Introduction

The Yangtze and Yellow river basins are considered to be the cradles of Chinese civilization, and together with the Lantsang river (known as the Mekong outside China), originate in the center of the Tibet Plateau. The plateau covers an area of 312,930 km2, has an average altitude of 4,200 m, and provides 25, 49, and 15 % of the water for the Yangtze, Yellow, and Lantsang rivers, respectively (Qin et al. 2008; Zhao 2011; Fig. 1 and S1). Due to its importance to water resources in China and South Asia, this region is known as China’s or Asia’s Water Tower. The ecological and environmental evolution of this area can be related to global climate change (Arnell 1999) and to the environmental security of South Asia and western China (Gautam 2010; Qian et al. 2010), and consequently, has attracted the attention of the public, in addition to researchers, government officials, and international organizations (Shen and Tan 2012). In addition, this region has the largest wetland of China, and with its high altitude and great biodiversity, these fragile ecosystems are highly sensitive to climate change (Tang et al. 2006; Fan et al. 2010; Li et al. 2011, 2012) and human activities (Chen 2005; Zhao and Zhou 2005).

Fig. 1
figure 1

Location map showing the source areas of the Yellow, Yangtze, and Lantsang rivers, and the locations of meteorological stations (black circles). Their locations in China are shown in Fig. S1 of the supplemental material

Full size image

Over recent decades, desertification has become an important issue in this region with its effects including grassland degradation (Li 1997; Wang et al. 2001; Wang and Cheng 2001; Yan et al. 2003; Liu 2007), decreased biodiversity (Wen et al. 2011), wetland degradation (Zheng et al. 2012), and sandy desertification (Fang et al. 1998; Han et al. 2004; Feng et al. 2005; Li et al. 2006) occurring in different periods. For instance, Liu et al. (2008) suggested that from the late 1970s to the early 2000s, severe grassland degradation occurred in this region. In addition, between the 1980s and the late 1990s, severe sandy desertification occurred in the region (Zou et al. 2002), and between 1975 and 2005, the areas of sandy desertification in the Yangtze river basin increased continuously (Dong et al. 2012), and from the mid-1970s to the early 2000s the supply of water from rivers in the region showed a continuous decline (Li et al. 2010).

In 2005, with the aim of combating desertification and land degradation, the Chinese government initiated a large-scale ecological migration program in the source regions of the Yellow, Yangtze, and Lantsang rivers; i.e., the Ecological Protection and Restoration Program (EPRP) (Qinghai Provincial Government 2005; Chen et al. 2007; Wang et al. 2010). The key measures of this program included reform of the traditional agricultural and animal husbandry practices, limiting the scale of herding in the region, removal of the bulk of the population from regions with fragile ecological environments, and to propose legislation for procuring regional environments (Qinghai Provincial Government 2007a, b). However, the effects of this program and the impact of climate change on ecological restoration in this region remain poorly understood. Therefore, in this study, we combine satellite-derived vegetation indices, field surveys, and surface meteorological data to evaluate the effects of the EPRP and the influence of climate change on desertification in this region. The main aims of this paper are to determine whether desertification has occurred over the past decade and to assess the roles of climate change and human activities on desertification processes in this region.

Methods

Satellite and meteorological data

Moderate-resolution imaging spectroradiometer (MODIS) surface reflectance daily L2G global 250 m data (MOD09GQ, Collection 5), the Landsat TM/ETM+ data, and meteorological data (provided by China National Meteorological Administration) from 2000 to 2012 were used in this study. The MODIS data (h26v05, h25v05) covering the study area were obtained from NASA (National Aeronautics and Space Administration) and can be accessed at http://reverb.echo.nasa.gov/reverb. MODIS images with cloud cover and pixels with values outside the valid range specified by the MOD09GQ.005 quality control descriptions were excluded from the analysis. The MODIS Reprojection Tool (MRT) was used to mosaic every two scenes and convert the sinusoidal projection into a WGS 1984 Albers projection. Subsets covering only the study area were then extracted from each image, and values were scaled between −1.0 and 1.0 by ENVI 4.7 to acquire daily normalized difference vegetation index (NDVI) data. Lastly, the maximum value composites (MVC) were processed by month to obtain the monthly NDVI.

The Landsat TM data with a spatial resolution of 30 m were obtained from the US Geological Survey (USGS) and can be accessed at http://earthexplorer.usgs.gov/. Landsat TM/ETM+ scenes obtained in 2000, 2005, and 2010 were selected for further analysis of the variations in trends in mobile sandy areas using time series obtained from May to September (MJJAS) images when vegetation cover reaches its maximum each year. More details of our analysis of the trends in the mobile sandy lands are provided in section S2 of the supplemental material. Due to the heavy cloud cover on the Tibetan Plateau, it was necessary to replace several images with an image from the same month of the previous or following year, but this had no significant effect on the interpretation of the trends in the mobile sandy lands because here we only compare trends in mobile sandy lands from 2000, 2005, and 2010. Radiometric calibration and atmospheric corrections were performed using ENVI4.7. Histogram normalization was applied to yield normalized radiometric data and so remove or normalize the reflectance variation between images acquired at different times. Scenes were mosaicked to obtain intact images. For comparative analysis among the images, image-to-image registration was used, and images were geo-referenced to the WGS 1984 Albers Equal-Area Conic coordinate system.

The meteorological data, which include monthly precipitation, temperature, and mean wind speed, were obtained from the China Meteorological Administration. Due to the harsh environments and arduous work conditions, there are only a few meteorological stations scattered across this region, and they provide insufficient data for high-resolution spatial analysis. Therefore, in each source area (Yellow, Yangtze, and Lantsang rivers), we integrated the meteorological data from all stations to assess the extent of regional climate change between 2000 and 2012.

Data processing

NDVI data processing

Variations in NDVI values usually represent changes in the vigor and photosynthetic capacity (or greenness) of the vegetation canopy (Chen and Brutsaert 1998), whose values are positively correlated with vegetation coverage with the desertification status of the region (e.g., Wang et al. 2003; Piao et al. 2005). Here, the maximum NDVI of the growing season (MJJAS) was defined as the annual NDVI values (e.g., Pettorelli et al. 2005; Verbyla 2008) and was employed in further analysis.

NDVI time series were evaluated using a linear regression model to obtain the temporal trend of every pixel. This method fits a linear equation to the NDVI as a function of the variable YEAR, to acquire the slope of the images. For every pixel, the linear relationship between NDVI and YEAR was determined using the ordinary least squares (OLS) method:

$$ {\text{NDVI}} = {\text{SLOPE}} \times {\text{YEAR}} + b $$
(1)
$$ {\text{SLOPE}} = \frac{{n \times \sum\nolimits_{i = 1}^{n} {i \times {\text{NDVI}}_{i} } - \left( {\sum\nolimits_{i = 1}^{n} i } \right)\left( {\sum\nolimits_{i = 1}^{n} {{\text{NDVI}}_{i} } } \right)}}{{n \times \sum\nolimits_{i = 1}^{n} {i^{2} } - \left( {\sum\nolimits_{i = 1}^{n} i } \right)^{2} }} $$
(2)

here, SLOPE represents the NDVI trend, with values greater than zero indicating an increasing NDVI and high vegetation productivity (e.g., Fang et al. 2004); conversely, values less than zero suggest a decreasing NDVI trend and low vegetation productivity in the region (see section S3 of the supplemental materials).

Actual and potential net primary productivity data processing

Net primary productivity (NPP) estimations have been used previously to investigate the magnitude and geographical distribution of primary productivity (Field et al. 1998; Cramer et al. 1999; Grosso et al. 2008) and have been successfully used to assess land degradation at various spatial and temporal scales (Wessels et al. 2003; Prince et al. 2009). More recently, NPP models have been widely used to quantitatively assess the relative roles of climatic and human activities in desertification or land degradation (Evans and Geerken 2004; Potter 2004; Wessels et al. 2004; Tao et al. 2005).

Here, we employ the Carnegie–Ames–Stanford Approach (CASA) model to evaluate the actual NPP variations in the region. In the CASA model, the NPP is determined mainly from the product of the absorbed photosynthetically active radiation (APAR) and light use efficiency (ε):

$$ {\text{NPP}}(x,t) = {\text{APAR}}(x,t) \times \varepsilon (x,t) $$
(3)

The data processing related to NPP outputs includes mean monthly temperature, evapotranspiration, sunshine duration, vegetation index, vegetation type, and soil type, about which additional information is provided in section S4 of the supplemental materials.

We also employ a similar method to the CASA model to assess the regional potential NPP, and more details of these methods are provided in the supplemental materials. Similarly to the NDVI slope, the temporal trends of the potential NPP slope can be taken to be representative of the role of climate change in desertification in the region, with values greater than zero indicating an increasing potential NPP trend and self-restoration of vegetation. We also employ the differences between the potential and actual NPPs (i.e., the residuals) to evaluate the impacts of human activities on the regional ecology and environment. Following previous studies (Evans and Geerken 2004), residuals with values of >0 and <0 were used to indicate whether human activities had negative or positive effects, respectively, on primary productivity. To further analyze the contribution of human activities to desertification in the region, the temporal trends in the residuals were considered. Negative trends indicate that human activities during a particular period had a positive influence on vegetation rehabilitation (Wessels et al. 2004), and vice versa. More details of these methods can be found in section S5 of the supplemental materials.

Results

NDVI trends over the past decade

The averaged NDVI spatial and slope trends (per year) between 2000 and 2010 in these source regions (Fig. 2a) show that there was relatively high vegetation coverage in the northeast source regions of the Yellow and Yangtze rivers, and along the Lantsang river. Regions with relatively low vegetation cover only appeared in the northeast source regions of the Yellow and Yangtze rivers. In addition, spatial trends in NDVI slope (Fig. 2b) show that the increases in vegetation productivity between 2000 and 2010 mainly occurred in the northern areas of these source regions, and the average NDVI slope (per year) acquired employing Arcgis zonal statistics functions for the Yellow, Yangtze, and Lantsang rivers was 0.00353, 0.00239, and 0.00026, respectively.

Fig. 2
figure 2

Spatial trends in NDVI (a) and NDVI slope (b) between 2000 and 2010 in the source areas

Full size image

In most of these source areas, there were no significant variations in NDVI from before and after 2005, when the EPRP was initiated (Fig. 3). For instance, during the period 2000–2005, the proportion of the total area in which vegetation productivity increased (areas with NDVI slope >0) for the Yellow, Yangtze, and Lantsang rivers was 67.62, 62.95, and 57.38 %, while from 2005 to 2010, it was 62.11, 74.52, and 64.74 %, respectively. For the total source regions, it was 64.03 and 68.73 % for the two periods, respectively (Table 1). The NDVI slope from 2000 to 2005 in the three source regions was 0.00473, 0.00244, and 0.00184, and from 2005 to 2010 was 0.00313, 0.00493, and 0.00357, respectively.

Fig. 3
figure 3

Comparisons of spatial trends in NDVI slope in the periods 2000–2005 (a) and 2005–2010 (b)

Full size image
Table 1 NDVI slope variations from 2000 to 2010 (km2)
Full size table

Potential NPP trends

If we exclude the impacts of human activity on vegetation productivity, we observe some variations in potential NPP values in these regions before and after 2005. For instance, between 2000 and 2005, the potential NPP decreased in most of the source areas of the Yellow and Lantsang rivers, while from 2005 to 2010, the potential NPP increased in most of these areas, and across most of the source areas of the Yangtze river, these values decreased significantly (Fig. 4). These results indicate that in the period 2005–2010, climate change promoted an increase in vegetation cover in the source areas of the Yellow, Yangtze, and Lantsang rivers. It is also apparent that in the period 2005–2010, climate change played a more important role in vegetation rehabilitation in the region compared with the period 2000–2005. For instance, from 2000 to 2005, the proportion of the source areas of the Yellow, Yangtze, and Lantsang rivers where the slope of the potential NPP was greater than zero was 13.50, 13.82, and 14.86 %, respectively, while from 2005 to 2010, it was 65.41, 50.87, and 52.41 %, respectively (Table 2). For the total source areas, these values were 13.83 % before 2005 and 56.48 % after 2005. These results show that over the past decade, the impact of climate change on vegetation rehabilitation in the region has varied.

Fig. 4
figure 4

Comparisons of spatial trends in potential NPP slope in the periods 2000–2005 (a) and 2005–2010 (b)

Full size image
Table 2 Potential NPP slope variations from 2000 to 2010 (km2)
Full size table

The residuals between the potential and actual NPP values also show that before and after the start of the EPRP, human activities had no appreciable impact across most of the Water Tower (Fig. 5). For instance, in the headwaters of the Yellow, Yangtze, and the Lantsang rivers from 2000 to 2005, the area with residuals of potential and actual NPPs >0 covered 7.38, 16.30, and 9.72 % of the total area, respectively, and from 2005 to 2010 covered 7.48, 15.97, and 11.55 %, respectively. For the whole source regions, these proportions were 12.19 % before 2005 and 12.28 % after 2005 (Table 3). In addition, the statistics show that after the EPRP was launched in the region, the negative effects of human activities continued to increase: in the source areas of the Yellow, Yangtze, and Lantsang rivers from 2000 to 2005, the areas with residual slope >0 covered 38.63, 45.68, and 46.97 % of the total area, while from 2005 to 2010, the proportions were 52.66, 56.43, and 58.52 %, respectively. For the total source areas, the area covered was 43.20 % before 2005 and 55.27 % after 2005. These results show that over the past decade, human activities had a negative effect on about 50 % of the source areas, and even after the EPRP began, there were still some slight increases.

Fig. 5
figure 5

Comparisons of the residuals between the potential and actual NPPs in the periods 2000–2005 (a) and 2005–2010 (b)

Full size image
Table 3 Residual variations from 2000 to 2010 (km2)
Full size table

Discussion

At present, the United Nations Environment Programme (UNEP) defines desertification as “land degradation in arid, semi-arid, and dry sub-humid areas resulting from various factors including climatic variations and human activities” (UNEP 1992; United Nations 1992; Wessels et al. 2008). In particular, sandy desertification is characterized by high wind activity in regions with sandy surface sediments and is accompanied by signs of strengthening wind activity coincident with the onset of land degradation (Zhu and Chen 1994; Wang et al. 2008). The main processes of sandy desertification include dune reactivation, surface coarsening, grassland degradation, and other related processes (Wu 2009). Over recent decades, several studies have suggested that severe land degradation/desertification has occurred in the source areas of the three rivers (Wang et al. 2002; Li et al. 2010), and the major forms of desertification during this period included grassland degradation (Liu et al. 2006; Yang et al. 2006; Zhang et al. 2006) and sandy desertification (Dong and Chen 2002; Feng et al. 2004). Due to the appearance of severe sandy desertification issues (Dong et al. 2012), this region has attracted the attention of scholars, governments, and international organizations (Foggin 2011).

In their source regions, the tributaries of the Yellow, Yangtze, and Lantsang rivers flow through grasslands underlain by sandy sediments (Fig. 6a), and the area covered by mobile sandy lands is 43,635 km2, which accounts for 13.94 % of the total area of the source regions. In addition, although there are no large areas of mobile sandy lands, there are many small patches scattered across the grassland surface, and under the influence of climate change or human activities, sandy desertification can, and has, occurred (Fig. 6b). However, the commencement of the EPRP was not marked by significant variations in the area covered by mobile sandy lands. For instance, in 2005, the area of mobile sandy lands (S2 of the supplemental materials) in the source regions of the Yellow, Yangtze, and Lantsang rivers was 10,844, 26,047, and 6,797 km2, respectively, and in 2010, it was 10,790, 25,873, and 6,786 km2, respectively; i.e., only a 0.73 % variation in the area covered by mobile sandy land (Fig. 7). Taking into account the errors associated with the interpretation of remote sensing imagery, we suggest that there was no increase in sandy desertification over the past decade in the source areas of the three rivers.

Fig. 6
figure 6

Rivers developed within the grasslands (a) and one of the forms of sandy desertification (b) in the source regions. The exact locations of the photographs are unknown

Full size image
Fig. 7
figure 7

Comparisons of spatial trends in mobile sandy lands in the periods 2000–2005 (a) and 2005–2010 (b)

Full size image

Over the past decade, the average annual temperature has increased by nearly 0.5 °C, and there has been an increase in precipitation of nearly 70 mm (Fig. 8) across the study area. Although we cannot determine the specific contributions made by the rising temperatures and increased precipitation to vegetation rehabilitation, previous studies (e.g., Li et al. 2011) suggest that in this region, there are strong positive relationships between vegetation productivity and precipitation/temperature variations. In addition, the annual mean wind speed also increased over the past decade by about 0.24 m/s, which encourages the development of sandy desertification (Wang et al. 2007); however, there were some slight decreases in the area of sandy desertification across the three source areas. At present, although our study is based on current data and field surveys, we cannot be sure that the apparent reversal in sandy desertification over the past decade was due to the positive impacts of human activities or to other factors (Wang et al. 2013) that reduced eolian activity in the region.

Fig. 8
figure 8

Temporal trends in annual temperature, precipitation, and wind speed from 2000 to 2012 in the source regions of the Yellow, Yangtze, and Lantsang rivers. Curves were smoothed with a 13-year (2000–2012) moving average. The locations of meteorological stations are shown in Fig. 1

Full size image

Human activity in the Water Tower region mainly involves land reclamation for cultivation and grazing, which may have negative effects on vegetation rehabilitation (Wang et al. 2006). However, even after the EPRP was implemented in 2005, the land carrying capacities (sown area, number of large animals, and number of goats and sheep) have still increased (Fig. 9). The reason that the increasing impacts of human activities have not triggered desertification in these source regions may be that the human activities may still not exceed the carrying capacities of the land, or the NPP increases driven by climate change have provided enough biomass to support the increased human activity, and therefore, there was no increase in desertification across the source areas of the three rivers.

Fig. 9
figure 9

Temporal trends in human activity (indicated by the sown area, numbers of large animals, numbers of goats and sheep, and total output of meat). Large animals include horses, camels, cattle, donkeys, and mules. Data are from the Qinghai Province Statistical Yearbook

Full size image

Conclusions

Over the past decade, there have been no problems with desertification in the source areas of the Yellow, Yangtze, and Lantsang rivers. Although human activity may have affected about 50 % of the study area, vegetation rehabilitation has occurred from the beginning of the century to the present day. After the EPRP was launched in 2005, the negative effects of human activities, such as agriculture, also increased, but have not triggered desertification in this region. Although we cannot determine whether it was the rising temperature or increased precipitation that enhanced vegetation productivity, our analysis shows that the dominant factors controlling vegetation rehabilitation were related to climate change. In the early twenty-first century, human activities have not played an important role in the ecological rehabilitation/destruction of China’s Water Tower.