Many public and privately owned archives are replete with documents containing information about past land use and land management. Historical records provide land managers with information that could be used to understand trajectories of land use/cover change. From the domestic and foreign research, the related historical documents mainly include chronicles, timber ledgers, tax record, census data, land survey records (e.g., Public Land Survey (PLS) records), forest inventories, paleoecological records, government report,gazetteer, notes of surveyors, statistical inventories, genealogy, general description by travelers, paintings, farmers’ diaries, cultural histories, etc., such as Chinese agricultural profile estimates, Statistical compilation of Chinese forest resource, Literature of Qing Dynasty, Statistical Yearbook. Official documents often record national historical events. Researchers generally could collect some great happenings of that time concerning the overall situation, but indirectly understand the location, size of land reclamation from the genealogy. Besides, for example, a farmer’s diary may be as valuable as contemporary newspaper reports or official agricultural statistics in contribution towards the reconstruction of past land-use practices and human impacts on the land (Russell, 1997). By combining different source types, a more complete picture of historical land-cover can often be obtained (Bürgi, 2007). Based on the data characteristics from historical documents, there are mainly two reconstruction methods. Firstly, researchers could extract directly useful information about historical land cover from these old documents after qualitative or semi-quantitative analysis and then form the uniform tabular data. Before calculating the quantity of land cover in study period, they must admit that the records and other spatial data could reflect the overall trends and regional differences of land-cover area (Ge, 2003). Ge (2003) revised the “Cropland Taxes” recorded in historical documents of Qing Dynasty and then studied the change characteristics in arable quantity and the driving factors in 18 provinces of China in the last 300 years. He (2011) speculated cropland area and population of each Lu (administrative region of Northern Song Dynasty) during the mid-Northern Song Dynasty by analyzing some society factors according to “Cropland Taxes” and “the Number of Households” data recorded in historical documents. By that way he reconstructed the gridding spatial distribution pattern of cropland of that dynasty. Taking Zhenlai County of Jilin Province as the study area, Lv (2010) discussed the application of toponymy and other correlative historical data to acquire the beginning period of large-scale land reclamation and analyzed the characteristics of residential area evolvement and land development in the study area. Zhang (2011a; 2011b) collected 197 records from Chinese historical documents including gazettes, government archives, traveling notes, etc., to reveal the primary natural vegetation pattern prior to significant agriculture (i.e., the late 17th century). Fensham (1997) used land survey record produced during the late 19th and early 20th centuries to reconstruct pre-European original vegetation patterns in the Darling Downs, Queensland, Australia. Bolliger et al. (2004) assessed ecological restoration potentials of Wisconsin, USA, using the data of U.S. General Land Office original PLS. The PLS datasets are widely recognized to provide reliable quantitative and qualitative information over landscapes, carried out in the 19th century from OH to the west coast of USA. The data also have been used to construct presettlement vegetation maps in many mid-western states (White and Mladenoff, 1994; Brown, 1998; Batek et al., 1999; Schulte et al., 2002). The records of land surveyors from periods of initial settlement were used by Fensham (1997) in North America to reconstruct bygone vegetation patterns. Foster (1992) used the proprietors’ grants, deeds and tax valuation lists, sale records, Harvard Forest (HF) Archives etc., to analyze the land-use history (1730-1990) and vegetation dynamics in central New England, USA. Besides, Jönsson (2009) used the timber surveys to reconstruct the approximate landscape-level abundance of large and intermediate diameter trees of Pinus and Picea. However, there are often big differences between historical data and current statistics due to the impact of economies, policies, wars and other factors, such as statistical standards, digital characteristics, richness (Ge, 2008), therefore revisions of these historical data were usually made before they could be used (He, 2011). And in this context, the second method is proposed: under some relatively reasonable assumptions, circumstantial evidence could be obtained from historical documents and then a quantitative land-cover analysis could be done based on its relationships with other variables, such as the relationship between cropland area and the rate of population change, urban-sprawl speed and the increasing population, or the pastural area and animals numbers. Some current related research methods generally incorporate assuming-index method, analogy and proxy method, method of quantitative section-comparison and adjustment with information symmetry incorporating multi-temporal data, and then the results are adjusted according to social, political and economic factors. The analogy method means comparing the similar data from a variety of sources to distinguish their connections and differences. The proxy method means that the data recorded indirectly could be replaced by similar ones according to different statistical standards. Method of quantitative section-comparison and adjustment with information symmetry incorporating multi-temporal data, obtains the land-use and land-cover data with the same statistical standards and characteristics from the different periods and types by finding the joint study range and data normalization treatment, as described in detail in Ge’s research (2008). Ye et al. (2009) reconstructed the reclamation changes in the past 300 years in Northeast China using the methods of adjusting the data from historical literatures and the model of relationships between multi-source cultivated-land data. Goldewijk (2001) assumed that there always should be some kinds of agricultural activities where people are living, especially in the past, and therefore he used population density as an acceptable proxy for the allocation of cropland and pasture when estimating global land use change over the past 300 years. Dahlström (2008) converted the numbers of livestock into livestock equivalents firstly and then into grazing equivalents in order to estimate the pasture area.As for the missing data in the past time series, researchers often fill the data gaps using backtracking model, retrospective interpolation method, and random Markov chain model. For example, based on the history of human settlement and patterns of economic development, Ramankutty and Foley (1999a) used the cropland data within a simple land cover change model, along with the historical inventory data, to reconstruct the global data on permanent cropland areas in 5 minute resolution from 1992 to 1700. Ge (2008) analyzed the spatial pattern change of arable land in Northeast China at regular time intervals based on retrospective interpolation method using the data in the basic year as the standard. Ramankutty and Foley (1999b) used the percentage of arable land as the cellular value to study the historical land changes. In conclusion, the general process to reconstruct land use and land cover based on historical documents is shown as follows (Figure 1).

2.2.1 Reconstruction based on historical maps

Historical cartographic materials are useful for retrospective analysis of land cover patterns and their change over time (e.g., Kienast, 1993; Petit and Lambin, 2002). Historical maps (e.g., old cadastral maps, topographic maps, military maps, landscapes maps, thematic maps) containing land-cover information with temporal layer, can serve as a basis for the reconstruction of past land use and land cover, especially when they are used in a geographic information system (GIS). Comparisons of historical and modern maps highlight the major changes in land use and land cover. For some locations, historical land-cover data can be generated directly from historical maps. In general, the selection among old maps for land-use and land-cover change studies depends on when they were elaborated, their scale and the purpose of the study.

Cadastral maps were widely used in land-cover analysis, which were generated in almost all of the European countries in the 19th century. They are reliable and detailed, and can be used for evaluating the landscape microstructure. Historical cadastral maps have been used in Slovenia (Petek, 2004), in Germany (Bender et al., 2005), in Sweden (Cousins, 2002), in Norway (Domaas, 2007; Hamre, 2007), and also in the Czech Republic (Sklenicka et al., 2009). Cousins (2001) explored the possibility of using non-geometric cadastral maps from the 17th and 18th centuries to carry out land redistribution. Petek and Urbanc (2004) accomplished the reconstruction of historical rural landscape in Slovenia by taking advan-tage of maps from stable cadastre. Skaloš (2007) carried out a comparison of landscape developments in Sweden and in the Czech Republic by means of old cadastral maps. Historical cadastre maps were also used in studies of Domaas (2007) and Trpák (2002).

Besides, historical topographic maps, old vegetation maps and military survey maps are also common ones used for investigation of land-cover distribution. Stäuble et al. (2008) used the accurate topographic maps (e.g., the Napoleon map, the Dufour map and the National maps) since the mid-19th century to reconstruct historical landscape change in the Canton of Valais, Switzerland. Gimmi et al. (2011) analyzed wetland cover change over the past 150 years for the Canton Zurich, Switzerland, using information from historical and current topographical maps. The topographic sheets (“t-sheets”) of the US Coast and Geodetic Survey underline historical mapping of coastal wetlands and estuaries (e.g., Borde et al., 2003; Van Dyke and Wasson, 2005; Kearney et al., 1988). McLure and Griffiths (2002) input the data directly from historical maps and aerial photography into a digital elevation model (DEM), enabling users to create perspective views of the historical landscape of West Oxfordshire, England, from any elevation and view angle. The unique potential of the old maps was also used by the archeologists Williams (2007) and Barclay (2005). Skaloš et al. (2011) similarly researched long-term changes in the land cover of the Czech Republic from old military survey maps. Moreover, thematic maps also play essential roles in reconstructing historical specific vegetations. For example, drainage maps provide vital information on when and where wetlands vanished in the study area (Gimmi, 2011).

2.2.2 Reconstruction based on historical pictures

Pictures, especially photographs, often provide greater detail and realism about the land-cover than maps. Repeated photographs from the air or space can provide broad-scale perspectives of landscape to regional-scale land use/land cover change, thereby can overcome some biases of selecting specific and smaller localities. In the meantime, they are subject to distortion, since photographs are taken for specific purposes that may contain bias from the photographers. The use of historical aerial photographs to study land-use and cover change (LUCC) has been a well-established method in vegetation science (e.g., Swetnam et al., 1999) and has also been applied in erosion monitoring. Boucher et al. (2009) used a series of forest inventory maps, produced from aerial photographs during the past forest inventories, to reconstruct the transformation of the landscape in the 20th century. Schiefer (2007) reconstructed morphometric change in a proglacial landscape using historical aerial photography and automated DEM generation.

In conclusion, the general process (Figure 2) to make these data more accessible is firstly to scan original maps and aerial photos at high resolution, which is necessary due to high requirements for interpretation purposes and the quality of the historical maps and pictures. Secondly, the maps and pictures are rectified by choosing control points and then resampling (e.g., using polynominal equations) is carried out in a raster-based GIS to get the best fit. Finally, the maps and pictures are digitized in vector format and thus land-cover data are extracted. If the result proves to be unsatisfactory, new control points should be chosen or ncertain control points are omitted, and the transformation is repeated. The maps and pictures with GIS methods have been widely used in digitization due to the general availability of these maps and pictures in electronic form. The quality of the research depends on the accuracy of the historical maps and their potentiality to be integrated in a GIS.

The natural archives, which are also called “Geoarchives” or “proxy records”, are those “recorded” by earth-system processes, such as colluvial sediments (e.g., pollen, charcoal, plant macrofossils, and phytoliths, in ponds, lakes, bogs, and soils; Jackson et al., 1997), animal deposits (packrat middens; Betancourt et al., 1990), annual plant and animal growth cycles (e.g., tree rings and coral layers; Fritts, 1989), and other layered records (e.g., ice cores; Thompson et al., 1998) (Swetnam, 1999). They are subject to the filtering of past environmental information through physical and biological variables. This method has been widely used in reconstructing historical land use and land cover in the time scale of thousand years or even longer by palynologist, ecologist, archaeologist and geologist. Especially, the pollen and tree-ring analyses, which can allow the reconstruction of the local and regional vegetation composition and can reflect a region’s changing human activities, have been widely used. Fossil pollen, as the relic of historical plants, has been used extensively as an indicator of historical vegetation (e.g., Davis, 1973; Chen and Ni, 2008; Broström et al., 2008; Scharf, 2010). Reconstruction of vegetation from pollen analysis assumes that the spatial scale of vegetation represented by pollen assemblages can be understood, and that the inferred vegetation abundances can be translated into a vegetation map or other vegetation characteristics in specific applications. Over the last decades the development of a theoretical framework and models of pollen-vegetation relationships has made such spatial vegetation/landscape reconstruction feasible (Sugita, 1994; Dahlström, 2008). Research on modern relationships between pollen assemblages and human-induced vegetation related to various traditional human activities provides the foundation for an accurate interpretation of past human impact on vegetation and quantitative reconstruction of human-induced landscapes from fossil pollen data. There have been many examples of such research (Edwards, 1991; Hjelle, 1997; Gaillard et al., 2008; Brun, 2011; Li, 2013). The prime sources of information about forest development are tree rings (Esper et al., 2007), which allow precise dating of events. Researchers have gradually discovered that tree-ring index could evaluate the Normalized Difference Vegetation Index (NDVI) which reflects vegetation productivity, indicating that using tree-ring data could reconstruct historical land-cover. Malmstrom et al. (1997) had previously shown the potential of using the tree-ring width parameter for comparison with NDVI-based NPP estimates. Yuan (1989) reconstructed the long sequence of fodder grass output from tree-ring chronologies based on the relationships among tree-ring chronologies, fodder, grass and summer precipitation.

The early land use and land cover models are mainly the traditional natural cover models, carrying out climate-related method to project the vegetation distribution, such as General Circulation Models (GCMs) (Goldewijk, 2004) and Lund-Potsdam-Jena Dynamic Global Vegetation Model (LPJ) (Kaplan, 2012). These models project vegetation distribution mainly based on the relationship between temperature, precipitation and other climatic factors, but they do not fully consider the other factors such as soil, topography and human disturbance, which limit the system’s accuracy. Subsequently species-statistics method and growth-model have gradually replaced the climate-related method, such as the Gap model developed by Botkin which could simulate the biological changes in vegetation dynamics while considering non-climatic factors. Except for the aforementioned studies, some researchers have developed biophysics models based on the crop physiology while others developed crop-rotation models according to the interrelationship among various land-cover types (Benitez, 2004).

With the leading role of human activities on land use/cover change becoming increasingly prominent, historical reconstruction models of land use also give more consideration to both social and cultural factors. Initial studies mainly used a simple mechanical model to generate past projection at some certain assumptions without considering the driving forces and driving mechanism in the process of land use and land cover change. For example, Petit (2002) produced backward projections based on a mechanistic model which computes the demand for different land uses under the equilibrium assumption between the production and consumption of resources. Besides, he also believed landscape change processes can be simulated using linear, stochastic techniques such as Markov chains when a time series of land-cover maps is available. Ramankutty and Foley (1999a) used a ‘‘hindcast’’ modeling technique to extrapolate the data backwards in time by combining it with a compilation of historical cropland inventory data to create a dataset of croplands from 1700 to 1992.

Since the 1990s, land use/land cover reconstruction models have been developed along three important trends. Firstly, time dynamic simulation has been combined with spatial pattern analysis and geographic information systems. Secondly, remote sensing data have been widely used. Thirdly, natural elements have been integrated with social, economic and cultural elements (Chen, 2002).

Historical reconstruction of land use/land cover contains two important parts: the first part is to make clear the quantity of each land-cover and the second is to reconstruct the spatial pattern which is a process of spatial allocation. Numerical reconstruction could provide quantitative information for historical land-covers’ areas, which is the basis of land use reconstruction and also the premise of spatial reconstruction. Spatial reconstruction aims to distribute statistical data from numerical reconstruction into the geographical locations according to certain principles or approaches. Studies on historical reconstruction of land use/land cover usually aim to make clear the historical land-cover spatial distribution. As it is very difficult to obtain the land use/land cover data in the past, especially before the advent of satellite remote sensing, many researchers prefer to carry out land suitability evaluation using environmental background data (e.g., climate, soil, topography) and socioeconomic data (e.g., population and crop yield). Then they use land suitability evaluation map to exhibit the possible spatial distribution of historical land-cover. For specific spatial allocation, downscaling can be used in the descending order of distribution probability under the control of total land-cover area and other distribution factors (Figure 3).

With modern remote sensing technology developing quickly, it becomes feasible and convenient to carry out earth observation in large-scale with high resolution. Thus an important reconstruction method has been developed to produce backward projections by nalyzing the land use and its change since the advent of remote sensing. This method is under the assumption of similarities between historical spatial pattern of land use and the present one. There are different approaches in this method. Zhu (2012) believed the approaches had various effects on controlling the similarities and summarized them to three types: totally dependent, partially dependent and dynamically dependent. Totally dependent approach was reasonably used in land-cover reconstruction closely related with human activity, such as arable land and pasture land in centennial time-scale. In order to avoid “using today’s pattern for historical reference”, partially dependent approach was based on an assumption that the area of historical agricultural activities could not exceed the present one. Dynamically dependent approach had the most sophisticated algorithms taking the time effect into consideration, firstly proposed by Goldewijk (2011) in his HYDE 3.1 (History Database of the Global Environment) application. The driving force of the land use process has to be clarified and the land use change trajectories have to be analyzed before using the partially dependent approach. Then a land use forecasting model (e.g., cellular automata (CA), Geomod) as a carrier can be used to produce backward simulation. The difference between historical reconstruction of land use/land cover and land-use change prediction lies in that the former could use historical maps, historical data and literatures, to revise the simulation model results. Besides, the former is a backward process while the latter, land-use change prediction, is a forward process to the future. Yet, both of them describe land-cover change over time and thus have the similar modeling theories and methods. Poska (2008) reconstructed the proportions of four classes of land-cover characteristics of cultural landscape: habitation area, arable land, grassland and woodland at Rõuge, South Estonia, for seven periods. Mao (2010) also used this model to rebuilt land use and land cover of Baicheng city, China, from 1890 to 2029 and then studied the soil organic carbon pool. Bai (2004, 2007) used TM (thematic mapper) and MSS (multispectral scanner) images, physical environmental background maps and socioeconomic data to design a new diagnostic model of land utilization spatial information in historical period based on CA model. Pontius (2003) reconstructed the landscapes of Massachusetts Ipswich River basin in 1951 using the Geomod model. Lv (2012) also discussed the historical LUCC researches in Northeast China using this model based on the application of toponymy and other correlative historical data to acquire the information of the beginning period of large-scale land reclamation. Pontius (2001) simulated the progressive loss of closed-canopy forest in Costa Rica for 1940, 1961 and 1983 with Geomod2 and then simulated the location of land-use change over several decades.

Studying land-reconstruction methods in depth becomes the key of studying the long-period land use and land cover dynamics. Especially, studies by modeling in particular may be prone to promoting proper forecasts, projections or predictions (Veldkamp and Lambin, 2001). The more complex the system is and the more factors are involved, the more difficult it is to make acceptable predictions.

Each single method has limitations and advantages, and subsequently most studies benefit from using multiple approaches (Foster et al., 1996). Researchers usually combined several of the above methods to reconstruct the composition and structure of land cover. By combining different data sources, they often obtained a more complete picture of land-use and land-cover change. For example, historical rural land use and settlement structure have large regional and temporal variations and the reconstruction of a specific historical land-cover can only be done by a high-resolution local case study (e.g. Sarmaja-Korjonen, 1992). The critical assumption for such reconstruction is the availability of relevant and reliable historical documents. As historical sources are chronologically and regionally unevenly preserved, the combined usage of different records is inevitable (Petit and Lambin, 2002; Nielsen and Odgaard, 2004). Large-scale cadastral maps, historical census data, taxpayers lists and plough-land revisions contain valuable information about historical land use, settlement structure, demographic and economic parameters (Foster et al., 1998; Fuller et al., 1998; Cousins, 2001). Recognizing the challenges of assessing the accuracy of historical mapping (Harley, 1989), Robin (2007) used multiple document types to enable triangulation, or cross-referencing, of independent data sources to reconstruct historical land cover in California’s Santa Clara Valley and the findings suggested that the triangulation of overlapping from different data sources might be more useful than the independent data sources. Zhang et al. (2006) rebuilt LUCC spatial-temporal distribution over the past century in Northeast China by using many sources of data and by modeling, such as TM and MSS images, physical environmental background maps including terrain, climate, geology, soil, vegetation and hydrology as well as socioeconomic statistical data. To reconstruct the historical landscape ecology of an urbanized California valley, Grossinger (2007) examined several thousand historical records at over two dozen archival institutions and collected materials and their archival provenance, including Spanish explorers’ accounts, Mexican land grant maps and court testimony, PLS records, early city and county maps, early American journals and histories, 19th-century paintings and drawings, US Geological Survey maps, US Coast and Geodetic Survey maps, landscape paintings and photography, aerial photography, soil surveys and ecological research/collections. Hamre (2007) used an old cadastral map, a field survey of the present landscape and GIS to quantify and analyze land-cover changes in a western Norwegian cultural landscape since the time of land consolidation (1865) until 2002; Skaloš (2010) tested a method for analyzing long-term structural changes in non-forest wood elements using five main sources of data: official statistical data on land cover in the Czech Republic, stable cadastre maps from 1839 and 1843, aerial photographs, field maps and the digital map of the territory (DMU) on a scale of 1:25,000.

Moreover, researchers face the difficulties like inaccuracy reconstruction and unverified historical data due to the limits of their own disciplinary knowledge. With the widely application of multidisciplinary studies, they gradually become aware of the fact that different types of historical data might provide useful information for different land cover types, and then studied the historical land use reconstruction by integrating paleoecology, paleogeography, historical landscape science, palaeobotany, palaeophysiography, history and other disciplines. By this way they could reconstruct historical land-cover from multiple perspectives, complement the missing data, verify reconstruction results and then improve reconstruction results. For example, in southern Sweden, information about land-cover collected from different historical sources including old maps was compared with pollen diagrams from ten sites and other palaeoecological records (Berglund, 1991; Nielsen, 2004); Wulf (2010) used a combined approach of GIS-techniques and historical reconstructions from archives to study forest cover changes in the Prignitz region, Northeast Germany, between 1790 and 1960 in relation to soils and other diving forces. “BIOME 300” project, a joint LUCC-PAGES (Past Global Changes) initiative (Leemans et al., 2000), was initiated to reconstruct historical land use/land cover datasets for the past 300 years using data from several disciplines such as paleo-data from pollen records. The participation of scientists from around the world with expertise in their own particular regions of study is required in these global efforts. In short, integration of different methods and data sources could extend the environmental change information in different spatial and temporal scales, which has good prospects in future applications.