A review on historical trajectories and spatially explicit scenarios of land-use and land-cover changes in China

ABSTRACT

Land use in China has changed remarkably over the past centuries with manifold implications for sustainable development. Investigation and projection of the land-use and land-cover changes (LUCC) have therefore become critical to promote the understanding of the LUCC process and the interactions with human societies. Scientists had devoted great efforts to create, process, and interpret comprehensive historical land-use/cover datasets, and to simulate future land-use systems under different scenarios. We synthesize the literature on the historical trajectories of LUCC in China, and summarize existing efforts of scenario developments for the projection of spatially explicit LUCC. Our review therefore provides salient suggestions for future land-use change analysis.

KEYWORDS:

• Land-use change
• remote sensing
• geographic information system
• scenario analysis
• China

1. Introduction

During the past decades, human activities have greatly transformed the terrestrial surface through changes in land-use and land-cover (Brown, Walker, Manson, & Seto, 2012; Foley et al., 2005). Spatially explicit identification of historical trajectories of land-use and land-cover changes (LUCC) and clarification of the interactions between human activities and natural process will provide important lessons for better understanding of future changes in the Earth system, including land-use change, climate change, and associated changes in human societies and economy (Lambin & Meyfroidt, 2011).

A crucial step for attaining sustainable land-use management is to derive and interpret accurate data resource based on integrated methodology and technology that support spatial monitoring and investigation of LUCC at different scales (Liu, 2001). Then, the next important step is to investigate the drivers of LUCC and the interactions with human activities. Subsequently, scenarios development combined with land-use models can provide spatially explicit projection of LUCC, which is critical for the identification and assessment of potential land-use management and policies (Liu & Tian, 2010).

China, as the most populous and one of the most rapidly developing countries, has experienced dramatic LUCC over the past four decades, and has been confronted with the challenges of supporting the growing population whilst easing the negative impacts of LUCC (Deng, Huang, Rozelle, & Uchida, 2010; Liu et al., 2010). In China, lots of research on LUCC have been conducted at regional scale (Wang, Liu, & Ma, 2010; Zhang, Sun, Zhang, & Tong, 2008), while in order to implement reasonable land management for long-term sustainable economic development, it is crucial to conduct LUCC monitoring at a whole nation scale based on accurate and timely information on land dynamics. Thus, this paper focuses on summarizing the historical trajectories of nationwide LUCC in China, and comparing different scenarios for the simulation of spatial patterns of future land-use and land-cover. Based on such review, we aim to supply comprehensive information for historical land-use and land-cover data and the methods used to derive these, and also to stimulate future research in analyzing and projecting land-use changes from the perspective of China’s sustainable land-use management and development.

2. Historical trajectories of land-use and land-cover change in China

Land use within China has undergone substantial transformations mainly by human activities for socioeconomic development, and the magnitude and patterns of China’s historical LUCC have been regarded as an important part in global change studies (Deng, Huang, Rozelle, Zhang, & Li, 2015; Liu & Tian, 2010; Miao et al., 2013). For better understanding of the dynamics and change trends of land in China, research on historical LUCC have attracted lots of attention from the scientific community, and significantly promoted the understanding of the LUCC process and the interactions with human societies (Miao et al., 2013).

2.1. Data and method for monitoring

2.1.1. National and global land-cover datasets

1. National land-cover datasets

Reliable datasets play a critical role in the processes of monitoring LUCC, including land-use mapping, change detection, and dynamic analysis. In a traditional way that is based on field surveys, there are few representative studies on nationwide LUCC in China. Though Wu and Guo (1994) conducted the first nationwide land-use survey and mapping through extensive field surveys and interpretation of aerial photographers and Landsat images during the late 1970s and early 1980s (Wu & Guo, 1994), national surveys of land use were very limited in the past due to China’s vast territory and geographical complexity, thus data quality and reliability have been the major problems faced by historical LUCC analysis in China (Liu & Tian, 2010; Liu et al., 2003). However, with the development of high-resolution remote sensing (RS) and geographic information systems (GIS) technologies, LUCC analysis made headways in China. Compared to traditional manual LUCC monitoring, combining the high-resolution RS images with GIS technology can provide more accurate, large-scale, time-saving datasets with great ability to be updated (Li, Yin, & Liu, 2011).

RS technology has been widely used to obtain accurate and timely information on LUCC in China since the late 1970s, and has played a significant role in understanding the nationwide land-use and environmental situations (Deng & Liu, 2012; Liu, 2001; Zhang & Zhang, 2007). Several national and global datasets with various spatiotemporal resolutions have been obtained based on multi-temporal and multi-sensors satellite images, and have been applied to reveal historical LUCC in China. To meet the Chinese government’s requirement for better understanding China’s role in global environmental changes, the Chinese Academy of Sciences (CAS) and the Ministry of Science and Technology of China (MOST) implemented a series of key science and technology programs, which involved many outstanding scientists in the dynamic monitoring on national-scale LUCC. In the middle of 1990s, Liu et al. implemented the project entitled ‘Remote Sensing Investigation and Dynamic Study on Resources and Environment in China’, which first integrated the RS and GIS technologies to establish the nationwide multi-temporal land-use datasets and put forward the land-use and land-cover combined classification system with six first level types and 25 s sub-level types considering the demands of macroscale investigation and RS characteristics (Liu, 2001; Zhuang, Liu, & Liu, 1999). Up to present time, a national land-use/cover database of China (NLCD) at 1:100000 scale has been constructed, which contains nationwide land-use/cover data of five periods (the late 1980s, 1995, 2000, 2005, 2010) from the late 1980s to 2010 using the same data sources and methodologies based on the spatiotemporal data platform supported by the National Resources and Environment Database (NRED) (Liu, 2001; Liu, Liu, Zhuang, Zhang, & Deng, 2003; Liu et al., 2010; Liu et al., 2014). To certify and rectify the interpreted datasets, the field survey and random sample check were correspondingly conducted, and the overall accuracy of interpretation for all the five periods was around 95%.

In addition to monitor the LUCC with all types of land use, many researches have been conducted to investigate LUCC of China’s major land-use types, such as urban expansion (Ji et al., 2001; Liu et al., 2016), cultivated land change (Zhang & Zhang, 2007) and forest change (Deng, Yin, Uchida, & Rozelle, 2012; He, Ge, Dai, & Rao, 2008; Song, Huang, Noojipady, Channan, & Townshend, 2014). The China State Land Administration funded a project in 1997 to monitor the dynamics of urban expansion in 100 municipalities throughout China based on Landsat TM images acquired for 1989/1992 and 1996/1997 (Ji et al., 2001). Liu et al. (2016) analyze the urban expansion process in 57 sample cities across China based on RS images with different sensors and spatial resolution (Liu et al., 2016). Similarly, from 1999 to 2005, the Ministry of Land and Resources (MLR, successor of CSLA) of China implemented the annual national land-use change monitoring program to monitor LUCC with RS technology. This program aimed to investigate the scales and patterns of the transition from cultivated land to urban area based on Landsat TM and SPOT images, which contributed fundamental information for policy making on land-use management and planning on LUCC at a national scale (Zhang & Zhang, 2007). Forest is a main part of the terrestrial ecosystem in China, which contributes greatly to the carbon balance. Forest cover change monitoring has been conducted based on multi-source datasets. He et al. (2008) linked the National Forest Inventory (NFI) data with the historical archives of documents and statistics to investigate the spatiotemporal forest dynamics in China from 1700 to 2000 (He et al., 2008). Deng et al. (2012) analyzed the changes in forestry area for the entire China based on RS information (Landsat TM/ETM images) from 1980s to 2000 (Deng et al., 2012). Song et al. (2014) further compared the Global Land Cover Facility (GLCF) forest cover change (GFCC) map against the NFI dataset, which showed that there existed inconsistencies concerning magnitude (Song et al., 2014).

1. Global land-cover datasets

Global land-cover datasets with high accuracy are also crucial data sources for China’s historical LUCC analysis. For the past decades, along with the promotion of global land projects, seven sets of global land-use and land-cover data have been established, including:

1. The 1 km resolution International Geosphere-Biosphere Programme DISCover land-cover dataset (IGBP-DISCover), which consist of 17 land-cover types and cover the period of 1992–1993, was developed continent-by-continent in an unsupervised classification algorithm based on 12-monthly 1 km Advanced Very High Resolution Radiometer (AVHRR) Normalized Difference Vegetation Index (NDVI) and supplemented ancillary data analysis (Loveland et al., 2000);

2. The 1 km resolution UMd land-cover dataset from the University of Maryland, which consist of 14 land-cover types and cover the period of 1992–1993. The UMd land-cover dataset was an alternative 1 km land-cover dataset that also developed based on the AVHRR individual spectral bands as well as NDVI values. Compared to the classification algorithm applied in the IGBP-DISCover land-cover dataset, the UMd land-cover dataset was developed with a supervised method where the entire globe is classified using a classification tree algorithm (Hansen, DeFries, Townshend, & Sohlberg, 2000);

3. A new global land-cover dataset for the year 2000 (GLC2000) with 1 km resolution, which was provided by international research groups coordinated by the European Commission’s Joint Research Center (JRC). This dataset was produced based on daily data from the SOPT4 Vegetation and consist of 22 land-cover types according to the Land Cover Classification System (LCCS) developed by the UN Food and Agricultural Organization (FAO) (Bartholomé & Belward, 2005); As a part of GLC2000, the land-cover data in China has been developed with the 10-day composite SPOT4 Vegetation NDVI data, DEM, and the Meteorological data, and has been delivered to JRC as a regional component of GLC2000 product (Wu, Xu, Huang, Yan, & Xu, 2004);

4. The 500 m resolution Moderate Resolution Imaging Spectro-radiometer (MODIS MCD12Q1) annual land-cover dataset, which consist of five different classification systems and cover each year from 2001 to 2012. This dataset was produced with an ensemble supervised classification algorithm based on the MODIS on-board Terra data (Friedl et al., 2002, 2010),

5. The 300 m resolution GLOBCOVER land-cover dataset, which consist of 22 land-cover types defined by LCCS and cover the periods of 2005–2006 and 2009, was developed by the European Space Agency (ESA) based on the Medium Resolution Imaging Spectrometer (MERIS) 300 m Full resolution Full Swath (FRS) products (Arino et al., 2007; Bicheron, Huc, Henry, Bontemps, & Lacaux, 2008; Bontemps et al., 2011);

6. The Global Land Cover by National Mapping Organizations (GLCNMO) dataset, which consist of 22 land-cover types defined by LCCS and cover the periods of 2003 (1 km resolution), 2008, and 2013 (500 m resolution), was developed by the Global Mapping Project organized by International Steering Committee for Global Mapping (ISCGM) using MODIS data (Tateishi et al., 2014, 2011);

7. Two sets of the 30 m resolution global land-cover dataset. The first set of 30 m resolution global land-cover dataset was named as Finer Resolution Observation and Monitoring-Global Land Cover (FROM-GLC), which comprise 10 first level types and 29 s sub-level types of land cover for the year 2010, was developed by Gong et al. (2013) using automated classification approaches based on Landsat TM/ETM+ data (Gong et al., 2013). The FORM-GLC dataset was further improved in FROM-GLC-seg, which complemented Landsat TM/ETM+ with 250 m spatial resolution time series MODIS Enhanced Vegetation Index (EVI) and other auxiliary datasets using a segmentation-based approach (Yu, Wang, & Gong, 2013). In order to further improve the accuracies for the land-cover types, Yu et al. (2014) produced an improved 30 m global land-cover map-FROM-GLC-agg (Aggregation) through aggregating the FROM-GLC and FROM-GLC-seg with two coarser resolution global maps on development, i.e., Nighttime Light Impervious Surface Area (NL-ISA) and MODIS urban extent (MODIS-urban) (Yu et al., 2014). The second high-resolution map (30-m resolution) of Earth’s land cover (GlobalLand30 dataset) that comprise 10 types of land cover for the years 2000 and 2010 was produced based on a pixel-object-knowledge-based (POK-based) classification approach (Chen et al., 2015).

As a result, a number of national and global land-cover datasets have been developed with resolution ranging from 30 m to 1 km using different classification algorithms based on satellite data derived from different sensors. However, due to the variations in spatial resolutions, temporal scales, sensor types, classification techniques, and other aspects of the data products, there are large discrepancies in the datasets for monitoring the historical LUCC change at global or regional level (Giri, Zhu, & Reed, 2005; Gong et al., 2013; Herold, Mayaux, Woodcock, Baccini, & Schmullius, 2008). For better understanding of the consistency and application of global land-cover datasets in China, many researches have been conducted to investigate the consistencies and discrepancies among the national and global land-cover datasets. For example, Liu, Liang, Liu, and Zhuang (2006) compared the distribution of forest land from the NLCD and the MODIS datasets in the year 2000 and showed that significant discrepancies exist (Liu et al., 2006). Wu et al. (2008) validated and compared the four global land-cover datasets, including IGBP-DISCover, UMd, MODIS and GLC2000 land-cover datasets, with the NLCD 2000 to evaluate the accuracy of estimates of spatial distribution and aggregated areas of cropland across China, which revealed that there exist varying discrepancies among the datasets and the MODIS dataset is best fitted for cropland estimation in China (Wu et al., 2008). Similarly, Ran, Li, and Lu (2010) investigated the similarities and differences among IGBP-DISCover, UMd, GLC2000 and MODIS land-cover datasets over China and compared them with the NLCD 2000 dataset, and showed that the GLC2000 represents the highest accuracy (Ran et al., 2010). They further produced a Multi-source Integrated Chinese Land Cover (MICLCover) map for China by combining multi-source data including NLCD2000 and MODIS2001 land-cover datasets (Ran, Li, Lu, & Li, 2012). He & Bo (2011) analyzed the consistency between MODIS2005 and GLOBCOVER (2005–2006) land-cover datasets, which showed that there exit reasonable total type area consistency at a national level while limited spatial consistency between the two datasets (He & Bo, 2011). Bai et al. (2014) assessed consistency of five global land-cover datasets (including IGBP-DISCover, UMd, GLC2000, MODIS, and GLOBCOVER datasets) with the NLCD2005 as reference data in China, and the results showed that GLC2000 has the highest agreement with NLCD2005, and the consistency in grassland and cropland is relative high while the discrepancies within forest classes is large among the five datasets (Bai et al., 2014). In a word, the consistency and accuracy of the land-cover datasets are still big concerns for the users to apply the datasets to analyze the historical trajectories of LUCC in China.

2.1.2. Land-use and land-cover change monitoring methods

1. Change detection

To monitor the historical trajectories of LUCC, change detection is an important process to identify the spatial location and types of change and quantify the changes of different periods using multi-temporal remote sensed land-cover datasets (Hussain, Chen, Cheng, Wei, & Stanley, 2013; Singh, 1989). Different land-use change detection methods have been put forward, Hussain et al. (2013) classified them into pixel-based and object-based techniques (Hussain et al., 2013).

The traditional pixel-based techniques are commonly classified into pre-classification and post-classification methods (Chen, Chen, Shi, & Yamaguchi, 2012). The pre-classification methods consists of various algorithms including image differencing, image ratioing, image regression, vegetation index differencing, change vector analysis (CVA), principal component analysis (PCA), Tasselled cap transformation (KT), chi-square transformation, and texture analysis (Alqurashi & Kumar, 2013; Hussain et al., 2013). The post-classification methods are implemented based on independently classified remotely sensed images, including post-classification comparison (PCC) and CVA in posterior probability space (CVAPS) (Chen, Chen, Cui, & Chen, 2011). These two sets of change detection methods have their own merits and drawbacks. For the pre-classification methods, it is less labor intensive to interpret the images, while some algorithms including image differing, image ratioing, vegetation index differencing and PCA methods can only generate ‘change’ or ‘no-change’ binary information without providing sufficient change trend information, it is difficult and time-consuming to select suitable thresholds for image differencing, image ratioing, PCA, chi-square transformation, and CVA, and it strictly requires the used images to be acquired in the same phenological period and from the same remote sensors (Alqurashi & Kumar, 2013; Chen, Chen et al., 2012). In contrast, the post-classification analysis provide more detailed change matrix (change direction), has no critical requirements for sensors consistency, while it always contains cumulative errors resulted from classification errors in the mixed pixels and thus requires accurate classification of individual images (Castellana, D’Addabbo, & Pasquariello, 2007; Chen et al., 2011; Chen, Chen et al., 2012). In addition, some pixel-based machine-learning algorithms, such as support vector machine (SVM), artificial neural net (ANN), decision tree and GIS-based methods have been frequently used for change detection (Alqurashi & Kumar, 2013; Dixon & Candade, 2008).

Compared to the pixel-based techniques, object-based techniques, including direct object comparison, object classification comparison and multi-temporal object change detection, are considered more appropriate for very high resolution (VHR) RS data (Blaschke, 2010; Chen, Hay, Carvalho, & Wulder, 2012; Lang, 2008). In addition, with the development of multi-scale and multi-spatiotemporal RS datasets, change detection is becoming more data driven, which inspire the emerging of data mining techniques in change detection (Boulila, Farah, Ettabaa, Solaiman, & Ghézala, 2011). Both object-based and data mining techniques have been frequently applied recently and have become the most potential alternatives of the traditional pixel-based techniques for accurate change detection (Hussain et al., 2013).

1. Land-use dynamics

Apart from the change detection methods, a dynamic degree model of land use has been proposed to analyze the spatial land-use change patterns, in which the 1 km grid percentage dataset, reflecting area proportion for each type of land-use, was generated as an effective support for the study of regional land-use dynamics to monitor the regional land use, analyze the driving forces and predict the future land-use change (Liu, Tang, Liu, & Zhuang, 2001; Liu, Liu et al., 2003). Based on 1 km grid percentage data, the rates of land conversion among different land-use types can be calculated in each grid, and the dominated conversion in each grid was identified as the conversion type of the grid (Liu, Liu et al., 2003). Further, the calculated 1 km Grid land conversion data can be aggregated to 10 km grid data frame to realize the dynamic regionalization of land-use change, which follows the primary principles: (a) land-use conversions are consistent in the same regionalization; (b) the succession of land-use change for neighboring regions was considered, involving topographical and macroeconomic differences; (c) land-use change and natural/socioeconomic conditions should be consistent in a certain region (Liu et al., 2010). Based on the dynamic degree model of land use, many researches have been conducted to detect the spatial patters of land-use change in China of different periods. For example, during 1995–2000, the regionalization of land-use change included 12 zones across China with respective consistency of land use and land-use change features, and consistency of regional physical and macroeconomic environments (Liu, Liu et al., 2003). Similarly, 15 zones of land-use changes were identified that characterize the spatial pattern of land-use changes during 2000–2005 (Liu et al., 2010), and 13 and 15 zones were identified for the two periods from the late 1980s to 2000 and from 2000 to 2010, respectively (Liu et al., 2014).

2.2. Empirical evidence of historical land-use and land-cover change

2.2.1. Short-term land-use and land-cover change analysis

China has gone through significant LUCC in the context of rapid socioeconomic development with intensive industrialization and urbanization for the past decades (Ge et al., 2004; Houghton & Hackler, 2003b; Liu et al., 2014, 2016). Based on the multi-temporal and multi-source land-cover datasets and change detection techniques, the analyses on historical trajectories of LUCC in China have been implemented during different periods. For example, Lin et al. identified the general trend and pattern of land-use change from 1949 to 1996 based on the land survey datasets, which revealed that cultivated expanded at the expense of pasture and unused land in environmentally fragile frontier regions, while it decreased as a result of rapid urbanization since the 1990s (Lin & Ho, 2003). In addition, supported by RS and GIS technologies and based on the NLCD land-cover datasets, several researches have been conducted to investigate the spatiotemporal characteristics and differences of LUCC in China at a national form 1990–2010, which revealed that cropland increased before 2000 while decreased after 2000, while woodland showed a decreasing trend before 2000 and a recovery trend after 2000, and built-up land continuously expanded while grassland kept decreasing (Deng & Liu, 2012; Liu, Liu et al., 2003; Liu, Liu et al., 2005; Liu, Tian et al., 2005; Liu et al., 2014, 2002, 2010). The main features of LUCC during each period are summarized as Table 1.

Table 1. Spatiotemporal characteristics of LUCC in China, 1990–2010.

Research periods	Characteristics of magnitudes of LUCC	Characteristics of spatial patterns of LUCC
1990–2000	Cropland increased by 2.99 × 10^4 km^2 (with about 3.2 × 10^4 km^2 of croplands being converted to other land covers including 1.5 × 10^4 km^2 to built-up land, while about 6.2 × 10^4 km^2, mostly woodlands, grasslands, and previously unused lands, were converted to croplands), built-up land increased by 0.82 × 10^4 km^2; and woodland and grassland experienced a net loss of 1.1 × 10^4 km^2 and 3.4 × 10^4 km^2, respectively (Liu, Tian, et al., 2005); and unused land decreased by 0.375 × 10^4 km^2 (Liu et al., 2014).	Large areas of woodlands, grasslands, and wetlands were converted to croplands in Northeast China, and croplands also increased in the north and northwestern regions mostly through the conversion of grasslands; while large areas of croplands were converted to urban areas in Southern China; urban land expansion differed among regions across China, with about one-third of the increase occurring in the north region, and almost another third occurring in the east region (Liu, Tian, et al., 2005); woodland decreased mainly in the major forests of Northeast China due to the harvesting of closed forests and the expansion of cropland into woodland; grassland reduced mainly due to the reclamation of grassland for cropland in Northeast, North, and Northwest China, as well as the reforestation of grassland in the south (Liu et al., 2014).
1995–2000	Cropland increased by 1.69 × 10^4 km^2 and were mainly converted from natural land-cover types such as forest land and grassland (Liu et al., 2002); built-up land expanded by 0.533 × 10^4 km^2 mostly at the cost of cropland; woodland, and grassland decreased by 1.30 × 10^4 km^2, among which a large majority of land is converted from woodland with canopy density greater than 30% and high dense grassland (canopy density greater than 50%); and water areas and unused land are increased with a small scale (Liu, Liu et al., 2003).	Large areas of cropland increased in Northeast China, North China, and Xinjiang oases while decreased in the North China Plain, Yangtze River Delta, Huanghe River band in the vicinity of Baotou and Datong section and Sichuan Basin; woodland seriously decreased in Northeast China, Southwest China and southeast coastal regions; grassland decreased mainly Central China and Northwest China; urban expansion mainly happened in the North China Plain, Beijing-Tianjin-Tangshan, central part of Gansu, southeast coastal region, Sichun Basin, and Xinjiang oases (Liu, Liu et al., 2003).
2000–2005	Cropland experienced a net decrease of 0.687 × 10^4 km^2, with decrease derived from built-up land occupancy and increase derived from unused land reclamation; built-up land increased by about 1.71 × 10^4 km^2, of which about 75% were converted from cropland; woodland increased by 0.23 × 10^4 km^2, while grassland still decreased by about 1.18×10^4 km^2 with grassland being reclaimed to cropland; water area increased by 0.16 × 10^4 km^2, and unused land decreased by 0.19 × 10^4 km^2, with decrease derived from cropland reclamation (accounted for over 42%) and increase derived from grassland degradation (accounted for 67%) (Liu et al., 2010).	Cropland decreased mainly in traditional farming regions such as southeastern coastal area and the North China Plain, slightly increased in farming-grazing transitional zone, farming-forest transitional zone, and oasis across Northeast China, Northwest China, and North China; built-up land expansion concentrated in Eastern China, with the southeast coastal areas and plain regions the hot spots for urban sprawl; woodland increased in Guizhou, Chongqing, Shaanxi, Ningxia, and southwestern mountains of Inner Mongolia, while decreased in the east of China; grassland decrease occurred in central steppe of Inner Mongolia, oasis in desert of Xinjiang, farming-grazing transitional zone of the Loess Plateau as well as Guizhou and Chongqing in western China, while it also experienced expansion in Southern Gansu, Northern Shaanxi, and northern part of Sichuan Basin mainly due to the implementation of the ‘Grain for Green’ policy (Liu et al., 2010).
2000–2010	Cropland decreased by 1.02 × 10^4 km^2, which were exploited for built-up land expansion and ecological reconstruction; built-up land increased by 3.76 × 10^4 km^2 mainly with the occupation of cropland; woodland increased by 0.237 × 10^4 km^2, while grassland decreased by 1.89 × 10^4 km^2; and unused land decreased by 0.779 × 10^4 km^2 (Liu et al., 2014)	A large amount of cropland were exploited due to built-up land expansion and the implementation of ‘Grain for Green’ project in ecologically fragile areas of Central and Western China; built-up land expansion still concentrated in the populated regions in Eastern China, with the growth rate 2.14 times of that during 1990–2000; woodland increased mainly in the Loess Plateau and the southern hilly areas due to the implementation of the six key forestry projects and particularly the ‘Grain for Green’ project; grassland increased mainly due to the reclamation of grassland in Northwest China and the regeneration of grassland in South China; unused land decreased mainly due to the reclamation of unused lands in Xinjiang and Heilongjiang and increased as a result of grassland degradation in Inner Mongolia (Liu et al., 2014).
1990–2010	Cropland increased by 1.82 × 10^4 km^2, built-up land increased by 5.52 × 10^4 km^2, woodland decreased by 0.852 × 10^4 km^2, grassland decreased by 5.32 × 10^4 km^2, and unused lands decreased by 1.15 × 10^4 km^2 (Liu et al., 2014).	Within two decades, the focus of cropland reclamation gradually shifted from the northeast to the northwest; built-up land expansion showed accelerated growth rate in the east, and such situation was spread to the central and western regions; the spatial patterns of woodland and grassland changes were strongly influenced by the national polices during different periods (Liu et al., 2014).

2.2.2. Long-term land-use and land-cover change analysis

Due to the short time span of the remotely sensed land-cover datasets, most of the researches on the historical trajectories of LUCC only cover the past four decades (Yang et al., 2014). While, human activities have affected the historical change in land cover over several centuries (Schneider & Eugster, 2007), and those centennial historical land-use changes still have strong or even overriding influence on the present earth systems with time-lagged response (Gragson & Bolstad, 2006). In this case, analyses of historical trajectories of LUCC for centennial periods are vital to obtain comprehensive understanding of LUCC in China from a historical perspective. Combined various historical data sources, such as land surveys, historical maps, aerial photographs, and various statistical reports, with various historical reconstruction methods, many researchers have examined the historical LUCC in China at a nation scale for the past centuries (Ge, Dai, He, Pan, & Wang, 2008; He, Li, Li, & Xiao, 2015; He et al., 2008; Liu & Tian, 2010; Miao et al., 2013; Yang et al., 2014). The main characteristics of the long-term historical trajectories of LUCC are summarized as Table 2.

Table 2. Spatiotemporal characteristics of long-term LUCC in China.

Research periods	Characteristics of magnitudes of LUCC	Characteristics of spatial patterns of LUCC
1960s–2000	Ge et al. reported that cropland increased from 60.78 × 10^4 km^2 in 1661 to 96.09 × 10^4 km^2 in the late 1990s with an overall increasing trend (Ge et al., 2008), while Li, He and Zhang. (2016) reported that cropland area of China increased from 55.5 × 10^4 km^2 in 1661 to 130.0 × 10^4 km^2 in 1996 (Li, He, & Zhang, 2016); He et al. reported that forest land declined by 95 × 10^4 km^2 in total from 1700 to 1998, with a trend of accelerated descending before the 1960s (reduced by 1.66 × 10^6 km^2) and a rapid increase after the 1960s, there has been a rapid increase (He et al., 2008), Houghton et al. also reported a 180 (range: 80–200) ×10^4 km^2 of forest loss during 1700–2000 (Houghton & Hackler, 2003a); grassland decreased from 258.00 × 10^4 km^2 in 1700 to 218.00 × 10^4 km^2 in 1990 with the grassland coverage rate decreased by about 4.17% (He, Li, Li et al., 2015).	Cropland expansion mainly concentrated in Inner Mongolia, Xinjiang, and traditional cultivated regions in the northeast of China due to intensive land reclamation (Ge et al., 2008; He et al., 2015); There was a significant spatial difference in the dynamics of forest. The amplitudes of increase and decrease in Western China are both smaller than the ones in Eastern China (He et al., 2008). Specifically, according to a spatially explicit reconstruction of forest cover in China, the deforestation mainly occurred in Southwest China, the hilly regions of South China, the southeast of Gansu province, and Northeast China from 1700 to the 1960s, while the reforestation occurred in most traditional forested regions of China, particularly in the Tibet Plateau, hilly regions of South China and the Greater Khingan Mountains from the 1960s to 2000 (He, Li, & Zhang, 2015).
1700–2005	Cropland increased by 39.7 × 10^4 km^2 with a growth rate of 41.8%, with a relatively slow increase from 1700 to the mid-1930s, substantial increase in the mid-1950s, followed by a steady decrease through 1990, a recovery trend in the 1990s, and rapid loss from 2000 to 2005; urban land increased by 17.1 × 10^4 km^2, with quick expansion during the periods from 1700 to the mid-1810s and from 1950 to 2005; while forest land decreased by 38.4 × 10^4 km^2 (21.8%), with a decreasing trend from 1700 to 1950 and a recovery trend since the 1950s due to an increase in plantations and regeneration of natural forest; grassland decreased by 15.0 × 10^4 km^2 (5.3%) with significant changes from the 1930s to the 1960s, and shrub land decreased slightly by a total of 3.6 × 10^4 km^2 (Liu & Tian, 2010).	Cropland expansion was distributed across China and mainly concentrated in the southwest (mostly converted from non-forest natural land cover) and northeast (mostly converted forest land); urban land experienced the largest expansion in the Eastern Plain and the southeast; forested land experienced the largest loss in the northeast (e.g., Liaoning, Jilin, and Heilongjiang), followed by the southwest (e.g., Sichuan, Chongqing, Guizhou, Yunnan, and Tibet), while had slightly increase in the north (e.g., Inner Mongolia, Shannxi, Shanxi, and Ningxia) and some parts of the Eastern Plain (e.g., Hebei, Beijing, Tianjin, Shandong, Henan, Jiangsu, Shanghai, and Anhui); Especially during 1980–2005, there was a serious cropland loss and urban expansion in the eastern plain, north, and southeast regions of China and a large increase in forest land in the southeast and southwest regions (Liu & Tian, 2010).

In addition to the above mentioned researches on the long-term historical LUCC analysis at a nation scale, several global historical land-cover change datasets for the past centuries have been developed, such as the History Database of the Global Environment (HYDE) dataset (Goldewijk, 2001) and the Center for Sustainability and the Global Environment (SAGE) datasets (Ramankutty & Foley, 1999), and the first global gridded datasets of the underlying land conversions (land-use transitions), wood harvesting, and resulting secondary lands annually for the period 1700–2000 (Hurtt et al., 2006). In order to investigate discrepancies in estimating the long-term historical LUCC in China with different datasets, Liu et al. (2010) conducted the comparisons between their new dataset with other different datasets in their estimations of historical forest, cropland, and urban land changes. The results showed that estimations of changes in cropland, forest, and urban areas range widely among different data sources, and indicated that using high-resolution satellite data as a baseline is essential for accurately reconstructing historical LUCC (Liu & Tian, 2010). Miao et al. (2013) reviewed relevant researches on China’s land use during 1700–2000, which showed discrepancies in the magnitude and rate of changes and spatial distributions of LUCC. Explicitly, all studies showed cropland increased with different magnitudes and rates, forest land overall showed a decreasing trend, and during 1950–2000, rapid urban expansion is predominant in China (Miao et al., 2013).

In sum, LUCC processes manifested itself with different characteristics in different periods. In each period, different factors, such as biophysical process, socioeconomic development characteristics and land-use management policies, drive the LUCC to different directions. However, accurate historical trajectories of LUCC still face many challenges to improve the estimation accuracy and eliminate uncertainties and discrepancies in the temporal and spatial change detections. Future research is encouraged to integrate various methods and datasets, particularly to incorporate global datasets into China’s studies, to monitor the national-scale LUCC, which is of great importance for a better understanding of the relationship between human activity and environmental conditions to formulate land-use policies.

3. Spatially explicit scenarios of land-use and land-cover changes in China

3.1. Overview of scenario-based land-use change projections

For the past decades, massive researches have been conducted to examine the regional or global historical LUCC all over the world, and further to investigate the driving forces of change and its impact on earth systems (Serra, Pons, & Saurí, 2008; Veldkamp, Wada, Ward, & Aerts, 2016). With the accumulated knowledge of the LUCC process and the driving mechanism, land-use policymakers are further interested in potential future pathways of land-use change and its spatiotemporal dynamics under different assumption about socioeconomic development, climatic change and environmental change. In this case, scenario-based analysis of future LUCC has become popular and been widely used because scenarios provide a tool to describe the alternative future situations that is driven by various factors across socioeconomic, biophysical, and geographical domains (Rounsevell et al., 2006; Veldkamp & Lambin, 2001; Veldkamp, Verburg, Kok, De Koning, & Soepboer, 2002). Based on scenario construction and analysis, the range of uncertainties, feedbacks, and thresholds that are associated with the future state of a system can be explored, which will help policymakers and researchers to cope with the uncertainties and device measures to mitigate the potential negative impacts (Havlík et al., 2011; Mahmoud et al., 2009; Mancosu et al., 2015).

Spatially explicit scenarios of LUCC have been widely applied as basic inputs for a series of natural process modeling, such as hydrological modeling (Tong, Sun, Ranatunga, He, & Yang, 2012), ecosystem services evaluation (Lawler et al., 2014), and climatic modeling (Brovkin et al., 2013). In addition, quantitative analysis of spatiotemporal dynamics of land demand and supply under different scenarios has applied to assist regional land-use planning (Zheng et al., 2012). Knowledge about possible response of the future land use is also important to regulate the land-use management policies that aim to secure food production (Fischer & Sun, 2001; Zhen et al., 2014) and to provide perceptions about future environmental problems (Sohl et al., 2012; Verburg, Rounsevell, & Veldkamp, 2006; Verburg & Van Der Gon, 2001). Overall, scenario-based projections on the spatiotemporal processes and trends of LUCC are crucial to achieve efficient utilization of regional resources, and for a sustainable management of the environment (Luo, Yin, Chen, Xu, & Lu, 2010; Ojima, Lavorel, Graumlich, & Moran, 2002).

3.2. Scenario-based future land use in China

Scenarios for LUCC projections can elucidate multiple potential pathways of future LUCC with thematic detail that represents the overall scope of land use (Sohl et al., 2012). Many scenario-based LUCC projections have been developed following the concept storylines provided by the Inter-governmental Panel on Climate Change (IPCC) in its Special Report on Emission Scenarios (SRES) (Mancosu et al., 2015; Schulp, Nabuurs, & Verburg, 2008; Sohl et al., 2012; Verburg, Schulp, Witte, & Veldkamp, 2006). Based on the SRES scenarios, scenarios of land cover in China have also been developed at a national scale. Yue, Fan, and Liu (2007) first downscaled the three Hadley Centre Coupled Model version 3 (HadCM3) climate scenarios, in zonal model of spatial change in China, based on which further derived the corresponding Holdridge Life Zone (HLZ) scenarios; and finally the land-cover scenarios of China for the years 2039, 2069, and 2099 were derived based on the method for surface modeling of land cover change (SMLC), which is developed on the basis of establishing transition probability matrixes between land-cover types and HLZ types. The results showed that the cultivated land will likely decrease and woodland will increase. These changes are associated with expected climatic changes and will also be driven by results of the Grain-for-Green policy (Yue et al., 2007). Similarly, Sun, Yue, and Fan (2012) also applied downscaled HadCM3 climate scenarios to simulate future land use in China from 2005 to 2100. They used the SMLC model to calculate area changes of each land-use type at country level and then applied the Dyna-CLUE (Dynamic Conversion of Land Use and its Effects) model to spatially simulate land-use pattern at 2 km² resolution based on the country level areas demands for each land-cover type (Sun et al., 2012).

Moreover, as the Chinese government is dedicated to pursue the environmentally friendly economic development, scenario development that links biophysical process and economic activities is necessary to help the decision-makers to evaluate the potential impacts on the environment of LUCC resulted from future socioeconomic development and land-use management. To achieve this, Xu, Jiang, Cao, Li, and Deng (2013) had developed three socioeconomic development scenarios, including Business As Usual (BAU), Rapid Economic Growth (REG) and Cooperate Environmental Sustainability (CES) scenarios, based on which they further simulated structural land-use change with the Agriculture and Land Use module (AgLU) in the Global Change Assessment Model (GCAM) from 2008 to 2100 (Xu et al., 2013), then Yuan, Zhao, Wang, Chen, and Wu (2013) simulated the spatial patterns of LUCC with Dynamics of Land System (DLS) model across China during 2010–2100 under each scenario. The simulation results showed the built-up land will expand rapidly with different growth rate, while spatial distributions of land-use changes will be consistent with each other on the whole with some regional differences under the three scenarios (Yuan et al., 2013). The scenario-based projection reflects the spatial pattern of land use/cover of China in the future, which is of importance to policy implications and scientific supporting for land-use planning and sustainable land-use management at both regional and national scales.

In addition to examine future impacts of LUCC on the long-term dimensions up to a century, short-term investigations on the impacts of LUCC have profound and central implications for the land system sustainability in the near future. For example, Verburg, Veldkamp, and Fresco (1999) developed the scenarios of land-use changes at a national scale in China for the period 1990–2010. Using the demands for the different land-use types based upon an extrapolation of trends and estimates of land-use change at a national scale for the period 1990–2000, which is developed by Smil (1993) (Smil, 1993), they applied the Conversion of Land Use and its Effects (CLUE) model that integrated both socioeconomic and geophysical factors to simulate the spatial dynamics of land-use change (Verburg & Van Der Gon, 2001; Verburg & Veldkamp, 2001; Verburg et al., 1999). Fischer and Sun (2001) spatially evaluated land productivity based on an enhanced agro-ecological zones (AEZ) model, and developed land-use change scenarios with an extended input-output model that took into account of the scenarios of future demographic and economic trends in China. The authors also analyzed policy implications based on the projection of future land-use alternatives (Fischer & Sun, 2001).

At a national scale, scenarios of LUCC can be developed through different methodologies from both biophysical and economic perspective, including climate scenarios, demographic and economic development scenarios, and varying changes in land-use demands. Spatial simulations for the different scenarios can yield diverging spatially explicit future land-use arrangements and policies that foster desired land-use change and contribute to attaining more sustainable land development. However, scenario development and simulation of LUCC at a national scale still lacks in China, meanwhile current developed scenarios mostly fixed their scope within one specific discipline, thus there should be more scenarios developed with biophysical process, socioeconomic development and land-use demands integrated, to provide comprehensive envisions of the future land-use changes for policy analysis in China.

4. Conclusions and discussion

Research on historical trajectories of LUCC at a national scale in China has greatly advanced in accuracy and coverage during the past decades. Several national and global land-cover datasets have been constructed based on RS and GIS technologies, which served to examine the spatiotemporal land dynamics and transition patterns across China from the late 1980s to 2010. The results showed that cropland increased before 2000 and decreased after 2000, while woodland showed a decreasing trend before 2000 and a recovery trend after 2000, and built-up land continuously expanded while grassland kept decreasing. In addition, based on historical archives of land surveys, maps, aerial photographs and various statistical reports, the historical trajectories of LUCC have been investigated for centennial periods, which revealed an overall increase in cropland and urban areas, while decrease in forest land during 1700–2005. Based on the historical insights, different scenarios have been developed to better understand alternative future pathways to help coping with systemic uncertainties that may be posed by climate change, economic volatility, and unexpected land system responses.

However, measuring and analyzing historical trajectories of LUCC and development of spatially explicit scenarios both face challenges. Accurate historical trajectories of LUCC still need to deal with the issues of integration of data at different spatial and temporal scales and to improve the estimation accuracy and eliminate uncertainties and discrepancies in the temporal and spatial change detections. Yet, more efforts should be focused on comprehensive scenario developments that better integrate different scientific disciplines. This review demonstrated the great value of national-scale examinations of historical trajectories of LUCC in China, while regional-scale analyses should be investigated, which are also of eminent value for informing land-use management as higher-resolution and more targeted data resources can be taken into consideration regarding geographical and socioeconomic heterogeneity.

Acknowledgments

The authors extend their gratitude to four anonymous reviewers for thorough comments that have helped improve the article. Any remaining errors of fact, argument or interpretation are our fault, and not theirs.

Disclosure statement

No potential conflict of interest was reported by the authors.

Additional information

Funding

This work is supported by National Natural Science Foundation of China for Distinguished Young Scholar: [Grant Number 71225005].