Vegetation response to 30 years hydropower cascade exploitation in upper stream of Yellow River

Commun Nonlinear Sci Numer Simulat 15 (2010) 1928–1941

Contents lists available at ScienceDirect

Commun Nonlinear Sci Numer Simulat journal homepage: www.elsevier.com/locate/cnsns

Vegetation response to 30 years hydropower cascade exploitation in upper stream of Yellow River W. Ouyang a,b,*, F.H. Hao a, C. Zhao a, C. Lin c a

School of Environment, State Key Laboratory of Water Environment Simulation, Beijing Normal University, Beijing 100875, China International Institute for Geo-Information Science and Earth Observation (ITC), Hengelosestraat 99, P.O. Box 6, Enschede 7500 AA, The Netherlands c Key Laboratory of Agricultural Bio-Environmental Engineering Ministry of Agriculture, China Agricultural University, Beijing 100083, China b

a r t i c l e

i n f o

Article history: Received 3 May 2009 Received in revised form 21 July 2009 Accepted 22 July 2009 Available online 28 July 2009 Keywords: Vegetation response survey Grassland degradation Hydropower cascade exploitation Yellow River basin

a b s t r a c t The accumulated response of vegetation successive dam constructions and operations is an important concern, but the systematic assessment of impacts induced by cascade hydropower exploitation over long periods are seriously lacking. Using remote sensing data, the variations in grassland, the principal land cover in the upper catchment of the Yellow River, were investigated for eight dams constructed during the period 1977–2006. Two different scales—watershed scale and on-site area—were used to compare the changes in grassland and water area. Correlation coefficients from regression analyses showed that grassland area had more significant interactions with hydropower exploitation indicators in on-site scale than in watershed scale. The hydropower exploitation indicators had a more complex correlation with water area in watershed scale than in on-site scale. Consequently, observations of grassland area responses to successive hydropower exploitations were focused on the on-site region. The Normalized Difference Vegetation Index (NDVI) and the standardized NDVI, which can be used to analyze inter-annual climatic differences, were applied to identify the most heavily influenced vegetation zones. For different hydrological and micro-climatic conditions, the vegetation zones around reservoirs and along the main stream of Yellow River were analyzed, respectively. Two NDVI spatial principles at varied distances from the water demonstrated that the vegetation NDVI was recovering from 1994 to 2006. For distance of less than 10 km from water, the vegetation around reservoirs was better as the higher NDVI in 2006 than in 1977. The inter-annual NDVI comparison demonstrated that the critically affected vegetation zone was concentrated at distances of 0.1–0.4 and 1–6 km from the water. In on-site region, the grassland was further analyzed with elevation and aspect information, which indicated that grassland in sunny aspects was much disturbed. Detailed information about grassland response with water distance and the degradation characteristics provide the comprehensive assessment by cascade hydropower exploitation. Ó 2009 Published by Elsevier B.V.

1. Introduction More than 40,000 dams have been constructed on the world’s rivers, nearly 29,000 of which are located in China. These dams can provide extensive economic benefits, but they also disturb fluvial processes in the rivers and the terrestrial status of the associated watersheds. In many of the world’s rivers eco-environmental aggradations have taken place [1,2]. * Corresponding author. Address: School of Environment, State Key Laboratory of Water Environment Simulation, Beijing Normal University, Beijing 100875, China. E-mail addresses: [email protected] (W. Ouyang), [email protected] (C. Lin). 1007-5704/$ - see front matter Ó 2009 Published by Elsevier B.V. doi:10.1016/j.cnsns.2009.07.021

W. Ouyang et al. / Commun Nonlinear Sci Numer Simulat 15 (2010) 1928–1941

1929

The Yellow River in China has experienced a dramatically decreasing trend in water resource discharge since the construction and operation of large reservoirs. Most of these dams and reservoirs have been built in the river’s upstream catchment area during the last several decades [3]. Not only is the land cover in the upper stream portion of the Yellow River affected by the construction of eight successive dams, but also recovery is difficult. This regional vegetation is extremely fragile as the result of typical continental season climate. Understanding the long-term variation in the vegetation recovery process, and its principles, resulting from hydropower cascade exploitation (HCE) can help explain the spatial extent and the degree of influence from human activities. Equally important, it can help in predicting the impact of future HCE plans and assist in preventing environmental degradation from such projects. Dams are constructed for diverse purposes, including seasonal flood control, generation of hydroelectric power, and improvement in the supply of water resources. Although dams can provide various economic benefits, they can also be the source of negative impacts on the natural environment [4,5]. Hydropower dam constructions affect the status of the regional land cover, which was the result of flooding and alteration of river. After dams are constructed and under operation, irrigation conditions improve and water table depths increase, leading to more convenient access to groundwater. As a result, farmland areas will increase and cropping patterns will shift [6]. In most cases, communities in close proximity to large dams are displaced or affected by explorations of hydropower resources [7]. As a consequence of hydrological and socio-economic developments associated with dam construction, watershed land cover changes occur. However, during the environmental impact assessment and management process, the principal problem is to determine an appropriate definition of the assessment region. In this paper, the affected land cover characteristics are examined over two regions; a watershed region and the other by an on-site region. Thus, land cover characteristics are analyzed from the perspective of two different scales. Using a standard hydrological concept, the watershed region delineates the geographic area from which the run-off water flows into the portion of the Yellow River under study. The on-site region of the HCE delineates a smaller region which is defined in a more complicated manner. The first step is to construct a buffer zone of 25 km along each side of the riverbank, a designation that covers almost all construction fields associated with real hydropower engineering. In the second step, the main portion of the Yellow River watershed is identified. For example, when a mountain intrudes into the 25 km buffer zone in such a way that the water on the side of the mountain away from the river does not drain directly into the river and construction activities do not occur there, the mountainous portion is excluded from consideration. Finally, the on-site region of the HCE is defined as the intersection of the areas determined in steps one and two. Whether or not hydropower explorations should be continued has become an increasingly important topic and is widely debated in developing countries, as well as developed countries [8]. For the most part, these discussions have been about the impact identifications and assessments for single dam constructions. Pamo and Tchamba studied land changes resulting from dams and its impact on animals [9]. Gordon and Ross studied dam operation effects on land use and riparian vegetation [10]. However, there is a widespread lack of detailed data to support systematic assessments of HCE induced impacts. With the advance of remote sensing (RS) and Geographic Information System (GIS) technologies, the study of land cover variation, and its principles, resulting from HCE has become feasible and reliable. Many studies have demonstrated that remotely sensed data can provide both actual and spatially distributed information for land cover variation monitoring and analysis, especially on a watershed scale over long time periods [11,12]. Such analyses are difficult to monitor by conventional techniques [13]. The Landsat MSS data has been widely applied to multi-temporal land cover analyses because it has been available for quite some time. The later launched Landsat TM can also provide high spatial resolution and frequent time-series data for a wide variety of environmental applications, ranging from regional land use analyses to vegetation simulation [14]. With the Normalized Difference Vegetation Index (NDVI), the temporal–spatial variation principle of vegetation status can be achieved. Fabio successfully used the NOAA-AVHRR, Landsat TM, and ETM+ images to produce a long-term NDVI data series for coniferous and broadleaved forestland assessment in Tuscany (Central Italy) [15]. The GIS tools treat the information extracted from remotely sensed data and the analytical results can be used as the primary base for regional environmental impact identification and management [16,17]. Land cover variation is one of the most obvious impacts resulting from dam construction. Temporal changes in land cover have been studied with the aid of satellite images. However, no studies are available that identify the range of land cover affected by HCE over a long time period. This article presents the temporal–spatial variation principle of land cover in the Longliu Section in the upper stream of the Yellow River during the period 1977–2006. In this period eight reservoirs were built along the river; these constructions were the most critical cause of regional land cover transformation. The objectives of this paper are summarized as follows: (1) Define the area directly influenced from dam construction and operation by comparing the correlation of HCE indicators with grassland and water areas in the on-site and watershed regions. (2) Determine grassland (the dominant land cover) degradation characteristics within the on-site region at different elevations and their relationship with HCE. (3) Demonstrate the differences in spatial principles of on-site vegetation NDVI variation by the data in last 30 years. These observations are based on the water distance along the riverbank and around reservoirs, thereby identifying the detailed impacted zone from HCE.

1930

W. Ouyang et al. / Commun Nonlinear Sci Numer Simulat 15 (2010) 1928–1941

2. Methodology 2.1. Study area The study area is a part of the upper stream portion of the Yellow River watershed. Its lower control point is the outlet of the Liujiaxia Reservoir and the upper control point is the end of the Longyangxia Reservoir. The study area between these two points is named the Longliu Section (Fig. 1). This area lies on the conjunction of the Qinghai–Tibet Plateau and the Loess Plateau. This is the most important water holding area for the entire Yellow River watershed and about 49.2% of the water flow comes from here. The typical continental climate in this area is cold and dry, and obviously varies with the season. The average annual temperature is 2.3 °C [18]. The temperature and precipitation observed the stable status in recent three decades. The watershed land cover consists largely of high-cold meadows, steppes, and forestlands. Grassland occupies more than half of the total area and appears throughout the region. The deciduous forest is located mainly in mountainous areas at higher altitudes [19]. The steppes are situated in most parts of the watershed with desert steppes appearing in some places in the western part of the research area. 2.2. Hydropower cascade exploitation retrospection In order to control regular floods and generate hydropower resources, 22 dams have been or are under construction on the entire Yellow River watershed. With the exception of the Sanmenxia and Xiaolangdi stations, which are located in the middle reach, all the remaining ones are spread out in the upper reach [20,21]. In the upper stream portion of the Yellow River, located in the Gansu and Qinghai Provinces, 14 dams are planned along the river (Fig. 1). Among the planned hydro-stations, eight have been constructed in the Longliu Section, which is the most intensely exploited area and has the longest construction history (Fig. 2). The Liujiaxia Reservoir was the first to be constructed in this section and is located at the outlet of the study watershed. After that, the Longyangxia Reservoir, which lies at the inlet of the section, came on line with construction beginning in 1978. Between these two dams, another six other dams came into service. Except for the Liujiaxia Reservoir, the other five dams are run-off hydropower dams and cannot contain water flow. With all eight dams completed and in operation beginning in 2006, the accumulated reservoir capacities are 32.4559 billion cubic meters and the total hydropower generators are capacity of 6864 MW, can provide 25.856 billion Kwh of electricity annually. 2.3. Land cover classification process Based on the HCE characteristics and availability of remote sensing data covering the last three decades, the years 1977, 1996, 2000, and 2006 were chosen for our study purposes with the goal of identifying the land cover transformation principle. The Chinese Academy of Sciences (CAS) has developed national land cover databases in 1996 and 2000, which are based on Landsat series images [22]. In this study, the land cover data for 1996 and 2000 were extracted from national databases. The land cover data for 1977 and 2006 were developed from Landsat MSS and TM data. Details about satellite sensor, track and frame, and date are listed in Table 1. From the available images, those with no cloud cover were selected. The selection of images and same interpretation procedure with national database aimed to minimize the system error.

Fig. 1. Study area and cascade hydropower station distribution.

1931

W. Ouyang et al. / Commun Nonlinear Sci Numer Simulat 15 (2010) 1928–1941

2500

200 178

180

155 2000

147

139

140 Dam height / m

Hydropower capacities

1500

120 100

1000

80 60

43

42.5

45

51.2 500

40

Hydropower capacities /M W

160

Dam height

20 0

0 1968/10 1986/10 1996/12 2003/02 2004/08 2005/04 2006/08 2006/09 LiujiaxiaLongyangxia Lijiaxia Nina GongboxiaZhiganlaka Kanyang Suzhi Fig. 2. Hydropower cascade exploitation characteristics and construed time.

Table 1 Description of Landsat images applied in watershed land cover classification. Path/row

MSS

TM

131/35 131/36 132/35 132/36 133/35

04 14 15 15 22

05 05 10 10 20

Jan. 1974 Jul. 1977 Jul. 1977 Jul. 1977 Feb. 1977

Aug. 2006 Aug. 2006 Sep. 2005 Sep. 2005 Sep. 2006

The 10 MSS and TM satellite images were geo-referenced to the Universal Mercator (UTM) projection system by the nearest-neighbor re-sampling method [23]. The image data was subsequently re-sampled to the same scale as those in the available databases for 1996 and 2000. In order to improve classification accuracy, both the unsupervised and supervised classification methods were employed. First, we employed the unsupervised spectral clustering of the satellite image, and then we moved to the supervised classification referring to ground truth control points, the digital elevation model, and land cover information for 1996 and 2000. As Liu has cited the procedure employed in the national database, the images were classified following the second-level classes and then were integrated into the first level [22]. All second-level classes were interpreted after the supervised classification was employed; these classes were then combined and regrouped into six first level classes. Positional accuracy refers to the accuracy of a geometrically rectified image; an error less than the grid cell size can be ignored [24]. The thematic accuracy refers to the interpretative characteristics of spatial data [25]. No reference data were available for the interpretation of 1977 images; therefore no accuracy estimation was carried out for that year. The accuracy assessment of TM based images in 2006 was conducted by the site-specific method in which the land cover map was compared to ground control points [26]. In Table 2, the error matrix and the accuracy assessment for land cover classification in 2006 are presented. The 88 control points were sampled randomly across the studied watershed and were collected by the Global Positioning System (GPS) in August 2007. The overall accuracy of the classification is approximately 86.34%, which indicated the interpreted data was reliable. 2.4. Vegetation NDVI data procedure We applied the Normalized Difference Vegetation Index (NDVI) to analyze on-site vegetation biomass characteristics over three decades. Because different land covers reflect different NIR and Red bands, water areas, bare land, and vegetation can be effectively distinguished by NDVI information [27]. The on-site study area involves three frames of Landsat series images; Table 2 Classification accuracy evaluation of land cover in 2006. Class name

Reference totals

Classified totals

Number correct

Producers accuracy (%)

Users accuracy (%)

Farmland Forestry Grassland Water area Construction land Bare land

10 12 32 15 8 11

13 13 36 13 11 13

9 11 30 11 7 9

90.00 91.67 93.75 73.33 87.50 81.82

69.23 84.62 83.33 84.62 63.64 69.23

1932

W. Ouyang et al. / Commun Nonlinear Sci Numer Simulat 15 (2010) 1928–1941

each frame occurring on different dates (Table 3). With the aid of temporal NDVI regression equation, watershed NDVI vegetation data was extracted from MODIS images and the NDVI relationship on different dates was calculated using Arcgis 9.2 [28]. The images for the 4 years were related to three kinds of sensor systems and, as a result, systematic differences occurred in NDVI comparisons. Fortunately, with the aid of the following equations, the NDVI differences from different Landsat sensor systems could be summarized and standardized [29]:

NDVITM ¼ 1:052NDVIMSS  0:021

ð1Þ

NDVITM ¼ 0:979NDVIETMþ þ 0:002

ð2Þ

With these equations, the NDVI data from MSS and EMT+ was transferred into TM sensor format. Following this procedure, the vegetation NDVI data from multiple satellite systems on different dates over three decades was integrated into a comparable format. This integration provided the base for the temporal–spatial vegetation analysis for the years 1977, 1994, 2000, and 2006. 3. Results 3.1. Watershed land cover variation from 1977 to 2006 Maps of the land cover distribution for the years 1977, 1996, 2000, and 2006 are shown in Fig. 3a–d. The natural vegetation, consisting mainly of forestland and grassland are the dominant land cover categories in the study area. In 1996, forestland and grassland occupied 15.49% and 59.57% of the study area, respectively. Farmland, which accounted for another 12.82% of land cover in 2006, was most commonly found in the northeastern portion of the study area. Most of the forestland is situated in the central and southern parts of the watershed where the mountains are located. A high proportion of the bare land is located in the western part. By evaluating land cover distribution maps in the four observed years, the transformational processes occurring during the period 1977–2006 were identified; the statistical results are presented in Table 4. The most intensive reduction occurred in the grassland category. In 1977, grassland occupied 63.07% of the study area; by 2006 it was reduced to 59.57%. This change represents a loss of 120,655 ha. As eight reservoirs came into service, the water area increased continually, growing from 53,919 ha in 1977 to 78,124 ha in 2006. The forestland area decreased from 527,936 ha in 1977 to 489,799 ha in 2000. However, the forested area climbed dramatically in the last 7 years of the study period, rising to 530,756 ha, a figure that is even greater than corresponding one for 1977. The reduction in grassland coincided with the increase in the other five types of land cover. A remarkable increase occurred in the bare land category during the period 1977–1996 when its area changed from 271,427 to 332,419 ha. However, over the time span from 1996 to 2006, the area of bare land decreased to 316,043 ha. Farmland had a net decrease in area from 1977 to 2000, but the decline occurred during the period 1977–1996. A recovery of farmland area that began in 1996 was sustained through 2006, resulting in an overall increase of 35,076 ha from 1977 to 2006. Finally, a dramatic climb in land area devoted to construction occurred between 1977 and 1996, rising from 2464 to 21,588 ha. However, the area for this category continually dropped thereafter, declining to 15,349 ha in the last decade of the study period. 3.2. On-site land cover variation from 1977 to 2006 In this study area, the HCE is the priority human activity, which induced local urbanization, farming, and deforestation. The boundary for the on-site region defines the area directly influenced by the hydropower station construction and operation. Based on the land cover distribution shown in Fig. 3, the six land cover categories were delineated and their long-term variations summarized in Table 5. Once again, the dominant land cover category is grassland, which covered 59.33% of the study area in 2006, a figure that is similar to the one for the watershed scale. Because most of the forestland is situated in mountainous parts of the study site, its area in the on-site region occupied just 6.5% of the total in 2006. However, farmland occupied 20.01% of the total in 2006, a figure that is much higher than 12.82% amount in the watershed scale. We concluded that most of the farmland was located near the reservoirs and the main stream because of the availability of water resources. A similar distribution principle was observed after comparing the construction land and bare land percentages. The noted vegetation differences are consistent with observation that land cover in the on-site area has a closer relationship with human activities. The land cover transformational principles for the on-site region are similar to those for the watershed region. The grassland area had the greatest decrease of all the land cover categories, declining from 63.82% in 1977 to 59.33% in 2006. The on-

Table 3 Description of Landsat images applied in on-site vegetation NDVI analysis. Path/row

MSS

TM

ETM+

TM

131/35 132/35 133/35

04 Jan. 1974 15 Jul. 1977 22 Feb. 1977

19 Jul. 1994 16 Aug. 1996 15 Aug. 1994

14 Jul. 2001 09 Oct. 2001 12 Jul. 2001

05 Aug. 2006 10 Sep. 2005 20 Sep. 2006

1933

W. Ouyang et al. / Commun Nonlinear Sci Numer Simulat 15 (2010) 1928–1941

Fig. 3. Time series land cover distribution of watershed and on-site scale in 1977, 1996, 2000 and 2006.

Table 4 Watershed scale land cover transformation. Land cover

Farmland Forestland Grassland Water area Construction land Bare land

1977

1996

2000

2006

ha

%

ha

%

ha

%

ha

%

410,355 527,936 2,162,346 53,919 2464 271,427

11.97 15.40 63.07 1.57 0.07 7.92

384,582 491,598 2,136,788 60,597 21,558 332,419

11.40 14.34 62.34 1.77 0.45 9.70

392,696 489,799 2,125,454 72,405 19,701 327,487

11.46 14.29 62.01 2.11 0.57 9.55

445,431 530,756 2,041,691 78,124 15,349 316,043

12.82 15.49 59.57 2.28 0.63 9.22

Table 5 Land cover transformation in the on-site region. Land cover

Farmland Forestland Grassland Water area Construction land Bare land

1977

1996

2000

2006

ha

%

ha

%

ha

%

ha

%

270,614 79,365 885,161 47,048 1226 103,516

19.51 5.72 63.82 3.39 0.09 7.46

260,952 82,928 865,766 52,282 15,045 109,959

18.82 5.98 62.42 3.77 1.08 7.93

267,387 82,271 853,265 65,005 15,256 103,746

19.28 5.93 61.52 4.69 1.1 7.48

277,540 90,120 822,886 70,355 15,604 110,426

20.01 6.50 59.33 5.07 1.13 7.96

site scale water area grew continually during the study period, occupying 47,048 ha in 1977, and then, in 2006, with eight reservoirs in operation, rose to 70,355 ha. In contrast to the fluctuation in forestland in the watershed scale, the forestland area for the on-site scale increased continually from 5.72% to 6.50% during the period 1977–2006. The conversion of bare land area in the on-site region was not as pronounced as in the watershed region. Bare land area increased continually from 103,516 ha in 1977 to 110,426 ha in 2006, a change that was considerably smaller than the overall increase of 44,616 ha in the watershed region. With the construction of hydropower stations, the construction land climbed steadily from 0.09% in 1977 of the total land cover to 1.13% in 2006.

1934

W. Ouyang et al. / Commun Nonlinear Sci Numer Simulat 15 (2010) 1928–1941

4. Relationships between land cover and hydropower cascade exploitation 4.1. Watershed transformations correlated with hydropower cascade exploitation Human activities, rather than natural factors, provide the main trigger for most land cover changes. With this as a premise, an attempt to separate natural variations due to climate change from those due to human activities is a major research focus [30]. Although the regional climate varied lightly as the global climate change, there is, no essential change occurred in precipitation and temperature over the last 40 years [31]. After field trip, the HCE has been recognized as the most intensive human activity and the only industry in this under development region over the last three decades [32]. Therefore, HCE is treated as the direct cause of regional land cover transformation. The total dam heights and total hydropower capacity have been selected to describe HCE characteristics in the four observed years. With the conclusions derived from the land cover analysis, the grassland and water areas have been chosen as indicators of typical land cover transformation over the last 30 years. Using simple regression methods, we first calculated the associations between the HCE indicators and the land cover indicators for the watershed region. During the period 1977–2006, the total dam heights increased to 800.7 m. For the same period, the water area climbed from 53,919 to 78,124 ha and grassland coverage decreased from 2,162,346 to 2,041,691 ha. The dam height total shows a strong association ðR2 ¼ 0:9508Þ with grassland area for the study period 1977–2006. The water area is also observed to have a significant positive relationship with dam height total ðR2 ¼ 0:9231Þ (Fig. 4). The association between land cover area and dam height total shows that each meter added to the dam height total will take away 183.7 ha of grassland. Similarly, each meter added to the dam height total will result in an additional 39.1041 ha of water area. Another set of correlations was derived by using hydropower capacity as the HCE indicator and relating it to grassland and water areas. In a manner similar to the increase in the dam height total, the total hydropower capacity increased steeply from 1225 MW in 1977 to 6864 MW in 2006. The results shown in Fig. 4 demonstrate a strong negative association ðR2 ¼ 0:9052Þ between grassland area and the total hydropower capacity. The association between the increase in water area and the hydropower capacity total also has a strong correlation index ðR2 ¼ 0:9646Þ. These correlation principles can be used to estimate the decrease in grassland area and increase in water area resulting from an increase in one million watts of hydropower: about 20 ha in grassland will disappear and approximately 4 ha of water will appear. With the aid of these four strong correlation models, watershed grassland and water area responses can be predicted for future HCE scenarios in the Longliu Section. 4.2. On-site transformations correlated with hydropower cascade exploitation Following the same procedures as those used in the previous section, the interaction between HCE indicators and grassland and water areas in the on-site region were analyzed. Compared with the correlations for the watershed region, the HCE indicators had a much more significant correlation with land cover in the on-site region. We obtained the clear conclusion that the two exploitation indicators had significant associations with two land cover area variables (Fig. 5). The dam height total had an extremely strong correlation ðR2 ¼ 0:9981Þ with on-site grassland area during the study period 1977–2006. For the same period, there was also a close positive interrelationship between dam height total and water area ðR2 ¼ 0:9108Þ. When the total hydropower capacity is used as the HCE indicator, strong associations were observed once again; a negative 8.0

8.0

2.20

Area Grass= -20.188*P + 2E+06 R2 = 0.9052

Area Grass= -183.7*D + 2E+06 R2 = 0.9508

7.5

7.5

7.0

6.0

Grass land /106 ha

Water area /10 4 ha

2.10

6.5

6.5

6.0

2.05 5.5

5.0 0

200

Water area

400 600 800 T otal dam heights (D) /m

Grassland

5.5

Area Water = 4.3867*P + 49702 R2 = 0.9646

Area Water = 38.101D + 49566 R2 = 0.9231 2.00 1000

0

2000

4000

6000

5.0 8000

T otal hydropower capacities (P) /MW

Correlation of water area

Correlation of grassland

Fig. 4. Correlations between HCE indicators and watershed land cover indicators.

Water area /10 4 ha

2.15 7.0

1935

W. Ouyang et al. / Commun Nonlinear Sci Numer Simulat 15 (2010) 1928–1941

7.5

7.5

0.90 Area Grass = -1E-05*P+ 0.8964 R2 = 0.9801

Area Grass = -9E-05*D + 0.898 R2 = 0.9981 0.88

7.0

6.0 0.84

Grass land /106 ha

Water area /10 4 ha

6.5

0.86

6.0

5.5

Water area /104ha

6.8

5.3 0.82 AreaWater = 0.0037*D + 4.2328 R2 = 0.9108 4.5 0

200

Water area

400 600 800 T otal dam heights (D) /m

Grassland

Area Water= 0.0004*P + 4.2389 R2 = 0.9601

0.80 1000

0

2000 4000 6000 T otal hydropower capacities (P) /MW

Correlation of water area

5.0

4.5 8000

Correlation of grassland

Fig. 5. Correlation between HCE indicators and on-sites land cover indicators.

correlation for grassland area ðR2 ¼ 0:9801Þ and a positive one for water area ðR2 ¼ 0:9016Þ. The R2 value indicated that a much closer relationship existed between grassland area and the two HCE indicators in the on-site scale than in the watershed scale. It can be seen that the grassland effects were concentrated within the on-site region. Stated briefly, the HCE had a much more direct impact on the grassland area in the on-site region than on the grassland area in the watershed region. However, the water area in the on-site region had a more complicated correlation trend. The total hydropower capacity had very close R2 values in the two studied scales. For the total dam height variable, the watershed scale correlation was higher than the onsite scale correlation. Nevertheless, the impact of the water area affected the entire watershed after the dams were constructed. In conclusion, the observation scale of grassland impact from HCE should be focused on the on-site region. 4.3. On-site grassland variation characteristics We have found that the on-site grassland transformations was affected by the HCE. In order to understand the variations of land cover in detail, the response of on-site grassland (the most dominant form of vegetation) in different aspects and at different elevations were analyzed. A grassland degradation rate was employed to describe grassland variation due to hydropower exploitation. The grassland degradation rate was calculated with the following equation:

D ¼ Ai =Aj

ð3Þ

where D is the degradation rate in an elevation zone with a given aspect; and, for the specified elevation zone and aspect, Aj is the original amount of grassland and Ai is the amount of grassland converted to bare land, construction land, water area, or farmland. The Arcgis 9.2 matrix tool was applied to explore on-site grassland transformation principles in relation to the other five land cover types, and identify variations in location, for three time periods. The grassland transferred to bare land, forestland, construction land, water area, and farmland was analyzed at pixel scale. The pixel aspect of grassland was analyzed based on nine different aspects that were defined for the study. After that, the varied pixels were categorized into different elevation conditions. The grassland in every aspect and elevation range was divided by the original grassland area in these locations. Finally, the grassland degradation rate was calculated and entered into the Origin statistical tool to determine the elevation and aspect (Fig. 6). The on-site grassland had a sustained degradation in the period 1977–1996 (Fig. 6a). The most remarkable degradation rate was 0.430, a result that occurred when the elevation was in the range of 1700–2000 m and the aspect was east. For grassland patches with the west and northwest aspects, light degradation occurred; the degradation rate was less than 0.054. The grassland in other aspects had a higher degradation rate than in the west; and the degradation was noticeably more serious for elevations greater than 3300 m and lower than 2900 m. The higher elevation area occurred in the vicinity of the Longyangxia hydropower station, a facility that was constructed during this time period. In this time period also, the increase in human activity around the Liujiaxia Reservoir caused grassland degradation in lower elevations. The grassland degradation in on the period 1996–2000 was not particularly notable; the maximum degradation rate was 0.130. However, some spatial distribution principles were still observable (Fig. 6b). Most of the degradation appeared in the south, southwest, and north aspects. The grassland patch located in the elevation range 1700–2300 m with the south and southwest aspects experienced the most sustained degradation. When the location of the HCE for this period was considered,

1936

W. Ouyang et al. / Commun Nonlinear Sci Numer Simulat 15 (2010) 1928–1941

Fig. 6. Results from the analysis of grassland degradation rates in different aspects and elevations.

it was clear that the degraded grassland patch was the result of operations at the Liujiaxia station. During the hydropower station operation, the water level in the reservoir rose, with the result that grassland along the banks was inundated. Grassland patches at lower elevations (below 2000 m) having the north aspect, also experienced obvious grassland degradation. The degraded areas, identified as locations near the Liujiaxia Reservoir, were the result of an increased community size related to station construction and maintenance. Moderate and more complex grassland degradation was observed during the period from 2000 to 2006 (Fig. 6c). Grassland degradation rates were relatively lower in the flat, north, and west aspects; all were less than 0.086. However, in the east, south, and southwest aspects, the degradation rates were in a middle range, the lowest value of which was 0.115. Grassland located at elevations in the range of 2500–3500 m experienced the highest degradation rates, with a maximum of 0.230. Two patches of grassland in the east and southwest aspects also experienced significant degradation. The elevation of one patch was around 1700 m and the other one was located at an elevation between 3300 and 3700 m. By analyzing grassland degradation characteristics in the on-site region, we determined that HCE had completely different influences on grassland depending on the elevation level and aspect features, which are important factors in grassland degradation studies. The grassland in the south and east aspects were sensitive and had high degradation rates. 4.4. Vegetation impact distance analysis After summarizing land cover impact ranges and grassland characteristics in HCE areas, vegetation response principles were analyzed based on distance from water area. Vegetation zones at distances of 0–0.1, 0.1–0.2, 0.2–0.3, 0.3–0.4, 0.4–0.5, 0.5–0.6, 0.6–0.7, 0.7–0.8, 0.8–0.9, 0.9–1, 1–2, 2–3, 3–4, 4–5, 5–6, 6–7, 7–8, 8–9, and 9–10 km from the central line of the main stream were delineated with the aid of buffer zone tools. Using the Arcgis Intersect tool, the vegetation NDVI in these zones was averaged; this information was used to identify fluctuation principles based on distance from water. This procedure enabled us to identify impact range mechanisms. Under different hydrological characteristics, vegetation based on distance from water, were analyzed in two groups: the first one focused on the vegetation around the three largest reservoirs (the five run-off hydropower stations were excluded) and the second focused on the main stream of the Yellow River (Fig. 7). After obtaining the vegetation NDVI values in different buffer zones, the spatial NDVI variation principle was mapped based on the distance from water. However, in addition to human activities, climatic features are also a critical factor [33]. In order to observe absolute differences in inter-annual NDVI variations and neglect climatic inputs, the standardized NDVI (SNDVI) was employed to analyze the vegetation impact mechanism and the equation used in performing these calculations was:

.X SNDVIi ¼ ðNDVIi  NDVI100 Þ ðNDVIi Þ

ð4Þ

where SNDVIi is the standardized NDVI value at a distance i meters from the water, NDVIi is the calculated NDVI value at a distance of i meters from the water, and NDVI100 is the NDVI value at a distance of 100 m from the water. 4.4.1. Spatial NDVI distribution in water distance around reservoir The vegetation NDVI in zones at various distances around the three dominant reservoirs in four observed years, 1977, 1994, 2000, 2006, are shown in Fig. 8. The four curves displayed there reveal similar variation principles, especially at water

W. Ouyang et al. / Commun Nonlinear Sci Numer Simulat 15 (2010) 1928–1941

1937

Fig. 7. Variation in NDVI vegetation distribution around reservoirs and along the Yellow River based on distance.

distances greater than 0.4 km. The NDVI values are the lowest for the year 2000 because it is the year that had the least amount of precipitation. In most vegetation zones where the distance to the water’s edge is less than 3 km, the NDVI values for the years 1994 and 2006 were nearly the same. Similar weather conditions occurred in those 2 years. However, the NDVI values at water distances greater than 3 km were separated from the 1994 values by increasingly large amounts, a result due to HCE that occurred during the 10-year time period. As the distance from the water’s edge increases over the range of 1– 10 km, the NDVI values changed more rapidly in 1977 than in the other years. This observation indicates that more intensive influences occurred in this range in 1977. In contrast, the areas whose distances from the water’s edge are in the range 0.4– 1 km were covered with dense grass and, in some cases, moderate amounts of grass along with farmland and villages. Relatively light impacts occurred there. In areas where the distance from the water’s edge was less than 0.4 km, especially in the first 0.1 km, the gap in vegetation NDVI values with those at the 0.2 distance widened in 1977. Nevertheless, a decrease occurred in the NDVI values as the distance from the water’s edge increased over the range of 0–0.2 km, an indication that the vegetation in this zone experienced a positive impact from hydropower exploitation. Thus, over the three-decade period a stable riparian vegetation system came into existence. An examination of the spatial distribution of the SNDVI values around the three reservoirs indicates that for most distances (all except the very nearest) from the water’s edge the highest SNDVI values occurred in the year 2006; these values

1938

W. Ouyang et al. / Commun Nonlinear Sci Numer Simulat 15 (2010) 1928–1941

0.25 1977

1994

2000

2006

NDVI value

0.15

0.05

-0.05

-0.15 0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

2 1 1 Distance /Km

3

4

5

6

7

8

9

10

Fig. 8. Analysis of vegetation NDVI in the vicinity of reservoirs from 1977 to 2006.

were even higher than those for 1977 (Fig. 9). This comparison indicates that the vegetation around the reservoirs benefited from the three-decade HCE, a positive result from micro-climatic improvement. At distances from the water’s edge that were less than 0.3 km, the vegetation SNDVI values for 2006 were less than the corresponding values for 1977. These areas were permanently affected by HCE and the recovery of vegetation in these locations was much more difficult. In most zones the SNDVI values for the years 1994 and 2000 were nearly the same, especially at distances from the water’s edge in the range 0.5–1 km. In the range 0.2–0.5 km, the SNDVI values for 2000 were higher than those for 1994, indicating that the vegetation was positively impacted from HCE. In the range 2–9 km, a reverse phenomenon occurred; the values for 1994 were higher than those for 2000, indicating that vegetation in this zone was disturbed by HCE during the 6-year period following 1994. Therefore, influences from HCEin the period from 1994 to 2000 occurred at distances from the water’s edge in the range of 2– 9 km. In brief, the SNDVI values demonstrated that vegetation around reservoirs at distances from the water’s edge in the range 0.4–1 km developed well during the three-decade period of hydropower exploitation. A comparison of inter-annual SNDVI values indicates that the affected vegetation zones around the reservoirs were concentrated at distances from the water’s edge in the ranges 0.1–0.4 and 1–6 km. 4.4.2. Spatial NDVI distribution in water distance along riverbank There is a similar spatial distribution principle of vegetation NDVI along the riverbank as in the vicinity of reservoirs (Fig. 10). The vegetation along the riverbank at distances of 0.4–1 km, where dense grass and farmlands existed, was not subject to any obvious spatial change principles. In the spatial curve for the year 1977 there was a sustained decrease at distances in the range 1–3 km; no such change was observed for the other studied 3 years. This change indicates that the HCE affected land cover in 1977, but recovery occurred in the following three decades, and the area was resistant to exploi-

1.00 1977

1994

2000

2006

Standardized NDVI

0.80

0.60

0.40

0.20

0.00 0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1 1

2

3

4

5

6

7

8

Distance /Km

Fig. 9. Standardized NDVI values around reservoirs at various distances from the water’s edge from 1977 to 2006.

9

10

1939

W. Ouyang et al. / Commun Nonlinear Sci Numer Simulat 15 (2010) 1928–1941 0.22 1977

1994

2000

2006

NDVI value

0.12

0.02

-0.08

-0.18

-0.28 0.0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1.0 1

2

3

4

5

6

7

8

9

10

Distance /Km Fig. 10. Vegetation NDVI distributions at varied distances from the river bank from 1977 to 2006.

tation impacts thereafter. In the zone with water distance between 0.1 and 0.2 km, the vegetation NDVI improved significantly. However, in the zone from the water’s edge between 0.2 and 0.4 km, the vegetation NDVI underwent a sustained decrease; this was the key area influenced by exploitations. Another disturbed area was the zone bounded by distances from the water’s edge of 3 and 10 km. The NDVI curves for the years 1977 and 2000 are parallel throughout this zone. However, the spatial distribution curve for the year 2006 shows an obvious decrease there, a result that reveals an impact from HCE in the period 2000–2006. An observation of SNDVI values along the river displayed in Fig. 11 shows that the vegetation experienced a sustained degeneration from 1977 to 2000, but the recovered somewhat by 2006. A comparison with Fig. 9 indicates that the vegetation around the reservoirs had a completely different spatial variation principle than is the case near the riverbank. The vegetation NDVI at distances from the water between 0.4 and 1 km was steady and similar inter-annual differences are evident. We concluded that the vegetation in this zone experienced similar influences over three decades. The curve for the year 1977 shows a very notable decrease in the SNDVI value at the 2 km mark, which reveals a critical impact from hydropower exploitation. However, there was no obvious variation at this distance in other observed years, indicating that the disturbance was under control. In the area near the riverbank with distance less than 0.4 km, the gap between the SNDVI values for 1977 and each of the other three observed years was greater than the values at any other distance. It was noted that this zone is most significantly affected. A comparison of the NDVI curves for 1994 and 2006 in Fig. 10 shows similar changes occurring throughout most zones, but the SNDVI values are higher for the year 2006. This observation indicates that the HCE in the recent 12 years did not harm the vegetation in on-site region. In contrast, the vegetation SNDVI at distances from the water of less than 2 km increased steadily and, therefore, the local ecosystem recovered. Briefly, the SNDVI analysis indicates that the most critically affected area along the main stream was at distances in the range 0.3–3 km, but over time the impacts from HCE were increasingly diminished and recovery was occurring over the most recent 12 years. 1.0

Standardized NDVI

0.8

0.6

0.4

0.2

1977

1994

2000

2006

0.0 0.0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1.0 1

2

3

4

5

6

Distance /Km

Fig. 11. Standardized SNDVI analysis at distances from the river bank from 1977 to 2006.

7

8

9

10

1940

W. Ouyang et al. / Commun Nonlinear Sci Numer Simulat 15 (2010) 1928–1941

The study of spatial distribution principles based on varied distances from water’s edge provided detailed information about impacts resulting from hydropower exploitation. The SNDVI filtered out annual climatic differences and quarantined the inter-annual comparisons provide reliable conclusions. The two analytical procedures demonstrated that the most directly affected area were the zones in water distances of 0.2–0.4 and 1–6 km. The vegetation along the riverbank located at distances less than 3 km experienced the most intensive influences from HCE before 2000, but then recovered in next 6 years. The restored vegetation had higher NDVI value than in 1994, but they were still lower than the corresponding values in 1977. However, the vegetation around the reservoirs recovered and was better than its original status. 5. Summary and conclusions Two land scales were used to assess the correlation between HCE indicators and types of land cover in an upper stream portion of the Yellow River. One scale was based on a watershed region and the other on an on-site region. The R2 values for the correlations of grassland area with both HCE indexes were higher for the on-site region than for the watershed region. The indicators had much more complicated correlations with water area in watershed region than in on-site region. After comparing coefficients of four correlation equations, it was concluded that the HCE had closer interaction with variation in water area within the watershed region. Consequently, observations of long-term grassland variations brought about by HCE should be focused on the on-site region. The analytical technique in this paper has proven to be an effective procedure for identifying long-term grassland impacts induced by HCE. In the case of the distance from the riverbank, the SNDVI spatial distribution revealed that the most directly affected areas were in the zones where the water distances were 0.2–0.4 and 3–10 km. It was also concluded that the vegetation along the riverbank was recovering, but that it had not yet returned to the original levels. In the vicinity of reservoirs, however, our analysis revealed that at distances greater than 0.4 km from the water’s edge, the SNDVI values in the years after 1977 were greater. Therefore, positive impacts occurred during the HCE. The inter-annual SNDVI comparisons showed that the affected vegetation zones around the reservoirs were concentrated at distances from the water’s edge in the ranges 0.1–0.4 and 1–6 km. The analysis of grassland variation in different aspects indicated that grasslands in the south and east aspects were more sensitive and experienced higher degradation rates. With the Westward Development Plan that China has undertaken, great opportunities are provided for the development of water resources in other rivers. This study provided the first lines of evidences of the durable and extensive environmental transformations that are induced HCE. The proposed procedure assists to forecast the impact by dams in the near future. The findings and principles learned from this study will be applicable to the development of processes for preserving land cover and, in addition, provide fundamental guidance for minimizing environmental impacts especially in the developing countries. Acknowledgements This paper would not have been possible without the help and assistance provided by Prof. Andrew K. Skidmore at the International Institute for Geo-Information Science and Earth Observation (ITC) and Professor Robert Wenger of the University of Wisconsin-Green Bay. We are grateful for assistance with data requirements, particularly spatial data, provided by personnel at the Data Center for Resources and Environmental Sciences, Chinese Academy of Sciences. The research project discussed in this paper benefited from financial support from the National Natural Science Foundation (Grant No. 40771192) and the International Institute for Geo-Information Science and Earth Observation (ITC) in The Netherlands. The authors would like to thanks the anonymous reviewers for their helpful comments. References [1] Wang ZY, Wu BS, Wang GQ. Fluvial processes and morphological response in the Yellow and Weihe Rivers to closure and operation of Sanmenxia Dam. Geomorphology 2007;91(1–2):65–79. [2] Zhang SR, Lu XX, Higgitt DL, et al. Recent changes of water discharge and sediment load in the Zhujiang (Pearl River) Basin, China. Global Planet Change 2008;60(3–4):365–80. [3] Burke M. Managing China’s water resources. Environ Sci Technol 2000;34(9):219–21. [4] Coker EH. Conversion of a flood control system to a sustainable system: the energy requirements for pipeline transport of silt. Environ Sci Technol 2000;34(17):3730–6. [5] Sahin S, Kurum E. Erosion risk analysis by GIS in environmental impact assessments: a case study—Seyhan Köprü Dam construction. J Environ Manage 2002;66(3):239–47. [6] Ashraf M, Kahlown MA, Ashfaq A. Impact of small dams on agriculture and groundwater development: a case study from Pakistan. Agric Water Manage 2007;92(1–2):90–8. [7] Lerer LB, Scudder T. Health impacts of large dams. Environ Impact Assess 1999;19(2):113–23. [8] Han SY, Kwak SJ, Yoo SH. Valuing environmental impacts of large dam construction in Korea: an application of choice experiments. Environ Impact Assess 2008;28(4–5):256–66. [9] Pamo ET, Tchamba MN. Elephants and vegetation change in the Sahelo-Soudanian region of Cameroon. J Arid Environ 2001;48(3):243–53. [10] Gordon E, Meentemeyer RK. Effects of dam operation and land use on stream channel morphology and riparian vegetation. Geomorphology 2006;82(3–4):412–29. [11] Paloscia S, Pampaloni P, Macelloni G, Sigismondi S. Microwave remote sensing of hydrological parameters on the NOPEX area. Agric For Meteorol 1999;98-99(31):375–87.

W. Ouyang et al. / Commun Nonlinear Sci Numer Simulat 15 (2010) 1928–1941

1941

[12] Garrigues S, Allard D, Baret F, Weiss M. Influence of landscape spatial heterogeneity on the non-linear estimation of leaf area index from moderate spatial resolution remote sensing data. Remote Sens Environ 2006;105(4):286–98. [13] Helmschro J, Flügel WA. Land use characterisation and change detection analysis for hydrological model parameterisation of large scale afforested areas using remote sensing. Phys Chem Earth Parts A/B/C 2002;27(9–10):711–8. [14] Busetto L, Meroni M, Colombo R. Combining medium and coarse spatial resolution satellite data to improve the estimation of sub-pixel NDVI time series. Remote Sens Environ 2008;112(1):118–31. [15] Maselli F. Monitoring forest conditions in a protected Mediterranean coastal area by the analysis of multiyear NDVI data. Remote Sens Environ 2004;89(4):423–33. [16] Ning SK, Chang NB, Jeng KY, Tseng YH. Soil erosion and non-point source pollution impacts assessment with the aid of multi-temporal remote sensing images. J Environ Manage 2006;79(1):88–101. [17] Bonazountas M, Kallidromitou D, Kassomenos P, Passas N. A decision support system for managing forest fire casualties. J Environ Manage 2007;84(4):412–8. [18] Feng JM, Wang T, Qi SZ, Xie CW. Land degradation in the source region of the Yellow River, northeast Qinghai–Xizang Plateau: classification and evaluation. Environ Geol 2005;47(4):459–66. [19] Wang XD, Li MH, Liu SZ, Liu GC. Fractal characteristics of soils under different land-use patterns in the arid and semiarid regions of the Tibetan Plateau, China. Geoderma 2006;134(1–2):56–61. [20] Xu GG, Ma GY, Li QB, Cui ZF. Underground excavation in Xiaolangdi project in Yellow River. Eng Geol 2004;76(1–2):129–39. [21] Wang HJ, Yang ZS, Yoshiki Saito, Liu JP, Sun XX, Wang Y. Stepwise decreases of the Huanghe (Yellow River) sediment load (1950–2005): impacts of climate change and human activities. Global Planet Change 2007;57(3–4):331–54. [22] Liu JY, Liu ML, Tian HQ. Spatial and temporal patterns of China’s cropland during 1990–2000: an analysis based on Landsat TM data. Remote Sens Environ 2005;98(4):442–56. [23] Angelis HD, Rau F, Skvarca P. Snow zonation on Hielo Patagónico Sur, Southern Patagonia, derived from Landsat 5 TM data. Global Planet Change 2007;59(1–4):149–58. [24] Bach M, Breuer L, Frede HG, Huisman JA, Otte A, Waldhardt R. Accuracy and congruency of three different digital land-use maps. Landscape Urban Plan 2006;78(4):289–99. [25] Tømmervik H, Høgda KA, Solheim I. Monitoring vegetation changes in Pasvik (Norway) and Pechenga in Kola Peninsula (Russia) using multi-temporal Landsat MSS/TM data. Remote Sens Environ 2003;85(3):370–88. [26] Chen XL, Zhao HM, Li PX, Yin ZY. Remote sensing image-based analysis of the relationship between urban heat island and land use/cover changes. Remote Sens Environ 2006;104(2):133–46. [27] Southworth J, Munroe D, Nagendra H. Land cover change and landscape fragmentation—comparing the utility of continuous and discrete analyses for a western Honduras region. Agric Ecosyst Environ 2004;101(2–3):185–205. [28] Wei OY, Skidmore AK, Hao FH, Chen Z, Yao HP, Zhang X. Land cover transformation feature understanding by MODIS and response to climate upper stream of Yellow river. In: ISPRS 2008, Proceedings of the XXI congress: silk road for information from imagery, 3–11 July, Beijing, China. Comm. VIII, WG VIII/3. p. 619–25. Available from: http://www.isprs.org/congresses/beijing2008/proceedings/8_pdf/3_WG-VIII-3/26.pdf. [29] Steven MD, Malthus TJ, Baret F. Intercalibration of vegetation indices from different sensor systems. Remote Sens Environ 2003;88(4):412–22. [30] Raumann CG, Cablk ME. Change in the forested and developed landscape of the Lake Tahoe basin, California and Nevada, USA, 1940–2002. For Ecol Manage 2008;255(8–9):3424–39. [31] Zhao L, Ping CL, Yang DQ, Cheng GD, Ding YJ, Liu SY. Changes of climate and seasonally frozen ground over the past 30 years in Qinghai–Xizang (Tibetan) Plateau, China. Global Planet Change 2004;43(1–2):19–31. [32] Xu JX. Historical bank-breachings of the lower Yellow River as influenced by drainage basin factors. Catena 2001;45(1):1–17. [33] Weiss JL, Gutzler DS, Coonrod JEA, Dahm CN. Seasonal and inter-annual relationships between vegetation and climate in central New Mexico, USA. J Arid Environ 2004;57:507–34.