Global Change Biology Global Change Biology (2015) 21, 1197–1212, doi: 10.1111/gcb.12776

From forest to farmland: pollen-inferred land cover change across Europe using the pseudobiomization approach R A L P H M . F Y F E , J E S S I E W O O D B R I D G E and N E I L R O B E R T S School of Geography, Earth and Environmental Sciences, Plymouth University, Plymouth PL4 8AA, UK

Abstract Maps of continental-scale land cover are utilized by a range of diverse users but whilst a range of products exist that describe present and recent land cover in Europe, there are currently no datasets that describe past variations over long time-scales. User groups with an interest in past land cover include the climate modelling community, socioecological historians and earth system scientists. Europe is one of the continents with the longest histories of land conversion from forest to farmland, thus understanding land cover change in this area is globally significant. This study applies the pseudobiomization method (PBM) to 982 pollen records from across Europe, taken from the European Pollen Database (EPD) to produce a first synthesis of pan-European land cover change for the period 9000 BP to present, in contiguous 200 year time intervals. The PBM transforms pollen proportions from each site to one of eight land cover classes (LCCs) that are directly comparable to the CORINE land cover classification. The proportion of LCCs represented in each time window provides a spatially aggregated record of land cover change for temperate and northern Europe, and for a series of case study regions (western France, the western Alps, and the Czech Republic and Slovakia). At the European scale, the impact of Neolithic food producing economies appear to be detectable from 6000 BP through reduction in broad-leaf forests resulting from human land use activities such as forest clearance. Total forest cover at a pan-European scale moved outside the range of previous background variability from 4000 BP onwards. From 2200 BP land cover change intensified, and the broad pattern of land cover for preindustrial Europe was established by 1000 BP. Recognizing the timing of anthropogenic land cover change in Europe will further the understanding of land cover-climate interactions, and the origins of the modern cultural landscape. Keywords: Europe, holocene, human impacts, land cover, pollen, pseudobiomization, vegetation Received 1 July 2014; revised version received 10 October 2014 and accepted 16 October 2014

Introduction Mapping land cover at a continental and global scale has been driven by the needs of diverse user-groups within the scientific and practitioner communities. Mapping modern land cover is not straightforward, but methods drawing on remote sensing of the environment (e.g., CORINE land cover maps, Bossard et al., 2000) have resulted in global and regional-scale products that are now widely used (e.g., in environmental protection and management, mapping species population densities, and crop yield monitoring and forecasting). The understanding of past variations in land cover over long time periods is desirable by an equally wide group of researchers. They include those with an interest in land cover – climate feedback mechanisms (e.g., Gaillard et al., 2010; Goldewijk et al., 2011; Strandberg et al., 2014), groups seeking to understand long-term socio-ecological histories (e.g., Pongratz et al., 2008; Correspondence: Ralph M. Fyfe, tel. +44 1752 585929, fax +44 1752 585565, e-mail: [email protected]

© 2014 John Wiley & Sons Ltd

Ellis et al., 2013) and those analysing the role of land cover change within earth system science, such as carbon storage and release (e.g., Olofsson & Hickler, 2008; Kaplan et al., 2011; Williams et al., 2011; Ruddiman, 2014), and erosion histories and long-term processes of change in geomorphic systems (e.g., Foulds & Macklin, 2006; Macklin & Lewin, 2008; Van De Wiel et al., 2011). The production of detailed land cover maps for the period before the late 20th century must move beyond remote sensing methods. Cadastral maps and land use surveys may extend mapping back to the 18th/19th centuries (e.g., The UK National Archives), but these are patchy in spatial coverage and do not facilitate greater time-depth to land use and land cover change. The challenges in producing land cover maps that cover Holocene time scales (the last ~12 000 years) can be addressed through the use of proxies, and attempts at mapping vegetation patterns draw primarily on compilations of dated pollen records from multiple sedimentary records (e.g., Huntley, 1990). While individual plant taxa and biomes have been mapped at the continental scale from pollen data (e.g., Huntley & 1197

1198 R . M . F Y F E et al. Birks, 1983; Prentice et al., 1996; Williams et al., 2011), these efforts have not, to date, resulted in broad-scale reconstructions of land cover. Broad-scale land cover mapping through quantification of pollen datasets has been the focus of an international working group (LANDCLIM: Gaillard et al., 2010). The first results of the LANDCLIM project are now published for selected time periods during the Holocene (Trondman et al., in press). These results will make it possible to evaluate the timing and scale of long-term human modification of natural vegetation beyond the local or regional scale, for the first time. Natural vegetation is considered the cover that existed before people became a significant ecological factor (‘original-naturalness’ sensu Peterken, 1996). The transformation of Europe’s landscapes has been long and complex, linked to the conversion of ‘wildwood’ into a mosaic of pasture, farmland, forest and heath. While this process has involved multiple agencies, the major long-term trend has been from a naturedominated to a more human-dominated state. This was associated with growth of human populations relying on agricultural economies, and their need for arable and grazing land, along with timber and other forest products. Along with east Asia, Europe is the continent with the longest history of land conversion from forest

to farmland, and as such, is globally significant (Ruddiman, 2014). In this article, we present a first synthesis of pan-European land cover change from pollen data for the period 9000 BP [years Before Present (AD 1950)] to the present day. To generate natural and anthropogenic land cover classes (LCCs) we have employed the pseudobiomization method (PBM; Fyfe et al., 2010). Woodbridge et al. (2014b) tested the method by applying it to 2471 modern pollen samples and comparing the results to remote sensed data taken from the CORINE land cover map of Europe. The PBM produced results for modern land cover that are comparable to those from remotely sensed maps at a regional and European scale, and a full discussion of the results of this comparison is included within Woodbridge et al. (2014b). As a method of intermediate complexity, the PBM has the important advantage of being applicable to the large number of published pollen records distributed across the European continent (Fig. 1).

Mapping regional and continental scale vegetation from pollen data Attempting to synthesize pollen data to assess vegetation patterning at the continental scale is not new, with a notable first attempt for Europe by Huntley & Birks

Fig. 1 Map showing European Pollen Database (EPD) sites. Dark grey circles represent sites included in the spatially aggregated analysis for temperate and northern Europe. Symbols represent case study areas (dark grey circles/light grey pentagons/triangles/squares: temperate Europe, pentagons: western France, triangles: western Alps, squares: Czech/Slovakia, and light grey circles: sites not included in regional case studies. Spatial interpolation is based on all sites, but area presented clipped to hashed rectangle. © 2014 John Wiley & Sons Ltd, Global Change Biology, 21, 1197–1212

E U R O P E A N L A N D C O V E R C H A N G E 1199 (1983). Their work presented major patterns of spatial variation via isopoll mapping of numerous individual pollen types, and drew on ‘digitized’ data from published pollen diagrams. Huntley & Birks (1983) mapped at both the taxon level and groups of taxa to produce semiquantitative palaeo-vegetation maps of Europe; this was subsequently developed through identification of ‘communities’ within the pollen data (Huntley, 1990). More recent studies have focussed on specific plant taxa at the European scale (e.g., Giesecke & Bennett, 2004; Giesecke et al., 2007). Pollen data have also been used to establish the distribution of the biomes of Europe in the past, although this was initially undertaken only for key time intervals: notably at 18 000 BP, 6000 BP and present (Prentice et al., 1996, 1998; Peyron et al., 1998). More recently, Davis et al. (in press) have converted European pollen data into plant functional type (PFT) assemblages and used a square chord distance (SCD) measure of ‘difference from modern’ on biome scores for 500-year time windows during the Holocene. During the early 2000s a renewed interest in quantification of vegetation cover from pollen data resulted in the Landscape Reconstruction Algorithm (LRA; Sugita, 2007a,b), which is a generalized version of the r-value model of Davis (1963). The LRA is a two-part modelling process that first quantifies regional vegetation via the REVEALS (Regional Estimates of VEgetation Abundance from Large Sites) model, which can then be used to produce LOcal VEgetation estimates (the LOVE model). This quantification of regional vegetation has resulted in various national and regional syntheses of Holocene land cover for particular time intervals (e.g., Gaillard et al., 2010; Nielsen et al., 2012; Trondman et al., in press) and through the Holocene in 500-year time windows (e.g., Fyfe et al., 2013; Marquer et al., 2014). One advantage of the REVEALS approach over biomization methods is that it is possible to determine the relative composition of taxa within biomes (Trondman et al., in press). Considerable compositional changes can occur within a biome, which are not reflected at the coarse ‘biome’ level. Whilst there is good confidence in the results of the REVEALS approach, it is a complex method and relies on the availability and reliability of a key input parameter, the region-specific pollen productivity estimate (PPE) of the various taxa under study (Brostr€ om et al., 2008). The application of the REVEALS approach to the whole of Europe represents a research goal that is still in its early stages. This article draws upon an alternative approach to classification of land cover from pollen data, one developed in Fyfe et al. (2010) and Woodbridge et al. (2014b). It seeks to classify pollen samples using a pseudobiomization approach, rather than by quantifying taxon © 2014 John Wiley & Sons Ltd, Global Change Biology, 21, 1197–1212

abundance. The PBM incorporates the indicator species approach (Behre, 1986) employing biomization techniques (Prentice et al., 1996; Peyron et al., 1998; Prentice & Webb, 1998) and has previously been successfully applied to Holocene pollen datasets from Britain (Fyfe et al., 2010; Woodbridge et al., 2014a) and south Germany (Lechterbeck et al., 2014).

Materials and methods

Selection and formatting of pollen data The data used to generate Holocene land cover have been taken from the European Pollen Database (EPD: Fyfe et al., 2009; Leydet, 2007-2013). The EPD is an open data repository for original (rather than digitised) pollen counts on sites from Europe, constructed and maintained by the pollen community for the use in wider scientific endeavour. The version of the database used (version released in September 2013) contains 1363 records, although not all of these are dated by radiocarbon or similar techniques. 982 records from 805 sites have been selected for this analysis based on their chronologies, time period covered and location. Earlier work has sought to ensure that errors within the EPD are managed, particularly with older data (Fyfe et al., 2009) and considerable effort has gone in to developing reliable chronologies for the dated records. Data were taken only from sites with robust chronologies selected using Giesecke et al.’s (2014) uncertainty classification; all pollen samples with a ‘0’ star rating (greater than 2000 years from the nearest 14C date) were removed from the dataset. Chronologies were based on the calibrated age-depth models prepared by Giesecke et al. (2014). For the small number of pollen count data added to the EPD more recently than Giesecke et al. (2014), the contributing author’s calibrated chronology was used. Some synthetic studies have sought to augment publicly available datasets (e.g. EPD) with pollen data from additional sites for improved spatial coverage (e.g. Collins et al., 2012). This was not undertaken in our analysis to ensure that other studies could replicate our findings, or draw comparisons based on different methodological approaches applied to the same data. Previous work has demonstrated that, following the end of the last glacial period, Holocene vegetation cover in most regions of Europe was established by ~8000 BP (Marquer et al., 2014); consequently, for the purpose of this study, the EPD was filtered to only include samples dating between 9000 BP and present. The PBM as employed here has been optimised to reconstruct anthropogenic as well as natural LCCs, making it unsuitable for study of Late-Glacial and early Holocene vegetation change (Fyfe et al., 2010). All pollen data underwent taxonomic harmonization using established European Pollen Database (EPD) protocols for nomenclature standardization (Leydet, 2007-2013). For the PBM, taxa were grouped according to the taxonomic level required for assigning to different LCCs (Table 1 and Data S1 and S2). To facilitate comparison between sequences, pollen count data were summed into contiguous 200-year time intervals. Previous studies have shown that 200-year time windows are sufficient to identify

1200 R . M . F Y F E et al. Table 1 Land cover classes (LCCs) defined for the pseudobiomization (PBM) method Land Cover Class Closed LCCs LCC1 Needle-leaf forest LCC2 Broad-leaf forest LCC3

Mixed forest

Semiopen LCC LCC4 Semiopen vegetation Open LCCs LCC5 LCC6

Heath/scrubland Pasture/natural grassland

LCC7

Arable/disturbed land

LCC8

Mixed open vegetation

Description

Defined by a small number of needle-leaf trees (typically high pollen producers) Defined by a large number of broad-leaf trees (e.g. deciduous Quercus), including a small number of epiphytes (e.g. Hedera), and some fruit trees (e.g. Olea) Dominated by a mix of needle-leaf and broad-leaf forest Defined by thresholds between different forest types Mixed land cover including both forest and open vegetation Defined by thresholds between other classes Defined by a mixture of heath/scrub type taxa, including evergreen Quercus. Predominantly defined by a large mixture of grassland herbs including a number of taxa associated with pasture Defined by a large number of herbaceous taxa indicative of arable and disturbed land (excluding fruit trees) Dominated by a combination of heath/scrubland, pastures/natural grassland and arable land Defined by thresholds between open classes

human-induced land use changes (Woodbridge et al., 2014a; Lechterbeck et al., 2014), and avoid any potential temporal smoothing that might result when using longer time windows.

Transformation of pollen data to landcover classes (LCCs) The transformation from pollen proportions to land cover classes (LCCs) for each pollen sample followed the approach outlined in Fyfe et al. (2010), using the refinements applied in Woodbridge et al. (2014b) (Data S1). Eight LCCs were established to reflect those defined by the CORINE land cover classification (Bossard et al., 2000) (Table 1). This includes two ‘mixed’ vegetation LCCs, one semiopen LCC and five ‘pure’ classes (e.g., broad-leaf forest). ‘Pure’ does not imply naturalness, but is a technical term reflecting the nature of LCCs which are defined as dominated by one land cover type and not a combination of more than one class. Applying the PBM to the EPD data involved assigning taxa to LCCs (Data S2), which was initially informed by taxa grouping according to Principle Components Analysis (PCA) (see Woodbridge et al., 2014b), square root transforming pollen percentage data, grouping, summing and normalizing the values for different LCCs, calculating an affinity score based on the modified pollen sum and assigning the ‘winning’ LCC (see Woodbridge et al., 2014b). The LCC with the highest score (sum of the square root transformed percentages for all species attributed to the class) is assigned to the sample. The square root transformation allows high pollen producing taxa, such as Pinus to be down-weighed and low pollen producing herbaceous taxa to be up-weighed in the dataset. Additionally, broad-leaf forest was down-weighted (9 0.6) and arable/disturbed land was up-weighted (9 1.3) in the dataset. These values were determined through a process of optimisation, based on comparison of modern pollen assemblages with remote-sensed land cover data (Woodbridge et al., 2014b). Semiopen and mixed LCCs are defined using a threshold value between the

sum of closed and open land cover (see Fyfe et al., 2010; Woodbridge et al., 2014b).

Spatial aggregation of pollen records The proportion of each LCC represented in each time window was calculated to obtain a spatially aggregated record of landcover change in Europe. Parts of Europe where testing of the method using modern pollen and remotely sensed land cover data indicated a poorer match (Woodbridge et al., 2014b) have been excluded from the synthesis of the pollen dataset shown in Fig. 1. This included most of the Mediterranean where differentiating natural LCCs from anthropogenic classes is likely to require a land cover classification somewhat different to that used for temperate and northern Europe. This filtering exercise nonetheless included 813 suitable records covering the rest of Europe (Fig. 1), with 300 – 500 sites included for each 200-year time window. We also use a series of case-study regions to explore spatial and temporal patterns of land cover change in regions of temperate Europe, representative of the diverse landscapes of this part of the continent. The case study areas used are western France, the western (i.e., Swiss and French) Alps, and the Czech Republic and Slovakia (Fig. 1). These represent a gradient from more oceanic to more continental climatic settings, have sufficient density of available pollen data for generalization of past regional land use.

Spatial interpolation of LCC scores to generate landcover maps In addition to spatially aggregated land cover reconstructions for Europe and case-study areas, spatially interpolated maps have been constructed to produce continuous representations of land cover across Europe (including the Mediterranean) for all 45 time intervals. This was undertaken using a thin plate © 2014 John Wiley & Sons Ltd, Global Change Biology, 21, 1197–1212

E U R O P E A N L A N D C O V E R C H A N G E 1201 spline (TPS) method executed in R (R Core Team, 2013) using the package Fields (Furrer et al., 2013). Interpolation was based on the score for each LCC, the penultimate stage of the PBM, at a spatial resolution of 20 km. Each interpolated surface thus represents an estimate of the proportion of each LCC within each grid cell. It is not possible, using this method, to represent semiopen or mixed LCCs, as these do not have a numerical ‘score’ that can be interpolated, but are assigned using thresholds between the different ‘pure’ LCCs. Elevation was used as a covariate, and extracted from the GTOPO30 global digital elevation dataset (USGS, 1996). The TPS interpolation was based on all sites (982 records), but clipped to the area of Europe with best spatial coverage of sites (886 records) where the PBM approach appears to perform most efficiently.

Results

Spatially aggregated land cover change The percentage of pollen samples assigned to each LCC for the last nine millennia is shown in Fig. 2.

The period 8600–6000 BP marks a relatively stable maximum in forest cover, after which the forest sum begins a very gradual decline. Broad-leaf forest gradually increases to a maximum at 6000 BP, subsequently showing an almost linear decrease to very low levels at the present-day. Broad-leaf trees remain an important component of Europe’s vegetation today, but in the classification used here they form part of either mixed forests or semiopen vegetation, the latter typically representing a mosaic of deciduous woodland alongside more open land cover types. Pure homogeneous broad-leafed forest, which was Europe’s most common LCC 6000 years ago, is now greatly reduced in extent. By contrast, needle-leaf forest has shown only minor changes in extent over the last 9000 years. The reduction in total forest cover since the mid-Holocene has been mirrored by a corresponding increase in the proportion of sites classified as open LCCs, with a significant acceleration after 1400 BP. The most important of these open LCCs are pasture/grassland,

Fig. 2 Spatially aggregated pseudobiomization (PBM) results for temperate and northern Europe (813 sites): percentage of sites assigned to each LCC per 200 year time interval (9000–0 BP). Grey box shows forest maximum. © 2014 John Wiley & Sons Ltd, Global Change Biology, 21, 1197–1212

1202 R . M . F Y F E et al. arable/disturbed land or a combination of the two. Within the broad trend towards more open landscapes from 6000 BP, there are also periods where the forest sum remained stable across several time windows, notably between 3000–2400 BP and 2000–1400 BP. In the earlier period needle-leaf forest expanded slightly whilst broad-leaf forest declined; in the latter both forest types remained stable.

Variability in land cover change across Europe Spatial variability in land cover change can be explored through examination of LCC curves for the three case study regions (Figs. 3 and 4). All time intervals contain at least 15 sites per region to ensure that they are representative of the overall proportions of regional land cover. Even so, changes in the proportions of sites in

Fig. 3 Spatially aggregated pseudobiomization (PBM) results for case study areas (western France, western Alps, and Czech/Slovakia): percentage of samples assigned to each LCC per 200 year time interval (9000–0 BP). © 2014 John Wiley & Sons Ltd, Global Change Biology, 21, 1197–1212

E U R O P E A N L A N D C O V E R C H A N G E 1203 each LCC between adjacent time windows are more abrupt than for the whole of temperate Europe, this being a reflection of having fewer sites contributing to the records, and analysis at a finer spatial scale. There are clear differences in land cover change between the three case study areas and in the duration of forest dominance. Western France has significantly more sites classified as ‘semiopen’ and open, not only in recent millennia but throughout the Holocene. In contrast with many regions in central and eastern Europe, the sum of forest LCCs never exceeded 60%, indicating that Atlantic seaboard landscapes were always less densely wooded. Forests in western France were dominated by broad-leaf trees and their long-term decline began at 4500 BP, around the same time as they did in Britain (Woodbridge et al., 2014a). The forest sum had reached stable low levels by 2200 BP, by which time western France would have been a predominantly open landscape. Part-wooded semiopen vegetation continued to decline until the most recent time interval, as open LCCs increased in importance. The latter comprised a mix of heathland, which reached a maximum 2000–600 BP, pasture/grassland, and arable/disturbed land, which reached a maximum in early modern times and has declined in the last four centuries. The western Alps region stayed densely forested until ~4000 BP, although the proportion of sites classified as needle-leaf forest steadily increased during the mid-Holocene at the expense of broad-leaf trees. Forest cover started to decline during Bronze Age times, but the main fall in forest LCC in this region occurred rather abruptly ~2200 BP, at the end of the Iron Age. The proportion of sites classed as forest declined thereafter, reaching a minimum only in the most recent time interval. Open LCCs in the Alps were dominated by grassland, which had been rare in mid-Holocene times, while arable/disturbed land peaked in Medieval-early modern times, as in western France. In the Czech/Slovakia region the forest sum reached values even higher than in the Alps, but here its composition has been largely mixed needle-leaf/broad-leaf forest. This has therefore always been a densely forested region, and the forest maximum persisted until 1200 BP, before rapidly declining to lower levels around 600 BP, that is, during Medieval times. The proportion of sites classified as open vegetation has decreased in the two most recent time intervals (400–0 BP). The Czech/Slovakia region remains the study area with the largest proportion of sites still classified as forest, 13% in the current (0–200 BP) time interval. Heath/scrubland is not a dominant LCC in any of the records for the last 9000 years. The only case study region where this LCC is significant is in western © 2014 John Wiley & Sons Ltd, Global Change Biology, 21, 1197–1212

France, whereas in Czech/Slovakia heathland is never recorded as the dominant LCC.

Spatially interpolated land cover change Land cover classes scores can be described as the penultimate stage of the PBM method (Fyfe et al., 2010). Whilst they are not a direct measure of the proportions of each LCC within the pollen source area of a site, they provide a useable value for interpolation between sites and hence for mapping past LCCs. Whilst the EPD dataset is not sufficiently dense to be able to capture the full pattern of spatial variability in European land cover, it does facilitate a first-order approximation of the general patterns of land cover at the continental scale. Interpolated maps for each of the main ‘pure’ LCCs have been produced for every 200-year time interval, and selected time windows, every 1000 years, are shown in Figs 5–8 to illustrate spatial and temporal patterns of land cover change. When interpreting changes in the interpolated land cover maps, patterns are most appropriately interpreted in terms of relative changes between different time intervals, as interpolated values do not provide fully quantified land cover extent, but rather indicate where and when changes in land use took place. The earliest time intervals show clear south-north and elevation gradients in the broad-leaf and needle-leaf forest classes, and these gradients persist through all time intervals (Figs 5 and 6). The Alps and other mountainous regions are clearly visible, having high proportions of conifer woodland, while the Mediterranean and parts of eastern Europe have rather low broad-leaf forest scores. Declines in interpolated broad-leaf forest scores are most evident from 4000 BP onwards, notably in northwest Europe, and these forests appear to have survived longest in the North European plain (e.g., northern Germany, southern Scandinavia), only being cleared in Medieval times. The highest levels of the pasture/natural grassland LCC occurred in the south of Europe during mid-Holocene times, while the lowest levels are found around the Baltic, particularly around the Gulf of Bothnia (Fig. 7). From 4000 BP increases in pasture/grassland scores are clearly evident on the Atlantic coast, particularly in Britain and Ireland, and there is a clear west-east trend, with scores declining towards eastern Europe. By modern times, grassland as a land cover type expanded its spatial coverage across all of Europe except in the far northeast. The highest scores for arable/disturbed land were initially found in the Mediterranean and Black Sea regions, with the lowest scores being found at higher latitudes, notably in northern Britain and Scandinavia (Fig. 8). After 3000 BP scores for this LCC began to increase in western and

1204 R . M . F Y F E et al.

Fig. 4 Summed percentage of samples assigned to forest, semiopen and open vegetation LCCs in each 200 year time interval (9000-0 BP) for case study regions (western France, western Alps, and Czech/Slovakia) with total number of sites in time interval.

central Europe, and across parts of northern Europe after 1000 BP. The expansion of arable/disturbed land into higher latitudes (northern Britain and most of Scandinavia) has remained limited, even today. Overall these maps correspond well to Europe’s modern and historical plant biogeography and land use. The pollen-based PBM land cover reconstructions correctly place needle-leaf forests as dominant in mountain regions, such as the Alps, and in the Boreal forest zone of northern and northeast Europe. The PBM reflects the maximum extent of pasture land in northwest Europe in recent millennia, notably in Ireland’s ‘emerald isle’, while it shows arable land as occupying a wide swathe of temperate Europe that once was occupied by mainly deciduous forests. Today this is one of Europe’s most important areas of cereal and other crop cultivation.

Discussion

When did people start to change Europe’s land cover? It is widely recognized that the vegetation and land cover of Europe have been substantially modified through human management and intervention. Land

cover changes resulting from the spread of agriculture are discernible in Europe from the mid-Holocene both in individual pollen records and from excavated Neolithic sites (Whitehouse et al., 2014). However, disentangling human-induced land cover change from climate-driven vegetation forcing is not straightforward. The results of the transformation of pollen data presented here using the PBM method across temperate Europe have allowed temporal and spatial patterns of land cover change to be identified. This also provides a spatially explicit, data-led approach that can be compared with computer model simulations of anthropogenic land cover change (e.g., Kaplan et al., 2011). In previous regional-scale studies using the PBM we have been able to identify temporal correlations between the emergence of agriculture during the Neolithic period and land cover change associated with forest conversion to cropland and pasture (Woodbridge et al., 2014a; Lechterbeck et al., 2014). At the European scale we can similarly identify mid-Holocene land cover changes that may have been related to early farming activity. For example, there is a reduction in total number of pollen records classified as broad-leaf forest from 6000 BP onwards, coinciding with a major rise in human populations inferred from summed 14C date probability © 2014 John Wiley & Sons Ltd, Global Change Biology, 21, 1197–1212

E U R O P E A N L A N D C O V E R C H A N G E 1205

Fig. 5 Spatially interpolated pseudobiomization (PBM) results for needle-leaf forest. Selected 200 year time intervals presented. Black circles represent pollen site locations.

distributions from central and western European archaeological sites (Shennan et al., 2013). Neolithic food producing economies involved clearance of Europe’s broad-leaf forests, and even if this clearance was initially limited in scope, its impact appears to be detectable at a continental scale when pollen records are aggregated and converted into past land cover. The timing and pattern of prehistoric land cover change is therefore consistent with the hypothesis that © 2014 John Wiley & Sons Ltd, Global Change Biology, 21, 1197–1212

the onset of forest decline in temperate Europe was a direct result of human modification of land cover, relating to the adoption and spread of agriculture. This was a time-transgressive process, which started around 10000 BP in Southwest Asia (Simmons, 2007) and reached its geographic extent in northern Europe ~6000 BP (Whitehouse et al., 2014). Thus, the changes seen by aggregating data for the whole of Europe obfuscate patterning that exists at finer spatial scales, as

1206 R . M . F Y F E et al.

Fig. 6 Spatially interpolated pseudobiomization (PBM) results for broad-leaf forest. Selected 200 year time intervals presented. Black circles represent pollen site locations.

demonstrated in the regional patterns for western France, the western Alps and Czech/Slovakia. More detailed local-regional analysis can be accomplished using the REVEALS-based approach on single, regionally representative sites, such as by Marquer et al. (2014), which has offered insights into the development of the cultural landscape of northern Germany and Poland, Scandinavia and Britain and Ireland. They

recognize the first Neolithic impacts for northern Germany and Poland around 6700 BP, marked by the emergence of arable indicators, while Nielsen et al. (2012) recognize similar early indicators of arable cultivation in their synthesis for Germany and Denmark. Support for this can be found in Lechterbeck et al. (2014) which, using the PBM approach, has recognized regional changes in woodland composition in southern © 2014 John Wiley & Sons Ltd, Global Change Biology, 21, 1197–1212

E U R O P E A N L A N D C O V E R C H A N G E 1207

Fig. 7 Spatially interpolated pseudobiomization (PBM) results for pasture/natural grassland. Selected 200 year time intervals presented. Black circles represent pollen site locations.

Germany, attributed to growth of secondary woodland following clearance during an early wave of Neolithic agriculture. The time at which large-scale human disturbance starts to become recognizable is an important consideration for understanding nature/culture relationships. There is some variation between the few studies that have attempted to address this question for Europe. Marquer et al. (2014) argue that while agriculture can © 2014 John Wiley & Sons Ltd, Global Change Biology, 21, 1197–1212

be recognized in pollen records before 5200 BP, its impact on landscape openness was minimal, and that deforestation became significant only after this time. Davis et al. (in press) identify forest decline at 4000 BP in their square chord distance analysis, and at 5000 BP in the falling proportion of %APFT (Arboreal Plant Functional Types). The pan-European synthesis of temperate and northern Europe by Trondman et al. (in press) demonstrates a significant increase in open land

1208 R . M . F Y F E et al. and grassland between 6000 and 3000 BP, as a consequence of increases in grazing land and the need to provide winter fodder for animals (particularly hay meadows). The precise time at which the first deforestation occurs cannot, however, be established from their study as they do not assess any time intervals between 6000 and 3000 BP. Our analysis shows a progressive decline in broad-leaf forest from 6000 BP, but this is partly compensated by an expansion of mixed forest, and total forest cover at a pan-European scale only falls outside the range of previous background variability from 4000 BP onwards. Thus, while the impact of Neolithic agriculture is certainly detectable, it was during the Bronze Age that the first large-scale deforestation of Europe took place. These variations in the inferred timing for the onset of forest decline in temperate Europe are a consequence of several key factors. Firstly, the time-averaging approach to pollen data means that studies using longer time windows (e.g., Marquer et al., 2014; Davis et al., in press) are more likely to ‘smooth’ patterns that are evident at shorter time intervals, a temporal example of the scale effect in the modifiable areal unit problem (Openshaw, 1983). Secondly, the spatial extent of analysis varies between studies. Finally, there are methodological differences in approach to transformation of the data. In our current study using the PBM, for example, we have produced LCCs which include mixed and semiopen classes, while Davis et al. (in press) produced climate-induced vegetation biomes (plant functional types), although a stable biome reconstruction does not preclude significant changes in the composition of vegetation within that biome. Including mixed and semiopen classes overcomes this limitation to an extent. Trondman et al. (in press) describe a further limitation, in that the biomization process does not, per se, represent anthropogenic open land cover. In consequence, much of northwest Europe (e.g., Britain and France) in the 0–500 BP time interval remain classified as temperate deciduous forest in Davis et al. (in press: Fig. 6), even though historical sources show clearly that most of these forests had been cleared by this time (Williams, 2003). The PBM attempts to resolve this by explicitly considering anthropogenically modified LCCs. It is not possible palynologically to distinguish easily between natural grassland and managed pasture-land, or between arable and other forms of disturbed open land cover. However, because the contribution of LCCs 6–8 to overall land cover has increased markedly over time (from

From forest to farmland: pollen-inferred land cover change across Europe using the pseudobiomization approach.

Maps of continental-scale land cover are utilized by a range of diverse users but whilst a range of products exist that describe present and recent la...
1MB Sizes 0 Downloads 8 Views