- Email: [email protected]
Landscape pattern and transition under natural and anthropogenic disturbance in an arid region of northwestern China
International Journal of Applied Earth Observation and Geoinformation 44 (2016) 1–10
Contents lists available at ScienceDirect
International Journal of Applied Earth Observation and Geoinformation journal homepage: www.elsevier.com/locate/jag
Landscape pattern and transition under natural and anthropogenic disturbance in an arid region of northwestern China Yu Zhang a , Tianwei Wang a,∗ , Chongfa Cai a , Chongguang Li b , Yaojun Liu a,c , Yuze Bao b , Wuhong Guan a a
College of Resources and Environment, Huazhong Agricultural University, Wuhan 430070, China College of Economics & Management, Huazhong Agriculture University, Wuhan 430070, China c Jiangxi Institute of Soil and Water Conservation, Nanchang 330029, China b
a r t i c l e
i n f o
Article history: Received 19 July 2014 Accepted 15 June 2015 Keywords: Landscape transition Driving force Anthropogenic disturbance Redundancy analysis (RDA) Variation partitioning
a b s t r a c t There is a pressing need to determine the relationships between driving variables and landscape transformations. Human activities shape landscapes and turn them into complex assemblages of highly diverse structures. Other factors, including climate and topography, also play significant roles in landscape transitions, and identifying the interactions among the variables is critical to environmental management. This study analyzed the configurations and spatial-temporal processes of landscape changes from 1998 to 2011 under different anthropogenic disturbances, identified the main variables that determine the landscape patterns and transitions, and quantified the relationships between pairs of driver sets. Landsat images of Baicheng and Tekes from 1998, 2006 and 2011 were used to classify landscapes by supervised classification. Redundancy analysis (RDA) and variation partitioning were performed to identify the main driving forces and to quantify the unique, shared, and total explained variation of the sets of variables. The results indicate that the proportions of otherwise identical landscapes in Baicheng and Tekes were very different. The area of the grassland in Tekes was much larger than that of the cropland; however, the differences between the grassland and cropland in Baicheng were not as pronounced. Much of the grassland in Tekes was located in an area that was near residents, whereas most of the grassland in Baicheng was far from residents. The slope, elevation, annual precipitation, annual temperature, and distance to the nearest resident were strong driving forces influencing the patterns and transitions of the landscapes. The results of the variation partitioning indicated complex interrelationships among all of the pairs of driver sets. All of the variable sets had significant explanatory roles, most of which had both unique and shared variations with the others. The results of this study can assist policy makers and planners in implementing sustainable landscape management and effective protection strategies. © 2015 Elsevier B.V. All rights reserved.
1. Introduction Land use in China has changed dramatically since the adoption of economic reforms and the opening-up policy in 1978 (Xu, 2004). Unprecedented urban expansion represents an important type of land transition (Seto and Fragkias, 2005). A large proportion of the land has been degraded by agricultural enterprises and overgrazing. Fragmentation of cropland, due to the construction of the
∗ Corresponding author. E-mail addresses: [email protected] (Y. Zhang), [email protected] (T. Wang), [email protected] (C. Cai), [email protected] (C. Li), [email protected] (Y. Liu), [email protected] (Y. Bao), gwh [email protected] (W. Guan). http://dx.doi.org/10.1016/j.jag.2015.06.013 0303-2434/© 2015 Elsevier B.V. All rights reserved.
countryside (Long and Li, 2005), has caused many concerns regarding China’s food security. The principal consequences of landscape changes are resource depletion and environmental and ecological problems (Imhoff et al., 2004), such as the loss, fragmentation and degradation of the available habitat for most species, which are among the major threats to biodiversity worldwide (Johnson and Zuleta, 2013; Maeda et al., 2011). These problems have had complicated influences on the modernization of China (Liu et al., 2010a). Therefore, efforts to understand the processes of landscape dynamics and driving forces are critical. Human-induced changes to natural landscapes constitute the main driving forces of land-cover change at the local, regional and global scales (Walker et al., 2004; Etter et al., 2006; Wyman and Stein, 2010). Human disturbances, such as pollution, alteration, fire, grazing, cutting, and cultivation, in particular, are complex and
2
Y. Zhang et al. / International Journal of Applied Earth Observation and Geoinformation 44 (2016) 1–10
Fig. 1. Location of the study area.
reflect temporal and spatial shifts in systems of land use (Paudel and Yuan, 2012; Tzanopoulos and Vogiatzakis, 2011). The Random House dictionary (1987) defines culture as the sum total of ways of living developed by a group of human beings and transmitted from one generation to another. Ethnic culture is a complicated entity comprising knowledge, belief, art, morality, regulations, conventions and all types of abilities and customs that the members have acquired (Shi, 1995). In a context where one ethnic group currently represents a small minority of the residents and the residents of another ethnic group make new demands on the environmental, social and cultural functions of landscapes, landscape transitions are often at the root of the tensions between different ethnic groups. Some researchers have attempted to depict the nature, direction, frequency, and rate of change (Krausmann et al., 2003; Lambin et al., 2003); others have focused on describing the driving forces from a historical point of view (Bicík et al., 2001; Bürgi, 1999). However, studies that use systematic methods to investigate how anthropogenic disturbances influence the pattern and change of landscapes with populations of different cultural backgrounds are rare. There are 56 ethnic groups in China, and 47 in Xinjiang Province, including 12 permanent minorities. Baicheng and Tekes are located in Xinjiang Province. Each ethnic group has its own cultural background, making these two counties typical and ideal places for studying human-caused disturbances. The drivers of landscape changes, which are always interrelated and covarying, can be grouped into logical driver sets. Each driver set can be identified as having both unique influences that cannot be attributed to other factors and shared explanatory power due to their interrelated roles (Fisichelli et al., 2013). Attention can be paid to single variables when other covarying factors are partially driving trends (Legendre and Legendre, 1998). The confounded relationships among the factors of landscape changes need to be quantified in order to better understand the mechanism of the variables. In the present study, multi-temporal Landsat images were used to explore landscape changes through analysis of patterns, rates of transition and drivers and quantification of the relationships between pairs of driver sets. The specific objectives were as follows: (a) to characterize the configuration and the main spatial-temporal processes of landscape changes from 1998 to 2011 under different
anthropogenic disturbances, (b) to identify the main variables that determine landscape patterns and transitions, and (c) to examine and quantify the explanatory relationships between anthropogenic disturbances and other driver sets.
2. Materials and methods 2.1. Characteristics of the study sites The study sites were Baicheng and Tekes, two adjacent counties in Xinjiang Province in northwestern China (Fig. 1) and have similar geographical conditions. Baicheng is a county in the northwestern Akesu Prefecture, which is located in the southern foothills in the middle of the Tianshan Mountains. Baicheng lies between latitude 41◦ 24 and 42◦ 51 N and stretches between longitude 80◦ 37 and 83◦ 03 E, with an area of 15,917 km2 and elevations ranging from 1041 to 6289 m. Baicheng has a temperate continental monsoon climate with cold winters and cool summers. According to weather records, the average annual temperature is 7.6 ◦ C and the daily minimum and maximum temperatures are −28 ◦ C and 38.3 ◦ C, respectively. The average annual rainfall is 171.13 mm. Baicheng is surrounded by mountains, forming a banded basin. The geographic conditions are complex, with natural steep slopes. The total population of Baicheng County was 189,000 in 1998 and 230,900 in 2011 (Xinjiang Province Statistical Bureau, 1999, 2012). Tekes is a county in the southwestern Ili Kazak Autonomous Prefecture, which is situated in the northern foothills of the fold belt of Tianshan Mountain. Tekes lies between latitude 42◦ 22 and 43◦ 25 N and stretches between longitude 81◦ 19 and 82◦ 37 E, with an area of 8080 km2 and elevations ranging from 924 to 4853 m. Tekes belongs to the inversion belt control area and has a typical temperate continental climate with an average annual rainfall of 383 mm. The average annual temperature is 5.3 ◦ C, and the daily minimum and maximum temperatures are −32 ◦ C and 33.5 ◦ C, respectively. Three major rivers flow across the county, and three mountains lie from west to east, constituting approximately 94% of the total area. It has rich pasture resources, a prerequisite for the development of
Y. Zhang et al. / International Journal of Applied Earth Observation and Geoinformation 44 (2016) 1–10
animal husbandry. Tekes had a population of 148,600 in 1998 and 167,800 in 2011 (Xinjiang Province Statistical Bureau, 1999, 2012). 2.2. Data acquisition and processing To identity landscape changes from 1998 to 2011 in the two settings, we used a variety of satellite image sources (Table 1). Phenological variation can complicate consistency in image classification between scenes (Shi et al., 2011). Different phenological stages of plant cover have different spectral signatures, and the vegetation is vigorous before harvest. We therefore classified the landscapes of the study area and selected images from 1998, 2006 and 2011 (TM) from approximately September–October, which is the best time of the year for distinguishing between land-cover types (Brinkmann et al., 2012). All of the images were terraincorrected with less than 5% cloud cover. To ascertain the spectral and texture characteristics of the different landscapes of the study area, a field survey was carried out in 2011, and training samples for different landscapes were constructed. The land-cover maps were generated using the supervised maximum likelihood classification in ERDAS. The following six landscape classes were identified: water, forest, cropland, grassland, urban, and barren land. In this study, urban included cities, industrial facilities, settlements, and single houses, and barren land included non-vegetated and sparsely vegetated areas. Because of the large geographic extent of the study area, substantial time differences in the satellite images, and especially the complicated land-use structure and steep terrain in the study sites (Zhang et al., 2009), some classes had similar spectral characteristics (Brinkmann et al., 2012) and a single type of object might have different spectral characteristics. It will be difficult to meet accuracy requirements using the supervised classification method. Therefore, to improve accuracy, we used a strategy that combines visual interpretation techniques with high-resolution satellite images from Google Earth together with ground truthing from field surveys to classify the landscape types after the process of supervised classification. To assess accuracy, 300 independent validation plots were randomly selected for each county. The observed landscape in these locations was visually classified using recent high-resolution Google Earth satellite images (Brinkmann et al., 2012). The overall accuracy of the 2011 image classification was 93.7% for Tekes and 91.3% for Baicheng, with a kappa statistic of 0.92 and 0.87, respectively. A total of eight environmental variables were included in our study, divided into three groups (topography, anthropogenic disturbance, and climate) according to their attributes (Hietel et al., 2004). The variables are described in Table 2. The statistical yearbooks of Xinjiang Province for the years 1999, 2007, and 2012 provided data on the population of the main ethnic groups of each study site. A digital elevation model (DEM) of the study sites with a spatial resolution of 30 m was used to produce the elevation, slope and aspect variables and to evaluate the conditions of topography between areas that experienced different landscape transitions. The values of elevation and slope from the DEM can be used in the models directly, while the values of aspect layer cannot indicate the solar radiation an area place can receive. As a result, the aspect measurements were classified from 1 to 8 as follows: 1 (337.6◦ –22.5◦ ), 2 (22.6◦ –67.5◦ ), 3 (292.6◦ –337.5◦ ), 4 (67.6◦ –112.5◦ ), 5 (247.6◦ –292.5◦ ), 6 (112.6◦ –157.5◦ ), 7 (202.6◦ –247.5◦ ), and 8 (157.6◦ –202.5◦ ). The greater the value is, the sunnier and drier the area will be (Zhang et al., 2012). Because distantial attenuation is a significant characteristic of the residential environment (Liu et al., 2010b), we selected distance to the nearest resident as a variable of anthropogenic disturbance. The distance variable, referring to the distance-cost relationship in the Von Thünen model (Von Thünen, 1826), was calculated using the nearest villages. The layer of the vil-
3
lages, which were revised according to the physical truthing, was derived from the Landsat images from 2006. Then, raster images for the two study sites were generated to depict the Euclidean distance to the nearest village. All of the environmental variable layers were resampled to a 30-m resolution for modeling. The selected factors were applied as stable discriminant factors without variation in time. All of the spatial analyses were performed using ArcGIS 10.0. 2.3. Detection of landscape changes Short-term changes, swaps between locations and backward transitions can be analyzed by studying each transition between one land-use type and another (Mottet et al., 2006). The importance of studying not only the net changes but also the systematic transitions between the land-use types when trying to link patterns and processes in the research of landscape dynamics has been emphasized (Pontius et al., 2004). Changes were detected by a post-classification comparison of bi-temporal maps (Shoyama and Braimoh, 2011) of the study sites to identify the interlinkages of different landscapes. The dynamic changes of these landscapes were analyzed in terms of the total area, total change and change rate (%/year) over the two periods (1998–2006, 2006–2011). To characterize the landscape changes during the study periods in detail, two cross-tabulation matrices were computed separately for each area using the surface occupied by each land cover. Each cross-tabulation matrix consisted of six columns representing the area of each land cover at time t1 converted to the different land covers at time t2 (Silva et al., 2011). Entries along the diagonal of the matrices indicate the persistence of the corresponding land covers. 2.4. Statistical analyses A multivariate method of redundancy analysis (RDA) was used to explore the relationships among the environmental variables and the landscape changes. The RDA, a constrained multivariate ordination technique that is essentially the multivariate extension of multiple regressions (Legendre and Legendre, 1998; Fisichelli et al., 2013), needs a response matrix and an explanatory variable matrix. In this study, the response matrixes were the landscapetransition matrixes of different study periods, and the explanatory variable matrix contained all of the drivers selected for the analysis. The RDA can effectively identify the relationship between one or a group of variables and another group in the sight of statistics among ˇ all the of the gradient analysis methods (ter Braak and Smilauer, 2002). In our study, we focused only on the patterns and changes of three landscapes: cropland, grassland, and forest. To explore the relationships between the patterns and environmental variables, 50 plots (30 m × 30 m) were randomly placed within each studied landscape in Tekes and Baicheng where the landscape did not change during the successive periods. The random placement of the plots was achieved with Hawth’s Analysis Tools for ArcGIS (Beyer, 2007). Regarding the landscape changes, 8–17 patches with the largest areas in each transition were selected, and 2–6 plots were randomly placed within each patch. We chose transitions that had an area greater than one-thousandth of the total area in each county. The corresponding environmental variables of each plot constitute the data matrix of the environment. Because many of the explanatory variables were strongly interrelated, the variables with a variance inflation factor (VIF) greater than 20 were excluded, and the remaining variables were used in the RDA analyses. The Monte Carlo permutation test was used to test the significance of the eigenvalues for the first canonical axis and for all of the canonical axes.
4
Y. Zhang et al. / International Journal of Applied Earth Observation and Geoinformation 44 (2016) 1–10
Table 1 Satellite images used for landscape change analysis of the study sites. Path/row
Observation date (y/m/d)
Number of bands
Resolution (m)
Satellite (sensor)
145/031 146/030 146/031 145/031 146/030 146/031 145/031 146/030 146/031
1998/10/04 1998/09/25 1998/09/25 2006/10/10 2006/10/01 2006/10/17 2011/09/06 2011/09/13 2011/09/13
7 7 7 7 7 7 7 7 7
30 30 30 30 30 30 30 30 30
Landsat 5 (TM) Landsat 5 (TM) Landsat 5 (TM) Landsat 5 (TM) Landsat 5 (TM) Landsat 5 (TM) Landsat 5 (TM) Landsat 5 (TM) Landsat 5 (TM)
Table 2 Description of the environmental variables selected for the model. Variables
Description
Unit
Source of data
Topography variables Elev Slp Asp
Elevation Slope Aspect
m % ◦
Digital elevation model (cell size 30 m) Digital elevation model (cell size 30 m) Digital elevation model (cell size 30 m)
Anthropogenic disturbance Distr Pop GDP
Distance to the nearest resident Population density Gross domestic product
m no./km2 Billion
Base cartography of the study sites Database of natural resources of China (2003) Database of natural resources of China (2003)
Climate variables Temp Pre
Annual temperature Annual precipitation
◦
Database of natural resources of China Database of natural resources of China
2.5. Variation partitioning Variation partitioning was used in our study to further analyze the interrelationships among the different driver sets. This method can be used to identify unique, shared, and total contributions to model prediction in multiple regression analyses (Kerlinger and Pedhazur, 1973), which is more suitable than analyzing the individual contributions via their partial correlation coefficients (Peres-Neto et al., 2006). Shared variation is the confounded portion that is common to more than a single variable or a set of variables. The unique variation is the portion that is explained by a variable or a set of variables after removing the effects of all of the other measured variables. The total variation of a set of variables is the sum of the unique and shared portions (Borcard et al., 1992; Fisichelli et al., 2013). In this study, variation partitioning was conducted between pairs of values for anthropogenic disturbance, topography, and climate to identify the individual contributions and the interrelationships among the driver sets.
C mm
from 8.2% to 9.2% over the same period. In 1998, grassland covered 37.8% of the total area of Tekes and increased to 39.5% in 2006 at a rate of 0.574%/year and to 40.7% in 2011 at a rate of 0.566%/year. However, in Baicheng, the proportions of grassland were 11.2%, 7.7%, and 10.6% in 1998, 2006, and 2011, respectively. The rate of decrease during period 1 (1998–2006) was −4.666%/year, and the rate of increase was 6.458%/year during period 2 (2006–2011). As observed in Fig. 2, the area of grassland in Tekes was approximately 5-fold larger than that of cropland; however, in Baicheng, the area of grassland was slightly larger than the area of cropland in 1998 and 2011, and the area of grassland was even smaller than that of cropland in 2006, as a result of degradation. The greatest loss of grassland was in Baicheng, with a massive loss of 55,259 ha during period 1; however, there was an increase of 46,541 ha in the same county during the subsequent studied period. In terms of forest, the most marked change since 1998 was in Baicheng, with a deforestation of 27,072 ha during period 2. 3.2. Landscape transitions
3. Results 3.1. Changes in the composition of the landscape The total area of the different landscapes of Tekes and Baicheng and the corresponding proportions in 1998, 2006 and 2011 are presented in Table 3 and Fig. 2. As shown in Fig. 2, Tekes was dominated by barren land and grassland that in 1998 represented 78.4% of the total area and reached 79.5% and 78.3% in 2006 and 2011, respectively. The predominant landscape of Baicheng was barren land, which covered 73.9%, 76.0%, and 74.2% of the total land in the three study years, respectively. The proportion of forest in Tekes decreased from 14.5% in 1998 to 12.7% in 2006 and increased to 13.7% in 2011, whereas the proportion of forest in Baicheng increased from 5.6% in 1998 to 6.5% in 2006 and decreased to 4.8% in 2011. The areas of cropland in the two counties showed an increasing trend in the two periods. For Tekes, the proportion of cropland was 5.9% in 1998 and increased to 6.1% in 2011, whereas the proportion of cropland in Baicheng increased
The landscape transitions that occurred during the two periods in Tekes and Baicheng are presented in Table 4, providing insight into the explicit changes between different landscapes. The results confirm that barren land remained the dominant landscape at 34.8% and 32.1% of the study area in Tekes and 71.7% and 72.0% of the study area in Baicheng during the two periods, respectively. The transition matrixes also indicate many types of transitions between different landscapes. The transition of cropland into other landscapes in Tekes was higher during period 1 than during period 2. During period 1, cropland in Tekes suffered a great loss of coverage; 0.2% was turned into grassland and barren land, respectively, 0.1% to urban area and 0.4% to water. During period 2, the main losses of cropland in Tekes were due to conversion to urban (0.2%) and barren land (0.3%). In Baicheng, 7610 ha (0.5%) were transformed from cropland to barren land during period 1, and the area of cropland that was transformed to urban and barren land was 1707 ha (0.1%) and 5720 ha (0.4%), respectively, during period 2.
Y. Zhang et al. / International Journal of Applied Earth Observation and Geoinformation 44 (2016) 1–10
5
Table 3 Total area (ha) and change rate (%/year) of different landscapes in the two periods. Landscape category
Total area 1998
2006
2011
Period (1998–2006)
1
Period (2006–2011)
2
Total change
Rate of change
Total change
Rate of change
Tekes Cropland Grassland Forest Urban Barren land Water
49,420 314,394 120,542 5,221 337,921 5,193
49,405 329,170 105,851 5,877 333,445 8,943
50,625 338,615 114,485 6,915 312,920 9,131
−15 14,776 −14,691 656 −4,476 3,750
−0.004 0.574 −1.625 1.480 −0.167 6.794
1,220 9,445 8,634 1,037 −20,525 188
0.488 0.566 1.568 3.251 −1.271 0.416
Baicheng Cropland Grassland Forest Urban Barren land Water
130,267 177,380 89,446 4,994 1,170,551 11,823
138,886 122,121 102,743 5,147 1,204,607 10,957
145,049 168,662 75,671 6,916 1,175,884 12,279
8,619 −55,259 13,297 153 34,056 -866
0.801 −4.666 1.732 0.378 0.358 −0.951
6,163 46,541 −27,072 1,769 −28,723 1,322
0.868 6.458 −6.117 5.907 −0.483 2.279
Fig. 2. Proportions of different landscapes in 1998, 2006 and 2011 in the study area. Table 4 Area of landscape (ha) and percentage (%) converted from one type to the next for the period 1998–2006 and 2006–2011. %
Grassland
%
Tekes 1998–2006
Cropland Grassland Forest Urban Barren land Water
Cropland 41,783 1,832 205 569 4,877 139
5.0 0.2 0.0 0.1 0.6 0.0
1,526 271,496 15,630 18 40,438 61
0.2 32.6 1.9 0.0 4.9 0.0
Forest 372 3,916 99,012 0 2,300 252
% 0.0 0.5 11.9 0.0 0.3 0.0
Urban 1,151 78 72 4,255 209 112
% 0.1 0.0 0.0 0.5 0.0 0.0
Barren land 1,273 36,973 4,842 71 289,710 576
% 0.2 4.4 0.6 0.0 34.8 0.1
Water 3,315 99 781 308 388 4,052
% 0.4 0.0 0.1 0.0 0.0 0.5
Tekes 2006–2011
Cropland Grassland Forest Urban Barren land Water
43,781 3,512 245 712 2,092 283
5.3 0.4 0.0 0.1 0.3 0.0
658 271,646 8,152 161 57,994 5
0.1 32.6 1.0 0.0 7.0 0.0
764 12,215 95,971 79 5,008 447
0.1 1.5 11.5 0.0 0.6 0.1
1,601 222 5 4,704 360 23
0.2 0.0 0.0 0.6 0.0 0.0
2,346 41,442 1,256 191 267,310 375
0.3 5.0 0.2 0.0 32.1 0.0
255 132 223 30 681 7,810
0.0 0.0 0.0 0.0 0.1 0.9
Baicheng 1998–2006
Cropland Grassland Forest Urban Barren land Water
121,725 3,231 0 211 12,518 1,202
7.7 0.2 0.0 0.0 0.8 0.1
25 109,426 6,777 0 5,790 103
0.0 6.9 0.4 0.0 0.4 0.0
13 9,131 77,312 0 16,287 0
0.0 0.6 4.9 0.0 1.0 0.0
273 0 0 4,780 94 0
0.0 0.0 0.0 0.3 0.0 0.0
7,610 55,195 5,358 1 1,135,452 990
0.5 3.5 0.3 0.0 71.7 0.1
622 396 0 2 409 9,528
0.0 0.0 0.0 0.0 0.0 0.6
Baicheng 2006–2011
Cropland Grassland Forest Urban Barren land Water
130,140 219 0 98 13,363 1,230
8.2 0.0 0.0 0.0 0.8 0.1
154 108,377 15,317 0 44,788 27
0.0 6.8 1.0 0.0 2.8 0.0
0 5,093 67,500 0 3,079 0
0.0 0.3 4.3 0.0 0.2 0.0
1,707 0 0 5,049 154 5
0.1 0.0 0.0 0.3 0.0 0.0
5,720 8,362 19,927 0 1,141,087 787
0.4 0.5 1.3 0.0 72.0 0.0
1,165 70 0 0 2,136 8,907
0.1 0.0 0.0 0.0 0.1 0.6
6
Y. Zhang et al. / International Journal of Applied Earth Observation and Geoinformation 44 (2016) 1–10
Fig. 3. Biplot of RDA ordination of the distribution of landscapes and environmental variables in Tekes and Baicheng: T means Tekes, B means Baicheng, and 1–3 represent cropland, grassland and forest, respectively. Abbreviations for the environmental variables are listed in Table 2.
During period 1, the area of grassland that was transformed to forest was 3916 ha (0.5%) and 9131 ha (0.6%), and during period 2 it was 12,215 ha (1.5%) and 5093 ha (0.3%), in Tekes and Baicheng, respectively. However, as shown in Table 4, grassland experienced the largest area loss due to the transition to barren land (4.4% and 5% in Tekes, 3.5% and 0.5% in Baicheng during the two periods, respectively). Forest suffered a severe degradation in Tekes: 15,630 ha (1.9%) transitioned to grassland and 4842 ha (0.6%) to barren land during period 1, and 8152 ha (1%) and 1256 ha (0.2%) transitioned to grassland and barren land, respectively, during period 2. In Baicheng, the main loss of forest involved the transition to grassland (6777 ha) and barren land (5358 ha) during period 1 and to grassland (15,317 ha) and barren land (19,927 ha) during period 2. During period 1, the main contributor to cropland and grassland gains was barren land (4877 ha and 40,483 ha, respectively) in Tekes; in Baicheng, the area of barren land that transitioned to cropland, grassland, and forest was 12,518 ha (0.8%), 5790 ha (0.4%), and 16,287 ha (1.0%), respectively. However, during period 2, the largest transformation to grassland took place in Tekes, with 57,994 ha (7.0%) transitioning from barren land to grassland. 3.3. Drivers of the distribution and change of landscapes The biplot of the RDA ordination of the distribution of landscapes and environmental variables in Tekes and Baicheng is illustrated in Fig. 3, and the correlation coefficients between the environmental factors and the first four RDA axes are shown in Table 5. Although eight axes were used in the imputation, the last four axes have very little influence on neighbor selection because the distances are weighted by the eigenvalues (Ohmann et al., 2011). The eigenvalues for the first four axes were 0.173, 0.151, 0.051, and 0.013. The landscape-environment correlations of the first four RDA axes were 0.929, 0.870, 0.506, and 0.250, and the cumulative percentage variances of the landscape-environment correlation for the first four axes were 98.5%, indicating that RDA performed well in describing the relationships between the landscapes and the environmental gradients (Zhang et al., 2008). The Monte Carlo permutation test indicated that the eigenvalues for the first canonical axis and for all of the canonical axes were significant (p < 0.01). As shown in Table 5
and Fig. 3, the first axis had significant correlations with elevation (Elev), slope (Slp), distance to the nearest resident (Distr), temperature (Temp), population density (Pop), gross domestic product (GDP), and precipitation (Pre). Axis 2 was significantly related to all eight variables. The third axis was significantly correlated with aspect (Asp) and was also related to Pop and Temp, and the fourth axis was related to Pop and Pre. The biplots of the RDA ordination of the transition of landscapes and the environmental variables in Tekes and Baicheng during the two periods are illustrated in Fig. 4, and the correlation coefficients between the environmental factors and the first four RDA axes are shown in Table 6. During the period 1998–2006, the eigenvalues for the first four axes were 0.044, 0.032, 0.014, and 0.012. The landscape transition-environment correlations of the first four RDA axes were 0.937, 0.785, 0.522, and 0.493, and the cumulative percentage variances of the landscape transition-environment correlation for the first four axes were 87.0%, indicating that the RDA performed well in describing the relationships between the landscape transitions and the environmental gradients during period 1. The Monte Carlo permutation test showed that the eigenvalues for the first canonical axis and for all of the canonical axes were significant (p–< 0.01). Axis 1 had significant correlations with Elev, Slp, Distr, Pre, Temp, GDP, and Pop. Axis 2 was significantly correlated with Pre, Temp, and Slp and also related to Distr. Axis 3 was significantly related to Temp, Pre, Asp, and Slp and was also related to GDP. Axis 4 had significant correlations with Pre and Slp and was also related to Distr. During the period 2006–2011, the eigenvalues for the first four axes were 0.047, 0.035, 0.015, and 0.014. The landscape transitionenvironment correlations of the first four RDA axes were 0.900, 0.791, 0.537, and 0.505, and the cumulative percentage variances of the landscape transition-environment correlation for the first four axes were 83.60%, indicating that the RDA performed well in describing the relationships between the landscape transitions and the environmental gradients during period 2. The Monte Carlo permutation test indicated that the eigenvalues for the first canonical axis and for all of the canonical axes were significant in this ordination (p < 0.01). The first axis had significant correlations with all the variables, and the second axis had significant correlations with the same variables, except for Elev and Pop. The third axis was significantly correlated with Distr, Slp, Temp, Pre, and Asp and also related to GDP. The fourth axis was significantly related to Distr, Slp, and GDP and also related to Pop and Elev. 3.4. Variation partitioning The results of the variation partitioning for the three RDA processes are illustrated in Fig. 5. In the RDA of the distribution of the landscape and the environmental variables, the shared variation of topography with the other sets was 2% and 35%, respectively (Fig. 5(a)). Anthropogenic disturbance shared 10% and 61% of its explanatory power with climate and topography. In the transition of the landscapes and the environmental variables during period 1, there was no common portion between anthropogenic disturbance and climate, whereas topography shared 6% and 45% of its explained variation with climate and anthropogenic disturbance. During period 2, anthropogenic disturbance shared levels of explained variation with topography and climate similar to those in the first analysis process (the RDA of the distribution of landscapes and the environmental variables). The shared explanatory power between the climate and the other two driver sets was higher than that of the first two processes (21% for topography and 15% for anthropogenic disturbance). In the three RDA processes, topography, which explained the greatest variation, shared a large portion with anthropogenic disturbance but shared little explained variation with climate. The extent of the shared explanatory powers
Y. Zhang et al. / International Journal of Applied Earth Observation and Geoinformation 44 (2016) 1–10
7
Table 5 Correlation coefficients between environmental factors and the first four RDA axes in the distribution of the landscapes in the study area. Variables
Axis 1
Axis 2
Axis 3
Axis 4
Elev Slp Asp Distr Pop GDP Pre Temp
0.757** 0.640** −0.108 0.410** −0.271** −0.252** 0.242** 0.304**
0.456** 0.468** −0.329** 0.597** −0.315** −0.310** −0.466** −0.774**
0.104 −0.047 0.414** −0.012 −0.113* −0.077 0.039 −0.117*
0.005 −0.050 −0.054 −0.010 −0.114* 0.018 0.143* −0.002
Abbreviations for the variables are listed in Table 2. * p < 0.05. ** p < 0.01.
Fig. 4. Biplots of RDA ordination of the transition of landscapes and environmental variables in Tekes and Baicheng in period 1 (a) and period 2 (b): T means Tekes, B means Baicheng, and 1–5 represent cropland, grassland, forest, urban and barren land, respectively. Abbreviations for the environmental variables are listed in Table 2.
Table 6 The correlation coefficients between environmental factors and the first four RDA axes in the landscape transitions of the study area. Variables
Elev Slp Asp Distr Pop GDP Pre Temp
Period 1 (1998–2006)
Period 2 (2006–2011)
Axis 1
Axis 2
Axis 3
Axis 4
Axis 1
Axis 2
Axis 3
Axis 4
0.919** 0.802** -0.063 0.812** −0.211** −0.294** −0.327** 0.307**
0.037 0.160** −0.024 0.084* 0.038 0.008 0.514** −0.479**
0.049 −0.149** −0.197** −0.038 0.051 0.080* −0.240** −0.309**
−0.074 0.153** −0.049 0.096* 0.038 0.062 −0.164** 0.027
0.858** 0.732** −0.129** 0.706** −0.123** −0.231** −0.322** 0.526**
0.071 0.280** −0.147** 0.241** −0.027 −0.137** 0.620** −0.284**
−0.064 0.146** 0.122** 0.218** 0.045 0.093* −0.128** 0.129**
0.096* −0.147** 0.053 0.174** −0.105* −0.113** 0.057 0.070
Abbreviations for the variables are listed in Table 2. * p < 0.05. ** p < 0.01.
indicates complex relationships among all of the pairs of driving forces. 4. Discussion 4.1. The role of anthropogenic disturbance The Distr variable, as well as GDP and Pop, had different effects on the patterns and transitions in Tekes and Baicheng. These differences were attributable to the cultural backgrounds of residents of the two counties. Areas that share common ethnic and cultural features would have similar anthropogenic activities (Brinkmann et al., 2012). The populations of the main ethnic groups of Tekes and Baicheng are illustrated in Fig. 6. Although each ethnic group had a
fluctuating population in these two counties, the dominant groups did not change. Kazak, Han, Hui, Uygur, and Kirgiz were the main ethnic groups in Tekes. The population of Kazak increased from 59,975 (40.4%) in 1998 to 71,683 (42.7%) in 2011. The Uygur population showed a similar trend, with a population of 14,715 (9.9%) in 1998 and 19,044 (11.3%) in 2011. The Han population decreased from 36,531 (24.6%) in 1998 to 36,181 (21.6%) in 2011, and the population of Hui and Kirgiz decreased during 1998–2006 and increased during 2006–2011. The proportions of Hui were 13.0%, 12.6%, and 13.8%, and the proportions of Kirgiz were 6.0%, 5.7%, and 6.1%, in 1998, 2006, and 2011, respectively. The main ethnic groups in Baicheng were Uygur and Han. Uygur constituted 87.0% of the population of Baicheng in 1998, and the proportions in 2006 and 2011 were 85.8% and 87.1%. From 1998 to 2006, the proportion
8
Y. Zhang et al. / International Journal of Applied Earth Observation and Geoinformation 44 (2016) 1–10
Fig. 5. The results of variation partitioning for pairs of driver sets in the RDA of (a): the distribution of landscapes and environmental variables, (b): the transition of landscapes and environmental variables in Tekes and Baicheng in period 1 and (c): the transition of landscapes and environmental variables in Tekes and Baicheng in period 2. The overlapping middle sections represent the shared variation explained by both driver sets, and the left and right sections are the unique portions explained by the corresponding driver set.
Fig. 6. Populations of Tekes and Baicheng in 1998, 2006 and 2011.
of Han increased from 12.1% to 13.4%, but in 2011 it decreased to 11.9%. Different ethnic groups have their own production and living styles, and their land-use needs differ greatly. The places where agriculture-focused populations gathered have stronger farming activities than animal husbandry populations do. Much more cropland is needed in such areas, which leads people to reclaim cropland as the population increases; as a result, other types of landscape will be transformed to cropland. By contrast, grassland nearby is used for grazing, and when grassland degenerates, herdsmen find new grazing sites far from their houses. Landscapes that are near residents are more likely to be affected by people because of their reconstruction of the land. Bceause the main foods of ethnic Han, Uygur, and Hui people are grain crops, their activities greatly influence croplands, including paddy fields and dry land. Ethnic Kazak and Kirgiz people are engaged mainly in animal husbandry, and most of them move their pastures as the seasons change (Li, 2011).
The areas and patterns of grassland would be influenced more in the places where these two ethnic groups gather than in other places. According to the description and analysis above, the total proportions of ethnic Uygur and Han in Baicheng were 99.1%, 99.2%, and 99.0% in 1998, 2006, and 2011, respectively, and most of these populations are engaged mainly in agriculture. However, the total proportions of Uygur, Han, and Hui were 47.5% in 1998, 46.9% in 2006, and 46.7% in 2011, which were much lower than those in Baicheng. Few people rely on animal husbandry for living in Baicheng, whereas in Tekes the total proportions of Kazak and Kirgiz engaged mainly in animal husbandry were 46.4%, 48.2%, and 48.8% in 1998, 2006 and 2011, respectively. As a result, the patterns and changes in landscapes at the same distance to the nearest residents in Tekes and Baicheng were different. Accelerated repopulation increased the demand for land resources during the study periods. The use of wood for economic purposes and the subsistence needs of the local people resulted in a reduction in forest areas in these two counties. The deforested region was subsequently replaced by grassland or barren land. Livestock production in the study area was based primarily on cattle, buffaloes, goats, and sheep. Some areas were subjected to overgrazing; as a result, large amounts of grassland were degraded to barren land. When the remaining land resources could no longer meet the farmers’ needs, they sought new resources. With the increased pressure of the population and the loss of grassland, barren land nearby was turned into rangeland for livestock grazing. Intensive agricultural activities are important causes of the decline in the quality of farmland, which is therefore abandoned after several decades of cultivation. Because most of the people in Baicheng were engaged in agriculture, larger areas of barren land and grassland were converted to cropland than in Tekes. The variation of cultivation and grazing intensity in Tekes and Baicheng mirrored the differentiating disturbance by ethnic groups.
4.2. Landscape transitions and their drivers The croplands in the two counties were located in areas that had higher Pop and GDP. Grassland and forest were found at higher ele-
Y. Zhang et al. / International Journal of Applied Earth Observation and Geoinformation 44 (2016) 1–10
vations and steeper areas, and the distances to the nearest resident were much longer than those of cropland. As observed in Fig. 3, the grassland in Baicheng was more strongly correlated with Distr than that in Tekes, indicating that some of the grassland in Tekes was near residents and that most of the grassland in Baicheng was far from residents. Almost all of the people in Baicheng are engaged in agriculture and require more cropland for living. However, in Tekes, 48% of the total population is occupied in animal husbandry; therefore, a large amount of grassland is near their houses, which is convenient for pasture. During period 1, a large amount of cropland in Baicheng was transformed to barren land and much of the grassland was transformed into cropland, indicating that when the cropland started to degenerate, some grassland was gradually converted to cropland to meet the farmers’ needs. However, in Tekes, much of the cropland was transformed to grassland. During this period, the degradation of cropland in Baicheng was severe owing to the high intensity of agricultural production, and grassland was used to make up for the lack of cropland. Both of the transitions in Baicheng had negative correlations with slope, elevation, and distance to the nearest resident. In period 2, the transition of grassland to barren land in Tekes had a negative correlation with Distr, whereas in Baicheng, the same transition had a positive correlation with Distr. This phenomenon was related to anthropogenic activities and cultural backgrounds. The people in Tekes had more demand for grassland; therefore, much of the grassland degradation occurred near their living areas. The increase in barren land, which indicates a decrease in grassland, might also be a result of over-exploitation by grazing animals (Brinkmann et al., 2012). However, in Baicheng, the grassland was far from the residents; as a result, grassland degradation was influenced mainly by natural variables, such as Temp and Elev (Fig. 4(b)), and was weakly related to anthropogenic activities. 4.3. Variation partitioning The results of variation partitioning show the interrelationships among pairs of driver sets in the identification of the drivers of the distribution and change of landscapes. Our emphasis was on anthropogenic disturbance, and the relative strengths of the interrelated variables must be taken into account when attempting to project the influence of the target driver set (Fisichelli et al., 2013). The central role of the topography has strong influences on the distribution and the transition of landscapes. GDP and Pop were negatively related to Distr, which was due mainly to the effects of human activities, such as grazing and cultivation. The land far away from residents was less likely to be reclaimed or utilized. Only when the land nearby no longer satisfied the needs of the people would they consider further places. Anthropogenic disturbance is closely related to topography, and the strength of human activities may vary with topography because higher elevation and steeper slope were not suitable for people and would limit their production and lives. Climate was related to geographical location, topography, and certain other complicated factors; as a result, Temp and Pre had little shared variation with the anthropogenic disturbance set, but they all play important roles in the distribution and change of landscapes. 5. Conclusion This study detected the patterns and transitions of landscapes, as well as their drivers, in Tekes and Baicheng, where the configurations of the ethnic groups varied greatly. There were considerable variations between the two counties in the characteristics of the landscape patterns and their transitions. The amount of arable land
9
in Baicheng was much greater than that in Tekes. By contrast, Tekes had larger areas of grassland than Baicheng, most of which was located near the residents. The significant drivers that affected the configuration and transitions of landscape were Pre, Temp, Slp, Distr, and Elev. There are different ethnic groups in the two counties, and people of different ethnic groups have their own means of production, resulting in the different proportions of landscapes and landscape transitions in the study area. The results of the variation partitioning indicate complex interrelationships among all of the pairs of driver sets. All the variable sets had significant explanatory roles, most of which had both unique and shared variations with the others. Anthropogenic activities could strongly influence landscape patterns and likely accelerated the transitions of the landscapes. The dynamics of the landscapes became increasingly complex with increasing populations and economic development. Policy, technology, and socio-economic variables also significantly impact landscape transitions. In future studies, more attention should be paid to the impact and interrelationships of policy, technology, and socioeconomic variables. The results of this study can assist policy makers and planners in implementing effective protection strategies and sustainable landscape management.
Acknowledgement Financial support for this research was provided by the Project of the Special Research Foundation of Public Welfare (no. 201203005), the Special Program for National Science & Technology Basic Work (no. 2014FY110200A16), and the Fundamental Research Funds for the Central Universities (no. 2662015JC007).
Appendix A. Supplementary data Supplementary data associated with this article can be found, in the online version, at http://dx.doi.org/10.1016/j.jag.2015.06.013
References Beyer, H.L., 2007. Hawth’s analysis tools for ArcGIS. Available at http://www. spatialecology.com/htools Bicík, I., Jelecek, L., Stepanek, V., 2001. Land-use changes and their social driving forces in Czechia in the 19th and 20th centuries. Land Use Policy 18, 65–73. Borcard, D., Legendre, P., Drapeau, P., 1992. Partialling out the spatial component of ecological variation. Ecology 73, 1045–1055. Brinkmann, K., Schumacher, J., Dittrich, A., Kadaore, I., Buerkert, A., 2012. Analysis of landscape transformation processes in and around four West African cities over the last 50 years. Landsc. Urban Plann. 105, 94–105. Bürgi, M., 1999. A case study of forest change in the Swiss low-lands. Landsc. Ecol. 14, 567–575. Etter, A., McAlpine, C., Pullar, D., Possingham, H., 2006. Modelling the conversion of Columbian lowland ecosystems since 1940: drivers, patterns and rates. J. Environ. Manage. 79, 74–87. Fisichelli, N.A., Frelich, L.E., Reich, P.B., 2013. Climate and interrelated tree regeneration drivers in mixed temperate-boreal forests. Landsc. Ecol. 28, 149–159. Hietel, E., Waldhardt, R., Otte, A., 2004. Analysing land-cover changes in relation to environmental variables in Hesse, Germany. Landsc. Ecol. 19, 473–489. Imhoff, M.L., Bounoua, L., Defries, R., Lawrence, W.T., Stutzer, D., Tucker, C.J., Ricketts, T., 2004. The consequences of urban land transformation on net primary productivity in the United States. Remote Sens. Environ. 89, 434–443. Johnson, B.G., Zuleta, G.A., 2013. Land-use land-cover change and ecosystem loss in the Espinal ecoregion, Argentina. Agric. Ecosyst. Environ. 181, 31–40. Kerlinger, F.N., Pedhazur, E.J., 1973. Multiple Regression in Behavioral Research. Holt, Rinehart and Winston, New York, USA. Krausmann, F., Haberl, H., Schulz, N.B., Erb, K.H., Darge, E., Gaube, V., 2003. Land-use change and socio-economic metabolism in Austria, Part I: driving forces of land-use change 1950–1995. Land Use Policy 20, 1–20. Lambin, E.F., Geist, H.J., Lepers, E., 2003. Dynamics of land-use and land-cover change in tropical regions. Ann. Rev. Environ. Res. 28, 205–241. Legendre, P., Legendre, L., 1998. Numerical Ecology, 2nd English ed. Elsevier Science, Amsterdam. Li, C.H., 2011. Xinjiang Scenery Blog. Xinjiang Province, China.
10
Y. Zhang et al. / International Journal of Applied Earth Observation and Geoinformation 44 (2016) 1–10
Liu, J., Zhang, Z., Xu, X., Kuang, W., Zhou, W., Zhang, S., et al., 2010a. Spatial patterns and driving forces of land use change in China during the early 21st century. J. Geogr. Sci. 20 (4), 483–494. Liu, Y.L., Zheng, B., Turkstra, J., Huang, L., 2010b. A hedonic model comparison for residential land value analysis. Int. J. Appl. Earth Obs. Geoinf. 12S, S181–S193. Long, H.L., Li, X.B., 2005. Rural housing land transition in transect of the Yangtze River. Acta Geogr. Sinica 60, 179–188 (in Chinese). Maeda, E.E., De Almeida, C.M., De Carvalho Ximenes, A., Formaggio, A.R., Shimabukuro, Y.E., Pellikka, P., 2011. Dynamic modeling of forest conversion: simulation of past and future scenarios of rural activities expansion in the fringes of the Xingu National Park, Brazilian Amazon. Int. J. Appl. Earth Obs. Geoinf. 13, 435–446. Mottet, A., Ladet, S., Coqué, N., Gibon, A., 2006. Agricultural land-use change and its drivers in mountain landscapes: a case study in the Pyrenees. Agric. Ecosyst. Environ. 114, 296–310. Ohmann, J.L., Gregory, M.J., Henderson, E.B., Roberts, H.M., 2011. Mapping gradients of community composition with nearest-neighbour imputation: extending plot data for landscape analysis. J. Veg. Sci. 22, 660–676. Paudel, S., Yuan, F., 2012. Assessing landscape changes and dynamics using patch analysis and GIS modeling. Int. J. Appl. Earth Obs. Geoinf. 16, 66–76. Peres-Neto, P.R., Legendre, P., Dray, S., Borcard, D., 2006. Variation partitioning of species data matrices: estimation and comparison of fractions. Ecology 87, 2614–2625. Pontius Jr, R.G., Shusas, E., McEachern, M., 2004. Detecting important cate-gorical land changes while accounting for persistence. Agric. Ecosyst. Environ. 101 (2/3), 251–268. Random House, 1987. The Random House dictionary of the English language, 2nd ed. Unabridged. Seto, K.C., Fragkias, M., 2005. Quantifying spatiotemporal patterns of urban land-use change in four cities of China with time series landscape metrics. Landsc. Ecol. 20, 871–888. Shi, L., 1995. The theory of culture-ethnic culture and market culture. Ethno.-National. Stud. 1, 20–26 (in Chinese). Shi, Z.H., Li, L., Yin, W., Ai, L., Fang, N.F., Song, Y.T., 2011. Use of multi-temporal Landsat images for analyzing forest transition in relation to socioeconomic factors and the environment. Int. J. Appl. Earth Obs. Geoinf 13, 468–476. Shoyama, K., Braimoh, A.K., 2011. Analyzing about sixty years of land-cover change and associated landscape fragmentation in Shiretoko Peninsula, Northern Japan. Landsc. Urban Plann. 101, 22–29.
Silva, J.S., Vaz, P., Moreira, F., Catry, F., Rego, F.C., 2011. Wildfires as a major driver of landscape dynamics in three fire-prone areas of Portugal. Landsc. Urban Plann. 101, 349–358. ˇ ter Braak, C.J.F., Smilauer, P., 2002. Canoco Reference Manual and User’s Guide to CanoDraw for Windows. Software for Canonical Community Ordination (Version 4.5). Centre for Biometry Wageningen/Microcomputer Power, Wageningen, NL/Ithaca NY, USA, pp. 352 pp. Tzanopoulos, J., Vogiatzakis, I.N., 2011. Processes and patterns of landscape change on a small Aegean island: the case of Sifnos, Greece. Landsc. Urban Plann. 99, 58–64. Von Thünen, J.H., 1826. Der isolierte Staat in Beziehung auf Landwirtschaft und Nationalökonomie. Hamburg, Germany. Walker, R., Drzyzga, S.A., Li, Y., Qi, J., Caldas, M., Arima, E., Vergara, D., 2004. A behavioral model of landscape change in the Amazon Basin: the colonist case. Ecol. Appl. 14, 299–312. Wyman, M.S., Stein, T.V., 2010. Modeling social and land-use/land-cover change data to assess drivers of smallholder deforestation in Belize. Appl. Geogr. 30, 329–342. Xinjiang Province Statistical Bureau, 1999. Statistical yearbook of Xinjiang Province 1999. Xinjiang Province, China. Xinjiang Province Statistical Bureau, 2012. Statistical yearbook of Xinjiang Province 2012. Xinjiang Province, China. Xu, W., 2004. The changing dynamics of land-use change in rural China: a case study of Yuhang, Zhejiang Province. Environ. Plann. A 36, 1595–1615. Zhang, J.T., Dong, Y.R., Xi, Y.X., 2008. A comparison of SOFM ordination with DCA and PCA in gradient analysis of plant communities in the midst of Taihang Mountains, China. Ecol. Inf. 3, 367–374. Zhang, J.T., Xiang, C.L., Li, M., 2012. Integrative ecological sensitivity (IES) applied to assessment of eco-tourism impact on forest vegetation landscape: a case from the Baihua Mountain Reserve of Beijing, China. Ecol. Indic. 18, 365–370. Zhang, J.X., Liu, Z.J., Sun, X.X., 2009. Changing landscape in the three Gorges Reservoir Area of Yangtze River from 1977 to 2005: land use/land cover, vegetation cover changes estimated using multi-source satellite data. Int. J. Appl. Earth Obs. Geoinf. 11, 403–412.
Contents lists available at ScienceDirect
International Journal of Applied Earth Observation and Geoinformation journal homepage: www.elsevier.com/locate/jag
Landscape pattern and transition under natural and anthropogenic disturbance in an arid region of northwestern China Yu Zhang a , Tianwei Wang a,∗ , Chongfa Cai a , Chongguang Li b , Yaojun Liu a,c , Yuze Bao b , Wuhong Guan a a
College of Resources and Environment, Huazhong Agricultural University, Wuhan 430070, China College of Economics & Management, Huazhong Agriculture University, Wuhan 430070, China c Jiangxi Institute of Soil and Water Conservation, Nanchang 330029, China b
a r t i c l e
i n f o
Article history: Received 19 July 2014 Accepted 15 June 2015 Keywords: Landscape transition Driving force Anthropogenic disturbance Redundancy analysis (RDA) Variation partitioning
a b s t r a c t There is a pressing need to determine the relationships between driving variables and landscape transformations. Human activities shape landscapes and turn them into complex assemblages of highly diverse structures. Other factors, including climate and topography, also play significant roles in landscape transitions, and identifying the interactions among the variables is critical to environmental management. This study analyzed the configurations and spatial-temporal processes of landscape changes from 1998 to 2011 under different anthropogenic disturbances, identified the main variables that determine the landscape patterns and transitions, and quantified the relationships between pairs of driver sets. Landsat images of Baicheng and Tekes from 1998, 2006 and 2011 were used to classify landscapes by supervised classification. Redundancy analysis (RDA) and variation partitioning were performed to identify the main driving forces and to quantify the unique, shared, and total explained variation of the sets of variables. The results indicate that the proportions of otherwise identical landscapes in Baicheng and Tekes were very different. The area of the grassland in Tekes was much larger than that of the cropland; however, the differences between the grassland and cropland in Baicheng were not as pronounced. Much of the grassland in Tekes was located in an area that was near residents, whereas most of the grassland in Baicheng was far from residents. The slope, elevation, annual precipitation, annual temperature, and distance to the nearest resident were strong driving forces influencing the patterns and transitions of the landscapes. The results of the variation partitioning indicated complex interrelationships among all of the pairs of driver sets. All of the variable sets had significant explanatory roles, most of which had both unique and shared variations with the others. The results of this study can assist policy makers and planners in implementing sustainable landscape management and effective protection strategies. © 2015 Elsevier B.V. All rights reserved.
1. Introduction Land use in China has changed dramatically since the adoption of economic reforms and the opening-up policy in 1978 (Xu, 2004). Unprecedented urban expansion represents an important type of land transition (Seto and Fragkias, 2005). A large proportion of the land has been degraded by agricultural enterprises and overgrazing. Fragmentation of cropland, due to the construction of the
∗ Corresponding author. E-mail addresses: [email protected] (Y. Zhang), [email protected] (T. Wang), [email protected] (C. Cai), [email protected] (C. Li), [email protected] (Y. Liu), [email protected] (Y. Bao), gwh [email protected] (W. Guan). http://dx.doi.org/10.1016/j.jag.2015.06.013 0303-2434/© 2015 Elsevier B.V. All rights reserved.
countryside (Long and Li, 2005), has caused many concerns regarding China’s food security. The principal consequences of landscape changes are resource depletion and environmental and ecological problems (Imhoff et al., 2004), such as the loss, fragmentation and degradation of the available habitat for most species, which are among the major threats to biodiversity worldwide (Johnson and Zuleta, 2013; Maeda et al., 2011). These problems have had complicated influences on the modernization of China (Liu et al., 2010a). Therefore, efforts to understand the processes of landscape dynamics and driving forces are critical. Human-induced changes to natural landscapes constitute the main driving forces of land-cover change at the local, regional and global scales (Walker et al., 2004; Etter et al., 2006; Wyman and Stein, 2010). Human disturbances, such as pollution, alteration, fire, grazing, cutting, and cultivation, in particular, are complex and
2
Y. Zhang et al. / International Journal of Applied Earth Observation and Geoinformation 44 (2016) 1–10
Fig. 1. Location of the study area.
reflect temporal and spatial shifts in systems of land use (Paudel and Yuan, 2012; Tzanopoulos and Vogiatzakis, 2011). The Random House dictionary (1987) defines culture as the sum total of ways of living developed by a group of human beings and transmitted from one generation to another. Ethnic culture is a complicated entity comprising knowledge, belief, art, morality, regulations, conventions and all types of abilities and customs that the members have acquired (Shi, 1995). In a context where one ethnic group currently represents a small minority of the residents and the residents of another ethnic group make new demands on the environmental, social and cultural functions of landscapes, landscape transitions are often at the root of the tensions between different ethnic groups. Some researchers have attempted to depict the nature, direction, frequency, and rate of change (Krausmann et al., 2003; Lambin et al., 2003); others have focused on describing the driving forces from a historical point of view (Bicík et al., 2001; Bürgi, 1999). However, studies that use systematic methods to investigate how anthropogenic disturbances influence the pattern and change of landscapes with populations of different cultural backgrounds are rare. There are 56 ethnic groups in China, and 47 in Xinjiang Province, including 12 permanent minorities. Baicheng and Tekes are located in Xinjiang Province. Each ethnic group has its own cultural background, making these two counties typical and ideal places for studying human-caused disturbances. The drivers of landscape changes, which are always interrelated and covarying, can be grouped into logical driver sets. Each driver set can be identified as having both unique influences that cannot be attributed to other factors and shared explanatory power due to their interrelated roles (Fisichelli et al., 2013). Attention can be paid to single variables when other covarying factors are partially driving trends (Legendre and Legendre, 1998). The confounded relationships among the factors of landscape changes need to be quantified in order to better understand the mechanism of the variables. In the present study, multi-temporal Landsat images were used to explore landscape changes through analysis of patterns, rates of transition and drivers and quantification of the relationships between pairs of driver sets. The specific objectives were as follows: (a) to characterize the configuration and the main spatial-temporal processes of landscape changes from 1998 to 2011 under different
anthropogenic disturbances, (b) to identify the main variables that determine landscape patterns and transitions, and (c) to examine and quantify the explanatory relationships between anthropogenic disturbances and other driver sets.
2. Materials and methods 2.1. Characteristics of the study sites The study sites were Baicheng and Tekes, two adjacent counties in Xinjiang Province in northwestern China (Fig. 1) and have similar geographical conditions. Baicheng is a county in the northwestern Akesu Prefecture, which is located in the southern foothills in the middle of the Tianshan Mountains. Baicheng lies between latitude 41◦ 24 and 42◦ 51 N and stretches between longitude 80◦ 37 and 83◦ 03 E, with an area of 15,917 km2 and elevations ranging from 1041 to 6289 m. Baicheng has a temperate continental monsoon climate with cold winters and cool summers. According to weather records, the average annual temperature is 7.6 ◦ C and the daily minimum and maximum temperatures are −28 ◦ C and 38.3 ◦ C, respectively. The average annual rainfall is 171.13 mm. Baicheng is surrounded by mountains, forming a banded basin. The geographic conditions are complex, with natural steep slopes. The total population of Baicheng County was 189,000 in 1998 and 230,900 in 2011 (Xinjiang Province Statistical Bureau, 1999, 2012). Tekes is a county in the southwestern Ili Kazak Autonomous Prefecture, which is situated in the northern foothills of the fold belt of Tianshan Mountain. Tekes lies between latitude 42◦ 22 and 43◦ 25 N and stretches between longitude 81◦ 19 and 82◦ 37 E, with an area of 8080 km2 and elevations ranging from 924 to 4853 m. Tekes belongs to the inversion belt control area and has a typical temperate continental climate with an average annual rainfall of 383 mm. The average annual temperature is 5.3 ◦ C, and the daily minimum and maximum temperatures are −32 ◦ C and 33.5 ◦ C, respectively. Three major rivers flow across the county, and three mountains lie from west to east, constituting approximately 94% of the total area. It has rich pasture resources, a prerequisite for the development of
Y. Zhang et al. / International Journal of Applied Earth Observation and Geoinformation 44 (2016) 1–10
animal husbandry. Tekes had a population of 148,600 in 1998 and 167,800 in 2011 (Xinjiang Province Statistical Bureau, 1999, 2012). 2.2. Data acquisition and processing To identity landscape changes from 1998 to 2011 in the two settings, we used a variety of satellite image sources (Table 1). Phenological variation can complicate consistency in image classification between scenes (Shi et al., 2011). Different phenological stages of plant cover have different spectral signatures, and the vegetation is vigorous before harvest. We therefore classified the landscapes of the study area and selected images from 1998, 2006 and 2011 (TM) from approximately September–October, which is the best time of the year for distinguishing between land-cover types (Brinkmann et al., 2012). All of the images were terraincorrected with less than 5% cloud cover. To ascertain the spectral and texture characteristics of the different landscapes of the study area, a field survey was carried out in 2011, and training samples for different landscapes were constructed. The land-cover maps were generated using the supervised maximum likelihood classification in ERDAS. The following six landscape classes were identified: water, forest, cropland, grassland, urban, and barren land. In this study, urban included cities, industrial facilities, settlements, and single houses, and barren land included non-vegetated and sparsely vegetated areas. Because of the large geographic extent of the study area, substantial time differences in the satellite images, and especially the complicated land-use structure and steep terrain in the study sites (Zhang et al., 2009), some classes had similar spectral characteristics (Brinkmann et al., 2012) and a single type of object might have different spectral characteristics. It will be difficult to meet accuracy requirements using the supervised classification method. Therefore, to improve accuracy, we used a strategy that combines visual interpretation techniques with high-resolution satellite images from Google Earth together with ground truthing from field surveys to classify the landscape types after the process of supervised classification. To assess accuracy, 300 independent validation plots were randomly selected for each county. The observed landscape in these locations was visually classified using recent high-resolution Google Earth satellite images (Brinkmann et al., 2012). The overall accuracy of the 2011 image classification was 93.7% for Tekes and 91.3% for Baicheng, with a kappa statistic of 0.92 and 0.87, respectively. A total of eight environmental variables were included in our study, divided into three groups (topography, anthropogenic disturbance, and climate) according to their attributes (Hietel et al., 2004). The variables are described in Table 2. The statistical yearbooks of Xinjiang Province for the years 1999, 2007, and 2012 provided data on the population of the main ethnic groups of each study site. A digital elevation model (DEM) of the study sites with a spatial resolution of 30 m was used to produce the elevation, slope and aspect variables and to evaluate the conditions of topography between areas that experienced different landscape transitions. The values of elevation and slope from the DEM can be used in the models directly, while the values of aspect layer cannot indicate the solar radiation an area place can receive. As a result, the aspect measurements were classified from 1 to 8 as follows: 1 (337.6◦ –22.5◦ ), 2 (22.6◦ –67.5◦ ), 3 (292.6◦ –337.5◦ ), 4 (67.6◦ –112.5◦ ), 5 (247.6◦ –292.5◦ ), 6 (112.6◦ –157.5◦ ), 7 (202.6◦ –247.5◦ ), and 8 (157.6◦ –202.5◦ ). The greater the value is, the sunnier and drier the area will be (Zhang et al., 2012). Because distantial attenuation is a significant characteristic of the residential environment (Liu et al., 2010b), we selected distance to the nearest resident as a variable of anthropogenic disturbance. The distance variable, referring to the distance-cost relationship in the Von Thünen model (Von Thünen, 1826), was calculated using the nearest villages. The layer of the vil-
3
lages, which were revised according to the physical truthing, was derived from the Landsat images from 2006. Then, raster images for the two study sites were generated to depict the Euclidean distance to the nearest village. All of the environmental variable layers were resampled to a 30-m resolution for modeling. The selected factors were applied as stable discriminant factors without variation in time. All of the spatial analyses were performed using ArcGIS 10.0. 2.3. Detection of landscape changes Short-term changes, swaps between locations and backward transitions can be analyzed by studying each transition between one land-use type and another (Mottet et al., 2006). The importance of studying not only the net changes but also the systematic transitions between the land-use types when trying to link patterns and processes in the research of landscape dynamics has been emphasized (Pontius et al., 2004). Changes were detected by a post-classification comparison of bi-temporal maps (Shoyama and Braimoh, 2011) of the study sites to identify the interlinkages of different landscapes. The dynamic changes of these landscapes were analyzed in terms of the total area, total change and change rate (%/year) over the two periods (1998–2006, 2006–2011). To characterize the landscape changes during the study periods in detail, two cross-tabulation matrices were computed separately for each area using the surface occupied by each land cover. Each cross-tabulation matrix consisted of six columns representing the area of each land cover at time t1 converted to the different land covers at time t2 (Silva et al., 2011). Entries along the diagonal of the matrices indicate the persistence of the corresponding land covers. 2.4. Statistical analyses A multivariate method of redundancy analysis (RDA) was used to explore the relationships among the environmental variables and the landscape changes. The RDA, a constrained multivariate ordination technique that is essentially the multivariate extension of multiple regressions (Legendre and Legendre, 1998; Fisichelli et al., 2013), needs a response matrix and an explanatory variable matrix. In this study, the response matrixes were the landscapetransition matrixes of different study periods, and the explanatory variable matrix contained all of the drivers selected for the analysis. The RDA can effectively identify the relationship between one or a group of variables and another group in the sight of statistics among ˇ all the of the gradient analysis methods (ter Braak and Smilauer, 2002). In our study, we focused only on the patterns and changes of three landscapes: cropland, grassland, and forest. To explore the relationships between the patterns and environmental variables, 50 plots (30 m × 30 m) were randomly placed within each studied landscape in Tekes and Baicheng where the landscape did not change during the successive periods. The random placement of the plots was achieved with Hawth’s Analysis Tools for ArcGIS (Beyer, 2007). Regarding the landscape changes, 8–17 patches with the largest areas in each transition were selected, and 2–6 plots were randomly placed within each patch. We chose transitions that had an area greater than one-thousandth of the total area in each county. The corresponding environmental variables of each plot constitute the data matrix of the environment. Because many of the explanatory variables were strongly interrelated, the variables with a variance inflation factor (VIF) greater than 20 were excluded, and the remaining variables were used in the RDA analyses. The Monte Carlo permutation test was used to test the significance of the eigenvalues for the first canonical axis and for all of the canonical axes.
4
Y. Zhang et al. / International Journal of Applied Earth Observation and Geoinformation 44 (2016) 1–10
Table 1 Satellite images used for landscape change analysis of the study sites. Path/row
Observation date (y/m/d)
Number of bands
Resolution (m)
Satellite (sensor)
145/031 146/030 146/031 145/031 146/030 146/031 145/031 146/030 146/031
1998/10/04 1998/09/25 1998/09/25 2006/10/10 2006/10/01 2006/10/17 2011/09/06 2011/09/13 2011/09/13
7 7 7 7 7 7 7 7 7
30 30 30 30 30 30 30 30 30
Landsat 5 (TM) Landsat 5 (TM) Landsat 5 (TM) Landsat 5 (TM) Landsat 5 (TM) Landsat 5 (TM) Landsat 5 (TM) Landsat 5 (TM) Landsat 5 (TM)
Table 2 Description of the environmental variables selected for the model. Variables
Description
Unit
Source of data
Topography variables Elev Slp Asp
Elevation Slope Aspect
m % ◦
Digital elevation model (cell size 30 m) Digital elevation model (cell size 30 m) Digital elevation model (cell size 30 m)
Anthropogenic disturbance Distr Pop GDP
Distance to the nearest resident Population density Gross domestic product
m no./km2 Billion
Base cartography of the study sites Database of natural resources of China (2003) Database of natural resources of China (2003)
Climate variables Temp Pre
Annual temperature Annual precipitation
◦
Database of natural resources of China Database of natural resources of China
2.5. Variation partitioning Variation partitioning was used in our study to further analyze the interrelationships among the different driver sets. This method can be used to identify unique, shared, and total contributions to model prediction in multiple regression analyses (Kerlinger and Pedhazur, 1973), which is more suitable than analyzing the individual contributions via their partial correlation coefficients (Peres-Neto et al., 2006). Shared variation is the confounded portion that is common to more than a single variable or a set of variables. The unique variation is the portion that is explained by a variable or a set of variables after removing the effects of all of the other measured variables. The total variation of a set of variables is the sum of the unique and shared portions (Borcard et al., 1992; Fisichelli et al., 2013). In this study, variation partitioning was conducted between pairs of values for anthropogenic disturbance, topography, and climate to identify the individual contributions and the interrelationships among the driver sets.
C mm
from 8.2% to 9.2% over the same period. In 1998, grassland covered 37.8% of the total area of Tekes and increased to 39.5% in 2006 at a rate of 0.574%/year and to 40.7% in 2011 at a rate of 0.566%/year. However, in Baicheng, the proportions of grassland were 11.2%, 7.7%, and 10.6% in 1998, 2006, and 2011, respectively. The rate of decrease during period 1 (1998–2006) was −4.666%/year, and the rate of increase was 6.458%/year during period 2 (2006–2011). As observed in Fig. 2, the area of grassland in Tekes was approximately 5-fold larger than that of cropland; however, in Baicheng, the area of grassland was slightly larger than the area of cropland in 1998 and 2011, and the area of grassland was even smaller than that of cropland in 2006, as a result of degradation. The greatest loss of grassland was in Baicheng, with a massive loss of 55,259 ha during period 1; however, there was an increase of 46,541 ha in the same county during the subsequent studied period. In terms of forest, the most marked change since 1998 was in Baicheng, with a deforestation of 27,072 ha during period 2. 3.2. Landscape transitions
3. Results 3.1. Changes in the composition of the landscape The total area of the different landscapes of Tekes and Baicheng and the corresponding proportions in 1998, 2006 and 2011 are presented in Table 3 and Fig. 2. As shown in Fig. 2, Tekes was dominated by barren land and grassland that in 1998 represented 78.4% of the total area and reached 79.5% and 78.3% in 2006 and 2011, respectively. The predominant landscape of Baicheng was barren land, which covered 73.9%, 76.0%, and 74.2% of the total land in the three study years, respectively. The proportion of forest in Tekes decreased from 14.5% in 1998 to 12.7% in 2006 and increased to 13.7% in 2011, whereas the proportion of forest in Baicheng increased from 5.6% in 1998 to 6.5% in 2006 and decreased to 4.8% in 2011. The areas of cropland in the two counties showed an increasing trend in the two periods. For Tekes, the proportion of cropland was 5.9% in 1998 and increased to 6.1% in 2011, whereas the proportion of cropland in Baicheng increased
The landscape transitions that occurred during the two periods in Tekes and Baicheng are presented in Table 4, providing insight into the explicit changes between different landscapes. The results confirm that barren land remained the dominant landscape at 34.8% and 32.1% of the study area in Tekes and 71.7% and 72.0% of the study area in Baicheng during the two periods, respectively. The transition matrixes also indicate many types of transitions between different landscapes. The transition of cropland into other landscapes in Tekes was higher during period 1 than during period 2. During period 1, cropland in Tekes suffered a great loss of coverage; 0.2% was turned into grassland and barren land, respectively, 0.1% to urban area and 0.4% to water. During period 2, the main losses of cropland in Tekes were due to conversion to urban (0.2%) and barren land (0.3%). In Baicheng, 7610 ha (0.5%) were transformed from cropland to barren land during period 1, and the area of cropland that was transformed to urban and barren land was 1707 ha (0.1%) and 5720 ha (0.4%), respectively, during period 2.
Y. Zhang et al. / International Journal of Applied Earth Observation and Geoinformation 44 (2016) 1–10
5
Table 3 Total area (ha) and change rate (%/year) of different landscapes in the two periods. Landscape category
Total area 1998
2006
2011
Period (1998–2006)
1
Period (2006–2011)
2
Total change
Rate of change
Total change
Rate of change
Tekes Cropland Grassland Forest Urban Barren land Water
49,420 314,394 120,542 5,221 337,921 5,193
49,405 329,170 105,851 5,877 333,445 8,943
50,625 338,615 114,485 6,915 312,920 9,131
−15 14,776 −14,691 656 −4,476 3,750
−0.004 0.574 −1.625 1.480 −0.167 6.794
1,220 9,445 8,634 1,037 −20,525 188
0.488 0.566 1.568 3.251 −1.271 0.416
Baicheng Cropland Grassland Forest Urban Barren land Water
130,267 177,380 89,446 4,994 1,170,551 11,823
138,886 122,121 102,743 5,147 1,204,607 10,957
145,049 168,662 75,671 6,916 1,175,884 12,279
8,619 −55,259 13,297 153 34,056 -866
0.801 −4.666 1.732 0.378 0.358 −0.951
6,163 46,541 −27,072 1,769 −28,723 1,322
0.868 6.458 −6.117 5.907 −0.483 2.279
Fig. 2. Proportions of different landscapes in 1998, 2006 and 2011 in the study area. Table 4 Area of landscape (ha) and percentage (%) converted from one type to the next for the period 1998–2006 and 2006–2011. %
Grassland
%
Tekes 1998–2006
Cropland Grassland Forest Urban Barren land Water
Cropland 41,783 1,832 205 569 4,877 139
5.0 0.2 0.0 0.1 0.6 0.0
1,526 271,496 15,630 18 40,438 61
0.2 32.6 1.9 0.0 4.9 0.0
Forest 372 3,916 99,012 0 2,300 252
% 0.0 0.5 11.9 0.0 0.3 0.0
Urban 1,151 78 72 4,255 209 112
% 0.1 0.0 0.0 0.5 0.0 0.0
Barren land 1,273 36,973 4,842 71 289,710 576
% 0.2 4.4 0.6 0.0 34.8 0.1
Water 3,315 99 781 308 388 4,052
% 0.4 0.0 0.1 0.0 0.0 0.5
Tekes 2006–2011
Cropland Grassland Forest Urban Barren land Water
43,781 3,512 245 712 2,092 283
5.3 0.4 0.0 0.1 0.3 0.0
658 271,646 8,152 161 57,994 5
0.1 32.6 1.0 0.0 7.0 0.0
764 12,215 95,971 79 5,008 447
0.1 1.5 11.5 0.0 0.6 0.1
1,601 222 5 4,704 360 23
0.2 0.0 0.0 0.6 0.0 0.0
2,346 41,442 1,256 191 267,310 375
0.3 5.0 0.2 0.0 32.1 0.0
255 132 223 30 681 7,810
0.0 0.0 0.0 0.0 0.1 0.9
Baicheng 1998–2006
Cropland Grassland Forest Urban Barren land Water
121,725 3,231 0 211 12,518 1,202
7.7 0.2 0.0 0.0 0.8 0.1
25 109,426 6,777 0 5,790 103
0.0 6.9 0.4 0.0 0.4 0.0
13 9,131 77,312 0 16,287 0
0.0 0.6 4.9 0.0 1.0 0.0
273 0 0 4,780 94 0
0.0 0.0 0.0 0.3 0.0 0.0
7,610 55,195 5,358 1 1,135,452 990
0.5 3.5 0.3 0.0 71.7 0.1
622 396 0 2 409 9,528
0.0 0.0 0.0 0.0 0.0 0.6
Baicheng 2006–2011
Cropland Grassland Forest Urban Barren land Water
130,140 219 0 98 13,363 1,230
8.2 0.0 0.0 0.0 0.8 0.1
154 108,377 15,317 0 44,788 27
0.0 6.8 1.0 0.0 2.8 0.0
0 5,093 67,500 0 3,079 0
0.0 0.3 4.3 0.0 0.2 0.0
1,707 0 0 5,049 154 5
0.1 0.0 0.0 0.3 0.0 0.0
5,720 8,362 19,927 0 1,141,087 787
0.4 0.5 1.3 0.0 72.0 0.0
1,165 70 0 0 2,136 8,907
0.1 0.0 0.0 0.0 0.1 0.6
6
Y. Zhang et al. / International Journal of Applied Earth Observation and Geoinformation 44 (2016) 1–10
Fig. 3. Biplot of RDA ordination of the distribution of landscapes and environmental variables in Tekes and Baicheng: T means Tekes, B means Baicheng, and 1–3 represent cropland, grassland and forest, respectively. Abbreviations for the environmental variables are listed in Table 2.
During period 1, the area of grassland that was transformed to forest was 3916 ha (0.5%) and 9131 ha (0.6%), and during period 2 it was 12,215 ha (1.5%) and 5093 ha (0.3%), in Tekes and Baicheng, respectively. However, as shown in Table 4, grassland experienced the largest area loss due to the transition to barren land (4.4% and 5% in Tekes, 3.5% and 0.5% in Baicheng during the two periods, respectively). Forest suffered a severe degradation in Tekes: 15,630 ha (1.9%) transitioned to grassland and 4842 ha (0.6%) to barren land during period 1, and 8152 ha (1%) and 1256 ha (0.2%) transitioned to grassland and barren land, respectively, during period 2. In Baicheng, the main loss of forest involved the transition to grassland (6777 ha) and barren land (5358 ha) during period 1 and to grassland (15,317 ha) and barren land (19,927 ha) during period 2. During period 1, the main contributor to cropland and grassland gains was barren land (4877 ha and 40,483 ha, respectively) in Tekes; in Baicheng, the area of barren land that transitioned to cropland, grassland, and forest was 12,518 ha (0.8%), 5790 ha (0.4%), and 16,287 ha (1.0%), respectively. However, during period 2, the largest transformation to grassland took place in Tekes, with 57,994 ha (7.0%) transitioning from barren land to grassland. 3.3. Drivers of the distribution and change of landscapes The biplot of the RDA ordination of the distribution of landscapes and environmental variables in Tekes and Baicheng is illustrated in Fig. 3, and the correlation coefficients between the environmental factors and the first four RDA axes are shown in Table 5. Although eight axes were used in the imputation, the last four axes have very little influence on neighbor selection because the distances are weighted by the eigenvalues (Ohmann et al., 2011). The eigenvalues for the first four axes were 0.173, 0.151, 0.051, and 0.013. The landscape-environment correlations of the first four RDA axes were 0.929, 0.870, 0.506, and 0.250, and the cumulative percentage variances of the landscape-environment correlation for the first four axes were 98.5%, indicating that RDA performed well in describing the relationships between the landscapes and the environmental gradients (Zhang et al., 2008). The Monte Carlo permutation test indicated that the eigenvalues for the first canonical axis and for all of the canonical axes were significant (p < 0.01). As shown in Table 5
and Fig. 3, the first axis had significant correlations with elevation (Elev), slope (Slp), distance to the nearest resident (Distr), temperature (Temp), population density (Pop), gross domestic product (GDP), and precipitation (Pre). Axis 2 was significantly related to all eight variables. The third axis was significantly correlated with aspect (Asp) and was also related to Pop and Temp, and the fourth axis was related to Pop and Pre. The biplots of the RDA ordination of the transition of landscapes and the environmental variables in Tekes and Baicheng during the two periods are illustrated in Fig. 4, and the correlation coefficients between the environmental factors and the first four RDA axes are shown in Table 6. During the period 1998–2006, the eigenvalues for the first four axes were 0.044, 0.032, 0.014, and 0.012. The landscape transition-environment correlations of the first four RDA axes were 0.937, 0.785, 0.522, and 0.493, and the cumulative percentage variances of the landscape transition-environment correlation for the first four axes were 87.0%, indicating that the RDA performed well in describing the relationships between the landscape transitions and the environmental gradients during period 1. The Monte Carlo permutation test showed that the eigenvalues for the first canonical axis and for all of the canonical axes were significant (p–< 0.01). Axis 1 had significant correlations with Elev, Slp, Distr, Pre, Temp, GDP, and Pop. Axis 2 was significantly correlated with Pre, Temp, and Slp and also related to Distr. Axis 3 was significantly related to Temp, Pre, Asp, and Slp and was also related to GDP. Axis 4 had significant correlations with Pre and Slp and was also related to Distr. During the period 2006–2011, the eigenvalues for the first four axes were 0.047, 0.035, 0.015, and 0.014. The landscape transitionenvironment correlations of the first four RDA axes were 0.900, 0.791, 0.537, and 0.505, and the cumulative percentage variances of the landscape transition-environment correlation for the first four axes were 83.60%, indicating that the RDA performed well in describing the relationships between the landscape transitions and the environmental gradients during period 2. The Monte Carlo permutation test indicated that the eigenvalues for the first canonical axis and for all of the canonical axes were significant in this ordination (p < 0.01). The first axis had significant correlations with all the variables, and the second axis had significant correlations with the same variables, except for Elev and Pop. The third axis was significantly correlated with Distr, Slp, Temp, Pre, and Asp and also related to GDP. The fourth axis was significantly related to Distr, Slp, and GDP and also related to Pop and Elev. 3.4. Variation partitioning The results of the variation partitioning for the three RDA processes are illustrated in Fig. 5. In the RDA of the distribution of the landscape and the environmental variables, the shared variation of topography with the other sets was 2% and 35%, respectively (Fig. 5(a)). Anthropogenic disturbance shared 10% and 61% of its explanatory power with climate and topography. In the transition of the landscapes and the environmental variables during period 1, there was no common portion between anthropogenic disturbance and climate, whereas topography shared 6% and 45% of its explained variation with climate and anthropogenic disturbance. During period 2, anthropogenic disturbance shared levels of explained variation with topography and climate similar to those in the first analysis process (the RDA of the distribution of landscapes and the environmental variables). The shared explanatory power between the climate and the other two driver sets was higher than that of the first two processes (21% for topography and 15% for anthropogenic disturbance). In the three RDA processes, topography, which explained the greatest variation, shared a large portion with anthropogenic disturbance but shared little explained variation with climate. The extent of the shared explanatory powers
Y. Zhang et al. / International Journal of Applied Earth Observation and Geoinformation 44 (2016) 1–10
7
Table 5 Correlation coefficients between environmental factors and the first four RDA axes in the distribution of the landscapes in the study area. Variables
Axis 1
Axis 2
Axis 3
Axis 4
Elev Slp Asp Distr Pop GDP Pre Temp
0.757** 0.640** −0.108 0.410** −0.271** −0.252** 0.242** 0.304**
0.456** 0.468** −0.329** 0.597** −0.315** −0.310** −0.466** −0.774**
0.104 −0.047 0.414** −0.012 −0.113* −0.077 0.039 −0.117*
0.005 −0.050 −0.054 −0.010 −0.114* 0.018 0.143* −0.002
Abbreviations for the variables are listed in Table 2. * p < 0.05. ** p < 0.01.
Fig. 4. Biplots of RDA ordination of the transition of landscapes and environmental variables in Tekes and Baicheng in period 1 (a) and period 2 (b): T means Tekes, B means Baicheng, and 1–5 represent cropland, grassland, forest, urban and barren land, respectively. Abbreviations for the environmental variables are listed in Table 2.
Table 6 The correlation coefficients between environmental factors and the first four RDA axes in the landscape transitions of the study area. Variables
Elev Slp Asp Distr Pop GDP Pre Temp
Period 1 (1998–2006)
Period 2 (2006–2011)
Axis 1
Axis 2
Axis 3
Axis 4
Axis 1
Axis 2
Axis 3
Axis 4
0.919** 0.802** -0.063 0.812** −0.211** −0.294** −0.327** 0.307**
0.037 0.160** −0.024 0.084* 0.038 0.008 0.514** −0.479**
0.049 −0.149** −0.197** −0.038 0.051 0.080* −0.240** −0.309**
−0.074 0.153** −0.049 0.096* 0.038 0.062 −0.164** 0.027
0.858** 0.732** −0.129** 0.706** −0.123** −0.231** −0.322** 0.526**
0.071 0.280** −0.147** 0.241** −0.027 −0.137** 0.620** −0.284**
−0.064 0.146** 0.122** 0.218** 0.045 0.093* −0.128** 0.129**
0.096* −0.147** 0.053 0.174** −0.105* −0.113** 0.057 0.070
Abbreviations for the variables are listed in Table 2. * p < 0.05. ** p < 0.01.
indicates complex relationships among all of the pairs of driving forces. 4. Discussion 4.1. The role of anthropogenic disturbance The Distr variable, as well as GDP and Pop, had different effects on the patterns and transitions in Tekes and Baicheng. These differences were attributable to the cultural backgrounds of residents of the two counties. Areas that share common ethnic and cultural features would have similar anthropogenic activities (Brinkmann et al., 2012). The populations of the main ethnic groups of Tekes and Baicheng are illustrated in Fig. 6. Although each ethnic group had a
fluctuating population in these two counties, the dominant groups did not change. Kazak, Han, Hui, Uygur, and Kirgiz were the main ethnic groups in Tekes. The population of Kazak increased from 59,975 (40.4%) in 1998 to 71,683 (42.7%) in 2011. The Uygur population showed a similar trend, with a population of 14,715 (9.9%) in 1998 and 19,044 (11.3%) in 2011. The Han population decreased from 36,531 (24.6%) in 1998 to 36,181 (21.6%) in 2011, and the population of Hui and Kirgiz decreased during 1998–2006 and increased during 2006–2011. The proportions of Hui were 13.0%, 12.6%, and 13.8%, and the proportions of Kirgiz were 6.0%, 5.7%, and 6.1%, in 1998, 2006, and 2011, respectively. The main ethnic groups in Baicheng were Uygur and Han. Uygur constituted 87.0% of the population of Baicheng in 1998, and the proportions in 2006 and 2011 were 85.8% and 87.1%. From 1998 to 2006, the proportion
8
Y. Zhang et al. / International Journal of Applied Earth Observation and Geoinformation 44 (2016) 1–10
Fig. 5. The results of variation partitioning for pairs of driver sets in the RDA of (a): the distribution of landscapes and environmental variables, (b): the transition of landscapes and environmental variables in Tekes and Baicheng in period 1 and (c): the transition of landscapes and environmental variables in Tekes and Baicheng in period 2. The overlapping middle sections represent the shared variation explained by both driver sets, and the left and right sections are the unique portions explained by the corresponding driver set.
Fig. 6. Populations of Tekes and Baicheng in 1998, 2006 and 2011.
of Han increased from 12.1% to 13.4%, but in 2011 it decreased to 11.9%. Different ethnic groups have their own production and living styles, and their land-use needs differ greatly. The places where agriculture-focused populations gathered have stronger farming activities than animal husbandry populations do. Much more cropland is needed in such areas, which leads people to reclaim cropland as the population increases; as a result, other types of landscape will be transformed to cropland. By contrast, grassland nearby is used for grazing, and when grassland degenerates, herdsmen find new grazing sites far from their houses. Landscapes that are near residents are more likely to be affected by people because of their reconstruction of the land. Bceause the main foods of ethnic Han, Uygur, and Hui people are grain crops, their activities greatly influence croplands, including paddy fields and dry land. Ethnic Kazak and Kirgiz people are engaged mainly in animal husbandry, and most of them move their pastures as the seasons change (Li, 2011).
The areas and patterns of grassland would be influenced more in the places where these two ethnic groups gather than in other places. According to the description and analysis above, the total proportions of ethnic Uygur and Han in Baicheng were 99.1%, 99.2%, and 99.0% in 1998, 2006, and 2011, respectively, and most of these populations are engaged mainly in agriculture. However, the total proportions of Uygur, Han, and Hui were 47.5% in 1998, 46.9% in 2006, and 46.7% in 2011, which were much lower than those in Baicheng. Few people rely on animal husbandry for living in Baicheng, whereas in Tekes the total proportions of Kazak and Kirgiz engaged mainly in animal husbandry were 46.4%, 48.2%, and 48.8% in 1998, 2006 and 2011, respectively. As a result, the patterns and changes in landscapes at the same distance to the nearest residents in Tekes and Baicheng were different. Accelerated repopulation increased the demand for land resources during the study periods. The use of wood for economic purposes and the subsistence needs of the local people resulted in a reduction in forest areas in these two counties. The deforested region was subsequently replaced by grassland or barren land. Livestock production in the study area was based primarily on cattle, buffaloes, goats, and sheep. Some areas were subjected to overgrazing; as a result, large amounts of grassland were degraded to barren land. When the remaining land resources could no longer meet the farmers’ needs, they sought new resources. With the increased pressure of the population and the loss of grassland, barren land nearby was turned into rangeland for livestock grazing. Intensive agricultural activities are important causes of the decline in the quality of farmland, which is therefore abandoned after several decades of cultivation. Because most of the people in Baicheng were engaged in agriculture, larger areas of barren land and grassland were converted to cropland than in Tekes. The variation of cultivation and grazing intensity in Tekes and Baicheng mirrored the differentiating disturbance by ethnic groups.
4.2. Landscape transitions and their drivers The croplands in the two counties were located in areas that had higher Pop and GDP. Grassland and forest were found at higher ele-
Y. Zhang et al. / International Journal of Applied Earth Observation and Geoinformation 44 (2016) 1–10
vations and steeper areas, and the distances to the nearest resident were much longer than those of cropland. As observed in Fig. 3, the grassland in Baicheng was more strongly correlated with Distr than that in Tekes, indicating that some of the grassland in Tekes was near residents and that most of the grassland in Baicheng was far from residents. Almost all of the people in Baicheng are engaged in agriculture and require more cropland for living. However, in Tekes, 48% of the total population is occupied in animal husbandry; therefore, a large amount of grassland is near their houses, which is convenient for pasture. During period 1, a large amount of cropland in Baicheng was transformed to barren land and much of the grassland was transformed into cropland, indicating that when the cropland started to degenerate, some grassland was gradually converted to cropland to meet the farmers’ needs. However, in Tekes, much of the cropland was transformed to grassland. During this period, the degradation of cropland in Baicheng was severe owing to the high intensity of agricultural production, and grassland was used to make up for the lack of cropland. Both of the transitions in Baicheng had negative correlations with slope, elevation, and distance to the nearest resident. In period 2, the transition of grassland to barren land in Tekes had a negative correlation with Distr, whereas in Baicheng, the same transition had a positive correlation with Distr. This phenomenon was related to anthropogenic activities and cultural backgrounds. The people in Tekes had more demand for grassland; therefore, much of the grassland degradation occurred near their living areas. The increase in barren land, which indicates a decrease in grassland, might also be a result of over-exploitation by grazing animals (Brinkmann et al., 2012). However, in Baicheng, the grassland was far from the residents; as a result, grassland degradation was influenced mainly by natural variables, such as Temp and Elev (Fig. 4(b)), and was weakly related to anthropogenic activities. 4.3. Variation partitioning The results of variation partitioning show the interrelationships among pairs of driver sets in the identification of the drivers of the distribution and change of landscapes. Our emphasis was on anthropogenic disturbance, and the relative strengths of the interrelated variables must be taken into account when attempting to project the influence of the target driver set (Fisichelli et al., 2013). The central role of the topography has strong influences on the distribution and the transition of landscapes. GDP and Pop were negatively related to Distr, which was due mainly to the effects of human activities, such as grazing and cultivation. The land far away from residents was less likely to be reclaimed or utilized. Only when the land nearby no longer satisfied the needs of the people would they consider further places. Anthropogenic disturbance is closely related to topography, and the strength of human activities may vary with topography because higher elevation and steeper slope were not suitable for people and would limit their production and lives. Climate was related to geographical location, topography, and certain other complicated factors; as a result, Temp and Pre had little shared variation with the anthropogenic disturbance set, but they all play important roles in the distribution and change of landscapes. 5. Conclusion This study detected the patterns and transitions of landscapes, as well as their drivers, in Tekes and Baicheng, where the configurations of the ethnic groups varied greatly. There were considerable variations between the two counties in the characteristics of the landscape patterns and their transitions. The amount of arable land
9
in Baicheng was much greater than that in Tekes. By contrast, Tekes had larger areas of grassland than Baicheng, most of which was located near the residents. The significant drivers that affected the configuration and transitions of landscape were Pre, Temp, Slp, Distr, and Elev. There are different ethnic groups in the two counties, and people of different ethnic groups have their own means of production, resulting in the different proportions of landscapes and landscape transitions in the study area. The results of the variation partitioning indicate complex interrelationships among all of the pairs of driver sets. All the variable sets had significant explanatory roles, most of which had both unique and shared variations with the others. Anthropogenic activities could strongly influence landscape patterns and likely accelerated the transitions of the landscapes. The dynamics of the landscapes became increasingly complex with increasing populations and economic development. Policy, technology, and socio-economic variables also significantly impact landscape transitions. In future studies, more attention should be paid to the impact and interrelationships of policy, technology, and socioeconomic variables. The results of this study can assist policy makers and planners in implementing effective protection strategies and sustainable landscape management.
Acknowledgement Financial support for this research was provided by the Project of the Special Research Foundation of Public Welfare (no. 201203005), the Special Program for National Science & Technology Basic Work (no. 2014FY110200A16), and the Fundamental Research Funds for the Central Universities (no. 2662015JC007).
Appendix A. Supplementary data Supplementary data associated with this article can be found, in the online version, at http://dx.doi.org/10.1016/j.jag.2015.06.013
References Beyer, H.L., 2007. Hawth’s analysis tools for ArcGIS. Available at http://www. spatialecology.com/htools Bicík, I., Jelecek, L., Stepanek, V., 2001. Land-use changes and their social driving forces in Czechia in the 19th and 20th centuries. Land Use Policy 18, 65–73. Borcard, D., Legendre, P., Drapeau, P., 1992. Partialling out the spatial component of ecological variation. Ecology 73, 1045–1055. Brinkmann, K., Schumacher, J., Dittrich, A., Kadaore, I., Buerkert, A., 2012. Analysis of landscape transformation processes in and around four West African cities over the last 50 years. Landsc. Urban Plann. 105, 94–105. Bürgi, M., 1999. A case study of forest change in the Swiss low-lands. Landsc. Ecol. 14, 567–575. Etter, A., McAlpine, C., Pullar, D., Possingham, H., 2006. Modelling the conversion of Columbian lowland ecosystems since 1940: drivers, patterns and rates. J. Environ. Manage. 79, 74–87. Fisichelli, N.A., Frelich, L.E., Reich, P.B., 2013. Climate and interrelated tree regeneration drivers in mixed temperate-boreal forests. Landsc. Ecol. 28, 149–159. Hietel, E., Waldhardt, R., Otte, A., 2004. Analysing land-cover changes in relation to environmental variables in Hesse, Germany. Landsc. Ecol. 19, 473–489. Imhoff, M.L., Bounoua, L., Defries, R., Lawrence, W.T., Stutzer, D., Tucker, C.J., Ricketts, T., 2004. The consequences of urban land transformation on net primary productivity in the United States. Remote Sens. Environ. 89, 434–443. Johnson, B.G., Zuleta, G.A., 2013. Land-use land-cover change and ecosystem loss in the Espinal ecoregion, Argentina. Agric. Ecosyst. Environ. 181, 31–40. Kerlinger, F.N., Pedhazur, E.J., 1973. Multiple Regression in Behavioral Research. Holt, Rinehart and Winston, New York, USA. Krausmann, F., Haberl, H., Schulz, N.B., Erb, K.H., Darge, E., Gaube, V., 2003. Land-use change and socio-economic metabolism in Austria, Part I: driving forces of land-use change 1950–1995. Land Use Policy 20, 1–20. Lambin, E.F., Geist, H.J., Lepers, E., 2003. Dynamics of land-use and land-cover change in tropical regions. Ann. Rev. Environ. Res. 28, 205–241. Legendre, P., Legendre, L., 1998. Numerical Ecology, 2nd English ed. Elsevier Science, Amsterdam. Li, C.H., 2011. Xinjiang Scenery Blog. Xinjiang Province, China.
10
Y. Zhang et al. / International Journal of Applied Earth Observation and Geoinformation 44 (2016) 1–10
Liu, J., Zhang, Z., Xu, X., Kuang, W., Zhou, W., Zhang, S., et al., 2010a. Spatial patterns and driving forces of land use change in China during the early 21st century. J. Geogr. Sci. 20 (4), 483–494. Liu, Y.L., Zheng, B., Turkstra, J., Huang, L., 2010b. A hedonic model comparison for residential land value analysis. Int. J. Appl. Earth Obs. Geoinf. 12S, S181–S193. Long, H.L., Li, X.B., 2005. Rural housing land transition in transect of the Yangtze River. Acta Geogr. Sinica 60, 179–188 (in Chinese). Maeda, E.E., De Almeida, C.M., De Carvalho Ximenes, A., Formaggio, A.R., Shimabukuro, Y.E., Pellikka, P., 2011. Dynamic modeling of forest conversion: simulation of past and future scenarios of rural activities expansion in the fringes of the Xingu National Park, Brazilian Amazon. Int. J. Appl. Earth Obs. Geoinf. 13, 435–446. Mottet, A., Ladet, S., Coqué, N., Gibon, A., 2006. Agricultural land-use change and its drivers in mountain landscapes: a case study in the Pyrenees. Agric. Ecosyst. Environ. 114, 296–310. Ohmann, J.L., Gregory, M.J., Henderson, E.B., Roberts, H.M., 2011. Mapping gradients of community composition with nearest-neighbour imputation: extending plot data for landscape analysis. J. Veg. Sci. 22, 660–676. Paudel, S., Yuan, F., 2012. Assessing landscape changes and dynamics using patch analysis and GIS modeling. Int. J. Appl. Earth Obs. Geoinf. 16, 66–76. Peres-Neto, P.R., Legendre, P., Dray, S., Borcard, D., 2006. Variation partitioning of species data matrices: estimation and comparison of fractions. Ecology 87, 2614–2625. Pontius Jr, R.G., Shusas, E., McEachern, M., 2004. Detecting important cate-gorical land changes while accounting for persistence. Agric. Ecosyst. Environ. 101 (2/3), 251–268. Random House, 1987. The Random House dictionary of the English language, 2nd ed. Unabridged. Seto, K.C., Fragkias, M., 2005. Quantifying spatiotemporal patterns of urban land-use change in four cities of China with time series landscape metrics. Landsc. Ecol. 20, 871–888. Shi, L., 1995. The theory of culture-ethnic culture and market culture. Ethno.-National. Stud. 1, 20–26 (in Chinese). Shi, Z.H., Li, L., Yin, W., Ai, L., Fang, N.F., Song, Y.T., 2011. Use of multi-temporal Landsat images for analyzing forest transition in relation to socioeconomic factors and the environment. Int. J. Appl. Earth Obs. Geoinf 13, 468–476. Shoyama, K., Braimoh, A.K., 2011. Analyzing about sixty years of land-cover change and associated landscape fragmentation in Shiretoko Peninsula, Northern Japan. Landsc. Urban Plann. 101, 22–29.
Silva, J.S., Vaz, P., Moreira, F., Catry, F., Rego, F.C., 2011. Wildfires as a major driver of landscape dynamics in three fire-prone areas of Portugal. Landsc. Urban Plann. 101, 349–358. ˇ ter Braak, C.J.F., Smilauer, P., 2002. Canoco Reference Manual and User’s Guide to CanoDraw for Windows. Software for Canonical Community Ordination (Version 4.5). Centre for Biometry Wageningen/Microcomputer Power, Wageningen, NL/Ithaca NY, USA, pp. 352 pp. Tzanopoulos, J., Vogiatzakis, I.N., 2011. Processes and patterns of landscape change on a small Aegean island: the case of Sifnos, Greece. Landsc. Urban Plann. 99, 58–64. Von Thünen, J.H., 1826. Der isolierte Staat in Beziehung auf Landwirtschaft und Nationalökonomie. Hamburg, Germany. Walker, R., Drzyzga, S.A., Li, Y., Qi, J., Caldas, M., Arima, E., Vergara, D., 2004. A behavioral model of landscape change in the Amazon Basin: the colonist case. Ecol. Appl. 14, 299–312. Wyman, M.S., Stein, T.V., 2010. Modeling social and land-use/land-cover change data to assess drivers of smallholder deforestation in Belize. Appl. Geogr. 30, 329–342. Xinjiang Province Statistical Bureau, 1999. Statistical yearbook of Xinjiang Province 1999. Xinjiang Province, China. Xinjiang Province Statistical Bureau, 2012. Statistical yearbook of Xinjiang Province 2012. Xinjiang Province, China. Xu, W., 2004. The changing dynamics of land-use change in rural China: a case study of Yuhang, Zhejiang Province. Environ. Plann. A 36, 1595–1615. Zhang, J.T., Dong, Y.R., Xi, Y.X., 2008. A comparison of SOFM ordination with DCA and PCA in gradient analysis of plant communities in the midst of Taihang Mountains, China. Ecol. Inf. 3, 367–374. Zhang, J.T., Xiang, C.L., Li, M., 2012. Integrative ecological sensitivity (IES) applied to assessment of eco-tourism impact on forest vegetation landscape: a case from the Baihua Mountain Reserve of Beijing, China. Ecol. Indic. 18, 365–370. Zhang, J.X., Liu, Z.J., Sun, X.X., 2009. Changing landscape in the three Gorges Reservoir Area of Yangtze River from 1977 to 2005: land use/land cover, vegetation cover changes estimated using multi-source satellite data. Int. J. Appl. Earth Obs. Geoinf. 11, 403–412.