Environ Geol (2009) 57:1825–1837 DOI 10.1007/s00254-008-1470-2

ORIGINAL ARTICLE

Using Landsat data to determine land use changes in Datong basin, China Ziyong Sun Æ Rui Ma Æ Yanxin Wang

Received: 27 May 2008 / Accepted: 26 June 2008 / Published online: 18 July 2008 Ó Springer-Verlag 2008

Abstract The aim of this study was to determine land use changes in Datong basin using multitemporal Landsat data for the period of 1977–2006. Four dates of Landsat images from 1977, 1990, 2000, and 2006 were selected to classify the study area. Based on the supervised classification method of maximum likelihood algorithm, images were classified into six classes: water, urban, forest, agriculture, wetland, and barren land. A multidate postclassification comparison change detection algorithm was used to determine changes in land use in four intervals. It is found that (1) urban land area increased 213% due to urbanization that resulted from rapid increase of urban population and high-speed economic development, (2) agriculture area increased 34.0% due to land reclamation that resulted from rapid increase of rural population and improvement of irrigation capacity, (3) forest area decreased 20.9% due to deforestation for urban area and agricultural use, (4) barren land area decreased 78.2% due to cultivation for agricultural use, and (5) water and wetland decreased 39.1 and 67.1%, respectively, due to exploitation of surface water and decrease of recharge from groundwater to surface water that resulted from over exploitation of groundwater.

Z. Sun (&)
R. Ma
Y. Wang School of Environmental Studies and MOE Key Laboratory of Biogeology and Environmental Geology, China University of Geosciences, 430074 Wuhan, People’s Republic of China e-mail: [email protected] Y. Wang e-mail: [email protected] R. Ma Department of Geological Sciences, The University of Alabama, Tuscaloosa, AL 35487, USA e-mail: [email protected]

Keywords Land use
Remote sensing
Image classification
Change detection
China

Introduction In the last few decades, conversion of grassland and forest into cropland and residential areas has risen dramatically due to population growth, food scarcity, and urbanization (Houghton 1994; Meyer and Turner 1994; Turner et al. 1994, 1995; Squires 2002; Aboel Ghar et al. 2004). This acceleration has spurred concerns about land use/land cover change (LUCC) for its great effects on regional water cycle, biological diversity, terrestrial ecosystem productivity and adaption, soils and their fertility, water quality and air quality. Land use change, the role of human activities and institutions in bringing the change, and the consequences of the change require careful consideration by land managers and policy makers in order to preserve environmental resources (Naiman et al. 1997). To improve understanding of land changes under diverse cultural, economic, political, institutional, and environmental situations, a flurry of research on the causes and consequences of land use change spawned recently (Meyer and Turner 1994; Lambin et al. 1999; Lunetta et al. 2002). Collecting accurate and timely information on land use is crucial for land use change research (Giri et al. 2005). There are various methods that can be used in the collection of land use data but the use of satellite remote sensing technologies can greatly facilitate the process (Gautam et al. 2003). In contrast to traditional ground-based surveys, satellite remote sensing provides greater amounts of information on the geographic distribution of land use in a relatively cost and time savings way for regional size areas (Rogan and Chen 2004; Yuan et al. 2005). Importantly,

123

1826

repeated satellite images can allow for timely and consistent estimates of changes in land use trends over large areas and have the additional advantage of ease of data capture into Geographical Information Systems (GIS), which can facilitate greatly the analysis and presentation of such data (ESCAP 1997; Prakash and Gupta 1998; Hathout 2002; Shalaby and Tateishi 2007). The recent work has shown that satellite remote sensing is a strong tool for providing accurate and timely geospatial information describing most types of land use (Wolter et al. 1995; Lunetta and Balogh 1999; Oettera et al. 2000; Yang 2002; Alberti et al. 2004; Goetz et al. 2004; Yuan et al. 2005). Numerous methods have been developed for land use change detection. Yuan et al. (1998) divide the methods for change detection and classification into preclassification and postclassification techniques. The preclassification techniques apply various algorithms directly to multiple dates of satellite imagery to generate ‘‘change’’ versus ‘‘nochange’’ maps. These techniques locate changes but do not provide information on the nature of change (Singh 1989; Ridd and Liu 1998; Yuan et al. 1998). On the other hand, postclassification comparison methods use separate classifications of images acquired at different times to produce difference maps from which ‘‘from–to’’ change information can be generated (Jensen 2005). Although the accuracy of the change maps is dependent on the accuracy of the individual classifications and is subject to error propagation, the classification of each date of imagery builds a historical series that can be more easily updated and used for applications other than change detection. The postclassification comparison approach also compensates for variation in atmospheric conditions and vegetation phenology between dates, since each classification is independently produced and mapped (Yuan et al. 1998, 2005; Coppin et al. 2004). Land use patterns have been dramatically changed in China since the late 1980s, especially with regard to urbanization and loss of cultivated land. The national and regional scale land use change in China over the last two decades has attracted more and more attention to analyze the processes, driving forces, impacts, and future trends (Liu et al. 2003; Zhang and Zhang 2007). However, there are relatively few studies using a long time series of Landsat data to determine land use changes at local scale in China (Chen and Li 2007). Datong basin is located in the agro-pasture zigzag zone and the semiarid ecologically fragile zone in China (Zhao et al. 2002; Tan 2007). It is characterized by the transition and vibration of natural condition and the sensitivity of environment quality to land use change. Datong basin is one of the most important coal-producing areas in China, which accelerates greatly the economic development of the area. However, increasing population and urbanization

123

Environ Geol (2009) 57:1825–1837

have caused irregular land use in the last three decades (Zhang 2006). Up to now, uncontrolled and unregulated cultivation, construction, urbanization, and industrialization activities have presented serious environmental problems such as soil erosion, groundwater and surface water pollution, desertization, and salinization (Guo 1992; Ma and Su 1998; Yang and Jia 2001; Guo and Chen 2006). Even though these negative effects of land use change in Datong basin are concerned widely, accurate monitoring and quantitative evaluation of land use over time have not been reported. The main aim of this study was to determine land use changes in Datong basin using multitemporal Landsat data for the period of 1977–2006. A comparison will be drawn among land use patterns in different years (1977, 1990, 2000, and 2006) over the study area.

Study area The study area, Datong basin, is located in the northeastern part of Shanxi province, north China (Fig. 1) and covers mainly plain terrain of Datong city and Shuozhou city. The total area is about 634,220 ha, including the water bodies of the lakes and rivers. The basin is surrounded by mountains. The altitude varies between 863 and 1,266 m above sea level from northeast to southwest. Sanggan River, with the tributaries of Yu River, Hun River, Huangshui River, Kouquan River, Libazhuang River, Dayu River, etc., runs through the basin from southwest to northeast. Scattered hill, proluvial fan and alluvial flat occur ordinarily from the margin to the center of the basin. The alluvial flat of the central basin, mostly located in both sides of Sanggan River from Shuozhou City to Datong City, is mainly used for agricultural activities. The basin is mainly covered by chestnut soil, except that saline soil occurs in riparian lowlands of Sanggan River due to salinization. Datong basin is located in the arid/semiarid region of north China. Climate is mainland monsoon-type. The average annual temperature is about 7°C and average annual precipitation is about 400 mm, 80% of which occurs from June to September. The average evaporation rate is about 2,000 mm. There are two growing seasons annually. Main crops are wheat, millet, broom corn millet, and potato.

Data Four periods of Landsat images were selected to classify the study area: June 19 and July 8, 1977; July 6, 1990; July 1, 2000; May 23, 2006. The satellite data were chosen

Environ Geol (2009) 57:1825–1837

1827

Fig. 1 Location of the study area (Datong basin, China)

according to the following criteria: (1) a long time series of images should be available for the study area; (2) the images should be acquired from May to August, the growing season of the study area, to maximize separability of land use classes; (3) all images should have less than 10% cloud cover. So, the images of Landsat-2 MSS for 1977 and Landsat-5 TM for 1990 and 2006 were chosen. Landsat-7 ETM + images were selected for 2000, because no cloud-free Landsat-5 TM images were available from May to August in the year. The characteristics of the image data are presented in Table 1. The study area is entirely contained within path 134, row 32 and path 135, rows 32 and 33 for Landsat MSS images, and path 125, rows 32 and 33 for Landsat TM/ETM+ images (Fig. 2). The data used for reference mainly include the following: (1) detailed topographic maps in 1976, at scale of Table 1 Characteristics of the satellite data used for land use mapping in the study area

1/50,000; (2) land use maps in 1996 and 2002 based on traditional ground surveys, at scale of 1/20,000; (3) ground reference data obtained from land survey with hand-held GPS (global positioning system) in 2006.

Methods Image preprocessing Radiometric correction of the images had already been carried out; so, it was not applied. Accurate geometric rectification is essential for change detection, since the potential exists for registration errors to be interpreted as land use change, leading to an overestimation of actual change (Stow 1999). Change detection analysis is performed on a

Type of imagery

Date

Path

Row

Nominal spatial resolution (m)

Average cloud cover (%)

Landsat-2 MSS

June 19, 1977

134

32

79

10

July 8, 1977

135

32

79

10

July 8, 1977

135

33

79

10

July 6, 1990

125

32

30

0

July 6, 1990

125

33

30

0

July 1, 2000

125

32

30

0

July 1, 2000

125

33

30

0

May 23, 2006 May 23, 2006

125 125

32 33

30 30

10 0

Landsat-5 TM Landsat-7 ETM+ Landsat-5 TM

123

1828

Environ Geol (2009) 57:1825–1837

Fig. 2 Landsat satellite image frames. The study area is entirely contained within path 134, row 32 and path 135, rows 32 and 33 for Landsat MSS images (a), and path 125, rows 32 and 33 for Landsat TM/ETM+ images (b)

pixel-by-pixel basis; therefore, any misregistration greater than one pixel will provide an anomalous result of that pixel. To overcome this problem, the RMSE (root meansquare error) between any two dates should not exceed 0.5 pixel (Lunetta and Elvidge 1998). In this study, imageto-map rectification was done firstly for the image of 2006 using ground control points from topographic maps with scale of 1/50,000 in 1976. Then this registered image was used to perform image-to-image rectification for the images of 1977, 1990, and 2001. To avoid altering the original pixel values of the images, nearest neighbor resampling method was used. All images were rectified to UTM Zone 49 N, WGS 1984 using at least 25 well distributed GCPs (ground control points). The RMSE between each two images was less than 0.5 pixel, which is acceptable. After the rectification, the multiple images belonging to the same period were mosaic into a single seamless composite image using feathering strategy (ERDAS 1999). Then, a subarea covering the study area was extracted from the image. Image classification According to the land cover and land use classification system developed by Anderson et al. (1976) and Jensen and

Cowen (1999) for interpretation of remote sensor data at various scales and resolutions, Landsat data are only suitable for level I land use/land cover mapping. So, US Geological Survey level I was chosen and referred to for the classification system in this study. The study area was classified into six classes: water, urban, forest, agriculture, wetland, and barren land. Description of these land use classes are presented in Table 2. The training and testing data for the supervised classification and accuracy assessment were collected using false color composite, soil-adjusted vegetation index (SAVI), tasseled cap (TC) transform, topographic maps in 1976 (1/ 50,000), land use maps in 1996 and 2002, and fieldwork in 2006. The data samples were then split into two subsets: the training data and the test data. False color composites can help to visualize land use without any enhancement processes. The false color images were generated with red = band 4, green = band 3, blue = band 2 for Landsat ETM + and TM, and red = band 4, green = band 2, blue = band 1 for MSS images (Jensen 2005). The vegetation index indicates the amount of green vegetation present, which is useful and important for land use change identification. To minimize the soil ‘‘noise,’’ SAVI was selected. It is calculated by the following equation: SAVI ¼

ð1 þ LÞðqnir  qred Þ qnir þ qred þ L

where qnir is the near-infrared-reflected radiant flux, qred is the red-reflected radiant flux, and L is a canopy background adjustment factor that accounts for differential red and near-infrared extinction through the canopy (Huete 1998; Huete et al. 1992; Karnieli et al. 2001). An L value of 0.5 had been proved to minimize soil brightness variations and eliminate the need for additional calibration for different soils (Huete and Liu 1994), and was adopted in this study. The TC transformation can be used to disaggregate the amount of soil brightness, vegetation, and moisture content in individual pixels in a Landsat image (Kauth et al. 1979; Crist et al. 1986; Price et al. 2002). In this study, TC

Table 2 Land use classification scheme used in this study No.

Class

1

Water

All areas of open water, including rivers, streams, lakes, and reservoirs

2

Urban

Including residential, commercial, and industrial buildings as well as open transportation facilities, airports, highways, railways, and single/multiple family houses

3

Forest

Including dense forest, open forest, orchards, and nurseries

4

Agriculture

Areas cultivated with dense annual crops and vegetables, including dry land and irrigable land

5

Wetland

Lowland areas saturated with moisture all year or flooded seasonally, including riparian wetland, lakeshore wetland, and flat water with dense hydrophytes

6

Barren land

Uncultivated areas of sparse plant cover, including saline alkali land along river/lake beaches, barren rock or sand in sloping fields, bare land, and cultivated land without crops

123

Description

Environ Geol (2009) 57:1825–1837

transformation was performed using the coefficients developed by Kauth et al. (1979) for Landsat MSS image and Crist et al. (1986) for Landsat TM/ETM + image. Soil brightness image and greenness vegetation image were obtained for 1977, and soil brightness images, greenness vegetation images, and moisture content images were obtained for 1990, 2000, and 2006, respectively. There are topographic maps in 1976 (1/50,000) and land use maps in 1996 and 2002 (1/20,000) based on ground surveys covering the study area. They were used to collect training and testing data on screen for the supervised classification and accuracy assessment of Landsat images in 1977 and 2000, respectively. Field investigation was conducted in 2006, providing essential independent reference data for identifying land use types within the Landsat scenes as well as for accuracy assessment. In this study, Landsat data of four dates were independently classified using the supervised classification method of maximum likelihood algorithm. Firstly, the training sites, which represent homogeneous examples of prior known land cover types, were located in the remotely sensed data using the training data from false color composite, SAVI, TC transform, land use maps, topographic maps, and fieldwork. The spectral characteristics of these training sites were counted in the form of multivariate statistical parameters and used to define the classification signatures. After evaluating and adjusting the signatures, classification signature files were subsequently created and used by maximum likelihood classifier to automatically categorize every pixel in the entire image into the land use class, of which it had the highest likelihood of being a member. The classified images were further smoothed with a majority filter with a 3 9 3 kernel to reduce the number of misclassified pixels (ERDAS 1999). Accuracy assessment Accuracy assessment was performed for 1977, 1990, 2000, and 2006 land use maps. The number of reference pixels is an important factor in determining the accuracy of the classification. It has been shown that more than 250 reference pixels are needed to estimate the mean accuracy of a class to within ±5% (Congalton 1991). A stratified random sampling approach was adopted in the accuracy assessment, and 751, 700, 708, and 1,116 reference pixels were selected for the land use maps of 1977, 1990, 2000, and 2006, respectively. The overall accuracy and a Kappa analysis were used to perform classification accuracy assessment based on error matrix analysis. The overall accuracy is calculated by summing the number of pixels classified correctly and dividing by the total number of pixels. Kappa analysis is a discrete multivariate technique of use in accuracy assessment (Foody 2002). Kappa

1829

analysis yields a statistic, kappa coefficient of agreement (Khat), which is a measure of agreement or accuracy between the classification map and the reference data (Rosenfield and Fitzpatrick 1986; Congalton 1991). Khat is computed as following equation: P P N 2 ki¼1 xii  ki¼1 ðxiþ  xþi Þ Khat ¼ P N 2  ki¼1 ðxiþ  xþi Þ where k is the number of rows in the matrix, xii is the number of observations in row i and column i, xi+ and x+i are the marginal totals for row i and column i, respectively, and N is the total number of pixels. Khat values [0.80 represent strong agreement or accuracy between the classification map and the reference information. Khat values between 0.40 and 0.80 represent moderate agreement. Khat values \0.40 represent poor agreement (Landis and Koch 1977). At the same time, producer’s accuracy, user’s accuracy, and conditional kappa coefficient of agreement (Kc) were used to estimate the accuracy of each individual class. The producer’s accuracy is a measure indicating the probability that the classifier has labeled an image pixel into Class A given that the ground truth is Class A. The user’s accuracy is a measure indicating the probability that a pixel is Class A given that the classifier has labeled the pixel into Class A. The conditional kappa coefficient of agreement is a measure of agreement between the reference and classification data with change agreement eliminated for user’s accuracies using the following equation (Congalton and Green 1999): Kc ¼

Nðxii Þ  ðxiþ  xþi Þ Nðxiþ Þ  ðxiþ  xþi Þ

where xii is the number of observations correctly classified for a particular category, xi+ and x+i are the marginal totals for row i and column i associated with the category, and N is the total number of observations in the entire error matrix. Change detection Following the classification of imagery from the individual years, a multidate postclassification comparison change detection algorithm was used to determine changes in land cover in four intervals, namely, 1977–1990, 1990–2000, 2000–2006, and 1977–2006. Postclassification comparison is proved to be the most effective approach for change detection, because each data is separately classified, thereby minimizing the problem of normalizing for atmospheric and sensor differences between two dates (Jensen 2005). The postclassification approach provides ‘‘from–to’’ change information, and the kind of land use transformations that

123

1830

Environ Geol (2009) 57:1825–1837

Results and discussion

accuracy of 89.96% and a kappa coefficient of agreement of 0.8763 (Table 3d). Thus, all land use classes were accurately classified and the result is acceptable. In terms of the producer’s accuracies, each class was above 85%, while in terms of the user’s accuracy, each class was also above 85%. The conditional kappa statistics for each land use class, all exceeded 0.80.

Classification accuracy

Land use change

Error matrices used to assess classification accuracy are summarized for four periods in Table 3. For the 1977 land use map, accuracy assessment result shows an overall accuracy of 92.94% and a kappa coefficient of agreement of 0.9130 (Table 3a). According to Landis and Koch (1977), all land use classes were quite accurately classified. Either in terms of producer’s accuracy or user’s accuracy, all classes were above 85%. For all land use classes, the conditional kappa coefficients of agreement exceeded 0.85 with the exception of forest, which was 0.8403. This might be caused by confusion between grove and agriculture, because both have similar vegetation cover and spectrum characteristics in summer. For the 1990 land use map, accuracy assessment result shows an overall accuracy of 89.86% and a kappa coefficient of agreement of 0.8715 (Table 3b), which represents that the classification result is accurate (Landis and Koch 1977). In terms of producer’s accuracy, all classes were above 85% except wetland, while in terms of user’s accuracy, all classes were above 80% except water. For all land use classes, the conditional kappa coefficients of agreement exceeded 0.80 with the exception of water, which could be confused with wetland. The wetland and water classes showed confusion, because wetland has high water content and the two land use classes have a similar reflection value. For the 2000 land use map, accuracy assessment result indicates an overall classification accuracy of 89.41% and a kappa coefficient of agreement of 0.8652 (Table 3c). The classification result is accurate and acceptable (Landis and Koch 1977). When examining the producer’s accuracies, all land use classes were above 80% except wetland, which was confused with agriculture. It is postulated that a lot of riparian wetland had been cultivated by 2000 and the residual wetland patches became smaller and fragmented. In the case, it is difficult to distinguish them from the matrix of agriculture. This postulation is proved by the following land use change result from 1990 to 2000. In terms of the user’s accuracy, each class exhibited above 85%. The conditional kappa statistics for each land use class, all exceeded 0.80. For the 2006 land use map, a total of 1,116 pixels were selected. The result indicates an overall classification

The land use maps for 1977, 1990, 2000, and 2006 are presented in Fig. 3, and the area and area changed of the six land use classes during the four intervals are shown in Table 4. Results show that urban and agriculture area increased while forest, wetland, and barren land declined continuously over the study period, which is consistent with the increasing trend of population of the region (Fig. 4). Water area decreased during the first period (1977–1990) and third period (2000–2006), but increased during the second period (1990–2000). To further evaluate losses and gains among the six land use classes, matrices of land use changes from 1977 to 1990, 1990 to 2000, 2000 to 2006, and 1977 to 2006 were created in Table 5.

have occurred can be easily calculated. Then, crosstabulation analysis was carried out to analyze the spatial distribution of different land use classes and land use changes.

123

Urban Urban areas have increased from 13,694 ha in 1977 to 41,883 ha in 2006 for the study area (Table 4), thus representing an increase of 213% in land area. Based on Fig. 3, the spatial expansion of urban area is clearly visible. In 1977, the urban areas were small and mainly located in Datong city. By 1990, five new highly concentrated areas, namely Shuozhou, Shanyin, Huairen, Yingxian, and Hunyuan, had emerged. From 1990 to 2006, the six concentrated areas expanded further. Urban expansion can be also seen in rural villages and transportation areas had taken place since 2000. The observed expansion of urban area could be explained by the rapid increase of urban population of Datong basin (Fig. 4), which increased from 0.79107 million in 1982 to 1.373 million in 2004, representing a continuous increase of 74.6%. However, the urban population of the region increased less than the urban expansion (213% in area). Another major reason for urban expansion should be the rapid economic development since 1978 when the national policy of reform and opening was adopted in China. Table 5 indicates that increases in urban areas mainly came from conversion of agriculture and forest area to urban area. Of the 28,188 ha of total growth in urban land use from 1977 to 2006, 46.0% was converted from agriculture, 32.1% from forest, and 14.6% from barren land. From 1977 to 2006, 10,099 ha of forest was converted to urban area, whereas at the same time 1,045 ha of urban

Environ Geol (2009) 57:1825–1837

1831

Table 3 Accuracy assessment results of the land use map produced from Landsat data Classified data

Reference data Water

Urban

Forest

Agriculture

Wetland

Barren land

Row total

User’s accuracy (%)

Kc

(a) 1977 Water

83

0

0

0

1

0

84

98.81

0.9864

Urban

2

67

1

0

0

0

70

95.71

0.9529

Forest

2

0

136

12

6

0

156

87.18

0.8403

Agriculture

1

0

7

196

3

1

208

94.23

0.9198

Wetland

5

1

4

1

92

0

103

89.32

0.8762

Barren land Column total

3 96

0 68

0 148

2 211

1 103

124 125

130 751

95.38

0.9446

Producer’s accuracy (%)

86.46

98.53

91.89

92.89

89.32

99.20

(b) 1990 Water

38

0

0

1

6

3

48

79.17

0.7784

Urban

1

100

0

2

0

1

104

96.15

0.9544

Forest

1

6

127

15

2

0

151

84.11

0.8017

Agriculture

0

1

10

206

2

1

220

93.64

0.9044

Wetland

1

0

2

0

51

0

54

94.44

0.9389

Barren land

1

3

0

10

2

107

123

86.99

0.8451

Column total

42

110

139

234

63

112

700

Producer’s accuracy (%)

90.48

90.91

91.37

88.03

80.95

95.54

(c) 2000 Water

62

6

0

3

1

0

72

86.11

0.8461

Urban

3

95

2

5

2

3

110

86.36

0.8377

Forest Agriculture

2 0

3 6

132 6

9 217

1 5

1 1

148 235

89.19 92.34

0.8650 0.8834

Wetland

0

0

0

1

35

0

36

97.22

0.9703

Barren land

2

3

1

8

1

92

107

85.98

0.8376

Column total

69

113

141

243

45

97

708

Producer’s accuracy (%)

89.86

84.07

93.62

89.30

77.78

94.85

Water

125

5

2

0

6

2

140

89.29

0.8785

Urban

2

136

5

11

2

2

158

86.08

0.8383

Forest

1

2

180

12

3

1

199

90.45

0.8823

Agriculture

1

10

19

295

4

4

333

88.59

0.8388

(d) 2006

Wetland

3

2

5

0

130

0

140

92.86

0.9179

Barren land

0

0

0

8

0

138

146

94.52

0.9369

Column total

132

155

211

326

145

147

1,116

Producer’s accuracy (%)

94.70

87.74

85.31

90.49

89.66

93.88

area was converted to forest. The later change may seem to be classification errors, but forested areas are among some of the most sought after areas for developing new housing. Streets, highways, and parks were generally classified as urban, but when urban tree canopies along the streets and highways or in parks grow and expand, the associated pixels may be classified as forest. It is noted that the changes from urban to forest occurred almost entirely near highways and streets or in newly developed residential areas (Fig. 3). Classification errors may also cause other

unusual changes. For example, between 2000 and 2006, 11,014 ha of urban changed to agriculture. These changes are most likely associated with omission and commission errors in the land use classifications map. Agriculture Table 4 shows that agriculture had an area of 312,507 ha in 1977, 351,976 ha in 1990, 388,213 ha in 2000, and 418,697 ha in 2006, representing a net increase of 34.0%

123

1832

Environ Geol (2009) 57:1825–1837

Fig. 3 Land use classification maps of Datong basin

Table 4 Results of land use classification for 1977, 1990, 2000, and 2006 images showing area and area changed (ha) of each class Class name

Area 1977

Area changed 1990

2000

2006

1977–1990

1990–2000

2000–2006

1977–2006

Water

19,938

11,141

12,820

12,141

-8,797

+1,679

-679

-7,797

Urban

13,694

21,814

41,008

41,883

+8,120

+19,194

+875

+28,188

Forest

167,505

156,606

145,955

132,483

-10,899

-10,651

-13,472

-35,022

Agriculture

312,507

351,976

388,213

418,697

+39,469

+36,237

+30,484

+106,190

Wetland

24,858

17,110

12,842

8,170

-7,748

-4,267

-4,672

-16,688

-20,143

-42,192

-12,536

-74,871

Barren land

95,717

75,574

33,382

20,846

Total

634,220

634,220

634,220

634,220

(106,190 ha) from 1977 to 2006. Increases in agriculture area mainly came from conversion of barren land and forest to agriculture. From 1977 to 2006, agricultural land lost 9,853 ha to barren land, while it gained 66,102 ha from it, representing a net gain of 56,249 ha from barren land, which is 53.0% of net increase of agriculture areas. Net gain of agriculture from forest is 44,519 ha, which contributes 41.9% to net increase of agriculture areas. The increasing trend of agriculture could be attributed to the land reclamation that resulted from rapid increase of rural population (about 20% from

123

1982 to 2004, shown in Fig. 4) and improvement of irrigation capacity. Although there was a net gain of agriculture area from forest and barren land, a substantial proportion of agriculture lost to barren land and forest at the same time (Table 5). The loss to forest might have been resulted due to the change from cropland to man-made forest, including orchard and timber product forest. The loss to barren land might be due to the abandonment of some agricultural plots after a few seasons of cultivation. It is also possible that some agricultural plots were cultivated biennially, and they

Environ Geol (2009) 57:1825–1837

Fig. 4 Population of the Datong basin from 1982 to 2005

might be fallow and have not been recovered by vegetation during the time of satellite observation. As a result, there is a possibility that these lands were classified as barren land. Figure 3 shows that the changes from agriculture to barren land occurred almost entirely at the margins of the basin or near the hills, where the proluvial fan mouths are located. These areas, with coarse-grained lean soil, low soil water retention capacity, and groundwater found 100–300 m beneath the surface (Guo and Wang 2004), do not have enough soil fertilities and water resources to sustain cultivation in successive years. They are often cultivated biennially or depending on the availability of precipitation. Based on Fig. 3, the spatial conversion of barren land and forest to agriculture is clearly visible. The agricultural land was very scattered, mostly located in both sides of Sanggan River and its tributaries in 1977. However, the barren land in west and northeast basin and the forest in southeast basin had been cultivated for agriculture ninetenths by 2006, resulting in spatial expansion of agriculture from the center to the margin of the basin. At the same time, the forest in the central basin was cleared for cultivating in large amounts, which caused connection of scattered agriculture patches. In land use classification map of 2006 (Fig. 3), the agricultural land almost became one single patch in appearance.

1833

conversed to urban and agriculture, respectively, from 1977 to 2006, though 60,204 and 16,270 ha were gained from agriculture and barren land, respectively, at the same time. Although the conversions from forest to agriculture can be found throughout the basin in the study period, the most remarkable deforestation was made in the southeast basin (Fig. 3). The conversion from forest to urban area mainly occurred near six concentrated areas, namely Datong, Shuozhou, Shanyin, Huairen, Yingxian, and Hunyuan. It is noted that the agriculture round Datong city were reforested from 1977 to 1990, but then deforested for urban use from 1990 to 2006. Only forest farms in the north basin and on the south of the Shuozhou city kept unchanged in the study period. The gain of forest area mainly came from reforestation of agriculture and barren lands on the hills in the central basin and on the proluvial fans in the north basin. Barren land Barren land lost 92.4% (88,460 ha) of its 1976 area to other classes and gained 14.2% (13,589 ha) from other classes, resulting in a net 74,871 ha decrease (11.8% of the total study area) in barren land area during the study period (Table 4). Of the 88,460 ha area that barren land lost from 1977 to 2006, 74.7% (66,102 ha) was to agriculture and 18.4% (16,270 ha) to forest. Of the 13,589 ha area that barren land gained in the period, 72.5% (9,853 ha) was from agriculture and 19.7% (2,679 ha) from forest. Although the total barren land area gained from agriculture was much less than that lost to agriculture, transformation between the two classes was frequent (Table 5a–c), which is possibly because, as mentioned above, the barren land on the mouth of the proluvial fans was cultivated once in several years interval for lack of nutrient and water (Fig. 3). According to Table 5d, 7,258 ha barren land of 1977 kept unchanged until 2006. It is found in Fig. 3 that the unchanged barren land was mainly located at the lowlands and dominated by salt marsh, where groundwater is less than 3 m beneath the surface (Guo and Wang 2004). Wetland

Forest According to Table 4, forest areas decreased continuously from 1977 to 2006. In quantitative terms, 10,899 ha, 10,651 ha, and 13,472 ha net decreases were detected in forest areas from 1977 to 1990, from 1990 to 2000, and from 2000 to 2006, respectively. Only around 27.2% (45,519 ha) of forest in 1976 remained unchanged until 2006. The loss of forest area was mainly caused by deforestation for urban and agricultural use. According to Table 5d, 10,099 ha and 104,723 ha forest areas were

In the similar way, 16,688 ha consistent decrease was observed in wetland areas from 1977 to 2006, which represents that wetland lost 67.1% of its 1977 area by 2006 (Table 4). Based on Table 5d, only 5.3% (1,319 ha) wetland kept unchanged from 1997 to 2006. Wetland lost 63.9% (15,876 ha) to agriculture and 21.4% (5,317 ha) to forest in the same period. Meanwhile, 2,073, 2,195, and 2,067 ha wetland areas were gained from water, forest, and agriculture, respectively. It is important to note that land use change between water and wetland might be affected

123

1834

Environ Geol (2009) 57:1825–1837

Table 5 Matrices of land use changes from 1977 to 2006 (Unit: ha) (a) 1977–1990 1977 Water

1990 Total Urban

Forest

Agriculture

Wetland

Barren land

1990 Water

3,866

189

1,427

3,227

1,310

1,122

11,141

Urban

1,030

8,832

4,135

5,568

396

1,852

21,814

Forest

2,515

1,401

68,359

69,806

5,068

9,457

156,606

Agriculture

6,676

2,295

85,388

193,242

12,405

51,970

351,976

Wetland

3,836

152

2,722

5,409

3,684

1,307

17,110

Barren land

2,016

825

5,474

35,256

1,994

30,010

75,574

19,938

13,694

167,505

312,507

24,858

95,717

634,220

1977 Total (b) 1990–2000

1990 Water

2000 Total Urban

Forest

Agriculture

Wetland

Barren land

2000 Water

3,366

1,628

3,027

2,212

1,450

1,138

12,820

Urban

1,143

15,969

8,998

9,485

1,592

3,820

41,008

Forest

1,200

1,144

63,132

70,488

4,379

5,613

145,955

Agriculture

3,643

2,656

75,610

249,986

7,522

48,796

388,213

Wetland Barren land

1,109 680

103 313

3,778 2,061

5,669 14,137

1,771 395

411 15,796

12,842 33,382

11,141

21,814

156,606

351,976

17,110

75,574

634,220

1990 Total (c) 2000–2006

2000 Water

2006 Total Urban

Forest

Agriculture

Wetland

Barren land

2006 Water

4,779

1,351

2,471

2,477

745

318

12,141

Urban

3,168

23,796

6,087

7,124

731

977

41,883

Forest

1,828

4,082

63,506

57,916

2,200

2,950

132,483

Agriculture

2,053

11,014

70,341

304,451

6,871

23,968

418,697

Wetland

885

533

2,343

2,173

2,192

44

8,170

Barren land

108

233

1,206

14,071

103

5,125

20,846

12,820

41,008

145,955

388,213

12,842

33,382

634,220

2000 Total (d) 1977–2006

1977

2006 Total

Water

Urban

Forest

Agriculture

Wetland

Barren land

Water Urban

2,736 1,453

223 8,032

2,289 10,099

4,525 17,107

824 984

1,544 4,208

12,141 41,883

Forest

4,129

1,045

45,519

60,204

5,317

16,270

132,483

Agriculture

9,115

4,129

104,723

218,752

15,876

66,102

418,697

Wetland

2,073

181

2,195

2,067

1,319

336

8,170

Barren land

433

85

2,679

9,853

539

7,258

20,846

19,938

13,694

167,505

312,507

24,858

95,717

634,220

2006

1977 Total

123

Environ Geol (2009) 57:1825–1837

greatly by precipitation and thus presented instability. Wetland area lost to water is less than that gained from water during 1977–1990 and 2000–2006, but reversely during 1990–2000. Figure 3 shows that wetland loss mainly occurred along the rivers. Most of the wetland there, either existing before 1977 or gaining from water during 1977–1990, had been converted to cropland or river bank protection forest by 2006. Water Water had an area of 19,938 ha in 1977, 11,141 ha in 1990, 12,820 ha in 2000, and 12,141 ha in 2006. Though a general trend of decreasing areas was observed from 1977 to 2006, the change of water area varied greatly in different periods. Water lost 44.1% (8,797 ha) of its 1977 area by 1990, with the narrowing of rivers, reduction of reservoirs water area, and disappearance of many ponds and little lakes throughout the whole basin (Fig. 3). It was mainly caused by exploitation of surface water for domestic, irrigation, and industrial uses (Han et al. 2003). Decreases of recharge from groundwater to surface water due to over exploitation of groundwater might be another major reason (Han et al. 2003). Table 5a indicates that water areas in the period mainly lost to agriculture and wetland. It is presumed that water body was covered by hydrophytes and converted to wetland when it became shallow, and brought under cultivation and converted to agriculture when it dried completely. This is proved by the facts that the residual water body in 1990 was surrounded from inner to outer by the agricultural land and wetland converted from water of 1977 in turn (Fig. 3). Groundwater and surface water were overexploited further from 1990 to 2000 (Han et al. 2003), and as a result water body in the central basin was affected greatly. By 2000, Huangshui River had dried completely, lower reach of Hun River had also dried, middle reach of Sanggan River had narrowed further, and the water area of most major reservoirs except Cetian Reservoir had shrunk more or even disappeared completely. However, 1,679 ha net increase of water area in the basin was observed from 1990 to 2000 according to Table 4. The increase of water areas in the period could be explained by the following two main reasons. Firstly, the precipitation between 1991 and 2000 is more than that in the other two periods. Monthly precipitation records from 45 weather stations in Datong basin were used to discuss the relationship between precipitation and water areas. The accumulative precipitation from January to July was used for each year, because the four periods of Landsat images employed in this study were acquired from May to July. The result shows that the average value from 1991 to 2000 is about 370 mm, more than that from 1981 to 1990 (about 320 mm) and from

1835

2001 to 2003 (about 230 mm) (Fig. 5). The increased precipitation supplied more inflows to the rivers running through the basin, causing the expansion of water area in the upper reaches of these rivers (Fig. 3). Secondly, a lot of man-made lowlands came into being for soil excavation in company with rapid urban and road construction from 1990 to 2000. According to monthly precipitation records in Datong basin, a high proportion (37.8%) of the rain of 2000 (661 mm) fell in July (250 mm), when the Landsat images of 2000 were acquired. As a result, the lowlands might be ephemerally filled with water and thus classified as water area in land use map, which was presumed as another major contributor to the increase of water area. It could be found in Fig. 3 that the gains of water in 2000 mainly occurred near newly-built highways and urban areas. Water turned into the trend of decreasing its area again from 2000 to 2006 for overexploitation of water resources and less precipitation (Fig. 5). In the period, water kept 4,779 ha area unchanged, lost 8,042 ha area to other classes, and gained 7,362 ha area from other classes, which resulted in a net 679 ha decrease in water area. According to Fig. 3, though several reservoirs were newly built or expanded in the northwest basin, resulting in the gain of water area from agriculture and forest, the loss of water area dominated from 2000 to 2006. Most main tributaries of Sanggan River, including Huangshui River, Hun River, Yu River, Libazhuang River, Dayu River, and Qili River, and the most reaches of Sanggan River itself, had dried by 2006. At the same time, Shijiazhai Reservoir, one of the main reservoirs in the basin, had nearly dried (Fig. 3). In addition, the water areas in man-made lowlands found in land use map of 2000 were usually ephemeral and depended on precipitation greatly. By 2006, most watered lowlands near highways had dried for decreased precipitation and had

Fig. 5 Accumulative precipitation from January to July in different years from 1981 to 2003. The monthly precipitation is gotten by averaging the observed values from 45 weather stations in Datong basin. Records for 1977–1980 and 2004–2006 are not available

123

1836

Environ Geol (2009) 57:1825–1837

been converted to agriculture and forest, while those near urban areas had been converted to urban for further expansion of urban.

NSFC-40702042), Ministry of Education of PR China (111 project) and Research Foundation for Outstanding Young Teachers, China University of Geosciences (Wuhan) (CUGQNL0714).

References Conclusions Land use changes in Datong basin were determined using Landsat data for the period of 1977–2006. The results demonstrate the potential of multitemporal Landsat data to provide an accurate, economical means to map and analyze changes in land use over time that can be used as inputs to land management and policy decisions. General patterns and trends of land use change in Datong basin were drawn from this study: 1.

2.

3.

4.

5.

6.

Urban area increased 213% from 1977 (13,694 ha) to 2006 (41,883 ha), mainly caused by the conversion of agriculture and forest area to urban area. The expansion of urban area could be explained by the rapid increase of urban population and high-speed economic development. Agriculture area increased 34.0% (106,190 ha) from 1977 to 2006, mainly coming from the conversion of barren land and forest to agriculture. The increasing trend of agriculture could be attributed to the land reclamation that resulted from rapid increase of rural population and improvement of irrigation capacity. Forest area decreased continuously from 1977 to 2006. Around 27.2% (45,519 ha) of forest in 1976 remained unchanged until 2006. The loss of forest area was mainly caused by deforestation for urban and agriculture use. Barren land area decreased by 74,871 ha from 1977 to 2006, which is mainly because the barren land was cultivated for agriculture use. Although the total barren land area gained from agriculture was much less than that lost to agriculture in the study period, transformations between the two classes were frequent. Wetland lost 67.1% (6,688 ha) of its 1977 area by 2006, mainly due to conversion of wetland to agriculture and forest. Only 5.3% (1,319 ha) wetland kept unchanged from 1997 to 2006. Land use change between water and wetland might be affected greatly by precipitation and thus presented instability. Though the change of water area varied greatly in different periods, the trend of decreasing areas was observed from 1977 to 2006. It might be caused by exploitation of surface water and decrease of recharge from groundwater to surface water due to overexploitation of groundwater.

Acknowledgments This work was financially supported by National Natural Science Foundation of China (NSFC-40425001 and

123

Aboel Ghar M, Shalaby A, Tateishi R (2004) Agricultural land monitoring in the Egyptian Nile Delta using Landsat data. Int J Environ Stud 61:651–657 Alberti M, Weeks R, Coe S (2004) Urban land cover change analysis in Central Puget Sound. Photogr Eng Remote Sens 70:1043– 1052 Anderson JR, Hardy EE, Roach JT, Witmer RE (1976) A land use and land cover classification system for use with remote sensor data. US Geological Survey Professional Paper No. 964. USGS, Washington, DC Chen ZJ, Li S (2007) Advance on Chinese land use and land coverage change research. Resour Dev Market 23:898–902 [in Chinese] Congalton RG (1991) A review of assessing the accuracy of classifications of remotely sensed data. Remote Sens Environ 37:35–46 Congalton RG, Green K (1999) Assessing the accuracy of remotely sensed data: principles and practices. Lewis Publishers, Boca Raton Coppin P, Jonckheere I, Nackaerts K, Muys B, Lambin E (2004) Digital change detection methods in ecosystem monitoring: a review. Int J Remote Sens 25:1565–1596 Crist EP, Lauren R, Cicone RC (1986) Vegetation and soils information contained in transformed Thematic Mapper data. In: Proceedings of IGARSS 86 symposium. European Space Agency, Paris, pp 1465–1470 ESCAP (1997) Guidelines and manual on land-use planning and practices in watershed management and disaster reduction. ESCAP, New York ERDAS (1999) ERDAS field guide, 5th edn. ERDAS Inc., Atlanta Foody GM (2002) Status of land cover classification accuracy assessment. Remote Sens Environ 80:185–201 Gautam AP, Webb EL, Shivakoti GP, Zoebisch MA (2003) Land use dynamics and landscape change pattern in a mountain watershed in Nepal. Agric Ecosyst Environ 99:83–96 Giri C, Zhu Z, Reed B (2005) A comparative analysis of the Global Land Cover 2000 and MODIS land cover data sets. Remote Sens Environ 94:123–132 Goetz SJ, Varlyguin D, Smith AJ, Wright RK, Prince SD, Mazzacato ME, Tringe J, Jantz C, Melchoir B (2004) Application of multitemporal Landsat data to map and monitor land cover and land use change in the Chesapeake Bay watershed. In: Smits PC, Bruzzone L (eds) Analysis of multi-temporal remote sensing images. World Scientific Publishers, Singapore, pp 223–232 Guo ZX (1992) Water and salt dynamic of alkalize solonchaks and its improving measures in basin of Datong. J Shanxi Agric Sci 27:60–63 [in Chinese] Guo QX, Chen HW (2006) Temporal characteristics of rainfallerosivity of Datong city. Sci Soil Water Conserv 4:25–29 [in Chinese] Guo HM, Wang YX (2004) Specific vulnerability assessment using the MLPI model in Datong city, Shanxi province, China. Environ Geol 45:401–407 Han GC, Gao WS, Kang YM (2003) Crisis of water resources and countermeasures in Tongsuo district. Groundwater 25:115–118 [in Chinese] Hathout S (2002) The use of GIS for monitoring and predicting urban growth in east and west St Paul, Winnipeg, Manitoba, Canada. J Environ Manage 66:229–238

Environ Geol (2009) 57:1825–1837 Houghton RA (1994) The worldwide extent of land-use change. Bioscience 44:305–313 Huete AR (1998) A soil-adjusted vegetation index (SAVI). Remote Sens Environ 25:295–309 Huete A, Liu HQ (1994) An error and sensitivity analysis of the atmospheric- and soil-correcting variants of the NDVI for the MODIS-EOS. IEEE Trans Geosci Remote Sens 32:897–905 Huete AR, Hua G, Qi J, Chehbouni A, van Leeuwen WJD (1992) Normalization of multidirectional red and NIR reflectances with the SAVI. Remote Sens Environ 41:143–154 Jensen JR (2005) Introductory digital image processing: a remote sensing perspective (third edition). Prentice-Hall, Englewood Cliffs Jensen JR, Cowen DC (1999) Remote sensing of urban/suburban infrastructure and socio-economic attributes. Photogr Eng Remote Sens 65:611–622 Karnieli A, Kaufman YJ, Remer L, Wald A (2001) AFRI—aerosol free vegetation index. Remote Sens Environ 77:10–21 Kauth RJ, Lambeck PF, Richardson W, Thomas GS, Pentland AP (1979) Feature extraction applied to agricultural crops as seen by LANDSAT. In: Proceedings of LACIE Symposium of the technical session. National Aeronautics and Space Administration, Houston, pp 705–721 Lambin EF, Baulies X, Bockstael N, Fischer G, Krug T, Leemans R, Moran EF, Rindfuss SYRR, Sole DL, Turner BL II, Vogel C (1999) Land-use and land-cover change (LUCC) implementation strategy. IGBP Report No. 48 and HDP Report No. 10. IGBP, Stockholm Landis J, Koch GG (1977) The measurement of observer agreement for categorical data. Biometrics 33:159–174 Liu JY, Liu ML, Zhuang DF, Zhang ZX, Deng XZ (2003) Study on spatial pattern of land-use change in China during 1995–2000. Sci China Ser D 46:373–384 Lunetta RS, Balogh M (1999) Application of multi-temporal Landsat 5 TM imagery for wetland identification. Photogr Eng Remote Sens 65:1303–1310 Lunetta RS, Elvidge CD (eds) (1998) Remote sensing change detection: environmental monitoring methods and applications. Taylor & Francis, London Lunetta RS, Ediriwickrema J, Johnson DM, Lyon JG, McKerrow A (2002) Impacts of vegetation dynamics on the identification of land-cover change in a biologically complex community in North Carolina, USA. Remote Sens Environ 82:258–270 Ma YJ, Su ZZ (1998) A study on the land desertification process and sustainable development of agriculture in the northwest of Shanxi province. Arid Land Geogr 21:59–65 [in Chinese] Meyer WB, Turner BLII (1994) Change in land use and land cover: a global perspective. Cambridge University Press, Cambridge Naiman RJ, Bisson PA, Turner MG (1997) Approaches to management at the watershed scale. In: Kohm KA, Franklin JF (eds) Creating a forestry for the 21st century: the science of ecosystem management. Island Press, Washington, DC, pp 239–253 Oettera DR, Cohenb WB, Berterretchea M, Maierspergera TK, Kennedy RE (2000) Land cover mapping in an agricultural setting using multiseasonal Thematic Mapper data. Remote Sens Environ 76:139–155 Prakash A, Gupta RP (1998) Land-use mapping and change detection in a coal mining area: a case study in the Jharia coalfield, India. Int J Remote Sens 19:391–410

1837 Price KP, Guo X, Stiles JM (2002) Optimal Landsat TM band combinations and vegetation indices for discrimination of six grassland types in eastern Kansas. Int J Remote Sens 23:5031– 5042 Ridd MK, Liu J (1998) A comparison of four algorithms for change detection in an urban environment. Remote Sens Environ 63:95– 100 Rogan J, Chen DM (2004) Remote sensing technology for mapping and monitoring land-cover and land-use change. Prog Plan 61:301–325 Rosenfield GH, Fitzpatrick K (1986) A coefficient of agreement as a measure of thematic classification accuracy. Photogr Eng Remote Sens 52:223–227 Shalaby A, Tateishi R (2007) Remote sensing and GIS for mapping and monitoring land cover and land-use changes in the northwestern coastal zone of Egypt. Appl Geogr 27:28–41 Singh A (1989) Digital change detection techniques using remotely sensed data. Int J Remote Sens 10:989–1003 Squires GD (2002) Urban sprawl and the uneven development of metropolitan America. In: Squires GD (ed) Urban sprawl: causes, consequences and policy responses. Urban Institute Press, Washington, DC, pp 1–22 Stow DA (1999) Reducing mis-registration effects for pixel-level analysis of land-cover change. Int J Remote Sens 20:477–483 Tan KL (2007) Remote sensing theories and practices in ecologically fragile semi-arid zone China. Science Press, Beijing [in Chinese] Turner BLII, Meyer WB, Skole DL (1994) Global land use/land cover change: towards an integrated program of study. Ambio 23:91–95 Turner BL II, Skole D, Sanderson S, Fischer G, Fresco L, Leemans R (1995) Land-use and land-cover change, in science/research plan. IGBP Report No. 35 and HDP Report No. 7. IGBP, Stockholm Wolter PT, Mladenoff DJ, Host GE, Crow TR (1995) Improved forest classification in the northern Lake States using multi-temporal Landsat imagery. Photogr Eng Remote Sens 61:1129–1143 Yang GD, Jia CQ (2001) Study on present situation, problem and sustainable countermeasures of land eco-environment in Datong city. Agric Res Arid Areas 19:97–101 [in Chinese] Yang X (2002) Satellite monitoring of urban spatial growth in the Atlanta metropolitan area. Photogr Eng Remote Sens 68:725– 734 Yuan D, Elvidge CD, Lunetta RS (1998) Survey of multispectral methods for land cover change analysis. In: Lunetta RS, Elvidge C (eds) Remote sensing change detection: environmental monitoring methods and applications. Taylor & Francis, London, pp 21–39 Yuan F, Sawaya KE, Loeffelholz BC, Bauer ME (2005) Land cover classification and change analysis of the Twin Cities (Minnesota) metropolitan area by multitemporal Landsat remote sensing. Remote Sens Environ 98:317–328 Zhang JX, Zhang YH (2007) Remote sensing research issues of the National Land Use Change Program of China. ISPRS J Photogr Remote Sens 62:461–472 Zhang YQ (2006) Present situation of soil degradation and study of trend change in Datong city. Environ Sci Manage 31:36–38 [in Chinese] Zhao HL, Zhao XY, Zhang TH, Zhou RL (2002) Boundary line on agro-pasture zigzag zone in north China and its problems on ecoenvironment. Adv Earth Sci 17:739–747 [in Chinese]

123