundwater table is often just around 1 m as evidenced (Fig. 6) also indicate that human activities, particularly excessive agri- by our field excavations. culture practice, have the potential to affect the landscape spatial pat- We think that the occurrence of sand dunes in relatively high reliefs terns at a larger scale, and even result in regional desertification. might be related to deeper ground water tables. Clear evidence from

The GeogDetector revealed that there is a high similarity of the PD,R field observation showed that the lower sites are wetlands or covered values between local relief and river system and they have a non-signif- by dense vegetation but bare sand dunes appear in the higher elevations icant statistical difference in terms of the risk detector (Table 2). One nearby (Fig. 9d and e). In extremely arid regions such as the Taklimakan possible explanation is an inextricable link between river system and desert, oases surrounded by sand seas often occur along river channels local relief. River networks can be extracted from the Digital Elevation because of the better soil moisture conditions along river courses Model based on the relationship between river channel and topography (Yang, 2001). The changes of soil moisture in relation to the distance (O'Callaghan and Mark, 1984; Luo et al., 2014). Since the relief changes to the river channels can explain the reasons causing striped spatial just slightly in our study area, and thus the absolute altitude variation landscapes similar to the river networks, as well as the increasing trends cannot represent small topographic features, this causes uncertainty in of the Sand Cover Ratios on both sides of the river channels (Fig. 6). We the results of risk detector (Supp.-Fig. 1; Supp.-Table 8). The topograph- thus think that the river system and local relief control the landscape ic analysis and field investigation showed that local relief, river net- spatial pattern through the soil moisture and ground water availability. works and sand dunes have a highly similar spatial pattern, especially Exploring the Sand Cover Ratio change with wind speed is quite dif- in the west part of the Maowusu Sandy Land (Fig. 9). In arid and semi- ficult and the PD, R value of wind (10.2%) must be underestimated in the arid regions, the soil water content is particularly vital for vegetation GeogDetector, because the wind energy differs greatly from site to site growth. It is obvious that local relief controls soil water patterns and (Wiggs et al., 1995; Yizhaq et al., 2007) and it is influenced by the un- even the groundwater table through surface flow and subsurface lateral derlying relief (Yang et al., 2007). Thus, the relationships between movement (Burt and Butcher, 1985; Grayson et al., 1997; Chen and wind speed and the Sand Cover Ratio cannot be disclosed by the risk

Fig. 7. Human relics and trends of July 1–15 NDVI from 1981 to 2006 (a) and from 2005 to 2013 (b) in the study area. The relics of Neolithic sites and ancient cities of Han Dynasty (202 B.C. to A.D. 220) were redrawn from Huang et al. (2009). Black triangle symbolized the residual scarp of palaeosols where ceramics or copper coins found (based on our ﬁeld investigation in September 2013). Blue line marks the dividing line between level e and level f in climatic potential productivity (Fig. 2a). The Great Wall referring to Chinese Cultural Bureau (1998) (For interpretation of the references to color in this ﬁgure legend, the reader is referred to the web version of this article.). P. Liang, X. Yang / Catena 145 (2016) 321–333 329

Fig. 8. Landscape of Fig. 1c in winter season (a) and summer season (b). The circle farmlands are bare in winter, but covered by crops in summer. These two images were overlaid by bands 7, 5, 2 from Landsat 8. (c) Straight boundary between two landscapes under different grazing pressure, image from Google Earth. (d) Field observation in September 2013. detector because of the low spatial resolution of wind data acquired In many cases in the Maowusu Sandy Land, human activities from meteorological stations. On the other hand, the proxy of wind con- played a more critical role than climatic background in shaping the ditions we used, i.e., mean annual wind speed, cannot present the de- landscapes. This opinion, however, should be treated with cautions tails of local wind energy (Fig. 2b). In the world of aeolian when we take the geological histories and spatial scales of the geomorphology, the wind energy is often presented via drift potential study area into account. We found that human activities in the (DP) (Fryberger and Dean, 1979). But acquiring high spatial resolution Maowusu had a northwest boundary which is roughly consistent data showing DP variations on local scales is not realistic anyhow. with the dividing line between level e and level f of climatic potential productivity (Fig. 2a). The discrepancy between the Sand Cover Ratio and climatic background should be attributed to intensive human ac- 5.2. Relative significance of human activities and natural factors in forming tivities in the past in the eastern part. Actually, the dividing line be- the landscape of the Maowusu tween level e and level f of climatic potential productivity is roughly consistent with the border between Shaanxi Provence and Identifying the relative role of human activities and natural factors in the Autonomous Region of Inner Mongolia (Fig. 7), which means forming the landscape of drylands is crucial for understanding the land- that the level f and level g of climatic potential productivity take scape evolution and desertification processes. Some researchers place in the Shaanxi Province where human activities were intensive claimed that the desert-like conditions in the Maowusu Sandy Land in the past. The historical records showed that the large-scale agri- were caused by human activities over the past 2000 years (Hou, 1973; cultural reclamation in the Ming Dynasty (A.D. 1368–1644) took Zhu, 1979). But Wang et al. (2007) argued that there is no firm evidence placesouthoftheGreatWall(Han, 2003) and cultivation in the to defend the ‘human-made’ viewpoint and they concluded that climate areas north of the Great Wall occurred in the Qing Dynasty (A.D. fluctuations triggered the desertification of the Maowusu Sandy Land. 1644–1912), and the northwest fringe of the cultivation was extend- Our result of the GeogDetector showed that the average Sand Cover ed 29 km away from the Great Wall (Han, 2006; Wu et al., 2014; Ratio does not always decrease with the increase in climatic potential Zhang et al., 2015). In other words, the Shaanxi–Inner Mongolia bor- productivity, as shown by the risk detector (Fig. 6). It indicates that der (also the dividing line between level e and level f in Figs. 2aand the landscape of eastern Maowusu is not consistent with its relatively 7) was the northwest boundary of agricultural cultivation in history advantageous climatic background. It was initially pointed out by Zhu especially during the Qing Dynasty (Han, 2006; Wu et al., 2014). The et al. (1980) that the desertification was caused by human economic ac- historical records are consistent with archaeological evidence show- tivities such as over-grazing, over-reclamation and over-cutting before ing that most of the past human relics, such as palaeosol sites with founding of the People's Republic of China (A.D. 1949). This viewpoint copper coins and ceramics and ancient cities, are located southeast has been supported by some later studies (e.g., Sun, 2000; Wu and Ci, of the Shaanxi-Inner Mongolia border (Hou, 1973; Fig. 7). It confirms 2002; Zhang et al., 2003; Huang et al., 2009) and Huang et al. (2009) that agricultural activities mainly occurred in the eastern part rather claimed further that the whole Maowusu region's desertification during than in the entire Maowusu Sandy Land during historical times. the late period of the Ming Dynasty (the time around ca. A.D. 1500– Huang et al. (2009), from the stratigraphic records and archaeologi- 1600) was due to human activities. cal discoveries, emphasized that desertification in the Maowusu 330 P. Liang, X. Yang / Catena 145 (2016) 321–333

Fig. 9. The relationships between Sand Cover Ratio (SCR) and local relief. (a) Landsat 8 OLI images overlaid by bands 4, 3, 2. (b) Local relief map according to ASTER GDEM data (Blue dash lines mark intermittent river channels). (c) Sand Cover Ratio and local relief changes at selected cross sections in Fig. 9a or b. (d) Areas dominated by vegetation in relatively lower relief in Fig. 9a or b. (e) Areas dominated by sand dunes in relatively higher relief in Fig. 9a or b (For interpretation of the references to color in this ﬁgure legend, the reader is referred to the web version of this article.).

Sandy Land took place in two phases and the latter starting in the late After a detailed scrutiny of the previous studies and the NDVI trends Ming Dynasty (ca. A.D. 1500–1600) and ending in the 1980s, mainly in the Maowusu, we found out that the dividing line between increased due to human activities. This paper, however, did not contain any in- and unchanged NDVI areas is roughly consistent with the Shaanxi-Inner formation about the dimension of human activities. Mongolia border also (Fig. 7). We think that this phenomenon is not a Bearing in mind that the human activities have spatial coincidence. The eastern part of Maowusu suffered intensive human ac- dimensions, we investigated the trends of NDVI from 1981 to 2013 tivities, especially over-cultivation since the Qing Dynasty (A.D. 1644– in our study areas. Earlier Mason et al. (2008) reported that there 1912). This large-scale agricultural practices had a more severe impact were no extensive and consistent trends of NDVI from 1981 to on the ecosystem than grazing and transformed the landscape markedly 2003 in the Maowusu despite the weaker wind power after 1981. (Han, 2003; Zhang et al., 2015). Although covered and protected by Our spatiotemporal analysis on NDVI from 1981 to 2013 confirm crops in the summer, the farmland surface is exposed in winter and that the increase in NDVI occurred only in eastern Maowusu and spring, and then suffers intensive wind deflation (Fig. 8a and b). With the NDVI of the western part remains almost unchanged from 1981 the improvement of the consciousness of preventing desertification, ac- to 2006 (Fig. 7a). Why did the increasing NDVI trends just occur in tions to fight desertification, including grazing bans, ground and air- the eastern part? Mason et al. (2008) argued that deliberate stabili- seeding of trees, bushes and grasses over large areas since the 1980s zation programs along with weakening winds may explain this have taken effect (Karnieli et al., 2014). The area of cultivated lands, trend. However, the weakening winds and stabilization programs which could influence the landscape at a large scale than grazing, also have occurred in the entire Maowusu Sandy Land (Mason et al., decreased since the 1960s in the eastern part of the Maowusu (except 2008; Wang X. et al., 2010), and they could not explain the almost the Jingbian county after 2005) (Fig. 10). Additionally, both near-surface unchanged NDVI in the western part. wind power and days of strong wind have shown significant declining P. Liang, X. Yang / Catena 145 (2016) 321–333 331

Fig. 10. Annual changes of cultivated land areas in counties Yulin, Jingbian and Hengshan. Data derived from Statistical Bureau of Yulin (1996–2013). trends since the 1970s in the northern mid-latitudes as well as in the landscape spatial heterogeneity in the Maowusu on the local scale. Maowusu (Wang et al., 2007; Jiang et al., 2010; Vautard et al., 2010). Human activities disturb the landscape spatial patterns on a very fine Both ecological rehabilitation efforts and weakening of winds result in scale which belongs to the azonal geographical changes. However, the some stabilization of sand dunes (Wang et al., 2007; Xu et al., 2015) large-scale agricultural reclamation in the eastern Maowusu has made and biomass increase (Runnström, 2003; Yan et al., 2015)inthe asignificant impact on landscape spatial patterns on a regional scale. Maowusu in the past thirty years. Under the background of wind power decreasing in the past thirty We think that the different responses to external forcing between years, the NDVI trends presented a distinct increasing trend in the east- the western and eastern parts of the Maowusu are due to different ern part of the Maowusu but almost no change in the western part environmental carrying capacities and the different land use history until the year 2006. The dividing line of these two parts is consistent of these two regions. As wind energy reduced and grasses or trees with the Shaanxi–Inner Mongolia provincial border as well as the divid- were planted, the eastern part of Maowusu moved toward a better ing line between level e and level f of climatic potential productivity and condition quickly, resulting from its wetter climatic background the northwest boundary of the agricultural cultivation in the Qing Dynas- and large decrease of human interference. But as the wind power ty (A.D. 1644–1912). From the observed NDVI trends from 1981 to 2013 continues to decrease or maintain at a weak status, and without ex- and land use histories of the study region, and the results from risk detec- pansion of destructive human activities, the dune field can be poten- tor of climatic potential productivity, we conclude that the causes of the tially stabilized and the NDVI shows increase trends in the entire NDVI trends increasing only in the eastern part were primarily the recent Maowusu Sandy Land at the end (Fig. 7b). We think that the increas- ecological rehabilitation activities, supported by better moisture condi- ing NDVI trends in the entire Maowusu over the recent decades tions. The reduced wind power might be of relevance but not decisive. (Fig. 7b) can be attributed to three reasons, i.e., the reducing destruc- Our investigation of the landscape spatial patterns demonstrates that tive human interference, increasing constructive human endeavor the asynchronous responses to weakening winds between the western and decreasing wind energy. This phenomenon is similar to the oc- and eastern parts of the Maowusu are due to different climate back- currence in the Israel-Egypt border area where there is a difference ground conditions and land use histories. It is worth mentioning that between the bare sand on the Egyptian side and the vegetated and the scale effect should be taken into account when assessing the domi- fully fixed dunes on the Israeli side (Yizhaq et al., 2007). This contrast nant factors in shaping the landscape patterns even in desertification. was attributed to the anthropogenic pressure ceased in the Israeli Our research reconfirms that the success of human activities in fighting side since 1982, along with the weak wind power and above-thresh- desertification in the Maowusu Sandy Land is affected by natural factors. old precipitation (Yizhaq et al., 2007). Supplementary data to this article can be found online at http://dx. Our results from the GeogDetector and spatiotemporal analysis on doi.org/10.1016/j.catena.2016.06.023. NDVI suggested that the human impact has a northwest boundary, which is consistent with the boundary between level e and level f of cli- Acknowledgments matic potential productivity. The spatiotemporal pattern of NDVI suggests that there are out-of-step responses to external forcing change This study is supported by the Chinese Academy of Sciences (CAS between western and eastern parts of the Maowusu because of different Strategic Priority Research Program, Grant no.: XDA05120502), climatic background and land use histories. We think the range and the National Natural Science Foundation of China (Grant no.: scale of the human activities should be taken into account while 41430532) and the Alexander von Humboldt Stiftung/Foundation, assessing the role of human activities in desertification. Germany (Research Award). Sincere thanks are extended to Prof. Martin Williams for having dully corrected the entire manu- 6. Conclusions script, to Prof. Bernhard Eitel, Dr. Bertil Mächtle and Dipl.-Geol. Gerd Schukraft for constructive suggestions and discussions, to Our results show that the first three important natural factors in Dr. Deguo Zhang for technical help, to Prof. Jinfeng Wang for provid- shaping the landscape spatial patterns are climatic potential productiv- ing the Geographical Detector model, to Editor Prof. Markus Egli and ity, local relief and river system and their power of determinant two anonymous reviewers for helpful comments. values (PD, R value) are 26.4%, 22.5% and 20.1%, respectively. Our study suggests that landscape spatial patterns of different scales would have different controlling factors. Climatic potential productivity decides References the landscapes' distribution on a regional scale. Local relief and river fl Bayarsaikhan, U., Boldgiv, B., Kim, K.-R., Park, K.-A., Lee, D., 2009. Change detection and networks in uence the sand dunes spatial pattern through soil mois- classification of land cover at Hustai National Park in Mongolia. Int. J. Appl. Earth ture, groundwater and local wind. These two factors cause the Obs. Geoinf. 11, 273–280. 332 P. Liang, X. Yang / Catena 145 (2016) 321–333

Beveridge, C., Kocurek, G., Ewing, R.C., Lancaster, N., Morthekai, P., Singhvi, A.K., Mahan, Livingstone, I., Bristow, C., Bryant, R.G., Bullard, J., White, K., Wiggs, G.F.S., Baas, A.C.W., S.A., 2006. Development of spatially diverse and complex dune-field patterns: Gran Bateman, M.D., Thomas, D.S.G., 2010. The Namib Sand Sea digital database of aeolian Desierto Dune Field, Sonora, Mexico. Sedimentology 53, 1391–1409. dunes and key forcing variables. Aeolian Res. 2, 93–104. Blümel, W.D., 2013. Wüsten. Eugen Ulmer KG, Stuttgart. Luo, W., Li, X., Molloy, I., Di, L., Stepinski, T., 2014. Web service for extracting stream net- Bürgi, M., Hersperger, A.M., Schneeberger, N., 2004. Driving forces of landscape works from DEM data. GeoJournal 79, 183–193. change—current and new directions. Landsc. Ecol. 19, 857–868. Ma, L., Qi, X., Tang, H., Dai, W., 2004. Geological Atlas of China. Geological Publishing Burt, T.P., Butcher, D.P., 1985. Topographic controls of soil moisture distributions. J. Soil House, Beijing (in Chinese). Sci. 36, 469–486. Mann, H., 1945. Nonparametric tests against trend. Econometrica 13, 245–259. Campbell, J.B., Wynne, R.H., 2011. Introduction to Remote Sensing. fifth ed. The Guilford Mason, J.A., Swinehart, J.B., Lu, H., Miao, X., Cha, P., Zhou, Y., 2008. Limited change in dune Press, New York and London. mobility in response to a large decrease in wind power in semi-arid northern China Cao, F., Ge, Y., Wang, J.F., 2013. Optimal discretization for geographical detectors-based since the 1970s. Geomorphology 102, 351–363. risk assessment. GIScience Remote Sens. 50, 78–92. O'Callaghan, J.F., Mark, D.M., 1984. The extraction of drainage networks from Digital Ele- Chen, J., Kumar, P., 2001. Topographic influence on the seasonal and interannual variation vation Data. Computer Version, Graphics, and Image Processing. 28, pp. 323–344. of water and energy balance of basins in North America. J. Clim. 14, 1989–2014. Ohte, N., Koba, K., Yoshikawa, K., Sugimoto, A., Matsuo, N., Kabeya, N., Wang, L., 2003. Chinese Cultural Bureau, 1998. The Atlas of Chinese Cultural Sites: Shaanxi Province. Xi'an Water utilization of natural and planted trees in the semiarid desert of Inner Mongo- Cartographic Publishing House, Xi'an (in Chinese). lia, China. Ecol. Appl. 13, 337–351. Congalton, R.G., 1991. A review of assessing the accuracy of classifications of remotely Peel, M.C., Finlayson, B.L., McMahon, T.A., 2007. Updated world map of the Köppen-Geiger sensed data. Remote Sens. Environ. 37, 35–46. climate classification. Hydrol. Earth Syst. Sci. 11, 1633–1644. Deng, H., Shu, S.G., Song, Y.Q., Xing, F.L., 2007. Distribution of sand dunes and sand shifts Penman, L., 1948. Natural evaporation from open water, bare soil and grass. Proc. R. Soc. A along the southern fringe of the Mu Us Desert since the Ming Dynasty. Chin. Sci. Bull. 193, 120–145. 52, 3128–3138. R Core Team, 2016. R: A Language and Environment for Statistical Computing. R Founda- Di Gregorio, A., Jansen, L.J.M., 2000. Land Cover Classification System (LCCS): Classifica- tion for Statistical Computing, Vienna, Austria (URL: https://www.R-project.org/). tion Concepts and User Manual. Food and Agriculture Organization of the United Na- Rasmussen, K., Fog, B., Madsen, J.E., 2001. Desertification in reverse? Observations from tions, Rome. northern Burkina Faso. Glob. Environ. Chang. 11, 271–282. Domrös, M., Peng, G., 1988. The Climate of China. Springer, Berlin, Heidelberg. Ren, M., 1999. The Outline of Physical Geography of China. third ed. Commercial Publish- Eitel, B., Hecht, S., Mächtle, B., Schukraft, G., Kadereit, A., Wagner, G., Kromer, B., Unkel, I., er, Beijing (in Chinese). Reidel, M., 2005. Geoarchaeological evidence from desert loess in the Nazca–Palpa re- Richards, J.A., 2013. Remote Sensing Digital Image Analysis: An Introduction. fifth ed. gion, southern Peru: palaeoenvironmental changes and their impact on Pre-Columbi- Springer, Berlin and Heidelberg. an cultures. Archaeometry 47, 137–185. Runnström, M., 2003. Rangeland development of the Mu Us Sandy Land in semiarid ESRI, 2015. ArcMap for Desktop: Release 10.3. Environmental Systems Research Institute, China: an analysis using Landsat and NOAA remote sensing data. Land Degrad. Dev. Redlands, CA, USA (URL: https://www.esri.com/). 14, 189–202. Ewing, R., Kocurek, G., 2010. Aeolian dune-field pattern boundary conditions. Geomor- Scuderi, L., Weissmann, G., Kindilien, P., Yang, X., 2015. Evaluating the potential of data- phology 114, 175–187. base technology for documenting environmental change in China's deserts. Catena Forman, S.L., Oglesby, R., Webb, R.S., 2001. Temporal and spatial patterns of Holocene 134, 87–97. dune activity on the Great Plains of North America: megadroughts and climate Singhvi, A., Williams, M., Rajaguru, S., Misra, V., Chawla, S., Stokes, S., Chauhan, N., Francis, links. Glob. Planet. Chang. 29, 1–29. T., Ganjoo, R., Humphreys, G., 2010. A ∼200 ka record of climatic change and dune ac- Fryberger, S.G., Dean, G., 1979. Dune forms and wind regime. In: McKee, E.D. (Ed.), A tivity in the Thar Desert, India. Quat. Sci. Rev. 29, 3095–3105. Study of Global Sand Seas. United States Geological Survey, Washington, DC, Slaymaker, O., Spencer, T., 1998. Physical Geography and Global Environmental Change. pp. 137–169 (Volume 1052 of USGS Professional Paper). Longman, Harlow. Goudie, A.S., 2002. Great Warm Deserts of the World: Landscapes and Evolution. Oxford Statistical Bureau of Yulin, 1996–2013. Statistical Yearbook of Yulin (1996–2013). China University Press, New York. Statistics Press, Beijing (in Chinese). Grayson, R.B., Western, A.W., Chiew, F.H., Blöschl, G., 1997. Preferred states in spatial soil Stevens, T., Carter, A., Watson, T.P., Vermeesch, P., Andó, S., Bird, A.F., Lu, H., Garzanti, E., moisture patterns: local and nonlocal controls. Water Resour. Res. 33, 2897–2908. Cottam, M.A., Sevastjanova, I., 2013. Genetic linkage between the Yellow River, the Han, Z.Q., 2003. The evolution of the Maowusu Desert and the reclamation in the adjacent Mu Us desert and the Chinese Loess Plateau. Quat. Sci. Rev. 78, 355– 368. areas in the Ming Dynasty. Social Sciences in China, pp. 151–204 (in Chinese with En- Strahler, A.H., 1980. The use of prior probabilities in maximum likelihood classification of glish abstract). remotely sensed data. Remote Sens. Environ. 10 (2), 135–163. Han, Z.Q., 2006. Effect of opening western part of Inner Mongolia for cultivation on Sun, J.M., 2000. Origin of eolian sand mobilization during the past 2300 years in the Mu Us change of Mu Us Desert at the end of the Qing Dynasty. Sci. Geogr. Sin. 26, Desert, China. Quat. Res. 53, 78–88. 728–734 (in Chinese with English abstract). Tachikawa, T., Hato, M., Kaku, M., Iwasaki, A., 2011. Characteristics of ASTER GDEM ver- Hou, R., 1973. A research on the evolution of the Maowusu Desert as indicated by the sion 2. Geoscience and Remote Sensing Symposium (IGARSS), 2011 IEEE Internation- deserted ancient cities along the Hongliu River. Cult. Relics 1, 35–41 (in Chinese). al. IEEE, pp. 3657–3660. Huang, B.W., 1985. Potential agriculture productivity of China—potential photosynthesis Tsoar, H., Karnieli, A., 1996. What determines the spectral reflectance of the Negev-Sinai productivity. In: Institute of Geography, Chinese Academy of Sciences (Eds.), Agricul- sand dunes. Int. J. Remote Sens. 17 (3), 513–525. ture Productivity Research. Science Press, Beijing, pp. 15–22 (volume 17 of Geograph- Tucker, C.J., Pinzon, J.E., Brown, M.E., 2004. Global Inventory Modeling and Mapping Stud- ic Selections, (in Chinese)). ies, NA94apr15b.n11-VIg, 2.0. Global Land Cover Facility, University of Maryland, Col- Huang, B.W., Zheng, D., Zhao, M.C., 1999. Contemporary Physical Geography. Science lege Park, Maryland. Press, Beijing (in Chinese). Tucker, C.J., Pinzon, J.E., Brown, M.E., Slayback, D.A., Pak, E.W., Mahoney, R., Vermote, E.F., Huang, Y., Wang, N., He, T., Chen, H., Zhao, L., 2009. Historical desertification of the Mu Us El Saleous, N., 2005. An extended AVHRR 8-km NDVI dataset compatible with MODIS Desert, Northern China: a multidisciplinary study. Geomorphology 110, 108–117. and SPOT vegetation NDVI data. Int. J. Remote Sens. 26, 4485–4498. Hugenholtz,C.H.,Levin,N.,Barchyn,T.E.,Baddock,M.C.,2012.Remote sensing and Unkel, I., Kadereit, A., Mächtle, B., Eitel, B., Kromer, B., Wagner, G.A., Wacker, L., 2007. spatial analysis of aeolian sand dunes: a review and outlook. Earth Sci. Rev. Dating methods and geomorphic evidence of palaeo-environmental changes at the 111, 319–334. eastern margin of the South Peruvian coastal desert (14°30′S) before and during Jiang, Y., Luo, Y., Zhao, Z., Tao, S., 2010. Changes in wind speed over China during 1956– the Little Ice Age. Quat. Int. 175, 3–28. 2004. Theor. Appl. Climatol. 99, 421–430. Vautard, R., Cattiaux, J., Yiou, P., Thepaut, J.-N., Ciais, P., 2010. Northern Hemisphere atmo- Karnieli, A., Qin, Z., Wu, B., Panov, N., Yan, F., 2014. Spatio-temporal dynamics of land-use spheric stilling partly attributed to an increase in surface roughness. Nat. Geosci. 3, and land-cover in the Mu Us Sandy Land, China, using the Change Vector Analysis 756–761. technique. Remote Sens. 6, 9316–9339. Walter, H., Lieth, H., 1960. Klimadiagramm Weltatlas. G. Fischer, Jena, Germany. Kendall, M., 1970. Rank Correlation Methods. fourth ed. Griffin, London, UK. Wang, J.F., Hu, Y., 2012. Environmental health risk detection with GeogDetector. Environ. Kim, J., Mohanty, B.P., 2016. Influence of lateral subsurface flow and connectivity on soil Model. Softw. 33, 114–115. water storage in land surface modeling. J. Geophys. Res.-Atmos. 121. http://dx.doi. Wang, X., Hasi, E., Zhou, Z., Liu, X., 2007. Significance of variations in the wind energy en- org/10.1002/2015JD024067. vironment over the past 50 years with respect to dune activity and desertification in Köppen, W., 1931. Grundrisse der Klimakunde. Walter de Gruyter, Berlin. arid and semiarid northern China. Geomorphology 86, 252–266. Lancaster, N., Wolfe, S., Thomas, D., Bristow, C., Bubenzer, O., Burrough, S., Duller, J., Wang, J.F., Li, X.H., Christakos, G., Liao, Y.L., Zhang, T., Gu, X., Zheng, X., 2010. Geographical Halfen, A., Hesse, P., Roskin, J., Singhvi, A., Tsoar, H., Tripaldi, A., Yang, X., Zarate, M., detectors based health risk assessment and its application in the Neural Tube Defects 2015. The INQUA Dunes Atlas chronologic database. Quat. Int. http://dx.doi.org/10. study of the Heshun region, China. Int. J. Geogr. Inf. Sci. 24, 107–127. 1016/j.quaint.2015.10.044 (2015). Wieringa, J., 1986. Roughness—dependent geographical interpolation of surface wind Li, H., Yang, X., 2014. Temperate dryland vegetation changes under a warming climate speed averages. Q. J. R. Meteorol. Soc. 112, 867–889. and strong human intervention—with a particular reference to the district Xilin Gol, Wiggs, G.F.S., Thomas, D.S.G., Bullard, J.E., Livingstone, I., 1995. Dune mobility and Inner Mongolia, China. Catena 119, 9–20. vegetation cover in the Southwest Kalahari Desert. Earth Surf. Process. Landf. Lillesand, T.M., Kiefer, R.W., Chipman, J.W., 2004. Remote Sensing and Image Interpreta- 20, 515–529. tion. John Wiley & Sons Ltd. Williams, M., 2014. Climate Change in Deserts. Cambridge University Press, Cambridge. Liu, M., 2010. Atlas of Physical Geography of China. third ed. China Map Press, Beijing (in Williams, M., 2015. Interactions between fluvial and eolian geomorphic systems and pro- Chinese). cesses: Examples from the Sahara and Australia. Catena 134, 4–13. Liu, L.Y., Skidmore, E., Hasi, E., Wagner, L., Tatarko, J., 2005. Dune sand transport as influ- Williams, M., 2016. Nile Waters, Saharan Sands: Adventures of a Geomorphologist at enced by wind directions, speed and frequencies in the Ordos Plateau, China. Geo- Large. Springer, Dordrecht. morphology 67, 283–297. Williams, M., Balling, R., 1996. Interactions of Desertification and Climate. Arnold, London. P. Liang, X. Yang / Catena 145 (2016) 321–333 333

Wu, B., Ci, L.J., 2002. Landscape change and desertification development in the Mu Us Yizhaq, H., Ashkenazy, Y., Tsoar, H., 2007. Why do active and stabilized dunes coexist Sandland, Northern China. J. Arid Environ. 50, 429–444. under the same climatic conditions? Phys. Rev. Lett. 98, 188001. Wu, C., Deng, H., Shu, S., 2014. The study on the land development process in the border Zhang, L., Yue, L.P., Xia, B., 2003. The study of land desertification in transitional zones be- area between Shaanxi and Inner Mongolia. Geogr. Res. 33, 1579–1592 (in Chinese tween the Mu Us Desert and the Loess Plateau using RS and GIS—a case study of the with English abstract). Yulin region. Environ. Geol. 44, 530–534. Xu, Z., Mason, J.A., Lu, H., 2015. Vegetated dune morphodynamics during recent stabiliza- Zhang, X., Sun, T., Xu, J., 2015. The relationship between the spread of the Catholic Church tion of the Mu Us dune field, north-central China. Geomorphology 228, 486–503. and the shifting agro-pastoral line in the Chahar Region of northern China. Catena Yan, F., Wu, B., Wang, Y., 2015. Estimating spatiotemporal patterns of aboveground bio- 134, 75–86. mass using Landsat TM and MODIS images in the Mu Us Sandy Land, China. Agric. Zhao, H.L., Zhao, X.Y., Zhou, R.L., Zhang, T.H., Drake, S., 2005. Desertification processes due For. Meteorol. 200, 119–128. to heavy grazing in sandy rangeland, Inner Mongolia. J. Arid Environ. 62, 309–319. Yang, X., 2001. The oases along the Keriya River in the Taklamakan Desert, China, and Zhu, Z., 1979. Thirty years in research works on Chinese sandy deserts. Acta Geograph. their evolution since the end of the last glaciation. Environ. Geol. 41, 314–320. Sin. 34 (4), 305–314 (in Chinese with English abstract). Yang, X., Ding, Z., Fan, X., Zhou, Z., Ma, N., 2007. Processes and mechanisms of desertifica- Zhu, Z., Liu, S., 1981. Desertification Processes and their Control in Northern China. China tion in northern China during the last 30 years, with a special reference to the Forestry Press, Beijing (in Chinese). Hunshandake Sandy Land, eastern Inner Mongolia. Catena 71, 2–12. Zhu, Z., Wu, Z., Liu, S., Di, X., 1980. An Outline of Chinese Deserts. Science Press, Beijing (in Yang, X., Li, H., Conacher, A., 2012. Large-scale controls on the development of sand seas Chinese). in northern China. Quat. Int. 250, 74–83. Zuo, X., Zhao, X., Zhao, H., Zhang, T., Li, Y., Wang, S., Li, W., Powers, R., 2012. Scale depen- Yang, X., Wang, X., Liu, Z., Li, H., Ren, X., Zhang, D., Ma, Z., Rioual, P., Jin, X., Scuderi, L., dent effects of environmental factors on vegetation pattern and composition in 2013. Initiation and variation of the dune fields in semi-arid China—with a special ref- Horqin Sandy Land, Northern China. Geoderma 173-174, 1–9. erence to the Hunshandake Sandy Land, Inner Mongolia. Quat. Sci. Rev. 78, 369–380.