1 Introduction

Desertification has been defined as land degradation in arid, semiarid, and dry sub-humid areas as a result of various factors, including climatic change and human activities (UNCCD 1994). Globally, the total area of desertified land in 1996 amounted to 3.6 × 107 km2, amounting to 24.1 % of Earth’s land surface, and this affected on one-sixth of the world’s population, most of whom live in poverty (Middleton and Thomas 1997). China is one of the countries which has been most seriously affected by desertification (Wang and Zhu 2003); the area of desertified land totaled 1.84 × 107 km2 in 2005, distributed across 13 provinces in northwestern, northern, and northeastern China (CCICCD 2006).

Desertification is driven by several interconnected factors and triggered by the combination of the natural predisposition of the environment with human pressures and climate change (UNCCD 1994; Adamo and Crews-Meyer 2006). In general, climatic conditions are the dominant drivers of desertification in northern China (Wang et al. 2008; Han 2003). Specific temperature, drought, and wind regime trends have been associated with desertification in China (Wang et al. 2006, 2008). However, human activities such as inappropriate and unsustainable land use are the main cause of sandy desertification in some regions (Ge et al. 2013). As these previous results indicate, the relative roles and synergistic effects of climate conditions and human activities on desertification are still not clear. Recently, researchers have tried to clarify the relationships among these factors and the resulting impacts on desertification (Luo et al. 2005; Xu et al. 2011; Ge et al. 2013).

Desertification mainly occurs in dry lands (e.g., Dregne et al. 1991; Kassas 1995; Sivakumar 2007), but it was affected by a wide variety of environments, climates, and societies (e.g., Arnalds 2004). Different methods have been used to analyze desertification. These include empirical approaches (e.g., Liu et al. 2003; Geist and Lambin 2004), remote-sensing applications (e.g., Liu and Wang 2007; Shalaby and Tateishi 2007; Zhang et al. 2008; Hu et al. 2012; Li et al. 2013; Xue et al. 2013), and modeling (e.g., Salvati et al. 2008; Ding et al. 2009).

With the development of remote-sensing technology, especially when combined with geographic information systems (GIS) technology, researchers now have an effective method for conducting desertification research for large areas (Wang et al. 1998; Hu et al. 2012). By analyzing remote-sensing imagery over a long time period, it becomes possible to assess desertification trends accurately and in a timely manner, thereby supporting the prediction and management of this problem.

The Hexi region lies in Gansu Province of northeastern China. It is an arid to semiarid region and has one of the most sensitive ecological environments and most severe desertification in China (Han et al. 2013). The problems that have developed in this region have attracted the attention of China’s Central Committee and State Council, leading to much research on this region. Table 1 summarizes recent studies of the desertification that has occurred in the Hexi region. The area of desertification in the Hexi region varies among studies (Zhu and Liu 1989; Gao et al. 2001; Wang 2003; Wang et al. 2005; Shi et al. 2006) due to differences in the research methods and changes in the area of desertification over time. Differences in the indicators used to assess desertification were also responsible. Previous research mostly focused on the use of a single indicator or of the relationship between a small number of indicators (Nasrallah and Balling 1993; Dall-Olmo and Karnieli 2002; Niu 2005; Zeng et al. 2006; Ma and Xiao 2007; Zeng and Feng 2007).

Table 1 Area of desertified land in study region in different periods
Full size table

In this paper, we used Landsat TM and HJ-1A/B data obtained between June and August in 2000, 2005, and 2010 to analyze the spatial and temporal trends in desertification in the Hexi corridor. Satellite monitoring provides a rapid, frequent, and relatively inexpensive way to monitor such changes. The HJ-1A and 1B satellites are small Chinese Earth observation satellites that were launched on September 6, 2008. Among the payloads aboard these two satellites, their multispectral CCD cameras are important instruments that are widely used in eco-environmental monitoring. Each satellite carries two CCD cameras (CCD1 and CCD2), with a swath width of 700 km, a spatial resolution of 3,100 m, and a 48-h return period. The HJ-1A/1B CCDs have three visible bands (430–520, 520–600, and 630–690 nm) and one near-IR band (760–900 nm) (Wang et al. 2010; Sun et al. 2010). The satellite constellation is composed of three small satellites, supported by ground and application systems. It provides all-weather imagery. The two-satellite 1A and 1B constellation represents the first disaster monitoring spacecraft orbited by China.

We used five indicators in our analysis to assess the severity of desertification: the modified soil-adjusted vegetation index (MSAVI), surface albedo, land surface temperature (LST), fractional vegetation cover (FVC), and temperature–vegetation drought index (TVDI). Through comparisons of the five indicators and field investigations, we determined weights for the indicators and used them to define criteria for grading the severity of desertification. We also collected data on sandstorm frequency, temperature, relative humidity, and precipitation to analyze the influence of these factors on desertification. Simultaneously, to determine the effects of human activity on the desertification process, we analyzed the population, vegetation cover, cultivated land, and livestock. The results will help managers to assess regional desertification trends and develop more effective policies to combat desertification, changes in the regional ecological environment, and changes in the human environment.

2 Materials and methods

2.1 Study region

The Hexi corridor includes four cities: Zhangye, Jiuquan, Wuwei, and Dunhuang (Fig. 1). The corridor is bordered by the Tengger Desert in the east, the Badain Jaran Desert in the northeast, the Kumtagh Desert to the northwest, and the Qinghai-Tibetan Plateau to the southwest. It extends 800 km from east to west direction and 40–100 km from north to south (Gao et al. 2001). This region is far from the nearest sea, so there is little precipitation (an average annual total ranging from 50 to 200 mm), high temperatures (an annual average of 5–10 °C, with mean monthly values ranging from a minimum of −16 °C in January to a maximum of 27 °C in July), and strong evaporation, with pan evaporation averaging 2,000 mm annually (Ma et al. 2005). There are three continental rivers in the region: the Shiyang river, Heihe river, and Shulehe rivers in the eastern, middle, and western parts of the region (Fig. 1). The main types of desertified land are sand dunes (including mobile, semi-fixed, and fixed), wind-eroded land, saline and alkaline land, and water-eroded land (Wang 2003). Sand dunes and the desertified area are mainly found in the relatively flat land between Qilian Mountains in the south and the deserts to the north.

Fig. 1
figure 1

Location of the Hexi region

Full size image

2.2 Indicators of desertification

As noted earlier, we used five indicators for the degree of desertification: MSAVI (Li and Zhou 2001), surface albedo (Nasrallah and Balling 1993; Wang et al. 2013a, b; Zeng et al. 2006), LST (Han et al. 2014; Luo et al. 2010; Ma and Xiao 2007), FVC (Han et al. 2014; Wang et al. 2013a, b), and TVDI (Han et al. 2014; Li et al. 2008). We used maximum difference normalization method to normalize the value of each indicator, eliminate the subindex dimensional effects, weigh each indicator based on the average of the opinions of five experts (Table 2), and superimposing calculate the weighted five indexes, and used natural breaks (Jenks) classification method values to determine the degree of desertification by ArcGIS soft, and the natural breaks are based on natural groupings inherent in the data. ArcGIS identifies break points by picking the class breaks that best group similar values and maximizes the differences between classes. The features are divided into classes whose boundaries are set where there are relatively big jumps in the data values. To improve the accuracy of our monitoring, we used the same training samples. The data used to calculate the indicators were obtained by inversion of the same Landsat TM and HJ data with a spatial resolution of 30 m from 2000, 2005, and 2010, respectively. MSAVI had the highest precision and expressed the most information about desertification, followed by FVC, LST, Albedo, and TVDI (Table 3).

Table 2 Indicators of desertification used to determine the degree of desertification
Full size table
Table 3 Classification precision of different combination of desertification monitoring indexes
Full size table

2.3 Data acquisition, processing, and analysis

The Landsat images were obtained from the Global Land Cover Facility (http://glcfapp.umiacs.umd.edu/), and HJ-1A/B obtained from the satellite Environment Center Ministry of Environmental Protection (http://www.secmep.cn/secPortal/portal/index.faces). Because images acquired in different seasons will have different properties, and because this can influence the desertification monitoring results, we chose a single season for all of the data. According to previous research results, late summer is the best season for monitoring desertification in northern China (Wu 2001a, b). In the different period, remote-sensing image calculation processes, using the histogram equalization method to nonlinear expanding the three period images, make the pixel numbers in a given gray value are equal for each image. Finally, the images are enhanced.

The images were processed by a data supply agency, including systematic geometric correction and radiometric calibration. The data were geo-referenced by using ground control points collected in the field around the study site and based on 1:50,000 scale topographic maps provided by Gansu Province bureau of surveying and mapping (http://www.cehuiju.gansu.gov.cn/). The correction accuracy was better than 1 pixel. The resulting data were projected using an Albers equal-area conic coordinate system. All the images were of high quality (with <10 % cloud cover). The images were atmospherically corrected using the FLAASH module of the ENVI 4.8 software (Agrawal et al. 2011). FLAASH is a first-principles atmospheric correction that incorporates the MODTRAN4 radiation transfer code to correct the wavelengths in the visible, near-infrared, and shortwave infrared bands up to 3 nm (Sun et al. 2010; Wang et al. 2010).

We used version of ENVI 4.8 software to create false-color images obtained by stacking data from the near-infrared, red, and green bands to perform geometrical correction. More details of the methods for satellite image processing are described by Navas and Machin (1997), Valle et al. (1998), and Vasconcelos et al. (2002). Our evaluation included both the area of desertified land and the degree of degradation. We classified the desertified land into five grades (Table 4). We used field surveys on August, and 35 sites had surveyed to determine the actual degree of desertification and then related the correlation analysis to the characteristics of TM and HJ images from the August of 2010 to determine the actual status of the desertified land using TM and HJ images. We determined the location of the field survey sites with a 2D precision of 4 m using a global positioning system and recorded land surface features using a digital camera. All images were interpreted manually using version ENVI 4.8 software and 9.1 of ArcGIS (http://www.esri.com/software/arcgis) according to the criteria in Table 4. In the last step, we created maps of the distribution of desertified using ArcGIS and compared them between periods to determine the dynamics of the temporal changes and spatial patterns of desertification between 2000 and 2010. To detect transfers among different desertification categories and produce a land transfer matrix, we used the graphical superposition method of Li et al. (2007).

Table 4 Classification and indicators of desertification grades
Full size table

We used this data to assess the dynamics of desertification. We used the following equation (Zhao et al. 2001):

$$D_{i} = {\text{ST}}_{i} /{\text{DI}}_{i} \times 1/T \times 100\,\%$$
(1)

where i is the degree of desertification, with 0 representing no desertification, versus 1 for light, 2 for moderate, 3 for severe, and 4 for extremely severe desertification; D i is the dynamic response of desertification of degree i; DI i is the area of the ith degree of desertification in the initial year of a study period; ST i is the area of the ith degree of desertification that transferred to other grades of desertification during a study period; and T is the duration of the study period. The annual ratio of desertification was calculated using the following equation (Zhu and Chen 1994):

$$R_{i} = (\sqrt[T]{{{\text{DI}}_{i} /{\text{AI}}_{i} }} - 1) \times 100\,\%$$
(2)

where R i is the annual ratio, and AI i is the area of the ith degree of desertification after T years.

2.4 Meteorological and socioeconomic data

We collected and analyzed local meteorological and socioeconomic data to explore the mechanisms responsible for desertification. Meteorological data were obtained from four meteorological stations (Dunhuang, Jiuquan, Zhangye, and Wuwei) in the Hexi region. As proxies of human activity between 2000 and 2010, we chose the population in each year, livestock numbers (cattle, horses, camel, donkey, goats, and sheep), the area of cultivated land, and net primary productivity (NPP). We collected this data from the statistics departments of local governments and from the Gansu Development Yearbook (Gansu Development Yearbook). Thematic maps, geomorphologic maps, and vegetation maps (http://westdc.westgis.ac.cn/) were also used as supplementary data sources.

3 Results

3.1 Dynamic changes in the area of desertification from 2000 to 2010

Using the calculation method described in Sect. 2.2, we obtained weights of 0.30 for MSAVI, 0.25 for FVC, 0.20 for Albedo, 0.15 for LST, and 0.10 for TVDI. Figure 2 shows the spatial distribution of desertified land in the Hexi corridor in 2000, 2005, and 2010 based on the weighted sum of these indicators. The spatial distribution of the desertified land was similar in 2000, 2005, and 2010, with desertified land mainly distributed in the north and west, beside of oasis areas (cultivated land in Fig. 1) and bodies of water, and in regions with sand dunes (mobile dunes, semi-fixed dunes, and fixed dunes).

Fig. 2
figure 2

Distribution of the desertified land in 2000, 2005, and 2010

Full size image

Table 5 shows the area of each degree of desertification and its corresponding proportion of the total land area in each of the three years. The total area included in our monitoring was 27.59 × 104 km2, and the total area of desertified land was 22.02 × 104, 21.22 × 104, and 21.06 × 104 km2 in 2000, 2005, and 2010, respectively, which amounted to 79.8, 76.9, and 76.3 % of the total area. Most of the desertified land was extremely severely desertified in each year. The lightly, moderately, severely, and extremely severely desertified land accounted for 0.71 × 104, 1.55 × 104, 3.89 × 104, and 15.87 × 104 km2, respectively, of the total study area in 2000. However, the lightly, moderately, severely, and extremely severely desertified land accounted for 0.71, 1.38, 3.13, and 15.83 × 104 km2 of the total in 2010, respectively.

Table 5 Areas of desertified land in 2000, 2005, and 2010
Full size table

The area of land with no desertification increased at annual ratio was −2.65 and −0.62 % from 2000 to 2005 and from 2005 to 2010, respectively (Table 5). From 2000 to 2005, the areas of lightly, moderately, severely, and extremely severely desertified land decreased, with the largest decrease for severely and moderately desertified land, with values of 3.69 and 0.55 %, respectively. From 2005 to 2010, the areas of lightly and severely desertified land increased, by 0.03 and 0.14 %, respectively, but the areas of moderately and extremely severely desertified land decreased, by 0.10 and 0.80 %, respectively (Table 5).

3.2 Spatial changes, and transfers between different degrees of desertification

Based on change of the difference in the total desertification indicator between years, we defined five types of desertification pattern: strong development (−0.6 to −0.4), development (−0.4 to −0.2), stability (−0.2 to 0.2), reversal (0.2–0.4), and strong reversal (0.4–0.6). Figure 3 shows the spatial changes in desertification from 2000 to 2005 and from 2005 to 2010. In both periods, most of the study region showed a stability or reversal of desertification. From 2000 to 2005, the area that showed stability was largest (17.77 × 104 km2), followed by the areas with a reversal of desertification (1.53 × 104 km2), strong reversal (0.95 × 104 km2), development (0.16 × 104 km2), and strong development (<0.01 × 104 km2). From 2005 to 2010, the area with stability was also the largest (21.89 × 104 km2), followed by the areas with desertification development (1.02 × 104 km2), reversal (0.70 × 104 km2), strong reversal area (0.18 × 104 km2), and strong development (0.16 × 104 km2).

Fig. 3
figure 3

Trends of desertification development from 2000 to 2005 and from 2005 to 2010

Full size image

In 2005, the area of desertification decreased relative to that in 2000. Figure 3 shows that the reversal mainly occurred at the border of oasis areas and in the mountains. It is possible that rainfall increased in this region and that the rainfall was sufficient to support vegetation, but unfortunately, no meteorological stations provided data for most of these areas. However, the area of desertification development was mainly distributed in the southeastern region and between mountains and oasis areas. From 2000 to 2005, the area of desertified land increased by 0.31 × 104 km2, whereas desertification reversal occurred in 1.10 × 104 km2 (Table 6). More specifically, 0.45 × 104 km2 of severely desertified land changed to non-desertified land, and 0.18 × 104 km2 of non-desertified land change became lightly desertified land, accounting for 58.1 and 40.9 % of the total newly desertified land and reversal land, respectively.

Table 6 Transfer matrix for land that changed between different degrees of desertification
Full size table

In 2010, the area of desertified land also decreased relative to that in 2005. Figure 3 shows that the area in which reversal occurred was mainly distributed at the border of oasis areas, which because of the area of cultivated land decreased in this region (Sect. 4.2, Table 7). However, the area in which desertification developed was mainly distributed in the mountains, mainly because the snow area and the rainfall in the mountains decreased (Han et al. 2009). From 2005 to 2010, the area of desertified land increased by 0.35 × 104 km2, and the area where reversal occurred amounted to 0.55 × 104 km2 (Table 6). Moderate desertification changed to no desertification in 0.28 × 104 km2, whereas 0.15 × 104 km2 of non-desertified land changed to moderate desertification, accounting for 50.9 and 42.9 % of the desertified and reversal land, respectively.

Table 7 Dynamic response (D i ) of desertification during the two periods
Full size table

3.3 Desertification dynamics in land with different degrees of desertification

Between 2000 and 2010, the area of desertified land decreased by 1.0 × 104 km2, for an annual transformation rate of −3.25 %, and this decrease was almost the same as increase in the area of newly desertified land (0.96 × 104 km2). Table 7 shows the dynamic response of desertification during the two periods. From 2000 to 2005, the greatest dynamic response (D i ) was in land with severe desertification (25.2 %), followed by land with moderate, light, extremely severe, and no desertification. However, from 2005 to 2010, the greatest dynamic response was in land with moderate desertification (20.3 %) and then extremely severe, light, severe, and no desertification. From 2000 to 2010, the averaged dynamic response was largest for land with severe desertification (22.4 %), followed by moderate, light, extremely severe, and no desertification.

4 Discussion

Many factors can contribute to the observed desertification dynamics; these include climate, soil, vegetation, socioeconomic changes, and human activities. We summarized these variables as natural factors and human activities, each of which had different contributions to desertification processes.

4.1 Climate factors

Climate change is an important driving force for desertification in China (Zhu and Liu 1989; Wang et al. 2006). Climate affects on desertification mainly through its influence on air temperature, precipitation, relative humidity, and so on. Changes in air temperature, precipitation, relative humidity, and wind velocity change the threshold wind velocity for eolian entrainment of soil particles. High temperatures, low precipitation, and high wind velocity will exacerbate desertification. In this study, we used data from four meteorological stations to analyze the climate changes in our study area. We analyzed the air temperature, precipitation, relative humidity, and annual number of sandstorm days to represent climatic changes during the study period. We used linear regression (y = a + bx) to identify trends in these four parameters (Supplement Table S1). The mean annual temperature increased at all four stations, by an average of 0.6 °C from 2000 to 2010, and increased most strongly at the Zhangye station (R 2 = 0.57), by 1.1 °C (Fig. 4). Because increased air temperature could increase evaporation and decrease soil moisture content, leading to decreased vegetation cover and increased wind erosion, this trend indicates greater potential risk of desertification. The precipitation and relative humidity showed similar trends at all four sites; however, at Dunhuang and Wuwei, the precipitation and relative humidity decreased, whereas both increased at Jiuquan and Zhangye. The precipitation increased most strongly at the Wuwei station (R 2 = 0.17), and the relative humidity increased most strongly at the Zhangye station (R 2 = 0.43). The number of sandstorm days decreased at all four stations and decreased most strongly at the Wuwei station (R 2 = 0.35). This factor would tend to indicate a decreased risk of desertification in all four areas.

Fig. 4
figure 4

Changes in four key climatic parameters from 2000 and 2010 at Zhangye weather stations in the study area

Full size image

4.2 Human activities

Desertification is also the result of human factors. Human activities such as population growth, overgrazing, and overcutting of vegetation could increase desertification, whereas offsetting activities such as planting vegetation and various techniques to combat desertification could decrease desertification. In this paper, we evaluated human activities using four proxies: the human population, the number of livestock, the area of cultivated land, and changes in NPP, which we used as an indication of changes in vegetation cover (Table 8).

Table 8 Comparison of desertification land in two periods relative to the proxies of human activities
Full size table

The region’s population increased by 19.94 × 104 from 2000 to 2005 and decreased by 15.81 × 104 from 2005 to 2010 as a result of the implementation of an ecological migration policy imposed by the central government, which had the goal of moving residents to areas with a less fragile environment and higher carrying capacity. The increased population led to urban expansion and increased intensity of human activities, including an increased area of cultivated land and destruction of natural vegetation. Both factors would tend to promote desertification, as would grazing. In the oasis areas, increased populations combined with irrational land usage and overexploitation of water sources to increase the risk of desertification.

The livestock number increased by 3.01 × 107 and 2.14 × 107 from 2000 to 2005 and 2005 to 2010, respectively. More livestock need more grassland, and the resulting overgrazing destroyed grassland and exposed the top layer of the soil to wind erosion and nutrient loss that reduced the growth of the remaining vegetation. The government prohibited grazing in many areas to combat this process, leading to decreased livestock numbers, allowing the grassland to recover, and leading to reversal of desertification.

The area of cultivated land increased by 0.19 × 104 km2 from 2000 to 2005 and decreased by 0.11 × 104 km2 from 2005 to 2010. These changes corresponded to the aforementioned population changes, leading to increased reclamation of land for agriculture from 2000 to 2005 and an increased risk of desertification. However, from 2005 to 2010, the decreased population led to the abandonment of cultivated land, thereby decreasing pressure on the desertified land and allowing it to recover.

Kang et al. (2004) concluded that overexploitation and overutilization of water resources in the artificial oases in the middle reaches in three inland river basins have caused degradation of the ecological environment, leading to desertification of the natural oases in downstream reaches. The water resource problems related to desertification in the Shiyang River Basin, including competition for water resources among users in the upper, middle, and downstream parts of the basin, have increased the degradation of natural oases and increased desertification in downstream regions (Fang and Sun 2006; Dong 2009; Xie 2004; Du et al. 1996).

Recently, a series of key national projects have been initiated to combat desertification in the study area. These include the Three Norths Shelterbelt Project, the Conversion of Cropland to Forest or Grassland Program, and the Water and Soil Conservation Program (Xue et al. 2013). Under these programs, the NPP increased in 8.10 × 104 g a−1 of the study area from 2000 to 2005 and in 0.61 × 104 g a−1 of the study area from 2005 to 2010 (Table 8). The change of NPP directly reflects the desertification process. From 2000 to 2005, NPP increases more than it did during 2005 to 2010, leading to a greater reversal of desertification from 2000 to 2005 (0.82 × 104 km2) than from 2005 to 2010 (0.17 × 104 km2). Although the increase in NPP was greater in the first period, the ecological migration policy compensated for this somewhat by greatly decreasing human pressure on the land.

In our study area, the area in which reversal of desertification occurred responded to the observed changes in the population, livestock number, and vegetation cover (Table 8). Therefore, human factors were strong contributors to desertification.

5 Conclusions

We used remote-sensing imagery to calculate five indicators of desertification and monitor its spatial and temporal pattern in the Hexi corridor of northern China from 2000 to 2010. The area affected by significant desertification in the Hexi corridor was large, amounting to more than 76 % of the total land area. From 2000 to 2010, desertified land decreased by 0.06 × 104 km2, at an annual rate of 0.90 %. From 2000 to 2005 and from 2005 to 2010, most of the area of desertified land showed either a stability or reversal of desertification. Both natural and human factors contributed to the desertification dynamics during the study period. The trends in several natural factors contributed to a reversal of desertification in some areas. However, human factors such as population changes and increased cultivation of land appear to have been stronger driving forces that were directly responsible for the changes in desertification. Several national projects are currently underway to reverse desertification, and the present results show that it should be possible to monitor the effects of these policies by means of remote sensing.