- Email: [email protected]
Identification of the key ecological factors influencing vegetation degradation in semi-arid agro-pastoral ecotone considering spatial scales
Acta Oecologica 85 (2017) 62e68
Contents lists available at ScienceDirect
Acta Oecologica journal homepage: www.elsevier.com/locate/actoec
Identification of the key ecological factors influencing vegetation degradation in semi-arid agro-pastoral ecotone considering spatial scales Yu Peng*, Qinghui Wang, Min Fan College of Life & Environmental Sciences, Minzu University of China, Beijing 100081, China
a r t i c l e i n f o
a b s t r a c t
Article history: Received 3 February 2017 Received in revised form 17 September 2017 Accepted 19 September 2017
When assessing re-vegetation project performance and optimizing land management, identification of the key ecological factors inducing vegetation degradation has crucial implications. Rainfall, temperature, elevation, slope, aspect, land use type, and human disturbance are ecological factors affecting the status of vegetation index. However, at different spatial scales, the key factors may vary. Using Helin County, Inner-Mongolia, China as the study site and combining remote sensing image interpretation, field surveying, and mathematical methods, this study assesses key ecological factors affecting vegetation degradation under different spatial scales in a semi-arid agro-pastoral ecotone. It indicates that the key factors are different at various spatial scales. Elevation, rainfall, and temperature are identified as crucial for all spatial extents. Elevation, rainfall and human disturbance are key factors for small-scale quadrats of 300 m 300 m and 600 m 600 m, temperature and land use type are key factors for a medium-scale quadrat of 1 km 1 km, and rainfall, temperature, and land use are key factors for large-scale quadrats of 2 km 2 km and 5 km 5 km. For this region, human disturbance is not the key factor for vegetation degradation across spatial scales. It is necessary to consider spatial scale for the identification of key factors determining vegetation characteristics. The eco-restoration programs at various spatial scales should identify key influencing factors according their scales so as to take effective measurements. The new understanding obtained in this study may help to explore the forces which driving vegetation degradation in the degraded regions in the world. © 2017 Elsevier Masson SAS. All rights reserved.
Keywords: Vegetation index Ecological factor Correlation analysis Spatial scale effect Semi-arid area Ecotone
1. Introduction Derived from remote sensing images, Normalized Difference Vegetation Index (NDVI) is an accurate indicator of vegetation degradation detection (Evans and Geerken, 2004; Wessels et al., 2012; Karnieli et al., 2013) for riparian landscapes (Mcfarland et al., 2012), sandy land (Zhang et al., 2012), or savanna ecosystems (Jacquin et al., 2010). Several ecological factors can affect vegetation degradation, such as rainfall, temperature, elevation, topological features, and human disturbance (Zhou et al., 2014; Rishmawi and Prince, 2016; Sun et al., 2017). Identification of the main drivers determining vegetation degradation from these biophysical, geographical, and climatic factors has been a core research issue in recent decades. From rainforests to temperate forests,
* Corresponding author. E-mail address: [email protected] (Y. Peng). https://doi.org/10.1016/j.actao.2017.09.011 1146-609X/© 2017 Elsevier Masson SAS. All rights reserved.
semi-arid, or arid regions, the relationship between NDVI indices and ecological factors (including human activities, biophysical, geographical, and climatic factors) are used throughout the world to identify key factors(Pueyo et al., 2006; Paudel and Andersen, 2010; Wessels et al., 2012; Cao et al., 2014; Tian et al., 2014; Zhou et al., 2014). Under most circumstances, human activities are found to be primary factors underlying vegetation degradation, followed by climatic conditions. As in a central Mongolian grassland within the past decades, human impact (most probably, overgrazing and farmland abandonment) is the main factor for vegetation degradation at a large spatial scale, whereas climate change and soil erosion play subordinate roles (Tian et al., 2014). To the south of the Mongolian grassland, there is large-sized semi-arid vegetation in northwest China (five provinces), in which 65.75% of grassland degradation was caused by human activities, whereas 19.94% was caused by inter-annual climate change (Zhou et al., 2014) from 2001 to 2010. Human activities were the main factors inducing
Y. Peng et al. / Acta Oecologica 85 (2017) 62e68
NDVI variation in the Ordos arid region in south-northwest China (Zhang et al., 2014). In addition, a small-scale study in a Mongolian grassland indicated that as the distance from human settlements increased, the vegetation degradation gradient changed from medium to low (Jia et al., 2011), which demonstrated that human activities are the main factors driving vegetation degradation at a small spatial scale as well. However, human activities can enhance vegetation index, as seen in various case studies. Agricultural development in central Mongolia has greatly increased NDVI in the past decades (Tian et al., 2014). In northwest China, successful restoration projects can improve vegetation NDVI in numerous areas of the Ordos sandy land (Zhang et al., 2014). A study in northwest China (five provinces) indicated 32.32% of grassland restoration was caused by human activities, whereas 56.56% was caused by climatic factors. Obviously, human activities made more contributions to grassland NDVI increase compared with climate change (Zhou et al., 2014). Therefore, human activities result in positive (land restoration and re-vegetation) or negative (degradation) correlation with vegetation index. Among climatic factors, precipitation is the most influential factor determining vegetation coverage, as revealed by NDVI. In semi-arid grassland ecosystems, principal drivers of NDVI are precipitation amount and timing (He, 2014). Furthermore, the same trends are found in southern Africa (Wessels et al., 2012), Nepal (Paudel and Andersen, 2010), and the Qinghai Lake Basin in China from 2001 to 2010 (Guo et al., 2014). Other climatic factors determining vegetation degradation include average temperature, as in Qinghai Lake Basin (Guo et al., 2014). Moreover, elevation is a primary ecological factor determining vegetation characteristics, such as in the upper catchments of the Yellow River in China (Cao et al., 2014). Previous studies identified other ecological factors contributing to vegetation degradation, including slope, aspect, and soil properties (Pueyo et al., 2006; Cao et al., 2014). As for the above-mentioned studies, correlation of ecological factors with NDVI has been conducted in different study sites at different spatial scales, ranging from as small as a village to a median regional grassland scale or a large scale covering several provinces. However, few studies have been carried out to examine the effects of spatial scaling-up on the relationships between ecological factors and NDVI until date. Whether correlations depend on spatial scale and the behaviors and ecological factors within different spatial scales affecting NDVI are not well known. In numerous studies on landscape ecology, ecological effects caused € pfner and by spatial scale have been systematically evaluated (Ho Scherer, 2011; Cunningham et al., 2014). Stefanov and Netzband (2005) found that at spatial extents of 250, 500, and 1000 m, weak positive and negative correlations between NDVI and landscape metrics were present when arid landscape characteristics were assessed for metropolitan Phoenix. High spatial heterogeneity in eco-environmental factors such as slope, aspect, rainfall and land use categories contributed to spatial heterogeneity in ecological conditions. Besides, spatial scale effects have been taken into account for exploring the correlation between vegetation index and response factors, e.g., bird community (Cunningham et al., 2014), surface temperature (Chen et al., 2012), or vegetation pattern (Bradter et al., 2011; Danz et al., 2011). The results reveal that the relationship between vegetation index and ecological factors varies greatly with spatial scales. Spatial scale effects may affect the relationship between vegetation index and ecological factors, which have, however, been seldom mentioned in previous studies. The aim of the present study is to explore the correlations of vegetation index (indicated as NDVI) with ecological factors (annual average precipitation, annual average temperature, elevation, land slope, aspect, land curvature and such human activity
63
properties as land use type and human disturbance intensity) along a series of spatial scales. Among these analyses, the effects of spatial scales on the correlations between ecological factors and NDVI were identified, and the key ecological factors determining vegetation index at different spatial scales were determined in a semiarid agro-pastoral ecotone in northern China. 2. Study area and methods Through the application of various ecological factor maps, effects of spatial scale on the correlation of ecological factors with NDVI were tested for key factors determining vegetation degradation along spatial scale gradient. There were five steps in the process: (i) generating land use landscape and NDVI maps by combining remote sensing image interpretation and field surveying; (ii) based on collected data during step (i), ecological factor maps with the same grain size as NDVI map were generated; (iii) ecological factor and NDVI maps with spatial scales of 300, 600, 1000, 2000, and 5000 m respectively were generated, (iv) in the same spatial extent, NDVI map as overlaid with ecological factor maps and the correlations of ecological factors with NDVI were analyzed, and (v) main factors inducing vegetation degradation were identified. 2.1. Study area Located in the middle of an agro-pastoral ecotone, Helin County, central Inner Mongolia, China, the study region possesses several advantages that make it appropriate for the study of spatial scale effects. As the interface between farmland and pasture, this agropastoral ecotone with total area of 3401 km2 involves various ecological factors and vegetation with high heterogeneous spatial distribution, which is believed to be one of the most eco-sensitive regions responsive to climate and human disturbances. Therefore, it is an ideal area for studying the correlation between ecological factors and vegetation. With relatively complex composition, the region is characterized by a collection of flat plains, hills and mountains (Fig. 1). The highest elevation is 2031 m. With obvious wet and dry seasons, Helin County has a semi-arid temperate climate. Its annual average temperature is 5.6 C with a seasonal average temperature of 12.8 C in January and 22.1 C in July. Furthermore, the average annual precipitation is 417 mm with a precipitation of approximately 30 mm in January and 103 mm in July. When compared with summer and fall, the average wind speeds are slightly higher in spring and winter. Besides, no obvious seasonal changes are observed in the average relative humidity for the whole year. Sandy biological communities are supported by the semi-arid climate and grass and shrubs are predominant in this area. Helin County is composed of 9 towns with a population of 0.187 million people and the main income of local people comes from agricultural product and livestock resources. 2.2. Generating ecological factor and NDVI maps The data of the 2010 LULC of the Helin County area were employed, which were produced from a supervised classification model of atmospherically corrected and geo-rectified Landsat Thematic Mapper (hereafter referred to as TM) imagery. Based on field survey data and two Landsat images acquired in July of 2010, the model was originally developed. By adopting Maximum Likelihood Classification, a posteriori sorting of classes initially derived was performed by the classification system. Besides, geographical auxiliary map layers such as land-use maps, image textures and administrative maps were also applied. There were 7 categories in the final land use classification with a reported overall accuracy of
64
Y. Peng et al. / Acta Oecologica 85 (2017) 62e68
Fig. 1. Location and land use distribution of the study area in Helin Conty, northern China. The figure also shows spatial extents ranging from 300, 600, 1000, 2000e5000 m2.
Table 1 Seven LULC categories were adopted to show the land use/land cover in the study area in Helin County., Inner Mongolia, China. Categories
Descriptions
Grassland Farmland Shrub Forest Urban Water Barren land
Pasture, and high or short grassland Row crops, cereal grains, potato, feedlots, vineyards Shrub, semi-shrub, scatter shrub, bushwood, low or high shrub Open or closed forest, including young, old forest Residential, industrial, commercial and recreation areas, roads and railroads Rivers and streams, ponds, lakes and reservoirs Soil or sandy dunes without vegetation
~92% (Table 1). In order to estimate the spatial distribution of abundant vegetation, NDVI was computed from all raw Landsat images: NDVI ¼ (NIRRED)/(NIR þ RED). NDVI was calculated as the mean values between the onset of leaf development (May 1) and leaf senescence (September 30) during 2010, same year as land use map used. Annual average precipitation, temperature and humidity were calculated based on the corresponding yearly maps dating from 2000 to 2010 produced by the Meteorological Administration of Helin County. On account of their potential influences on NDVI, the following information was extracted: aspect (i.e., the slope facing direction), being divided in eight directions from northing to easting (i.e., respectively cosine and sine-transformed azimuth values); slope (i.e., steepness) and elevation and they were all
derived from the SRTM30 DEM (http://asterweb.jpl.nasa.gov/gde m.asp) in Helin County. The entire study area was located in the northern hemisphere, in which southern aspects received significantly higher radiation and more xeric conditions when compared with northern aspects, particularly when associated with steep slopes. Western aspects were dominantly exposed to westerly winds and this was related to higher precipitation and frosts frequency. Spatial distribution maps of human settlements and roads were digitized from the Helin Map produced by Sino Maps Press in 2010. Based on the experienced analysis of relationships between human activity intensity and ecological patterns or processes, human activities variables maps were created, of which, one illustrated the distance to human settlements and the other illustrated the
Y. Peng et al. / Acta Oecologica 85 (2017) 62e68
distance to roads. Since the data were provided in vector format, they were directly rasterized to match the grain size of each level of analysis. According to the aforementioned sources and creative method, eight map layers were created with the same grain size of 30 m, including NDVI, annual average precipitation, average temperature, slope, aspect, elevation, land use, and human disturbance maps. 2.3. Generating multiple spatial scale maps Concerning the whole Helin County area, all maps in vector format were converted into an Arc Grid format for a set of synoptic metric analyses that provided a basic representation of landscape pattern. In view of edge effects, grid datasets for the entire regional area and outer 5 km buffer zone were included in the analysis. Based on 30 m grain size maps, multiple spatial scales were respectively created by means of resample method under ArcGIS 10.0 software. The spatial scale was 300 300, 600 600, 1000 1000, 2000 2000 and 5000 5000 m2. When creating multiple spatial scale maps, the value for each quadrate was calculated as mean value yielded through resampling process. As for land use map, it was not reasonable to produce land use categories code value through mean value calculation, therefore the value was yielded by “majority” resample method. 2.4. Correlation analysis between ecological factors and NDVI The same scale maps and ecological factor and NDVI maps were overlapped with the same project system and coordination system on ArcGIS desktop. In order to extract attribute tables, resample maps were converted into vector format. Furthermore, attribute tables of ecological factors and NDVI maps were combined into an entire table, so as to guarantee that they were analyzed in the same spatial location for each quadrate (analyzing unit). In the next step, the combined dbf format table was converted into Microsoft Excel format for further calculation. In order to test for normality (a basic requirement for further application of parametric tests), all the variables in the exported xls database were submitted to a ShapiroeWilk test. And then, all the variables were logarithmically transformed. Consequently, units choice does not play any role (as long as various units are linearly related). In this study, Redundancy Analysis (hereafter referred to as RDA) was adopted to study the correlations between NDVI and ecological factors. As one form of asymmetric canonical analysis, RDA is widely employed by ecologists and palaeoecologists. ‘Asymmetric’ refers to that the two data matrices applied in the analysis do not play the same role: a matrix of response variables, which is denoted as Y and often contains vegetation index data and a matrix of explanatory variables (e.g. ecological factors), which is denoted as X and used to explain the variation in Y, as in regression analysis. In order to test the combined contribution of ecological variables to NDVI variability, RDA was conducted with NDVI as dependent variables, human disturbance as explanatory variables and ecological factor variables, such as annual average precipitation, annual average temperature, elevation, aspect, slope and land use. Moreover, RDA was performed through the application of Canoco software for Windows 4.5 (Leps and Smilauer, 2003). 3. Results Table 2 shows that rainfall and elevation have a significant positive relationship with NDVI at all spatial scales, whereas aspect has no significant relationship with NDVI across scales. Contrary to our expectation, temperature has a significant negative relationship with NDVI along all scales. However, the 300-m spatial scale is
65
Table 2 Pearson correlations between NDVI and ecological factors showing correlation coefficients and significance level along all spatial scales. NDVI
300 m
600 m
1000 m
2000 m
5000 m
Aspect Disturb Elevat Landuse Rainfall Slope Tempe
0.0657 0.2590* 0.1880** 0.0231** 0.2690** 0.0507 0.1590
0.0485 0.152* 0.0990* 0.1015 0.2150** 0.0750 0.1814*
0.0090 0.0963 0.1741** 0.2329** 0.1220* 0.1054* 0.2152*
0.0210 0.0343 0.2063* 0.1270 0.1662* 0.1670 0.2770**
0.0460 0.1092 0.2104* 0.0829 0.2808** 0.1535 0.3182**
**Correlation is significant at the 0.01 level (2-tailed), * Correlation is significant at the 0.05 level (2-tailed). Disturb ¼ Human disturbance, Elevat ¼ Elevation, Landuse ¼ Land use, Tempe ¼ Temperature. Same as in Fig. 2.
excluded. As demonstrated by the effect of distance from human settlements and land use categories, human activities have not reached a significant level to have an effect on vegetation degradation and can positively enhance vegetation index at a larger scale. RDA in Fig. 2 indicated that the main factors affecting NDVI varied along spatial scales. Within the spatial scales of 300 m and 600 m, human disturbance is the first main negative factor. With the increase in spatial scale, this negative factor weakens to serve as the second negative factor in an extent of 1000 m and nearly no correlation in the spatial scale of 2000 m. Furthermore, it turns into a considerable positive factor in a spatial scale of 5000 m. In the present study, disturbance factor is represented by distance from human settlements and roads, which demonstrates that vegetation is mainly affected by uncertain human disturbance at small spatial scales. As the analyzed spatial scale increases, such a disturbance becomes a minor factor affecting vegetation. In contrast to human disturbance, aspect shows an opposite trend, with a negative effect on vegetation. In addition, with an increase in spatial scale to 2000 and 5000 m, it moves towards positive correlation. Unlike the two ecological factors mentioned above, land use exhibits a different response to spatial scale: it has a considerably strong effect on NDVI at a moderate scale of 1000 m and low effect at smaller or larger scales. Elevation and rainfall have considerable effects on vegetation across all spatial extents and play a secondary role at an extent of 1000 m. Contrary to our expectation, increased temperature exerts a negative effect on vegetation significantly at all spatial extents. In short, elevation, rainfall, and temperature are the main factors at all spatial scales, and the significances differ with spatial scales. 4. Discussion The study area, a semi-arid temperate agro-pastoral ecotone was particularly favorable for exploring key factors determining vegetation index. In this area, rich spatial heterogeneity in landform (elevation, slope, and aspect), climatic conditions (temperature and precipitation), and human activities (distances from human settlements and land use types) were involved. With an elevation ranging from 1000 to 1977 m above mean sea level, the landscape was consisted of plains, valleys, hills, and mountains. Besides, high landforms were accompanied by different severities of slopes ranging from 0 to 34 . Except the diversity in natural ecological conditions, human activities such as agricultural development were dominant in this area. More than 80 human settlements (towns and villages) with a relatively even spatial distribution were present, which were connected by a complicated network of local roads and walking paths. High spatial heterogeneities in ecological factors existed in this area, which provided ideal conditions for analyzing correlations between NDVI and
66
Y. Peng et al. / Acta Oecologica 85 (2017) 62e68
Fig. 2. RDA graphics shows the relationships between ecological factors and NDVI in five spatial scales (300, 600, 1000, 2000 and 5000 m). The names of ecological factors in the figure are same as those in Table 2.
ecological factors. The usage of RDA on NDVI-land use pairs can overcome the variability among the various land use types, thus ensuring the credibility of the results. This study indicates that key factors affecting NDVI are different at various spatial scales. Elevation, rainfall, and temperature are crucial for all spatial extents. Previous studies have demonstrated that rainfall is a factor strongly affecting NDVI in semi-arid or arid areas (Guo et al., 2014; He, 2014). Because of rainfall, NDVI can be significantly enhanced by a margin of 30e40% in degraded arid areas (Wessels et al., 2012). Not only a fine-resolution NDVI image, but also a coarse one shows this positive effect (Paudel and Andersen, 2010). A case study argued that vegetation spatial distribution is highly correlated with elevation ranging from approximately 2000 to 3000 m (Cao et al., 2014). In the present study, high elevation and northerly or westerly aspects were associated with higher availability of moisture. Therefore, a number of locations across the study area possessed barren land, whereas others had fertile land. Under most conditions, different levels of soil moisture led to different vegetation cover, owing to different precipitation levels. Particularly in semi-arid areas, vegetation cover depended
heavily on the amount of precipitation, which was highly variable across different landscapes (slopes, landform, and elevations). In addition, the steepest slopes generally occurring at the high elevations, far from human disturbances, had a positive effect on tree growth. Therefore, NDVI values were usually the highest at locations with the highest elevations. Because of its effect on lateral water redistribution, water table, soil moisture, and plant growth, slope is an important topographical factor affecting vegetation index in mountainous areas. Characterized by steep slopes, wet and deep ditches, and a dense network of valleys, hilly terrains of this study region may have resulted in a large number of refuges for small groups of shrubs or forests, protected from heavy grazing or timber harvesting. Consequently, the present study shows a positive correlation with NDVI across all spatial scales. Rainfall and elevation affect vegetation index significantly across all spatial scales. Elevation, rainfall, and human disturbance are key factors for small-scale quadrats of 300 m 300 m and 600 m 600 m in this area. Human disturbance demonstrated negative effects on NDVI at these scales. Almost every quadrate was completely dominated by a
Y. Peng et al. / Acta Oecologica 85 (2017) 62e68
single ecological factor, for example human disturbance and land use type at a small quadrate, which resulted in a corresponding strong correlation with NDVI. At this scale, the plant habitat in one quadrate was randomly affected by various factors, such as wind, snow, fire, and livestock. Furthermore, the dominant ecological factor varied among quadrates owing to spatial heterogeneity in ecological conditions. In addition, plant species differed greatly among quadrates indicating that ecological conditions required by plant species were differed greatly. Therefore, rather than indicating the relationship with vegetation, the correlation mainly indicate the relationship with dominant species (Bradter et al., 2011). Consequently, one ecological factor could show a strong relationship with NDVI in one quadrate, which however hardly showed a strong influence on vegetation in other quadrates. At a small spatial scale of 300 m or 600 m, the dominant ecological factor was generally land use or land cover type. In the present study, land use was often shown as farmland or grassland (mainly controlled by human activities) and shrub land or forest land (mainly controlled by natural conditions). Land use or land cover type can determine the vegetation index directly. In grasslands studied, human disturbance at small scale (livestock grazing) can decrease the vegetation index directly (Peng et al., 2007). Dominant or combined effects of other ecological factors have been produced on vegetation within one quadrate. Gathering uneven stream flow through slope and gaining different radiation and precipitation through aspect, topography alters the water accumulation, and this spatial heterogeneity exerts indirect effects on the spatial pattern of vegetation index. At a spatial scale of 300 m, such effects are more obvious owing to high variance in topography. With increase in spatial scale, the effects of slopes decrease due to the average slope calculation and linear relationships are reduced. At a small spatial scale, standard deviations of the effects of ecological factors remain swing. As spatial scale increases, variations in the effects of ecological factors decrease consistently and reach a relative steady state. Furthermore, the dominant ecological factor can display its dominant effect on vegetation. In larger quadrates, ecological variables are likely to be correlated with surrounding dominant vegetation, and thus habitat requirements of vegetation can be represented (Bradter et al., 2011). From this point of view, it is likely that a moderate or large spatial scale is selected for the analysis of the correlation between ecological factors and NDVI. Nevertheless, land use type in one quadrate will increase, topological features will become more complicated and variations among or within different ecological factors increase with the increase in spatial scale. Correspondingly, the complication of exploring the correlation between ecological factors and NDVI is increased by these high variations. At a large spatial scale, human activities, such as urban or agricultural development, can determine vegetation directly, which shows that land use has a significant correlation with NDVI at a large spatial scale. In addition, rainfall and elevation strongly control vegetation at this scale. In the present study, human disturbance at a large spatial scale can enhance NDVI by means of revegetation. Southwest to our study area, the same correlations (areas of more than 5 km2) are demonstrated in Ordos Sandland (Zhang et al., 2014). However, forest plantation of coniferous species is one of main land uses in this study region, which is the main cause for vegetation expansion. By considering trade-offs among different ecological factors, the moderate scale of 1000 m may be generally suitable as an optional scale for ecological restoration or land development, in which NDVI and ecological factors reach a relatively stabilized correlation. At a large scale (five provinces in western China), approximately 20e30% of the vegetation possesses a significant correlation with
67
climatic factors, in which precipitation and temperature are regarded as key influencing factors. In addition, more than 30% of the vegetation has an insignificant correlation with climatic factors (Zhou et al., 2014). In the present study, the vegetation is concentrated on western and northern aspects. Besides, the areas of high elevations have local environmental conditions, such as higher levels of direct insulation and higher moisture, which may have facilitated plant seedling establishment and survival. In this study, main factors determining vegetation index were found to be mean annual temperature, elevation, and average annual precipitation, which showed minor differences as compared with that in previous studies, in which the main factors affecting vegetation were human activities (Pueyo et al., 2006; Tian et al., 2014; Zhou et al., 2014). Such differences occurred among studies conducted at different spatial scales, either in different areas or in the same area. Thus, it is crucial to consider the effects of spatial scale while identifying the key ecological factors affecting vegetation. Scale effects should be taken into consideration for vegetation or geographical unit management. At least for this area, at a small scale, measurements should focus on human disturbance, livestock grazing, and other random factors. At a moderate spatial scale, the focus should be transferred to human activities, such as land use (plantation, cultivation, and urban construction). At a larger scale, whether ecological factors such as rainfall, temperature, and elevation are suitable for vegetation restoration should be considered first before undertaking vegetation restoration projects. 5. Conclusion All ecological factors, including both natural and humaninduced disturbances, contribute to vegetation index and significant differences were found among them. Within all scales, rainfall, elevation, and temperature are the main factors determining the vegetation index. In a small scale, human disturbance produces a considerable negative effect on NDVI, which however exerts a positive effect in a large scale. In addition, different ecological factors have different effects on NDVI across spatial scales. Elevation, rainfall, and human disturbance are key factors for small-scale quadrats of 300 m 300 m and 600 m 600 m, temperature and land use type are key factors for a medium-scale quadrats of 1 km 1 km, and rainfall, temperature, and land use are key factors for large-scale quadrats of 2 km 2 km and 5 km 5 km. Due to the spatial quality of remote sensing data, processing methods, and the resolution and spatial scale, adopting remote sensing techniques to assess the scaling effects on the relationship between ecological factors and NDVI remains a challenging task. Our case study indicates that when compared with other spatial scales, employing the spatial extent of 1000 m can harvest relatively steady values with smaller scale effects and thus gain relatively reliable results in semi-arid ecotones. It is reasonable to consider the effects of spatial scale when key factors determining vegetation characteristics are identified. In addition, spatial scale effects should be considered in ecological restoration projects for sites with degraded vegetation. Author contributions Yu Peng designed and performed the study. All other authors collected and pre-processed the field data. All authors discussed the results and contributed to the manuscript. Acknowledgments The study was financially supported by the Top Discipline and
68
Y. Peng et al. / Acta Oecologica 85 (2017) 62e68
First-class University Construction Project (ydzxxk201618) of Minzu University of China; The 2011 project from the Collaborative Innovation Centre for Ethnic Minority Development in China (10301-017004040501); the Undergraduate Research and Training Program (URTP2017110024) of Minzu University of China. We want express our thanks to the two anonymous reviewers for their innovative and supportive comments. References Bradter, U., Thom, T.J., Altringham, J.D., Kunin, W.E., Benton, T.G., 2011. Prediction of national vegetation classification communities in the British Uplands using environmental data at multiple spatial scales, aerial images and the classifier random forest. J. Appl. Ecol. 48 (4), 1057e1065. Cao, R., Jiang, W., Yuan, L., Wang, W., Lv, Z., Chen, Z., 2014. Inter-annual variations in vegetation and their response to climatic factors in the upper catchments of the Yellow River from 2000 to 2010. J. Geog. Sci. 24, 963e979. Chen, X., Yamaguchi, Y., Chen, J., Shi, Y., 2012. Scale effect of vegetation-index-based spatial sharpening for thermal imagery: a simulation study by aster data. IEEE Geosci. Remote S 9 (4), 549e553. Cunningham, R., Lindenmayer, D., Barton, P., Ikin, K., Crane, M., Michael, D., Okada, S., Gibbons, P., Stein, J., 2014. Cross-sectional and temporal relationships between bird occupancy and vegetation cover at multiple spatial scales. Ecol. Appl. 24, 1275e1288. Danz, N.P., Reich, P.B., Frelich, L.E., Niemi, G.J., 2011. Vegetation controls vary across space and spatial scale in a historic grassland-forest biome boundary. Ecography 34 (3), 402e414. Evans, J., Geerken, R., 2004. Discrimination between climate and human-induced dryland degradation. J. Arid. Environ. 57 (4), 535e554. Guo, W., Ni, X., Jing, D., Li, S., 2014. Spatial-temporal patterns of vegetation dynamics and their relationships to climate variations in Qinghai Lake basin using Modis time-series data. J. Geogr. Sci. 24, 1009e1021. He, Y., 2014. The effect of precipitation on vegetation cover over three landscape units in protected semi-arid grassland: temporal dynamics and suitable climatic index. J. Arid. Environ. 109 (5), 74e82. €pfner, C., Scherer, D., 2011. Analysis of vegetation and land cover dynamics in Ho north-western Morocco during the last decade using MODIS NDVI time series data. Biogeosciences 8, 3359e3373. Jacquin, A., Sheeren, D., Lacombe, J.P., 2010. Vegetation cover degradation assessment in Madagascar savanna based on trend analysis of MODIS NDVI time series. Int. J. Appl. Earth Obs. 12 (S1), S3eS10. Jia, Z.Q., Wang, Y.G., Yang, X.H., 2011. Small-scale vegetation changes around a single
settlement site in a semi-arid steppe in China: a degradation gradient pattern. J. Food Agric. Environ. 9 (1), 671e675. Karnieli, A., Bayarjargal, Y., Bayasgalan, M., Mandakh, B., Dugarjav, C., Burgheimer, J., et al., 2013. Do vegetation indices provide a reliable indication of vegetation degradation? A case study in the Mongolian Pastures. Int. J. Remote Sens. 34, 6243e6262. Leps, J., Smilauer, P., 2003. Multivariate Analysis of Ecological Data Using CANOCO. Cambridge University Press, Cambridge, UK. Mcfarland, T.M., Van Riper III, C., Johnson, G.E., 2012. Evaluation of NDVI to assess avian abundance and richness along the upper San Pedro River. J. Arid. Environ. 77, 45e53. Paudel, K.P., Andersen, P., 2010. Assessing rangeland degradation using multi temporal satellite images and grazing pressure surface model in Upper Mustang, Trans Himalaya, Nepal. Remote Sens. Environ. 114 (8), 1845e1855. Peng, Y., Jiang, G.M., Liu, X.H., Niu, S.L., et al., 2007. Photosynthesis, transpiration and water use efficiency of four plant species with grazing intensities in Hunshandak Sandland, China. J. Arid. Environ. 70, 304e315. Pueyo, Y., Alados, C.L., Barrantes, O., 2006. Determinants of land degradation and fragmentation in semiarid vegetation at landscape scale. Biodivers. Conserv. 15, 939e956. Rishmawi, K., Prince, S.D., 2016. Environmental and anthropogenic degradation of vegetation in the Sahel from 1982 to 2006. Remote Sens. 8, 948e975. Stefanov, W.L., Netzband, M., 2005. Assessment of ASTER land-cover and MODIS NDVI data at multiple scales for ecological characterization of an arid urban center. Remote Sens. Environ. 99 (1e2), 31e43. Sun, L., Yang, Z.L., Hao, L., Fang, D., Jin, K.L., et al., 2017. Hydrological effects of vegetation cover degradation and environmental implications in a semiarid temperate steppe, China. Sustainability 9, 281e302. Tian, F., Herzschuh, U., Mischke, S., 2014. What drives the recent intensified vegetation degradation in Mongolia e climate change or human activity? Holocene 24, 1206e1215. Wessels, K.J., van den Bergh, F., Scholes, R.J., 2012. Limits to detectability of land degradation by trend analysis of vegetation index data. Remote Sens. Environ. 125 (4), 10e22. Zhang, G.L., Dong, J.W., Xiao, X.M., Hu, Z.M., Sheldon, S., 2012. Effectiveness of ecological restoration projects in Horqin Sandy Land, China based on SPOT-VGT NDVI data. Ecol. Eng. 38 (1), 20e29. Zhang, J., Niu, J., Bao, T., Buyantuyev, A., Zhang, Q., et al., 2014. Human induced dryland degradation in Ordos Plateau, China, revealed by multilevel statistical modeling of normalized difference vegetation index and rainfall time-series. J. Arid. Land 6 (2), 219e229. Zhou, W., Gang, C., Zhou, L., Chen, Y., Li, J., et al., 2014. Dynamic of grassland vegetation degradation and its quantitative assessment in the Northwest China. Acta Oecol 55, 86e96.
Contents lists available at ScienceDirect
Acta Oecologica journal homepage: www.elsevier.com/locate/actoec
Identification of the key ecological factors influencing vegetation degradation in semi-arid agro-pastoral ecotone considering spatial scales Yu Peng*, Qinghui Wang, Min Fan College of Life & Environmental Sciences, Minzu University of China, Beijing 100081, China
a r t i c l e i n f o
a b s t r a c t
Article history: Received 3 February 2017 Received in revised form 17 September 2017 Accepted 19 September 2017
When assessing re-vegetation project performance and optimizing land management, identification of the key ecological factors inducing vegetation degradation has crucial implications. Rainfall, temperature, elevation, slope, aspect, land use type, and human disturbance are ecological factors affecting the status of vegetation index. However, at different spatial scales, the key factors may vary. Using Helin County, Inner-Mongolia, China as the study site and combining remote sensing image interpretation, field surveying, and mathematical methods, this study assesses key ecological factors affecting vegetation degradation under different spatial scales in a semi-arid agro-pastoral ecotone. It indicates that the key factors are different at various spatial scales. Elevation, rainfall, and temperature are identified as crucial for all spatial extents. Elevation, rainfall and human disturbance are key factors for small-scale quadrats of 300 m 300 m and 600 m 600 m, temperature and land use type are key factors for a medium-scale quadrat of 1 km 1 km, and rainfall, temperature, and land use are key factors for large-scale quadrats of 2 km 2 km and 5 km 5 km. For this region, human disturbance is not the key factor for vegetation degradation across spatial scales. It is necessary to consider spatial scale for the identification of key factors determining vegetation characteristics. The eco-restoration programs at various spatial scales should identify key influencing factors according their scales so as to take effective measurements. The new understanding obtained in this study may help to explore the forces which driving vegetation degradation in the degraded regions in the world. © 2017 Elsevier Masson SAS. All rights reserved.
Keywords: Vegetation index Ecological factor Correlation analysis Spatial scale effect Semi-arid area Ecotone
1. Introduction Derived from remote sensing images, Normalized Difference Vegetation Index (NDVI) is an accurate indicator of vegetation degradation detection (Evans and Geerken, 2004; Wessels et al., 2012; Karnieli et al., 2013) for riparian landscapes (Mcfarland et al., 2012), sandy land (Zhang et al., 2012), or savanna ecosystems (Jacquin et al., 2010). Several ecological factors can affect vegetation degradation, such as rainfall, temperature, elevation, topological features, and human disturbance (Zhou et al., 2014; Rishmawi and Prince, 2016; Sun et al., 2017). Identification of the main drivers determining vegetation degradation from these biophysical, geographical, and climatic factors has been a core research issue in recent decades. From rainforests to temperate forests,
* Corresponding author. E-mail address: [email protected] (Y. Peng). https://doi.org/10.1016/j.actao.2017.09.011 1146-609X/© 2017 Elsevier Masson SAS. All rights reserved.
semi-arid, or arid regions, the relationship between NDVI indices and ecological factors (including human activities, biophysical, geographical, and climatic factors) are used throughout the world to identify key factors(Pueyo et al., 2006; Paudel and Andersen, 2010; Wessels et al., 2012; Cao et al., 2014; Tian et al., 2014; Zhou et al., 2014). Under most circumstances, human activities are found to be primary factors underlying vegetation degradation, followed by climatic conditions. As in a central Mongolian grassland within the past decades, human impact (most probably, overgrazing and farmland abandonment) is the main factor for vegetation degradation at a large spatial scale, whereas climate change and soil erosion play subordinate roles (Tian et al., 2014). To the south of the Mongolian grassland, there is large-sized semi-arid vegetation in northwest China (five provinces), in which 65.75% of grassland degradation was caused by human activities, whereas 19.94% was caused by inter-annual climate change (Zhou et al., 2014) from 2001 to 2010. Human activities were the main factors inducing
Y. Peng et al. / Acta Oecologica 85 (2017) 62e68
NDVI variation in the Ordos arid region in south-northwest China (Zhang et al., 2014). In addition, a small-scale study in a Mongolian grassland indicated that as the distance from human settlements increased, the vegetation degradation gradient changed from medium to low (Jia et al., 2011), which demonstrated that human activities are the main factors driving vegetation degradation at a small spatial scale as well. However, human activities can enhance vegetation index, as seen in various case studies. Agricultural development in central Mongolia has greatly increased NDVI in the past decades (Tian et al., 2014). In northwest China, successful restoration projects can improve vegetation NDVI in numerous areas of the Ordos sandy land (Zhang et al., 2014). A study in northwest China (five provinces) indicated 32.32% of grassland restoration was caused by human activities, whereas 56.56% was caused by climatic factors. Obviously, human activities made more contributions to grassland NDVI increase compared with climate change (Zhou et al., 2014). Therefore, human activities result in positive (land restoration and re-vegetation) or negative (degradation) correlation with vegetation index. Among climatic factors, precipitation is the most influential factor determining vegetation coverage, as revealed by NDVI. In semi-arid grassland ecosystems, principal drivers of NDVI are precipitation amount and timing (He, 2014). Furthermore, the same trends are found in southern Africa (Wessels et al., 2012), Nepal (Paudel and Andersen, 2010), and the Qinghai Lake Basin in China from 2001 to 2010 (Guo et al., 2014). Other climatic factors determining vegetation degradation include average temperature, as in Qinghai Lake Basin (Guo et al., 2014). Moreover, elevation is a primary ecological factor determining vegetation characteristics, such as in the upper catchments of the Yellow River in China (Cao et al., 2014). Previous studies identified other ecological factors contributing to vegetation degradation, including slope, aspect, and soil properties (Pueyo et al., 2006; Cao et al., 2014). As for the above-mentioned studies, correlation of ecological factors with NDVI has been conducted in different study sites at different spatial scales, ranging from as small as a village to a median regional grassland scale or a large scale covering several provinces. However, few studies have been carried out to examine the effects of spatial scaling-up on the relationships between ecological factors and NDVI until date. Whether correlations depend on spatial scale and the behaviors and ecological factors within different spatial scales affecting NDVI are not well known. In numerous studies on landscape ecology, ecological effects caused € pfner and by spatial scale have been systematically evaluated (Ho Scherer, 2011; Cunningham et al., 2014). Stefanov and Netzband (2005) found that at spatial extents of 250, 500, and 1000 m, weak positive and negative correlations between NDVI and landscape metrics were present when arid landscape characteristics were assessed for metropolitan Phoenix. High spatial heterogeneity in eco-environmental factors such as slope, aspect, rainfall and land use categories contributed to spatial heterogeneity in ecological conditions. Besides, spatial scale effects have been taken into account for exploring the correlation between vegetation index and response factors, e.g., bird community (Cunningham et al., 2014), surface temperature (Chen et al., 2012), or vegetation pattern (Bradter et al., 2011; Danz et al., 2011). The results reveal that the relationship between vegetation index and ecological factors varies greatly with spatial scales. Spatial scale effects may affect the relationship between vegetation index and ecological factors, which have, however, been seldom mentioned in previous studies. The aim of the present study is to explore the correlations of vegetation index (indicated as NDVI) with ecological factors (annual average precipitation, annual average temperature, elevation, land slope, aspect, land curvature and such human activity
63
properties as land use type and human disturbance intensity) along a series of spatial scales. Among these analyses, the effects of spatial scales on the correlations between ecological factors and NDVI were identified, and the key ecological factors determining vegetation index at different spatial scales were determined in a semiarid agro-pastoral ecotone in northern China. 2. Study area and methods Through the application of various ecological factor maps, effects of spatial scale on the correlation of ecological factors with NDVI were tested for key factors determining vegetation degradation along spatial scale gradient. There were five steps in the process: (i) generating land use landscape and NDVI maps by combining remote sensing image interpretation and field surveying; (ii) based on collected data during step (i), ecological factor maps with the same grain size as NDVI map were generated; (iii) ecological factor and NDVI maps with spatial scales of 300, 600, 1000, 2000, and 5000 m respectively were generated, (iv) in the same spatial extent, NDVI map as overlaid with ecological factor maps and the correlations of ecological factors with NDVI were analyzed, and (v) main factors inducing vegetation degradation were identified. 2.1. Study area Located in the middle of an agro-pastoral ecotone, Helin County, central Inner Mongolia, China, the study region possesses several advantages that make it appropriate for the study of spatial scale effects. As the interface between farmland and pasture, this agropastoral ecotone with total area of 3401 km2 involves various ecological factors and vegetation with high heterogeneous spatial distribution, which is believed to be one of the most eco-sensitive regions responsive to climate and human disturbances. Therefore, it is an ideal area for studying the correlation between ecological factors and vegetation. With relatively complex composition, the region is characterized by a collection of flat plains, hills and mountains (Fig. 1). The highest elevation is 2031 m. With obvious wet and dry seasons, Helin County has a semi-arid temperate climate. Its annual average temperature is 5.6 C with a seasonal average temperature of 12.8 C in January and 22.1 C in July. Furthermore, the average annual precipitation is 417 mm with a precipitation of approximately 30 mm in January and 103 mm in July. When compared with summer and fall, the average wind speeds are slightly higher in spring and winter. Besides, no obvious seasonal changes are observed in the average relative humidity for the whole year. Sandy biological communities are supported by the semi-arid climate and grass and shrubs are predominant in this area. Helin County is composed of 9 towns with a population of 0.187 million people and the main income of local people comes from agricultural product and livestock resources. 2.2. Generating ecological factor and NDVI maps The data of the 2010 LULC of the Helin County area were employed, which were produced from a supervised classification model of atmospherically corrected and geo-rectified Landsat Thematic Mapper (hereafter referred to as TM) imagery. Based on field survey data and two Landsat images acquired in July of 2010, the model was originally developed. By adopting Maximum Likelihood Classification, a posteriori sorting of classes initially derived was performed by the classification system. Besides, geographical auxiliary map layers such as land-use maps, image textures and administrative maps were also applied. There were 7 categories in the final land use classification with a reported overall accuracy of
64
Y. Peng et al. / Acta Oecologica 85 (2017) 62e68
Fig. 1. Location and land use distribution of the study area in Helin Conty, northern China. The figure also shows spatial extents ranging from 300, 600, 1000, 2000e5000 m2.
Table 1 Seven LULC categories were adopted to show the land use/land cover in the study area in Helin County., Inner Mongolia, China. Categories
Descriptions
Grassland Farmland Shrub Forest Urban Water Barren land
Pasture, and high or short grassland Row crops, cereal grains, potato, feedlots, vineyards Shrub, semi-shrub, scatter shrub, bushwood, low or high shrub Open or closed forest, including young, old forest Residential, industrial, commercial and recreation areas, roads and railroads Rivers and streams, ponds, lakes and reservoirs Soil or sandy dunes without vegetation
~92% (Table 1). In order to estimate the spatial distribution of abundant vegetation, NDVI was computed from all raw Landsat images: NDVI ¼ (NIRRED)/(NIR þ RED). NDVI was calculated as the mean values between the onset of leaf development (May 1) and leaf senescence (September 30) during 2010, same year as land use map used. Annual average precipitation, temperature and humidity were calculated based on the corresponding yearly maps dating from 2000 to 2010 produced by the Meteorological Administration of Helin County. On account of their potential influences on NDVI, the following information was extracted: aspect (i.e., the slope facing direction), being divided in eight directions from northing to easting (i.e., respectively cosine and sine-transformed azimuth values); slope (i.e., steepness) and elevation and they were all
derived from the SRTM30 DEM (http://asterweb.jpl.nasa.gov/gde m.asp) in Helin County. The entire study area was located in the northern hemisphere, in which southern aspects received significantly higher radiation and more xeric conditions when compared with northern aspects, particularly when associated with steep slopes. Western aspects were dominantly exposed to westerly winds and this was related to higher precipitation and frosts frequency. Spatial distribution maps of human settlements and roads were digitized from the Helin Map produced by Sino Maps Press in 2010. Based on the experienced analysis of relationships between human activity intensity and ecological patterns or processes, human activities variables maps were created, of which, one illustrated the distance to human settlements and the other illustrated the
Y. Peng et al. / Acta Oecologica 85 (2017) 62e68
distance to roads. Since the data were provided in vector format, they were directly rasterized to match the grain size of each level of analysis. According to the aforementioned sources and creative method, eight map layers were created with the same grain size of 30 m, including NDVI, annual average precipitation, average temperature, slope, aspect, elevation, land use, and human disturbance maps. 2.3. Generating multiple spatial scale maps Concerning the whole Helin County area, all maps in vector format were converted into an Arc Grid format for a set of synoptic metric analyses that provided a basic representation of landscape pattern. In view of edge effects, grid datasets for the entire regional area and outer 5 km buffer zone were included in the analysis. Based on 30 m grain size maps, multiple spatial scales were respectively created by means of resample method under ArcGIS 10.0 software. The spatial scale was 300 300, 600 600, 1000 1000, 2000 2000 and 5000 5000 m2. When creating multiple spatial scale maps, the value for each quadrate was calculated as mean value yielded through resampling process. As for land use map, it was not reasonable to produce land use categories code value through mean value calculation, therefore the value was yielded by “majority” resample method. 2.4. Correlation analysis between ecological factors and NDVI The same scale maps and ecological factor and NDVI maps were overlapped with the same project system and coordination system on ArcGIS desktop. In order to extract attribute tables, resample maps were converted into vector format. Furthermore, attribute tables of ecological factors and NDVI maps were combined into an entire table, so as to guarantee that they were analyzed in the same spatial location for each quadrate (analyzing unit). In the next step, the combined dbf format table was converted into Microsoft Excel format for further calculation. In order to test for normality (a basic requirement for further application of parametric tests), all the variables in the exported xls database were submitted to a ShapiroeWilk test. And then, all the variables were logarithmically transformed. Consequently, units choice does not play any role (as long as various units are linearly related). In this study, Redundancy Analysis (hereafter referred to as RDA) was adopted to study the correlations between NDVI and ecological factors. As one form of asymmetric canonical analysis, RDA is widely employed by ecologists and palaeoecologists. ‘Asymmetric’ refers to that the two data matrices applied in the analysis do not play the same role: a matrix of response variables, which is denoted as Y and often contains vegetation index data and a matrix of explanatory variables (e.g. ecological factors), which is denoted as X and used to explain the variation in Y, as in regression analysis. In order to test the combined contribution of ecological variables to NDVI variability, RDA was conducted with NDVI as dependent variables, human disturbance as explanatory variables and ecological factor variables, such as annual average precipitation, annual average temperature, elevation, aspect, slope and land use. Moreover, RDA was performed through the application of Canoco software for Windows 4.5 (Leps and Smilauer, 2003). 3. Results Table 2 shows that rainfall and elevation have a significant positive relationship with NDVI at all spatial scales, whereas aspect has no significant relationship with NDVI across scales. Contrary to our expectation, temperature has a significant negative relationship with NDVI along all scales. However, the 300-m spatial scale is
65
Table 2 Pearson correlations between NDVI and ecological factors showing correlation coefficients and significance level along all spatial scales. NDVI
300 m
600 m
1000 m
2000 m
5000 m
Aspect Disturb Elevat Landuse Rainfall Slope Tempe
0.0657 0.2590* 0.1880** 0.0231** 0.2690** 0.0507 0.1590
0.0485 0.152* 0.0990* 0.1015 0.2150** 0.0750 0.1814*
0.0090 0.0963 0.1741** 0.2329** 0.1220* 0.1054* 0.2152*
0.0210 0.0343 0.2063* 0.1270 0.1662* 0.1670 0.2770**
0.0460 0.1092 0.2104* 0.0829 0.2808** 0.1535 0.3182**
**Correlation is significant at the 0.01 level (2-tailed), * Correlation is significant at the 0.05 level (2-tailed). Disturb ¼ Human disturbance, Elevat ¼ Elevation, Landuse ¼ Land use, Tempe ¼ Temperature. Same as in Fig. 2.
excluded. As demonstrated by the effect of distance from human settlements and land use categories, human activities have not reached a significant level to have an effect on vegetation degradation and can positively enhance vegetation index at a larger scale. RDA in Fig. 2 indicated that the main factors affecting NDVI varied along spatial scales. Within the spatial scales of 300 m and 600 m, human disturbance is the first main negative factor. With the increase in spatial scale, this negative factor weakens to serve as the second negative factor in an extent of 1000 m and nearly no correlation in the spatial scale of 2000 m. Furthermore, it turns into a considerable positive factor in a spatial scale of 5000 m. In the present study, disturbance factor is represented by distance from human settlements and roads, which demonstrates that vegetation is mainly affected by uncertain human disturbance at small spatial scales. As the analyzed spatial scale increases, such a disturbance becomes a minor factor affecting vegetation. In contrast to human disturbance, aspect shows an opposite trend, with a negative effect on vegetation. In addition, with an increase in spatial scale to 2000 and 5000 m, it moves towards positive correlation. Unlike the two ecological factors mentioned above, land use exhibits a different response to spatial scale: it has a considerably strong effect on NDVI at a moderate scale of 1000 m and low effect at smaller or larger scales. Elevation and rainfall have considerable effects on vegetation across all spatial extents and play a secondary role at an extent of 1000 m. Contrary to our expectation, increased temperature exerts a negative effect on vegetation significantly at all spatial extents. In short, elevation, rainfall, and temperature are the main factors at all spatial scales, and the significances differ with spatial scales. 4. Discussion The study area, a semi-arid temperate agro-pastoral ecotone was particularly favorable for exploring key factors determining vegetation index. In this area, rich spatial heterogeneity in landform (elevation, slope, and aspect), climatic conditions (temperature and precipitation), and human activities (distances from human settlements and land use types) were involved. With an elevation ranging from 1000 to 1977 m above mean sea level, the landscape was consisted of plains, valleys, hills, and mountains. Besides, high landforms were accompanied by different severities of slopes ranging from 0 to 34 . Except the diversity in natural ecological conditions, human activities such as agricultural development were dominant in this area. More than 80 human settlements (towns and villages) with a relatively even spatial distribution were present, which were connected by a complicated network of local roads and walking paths. High spatial heterogeneities in ecological factors existed in this area, which provided ideal conditions for analyzing correlations between NDVI and
66
Y. Peng et al. / Acta Oecologica 85 (2017) 62e68
Fig. 2. RDA graphics shows the relationships between ecological factors and NDVI in five spatial scales (300, 600, 1000, 2000 and 5000 m). The names of ecological factors in the figure are same as those in Table 2.
ecological factors. The usage of RDA on NDVI-land use pairs can overcome the variability among the various land use types, thus ensuring the credibility of the results. This study indicates that key factors affecting NDVI are different at various spatial scales. Elevation, rainfall, and temperature are crucial for all spatial extents. Previous studies have demonstrated that rainfall is a factor strongly affecting NDVI in semi-arid or arid areas (Guo et al., 2014; He, 2014). Because of rainfall, NDVI can be significantly enhanced by a margin of 30e40% in degraded arid areas (Wessels et al., 2012). Not only a fine-resolution NDVI image, but also a coarse one shows this positive effect (Paudel and Andersen, 2010). A case study argued that vegetation spatial distribution is highly correlated with elevation ranging from approximately 2000 to 3000 m (Cao et al., 2014). In the present study, high elevation and northerly or westerly aspects were associated with higher availability of moisture. Therefore, a number of locations across the study area possessed barren land, whereas others had fertile land. Under most conditions, different levels of soil moisture led to different vegetation cover, owing to different precipitation levels. Particularly in semi-arid areas, vegetation cover depended
heavily on the amount of precipitation, which was highly variable across different landscapes (slopes, landform, and elevations). In addition, the steepest slopes generally occurring at the high elevations, far from human disturbances, had a positive effect on tree growth. Therefore, NDVI values were usually the highest at locations with the highest elevations. Because of its effect on lateral water redistribution, water table, soil moisture, and plant growth, slope is an important topographical factor affecting vegetation index in mountainous areas. Characterized by steep slopes, wet and deep ditches, and a dense network of valleys, hilly terrains of this study region may have resulted in a large number of refuges for small groups of shrubs or forests, protected from heavy grazing or timber harvesting. Consequently, the present study shows a positive correlation with NDVI across all spatial scales. Rainfall and elevation affect vegetation index significantly across all spatial scales. Elevation, rainfall, and human disturbance are key factors for small-scale quadrats of 300 m 300 m and 600 m 600 m in this area. Human disturbance demonstrated negative effects on NDVI at these scales. Almost every quadrate was completely dominated by a
Y. Peng et al. / Acta Oecologica 85 (2017) 62e68
single ecological factor, for example human disturbance and land use type at a small quadrate, which resulted in a corresponding strong correlation with NDVI. At this scale, the plant habitat in one quadrate was randomly affected by various factors, such as wind, snow, fire, and livestock. Furthermore, the dominant ecological factor varied among quadrates owing to spatial heterogeneity in ecological conditions. In addition, plant species differed greatly among quadrates indicating that ecological conditions required by plant species were differed greatly. Therefore, rather than indicating the relationship with vegetation, the correlation mainly indicate the relationship with dominant species (Bradter et al., 2011). Consequently, one ecological factor could show a strong relationship with NDVI in one quadrate, which however hardly showed a strong influence on vegetation in other quadrates. At a small spatial scale of 300 m or 600 m, the dominant ecological factor was generally land use or land cover type. In the present study, land use was often shown as farmland or grassland (mainly controlled by human activities) and shrub land or forest land (mainly controlled by natural conditions). Land use or land cover type can determine the vegetation index directly. In grasslands studied, human disturbance at small scale (livestock grazing) can decrease the vegetation index directly (Peng et al., 2007). Dominant or combined effects of other ecological factors have been produced on vegetation within one quadrate. Gathering uneven stream flow through slope and gaining different radiation and precipitation through aspect, topography alters the water accumulation, and this spatial heterogeneity exerts indirect effects on the spatial pattern of vegetation index. At a spatial scale of 300 m, such effects are more obvious owing to high variance in topography. With increase in spatial scale, the effects of slopes decrease due to the average slope calculation and linear relationships are reduced. At a small spatial scale, standard deviations of the effects of ecological factors remain swing. As spatial scale increases, variations in the effects of ecological factors decrease consistently and reach a relative steady state. Furthermore, the dominant ecological factor can display its dominant effect on vegetation. In larger quadrates, ecological variables are likely to be correlated with surrounding dominant vegetation, and thus habitat requirements of vegetation can be represented (Bradter et al., 2011). From this point of view, it is likely that a moderate or large spatial scale is selected for the analysis of the correlation between ecological factors and NDVI. Nevertheless, land use type in one quadrate will increase, topological features will become more complicated and variations among or within different ecological factors increase with the increase in spatial scale. Correspondingly, the complication of exploring the correlation between ecological factors and NDVI is increased by these high variations. At a large spatial scale, human activities, such as urban or agricultural development, can determine vegetation directly, which shows that land use has a significant correlation with NDVI at a large spatial scale. In addition, rainfall and elevation strongly control vegetation at this scale. In the present study, human disturbance at a large spatial scale can enhance NDVI by means of revegetation. Southwest to our study area, the same correlations (areas of more than 5 km2) are demonstrated in Ordos Sandland (Zhang et al., 2014). However, forest plantation of coniferous species is one of main land uses in this study region, which is the main cause for vegetation expansion. By considering trade-offs among different ecological factors, the moderate scale of 1000 m may be generally suitable as an optional scale for ecological restoration or land development, in which NDVI and ecological factors reach a relatively stabilized correlation. At a large scale (five provinces in western China), approximately 20e30% of the vegetation possesses a significant correlation with
67
climatic factors, in which precipitation and temperature are regarded as key influencing factors. In addition, more than 30% of the vegetation has an insignificant correlation with climatic factors (Zhou et al., 2014). In the present study, the vegetation is concentrated on western and northern aspects. Besides, the areas of high elevations have local environmental conditions, such as higher levels of direct insulation and higher moisture, which may have facilitated plant seedling establishment and survival. In this study, main factors determining vegetation index were found to be mean annual temperature, elevation, and average annual precipitation, which showed minor differences as compared with that in previous studies, in which the main factors affecting vegetation were human activities (Pueyo et al., 2006; Tian et al., 2014; Zhou et al., 2014). Such differences occurred among studies conducted at different spatial scales, either in different areas or in the same area. Thus, it is crucial to consider the effects of spatial scale while identifying the key ecological factors affecting vegetation. Scale effects should be taken into consideration for vegetation or geographical unit management. At least for this area, at a small scale, measurements should focus on human disturbance, livestock grazing, and other random factors. At a moderate spatial scale, the focus should be transferred to human activities, such as land use (plantation, cultivation, and urban construction). At a larger scale, whether ecological factors such as rainfall, temperature, and elevation are suitable for vegetation restoration should be considered first before undertaking vegetation restoration projects. 5. Conclusion All ecological factors, including both natural and humaninduced disturbances, contribute to vegetation index and significant differences were found among them. Within all scales, rainfall, elevation, and temperature are the main factors determining the vegetation index. In a small scale, human disturbance produces a considerable negative effect on NDVI, which however exerts a positive effect in a large scale. In addition, different ecological factors have different effects on NDVI across spatial scales. Elevation, rainfall, and human disturbance are key factors for small-scale quadrats of 300 m 300 m and 600 m 600 m, temperature and land use type are key factors for a medium-scale quadrats of 1 km 1 km, and rainfall, temperature, and land use are key factors for large-scale quadrats of 2 km 2 km and 5 km 5 km. Due to the spatial quality of remote sensing data, processing methods, and the resolution and spatial scale, adopting remote sensing techniques to assess the scaling effects on the relationship between ecological factors and NDVI remains a challenging task. Our case study indicates that when compared with other spatial scales, employing the spatial extent of 1000 m can harvest relatively steady values with smaller scale effects and thus gain relatively reliable results in semi-arid ecotones. It is reasonable to consider the effects of spatial scale when key factors determining vegetation characteristics are identified. In addition, spatial scale effects should be considered in ecological restoration projects for sites with degraded vegetation. Author contributions Yu Peng designed and performed the study. All other authors collected and pre-processed the field data. All authors discussed the results and contributed to the manuscript. Acknowledgments The study was financially supported by the Top Discipline and
68
Y. Peng et al. / Acta Oecologica 85 (2017) 62e68
First-class University Construction Project (ydzxxk201618) of Minzu University of China; The 2011 project from the Collaborative Innovation Centre for Ethnic Minority Development in China (10301-017004040501); the Undergraduate Research and Training Program (URTP2017110024) of Minzu University of China. We want express our thanks to the two anonymous reviewers for their innovative and supportive comments. References Bradter, U., Thom, T.J., Altringham, J.D., Kunin, W.E., Benton, T.G., 2011. Prediction of national vegetation classification communities in the British Uplands using environmental data at multiple spatial scales, aerial images and the classifier random forest. J. Appl. Ecol. 48 (4), 1057e1065. Cao, R., Jiang, W., Yuan, L., Wang, W., Lv, Z., Chen, Z., 2014. Inter-annual variations in vegetation and their response to climatic factors in the upper catchments of the Yellow River from 2000 to 2010. J. Geog. Sci. 24, 963e979. Chen, X., Yamaguchi, Y., Chen, J., Shi, Y., 2012. Scale effect of vegetation-index-based spatial sharpening for thermal imagery: a simulation study by aster data. IEEE Geosci. Remote S 9 (4), 549e553. Cunningham, R., Lindenmayer, D., Barton, P., Ikin, K., Crane, M., Michael, D., Okada, S., Gibbons, P., Stein, J., 2014. Cross-sectional and temporal relationships between bird occupancy and vegetation cover at multiple spatial scales. Ecol. Appl. 24, 1275e1288. Danz, N.P., Reich, P.B., Frelich, L.E., Niemi, G.J., 2011. Vegetation controls vary across space and spatial scale in a historic grassland-forest biome boundary. Ecography 34 (3), 402e414. Evans, J., Geerken, R., 2004. Discrimination between climate and human-induced dryland degradation. J. Arid. Environ. 57 (4), 535e554. Guo, W., Ni, X., Jing, D., Li, S., 2014. Spatial-temporal patterns of vegetation dynamics and their relationships to climate variations in Qinghai Lake basin using Modis time-series data. J. Geogr. Sci. 24, 1009e1021. He, Y., 2014. The effect of precipitation on vegetation cover over three landscape units in protected semi-arid grassland: temporal dynamics and suitable climatic index. J. Arid. Environ. 109 (5), 74e82. €pfner, C., Scherer, D., 2011. Analysis of vegetation and land cover dynamics in Ho north-western Morocco during the last decade using MODIS NDVI time series data. Biogeosciences 8, 3359e3373. Jacquin, A., Sheeren, D., Lacombe, J.P., 2010. Vegetation cover degradation assessment in Madagascar savanna based on trend analysis of MODIS NDVI time series. Int. J. Appl. Earth Obs. 12 (S1), S3eS10. Jia, Z.Q., Wang, Y.G., Yang, X.H., 2011. Small-scale vegetation changes around a single
settlement site in a semi-arid steppe in China: a degradation gradient pattern. J. Food Agric. Environ. 9 (1), 671e675. Karnieli, A., Bayarjargal, Y., Bayasgalan, M., Mandakh, B., Dugarjav, C., Burgheimer, J., et al., 2013. Do vegetation indices provide a reliable indication of vegetation degradation? A case study in the Mongolian Pastures. Int. J. Remote Sens. 34, 6243e6262. Leps, J., Smilauer, P., 2003. Multivariate Analysis of Ecological Data Using CANOCO. Cambridge University Press, Cambridge, UK. Mcfarland, T.M., Van Riper III, C., Johnson, G.E., 2012. Evaluation of NDVI to assess avian abundance and richness along the upper San Pedro River. J. Arid. Environ. 77, 45e53. Paudel, K.P., Andersen, P., 2010. Assessing rangeland degradation using multi temporal satellite images and grazing pressure surface model in Upper Mustang, Trans Himalaya, Nepal. Remote Sens. Environ. 114 (8), 1845e1855. Peng, Y., Jiang, G.M., Liu, X.H., Niu, S.L., et al., 2007. Photosynthesis, transpiration and water use efficiency of four plant species with grazing intensities in Hunshandak Sandland, China. J. Arid. Environ. 70, 304e315. Pueyo, Y., Alados, C.L., Barrantes, O., 2006. Determinants of land degradation and fragmentation in semiarid vegetation at landscape scale. Biodivers. Conserv. 15, 939e956. Rishmawi, K., Prince, S.D., 2016. Environmental and anthropogenic degradation of vegetation in the Sahel from 1982 to 2006. Remote Sens. 8, 948e975. Stefanov, W.L., Netzband, M., 2005. Assessment of ASTER land-cover and MODIS NDVI data at multiple scales for ecological characterization of an arid urban center. Remote Sens. Environ. 99 (1e2), 31e43. Sun, L., Yang, Z.L., Hao, L., Fang, D., Jin, K.L., et al., 2017. Hydrological effects of vegetation cover degradation and environmental implications in a semiarid temperate steppe, China. Sustainability 9, 281e302. Tian, F., Herzschuh, U., Mischke, S., 2014. What drives the recent intensified vegetation degradation in Mongolia e climate change or human activity? Holocene 24, 1206e1215. Wessels, K.J., van den Bergh, F., Scholes, R.J., 2012. Limits to detectability of land degradation by trend analysis of vegetation index data. Remote Sens. Environ. 125 (4), 10e22. Zhang, G.L., Dong, J.W., Xiao, X.M., Hu, Z.M., Sheldon, S., 2012. Effectiveness of ecological restoration projects in Horqin Sandy Land, China based on SPOT-VGT NDVI data. Ecol. Eng. 38 (1), 20e29. Zhang, J., Niu, J., Bao, T., Buyantuyev, A., Zhang, Q., et al., 2014. Human induced dryland degradation in Ordos Plateau, China, revealed by multilevel statistical modeling of normalized difference vegetation index and rainfall time-series. J. Arid. Land 6 (2), 219e229. Zhou, W., Gang, C., Zhou, L., Chen, Y., Li, J., et al., 2014. Dynamic of grassland vegetation degradation and its quantitative assessment in the Northwest China. Acta Oecol 55, 86e96.