Introduction

There is consensus that both natural (bio-physical, e.g., relief energy and soil type) and anthropogenic (socio-economic, e.g., population density and roads) factors are important for understanding land use/cover changes (LUCC) worldwide (Goldewijk 2001; Lambin and Geist 2006). It is evident that human impact in the past has led to extensive landscape alterations on a global scale (Houghton 1994; Goldewijk 2001) and, as a consequence, may have weakened the relationship between potential site conditions and the realised land use/cover (Foster et al. 1998; Jongman 2002). For example, the use of fertilisers enables agriculture at sites that would otherwise not be efficiently cultivable (Born 1989). Despite the huge and long-term impacts of humans on landscapes, relationships between natural factors and expected land use/cover can nonetheless be demonstrated (Cousins 2001; Hall et al. 2002; Zachwatowicz and Giętkowski 2010). In Sweden, semi-natural grasslands with shallow soils <50 cm deep, have been used for decades (Cousins 2001), and, in Poland, areas of afforestation/reforestation occurring within the last 150 years were mainly associated with sites not optimal for productive agricultural use (Zachwatowicz and Giętkowski 2010).

Humans will most likely continue to exert major influences on LUCC in the future. For scientific and planning purposes, particularly on a regional scale (Bender et al. 2005), it is of great interest to know how land use (LU) and land cover (LC) may appear in the coming decades (Lambin and Geist 2006). In this context, examining the past can help to detect fundamental human-environmental relationships, including continuities and discontinuities (Bork et al. 1998; Petit and Lambin 2002). Insights from the past (ca. the last 300 years) are necessary to predict future LUCC (Antrop 2005; Lambin and Geist 2006). With a long time span of more than 200 years, we typically face problems such as the lack of adequate historical maps and lack of historical data on natural and anthropogenic factors. Many studies on LUCC considering natural and anthropogenic factors have covered only a few decades (e.g., Hietel et al. 2005; Millington et al. 2007). However, several attempts have been made to reconstruct long-term LUCC on a global scale (Ramankutty and Foley 1999; Goldewijk 2001) and a regional scale over a time span of at least 100 years (Skånes and Bunce 1997; Foster et al. 1998; Petit and Lambin 2002; Bender et al. 2005).

Europe has undergone highly dynamic LUCC starting from the traditional landscapes in the eighteenth century, followed by the industrial revolution in the nineteenth century and further development to contemporary landscapes after the scientific and technical revolution in the twentieth century (Antrop 2005). LUCC in European regions can be best understood if we look back more than 200 years; otherwise, effects of the distinctive turning point from the eighteenth to the nineteenth century on the landscape can be overlooked, such as the changes in ownership structures in rural areas in east Elbian Prussia with its consequences on LUCC, as demonstrated in detail below.

The main driver of LUCC on a global scale was the enormous increase in the human population with a subsequent major expansion of urban areas and croplands (Goldewijk and Ramankutty 2004). The area of croplands doubled worldwide between 1850 and 1990 (Houghton 1994). Population increase was also an important driver of LUCC at smaller spatial scales, but there were regional differences in the drivers of LUCC, as regions have unique environmental conditions, population and ownership structures, economic activities, landscapes and social histories, among other factors (Ravichandran 1999; Antrop 2005).

Iverson (1988) found that land patches in a region of Illinois (USA) between 1820 and 1980 were controlled by human influences and relatively few were controlled by topographic and hydrologic features. Bender et al. (2005) observed that regions in southern Germany exhibited a weak relationship between cleared areas and soil fertility for the period from 1850 to 2000 as a result of land reallocation. For the Belgian Ardennes, Petit and Lambin (2002) found strong interactions between LU changes (1775–1929) and the history of societies. However, Hall et al. (2002) showed at regional scale that environmental conditions were primarily responsible for the broad pattern of variation in vegetation composition in Massachusetts (USA) over a 300-year period. In Hungary, Biró et al. (2013) observed that afforestation between 1783 and 2009 in a Pannonian region primarily occurred on sandy substrates, as expected. The varying outcomes of these previous studies underpin the need for studies on the dynamics of LUCC and their drivers on a regional scale (Zachwatowicz and Giętkowski 2010; Tian et al. 2014).

However, questions remain with regard to LUCC in our study area. The region is well known as the granary of Berlin (Enders 1992; Hagen 2002) since long before the end of the eighteenth century. It is remarkable that small sub-regions have undergone major changes in forest areas and open the LU/LC patterns did not create an equilibrium between profitable and unprofitable agricultural land until 1780? If so, why do high rates of LUCC occur in some sub-regions and not in others? Do anthropogenic and natural factors play fields (mainly arable fields) in the last 200–300 years. Does this mean that an equal role in these patterns?

Many approaches have been used to model LUCC (Jones 2005). These models cover a range of model types, including Markov models and logistic function and regression models. The latter has often been applied to explain LUCC over several decades (e.g., Millington et al. 2007; Bonilla-Moheno et al. 2012; Cabanillas et al. 2012) but rarely for periods of 50–100 years or longer (e.g., Aspinall 2004; Zak et al. 2008). In a given study area influenced for hundreds or even thousands of years, it is likely that the composition of LC types cannot be explained by only a few independent variables or one model (e.g., Aspinall 2004; Lambin and Geist 2006). Decisions on retaining or (re-)establishing specific LC types have always depended on the suitability of a site for a specific LC type/LU, including relief energy and nutrient and water availability, and on the feasibility of the LU, which may be based on the number of farmers/foresters living in the area, access to LC types via roads and agricultural equipment (e.g., Enders 1986; Aspinall 2004; Hawbaker et al. 2004). Aspinall (2004) has shown that it is important to consider as many factors as possible in such complex analyses. However, if a regression analysis is applied to such data, as typically performed (e.g., Bonilla-Moheno et al. 2012; Cabanillas et al. 2012), problems of collinearity and overfitting may arise (Legendre and Legendre 1998). Another problem is the difficulty in determining the minimum adequate model because it is likely that more than one model will fit the data well (Grueber et al. 2011). Instead of finding “the best model” (one hypothesis) implying the risk of overlooking important factors for LUCC, it is possible to apply model averaging (Grueber et al. 2011; Johnson and Omland 2004). This approach allows for the formulation of several hypotheses by generating a set of top models that are meaningful as long as each top model is reasonable (Grueber et al. 2011).

Our main aim was to analyse the determinants, namely, natural and anthropogenic factors, of the main LC types (arable fields, grasslands and forests) for three periods covering the time span before the industrial revolution (~1780), after the industrial revolution (~1890) and after the scientific and technological revolution (~2000). To facilitate these analyses, we considered landscape windows of equal size but differing in changes of forest-open land areas. Changes may have included conversion from forest to arable fields or grasslands and vice versa. We hypothesised the following:

(i) the main LC types (arable fields, grasslands and forests) were determined by different natural and anthropogenic factors during the three periods (1780, 1890 and 2000) and

(ii) the effects of natural and anthropogenic factors on the main LC types are different in highly dynamic and more stable landscape windows.

Methods

Study area

The Uckermark region (52°52′N–53°23′N, 13°20′E–14°12′E) is located in northeastern Germany (ca. 50 km north of Berlin) and encompasses approximately 305,800 ha. Brandenburg within the Uckermark was located in the eastern Elbian part of Prussia (1701–1918, Melton 2000). The system of land tenure in this part of Prussia was dominated by so-called manors owned by noblemen (“Gutsherrschaft”). These manors were characterised by a market-oriented estate agriculture carried out on a large scale by noble owners, and peasants provided large amounts of unpaid labour on these estates (Melton 2000). In the late eighteenth century, most of the Brandenburg estates were in the middle range of approximately 200–1000 ha (Melton 2000). Peasants also had estates for agriculture, with the farm size ranging from small to large (≤20 to 100 ha, Melton 2000; Hagen 2002). Together with cottagers and wageworkers, who owned no estates (Hagen 2002), three-fourths of the population lived in rural areas at least until the 1840s, while towns remained physically small (fixed town walls with town gates, Blackbourn 2003). A noble owner lived with his family and domestic servants in large premises (nearby or in a village), whereas peasants (farmers) lived in simple, small houses in the village. Cottagers and wageworkers often lived in colonies outside of the village. All these building complexes are considered as settlements.

Before the agrarian reforms started ca. 1850, many villages with surrounding land were under the rule of noblemen. Under the ownership of the noblemen, some villages (at least two) and the land tenanted by peasants were joined to form a commune (Melton 2000; Hagen 2002). However, particularly on historical maps, a commune was usually named as a manor. This practice had changed by the end of the nineteenth century, and the “Gutsherrschaft” had completely collapsed by the end of World War II. For that reason, we use both terms synonymously.

The region is characterised by a sub-oceanic to sub-continental climate with a mean annual temperature of approximately 8.3 °C (Angermünde 1961–1990) and 9.3 °C (Falkenberg 1981–2010), respectively. The mean annual precipitation is approximately 520 mm (Angermünde and Prenzlau 1981–2010) (Gränitz et al. 2008; DWD 2013). Glacial deposits of Weichsel age prevail, with ground moraines consisting of sand and loam. In the hilly region, there are many glacial relicts, e.g., kettle holes and lakes (Lippstreu et al. 1997). The main soil types at higher elevations are haplic luvisols (~40 %), calcaric regosols (~25 %), cambisols (~15 %), and albeluvisols (~6 %), with gleysols (~8 %) at lower elevations, according to the IUSS Working Group WRB (2006).

Our theoretical framework considered several facts well known for central Europe, where the area of this study was located; (1) the impact of humans on the landscape for at least 6000 years, (2) traditional LU/LC for at least 1000 years (e.g., wood pasture and three-field rotation), and (3) two major stages affecting LUCC in the past (Antrop 2005), which were the industrial revolution (the last 150 years) and the scientific and technological revolution (the last 50 years). The Uckermark region has experienced these impacts and the resulting LUCCs. Understanding the actual pattern of LC/LU requires considering these impacts over a long time period.

Selection of landscape windows

The approach of landscape windows or test sites of equal size enables the comparison of different (sub-)regions (Herzog et al. 2006). Several studies worked with 4 × 4 km or 5 × 5 km sites (e.g., Herzog et al. 2006; Biró et al. 2013). Because our basic units are manors/communes with a mean size of approximately 750 ha, windows of 25 km2, we had only 3.5 units, which was not sufficient for statistical analysis. However, there is a risk of losing the clear contrast in changes of LC types with an increased size of windows. Therefore, we selected landscape windows of 10 × 10 km in size (Fig. 1) as a trade-off to provide sufficient units (13.5 on average) and best contrasts. The selected four landscape windows are representative of sub-regions with both high and low percentages of forest-open land changes (Figs. 1, 2) and cover all main soil types occurring in the Uckermark region (Table 1). High changes mean that large areas of different LC types have been converted at least once throughout the last 220 years, which did not occur in more stable windows, as shown in Appendix A of ESM.

Fig. 1
figure 1

Development of forest-open land in the Uckermark region within the last 220 years. Ancient (forest continuity >220 years), old (forest continuity >150 yrs), and recent (forest continuity <100 years) forests and open land (mainly arable fields) in the Uckermark region. Landscape windows are shown as frames with number 1–4

Full size image
Fig. 2
figure 2

LC composition (in %) of the four landscape windows

Full size image
Table 1 Substrates in the four landscape windows
Full size table

Compiling data of LC

We used historical and recent maps from three discrete time periods, roughly corresponding to 1780, 1890, and 2000, to reconstruct long-term LCC during the past 220 years. Specifically, we rectified the hand-drawn Schmettau map (1767–1787, scale 1:50.000) and used the undistorted Prussian Land Survey map (originating from 1879 to 1902 for Brandenburg, scale 1:25,000). Herein, these maps approximately refer to 1780 and 1890. The hand-drawn Schmettau maps have been rectified, and the undistorted Prussian map has been used with permission from the Eberswalde Forestry State Center of Excellence (LFE) to reconstruct the locations of historical forest areas in Brandenburg (http://www.brandenburg-forst.de/webgis/). Because the maps of the LFE contain only forest areas, we digitised all other LC types for the four windows to make them complete. Both maps (Schmettau map and Prussian Land Survey) were georeferenced and vectorised in ArcMap 10.0 (ESRI, Kranzberg, Germany) for the extent of the windows. The LC derived from biotope mapping (scale 1:10,000, LUA 2007) was used as a recent reference; herein, this LC refers to approximately 2000. The different legends of the maps were aligned to one another and reduced to six LU types: arable fields, grasslands, forests, water bodies, settlements and others (e.g., churches with churchyards, large stables). Because we wanted to rely on comparable intervals between maps and because no reliable maps exist for the period from 1780 to 1890, we did not consider maps published between 1890 and 2000. The result of the intersection is shown in Fig. 1.

Compiling data of potentially explanatory variables

Traditionally, the Uckermark region primarily consisted of agricultural lands governed by manor owners. A comparison of the manor boundaries (rural communes) on Prussian maps revealed that the boundaries did not frequently change between 1608 and 1806 (Schulze et al. 1969), with some remaining even until the end of the nineteenth century. This stable state was proven by a comparison with the boundaries of rural communes on Prussian maps. Considerable changes in the administrative borders from the beginning of the twentieth century until the present make it difficult to obtain actual data at the commune level. However, for the old manors, we found so-called historical address books, municipality encyclopaedias and published agricultural statistics, which provided detailed data and information, e.g., on the total area of the manor, area of arable fields, grasslands and forests, and number of inhabitants and livestock (e.g., Bratring 1805; Boeck 1861; Fidicin 1864; Königlich Preußisches Statistisches Landesamt 1905; Enders 1986; Kölling 1999).

A particular challenge was the discovery of historical environmental data and information; we did obtain some information, but not all of it referred back to 1780. For example, we found a table on soil classes published in 1864 (Kölling 1999). This table contains information revealing that the parent material hardly changed over a period of 200 years; e.g., sandy substrate in the eighteenth century still remained as such until the nineteenth century. Therefore, we used these data for the 1780 period (Table 2). Detailed information on the original data, transformations and sources of all the variables described in Table 2 can be found in Appendix B of ESM. The values of the factors were corrected for the proportion of those communes not entirely located within the respective window.

Table 2 Natural and anthropogenic factors as potential explanatory variables for LCC from 1780 to 2000
Full size table

Data analysis

The definition of potentially explanatory variables was developed based on a comprehensive review of studies on this topic (e.g., Hietel et al. 2005; Bonilla-Moheno et al. 2012; Cabanillas et al. 2012) and the inclusion of expert knowledge sampled over the last 10–15 years (Flade 2006). Compilation of the data resulted in 10 variables for each of the three periods (Table 2). The observed long-term explanatory variables were in a few cases not identical for all periods. However, the number of natural and anthropogenic factors was nearly the same for all periods (Table 2).

To analyse the relationship between LC types and natural and anthropogenic factors for each of the three time periods, we used R (R Core Team 2015) and the function lme (package “nlme”, Pinheiro et al. 2015) to fit linear mixed-effect models. As fixed effects, we entered the 10 explanatory variables (Table 2) and changes (coded as 1/0 = high/low percentages of changes in forest-open land); we utilised “windows” as the random effect in random intercept models. Prior to the analyses, the LC data were arcsine transformed (Ahrens et al. 1990) and the area log transformed to meet the criteria of homoscedasticity and normality for the residuals in both the full models and each of the models selected for model averaging. Although high and low changes only became visible in 2000, we included changes in models for 1780 and 1890 as a proxy for the potential of LCC already completed in 1780 and 1890.

We aimed to reduce the computing time and fitted model sets starting with a maximum of three significant terms. If more variables were entered into the model, we proceeded when ΔAICc (“Akaike Information Criterion”, Sakamoto et al. 1986) was reduced by >2.5 compared to the former model with the lowest AICc. In most cases, we stopped this procedure with a maximum of five variables, although seven variables (including interactions) was the exception (see Appendix C of ESM). We selected the top models using AICc because it corrected for small sample sizes compared to the number of factors (herein, 65 communes and 10 factors), as recommended by Burnham and Anderson (2003).

Model averaging is recommended in cases in which models have a weight of <0.9 (Grueber et al. 2011). This was the case for all the dependent variables except for arable fields in 1780 (Table 3). Model averaging was carried out using the R function model.avg (package “MuMIn”, Bartoń 2015). Only models with ΔAICc <2.5 compared to the model with the lowest AICc value were considered (Grueber et al. 2011). To assess the model fit, we calculated the marginal and conditional R 2 according to Nakagawa and Schielzeth (2013). Because our focus was on the fixed factors, we calculated only the marginal R 2 for variance partitioning of the anthropogenic and natural factors.

Table 3 Summary results after model averaging: effect of parameters on LC of arable fields, forests and grasslands for 1780, 1890 and 2000
Full size table

Results

Land cover composition and its changes

In 1780, arable fields were the dominant LC type in all landscape windows, followed by forests, whereas grasslands played only a minor role, except in window 2 (>15 %). The situation did not change in 1890, with the exception of window 2, where forest became dominant. In 2000, arable fields were dominant only in windows 1 and 3, whereas forests reached higher proportions in windows 2 and 4. Grassland did not exhibit remarkable proportions throughout the periods but always exceeded 10 per cent at least in window 2 (Fig. 2).

The pattern of forest-arable field distribution is the result of a directed landscape development; the most productive soils in the northeast have been converted from forest to arable fields step by step over a long time period and were used as such several 100 years before 1780, whereas the rather sandy soils in the southwest are covered by large forests (Bayerl 2006). On this scale, we can confirm previous observations that forests have remained on less productive soils (Bamberg 1909; Foster 1992; Andersen et al. 1996). This general pattern remained more or less stable over the last 220 years (Fig. 1).

Top models

Model selection using AICc revealed two or three top models for all LC types, with the exception of arable fields in 1780 (Appendix C of ESM). In all but two models, at least three but not more than five variables were included (interactions not considered). For arable fields in 1780 and grasslands in 2000, only one or two variables were retained in the models, respectively (Appendix C of ESM).

LC of settlement and similar terrestrial LC types (e.g., church with churchyard, larger stables and “Vorwerke” as part of the manors outside villages, SEOT) were among the anthropogenic factors, including settlements and other related areas, important for approximately 78 % of nine models, with a positive effect on the cover of arable fields and grasslands but a negative effect on forest cover. Among the natural factors, cultivation unsuitability was important for approximately 56 % of all models, with a negative effect on the cover of arable fields and grasslands but a positive effect on forest cover. Change was an important factor for all LC types in 1780 and became less important in 1890 (in two models) and 2000 (in one model). This result was also valid for interactions of various variables with change.

Most important determinants for the cover of arable fields, forests and grasslands from 1780 to 2000

We found that both natural and anthropogenic factors influenced the LC composition for the three time periods, although natural factors were the most important determinants (Table 3). Specifically, Table 3 reveals that the unique variance explained by anthropogenic factors reached 2.1–18.7 %, whereas that of the natural factors reached 0.2–43.4 %. In four of six models with both types of factors, the natural factors reached higher values than the anthropogenic factors. However, for forest cover in 1780 and for grassland cover in 1890, the anthropogenic factors settlements and inhabitants exhibited a clearly higher explained variance than did the natural factors (Table 3).

Arable field and forest cover exhibited several important determinants in common in all three periods. Because the two LC types can be seen as counterparts (Pearson’s correlation in highly dynamic windows was −0.928, p value <0.001, and in more stable windows, it was −0.726, p value <0.001), it is not surprising that common determinants will be positively or negatively associated with arable field cover and vice versa with forest cover. Thus, whereas arable field cover increased with settlement areas and road density, forest cover decreased. In contrast, whereas forest cover increased with areas unsuitable for cultivation (cusuIn), the arable field cover decreased. The arable field cover was additionally negatively associated with the proportion of water bodies (WA) in 1890 and 2000, whereas forest cover was also positively associated with relief energy (relIn) in 1890 and 2000.

Change and its interactions with other factors had the largest effect on forest cover in all three periods but only in 1780 for arable field cover. Thus, arable field cover was greater in highly dynamic windows. In the case of forest cover, negative interactions for 1780 and 2000 indicated that cover was lower where the proportion of settlements (SEOT) and road density (ropa) was lower in both highly dynamic and more stable windows. The interactions for 1890 and 2000 revealed opposite effects for cultivation unsuitability (cusuIn) and indicated that forest cover was higher in more stable windows, where the proportion of areas unsuitable for cultivation was higher and vice versa.

Grassland cover was the least predictable of all LC types, as the marginal R 2 values of the averaged models were all relatively low (Table 3). Furthermore, there were few determinants in common among the periods, indicating that grassland as an LC type of minor importance was determined by different factors throughout the time period studied. However, it is expected that grassland cover increased with higher proportions of wet areas and those with high ground water levels (BOWE and gwIn). The larger gwIn range in highly dynamic windows (Appendix D of ESM) indicated more relatively dry soils limiting grassland cover. Wet areas (BOWE) reached higher cover in windows with high dynamic rather than more stable expressed as positive interactions between these variables for 1780 (Table 3). Due to the low explained variance and many non-significant factors, this LC type will be not discussed.

Discussion

Compared to Germany as a whole and the European Union, the Uckermark is an outstanding region (but not the only one), with a considerably higher proportion of arable fields (Table 4). This situation can be explained by the high proportion of very productive soils (Phaeozems) that were agriculturally exploited for thousands of years (since the Neolithic Age; Fischer-Zujkov et al. 1999).

Table 4 Proportions of LC types and number of inhabitants in Germany, the Uckermark and Berlin
Full size table

In Germany, until the industrial revolution, three-fourths of all people lived and worked in rural areas, whereas only one-fourth lived in urban areas (Blackbourn 2003). This is also true for the Uckermark region and Berlin. With the development of industry and agricultural machines, in parallel with a strong increase in the human population, the rural–urban-inhabitants relation has been inverted in that area (Enders 1986), which is in accordance with developments in other regions as well as worldwide (Ramankutty and Foley 1999; Petit and Lambin 2002; Blackbourn 2003). A strong increase in population is often accompanied by a decrease in forests and increase in agricultural land in order to produce sufficient food for the human population (e.g., Andersen et al. 1996; Tian et al. 2014). Nevertheless, not all regions changed greatly in their pattern of LC types due to population pressure. For example, the Pińczόw area in Poland remained relatively constant in its forest-open land distribution from 1840 to 2000 (Zachwatowicz and Giętkowski 2010). As in Poland, the Uckermark is a historically rural region, where a balance has been realised between natural conditions and LC composition on a broad scale for a very long time. However, despite this general condition, there are breaks in the landscape development, as parts of the region exhibited high changes in LCC after 1780.

At least since approximately 1750, the agricultural economy of east Elbian Prussia was based on market-oriented manors/communes (Melton 2000). The rural area was characterised by family-run farms, self-sufficient producers and, except in bad years, selling surplus agricultural products on nearby markets (Hagen 2002). This system worked because as noblemen were responsible for their commune but depended on the unpaid manual work and unpaid ox and cart work provided by farmers living in the commune. Farmers had to provide these services for several days per week and then had to cultivate their own piece of land (for those who did have their own land).

At this time, the level of self-sufficiency was very high in rural areas; thus, each commune depended on different resources for daily life during the early period (Kluge 2005). Therefore, in addition to arable fields, forest areas were of equal importance because they served as areas for (1) wood production for building houses and firewood and for creating various types of traditional tools (e.g., wooden rakes and pegs); (2) litter gathering for use in stables; (3) fruit and mushroom collection; (4) honey (nests of honey bees in old trees) collection; and (5) pasture for livestock, as grassland areas were small in the region (Enders 1986; Bayerl 2006). There was little progress in livestock farming at this time, except that indoor breeding became more common (Klemm et al. 1998).

Because self-sufficiency required high workloads and much time, people needed to settle near the places where they worked, resulting in a clear association between the cover of arable fields and settlements and roads (Hawbaker et al. 2004; Kluge 2005; Seidl 2006). In contrast, forest areas were located far from settlements on sites less suitable for agriculture, e.g., on sites with steep slopes or less productive sandy soils. This is expected and is well described for other European regions as well, e.g., in Poland (Zachwatowicz and Giętkowski 2010) and in Hungary (Biró et al. 2013).

With the increasing prices for cereals between 1796 and 1805, the situation changed as farmers wanted to generate profit from this development (Hagen 2002). However, their manpower and time were limited because they had to work on their own farm and for the noblemen. Thus, reforms were necessary, particularly as an increasing number of farmers began to fight for their rights. The reforms were realised over a period of approximately 50 years by giving farmers the chance to pay money and become free from unpaid manual work and unpaid ox and cart work. In approximately 1850, 80 % of all famers had realised the transfer fee to the noblemen (Melton 2000).

The level of self-sufficiency remained relatively high until the end of the nineteenth century, and even pasturing livestock in forest areas was common. At the same time, the application of fertilisers and the use of machines (plough and seed drill) led to yield increases. Nevertheless, because many farmers had to pay a portion of their profits to noblemen and machines were relatively expensive, only approximately 50 % of the farmers owned machines at the end of the nineteenth century (Kölling 1999; Seidl 2006). However, growing economic progress in Prussia was reflected in a comprehensive road construction programme, during which roads increased from ca. 3200 km in 1816–11,850 km in 1848. The expansion was crowned by the railway line between Berlin and Potsdam in the year 1838 and an increase in railway lines in all of Prussia from ca. 860 km in 1844 to ca. 2360 km in 1848 (Koch 1978). All these measures facilitated the transport of goods, and farmers were stimulated to increase yields and thus their incomes.

With an enormous decrease in prices for cereals due to imports from overseas, beginning in approximately 1890, livestock farming was supported in Germany (Klemm et al. 1998). As the study area was relatively poor in sites suitable to be cultivated as grassland, this national support had no clear effect. In addition, potato cultivation had increased beginning in ca. 1780, and potatoes were established as a common arable crop in the Uckermark region. Potatoes were used as fodder for animals, allowing a further increase in indoor breeding and reducing the need for permanent grassland. Potatoes were also used as food for humans, and with a much higher content of calories compared to cereals, they became very important during famines in 1816/1817, the 1840s and after 1873 (Enders 1986).

Our results showed that the cover of arable fields and forests was lowest in ca. 1780, indicating that the use of areas with productive soils as arable fields and leaving forest on less productive soils was not fully realised at that time. However, the explained variance for both LC types increased clearly for 1890 and 2000, indicating an increasing congruence of site conditions and LC types realised by human activities (Bayerl 2006). We assumed the dissolution of the interdependence between noblemen and farmers in the eighteenth century as a major driver of LCC. For landscape window 1, it is likely that the increase in arable fields to the costs of forest areas between 1780 and 1890 was possible due to renewed owner structures (Hagen 2002). In this window, productive soils prevailed, and it is not clear why they were not used as arable fields before the eighteenth century, as occurred in many other parts of the region with similar site conditions. Additionally, it is likely that the increased constructed roads and railway lines made the transport of wood easier and cheaper than previously, and the income helped farmers buy agricultural machines and fertilisers.

However, most farmers required many years to achieve full payment, by instalments, to the noblemen (Melton 2000; Hagen 2002). Thus, several farmers were not able to buy agricultural machines and fertilisers for the first time when they became free farmers. This situation hindered them from increasing the yield on arable fields and to convert less productive soils into forest areas. We assumed this was the main reason for the relatively late afforestation of less productive arable fields between 1890 and 2000 in landscape window 3.

As for landscape windows 2 and 4, which exhibited generally stable LC compositions, it is likely that the balance between site conditions and the most profitable LC/LU had already been established before 1780, as in the broad scale of the entire Uckermark region.

Conclusions

Because the main LC types were determined by different natural and anthropogenic factors during the three periods, our first hypothesis can be accepted and is in accordance with the results of Aspinall (2004). Although the Uckermark region has been impacted by humans for thousands of years, natural factors have been a major determinant of LC composition during the past 220 years. Nonetheless, anthropogenic factors have become more important in more recent times (Bayerl 2006).

Our second hypothesis can be partly accepted because anthropogenic and natural factors show opposite effects on LC types in highly dynamic and more stable windows, except in three cases (one-third of the nine total windows). We see changes as the potential for LCC per se, which only become visible when looking back on long-term LCC and are valuable for understanding landscape development against the background of interdependence between noblemen and farmers and its step-by-step dissolution from the late eighteenth to the mid-nineteenth century (Melton 2000; Hagen 2002).

The increasing number of independent farmers in parallel with an improved infrastructure has led to an optimal LC composition in parts of the Uckermark region where it was not realised previously. In fact, in one landscape window, forests have been cleared to make productive soil usable for agriculture, which was not possible before 1780, whereas in another window, less productive soils are no longer needed for the production of food and were afforested in the twentieth century.

Landscapes with a very long tradition in a given LU, as is typical for Europe, are affected by multiple factors, and it is likely that not all can be adequately simulated in models. We found that changing property situations in the rural landscape from the eighteenth to the nineteenth century is an important explanatory variable to understand LCC in sub-regions. Without looking back for more than 200 years, it would remain unclear why there had been extensive LCC after 1780 and even after 1890 in a region with a stable LC composition on a broad scale.