1 Introduction

Global land use/cover is undergoing rapid changes with human population growth and global warming (Pielke, 2005; Hanewinkel et al., 2013). Land management measures and land-cover changes have major effects on terrestrial ecosystems through variations in water cycle, surface temperature, biodiversity, primary productivity, carbon budget, and regional climate (Houghton et al., 1999; Kalnay and Cai, 2003; Foley et al., 2005; Liu et al., 2010; Pitman et al., 2011; Shannon et al., 2013; Sebastiaan et al., 2014). In recent decades, land use and land cover changes (LUCC) have taken place in China with the rapid development of the economy (Zhao et al., 2010; Li et al., 2015), especially in the Qinghai-Tibetan Plateau (QTP), where LUCC is strongly affected by both climate change and human activities (Wang et al., 2008; Li et al., 2009). However, land use changes have attracted less attention in the QTP than in more-developed areas in China. To date, the ecosystem in the QTP has suffered from severe degradation due to global warming and irresponsible human activities (Harris, 2010).

Maqu County is located in the source region of the Yellow River on the QTP, part of China's fragile environmental subzone. Because that ecosystem is extremely vulnerable and sensitive to global climate changes, it is often regarded as an ideal research region for studying the effects of global climate changes or global warming. In past decades, climate warming and livestock overgrazing have led to many eco-environmental problems in this region, such as land desertification (Dong et al., 2002), vegetation degradation (Yang et al., 2006), the degradation of frozen soil (Yang et al., 2004a), shrinkage of lakes and marshes (Qian et al., 2006), and increased soil erosion. Those studies showed that the LUCC was changing rapidly in Maqu County. A government project to restore grazing land to grassland was implemented in early 2004 and completed at the end of 2008 in Maqu County. The goal of that project was to fight grassland degradation and desertification, and to conserve biodiversity. Better understanding of the processes of LUCC and evaluation of the effects of that ecological restoration project may enable the improvement of future planning strategies (Shoshany and Goldshleger, 2002). Thus, study on LUCC and its driving forces in Maqu County is much needed.

In this study we used Landsat Thematic Mapper (TM) images from 1989, 2004, 2009, and 2014 to establish a high-precision land use and cover database, and we employed these data to detect and evaluate LUCC during the past 25 years. Our objectives were to: (1) explore the evolution of land use and cover in the past 25 years; (2) analyze the major driving forces responsible for LUCC in Maqu County; and (3) evaluate the effects of the ecological restoration project on the grassland ecosystem and provide some suggestions for future development and ecological environment protection.

2 Materials and methods 2.1 Study area

Maqu County is located at the source of Yellow River (Figure 1), with an area of 9.46×10⁵ ha; it is the main watershed for the Yellow River. The area has a continental monsoon climate typical of the QTP, with an annual mean temperature of 1.8 ℃. The annual mean precipitation is 600 mm, mostly concentrated in the summer; the annual evaporation is 1, 202 mm and the average relative humidity is 62%. The vegetation in the study area is primarily sub-alpine meadows. The soil is mainly alpine meadow soil, alpine steppe soil, and swamp meadow soils. The population density is low in Maqu County, averaging 5.8 people per km², and they are primarily Tibetan herdsmen.

2.2 Data sources

The data sources included TM images from 1989, 2004, 2009, and 2014, with a spatial resolution of 30 m. We selected images recorded between June and October when the vegetation was growing well. Because it was difficult to acquire cloud-free images that covered the whole study region within a given year, some images from adjacent years were chosen to replace the unusable images from those five years. During each study year at least four images (scenes) were obtained to cover the study area.

Digital topographic maps and other thematic maps such as maps of desertification, climatic zones, and vegetation were digitized from hardcopy topographic maps with a scale of 1:100, 000, and these images were used for geometric correction of the satellite images and to confirm the classification results.

Landscape images accurately positioned by means of GPS were acquired to assist in the interpretation of the established signs and to improve the classification accuracy. We also collected the relevant meteorological data for investigating the driving forces responsible for the LUCC.

2.3 Image pre-processing

The 2004 TM images were geo-referenced and ortho-rectified based on 30~40 ground control points (GCPs) using the Image Analyst function of the Modular GIS Environment (MGE) software (Geographic Resource Solutions, Arcata, CA, USA). GCPs were derived from the relief map at the scale of 1:100, 000 produced by the Chinese mapping agency in the early 1980s. The root-mean-square of the geometrical rectification between the two images error was limited to 1~2 pixels in the plain area and 2~3 pixels in the mountainous region. Images of 1989, 2009, and 2014 were matched with those of 2004 that had been precisely registered with GCPs; more than 50 GCPs were selected during the image-matching process in order to cover most of the area represented by the two sets of images.

2.4 Land use and cover classification

Maqu is solely an animal husbandry county; there is no farmland. Grassland degradation, marsh shrinkage, and desertification are the main ecological/environment problems. Therefore, we used an adjusted classification system (Table 1) based on a system of five primary land cover types (forest, grassland, water, built-up land, and unused land) that were divided into 23 secondary types. Relevant remote sensing interpretation marks for land use and land cover were as used in a previous study by Dong et al. (2012). Visual interpretation was used to derive the LUCC information. Although the visual interpretation of TM images is labor-intensive and time-consuming, the mapping accuracy of this method is higher than that of image classification using only the algorithms provided by image-processing software because of the low spatial and spectral resolution of TM images (Liu, 1996).

We used the freehand drawing function of the MGE software to delineate and label regions of the TM images by visually interactive interpretation to establish the LUCC databases for the four years studied. Based on the recognition ability of the TM images and the accuracy of the mapping, the manual visual interpretation and digitization of the TM images was carried out at a scale of 1:100, 000. During interpretation, we adopted the following mapping principles: (1) the minimum mapping patch was 7×7 pixels; (2) the deviation of the delineating locations was less than 1 pixel on the screen; and (3) the accuracy of the labeling patches was greater than 95% based on field investigation.

In addition to the TM images, we collected ancillary materials such as desertification maps, topographic maps, vegetation maps, meteorological data, and field investigation reports to assist in labeling the map patches during the interpretation process. After completing the manual visual interpretation, we derived the land use and cover dynamics by overlaying the data layers for each year using the GIS software, and calculated the data using spreadsheet software. We then created a matrix of land use changes in order to determine the patterns of change during the study period.

3 Results 3.1 Land use and cover dynamics since 1989

Figure 2 shows that grassland is the main land use and cover type in Maqu County, and the secondary types are marsh and forest (Table 2). From 1989 to 2009, high-coverage grassland, marsh, forest, water areas, and unused land decreased by 51, 607.1, 39, 430.8, 4, 463.9, 267.8, and 3, 217.3 ha, respectively. Moderate and low-coverage grassland, sand land, and built-up land showed the opposite trend, increasing by 43, 851.8, 53, 387.1, 1, 332, and 416.1 ha, respectively. These results suggest that degradation of the eco-environment was obvious from 1989 to 2009. From 2009 to 2014, high-coverage grassland, moderate-coverage grassland, and built-up land increased by 9, 192.3, 10, 904, and 50.8 ha, respectively, while low-coverage grassland, marsh, unused land, sand land, water areas, and forests decreased by 10, 568.3, 4, 917, 3, 968.7, 294.7, 315.5, and 82.9 ha, respectively.

3.2 LUCC processes

According to our cross-tabulation from 1989 to 2004 (Table 3), the total area of LUCC was 9.6% of the total area. There was 54, 058.1 ha of high-coverage grassland converted to other landscape categories, mainly to moderate-and low-coverage grassland, while 18, 807.8 ha of other landscape categories shifted to grassland, mainly from moderate and low-coverage grassland. The grassland conversions were mainly to different types of grassland coverage, while marshes mainly converted to moderate-and low-coverage grassland. These two types of changes were the main components of LUCC in Maqu County, accounting for 84.7% and 3.45%, respectively, of all the changed landscape area. With respect to grassland and marshes, the LUCC occurred primarily due to grassland degradation and marsh shrinkage caused by continuous overgrazing and increasing temperature. The other categories of LUCC were not obvious.

According to our cross-tabulation from 2004 to 2009 (Table 4), the total area of land use and land cover change was 17.5% of the total area. During this period, high-coverage grassland mainly converted to moderate-and low-coverage grassland, and marsh mainly converted to grassland; these changes accounted for 16.9%, 13.0%, and 23.3%, respectively, of all the changed landscape area. Water areas mainly converted to marsh and sand land, and the unused land converted to grassland and sand land. Changes in the built-up land were not obvious.

During the period of 2009-2014 (Table 5), the main LUCC type was moderate-coverage grassland converted to high-coverage grassland, comprising 18.4% of the total change area. Low-coverage grassland mainly converted to moderate-coverage grassland, comprising 24.2% of the total change area. Sand land mainly converted to moderate-and low-coverage grassland, and 6, 102.5 ha of marsh converted to different coverage grasslands, comprising 8.7% of the total change area.

3.3 Driving forces of land use and land cover change

Our results showed that the eco-environment of Maqu County seriously deteriorated before 2009, as manifested mainly in the continuous grassland degradation and rapid marsh degradation. Since 2009 the eco-environment has improved because of the ecological restoration project that was implemented from the beginning of 2004 to the end of 2008 in Maqu County. With reference to the natural characteristics of the study area and field investigation results, the main factors affecting the LUCC involved the regional climate change, human activity, and rodent damage.

3.3.1 Regional climate changes According to the meteorological data recorded at Maqu weather stations in the past 46 years, the average annual temperature has been rising significantly. The average mean temperature increased 0.35 ℃/decade, which was much larger than the global mean rate of increase (0.03 ℃ to 0.06 ℃/decade). Also, the average surface temperature increased at a rate of 0.36 ℃/decade, and this increasing trend has been more significant since 2000 (Ning et al., 2012). The annual precipitation has shown no significant change (Figure 3), being close to the multi-year average during the 2000s, higher in the 1960s to 1980s, and very much lower in the 1990s. Secondly, the annual precipitation fluctuated significantly in different years; for example, precipitation in 1981 was 1.8 times than 1996. Moreover, the difference between evaporation and precipitation increased, with a ratio of 12.1 mm/decade during the grass growing season (Wang, 2015), increased potential evapotranspiration and decreased precipitation have resulted in water deficit which was one of the most important factors in the grassland degradation and environmental deterioration. Such temperature and precipitation changes have delayed vegetation development and reduced plant productivity, and accelerated grassland degradation. More importantly, the increased temperature has resulted in degradation of seasonal frozen soil. It was reported that the thickness of the seasonal permafrost could be decreased by 12.6 cm/decade, and it has decreased from 1.2 m in 1970 to 0.7 m at present (Wang et al., 2006). Seasonal permafrost plays an important role in maintaining the meadow vegetation in the source of the Yellow River and other areas of the QTP, because permafrost layers can supply abundant moisture and special ecological environments for plant growing. Seasonal frozen soil degradation may cause the increase of the seasonal thawing depth and water infiltration, the increase of the surface layer ground temperature, and decreased soil water content. Eventually, it will result in the acceleration of desertification (Yang et al., 2004b). The factors driving LUCC in Maqu County are, in fact, interrelated. Of those factors, the temperature rise is predominant, directly resulting in water deficits and causing a series of ecological environment problems. 3.3.2 Human activity In the Maqu County region, livestock grazing on natural rangeland is the main economic activity. The total number of animal units (sheep units) was 192.16×10⁴in 2004, an overload of 33.34×10⁴ units. Seasonal grazing is very unbalanced; the winter and spring grazing area is smaller than the summer grazing area, but the grazing period is twice as long as in the summer grazing area. As a result, severe overgrazing takes place from October to May over a large part of the grassland. Desertification resulting from overgrazing is widespread in the winter and spring grazing area. Other human activities, such as gold mining and medicinal plant collecting, are also responsible for grassland degradation and desertification in Maqu County. Our study showed that grazing management measures, such as grazing prohibition and rest grazing have a significant effect on LUCC in the grassland ecosystem. 3.3.3 Rodent damage Previous studies have confirmed that rodent and insect damage is a severe problem for the grassland ecosystem in Maqu County. Field investigations have shown that this damage was mainly caused by the Plateau Pika (Ochotona curzoniae) and Plateau Zokor (Myospalax baileyi). These rodent-like mammals are widely distributed in Maqu County and in the source region of the Yellow River. They have severely damaged the vegetation, altered the community composition of dominant plants, and have even turned grasslands into "black soil flats, " creating breakthrough points where erosion begins, thereby accelerating the degradation of vegetation and desertification. In the early 2000s the rodent-affected grassland area in Maqu County reached 16.9×10⁴ ha, accounting for 19.5% of county's total grassland area. This was equivalent to 4.31×10⁴ha of grass yield (Feng et al., 2006). 4 Discussion and conclusions

Grasslands and marshes are the main land use and cover types in the study region, accounting for about 81.4% and 5.0% of the total area in 2014, respectively. The total amount of LUCC was 9.6% of the total area from 1989 to 2004, 17.5% of the total area from 2004 to 2009, and 7.4% of total area from 2009 to 2014. Before 2009, land use and cover changes were mainly manifested in degradation of vegetation, significant marsh shrinkage, land desertification, lake and other water body shrinkage, as well as a continuous decrease in the discharge of the Yellow River flowing out of Qinghai Province. From 2004 to 2008, a government project to restore grazing land to grassland was implemented in Maqu County. The trend of ecological environment deterioration has since been reversed, mainly embodied in grassland restoration. The marsh shrinkage rate is remarkably decreased compared with 1989 and 2004, and land desertification has been reversed. These results suggest that the project has had a positive effect on the grassland ecosystem.

Climate changes were main driving factors responsible for land use change in Maqu County. According to Maqu weather station data, on the whole, the average annual precipitation has had no significant change trend but the temperature has increased significantly; the average mean annual temperature increase has been 0.35 ℃/decade since 1967. As a result, the depth of frozen soil has become shallower and large areas of seasonal permafrost have thawed and some have even disappeared. Because seasonal permafrost plays an important role in maintaining alpine meadow ecosystems, the degradation of seasonal permafrost has led to the degradation of both soil and vegetation communities. The interaction between freeze-thaw processes and vegetation degradation creates large areas of exposed surfaces and causes ground temperatures to rise, the upper limit of frozen soil to become deeper, the surface soil to dry out, and the promotion of desertification.

Livestock husbandry has also exerted important effects on LUCC because of overgrazing and unbalanced seasonal grazing, as described above. Overgrazing has not only caused grass degradation but has also resulted in the increased pika and zoker populations (Xiao et al., 1982), which, along with insects, have also aggravated vegetation degradation. And finally, with the rapid human population growth and development of the economy, the impact of human activities on the eco-environment has also been exacerbated.

In recent years the Chinese government began to implement a series of ecological environment protection projects on the QTP, with goals of fighting desertification and grassland degradation, conserving water and soil, and protecting biodiversity. Those projects not only conserved grasslands and restored ecosystems, but also improved the quality of life for local people. Many studies have assessed those projects and have confirmed that ecological restoration projects have had great positive effects in restoring degraded vegetation; vegetation coverage, biomass, and the normalized difference vegetation index (NDVI) have clearly increased after implementing these projects (Wang et al., 2009; Guo et al., 2010). Our results also showed that ecological degradation in Maqu County was reversed after 2009, mainly because of the ecological restoration project that was implemented there from 2004 to 2008. These assessments may enable the improvement of future land use planning strategies.

Acknowledgments:

We gratefully acknowledge the funding received from the Natural Science Foundation of China (41301003, 41371026, and 31470480) and the Technology of the People's Republic of China (No. 2013CB956000).