- Email: [email protected]
Spatio-temporal analysis of vegetation variation in the Yellow River Basin
Ecological Indicators 51 (2015) 117–126
Contents lists available at ScienceDirect
Ecological Indicators journal homepage: www.elsevier.com/locate/ecolind
Spatio-temporal analysis of vegetation variation in the Yellow River Basin Weiguo Jiang a,b,∗ , Lihua Yuan a,b , Wenjie Wang c,∗ , Ran Cao a,b , Yunfei Zhang a,b , Wenming Shen d a
State Key Laboratory of Earth Surface Processes and Resource Ecology, Beijing Normal University, Beijing 100875, China Key Laboratory of Environmental Change and Natural Disaster, Beijing Normal University, Beijing 100875, China c Chinese Research Academy of Environmental Sciences, Beijing 100012, China d Satellite Environment Center, Ministry of Environmental Protection, Beijing 100097, China b
a r t i c l e
i n f o
Article history: Received 17 April 2014 Received in revised form 8 July 2014 Accepted 21 July 2014 Keywords: Yellow River Basin NDVI Spatial distribution Variation coefficient Trend analysis method Hurst index
a b s t r a c t To understand the variation and patterns of vegetation coverage in the Yellow River Basin, as well as to promote regional ecological protection and maintain ecological construction achievements, MOD13Q1 data at a resolution of 250 m were used to calculate the annual average normalised difference vegetation index (NDVI) in a time series from 2000 to 2010. Using a variation coefficient, a Theil–Sen Median trend analysis, the Mann–Kendall test, and the Hurst index method, this study investigated the temporal and spatial variations of vegetation coverage characteristics of the Yellow River Basin. The results showed that (1) the vegetation coverage of the Yellow River appeared to have an overall trend of high in the southeast and west and low in the northwest; (2) the averaged NDVI of the whole basin fluctuated in a range of 0.3 to 0.4 from 2000 to 2010 (from 2000 to 2004 there were larger variations and these have been growing rapidly since 2005); (3) the NDVI was stable, 73.4% of the vegetation-coverage area fluctuated with a low-to-medium amplitude, while 27.6% of the area varied by a large amplitude; (4) the regions with improved vegetation coverage (62.9%) were far greater than the degraded regions (27.7%), while the sustained invariant area accounted for 9.4% of the total vegetation coverage regions; and (5) 86% of the vegetation-covered area was positively sustainable. The areas with sustainable improvement accounted for 53.7% of the total vegetation coverage area; the invariant area accounted for 7.8%. The area with sustainable degradation was 24.5%; the future variation in trends of the residual (14%) could not be determined. Therefore, continuous attention must be given to the variation in trends of vegetation in the sustainably degraded and underdetermined regions. © 2014 Elsevier Ltd. All rights reserved.
1. Introduction Vegetation is characterised by inter-annual and seasonal variations. As a natural tie connecting atmosphere, water, and soil, vegetation plays a notably important role in soil conservation, atmosphere adjustment, and the maintenance of climatic and whole ecosystem stability (Sun et al., 1988). Because variations in the surface vegetation coverage affect the balance of regional ecosystems, studies on the variation in vegetation coverage is the
∗ Corresponding authors at: Key Laboratory of Environmental Change and Natural Disaster, Beijing Normal University, No.19 Xinjiekouwai Street, Haidia., China. Tel.: +86 10 58802923. E-mail addresses: [email protected] (W. Jiang), [email protected] (W. Wang). http://dx.doi.org/10.1016/j.ecolind.2014.07.031 1470-160X/© 2014 Elsevier Ltd. All rights reserved.
basis of the protection of the ecological environment (Fan et al., 2012; Peng et al., 2012; Zhang et al., 2013). With wide coverage, high temporal resolution rate, and free data, sensors such as National Oceanic and Atmospheric Administration/Advanced Very High Resolution Radiameter (NOAA/AVHRR), Systeme Probatoire d’Observation dela Tarre/VEGETATION (SPOT/VGT), and moderate resolution imaging spectroradiometer (MODIS) can provide large amounts of data for monitoring the variations of vegetation coverage over long time periods (Rigina et al., 1996; Tucker et al., 2005; Ma et al., 2006; Fensholt et al., 2009, 2012a,b; Fu et al., 2014). NDVI is functionally correlated with leaf area index (LAI) and vegetation coverage (Baret et al., 1991; Gutman et al., 1998); the higher the NDVI, the larger the LAI, and the higher the vegetation coverage. Therefore, NDVI can reflect the growth status of surface vegetation and also acts as an effective index for monitoring vegetation variations (Zhao, 2003; Tucker et al., 1985). For over 20 years, NDVI data
118
W. Jiang et al. / Ecological Indicators 51 (2015) 117–126
have been used to analyse spatial distribution characteristics in a large space scale and a long time scale (Tucker et al., 1985; Myneni et al., 1997; Senay et al., 2000; Stow et al., 2004); other methods, such as principal component analysis (Fan et al., 2012), unary linear regression (Piao et al., 2001; Song et al., 2011), variation vector analysis (Chen et al., 2002), Theil–Sen median trend analysis, the Mann–Kendall test (Pouliot et al., 2009; Fensholt et al., 2012a,b), Fourier transformation (Wang et al., 2006), and wavelet analysis method (Martínez and Gilabert, 2009) have also been employed to explore the temporal and spatial variation characteristics of the vegetation coverage. Moreover, correlation analysis has been used to investigate the correlations between inter-annual/annual variations and climate (Fensholt et al., 2012a,b; Kawabata et al., 2001; Blazkova and Beven, 2004; Zhang et al., 2010). The Yellow River Basin is located in arid, semi-arid, and semihumid areas in China. This river is endowed with a diversified climate, severely fluctuating topography, various types of landforms, and rich vegetation types (Liu et al., 2006). In recent years, with the change of climate and the constant intensification of human activities, this basin has exhibited vegetation changes. At present, a number of researchers have explored the vegetation changes in the Yellow River Basin. Sun et al. (2001) employed the NOAA AVHRR of 8 km to investigate the temporal and spatial distribution of vegetation from 1982 to 1999 and reported correlations of the NDVI with precipitation and temperature (Sun et al., 2001; Yang et al., 2003). Also using a NOAA AVHRR of 8 km, Li et al. (2004) studied the spatial distribution, the annual and seasonal variations of NDVI, as well as the inter-annual and annual correlations of NDVI with the precipitation and runoff in the Yellow River Basin from 1982 to 1999 (Li et al., 2004). Liu et al. (2006) used a 1 km of NOAA AVHRR to analyse the relationships of NDVI with temperature and rainfall in the Yellow River Basin. Based on the data of 1 km SPOT VGT, He et al. (2012) used the mean value method and the trend line analysis method to analyse the temporal and spatial distribution, time variation characteristics, and inter-annual variation in trends of the NDVI in the Yellow River Basin (He et al., 2012). Previous research has been primarily based on NOAA AVHRR data with spatial resolution of 1 km/8 km, or the SPOT VGT data with a 1 km spatial resolution; thus, the spatial resolution of the current research is relatively low. For analysing the inter-annual trends of vegetation, the method of linear regression analysis has been used with NDVI time series data. Using this method, vegetation trends are calculated by regression, which can easily be affected by outliers. Few studies using the more robust Theil–Sen trend analysis and Mann–Kendall tests have been conducted to explore the inter-annual trend of vegetation. Few of the studies on vegetation trends in the Yellow River Basin in the past have been focused on the vegetation trends for the future. This may be explained by the difficulty in simulating vegetation trends in the future through mathematical models between the NDVI and its related influencing factors. The Hurst exponent has been used to quantitatively detect the sustainability of time series data, but, to date, it has not been exploited for vegetation trend prediction in the Yellow River Basin. For our study, we choose MOD13Q1 data with a 250 m spatial resolution. Studies have indicated that there is a greater advantage on the spectral and spatial resolution using MODIS-250 m NDVI product data, and these data could provide more precise information on the land surface compared with the SPOT/VGT-1 km NDVI, and NOAA/AVHRR-8 km NDVI data (Muchoney et al., 2000). The data used in this research is more precise when compared with previous studies. The MOD13Q1 data were pre-processed and calculated to obtain the annual average NDVI time series in the Yellow River basin from 2000 to 2010. Next, using a variation coefficient, a Theil–Sen Median trend analysis, the Mann–Kendall test, and the Hurst exponent methods, this study investigated the temporal and
spatial variations, fluctuation characteristics, variation in trends, and sustainability of the vegetation coverage in the Yellow River basin. It is expected that understanding the variation characteristics and patterns of the vegetation coverage in the Yellow River Basin will promote the regional ecological protection and maintain ecological construction achievements. 2. Methods 2.1. Study area Originating from Bayankala Mountain in the Qinghai Province in China, the Yellow River flows through 9 Provinces, including Qinghai, Sichuan, Gansu, Ningxia, the Inner Mongolia Autonomous Region, Shaanxi, Shanxi, Henan, and Shandong. In Kenli County of the Shandong Province, it merges into the Bohai Sea. The Yellow River is 1,900 km long from east to west and 1100 km long from south to north and covers an area of 79.46 × 104 km2 . The river’s geographical coordinates are 39◦ 28 to 41◦ 05 N and 115◦ 25 to 117◦ 30 E under the WGS84/Albers Equal Area Conic projection (Fig. 1). The terrain of Yellow River basin is high in the west and low in the east. The western origination area lies at an average altitude of 4000 m and is composed of a series of mountains; the central region is in an altitude of 1000–2000 m and presents a loess landform with serious soil erosion; and the eastern area is 100 m below sea level and is mainly composed of the alluvial plain of the Yellow River (He et al., 2012). The area of the Yellow River has a continental climate. The south-eastern part belongs to semi-humid climate, the middle part has a semiarid climate, and the northwest part has a subordinate arid climate. The various landforms and complex habitats of the Yellow River Basin create favourable conditions for the development of various vegetation types (Liu et al., 2006); land-use types are mainly woodland, grassland, and farmland. 2.2. Methods 2.2.1. Data sources MODIS NDVI data were sourced from the MODIS vegetation index product data-MOD13Q1 on the website of the United States NASA. The data, collected from February 2000 to December 2010, were in Hdf format with a spatial resolution of 250 m, and a temporal time resolution of 16 days. The data were subjected to format and projection conversions using MODIS Reprojection Tool (MRT), which can be acquired from NASA Land Processes Distributed Active Archive Center (Sioux Falls, South Dakota, U.S.A.). The original Hdf format was transformed into the Geotiff format and the original Sinusoidal projection was transformed into the WGS84/Albers Equal Area Conic projection. The converted data were re-sampled using the adjacent natural method at a resolution of 250 m. The samples obtained were processed using the internationally accepted and commonly used maximum synthesis method to yield monthly NDVI data from 2000 to 2010, to avoid the influences of clouds, atmosphere, and solar altitude angle (Piao et al., 2001; Holben, 1986). Finally, using the mean value method, the annual average NDVI data were acquired to eliminate the influence of extreme yearly abnormal climate on the growth status of vegetation (Zhang et al., 2008). Land use/land cover data for the Yellow River Basin in 2010, at a resolution of 250 m, were provided by the project team, “The remote sensing investigation and assessment on the ten-year changes of the national ecological environment.” The land use types included forest, grassland, farmland, wetland, and artificial surface. 2.2.2. Methods The variation coefficient, Theil–Sen median trend analysis, Mann–Kendall, and Hurst exponent method were used to study the
W. Jiang et al. / Ecological Indicators 51 (2015) 117–126
Fig. 1. Location of the study area.
119
120
W. Jiang et al. / Ecological Indicators 51 (2015) 117–126
vegetation-covered regions of the Yellow River Basin, namely, the spatial distribution, temporal variation characteristics, volatility, variation in trends, and sustainability of the NDVI of the pixelcovered region with NDVI values greater than or equal to 0.1 (Piao et al., 2001). 2.2.2.1. Coefficient of variation. Coefficient of variation is commonly used in vegetation research and is mainly used to reflect the discrete degree of the NDVI data and the inter-annual/annual volatility of vegetation (Tucker et al., 2001; Milich, 2000). It is calculated by: CVNDVI =
NDVI NDVI
,
(1)
where, CVNDVI refers to the coefficient of variation of the NDVI value of each pixel in 2000–2010; NDVI represents the standard deviation of the NDVI value; NDVI is the mean NDVI value. A larger CVNDVI signifies that the NDVI time series shows larger data fluctuation, while a smaller CVNDVI denotes a more stable NDVI time series. 2.2.2.2. Theil–Sen Median trend analysis and the Mann–Kendall test. The Theil–Sen Median trend analysis method can be effectively combined with the Mann–Kendall test. These are important methods for judging the trend of long time series data, and this combination has been gradually used to analyse the long time series of vegetation reflecting the variation in trends of each pixel in a time series (Fensholt et al., 2012a,b; Milich, 2000; Lunetta et al., 2006). The Theil–Sen Median trend analysis is a robust trend statistical method (Theil, 1950; Sen, 1968; Hoaglin et al., 2000) and it calculates the median slopes between all n(n − 1)/2 pair-wise combinations of the time series data. It is based on non-parametric statistics and is particularly effective for the estimation of trends in small series (Hoaglin et al., 2000). The slope of Theil–Sen Median can represent the increase or decrease in the NDVI over the 11 years between 2000 and 2010 on a pixel scale. It is calculated by:
SNDVI = median
NDVIj − NDVIi j−i
,
2000 ≤ i < j ≤ 2010,
where, S =
j=1
i=j+1
sgn NDVIj − NDVIi =
0
s(S)
n−1 n
⎧ S−1 ⎪ , S>0 ⎪ ⎪ ⎪ ⎨ s(S) ⎪ ⎪ S+1 ⎪ ⎪ ⎩
,
S=0 .
,
S<0
(3)
sgn NDVIj − NDVIi ,
⎧ ⎨ 1
,
NDVIj − NDVIi > 0
0
,
NDVIj − NDVIi = 0 ,
−1
,
NDVIj − NDVIi < 0
⎩
2.2.2.3. Hurst exponent. The Hurst exponent is a method for distinguishing sustainability of time series data. It was proposed by British hydrologistHurst (1951), then Mandelbrot and Wallis (1969) improved the theory. This exponent has been widely used in the fields of hydrology, climatology, economics, geology, and geochemistry. Recently, it has been applied in the time series detection of vegetation variations (Fan et al., 2012; Hou et al., 2012; Wang et al., 2010). The primary calculation procedures are as follows (Sánchez Granero et al., 2008):
s (S) =
1 NDVI(t)
NDVI() =
= 1, 2, . . ., n
(4)
t=1
(3) To calculate the accumulated deviation, X(t,) =
t
NDVI(t) − NDVI()
1≤t≤
(5)
t=1
(4) To create the range sequence, R() = max X(t,) − min X(t,) 1≤t≤
1≤t≤
= 1, 2, . . ., n
(6)
(5) To create the standard deviation sequence,
S() =
2 1
NDVI(t) − NDVI()
1/2 = 1, 2, . . ., n
(7)
t=1
(6) To calculate Hurst exponent, R() = (c)H S()
(8)
The value of H is obtained by fitting the equation log(R/S)n = a + H × log(n), using the least squares method, where H is the Hurst exponent. According to previous research (Hurst, 1951; Mandelbrot and Wallis, 1969), the value of the Hurst exponent (H value) can be expanded from 0 to 1 and is valued in three types: when the H value is equal to 0.5, it means that the NDVI time series is a stochastic series without sustainability, indicating that the trend of the time series in the future would be unrelated with that of the study period; when the value is greater than 0.5, it means that the sustainability of the NDVI time series shows the same trend as the NDVI the time series in the future; and when the value is less than 0.5, it refers to the anti-sustainability of the NDVI time series, representing the opposite trend of the NDVI time series in the future.
n (n − 1) (2n + 5) 18
where, NDVIi and NDVIj represent the NDVI values of the pixels i and j; n is the length of the time series; sgn is a sign function; and the
(1) To define the time series NDVI(t) , t = 1, 2, . . ., n. (2) To define the mean sequence of the time series,
(2)
where, SNDVI refers to the Theil–Sen median, and NDVIi and NDVIj represent the NDVI values of years of i and j. In case of SNDVI > 0, the NDVI presents a rising trend, otherwise, the NDVI displays a decreasing trend. The Mann–Kendall test measures the significance of a trend. It is a non-parametric statistical test and it has the advantage that samples do not need to obey certain distributions and is free from ´ 2004). It has been the interference of outliers (Kendall, 1975; Toˇsic, widely applied to analyse the trends and variations at sites with hydrological and meteorological time series. Recently, this test has been used to study of vegetation variations over long time periods (Fensholt et al., 2009). The calculation formula is as follows: it is assumed that {NDVIi } , i = 2000, 2001, . . ., 2010
the statistics of Z is defined as Z =
Z statistic is valued in the range of (−∞, +∞). A given significance level, |Z| > u1–˛/2 , signifies that the times series shows significant variations on the level of ˛. Generally, the value of ˛ is 0.05. In this study, we choose ˛ = 0.05, meaning we measured the significance of the NDVI trend from 2000-2010 on pixel scale at a confidence level of 0.05.
W. Jiang et al. / Ecological Indicators 51 (2015) 117–126
121
Fig. 2. Spatial distribution of average NDVI in the Yellow River Basin from 2000 to 2010.
3. Results 3.1. The characteristics of the spatial distribution of vegetation coverage Fig. 2 shows the spatial distribution of average NDVI using the average NDVI for the 11 years from 2000 to 2010 and the spatial distribution characteristics of the NDVI in the Yellow River Basin. As shown in this figure, the NDVI spatial distribution in the Yellow River basin is high in the west and south-east, while it is low in the north. The woodlands, herbaceous wetlands, and crops have a high NDVI value, while the grassland etc. have a low NDVI value. The western area, located at a high altitude, is primarily composed of forest, grassland, and herbaceous wetlands and shows a high NDVI value. The southeastern part has a semi humid climate and has extensive forests and crops, therefore, the NDVI is relatively high and significant. The northern area is mainly comprised of the Erdos Plateau, the Loess Plateau, the Hetao Plain, and the Ningxia Plain. The Erdos and Loess Plateaus have low NDVI values because they are less covered by vegetation. However, the Hetao and Ningxia Plains belong to the alluvial plain of the Yellow River and are suitable for planting crops. The annual NDVI of the two plains is 0.3–0.4. The NDVIs of the non-vegetated areas (red part) are less than 0.1. These areas are mainly lake, glacier, bare rock, reservoir, desert, including the sandy lands and desert in the north of the basin. The results of the statistical classification of the average NDVI values over the 11 years of the study showed that non-vegetated area with a NDVI value under 0.1 accounted for 1.7% of the total basin area; the area with low NDVI values (0.1–0.4) accounted for 53.3%; the area with high NDVI values (above 0.1–0.4) accounted for 19.5%; the area with NDVI values from 0.5 to 0.6 accounted for 10.5%; the area with NDVI values above 0.6 accounted for 5%, and the areas with vegetation coverage were 98.3% of the total basin area. 3.2. The temporal variation characteristics of the vegetation coverage To study the temporal variation characteristics of the NDVI of the Yellow River Basin, the average NDVI of the pixels in the vegetationcovered areas in the images of annual average NDVI were used to represent the overall situation of the vegetation in the corresponding year. This study mainly analysed the NDVI variations of the vegetation-covered area. As shown in Fig. 3, the annual average
Fig. 3. Inter-annual variation of the NDVI in the Yellow River Basin from 2000 to 2010.
NDVI of the vegetation-covered area in the Yellow River Basin fluctuated from 0.3 to 0.4. The NDVI of the vegetation-covered area exhibited larger variations from 2000 to 2004. However, the variations did not show a significant trend. Since 2005, the NDVI of the vegetation-covered areas have been growing rapidly. 3.3. The volatility of the vegetation coverage According our calculations, the CVNDVI could be classified into 5 grades, including high volatility (CVNDVI ≥ 0.20), relatively high volatility (0.15 ≤ CVNDVI < 0.20), medium volatility (0.10 ≤ CVNDVI < 0.15), relatively low volatility (0.05 ≤ CVNDVI < 0.10), and low volatility (CVNDVI < 0.05). Table 1 shows the statistical classification results of the NDVI coefficient of variation on the pixel scale. As shown, 73.4% of the vegetationcovered areas have medium-high volatility, while merely 27.6% of the areas present high volatility. Therefore, it can be concluded that vegetation coverage is stable in the Yellow River Basin. Table 1 Coefficient of variation of NDVI in the Yellow River Basin. CVNDVI
Volatility degree
Area percentage (%)
≥0.20 0.15 ≤ CVNDVI < 0.20 0.10 ≤ CVNDVI < 0.15 0.05 ≤ CVNDVI < 0.10 <0.05
Low volatility Relatively low volatility Medium volatility Relatively high volatility High volatility
37.42 40.41 32.55 54.67 39.89
122
W. Jiang et al. / Ecological Indicators 51 (2015) 117–126
Fig. 4. Spatial distribution of the coefficient of variation of the inter-annul NDVI in the Yellow River Basin from 2000 to 2010.
Fig. 4 illustrates the spatial distribution of the coefficient of variation in the Yellow River Basin. Areas with high and relatively high volatility are mainly distributed in the Bayan Har Mountain, west of Amne Machin Mountain, Xining City, Lanzhou City, Longzhong Loess Plateau, Tengger Desert, Loess Plateau in central Ningxia, Ningxia Plain, Hedong Sand-land, Table Mountain, Langshan Mountain, the southern foot of Yinshan Mountain, Hetao Plain, Wula Mountain, the area from Baotou city to Hohhot City, the Daqing Mount area, Loess Plateau at the junction of Shanxi and Shaanxi Provinces, and the Taiyuan basin. The areas with medium volatility are mainly distributed in central Qinghai Province, east of Amne Machin Mountain, Sichuan Province, central Gansu Province, the junction of Gansu and Shaanxi Provinces, northern Shanxi Province, eastern Henan Province, and Shandong Province. The areas with medium volatility are mainly distributed in the Mu Us sand-land in the east of the Hedong sand-land, southern Shaanxi Province, southern Shanxi Province, and western Henan Province. The areas with low volatility are scattered along the basin. 3.4. The variation in trends of vegetation coverage The variation in trends of each pixel can be effectively characterised using the Theil–Sen Median trend analysis and the Mann–Kendall test to reflect the spatial distribution characteristics of the variation in the trend of the NDVI in the Yellow River Basin. Because areas with a SNDVI of 0 strictly do not exist, we made the following classifications according to the real conditions of the SNDVI of the NDVI. Areas with an SNDVI from −0.0005 to 0.0005 are classified as stable areas, areas with an SNDVI greater than or equal to 0.0005 are classified as improved areas, and areas with a SNDVI smaller than 0.0005 are classified as degraded areas. Moreover, significance test results of the Mann–Kendall test, at the confidence level of 0.05, are classified as significant variations (Z > 1.96 or Z < −1.96) or insignificant variations (−1.96 ≤ Z ≤ 1.96). By integrating the classification results of the Theil–Sen median trend analysis and the Mann–Kendall test, it is comparable to the data of trend variations of the NDVI. The results were categorised into five classes (Table 2). Table 2 shows the areas with vegetation coverage improvement, the areas with stable vegetation coverage (or without significant vegetation coverage variations), and the areas with vegetation degradation, which account for 62.9%, 9.4%, and 27.7%, respectively.
Table 2 Trend of the NDVI in the Yellow River Basin. SNDVI
Z
NDVI trend
Area percentage (%)
≥0.0005 ≥0.0005 −0.0005–0.0005 <−0.0005 <−0.0005
≥1.96 −1.96–1.96 −1.96–1.96 −1.96–1.96 <−1.96
Significantly improved Slightly improved Stable Slightly improved Severely degraded
32.8 20.1 9.4 21.9 5.8
Note: the pixels with SNDVI in −0.0005 and 0.0005 and Z > 1.96 or Z < −1.96 are classified as the stable.
As shown in Fig. 5, the area with improved surface vegetation is far larger than the area with degraded surface vegetation of the Yellow River Basin from 2000 to 2010. The areas with significantly improved vegetation coverage are mainly distributed in southern Gansu Province, northern Shaanxi Province, central and northern parts of Erdos City, Langshan Mountain, and western Shanxi Province, and both sides of the Yellow River Basin in Qinghai Province. The areas with slightly improved vegetation coverage are mainly distributed in the central areas of the junction of Sichuan Province, Gansu Province, and Qinghai Province, and Lvliang Mountain. The areas with stable vegetation coverage show a scattered distribution and the areas with slightly degraded vegetation are mainly located in Bayan Har Mountain, Amne Machin Mountain, Qilian Mountain, the area from Baotou City to Hohhot City, the southern foot of Yinshan Mountain, Ziwu Ridge, the north and west of Qingyang City in the Gansu Province, as well as the forestcovered areas in the west and south of Shaanxi Province. Areas with severe degradation are mainly observed in the Longzhong Loess Plateau, Hetao Plain, Taiyuan Basin, Daqing Mount, Jincheng City, and the two ends of the Guanzhong Basin. Areas with both slight and severe degradation mainly lie in Ningxia Plain, the west of Erdos City, Linfen Basin, Yuncheng Basin, Luoyang City, and the Shandong Province. 3.5. The sustainability of the vegetation coverage variations As shown in Fig. 6, the Hurst exponent of the NDVI of the Yellow River Basin is 0.5 on average. The regions with Hurst exponents lower than 0.5, which were sustainable, accounts for 14.0% of the total basin area. Those higher than 0.5, which were unsustainable,
W. Jiang et al. / Ecological Indicators 51 (2015) 117–126
123
Fig. 5. Trends of inter-annual NDVI change in the Yellow River Basin from 2000 to 2010.
Fig. 6. Sustainability of inter-annual NDVI change in the Yellow River Basin from 2000 to 2010.
account for 86.0% of the total basin area. There is little area with Hurst exponent equal to 0.5. This result suggests that the NDVI of the vegetation-covered areas is positively sustainable. To reveal the variation in trends and the sustainability of the vegetation, the data obtained using the Theil–Sen Median trend analysis and the Mann–Kendall test were superimposed with the Hurst exponent results to yield the coupled information of variation in trends and sustainability (Fig. 7). The coupling results were analysed on 6 cases: (1) sustainability and severe degradation; (2) sustainability and slight degradation; (3) sustainability and stability; (4) sustainability and slight improvement; (5) sustainability and significant improvement; and (6) undetermined variation in trends in the future. The latter includes five combinations, namely, unsustainability and severe degradation, unsustainability and slight degradation, unsustainability and stability, unsustainability and slight improvement, and unsustainability and significant improvement. Thus, the future variation in trends is difficult to determine.
As suggested by Fig. 7 and Table 3, the area showing sustainability and improvement accounts for 53.7% and is mainly distributed in the central and northern Shaanxi Province, southeastern Gansu Province, central Shanxi Province, and the east and north of Langshan Mountain and Erdos City. The area with sustainability and stability accounts for 7.8% and is primarily scattered in Inner Mongolia, Ningxia, Gansu, Qinghai, Shanxi, and Henan Provinces. The area with sustainability and degradation accounts for 24.5% and is mainly distributed in Bayan Har Mountain, south of Amne Machin Mountain, Qilian Mountain, Longzhong Loess Plateau, Ningxia Plain, Central Ningxia, the north of Qingyang City in the Gansu Province, Hetao Plain, the southern foot of Yinshan Mountain, Daqing Mount, the area from Baotou City to Hohhot City, Taiyuan Basin, Linfen Basin, Yuncheng Basin, Jincheng City, Guanzhong Basin, Luoyang City, and Shandong Province. The variation in trends of 14.0% of the basin is difficult to determine and these areas are mainly located in Amne Machin Mountain, Huangnan and Gannan Tibetan Autonomous Prefectures, Xining City, central
124
W. Jiang et al. / Ecological Indicators 51 (2015) 117–126
Fig. 7. Spatial distribution of the NDVI dynamics based on trends and the Hurst index. Table 3 Statistical result of trends and the Hurst index. SNDVI
Z
H
Variation types
Area percentage
<−0.0005 <−0.0005 −0.0005–0.0005 ≥0.0005 ≥0.0005 –
≥1.96 −1.96–1.96 −1.96–1.96 −1.96–1.96 <−1.96 –
>0.5 >0.5 >0.5 >0.5 >0.5 <0.5
Sustainability and severe degradation Sustainability and slight degradation Sustainability and stability Sustainability and slight improvement Sustainability and significantly improvement Undetermined future variation in trends
5.7 18.8 7.8 23.5 30.2 14
Gansu Province, and central Shaanxi Province. The vegetation conditions of the continued degradation and undetermined regions calls will require additional investigation.
4. Discussion The Yellow River Basin covers three steps of topography in China, and its terrain is highly variable and complex; however, the NDVI is rarely affected by topography, which is why it is possible to enhance its ability to respond to vegetation variation (Matsushita et al., 2007). Major land use/land cover types in the Yellow River Basin are grassland, farmland and forest (mainly deciduous), so the NDVI applied in this case study avoided asymptotic (saturated) signals over high biomass conditions (Huete, 1988). We selected the NDVI as the vegetation proxy and for this case study is reasonable; 250 m MODIS-NDVI data can provide more precise information regarding land surface than AVHRR-NDVI and VGTNDVI data, therefore, vegetation information was more accurately reflected by the MODIS-NDVI. The Theil + Sen Median Slope Analysis, together with the Mann–Kendall test is a new method for examining the trend analysis of NDVI time series data. This combination analysis is more advantageous than linear regression and correlation analysis and its main advantages include: data used does not need to obey a particular distribution, strong ability to avoid error and significance level testing also has a solid basis in statistical theory (Cai et al., 2009),
Therefore, this case study applied this combination, which made the examination of the vegetation trends in the Yellow River Basin more scientific and credible compared with past studies. Overlaying the results of the trend analysis and the results of the Hurst exponent calculations further quantified the future sustainability of the NDVI trends. Improving upon past studies on vegetation variations in the Yellow River Basin, this study was able to quantify future vegetation trends by overlaying the results of the trend analysis and the results of the Hurst exponent calculations. However, the Hurst exponent could not predict how long the sustainability of predicted vegetation trends would continue into the future, it was more accurate in determining the duration of the vegetation trends. There is an urgent need to develop methods to investigate the trend duration. Unfortunately, we were unable to find the methods needed to achieve this goal. Based on MODIS-250 m NDVI data from the Yellow River (2000 to 2010), in this case study applied a variation coefficient, the Theil–Sen Median trend analysis, the Mann–Kendall test, and the Hurst exponent method to investigate the temporal and spatial variations, fluctuation characteristics, variation in trends, and sustainability of the vegetation characteristics in the Yellow River Basin. However, more detailed work is necessary to analyse the vegetation variations of many vegetation types and to explore the cause of vegetation variations. The Yellow River Basin covers a large scale (79.46 × 104 km2 ) and various vegetation types are distributed
W. Jiang et al. / Ecological Indicators 51 (2015) 117–126
throughout the Basin. Different vegetation types demonstrate different inter-annual variations, therefore, it is necessary to analyse the variations of the different vegetation types. The variations of the NDVI were affected by climate change, human perturbations, and national/local policies; therefore, the driving forces associated with vegetation variations should be explored. Such explorations would be a great help in fully understanding vegetation variations, and would aid local government in adopting policies or projects that promote vegetation restoration.
125
Technology Support Program (2012BAH32B03 and 2012BAH33B05) and the Remote Sensing Investigation and Assessment Project for Decade-change of the National Ecological Environment (2000–2010). The authors sincerely thank Dr. Wang Mengjie and Dr. Tao Liangliang from Beijing Normal University, for the help in IDL programming and support in the Mann–Kendall test programming and Gao Mingliang from Capital Normal University, for providing the IDL program for the maximum synthesis algorithm.
5. Conclusions References In this study, NDVI time series data from 2000 to 2010 were obtained based on MOD13Q1 data at a resolution of 250 m. Variation coefficient, Theil–Sen Median trend analysis, the Mann–Kendall test, and the Hurst index method were used to investigate the temporal and spatial variations of the vegetation coverage characteristics of Yellow River Basin. (1) Regarding spatial distribution: the vegetation coverage of the Yellow River had an overall trend of high in the southeast and west while being low in northwest; the woodland, herb wetland, and crop areas showed high NDVI values, while the grassland and other types presented low NDVI values. (2) Regarding temporal variations: the annual average NDVI of vegetation coverage fluctuated in a range from 0.3 to 0.4 in 2000–2010. The NDVI in 2000–2004 exhibited larger variations, however, the variations did not present a significant trend. Since 2005, the NDVI has grown rapidly. (3) Regarding the volatility of the NDVI: the vegetation coverage in the Yellow River Basin was stable with 73.4% of the vegetationcovered areas fluctuating in a low-to-medium range, while 27.6% of the areas varied with a large amplitude. (4) Regarding the variation in trends of the NDVI: the regions with improved vegetation coverage were far greater than the degraded regions in the Yellow River basin in 2000–2010. The improved and degraded regions accounted for 62.9% and 27.7% of the total vegetation-covered region, respectively, while the rest, 9.4%, was undetermined. (5) The Hurst exponent-data analysis suggested that the NDVI of 86% of the vegetation-covered area was positively sustainable, namely, the NDVI showed strong sustainability. According to the superimposed variation in trends and sustainability data, the area showing sustainable improvement accounted for 53.7% of the total vegetation-covered area. The area with sustainability and stability was 7.8%; the area showing sustainable degradation accounted for 24.5%, however, the variation in trends of 14.0% of the basin was difficult to determine. The area with sustainable degradation was mainly distributed in areas of Bayan Har Mountain, the south of Amne Machin Mountain, Qilian Mountain, Longzhong Loess Plateau, Ningxia Plain, Central Ningxia, the north Qingyang City of Gansu Province, Hetao Plain, the southern foot of Yinshan Mountain, Daqing Mount, the area from Baotou City to Hohhot City, Taiyuan Basin, Linfen Basin, Yuncheng Basin, Jincheng City, Guanzhong Basin, Luoyang City, and Shandong Province. The undetermined area was mainly located in located in Amne Machin Mountain, Huangnan and Gannan Tibetan Autonomous Prefectures, Xining City, central Gansu Province, and central Shaanxi Province. The vegetation condition of the sustainably degraded and undetermined regions is unknown and warrants additional research.
Acknowledgments This research was financially supported by the National Natural Science Foundation of China (41171318), the National Key
Baret, F., et al., 1991. Potentials and limits of vegetation indices for LAI and APAR assessment. Remote Sens. Environ. 35 (2–3), 161–173. Blazkova, S., Beven, K., 2004. Flood frequency estimation by continuous simulation of subcatchment rainfalls and discharges with the aim of improving dam safety assessment in a large basin in the Czech Repulic. Journal of Hydrology 292, 153–172. Cai, B.F., et al., 2009. Advance and evaluation in the long time series vegetation trends research based on remote sensing. J. Remote Sens. 13 (6), 1170–1186 (in Chinese). Chen, Y.H., et al., 2002. The change of NDVI time series based on change vector analysis in China, 1983–1992. J. Remote Sens. 6 (1), 13–18 (in Chinese). Fan, N., et al., 2012. Spatial-temporal dynamic changes of vegetation cover in Lancang river basin during 2001–2010. Resour. Sci. 34 (7), 1222–1231 (in Chinese). Fensholt, R., et al., 2012a. Greenness in semi-arid areas across the globe 1981–2007—an Earth Observing Satellite based analysis of trends and drivers. Remote Sens. Environ. 121, 144–158. Fensholt, R., et al., 2012b. Evaluation of Earth Observation based global long term vegetation trends—comparing GIMMS and MODIS global NDVI time series. Remote Sens. Environ. 119, 131–147. Fensholt, R., et al., 2009. Evaluation of earth observation based long term vegetation trends—inter comparing NDVI time series trend analysis consistency of Sahel from AVHRR GIMMS, Terra MODIS and SPOTVGT data. Remote Sens. Environ. 113 (9), 1886–1898. Fu, X.Y., et al., 2014. An improved indicator of simulated grassland production based on MODIS NDVI and GPP data: a case study in the Sichuan Province, China. Ecol. Indic. 40, 102–108. Gutman, G., et al., 1998. The derivation of the green vegetation fraction from NOAA/AVHRR data for use in numerical weather prediction models. Int. J. Remote Sens. 19 (8), 1533–1543. He, Z., et al., 2012. Spatio-temporal variation of vegetation cover based on SPOT-VGT in the Yellow River Basin. Ecol. Environ. Sci. 21 (10), 1655–1659. Hoaglin, D.C., et al., 2000. Understanding Robust and Exploratory Data Analysis. Wiley, New York, NY. Holben, B.N., 1986. Characteristics of maximum-value composite images from temporal AVHRR data. Int. J. Remote Sens. 7 (11), 1471-1434. Hou, X.Y., et al., 2012. Characteristics of multi-temporal scale variation of vegetation coverage in the Circum Bohai Bay Region, 1999–2009. Acta Ecol. Sin. 32, 297–304. Huete, A., 1988. A soil-adjusted vegetation index (SAVI). Remote Sens. Environ. 25, 295–309. Hurst, H., 1951. Long term storage capacity of reservoirs. Trans. Am. Soc. Civil Eng. 6, 770–799. Kawabata, A., et al., 2001. Global monitoring of the inter annual changes in vegetation activities using NDVI and its relationships to temperature and precipitation. Int. J. Remote Sens. 22, 1377–1382. Kendall, M., 1975. Rank Correlation Methods. Charles Griffin, London. Li, C.H., et al., 2004. Spatio-temporal changes of NDVI and their relations with precipitation and runoff in the Yellow River Basin. Geog. Res. 23 (6), 753–759. Liu, L.L., et al., 2006. Spatial-temporal correlations of NDVI with precipitation and temperature in the Yellow River Basin. Chin. J. Ecol. 25 (5), 477–481 (in Chinese). Lunetta, R.S., et al., 2006. Land-cover change detection using multi-temporal MODIS NDVI data. Remote Sens. Environ. 105 (2), 142–154. Ma, M.G., et al., 2006. Advance in the inter-annual variability of vegetation and its relation to climate based on remote sensing. J. Remote Sens. 10 (3), 421–431. Mandelbrot, B., Wallis, J.R., 1969. Robustness of the rescaled range R/S in the measurement of noncyclic long-run statistical dependence. Water Resour. Res. 5, 967–988. Martínez, B., Gilabert, M.A., 2009. Vegetation dynamics from NDVI time series analysis using the wavelet transform. Remote Sens. Environ. 113, 1823–1842. Matsushita, B., et al., 2007. Sensitivity of the Enhanced Vegetation Index (EVI) and Normalized Difference Vegetation Index (NDVI) to topographic effects: a case Study in High-Density Cypress Forest. Sensors 7, 2636–2651. Milich, L., 2000. GAC NDVI inter annual coefficient of variation (CoV) images: ground truth sampling of the Sahel along north-south transects. Int. J. Remote Sens. 21 (2), 235–260. Muchoney, D., et al., 2000. Application of the MODIS global supervised classification model to vegetation and land cover mapping of Central America. Int. J. Remote Sens. 21 (6–7), 1115–1138. Myneni, R.B., et al., 1997. Increased plant growth in the Northern High Latitudes from 1981 to 1991. Nature 386, 698–702.
126
W. Jiang et al. / Ecological Indicators 51 (2015) 117–126
Peng, J., et al., 2012. Trend analysis of vegetation dynamics in Qinghai-Tibet Plateau using Hurst Exponent. Ecological Indicators 14, 28–39. Piao, S.L., et al., 2001. Dynamic vegetation cover change over last 18 years in China. Quat. Sci. 21 (4), 294–302 (in Chinese). Pouliot, D., et al., 2009. Trends in vegetation NDVI from 1 km AVHRR data over Canada for the period 1985–2006. Int. J. Remote Sens. 30, 149–168. Rigina, O., et al., 1996. Using trend line and principal component analysis to study vegetation changes in Senegal 1986–1999 from AVHRR NDVI 8 km data. Geografisk Tidsskrift 103 (1), 31–42. Sánchez Granero, M.A., Trinidad, J.E., García, P.J., 2008. Some comments on Hurst exponent and the long memory processes on capital markets. Physica A 387, 5543–5551. Sen, P.K., 1968. Estimates of the regression coefficient based on Kendall’s tau. J. Am. Stat. Assoc. 63 (324), 1379–1389. Senay, G.B., et al., 2000. Combining AVHRR-NDVI and landuse data to describe temporal and spatial dynamics of vegetation. For. Ecol. Manage. 128, 83–91. Song, F.Q., et al., 2011. Monitoring and assessment of vegetation variation in Northern Shaanxi based on MODIS/NDVI. Acta Ecol. Sin. 31 (2), 0354–0363. Stow, D., et al., 2004. Remote sensing of vegetation and land-cover change in Arctic Tundra ecosystems. Remote Sens. Environ. 89, 281–308. Sun, H.Y., et al., 1988. Analysis of the vegetation cover change and the relationship between NDVI and environmental factors by using NOAA time series data. J. Remote Sens. 2 (3), 204–210. Sun, R., et al., 2001. Relationship between the fractional vegetation cover change and rainfall in the Yellow River Basin. Acta Geog. Sin. 56 (6), 667–672 (in Chinese). Theil, H., 1950. A rank-invariant method of linear and polynomial regression analysis. I, II and III. Proc. K. Ned. Akad. Wet. 53, 386–392, 521- 525, 1397-1412.
´ I., 2004. Spatial and temporal variability of winter and summer precipitation Toˇsic, over Serbia and Montenegro. Theor. Appl. Climatol. 77, 47–56. Tucker, C.J., et al., 2001. Mean and inter-year variation of growing-season normalized difference vegetation index for the Sahel 1981–1989. Int. J. Remote Sens. 12 (6), 1133–1135. Tucker, C.J., et al., 1985. African land-over classification using satellite data. Science 227, 369–375. Tucker, C.J., et al., 2005. Analysis of Sahelian vegetation dynamics using NOAAAVHRR NDVI data from 1981–2003. J. Arid Environ. 63, 596–614. Wang, D., et al., 2006. Analyze the vegetation cover variation cycle of china from NOAA data. J. Image Graphics 11 (4), 516–520 (in Chinese). Wang, G.G., et al., 2010. Study on the vegetation dynamic change and R/S analysis in the past ten years in Xinjiang. Remote Sens. Technol. Appl. 25 (1), 84–90 (in Chinese). Yang, S.T., et al., 2003. The vegetation cover over last 20 years in the Yellow River Basin. Acta Geog. Sin. 57 (6), 679–684 (in Chinese). Zhang, F., et al., 2010. Study on quantitative method of spatial and temporal dynamic monitoring of vegetation cover in Hulun Buir Grassland. J. Nat. Resour. 25 (10), 1698–1708 (in Chinese). Zhang, Y.C., et al., 2008. Indicating variation of surface vegetation cover using SPOT NDVI in the northern part of North China. Geog. Res. 27 (4), 745–755 (in Chinese). Zhang, Z.M., et al., 2013. Mountain vegetation change quantification using surface landscape metrics in Lancang watershed, China. Ecol. Indic. 31, 49–58. Zhao, Y.S., 2003. Principles and Methods of Analysis of Remote Sensing Applications. Science Press, Beijing, pp. 374–376 (in Chinese).
Contents lists available at ScienceDirect
Ecological Indicators journal homepage: www.elsevier.com/locate/ecolind
Spatio-temporal analysis of vegetation variation in the Yellow River Basin Weiguo Jiang a,b,∗ , Lihua Yuan a,b , Wenjie Wang c,∗ , Ran Cao a,b , Yunfei Zhang a,b , Wenming Shen d a
State Key Laboratory of Earth Surface Processes and Resource Ecology, Beijing Normal University, Beijing 100875, China Key Laboratory of Environmental Change and Natural Disaster, Beijing Normal University, Beijing 100875, China c Chinese Research Academy of Environmental Sciences, Beijing 100012, China d Satellite Environment Center, Ministry of Environmental Protection, Beijing 100097, China b
a r t i c l e
i n f o
Article history: Received 17 April 2014 Received in revised form 8 July 2014 Accepted 21 July 2014 Keywords: Yellow River Basin NDVI Spatial distribution Variation coefficient Trend analysis method Hurst index
a b s t r a c t To understand the variation and patterns of vegetation coverage in the Yellow River Basin, as well as to promote regional ecological protection and maintain ecological construction achievements, MOD13Q1 data at a resolution of 250 m were used to calculate the annual average normalised difference vegetation index (NDVI) in a time series from 2000 to 2010. Using a variation coefficient, a Theil–Sen Median trend analysis, the Mann–Kendall test, and the Hurst index method, this study investigated the temporal and spatial variations of vegetation coverage characteristics of the Yellow River Basin. The results showed that (1) the vegetation coverage of the Yellow River appeared to have an overall trend of high in the southeast and west and low in the northwest; (2) the averaged NDVI of the whole basin fluctuated in a range of 0.3 to 0.4 from 2000 to 2010 (from 2000 to 2004 there were larger variations and these have been growing rapidly since 2005); (3) the NDVI was stable, 73.4% of the vegetation-coverage area fluctuated with a low-to-medium amplitude, while 27.6% of the area varied by a large amplitude; (4) the regions with improved vegetation coverage (62.9%) were far greater than the degraded regions (27.7%), while the sustained invariant area accounted for 9.4% of the total vegetation coverage regions; and (5) 86% of the vegetation-covered area was positively sustainable. The areas with sustainable improvement accounted for 53.7% of the total vegetation coverage area; the invariant area accounted for 7.8%. The area with sustainable degradation was 24.5%; the future variation in trends of the residual (14%) could not be determined. Therefore, continuous attention must be given to the variation in trends of vegetation in the sustainably degraded and underdetermined regions. © 2014 Elsevier Ltd. All rights reserved.
1. Introduction Vegetation is characterised by inter-annual and seasonal variations. As a natural tie connecting atmosphere, water, and soil, vegetation plays a notably important role in soil conservation, atmosphere adjustment, and the maintenance of climatic and whole ecosystem stability (Sun et al., 1988). Because variations in the surface vegetation coverage affect the balance of regional ecosystems, studies on the variation in vegetation coverage is the
∗ Corresponding authors at: Key Laboratory of Environmental Change and Natural Disaster, Beijing Normal University, No.19 Xinjiekouwai Street, Haidia., China. Tel.: +86 10 58802923. E-mail addresses: [email protected] (W. Jiang), [email protected] (W. Wang). http://dx.doi.org/10.1016/j.ecolind.2014.07.031 1470-160X/© 2014 Elsevier Ltd. All rights reserved.
basis of the protection of the ecological environment (Fan et al., 2012; Peng et al., 2012; Zhang et al., 2013). With wide coverage, high temporal resolution rate, and free data, sensors such as National Oceanic and Atmospheric Administration/Advanced Very High Resolution Radiameter (NOAA/AVHRR), Systeme Probatoire d’Observation dela Tarre/VEGETATION (SPOT/VGT), and moderate resolution imaging spectroradiometer (MODIS) can provide large amounts of data for monitoring the variations of vegetation coverage over long time periods (Rigina et al., 1996; Tucker et al., 2005; Ma et al., 2006; Fensholt et al., 2009, 2012a,b; Fu et al., 2014). NDVI is functionally correlated with leaf area index (LAI) and vegetation coverage (Baret et al., 1991; Gutman et al., 1998); the higher the NDVI, the larger the LAI, and the higher the vegetation coverage. Therefore, NDVI can reflect the growth status of surface vegetation and also acts as an effective index for monitoring vegetation variations (Zhao, 2003; Tucker et al., 1985). For over 20 years, NDVI data
118
W. Jiang et al. / Ecological Indicators 51 (2015) 117–126
have been used to analyse spatial distribution characteristics in a large space scale and a long time scale (Tucker et al., 1985; Myneni et al., 1997; Senay et al., 2000; Stow et al., 2004); other methods, such as principal component analysis (Fan et al., 2012), unary linear regression (Piao et al., 2001; Song et al., 2011), variation vector analysis (Chen et al., 2002), Theil–Sen median trend analysis, the Mann–Kendall test (Pouliot et al., 2009; Fensholt et al., 2012a,b), Fourier transformation (Wang et al., 2006), and wavelet analysis method (Martínez and Gilabert, 2009) have also been employed to explore the temporal and spatial variation characteristics of the vegetation coverage. Moreover, correlation analysis has been used to investigate the correlations between inter-annual/annual variations and climate (Fensholt et al., 2012a,b; Kawabata et al., 2001; Blazkova and Beven, 2004; Zhang et al., 2010). The Yellow River Basin is located in arid, semi-arid, and semihumid areas in China. This river is endowed with a diversified climate, severely fluctuating topography, various types of landforms, and rich vegetation types (Liu et al., 2006). In recent years, with the change of climate and the constant intensification of human activities, this basin has exhibited vegetation changes. At present, a number of researchers have explored the vegetation changes in the Yellow River Basin. Sun et al. (2001) employed the NOAA AVHRR of 8 km to investigate the temporal and spatial distribution of vegetation from 1982 to 1999 and reported correlations of the NDVI with precipitation and temperature (Sun et al., 2001; Yang et al., 2003). Also using a NOAA AVHRR of 8 km, Li et al. (2004) studied the spatial distribution, the annual and seasonal variations of NDVI, as well as the inter-annual and annual correlations of NDVI with the precipitation and runoff in the Yellow River Basin from 1982 to 1999 (Li et al., 2004). Liu et al. (2006) used a 1 km of NOAA AVHRR to analyse the relationships of NDVI with temperature and rainfall in the Yellow River Basin. Based on the data of 1 km SPOT VGT, He et al. (2012) used the mean value method and the trend line analysis method to analyse the temporal and spatial distribution, time variation characteristics, and inter-annual variation in trends of the NDVI in the Yellow River Basin (He et al., 2012). Previous research has been primarily based on NOAA AVHRR data with spatial resolution of 1 km/8 km, or the SPOT VGT data with a 1 km spatial resolution; thus, the spatial resolution of the current research is relatively low. For analysing the inter-annual trends of vegetation, the method of linear regression analysis has been used with NDVI time series data. Using this method, vegetation trends are calculated by regression, which can easily be affected by outliers. Few studies using the more robust Theil–Sen trend analysis and Mann–Kendall tests have been conducted to explore the inter-annual trend of vegetation. Few of the studies on vegetation trends in the Yellow River Basin in the past have been focused on the vegetation trends for the future. This may be explained by the difficulty in simulating vegetation trends in the future through mathematical models between the NDVI and its related influencing factors. The Hurst exponent has been used to quantitatively detect the sustainability of time series data, but, to date, it has not been exploited for vegetation trend prediction in the Yellow River Basin. For our study, we choose MOD13Q1 data with a 250 m spatial resolution. Studies have indicated that there is a greater advantage on the spectral and spatial resolution using MODIS-250 m NDVI product data, and these data could provide more precise information on the land surface compared with the SPOT/VGT-1 km NDVI, and NOAA/AVHRR-8 km NDVI data (Muchoney et al., 2000). The data used in this research is more precise when compared with previous studies. The MOD13Q1 data were pre-processed and calculated to obtain the annual average NDVI time series in the Yellow River basin from 2000 to 2010. Next, using a variation coefficient, a Theil–Sen Median trend analysis, the Mann–Kendall test, and the Hurst exponent methods, this study investigated the temporal and
spatial variations, fluctuation characteristics, variation in trends, and sustainability of the vegetation coverage in the Yellow River basin. It is expected that understanding the variation characteristics and patterns of the vegetation coverage in the Yellow River Basin will promote the regional ecological protection and maintain ecological construction achievements. 2. Methods 2.1. Study area Originating from Bayankala Mountain in the Qinghai Province in China, the Yellow River flows through 9 Provinces, including Qinghai, Sichuan, Gansu, Ningxia, the Inner Mongolia Autonomous Region, Shaanxi, Shanxi, Henan, and Shandong. In Kenli County of the Shandong Province, it merges into the Bohai Sea. The Yellow River is 1,900 km long from east to west and 1100 km long from south to north and covers an area of 79.46 × 104 km2 . The river’s geographical coordinates are 39◦ 28 to 41◦ 05 N and 115◦ 25 to 117◦ 30 E under the WGS84/Albers Equal Area Conic projection (Fig. 1). The terrain of Yellow River basin is high in the west and low in the east. The western origination area lies at an average altitude of 4000 m and is composed of a series of mountains; the central region is in an altitude of 1000–2000 m and presents a loess landform with serious soil erosion; and the eastern area is 100 m below sea level and is mainly composed of the alluvial plain of the Yellow River (He et al., 2012). The area of the Yellow River has a continental climate. The south-eastern part belongs to semi-humid climate, the middle part has a semiarid climate, and the northwest part has a subordinate arid climate. The various landforms and complex habitats of the Yellow River Basin create favourable conditions for the development of various vegetation types (Liu et al., 2006); land-use types are mainly woodland, grassland, and farmland. 2.2. Methods 2.2.1. Data sources MODIS NDVI data were sourced from the MODIS vegetation index product data-MOD13Q1 on the website of the United States NASA. The data, collected from February 2000 to December 2010, were in Hdf format with a spatial resolution of 250 m, and a temporal time resolution of 16 days. The data were subjected to format and projection conversions using MODIS Reprojection Tool (MRT), which can be acquired from NASA Land Processes Distributed Active Archive Center (Sioux Falls, South Dakota, U.S.A.). The original Hdf format was transformed into the Geotiff format and the original Sinusoidal projection was transformed into the WGS84/Albers Equal Area Conic projection. The converted data were re-sampled using the adjacent natural method at a resolution of 250 m. The samples obtained were processed using the internationally accepted and commonly used maximum synthesis method to yield monthly NDVI data from 2000 to 2010, to avoid the influences of clouds, atmosphere, and solar altitude angle (Piao et al., 2001; Holben, 1986). Finally, using the mean value method, the annual average NDVI data were acquired to eliminate the influence of extreme yearly abnormal climate on the growth status of vegetation (Zhang et al., 2008). Land use/land cover data for the Yellow River Basin in 2010, at a resolution of 250 m, were provided by the project team, “The remote sensing investigation and assessment on the ten-year changes of the national ecological environment.” The land use types included forest, grassland, farmland, wetland, and artificial surface. 2.2.2. Methods The variation coefficient, Theil–Sen median trend analysis, Mann–Kendall, and Hurst exponent method were used to study the
W. Jiang et al. / Ecological Indicators 51 (2015) 117–126
Fig. 1. Location of the study area.
119
120
W. Jiang et al. / Ecological Indicators 51 (2015) 117–126
vegetation-covered regions of the Yellow River Basin, namely, the spatial distribution, temporal variation characteristics, volatility, variation in trends, and sustainability of the NDVI of the pixelcovered region with NDVI values greater than or equal to 0.1 (Piao et al., 2001). 2.2.2.1. Coefficient of variation. Coefficient of variation is commonly used in vegetation research and is mainly used to reflect the discrete degree of the NDVI data and the inter-annual/annual volatility of vegetation (Tucker et al., 2001; Milich, 2000). It is calculated by: CVNDVI =
NDVI NDVI
,
(1)
where, CVNDVI refers to the coefficient of variation of the NDVI value of each pixel in 2000–2010; NDVI represents the standard deviation of the NDVI value; NDVI is the mean NDVI value. A larger CVNDVI signifies that the NDVI time series shows larger data fluctuation, while a smaller CVNDVI denotes a more stable NDVI time series. 2.2.2.2. Theil–Sen Median trend analysis and the Mann–Kendall test. The Theil–Sen Median trend analysis method can be effectively combined with the Mann–Kendall test. These are important methods for judging the trend of long time series data, and this combination has been gradually used to analyse the long time series of vegetation reflecting the variation in trends of each pixel in a time series (Fensholt et al., 2012a,b; Milich, 2000; Lunetta et al., 2006). The Theil–Sen Median trend analysis is a robust trend statistical method (Theil, 1950; Sen, 1968; Hoaglin et al., 2000) and it calculates the median slopes between all n(n − 1)/2 pair-wise combinations of the time series data. It is based on non-parametric statistics and is particularly effective for the estimation of trends in small series (Hoaglin et al., 2000). The slope of Theil–Sen Median can represent the increase or decrease in the NDVI over the 11 years between 2000 and 2010 on a pixel scale. It is calculated by:
SNDVI = median
NDVIj − NDVIi j−i
,
2000 ≤ i < j ≤ 2010,
where, S =
j=1
i=j+1
sgn NDVIj − NDVIi =
0
s(S)
n−1 n
⎧ S−1 ⎪ , S>0 ⎪ ⎪ ⎪ ⎨ s(S) ⎪ ⎪ S+1 ⎪ ⎪ ⎩
,
S=0 .
,
S<0
(3)
sgn NDVIj − NDVIi ,
⎧ ⎨ 1
,
NDVIj − NDVIi > 0
0
,
NDVIj − NDVIi = 0 ,
−1
,
NDVIj − NDVIi < 0
⎩
2.2.2.3. Hurst exponent. The Hurst exponent is a method for distinguishing sustainability of time series data. It was proposed by British hydrologistHurst (1951), then Mandelbrot and Wallis (1969) improved the theory. This exponent has been widely used in the fields of hydrology, climatology, economics, geology, and geochemistry. Recently, it has been applied in the time series detection of vegetation variations (Fan et al., 2012; Hou et al., 2012; Wang et al., 2010). The primary calculation procedures are as follows (Sánchez Granero et al., 2008):
s (S) =
1 NDVI(t)
NDVI() =
= 1, 2, . . ., n
(4)
t=1
(3) To calculate the accumulated deviation, X(t,) =
t
NDVI(t) − NDVI()
1≤t≤
(5)
t=1
(4) To create the range sequence, R() = max X(t,) − min X(t,) 1≤t≤
1≤t≤
= 1, 2, . . ., n
(6)
(5) To create the standard deviation sequence,
S() =
2 1
NDVI(t) − NDVI()
1/2 = 1, 2, . . ., n
(7)
t=1
(6) To calculate Hurst exponent, R() = (c)H S()
(8)
The value of H is obtained by fitting the equation log(R/S)n = a + H × log(n), using the least squares method, where H is the Hurst exponent. According to previous research (Hurst, 1951; Mandelbrot and Wallis, 1969), the value of the Hurst exponent (H value) can be expanded from 0 to 1 and is valued in three types: when the H value is equal to 0.5, it means that the NDVI time series is a stochastic series without sustainability, indicating that the trend of the time series in the future would be unrelated with that of the study period; when the value is greater than 0.5, it means that the sustainability of the NDVI time series shows the same trend as the NDVI the time series in the future; and when the value is less than 0.5, it refers to the anti-sustainability of the NDVI time series, representing the opposite trend of the NDVI time series in the future.
n (n − 1) (2n + 5) 18
where, NDVIi and NDVIj represent the NDVI values of the pixels i and j; n is the length of the time series; sgn is a sign function; and the
(1) To define the time series NDVI(t) , t = 1, 2, . . ., n. (2) To define the mean sequence of the time series,
(2)
where, SNDVI refers to the Theil–Sen median, and NDVIi and NDVIj represent the NDVI values of years of i and j. In case of SNDVI > 0, the NDVI presents a rising trend, otherwise, the NDVI displays a decreasing trend. The Mann–Kendall test measures the significance of a trend. It is a non-parametric statistical test and it has the advantage that samples do not need to obey certain distributions and is free from ´ 2004). It has been the interference of outliers (Kendall, 1975; Toˇsic, widely applied to analyse the trends and variations at sites with hydrological and meteorological time series. Recently, this test has been used to study of vegetation variations over long time periods (Fensholt et al., 2009). The calculation formula is as follows: it is assumed that {NDVIi } , i = 2000, 2001, . . ., 2010
the statistics of Z is defined as Z =
Z statistic is valued in the range of (−∞, +∞). A given significance level, |Z| > u1–˛/2 , signifies that the times series shows significant variations on the level of ˛. Generally, the value of ˛ is 0.05. In this study, we choose ˛ = 0.05, meaning we measured the significance of the NDVI trend from 2000-2010 on pixel scale at a confidence level of 0.05.
W. Jiang et al. / Ecological Indicators 51 (2015) 117–126
121
Fig. 2. Spatial distribution of average NDVI in the Yellow River Basin from 2000 to 2010.
3. Results 3.1. The characteristics of the spatial distribution of vegetation coverage Fig. 2 shows the spatial distribution of average NDVI using the average NDVI for the 11 years from 2000 to 2010 and the spatial distribution characteristics of the NDVI in the Yellow River Basin. As shown in this figure, the NDVI spatial distribution in the Yellow River basin is high in the west and south-east, while it is low in the north. The woodlands, herbaceous wetlands, and crops have a high NDVI value, while the grassland etc. have a low NDVI value. The western area, located at a high altitude, is primarily composed of forest, grassland, and herbaceous wetlands and shows a high NDVI value. The southeastern part has a semi humid climate and has extensive forests and crops, therefore, the NDVI is relatively high and significant. The northern area is mainly comprised of the Erdos Plateau, the Loess Plateau, the Hetao Plain, and the Ningxia Plain. The Erdos and Loess Plateaus have low NDVI values because they are less covered by vegetation. However, the Hetao and Ningxia Plains belong to the alluvial plain of the Yellow River and are suitable for planting crops. The annual NDVI of the two plains is 0.3–0.4. The NDVIs of the non-vegetated areas (red part) are less than 0.1. These areas are mainly lake, glacier, bare rock, reservoir, desert, including the sandy lands and desert in the north of the basin. The results of the statistical classification of the average NDVI values over the 11 years of the study showed that non-vegetated area with a NDVI value under 0.1 accounted for 1.7% of the total basin area; the area with low NDVI values (0.1–0.4) accounted for 53.3%; the area with high NDVI values (above 0.1–0.4) accounted for 19.5%; the area with NDVI values from 0.5 to 0.6 accounted for 10.5%; the area with NDVI values above 0.6 accounted for 5%, and the areas with vegetation coverage were 98.3% of the total basin area. 3.2. The temporal variation characteristics of the vegetation coverage To study the temporal variation characteristics of the NDVI of the Yellow River Basin, the average NDVI of the pixels in the vegetationcovered areas in the images of annual average NDVI were used to represent the overall situation of the vegetation in the corresponding year. This study mainly analysed the NDVI variations of the vegetation-covered area. As shown in Fig. 3, the annual average
Fig. 3. Inter-annual variation of the NDVI in the Yellow River Basin from 2000 to 2010.
NDVI of the vegetation-covered area in the Yellow River Basin fluctuated from 0.3 to 0.4. The NDVI of the vegetation-covered area exhibited larger variations from 2000 to 2004. However, the variations did not show a significant trend. Since 2005, the NDVI of the vegetation-covered areas have been growing rapidly. 3.3. The volatility of the vegetation coverage According our calculations, the CVNDVI could be classified into 5 grades, including high volatility (CVNDVI ≥ 0.20), relatively high volatility (0.15 ≤ CVNDVI < 0.20), medium volatility (0.10 ≤ CVNDVI < 0.15), relatively low volatility (0.05 ≤ CVNDVI < 0.10), and low volatility (CVNDVI < 0.05). Table 1 shows the statistical classification results of the NDVI coefficient of variation on the pixel scale. As shown, 73.4% of the vegetationcovered areas have medium-high volatility, while merely 27.6% of the areas present high volatility. Therefore, it can be concluded that vegetation coverage is stable in the Yellow River Basin. Table 1 Coefficient of variation of NDVI in the Yellow River Basin. CVNDVI
Volatility degree
Area percentage (%)
≥0.20 0.15 ≤ CVNDVI < 0.20 0.10 ≤ CVNDVI < 0.15 0.05 ≤ CVNDVI < 0.10 <0.05
Low volatility Relatively low volatility Medium volatility Relatively high volatility High volatility
37.42 40.41 32.55 54.67 39.89
122
W. Jiang et al. / Ecological Indicators 51 (2015) 117–126
Fig. 4. Spatial distribution of the coefficient of variation of the inter-annul NDVI in the Yellow River Basin from 2000 to 2010.
Fig. 4 illustrates the spatial distribution of the coefficient of variation in the Yellow River Basin. Areas with high and relatively high volatility are mainly distributed in the Bayan Har Mountain, west of Amne Machin Mountain, Xining City, Lanzhou City, Longzhong Loess Plateau, Tengger Desert, Loess Plateau in central Ningxia, Ningxia Plain, Hedong Sand-land, Table Mountain, Langshan Mountain, the southern foot of Yinshan Mountain, Hetao Plain, Wula Mountain, the area from Baotou city to Hohhot City, the Daqing Mount area, Loess Plateau at the junction of Shanxi and Shaanxi Provinces, and the Taiyuan basin. The areas with medium volatility are mainly distributed in central Qinghai Province, east of Amne Machin Mountain, Sichuan Province, central Gansu Province, the junction of Gansu and Shaanxi Provinces, northern Shanxi Province, eastern Henan Province, and Shandong Province. The areas with medium volatility are mainly distributed in the Mu Us sand-land in the east of the Hedong sand-land, southern Shaanxi Province, southern Shanxi Province, and western Henan Province. The areas with low volatility are scattered along the basin. 3.4. The variation in trends of vegetation coverage The variation in trends of each pixel can be effectively characterised using the Theil–Sen Median trend analysis and the Mann–Kendall test to reflect the spatial distribution characteristics of the variation in the trend of the NDVI in the Yellow River Basin. Because areas with a SNDVI of 0 strictly do not exist, we made the following classifications according to the real conditions of the SNDVI of the NDVI. Areas with an SNDVI from −0.0005 to 0.0005 are classified as stable areas, areas with an SNDVI greater than or equal to 0.0005 are classified as improved areas, and areas with a SNDVI smaller than 0.0005 are classified as degraded areas. Moreover, significance test results of the Mann–Kendall test, at the confidence level of 0.05, are classified as significant variations (Z > 1.96 or Z < −1.96) or insignificant variations (−1.96 ≤ Z ≤ 1.96). By integrating the classification results of the Theil–Sen median trend analysis and the Mann–Kendall test, it is comparable to the data of trend variations of the NDVI. The results were categorised into five classes (Table 2). Table 2 shows the areas with vegetation coverage improvement, the areas with stable vegetation coverage (or without significant vegetation coverage variations), and the areas with vegetation degradation, which account for 62.9%, 9.4%, and 27.7%, respectively.
Table 2 Trend of the NDVI in the Yellow River Basin. SNDVI
Z
NDVI trend
Area percentage (%)
≥0.0005 ≥0.0005 −0.0005–0.0005 <−0.0005 <−0.0005
≥1.96 −1.96–1.96 −1.96–1.96 −1.96–1.96 <−1.96
Significantly improved Slightly improved Stable Slightly improved Severely degraded
32.8 20.1 9.4 21.9 5.8
Note: the pixels with SNDVI in −0.0005 and 0.0005 and Z > 1.96 or Z < −1.96 are classified as the stable.
As shown in Fig. 5, the area with improved surface vegetation is far larger than the area with degraded surface vegetation of the Yellow River Basin from 2000 to 2010. The areas with significantly improved vegetation coverage are mainly distributed in southern Gansu Province, northern Shaanxi Province, central and northern parts of Erdos City, Langshan Mountain, and western Shanxi Province, and both sides of the Yellow River Basin in Qinghai Province. The areas with slightly improved vegetation coverage are mainly distributed in the central areas of the junction of Sichuan Province, Gansu Province, and Qinghai Province, and Lvliang Mountain. The areas with stable vegetation coverage show a scattered distribution and the areas with slightly degraded vegetation are mainly located in Bayan Har Mountain, Amne Machin Mountain, Qilian Mountain, the area from Baotou City to Hohhot City, the southern foot of Yinshan Mountain, Ziwu Ridge, the north and west of Qingyang City in the Gansu Province, as well as the forestcovered areas in the west and south of Shaanxi Province. Areas with severe degradation are mainly observed in the Longzhong Loess Plateau, Hetao Plain, Taiyuan Basin, Daqing Mount, Jincheng City, and the two ends of the Guanzhong Basin. Areas with both slight and severe degradation mainly lie in Ningxia Plain, the west of Erdos City, Linfen Basin, Yuncheng Basin, Luoyang City, and the Shandong Province. 3.5. The sustainability of the vegetation coverage variations As shown in Fig. 6, the Hurst exponent of the NDVI of the Yellow River Basin is 0.5 on average. The regions with Hurst exponents lower than 0.5, which were sustainable, accounts for 14.0% of the total basin area. Those higher than 0.5, which were unsustainable,
W. Jiang et al. / Ecological Indicators 51 (2015) 117–126
123
Fig. 5. Trends of inter-annual NDVI change in the Yellow River Basin from 2000 to 2010.
Fig. 6. Sustainability of inter-annual NDVI change in the Yellow River Basin from 2000 to 2010.
account for 86.0% of the total basin area. There is little area with Hurst exponent equal to 0.5. This result suggests that the NDVI of the vegetation-covered areas is positively sustainable. To reveal the variation in trends and the sustainability of the vegetation, the data obtained using the Theil–Sen Median trend analysis and the Mann–Kendall test were superimposed with the Hurst exponent results to yield the coupled information of variation in trends and sustainability (Fig. 7). The coupling results were analysed on 6 cases: (1) sustainability and severe degradation; (2) sustainability and slight degradation; (3) sustainability and stability; (4) sustainability and slight improvement; (5) sustainability and significant improvement; and (6) undetermined variation in trends in the future. The latter includes five combinations, namely, unsustainability and severe degradation, unsustainability and slight degradation, unsustainability and stability, unsustainability and slight improvement, and unsustainability and significant improvement. Thus, the future variation in trends is difficult to determine.
As suggested by Fig. 7 and Table 3, the area showing sustainability and improvement accounts for 53.7% and is mainly distributed in the central and northern Shaanxi Province, southeastern Gansu Province, central Shanxi Province, and the east and north of Langshan Mountain and Erdos City. The area with sustainability and stability accounts for 7.8% and is primarily scattered in Inner Mongolia, Ningxia, Gansu, Qinghai, Shanxi, and Henan Provinces. The area with sustainability and degradation accounts for 24.5% and is mainly distributed in Bayan Har Mountain, south of Amne Machin Mountain, Qilian Mountain, Longzhong Loess Plateau, Ningxia Plain, Central Ningxia, the north of Qingyang City in the Gansu Province, Hetao Plain, the southern foot of Yinshan Mountain, Daqing Mount, the area from Baotou City to Hohhot City, Taiyuan Basin, Linfen Basin, Yuncheng Basin, Jincheng City, Guanzhong Basin, Luoyang City, and Shandong Province. The variation in trends of 14.0% of the basin is difficult to determine and these areas are mainly located in Amne Machin Mountain, Huangnan and Gannan Tibetan Autonomous Prefectures, Xining City, central
124
W. Jiang et al. / Ecological Indicators 51 (2015) 117–126
Fig. 7. Spatial distribution of the NDVI dynamics based on trends and the Hurst index. Table 3 Statistical result of trends and the Hurst index. SNDVI
Z
H
Variation types
Area percentage
<−0.0005 <−0.0005 −0.0005–0.0005 ≥0.0005 ≥0.0005 –
≥1.96 −1.96–1.96 −1.96–1.96 −1.96–1.96 <−1.96 –
>0.5 >0.5 >0.5 >0.5 >0.5 <0.5
Sustainability and severe degradation Sustainability and slight degradation Sustainability and stability Sustainability and slight improvement Sustainability and significantly improvement Undetermined future variation in trends
5.7 18.8 7.8 23.5 30.2 14
Gansu Province, and central Shaanxi Province. The vegetation conditions of the continued degradation and undetermined regions calls will require additional investigation.
4. Discussion The Yellow River Basin covers three steps of topography in China, and its terrain is highly variable and complex; however, the NDVI is rarely affected by topography, which is why it is possible to enhance its ability to respond to vegetation variation (Matsushita et al., 2007). Major land use/land cover types in the Yellow River Basin are grassland, farmland and forest (mainly deciduous), so the NDVI applied in this case study avoided asymptotic (saturated) signals over high biomass conditions (Huete, 1988). We selected the NDVI as the vegetation proxy and for this case study is reasonable; 250 m MODIS-NDVI data can provide more precise information regarding land surface than AVHRR-NDVI and VGTNDVI data, therefore, vegetation information was more accurately reflected by the MODIS-NDVI. The Theil + Sen Median Slope Analysis, together with the Mann–Kendall test is a new method for examining the trend analysis of NDVI time series data. This combination analysis is more advantageous than linear regression and correlation analysis and its main advantages include: data used does not need to obey a particular distribution, strong ability to avoid error and significance level testing also has a solid basis in statistical theory (Cai et al., 2009),
Therefore, this case study applied this combination, which made the examination of the vegetation trends in the Yellow River Basin more scientific and credible compared with past studies. Overlaying the results of the trend analysis and the results of the Hurst exponent calculations further quantified the future sustainability of the NDVI trends. Improving upon past studies on vegetation variations in the Yellow River Basin, this study was able to quantify future vegetation trends by overlaying the results of the trend analysis and the results of the Hurst exponent calculations. However, the Hurst exponent could not predict how long the sustainability of predicted vegetation trends would continue into the future, it was more accurate in determining the duration of the vegetation trends. There is an urgent need to develop methods to investigate the trend duration. Unfortunately, we were unable to find the methods needed to achieve this goal. Based on MODIS-250 m NDVI data from the Yellow River (2000 to 2010), in this case study applied a variation coefficient, the Theil–Sen Median trend analysis, the Mann–Kendall test, and the Hurst exponent method to investigate the temporal and spatial variations, fluctuation characteristics, variation in trends, and sustainability of the vegetation characteristics in the Yellow River Basin. However, more detailed work is necessary to analyse the vegetation variations of many vegetation types and to explore the cause of vegetation variations. The Yellow River Basin covers a large scale (79.46 × 104 km2 ) and various vegetation types are distributed
W. Jiang et al. / Ecological Indicators 51 (2015) 117–126
throughout the Basin. Different vegetation types demonstrate different inter-annual variations, therefore, it is necessary to analyse the variations of the different vegetation types. The variations of the NDVI were affected by climate change, human perturbations, and national/local policies; therefore, the driving forces associated with vegetation variations should be explored. Such explorations would be a great help in fully understanding vegetation variations, and would aid local government in adopting policies or projects that promote vegetation restoration.
125
Technology Support Program (2012BAH32B03 and 2012BAH33B05) and the Remote Sensing Investigation and Assessment Project for Decade-change of the National Ecological Environment (2000–2010). The authors sincerely thank Dr. Wang Mengjie and Dr. Tao Liangliang from Beijing Normal University, for the help in IDL programming and support in the Mann–Kendall test programming and Gao Mingliang from Capital Normal University, for providing the IDL program for the maximum synthesis algorithm.
5. Conclusions References In this study, NDVI time series data from 2000 to 2010 were obtained based on MOD13Q1 data at a resolution of 250 m. Variation coefficient, Theil–Sen Median trend analysis, the Mann–Kendall test, and the Hurst index method were used to investigate the temporal and spatial variations of the vegetation coverage characteristics of Yellow River Basin. (1) Regarding spatial distribution: the vegetation coverage of the Yellow River had an overall trend of high in the southeast and west while being low in northwest; the woodland, herb wetland, and crop areas showed high NDVI values, while the grassland and other types presented low NDVI values. (2) Regarding temporal variations: the annual average NDVI of vegetation coverage fluctuated in a range from 0.3 to 0.4 in 2000–2010. The NDVI in 2000–2004 exhibited larger variations, however, the variations did not present a significant trend. Since 2005, the NDVI has grown rapidly. (3) Regarding the volatility of the NDVI: the vegetation coverage in the Yellow River Basin was stable with 73.4% of the vegetationcovered areas fluctuating in a low-to-medium range, while 27.6% of the areas varied with a large amplitude. (4) Regarding the variation in trends of the NDVI: the regions with improved vegetation coverage were far greater than the degraded regions in the Yellow River basin in 2000–2010. The improved and degraded regions accounted for 62.9% and 27.7% of the total vegetation-covered region, respectively, while the rest, 9.4%, was undetermined. (5) The Hurst exponent-data analysis suggested that the NDVI of 86% of the vegetation-covered area was positively sustainable, namely, the NDVI showed strong sustainability. According to the superimposed variation in trends and sustainability data, the area showing sustainable improvement accounted for 53.7% of the total vegetation-covered area. The area with sustainability and stability was 7.8%; the area showing sustainable degradation accounted for 24.5%, however, the variation in trends of 14.0% of the basin was difficult to determine. The area with sustainable degradation was mainly distributed in areas of Bayan Har Mountain, the south of Amne Machin Mountain, Qilian Mountain, Longzhong Loess Plateau, Ningxia Plain, Central Ningxia, the north Qingyang City of Gansu Province, Hetao Plain, the southern foot of Yinshan Mountain, Daqing Mount, the area from Baotou City to Hohhot City, Taiyuan Basin, Linfen Basin, Yuncheng Basin, Jincheng City, Guanzhong Basin, Luoyang City, and Shandong Province. The undetermined area was mainly located in located in Amne Machin Mountain, Huangnan and Gannan Tibetan Autonomous Prefectures, Xining City, central Gansu Province, and central Shaanxi Province. The vegetation condition of the sustainably degraded and undetermined regions is unknown and warrants additional research.
Acknowledgments This research was financially supported by the National Natural Science Foundation of China (41171318), the National Key
Baret, F., et al., 1991. Potentials and limits of vegetation indices for LAI and APAR assessment. Remote Sens. Environ. 35 (2–3), 161–173. Blazkova, S., Beven, K., 2004. Flood frequency estimation by continuous simulation of subcatchment rainfalls and discharges with the aim of improving dam safety assessment in a large basin in the Czech Repulic. Journal of Hydrology 292, 153–172. Cai, B.F., et al., 2009. Advance and evaluation in the long time series vegetation trends research based on remote sensing. J. Remote Sens. 13 (6), 1170–1186 (in Chinese). Chen, Y.H., et al., 2002. The change of NDVI time series based on change vector analysis in China, 1983–1992. J. Remote Sens. 6 (1), 13–18 (in Chinese). Fan, N., et al., 2012. Spatial-temporal dynamic changes of vegetation cover in Lancang river basin during 2001–2010. Resour. Sci. 34 (7), 1222–1231 (in Chinese). Fensholt, R., et al., 2012a. Greenness in semi-arid areas across the globe 1981–2007—an Earth Observing Satellite based analysis of trends and drivers. Remote Sens. Environ. 121, 144–158. Fensholt, R., et al., 2012b. Evaluation of Earth Observation based global long term vegetation trends—comparing GIMMS and MODIS global NDVI time series. Remote Sens. Environ. 119, 131–147. Fensholt, R., et al., 2009. Evaluation of earth observation based long term vegetation trends—inter comparing NDVI time series trend analysis consistency of Sahel from AVHRR GIMMS, Terra MODIS and SPOTVGT data. Remote Sens. Environ. 113 (9), 1886–1898. Fu, X.Y., et al., 2014. An improved indicator of simulated grassland production based on MODIS NDVI and GPP data: a case study in the Sichuan Province, China. Ecol. Indic. 40, 102–108. Gutman, G., et al., 1998. The derivation of the green vegetation fraction from NOAA/AVHRR data for use in numerical weather prediction models. Int. J. Remote Sens. 19 (8), 1533–1543. He, Z., et al., 2012. Spatio-temporal variation of vegetation cover based on SPOT-VGT in the Yellow River Basin. Ecol. Environ. Sci. 21 (10), 1655–1659. Hoaglin, D.C., et al., 2000. Understanding Robust and Exploratory Data Analysis. Wiley, New York, NY. Holben, B.N., 1986. Characteristics of maximum-value composite images from temporal AVHRR data. Int. J. Remote Sens. 7 (11), 1471-1434. Hou, X.Y., et al., 2012. Characteristics of multi-temporal scale variation of vegetation coverage in the Circum Bohai Bay Region, 1999–2009. Acta Ecol. Sin. 32, 297–304. Huete, A., 1988. A soil-adjusted vegetation index (SAVI). Remote Sens. Environ. 25, 295–309. Hurst, H., 1951. Long term storage capacity of reservoirs. Trans. Am. Soc. Civil Eng. 6, 770–799. Kawabata, A., et al., 2001. Global monitoring of the inter annual changes in vegetation activities using NDVI and its relationships to temperature and precipitation. Int. J. Remote Sens. 22, 1377–1382. Kendall, M., 1975. Rank Correlation Methods. Charles Griffin, London. Li, C.H., et al., 2004. Spatio-temporal changes of NDVI and their relations with precipitation and runoff in the Yellow River Basin. Geog. Res. 23 (6), 753–759. Liu, L.L., et al., 2006. Spatial-temporal correlations of NDVI with precipitation and temperature in the Yellow River Basin. Chin. J. Ecol. 25 (5), 477–481 (in Chinese). Lunetta, R.S., et al., 2006. Land-cover change detection using multi-temporal MODIS NDVI data. Remote Sens. Environ. 105 (2), 142–154. Ma, M.G., et al., 2006. Advance in the inter-annual variability of vegetation and its relation to climate based on remote sensing. J. Remote Sens. 10 (3), 421–431. Mandelbrot, B., Wallis, J.R., 1969. Robustness of the rescaled range R/S in the measurement of noncyclic long-run statistical dependence. Water Resour. Res. 5, 967–988. Martínez, B., Gilabert, M.A., 2009. Vegetation dynamics from NDVI time series analysis using the wavelet transform. Remote Sens. Environ. 113, 1823–1842. Matsushita, B., et al., 2007. Sensitivity of the Enhanced Vegetation Index (EVI) and Normalized Difference Vegetation Index (NDVI) to topographic effects: a case Study in High-Density Cypress Forest. Sensors 7, 2636–2651. Milich, L., 2000. GAC NDVI inter annual coefficient of variation (CoV) images: ground truth sampling of the Sahel along north-south transects. Int. J. Remote Sens. 21 (2), 235–260. Muchoney, D., et al., 2000. Application of the MODIS global supervised classification model to vegetation and land cover mapping of Central America. Int. J. Remote Sens. 21 (6–7), 1115–1138. Myneni, R.B., et al., 1997. Increased plant growth in the Northern High Latitudes from 1981 to 1991. Nature 386, 698–702.
126
W. Jiang et al. / Ecological Indicators 51 (2015) 117–126
Peng, J., et al., 2012. Trend analysis of vegetation dynamics in Qinghai-Tibet Plateau using Hurst Exponent. Ecological Indicators 14, 28–39. Piao, S.L., et al., 2001. Dynamic vegetation cover change over last 18 years in China. Quat. Sci. 21 (4), 294–302 (in Chinese). Pouliot, D., et al., 2009. Trends in vegetation NDVI from 1 km AVHRR data over Canada for the period 1985–2006. Int. J. Remote Sens. 30, 149–168. Rigina, O., et al., 1996. Using trend line and principal component analysis to study vegetation changes in Senegal 1986–1999 from AVHRR NDVI 8 km data. Geografisk Tidsskrift 103 (1), 31–42. Sánchez Granero, M.A., Trinidad, J.E., García, P.J., 2008. Some comments on Hurst exponent and the long memory processes on capital markets. Physica A 387, 5543–5551. Sen, P.K., 1968. Estimates of the regression coefficient based on Kendall’s tau. J. Am. Stat. Assoc. 63 (324), 1379–1389. Senay, G.B., et al., 2000. Combining AVHRR-NDVI and landuse data to describe temporal and spatial dynamics of vegetation. For. Ecol. Manage. 128, 83–91. Song, F.Q., et al., 2011. Monitoring and assessment of vegetation variation in Northern Shaanxi based on MODIS/NDVI. Acta Ecol. Sin. 31 (2), 0354–0363. Stow, D., et al., 2004. Remote sensing of vegetation and land-cover change in Arctic Tundra ecosystems. Remote Sens. Environ. 89, 281–308. Sun, H.Y., et al., 1988. Analysis of the vegetation cover change and the relationship between NDVI and environmental factors by using NOAA time series data. J. Remote Sens. 2 (3), 204–210. Sun, R., et al., 2001. Relationship between the fractional vegetation cover change and rainfall in the Yellow River Basin. Acta Geog. Sin. 56 (6), 667–672 (in Chinese). Theil, H., 1950. A rank-invariant method of linear and polynomial regression analysis. I, II and III. Proc. K. Ned. Akad. Wet. 53, 386–392, 521- 525, 1397-1412.
´ I., 2004. Spatial and temporal variability of winter and summer precipitation Toˇsic, over Serbia and Montenegro. Theor. Appl. Climatol. 77, 47–56. Tucker, C.J., et al., 2001. Mean and inter-year variation of growing-season normalized difference vegetation index for the Sahel 1981–1989. Int. J. Remote Sens. 12 (6), 1133–1135. Tucker, C.J., et al., 1985. African land-over classification using satellite data. Science 227, 369–375. Tucker, C.J., et al., 2005. Analysis of Sahelian vegetation dynamics using NOAAAVHRR NDVI data from 1981–2003. J. Arid Environ. 63, 596–614. Wang, D., et al., 2006. Analyze the vegetation cover variation cycle of china from NOAA data. J. Image Graphics 11 (4), 516–520 (in Chinese). Wang, G.G., et al., 2010. Study on the vegetation dynamic change and R/S analysis in the past ten years in Xinjiang. Remote Sens. Technol. Appl. 25 (1), 84–90 (in Chinese). Yang, S.T., et al., 2003. The vegetation cover over last 20 years in the Yellow River Basin. Acta Geog. Sin. 57 (6), 679–684 (in Chinese). Zhang, F., et al., 2010. Study on quantitative method of spatial and temporal dynamic monitoring of vegetation cover in Hulun Buir Grassland. J. Nat. Resour. 25 (10), 1698–1708 (in Chinese). Zhang, Y.C., et al., 2008. Indicating variation of surface vegetation cover using SPOT NDVI in the northern part of North China. Geog. Res. 27 (4), 745–755 (in Chinese). Zhang, Z.M., et al., 2013. Mountain vegetation change quantification using surface landscape metrics in Lancang watershed, China. Ecol. Indic. 31, 49–58. Zhao, Y.S., 2003. Principles and Methods of Analysis of Remote Sensing Applications. Science Press, Beijing, pp. 374–376 (in Chinese).