Introduction

In landscape archaeology, geophysical methods are increasingly being applied to conduct large area surveys1. Although these techniques allow obtaining high-resolution archaeological information at a landscape-scale2,3,4, they often neglect natural landscape variations. To fully understand the driving mechanisms behind human land-use, the integration of pedological and geomorphological information in these prospection stages is crucial.

In Europe, one of the most characteristic examples of past human-landscape interaction, is the reclamation of wetlands and forest that followed the urbanisation of the historical County of Flanders (Fig. 1b). During Medieval periods, these lowlands experienced a considerable population growth, making the County one of the most densely populated areas in Europe between the 11th and the 15th centuries5. To meet the demands of emerging cities, such as Ghent, Bruges and Ypres6, entire natural landscapes were reclaimed, transforming these into a landscape of dynamic exploitation7. The ruling Counts of Flanders spearheaded this evolution by endowing feudal lords with lands, while abbeys were deliberately installed on marginal land all over Flanders.

Figure 1
figure 1

Location and historical context of the site.

(a), The location and elevation32 of the study area with indication of the excavated Abbey of Boudelo (1). The dashed line represents the boundary between the palaeolake (to its south) and the coversand ridge (to its north). (b), Location of the studied area (red dot) within the late medieval County of Flanders (orange boundary) and the current state boundary of Belgium. Maps composed in ESRI ArcMAP 9.3.1.

Full size image

More recent, natural processes (e.g. flooding) and modernisation have altered the cultural landscape of medieval Flanders, leaving the nature and exact layout of these designed landscapes largely unknown8. Although social formation and economic exploitation of reclaimed landscapes during the High Middle Ages is, to some extent, documented in historical and cartographical sources, archaeological evidence remains scarce. One of the main reasons for this scarcity is that traditional investigations have focused on recovering architectural remains rather than situating the structures within their broader environmental contexts. By focusing on elite residences and monastic buildings, archaeological investigations have contributed to the biased image of medieval settlement landscapes that still prevails.

The situation was no different for the Cistercian abbey of Boudelo located in the north of the County of Flanders (Fig. 1b). In 1197, a small religious community settled in this area9, which at that time was an outback of the County, dominated by marshes and wetlands. Early historical accounts, such as those by monks from the Abbey of Clairvaux in Bourgogne (France), give testament to the harsh environmental conditions. They describe life at Boudelo as pauper and misserimus: poor and full of misery10. However, the community improved its property by cultivating the surrounding land. Between the 13th and 16th centuries, the monastic estate expanded to over 1000 ha, making the community one of the leading cultivators in the County. After religious as well as military struggles and successive floods, the monks were forced to abandon their grounds in Boudelo in 1578. They found refuge in Ghent and the monastic buildings were sold and dismantled.

Although the extensive reclamation strategy of the abbey is attested in historical records mentioning embankment, drainage, stockbreeding, extraction of peat and clay and the production of building ceramics, it remains hidden in the archaeological record. Excavations in the 1970s revealed the remnants of the cloister range and the abbey church, leading archaeologists to interpret the abbey grounds as limited to the coversand ridge11 (Fig. 1a). The border of the monastic precinct was believed to coincide with the edge of a Late Glacial palaeolake, which was and remains, a waterlogged environment dominated by peat and lime-rich lacustrine deposits (Fig. 1, Supplementary Fig. 1–2). However, in 2011 an electromagnetic induction (EMI) survey of this wetland area drastically altered this interpretation, as it unveiled traces of the abbey's outer court that once was part of the monastic precinct; a previously unknown designed landscape from which the monks directed their cultivation of the surrounding area.

The multi-receiver EMI sensor we used, allows simultaneously recording the apparent electrical conductivity (σa) and apparent magnetic susceptibility (κa) of four different soil volumes12. Whereas in non-saline environments σa can be directly related to soil texture (clay, silt, sand) and is influenced by soil organic matter and water content13,14, κa anomalies are often the result from ferrimagnetic soil enrichment15,16, the anthropogenic disturbance17 of top soils (e.g. ditches), or the heating of soil18 (e.g. hearths, bricks). This contrasts to other geophysical survey techniques, such as magnetometry19 and ground penetrating radar20, that each target only one specific variable and are inadequate to appreciate the full range of soil textural variability of surveyed sites. However, integrating electrical and magnetic soil properties in a single EMI survey allows targeting the pedological, geomorphological and archaeological variations jointly. By simultaneously mapping multiple soil volumes with a multi-receiver instrument, vertical discrimination potential is added to the electrical and magnetic data layers.

Results

The σa data revealed a complex environment with alternations of highly conductive lake deposits and peat accumulations (Fig. 2a). Apart from scarce sandy outcrops the measurements indicate a buried marshland influenced by long lasting, saturated conditions. Several ditches were revealed as highly conductive anomalies surrounding the entire survey area and enclosing two smaller zones. The high conductivity of the ditches is caused by high concentrations of fine-grained sediments (clay) and peat that contrast with the more resistive underlying sand. The extent and layout of the ditches and their persisting in the present-day drainage system indicates considerable medieval investment in reclaiming this wetland. The two smaller enclosed compounds appear to centre around sandy areas within the larger enclosure (S1 and S2 in Fig. 2). On the central platform within each compound, the κa measurements showed highly magnetic structures (Fig. 2b and d). At the easternmost site (S1), at least 18 features with a minimum diameter of 1 m, suggest the layout of a rectangular structure some 20 m by 30 m square (Fig. 2d). A central structure can be observed at the western site (S2) (Fig. 2b). Here, the elevated κa of the ditches indicates their infilling with brick and other magnetic debris, whereas a trail of high κa values leading north suggests displacing rubble towards the sand ridge.

Figure 2
figure 2

Electromagnetic induction data and excavated structures.

(a), 1 m HCP σa data of the study area, plotted on an orthophotograph of the study area33, with indication of a man-made ditch (1), a sandy outcrop (2) and an area with accumulated peat and lacustrine deposits (3). The dashed lines indicate both moated sites S1 and S2. (b), 2 m HCP κa data of the platform of S2, showing the central building and the magnetic disturbances caused by rubble in the ditches and leading towards the north (Supplementary Fig. 2). (c, e), Orthophotos of the brick foundations as revealed by the excavations at S2 (c) and S1 (e). (d), 2 m HCP κa data showing traces of 18 brick block foundations on the platform in S1. Maps composed in ESRI ArcMAP 9.3.1.

Full size image

To interpret and complement the geophysical data excavation trenches were laid out across the ditches and the central structure of each site (Supplementary Fig. 1–2). At both sites brick structures were found, as indicated by the κa measurements (Fig. 2b–e). These were identified as block foundations used to support a larger, most likely wooden, structure and to keep it away from the saturated soil. While a functional interpretation of S2 has been hampered by the poor conservation of the excavated building remains, the central structure at S1 has been interpreted as a large monastic barn. The morphology of the sandy outcrop, combined with elevated ground water levels caused instability of the foundations leading to the collapse of the building. The excavated material culture and construction materials indicate that both sites were occupied between the 13th and the early 14th centuries. This links their abandonment to a period of documented increased flooding in the region21.

Based on the σa and excavation data, we modelled the relative medieval topography of the site22,23. Although the heterogeneity of the ditch infillings, the peat layers and the targeted sandy layers limited the precision of this modelling procedure, an accurate model of the medieval surface was obtained. At 84 validation locations in the excavation trench at S2 a RMSEE of 0.37 cm and a Spearman correlation coefficient of 0.80 was obtained between the observed (Fig. 3a) and modelled (Fig. 3b) depths to the archaeological layers. Although the selection of the validation observations was not statistically random, the results confirmed the relative model correspondence to the medieval topography. A comparison of the modelled data to the observed surface along the validation transect (Fig. 3), shows the smoothing effect on the modelled data, but confirms the accurate representation of the targeted archaeological layers.

Figure 3
figure 3

Medieval terrain model evaluation.

(a), Geometrically correct visualisation of the western profile of the northern moat ditch in S2 (Supplementary Fig. 2b). 84 calibration points, 0.2 m apart, along the targeted archaeological surface (orange line), were used to validate the surface model. (b), modelled medieval surface . Note the difference in y-scale between (a) and (b), which indicates the smaller standard deviation of the modelled depths (standard deviation = 0.10 m) compared to the observed surface levels (standard deviation = 0.43 m).

Full size image

By integrating the κa data into the model, a comprehensive reconstruction of the reclaimed landscape was obtained (Fig. 4). Both S1 and S2 are located on prominently higher and sandier locations, which are connected to the bordering sand ridge in the north. However, while the barn (at S1) was built on an existing sandy outcrop, excavations showed that S2 was artificially raised (Supplementary Figure 2d) by removing sand from the ditches and the adjacent sand ridge. The design of both sites confirms a link to the nearby abbey buildings as their topographic position was chosen or altered in order to allow easy access to the sand ridge. The ditch enclosing the area was largest east of S1 with a width of up to 10 m and a depth of 1.5 m below the surface (Supplementary Figure 1d). Here, a larger drainage capacity was needed, as the ditch was part of the complex-wide drainage system. This large ditch also expressed a separation between the inside and outside worlds; a symbolic reminder of the separation between religious and secular life24. The ditches circling S2 and the remainder of S1 rarely surpassed a depth of 1 m and had a width ranging from 2 m to 8 m (Supplementary Fig. 1d and 2e). These wide ditches did not only supply additional drainage for the enclosed zones, but also formed a physical boundary that embodied aspects of identity and status25. Individual compounds, such as S1 and S2, with an (artificially) raised platform and enclosed by ditches, can be defined as late-medieval moated sites25. In addition, the surface model shows dykes neighbouring the ditches that add to the embankment and visibility of both moated sites. Remnants of these earthworks were attested in the excavation trenches, either still in situ or thrown into the moat ditches (Supplementary Figure 2e).

Figure 4
figure 4

Model of the medieval landscape.

Three-dimensional model of the depth to the medieval surface and schematic representation of the barn and central building remains at S1 and S2, derived from the κa data (Fig. 2). The topographical model shows the sandy outcrop bearing structure S1 (1) and the artificially raised terrain forming the platform at S2 (2). At both sites, a pathway to the coversand ridge can be seen (3). In the north of the site the edge of the coversand ridge itself is visible (4). Moat embankments are present around both S1 and S2 (5) and a large enclosure embankment (6) surrounds the entire area. Present-day parcel ditches separate the different agricultural fields in the area (7).

Full size image

Discussion

Supported only by very limited invasive research, this reconstruction gives a comprehensive and unique insight into a designed medieval environment and shows, despite its transient nature, the amount of effort that was put into it. Our results also underline the discrepancies that can exist between historical information and uncovered archaeological realities that can contribute to understanding the response of past societies to social and environmental changes. The 3-D mapping of multiple soil properties, combined with limited and directed invasive research, provides a broad foundation for further geoarchaeological research. In the future, this methodology could be optimized by integrating additional physical soil properties, such as dielectric permittivity, with the σa and κa data. With the incorporation of the presented approach into different studies and environments, our knowledge of past human-landscape interactions can be significantly improved.

Methods

EMI survey

We used a motorized setup of a Dualem-21S EMI sensor12 to survey the study area (Fig. 1). This EMI sensor simultaneously measures σa and κa with four coil configurations, i.e. with receiver coils at 1 m, 1.1 m, 2 and 2.1 m from the transmitting coil, in two orientations, i.e. horizontal coplanar (HCP) and perpendicular (PRP) orientation, resulting in 1 m HCP, 2 m HCP, 1.1 m PRP and 2.1 m PRP. The survey was conducted by driving across the study area along parallel lines, 0.75 m apart whereby measurements were taken every 0.25 m. During surveying, soil temperature was recorded at a depth of 0.3 m to correct the σa data for temperature variations26 between survey days. The resulting measurements were drift corrected12. These data were then interpolated to a 0.1 m by 0.1 m grid using ordinary kriging27 with the software Surfer v. 10 (Golden Software). All σa datasets were used but only the κa measurements from the HCP configurations were considered, as the κa data of the PRP configurations had a poor signal to noise ratio.

Excavations

In August 2011 and August 2012, two excavation trenches were laid out across the most apparent σa and κa anomalies (Supplementary Figures 1,2). All profiles and unearthed brick structures were recorded using computer vision-based 3-D registration28.

Depth modelling

The four σa measurements, representative for soil volumes down to 0.5 m, 1.0 m, 1.5 m and 3.2 m below the surface, were combined to model the depth to the soil layers below the more conductive plough layer and the peaty to clayey infilling of the ditches (Fig. 3). Based on the excavation data, the morphology of the targeted soil layers was considered representative for the medieval topography. We followed the procedure from Saey et al.22 and De Smedt et al.29, based on the depth response functions of the Dualem-21S coil configurations as described by Wait30 and McNeill31. Model calibration23 was performed, using the observed depth to these layers obtained at 34 locations in the excavation trench at S1.