J Archaeol Method Theory DOI 10.1007/s10816-017-9338-y

Integration of Complementary Archaeological Prospection Data from a Late Iron Age Settlement at Vesterager—Denmark Roland Filzwieser 1 & Lis Helles Olesen 2 & Geert Verhoeven 1 & Esben Schlosser Mauritsen 3 & Wolfgang Neubauer 1,4 & Immo Trinks 1 & Milena Nowak 1 & Rebecca Nowak 1 & Petra Schneidhofer 1 & Erich Nau 1 & Manuel Gabler 1

# Springer Science+Business Media New York 2017

Abstract The complementary use of various archaeological prospection data sets offers a series of new possibilities for the investigation of prehistoric settlements. In addition to the separate interpretations of the single methods, the implementation of image fusion provides an additional tool to obtain an even higher degree of data integration during the interpretation process. To investigate some possibilities and risks of image fusion, a procedure frequently used in the medical field but rarely applied in archaeology, various algorithms inside a dedicated MATLAB toolbox TAIFU (Toolbox for Archaeological Image FUsion) were tested on the geophysical prospection data from an Iron Age settlement near Vesterager in West Jutland, Denmark. The Ludwig Boltzmann Institute for Archaeological Prospection and Virtual Archaeology had conducted large-scale, high-resolution ground-penetrating radar (GPR) and magnetometry surveys at the site in 2014, based on its discovery by the Ringkøbing Museum through aerial photos and the results of a follow-up excavation in 2009. The aim was to determine if, and to what extent, geophysical prospection together with a novel integrative interpretational approach was able to add more detailed information to an

* Roland Filzwieser [email protected]

1

Ludwig Boltzmann Institute for Archaeological Prospection and Virtual Archaeology, Hohe Warte 38, 1190 Vienna, Austria

2

Holstebro Museum, Museumsvej 2B, 7500 Holstebro, Denmark

3

Ringkøbing-Skjern Museum, Bundsbækvej 25, 6900 Skjern, Denmark

4

Vienna Institute for Archaeological Science, University of Vienna, Franz-Klein-Gasse 1/III, 1190 Vienna, Austria

Filzwieser et al.

already known prehistoric settlement. Results yielded a variety of deeper insights into the separate farms (dated to around AD 400), including the discovery of several new structures and more information about the construction of the longhouses, as well as a first suggestion on how to implement image fusion into the process of analysis and archaeological interpretation of geophysical data sets. Keywords Image fusion . GPR . Magnetometry . Archaeological interpretation . Iron Age settlement . Denmark

Introduction In 2014, the Ludwig Boltzmann Institute for Archaeological Prospection and Virtual Archaeology (LBI ArchPro), together with the Holstebro Museum, conducted a geophysical survey to document an Iron Age settlement near Vesterager in West Jutland, Denmark (Fig. 1). The site had previously been investigated by the Holstebro Museum and Ringkøbing-Skjern Museum using aerial archaeology and excavation. The aim now, however, was an assessment of the possibilities derived from the integration of complementary archaeological survey methods as well as image fusion and data visualisation programs. It is expected that the hereby-proposed integrative approach will not only lead to a considerable increase of detailed insight into the settlement but also introduce a more efficient method of detecting archaeological structures in largescale prospection data of entire prehistoric landscapes. This paper will present the research history and describe the individual Vesterager data sets as well as their integrated archaeological interpretation. Additionally, the interpretation will be complemented by imagery that results from fusing various geophysical data sets, performed in order to investigate the possible advantages of implementing novel post-processual steps into a standardised interpretation process.

Background Since 2010, the LBI ArchPro has gathered considerable experience with the integrated interpretation of large-scale high-resolution prospection data (Trinks et al. 2012; Trinks et al. 2015). In the course of this work, many problems could be identified and solved through continuous development of filtering methods (Hinterleitner et al. 2013) and automated data classification (Pregesbauer et al. 2013; Sevara et al. 2016), improvement of the data acquisition software (Sandici et al. 2013) as well as the data processing software (ApMag, ApRadar) and through implementing the GIS extension ArchaeoAnalyst for improved analysis and interpretation of the processed images within a Geographical Information System (GIS). Yet, a truly integrated interpretation of different prospection data layers of the same site always remains to some extent restricted, especially if time is an issue and the interpretational work is shared amongst a team. In order to offer a possible solution for this data integration issue, TAIFU (the Toolbox for Archaeological Image FUsion—Verhoeven et al. 2016) was developed. Another LBI ArchPro development of the past years, which contributed significantly to the emergence of this article, was the ever-growing involvement in the non-invasive

Integration of Prospection Data from Denmark

Fig. 1 Map of the Holstebro region in West Jutland. The site is located some 15 km north of Ringkøbing

documentation and investigation of entire Iron Age and Viking Age landscapes in Scandinavia. Amongst these are the Norwegian sites of Borre (Draganits et al. 2015), Kaupang, Gokstad (Schneidhofer et al. 2016), and Oseberg, as well as the UNESCO World Heritage Site Birka-Hovgården (Trinks et al. 2010; Trinks et al. 2013; Trinks et al. 2014) and the prehistoric settlement of Uppåkra (Larsson et al. 2015) in Sweden. A collaboration with experts from the Holstebro Museum was established in order to widen the scope of this focus and to investigate the potential of an integrative archaeological prospection approach with near-surface geophysical prospection methods at corresponding sites in Denmark. The here-presented prospection measurements were carried out in the course of the ongoing aerial archaeological research project An aerial view of the past—Aerial archaeology in Denmark 1 (Olesen 2011). The Iron Age farmstead at Vesterager in West Jutland was amongst five sites of archaeological interest that were selected (Nau et al. 2015). Vesterager has already been investigated by aerial archaeological photography for many years, resulting in 1

http://www.fortidensetfrahimlen.dk

Filzwieser et al.

several very revealing aerial images (Eriksen and Olesen 2002). Following an excavation2 carried out in March 2009 (Olesen and Mauritsen 2015), this site was deemed to be ideal for this integrative investigation approach. The choice for the 0.8-ha large field to be surveyed was not only based on the density of well-observable archaeological structures in the aerial footage but was also motivated by the suitability of the present soil for geophysical prospection in the area. The expected high contrasts between anthropogenic structures and a rather homogenous soil matrix should also translate to a good visibility of relatively high reflective or absorbing deposits in the groundpenetrating radar (GPR) data (Conyers 2013) as well as different magnetic properties of the buried archaeology in regard to the magnetometry (Neubauer 2001). The aim of the project was to overcome some limitations of the aerial photography by providing new complementary data sets for the site and thus to derive an even more meaningful and integrated interpretation of the prehistoric farmsteads. In order to integrate the obtained geophysical data sets for enhanced analysis, an image fusion post-processing step was implemented using TAIFU.

Site Description Aerial Archaeology In 1992, a settlement dating from around AD 400 was discovered from the air in the flat landscape at Vesterager, only 4 km from the North Sea coast (Fig. 2) (Eriksen and Olesen 2002, 57–59). Since then, more of the site has been photographed from the air. The picture emerged of an elongated, east-west-oriented settlement area positioned on an extended elevation that rises only 5 m above sea level, and that is delimited to the north and to the south by lower-lying areas and fjords. When people decided to settle here in the Iron Age, it could very well have been because of the existence of extensive shore meadows at that time, favourable for cattle herds and the production of hay. The aerial photograph depicting the settlement best is shown in Fig. 2. Immediately to the right of the small modern buildings on the left side of the picture, the clear remains of a 29-m longhouse are visible, together with those of a smaller building and wells or pit houses on a rectangular farmyard enclosed by a post-built fence. However, the latter can only be seen to the north—to the right of the picture—and to the west. The byre end of the longhouse (to the east) clearly differs from the dwelling end (to the west) by having twice as many postholes due to the stall partitions. To the north of the farmyard, another possible longhouse as well as several ditches and further pits are visible. A second farmyard can be assumed immediately to the east. The strong, dark-green traces of trenches were interpreted as boundary ditches and the right-angled bends, or corners, support this assumption. In virtually treeless areas, such as along the fjords of West Jutland, ditches and dikes were often used around farms in the Viking Age and the Middle Ages, rather than wooden fences. For example, this situation was observed 10 km further north in 1994, during Ringkøbing Museum’s investigations of the Viking 2

Ringkøbing-Skjern Museum. RSM 10.033. Excavation by Palle Eriksen, Lis Helles Olesen, and Esben Schlosser Mauritsen.

Integration of Prospection Data from Denmark

Fig. 2 Oblique aerial image of Vesterager. The settlement was discovered and photographed for the first time on June 30th 1992. The area of the 2009 excavation is marked with a white polygon, measuring 38 m from north to south in its greatest extent. Photo: Palle Eriksen

Age settlement of Skadborg, near the Staby church, and incidentally also a settlement that was discovered from the air in 1992 (Eriksen and Henningsen 1995; Eriksen and Olesen 2000, 155–160; Eriksen and Olesen 2002, 101–102). Therefore, it was presumed that the boundary ditches at Vesterager were coeval with those at Staby, even though no corresponding Viking Age house remains could be spotted at Vesterager to match those at Staby. The general arrangement of Vesterager’s conspicuous boundary ditches suggests the presence of a road running north–south, which to the south (in the lowermost part of Fig. 2) runs alongside the second prehistoric farmyard enclosed by ditches. Inside the latter are two ring ditches, the largest of which with a diameter of 7.5 m. Even though this picture appears simple at first sight, there are nevertheless several places where house remains and fences cross, giving the impression of several settlement phases. Vesterager has been photographed from the air on numerous occasions and new traces continue to turn up and supplement the existing picture of the settlement. Excavation In connection with Holstebro Museum’s project An aerial view of the past—Aerial archaeology in Denmark, it was decided to carry out a small targeted excavation of the Vesterager settlement—first and foremost in order to determine the age of the boundary ditches and to elucidate the function of the ring ditches. The investigation also had to

Filzwieser et al.

respect this remarkable locality by disturbing as little of it as possible. The excavation took place in March 2009 as a joint undertaking between Holstebro Museum’s project and Ringkøbing-Skjern Museum (Fig. 3).3 An area of 500 m2 was uncovered, which included the boundary ditches and one of the ring ditches (Fig. 3). The 40–50-cm-thick layer of overlying soil was carefully removed using a mechanical excavator, but there was a 1-m-wide baulk over the chosen ring ditch left so that the central area of the structure could be excavated by hand. The number of features encountered within the excavation trench was not overwhelming and can be categorised as follows: two large boundary ditches and several smaller ditches, a ring ditch, a house, and two to three post-built fences, three possible hearths, three cremation graves or pits containing a little burnt bone, a well, field furrows from recent times, and modern disturbances (Fig. 4). The two large ditches actually consisted of at least two to three small superimposed ditches. Each of these smaller ditches was about 80 cm wide and 30–50 cm deep. The ring ditch was not completely circular as its diameter varied from 4.5–5 m. The ditch itself was up to 35 cm wide and in section, it could be seen that it had a depth of around 30 cm, including the layer of Iron Age soil through which it had been cut. There were no traces whatsoever of features in the centre of the ring ditch. In the northern part of the area, parts of a house site were uncovered—the southwestern corner of which was connected to a post-built fence that continued southwards, passing just to the west of the ring ditch. The house site, which was oriented in east-west direction, constituted the northern side of a fenced farmyard. The three other sides were bounded by conspicuous ditches, which framed the two ring ditches in the middle of the farmyard, and a well to the south. On the aerial photo (Fig. 2), traces of several postholes for the roof-bearing posts and the ditch belonging to the southern wall of the house can be seen. The two farmyards—one to the west with its house site and traces of postbuilt fences and one to the east with boundary ditches and the two ring-shaped structures—could very well have been in use simultaneously. However, it is also clear that there were several settlement phases, as a boundary ditch cuts across the western farmstead, and to the north of this farm, there are three to four boundary ditches that could not have been functional at the same time. Moreover, there are traces of further longhouses that have not yet been unambiguously matched with their respective farmyards (Fig. 9). One surprising result of the excavation was that all finds and all the features, with exception of the field furrows and the modern disturbances, are from the same period, i.e. the decades around AD 400. This means that the introduction of ditches instead of wooden fences in the region of West Jutland has to be pushed back 400 years in time, from the Viking Age to the Late Iron Age. No conclusive interpretation of the ring ditch could be made. But as there have been two ring ditches visible in the aerial photograph that lie together in the centre of the farmyard, this observation reinforces suspicions of a more practical 3

Ringkøbing-Skjern Museum. RSM 10.033. Excavation by Palle Eriksen, Lis Helles Olesen, and Esben Schlosser Mauritsen.

Integration of Prospection Data from Denmark

Fig. 3 Oblique aerial image of Vesterager. The exposed property boundaries (a) are clearly evident in the excavation trench, while the ring ditch (b) is fainter. Photo 216, 11.3.2009, Lis Helles Olesen

function, for example associated with a storage place for hay (i.e. possibly a drainage ditch around a hay rick), which in North German Iron Age settlements is referred to as Diemen (Zimmermann 1991). Similar ring-shaped structures, although smaller than those encountered at Vesterager, were found by Gudmund Hatt in 1946 at Røjklit, only 5 km to the south of Vesterager, when he investigated a settlement from the Early Roman Iron Age. Hatt suggested at the time that these ditches were traces of a cultic fertility-related construction associated with the harvesting of the last sheaf (Hatt 1953, 18–25).

Fig. 4 The most important structures are highlighted on this excavation plan from 2009. House site and fence (a), ditches (b), ring ditch (c), and well (d). Drawing: Palle Eriksen and Esben Schlosser Mauritsen

Filzwieser et al.

Methodology Geophysical Prospection The geophysical surveys at Vesterager were conducted by a team of the LBI ArchPro on the 1st and 2nd of September 2014. Within these 2 days of fieldwork, 5438 m2 had been covered with GPR measurements (Fig. 5) and 5775 m2 with magnetics (Fig. 6). The survey conditions were good and the field had been mowed just before the fieldwork took place. A first processing of both data sets was conducted immediately on site for quality control. On September 1st, a magnetometer system consisting of eight Förster FEREX CON650 probes mounted with 25 cm spacing on a custom-build non-magnetic cart towed by a quad bike was used; this system comprised a ten-channel Eastern Atlas analogue-to-digital converter and a ruggedised portable computer. LoggerVis, a software developed by the LBI ArchPro for magnetic data acquisition, permitted efficient visual guidance and navigation and close-to real-time data visualisation for quality control. This magnetometer system provides an average inline sample spacing of 10–15 cm at driving speeds of approximately 20 km/h. The positioning with centimetre accuracy was derived from a Real Time Kinematic Global Navigation Satellite System (RTK-GNSS) consisting of two Javad Triumph-1 receivers in base and rover configuration. On September 2nd, a 16-channel MALÅ Imaging Radar Array (MIRA) was used for the GPR survey; the MIRA system was housed in a plastic box mounted on the front hydraulics of a small tractor. Its nine-transmitter and eight-receiver antennae with a centre frequency of 400 MHz (Trinks et al. 2010) are arranged in two rows that are shifted by an offset of half an antenna width against each other. This antenna arrangement allows for a crossline channel spacing of 8 cm, while the average inline GPR trace spacing is 2 to 4 cm. The 16-channel array is therefore able to cover a 128-cm-wide swath with each driven line. On a ruggedised industrial computer installed inside the driver cabin, the software MIRAsoft was used to record the GPR data while LoggerVis

Fig. 5 GPR depth slice (0.5–0.7 m) of Vesterager superimposed on an aerial photograph, with the excavation trench from 2009 well visible as reflective (dark) deposit in the data

Integration of Prospection Data from Denmark

Fig. 6 How challenging the interpretation of large-scale GPR data can be becomes evident when comparing a magnetic data image (a) from Vesterager (−3 to +6 nT) with two respective GPR depth slices, 0.8–1.0 m (b), and 1.4–1.6 m (c)

was used for guidance and navigation. Positioning was achieved with a Leica 1200 GNSS, again in base and rover configuration. The final processing and visualisation steps for both the magnetometry and the GPR data sets were carried out using an in-house developed software package (ApMag and ApRadar, respectively) for large-scale high-resolution prospection data. For the magnetic data, several processing algorithms were applied (de-spiking, removal of the directional influence, GPS time stamp synchronisation, data interpolation, and GPS outlier removal) and optimised greyscale images were produced, depicting different magnetic data value ranges. A white-black clip-off range between −3 and +6 nT was found to be especially suitable for data interpretation. The

Filzwieser et al.

magnetic data from Vesterager shows a magnetically calm, homogeneous background. No naturally caused magnetic anomalies seem to be present. Furthermore, the homogenous background provides a good contrast in order to observe and interpret archaeological features—especially many larger pits, which can be observed better in the magnetic images than in any other data set. Several magnetic dipole anomalies, caused by buried iron objects, can be seen throughout the survey area. These, however, do not necessarily display relevant archaeological remains and usually originate from more recent manure containing household rubbish (Schofield 1989). For the GPR data, a 3D data block was created and, after time-to-depth conversion, cut into horizontal GPR depth slices of different thickness (5, 10, 20, 30, 40, and 50 cm). Prior to the generation of the 3D GPR data block, various standard processing steps were applied, such as constant time shift removal, frequency bandpass filtering, average-trace removal, 2D Kirchhoff migration, and time zero adjustment. For the gridding of the data, nearest position grid mapping was used and Hilbert transformed amplitudes computed. For the editing of the single depth-slice images, median cost distance removal, stripe removal, and interpolation of missing data were applied. Due to the effect that physical properties, such as soil moisture and clay content, have on the velocity of the GPR pulse, and in order to derive realistic average GPR velocity values for each survey site, it is crucial to analyse the actual pulse velocity. In the case of Vesterager, several reflection hyperbolae were used to this end, resulting in GPR pulse velocities between 9 and 16 cm/ns. The rather high velocities are caused by the well-drained glaciofluvial sands and grus that the Vesterager site is located in (GEUS 1989), which generally present favourable conditions for the propagation of electromagnetic waves. For the processing and time-to-depth conversion, an average velocity between 10 and 12 cm/ns was used. Therefore, it should be noted that the depth displayed in the resulting GPR images is only relatively correct, and deviations from the actual depth are possible throughout the entire data set. In general, the GPR data reveals much more detailed information about the archaeological structures in Vesterager than the magnetic data, which is, of course, mainly due to the much higher spatial measurement resolution of the GPR system used. All resulting images were automatically georeferenced in the coordinate reference system ETRS89/UTM zone 32N (EPSG 25832) using the acquired RTK-GNSS positions and subsequently imported into a geodatabase within a GIS project (Neubauer 2004). Further data analysis, archaeological interpretation and map creation were carried out within the framework of the GIS using the in-house developed ArcGIS extension ArchaeoAnalyst implemented in ArcMap 10.2. Image Fusion The continuously increasing number of various imaging techniques within archaeology, and especially the field of archaeological prospection, is also leading to an evergrowing demand for new techniques to analyse this multitude of different data. In the process of image fusion, two or more images are merged in a specific way, thereby generating a single composite image. This newly generated image then contains specific, respectively substantial information from the input images. Ideally, the output image should be more comprehensive and thus better suited for interpretation and further processing, while the fusion algorithm should be efficient and reliable. Image

Integration of Prospection Data from Denmark

fusion is frequently used in the medical field to improve image quality while simultaneously decreasing the amount of data and solving information redundancy. As such, it leads to a simplification of assessing medical problems (James and Dasarathy 2014). Image fusion can be applied at different levels, which can be categorised as signal, pixel, feature, and symbolic level. In this article, however, only pixel-based methods have been considered since image pixels are the final output of the geophysical data acquisition and processing chain. As image fusion has so far rarely been employed in archaeological research, a dedicated MATLAB toolbox TAIFU (Toolbox for Archaeological Image FUsion) was developed. The program functions as a platform for the examination of various well-established, state-of-the-art image fusion approaches (such as pansharpening or the plethora of well-known blending methods) as well as totally new algorithms that have been developed specifically for this task (e.g. distribution fitting). With this program, it should ideally be possible to fuse various geophysical data visualisations to aid the interpreter in locating and correlating features for a more reliable and targeted examination of hidden archaeological landscapes. In its current version 0.3, TAIFU allows the user to load two images that can then be edited separately (Figs. 7 and 8). Metadata of the loaded images, such as IPTC and Exif tags as well as georeferencing information, are verified and stored upon import. Additionally, there is a broad variety of pre-processing steps at the user’s disposal, i.e. visualisation and extraction of individual image channels and their histograms, colour-coding of single band imagery through various perception-based colour maps, and better readability of certain image features by different contrast enhancement algorithms. Furthermore, to fuse images for which the fusion methods expect an (un)equal number of bands, colour-to-greyscale conversion and vice versa was implemented. In a next step, one can choose from a large variety of fusion methods, of which many have additional options to fuse the loaded images into a single-output image. Once a useful result is achieved, the composite image with all the embedded metadata can be

Fig. 7 Screenshot of TAIFU, fusion (right) of two GPR depth slices, 0.8–1 m inverted (top left), and 1.4– 1.6 m (bottom left) of building H2 using the overlay blending mode

Filzwieser et al.

Fig. 8 Screenshot of TAIFU, fusion (right) of two fused GPR depth slices (top left) with the magnetics (bottom left) of building H2 using the grain merge blending mode

saved. These metadata, which can also be stored in a separate ASCII-file, originate from both input images but also contain any information about contrast enhancement, colour maps, channel conversion and fusion algorithms that have been applied to obtain the final fusion. To inform the user about ongoing and possible processes, all buttons feature tooltips while all fusion methods also come with explanations and separate notification windows. To investigate some of the possibilities, pitfalls, and limits of geophysical image fusion, a structure from the Vesterager data sets (house H2, see Fig. 9) was further processed in TAIFU. One problem commonly encountered during the interpretation of archaeological GPR data is the detection and understanding of small features, such as

Fig. 9 Interpretation map, combining the analysis of aerial, magnetic, GPR, and excavation data from Vesterager

Integration of Prospection Data from Denmark

postholes, throughout all depth slices. In the case of structure H2, these postholes can, as expected, be observed in the upper layers as absorbing deposits (depicting the less compressed absorbing backfill), while the packing stones or compressed soil on the bottom create strong reflective anomalies. However, these features can vary substantially in both intensity and coverage, and due to missing topographic correction, they can be distributed over several different depth slices, making the interpretation more challenging. Therefore, two different depth slices were loaded into TAIFU, each one depicting the absorbing respective to the reflective features of the building’s postholes best (i.e. depth slices 0.8–1 m and 1.4–1.6 m). In a first step, the upper depth slice, showing the postholes as absorbing (white) anomalies, was inverted, allowing for an image fusion of the now consistently black features using overlay. Overlay darkens the dark areas (by multiplying them) and lightens light areas (using the screen blending method). The latter first inverts the pixel values, multiplies them, and inverts them again. This yields the opposite effect to multiply. Moreover, a black layer does not affect the other layer, whereas screening with white produces white. The name comes from the fact that the resulting effect is similar to projecting multiple photographic slides on top of each other. The result is an image with enhanced contrast, while the initial shadows and the highlights of the base image are kept. As a side effect, certain parts of the resulting fused image might be washed out. In the case of the structure H2, the fused result displayed an increased visibility of the roof-bearing postholes (Fig. 7). The outer rows of postholes, however, show still much more distinctly in the magnetic data (Fig. 8). Therefore, in a second step, the overlay image created in step one was fused with the magnetics (visualised with a clip-off range between −3 nT for white and +6 nT for black) using the grain merge blending approach. The latter fusion method adds the pixel values of the upper and lower layers and subtracts 128 (0.5) to meancentre the data. This result is afterwards normalised to omit values outside the [0 1] range. This fusion yielded a composite image (Fig. 8), which superiorly renders the feature details of H2 over the entire extent of this structure. After presenting the overall archaeological interpretation of this case study in the next section, the benefits as well as the risks of such image fusion approaches will be discussed. Fusion between the aerial photographs and geophysical data images was not attempted in this paper, since an exhaustive treatise of merging such completely different image modalities would go beyond the scope of this paper.

Results and Discussion The geophysical data from Vesterager corroborate the impression of three neighbouring farmyards (Olesen and Mauritsen 2015, 118–121) from the Migration Period (Fig. 9). In the magnetic data, the remains of two longhouses (H1 and H2, Fig. 9), already known from the aerials, can be observed quite well. They are oriented in northwestsoutheast direction. Two rows of postholes and parts of the ditches alongside the walls build up the house structures. In both instances, the houses had a dwelling half with four pairs of larger postholes in the west, while to the east, the byre of one longhouse was comprised of eight pairs of smaller postholes, and the other longhouse had a byre half of five pairs. This division is quite common for longhouses of that period in western Denmark (Webley 2008, 51). The larger open space between the second and

Filzwieser et al.

third pair of postholes in the western half of both buildings can be interpreted as a central part of the dwelling areas with a possible fireplace; this occurrence is frequently observed in Iron Age longhouses (Eriksen and Olesen 2002, 46; Eriksen et al. 2009, 48–52). In the GPR data, the anomalies belonging to house H1 can be observed in its full extent. The building is about 30 m long and 6 m wide. The two inner rows of postholes are not as unambiguously defined as in the magnetic data. Yet, what appears in the magnetics in the cases of H1, as well as H2, as enclosing ditches, in the GPR data turns out to be rows of very densely placed individual postholes marking the outer walls (Fig. 10). Similar observations were made at the north-western corner of building H5. These discoveries correspond to the excavation results of house H4. There are, however, only a few individual postholes of the latter building to be found in the geophysical data. H2 is again best visible in the GPR data and can be observed quite well in a depth range of 0.8–1.6 m. The impression of H3, located about 8 m to the northeast of H1 and initially discovered in the aerial images, could also be expanded by the results of the geophysical survey. Here, the magnetic data provide only weak positive anomalies, interpreted as ditches surrounding the house, on its western, northern, and southern sides. This building shows roughly the same orientation as H1. In the eastern half of H3, no clear features of the building can be observed in the magnetics. The GPR data confirms its width of about 5 m and suggests that the house was up to 12 m long. Furthermore, the post-built fence structures to the western and northern end of the farmyard, attributed to H1 and H3, can be traced quite well on the radar images and might have been roofed, as known from other Iron Age sites in Denmark (Ethelberg et al. 2003, 232). In the magnetics, many of the elongated ditches, some already confirmed by the excavation and visible as positive magnetic anomalies, can be observed in the survey area. They are mainly oriented perpendicular to the longhouses. While some of them can be seen as contemporary boundary ditches surrounding the potential farmyards, others are clearly cutting the house structures (Figs. 6 and 10), suggesting multiple phases of the settlement. The ring ditch R1 can undoubtedly be observed on the magnetics, while ring ditch R2 cannot be seen at all. This may be due to the excavation in 2009, after which this structure was not visible for several years on aerial images either. On the GPR data, a third ring ditch (R3) with a diameter of about 4.7 m was discovered to the southeast of R1, slightly overlapping it. The ditches in the centre of the surveyed area match perfectly with the excavation results, as does the excavated ring ditch R2 (Fig. 4). Southwest of H2, a fourth (R4) of these structures, possibly the remains of so-called Diemen (Ethelberg et al. 2003, 152) that had been used for the stacking of hay can be assumed and probably two more (R5 and R6) to the very east of the investigated field (Fig. 9). Throughout the entire survey area, large circular pits can be observed. They vary in size between 1 and 4 m. The smaller ones could derive from individual postholes, whereas some of the larger ones could be interpreted as further wells or pit houses. Although these pits are present in the entire area, two clusters with larger numbers of pits are obvious. The first cluster is located to the northeast of H1, where six large pits, with diameters varying between 1.2 and 3.8 m, cause strong magnetic anomalies. Towards the eastern and southern side of H2, a similar concentration of eight larger and several smaller pits with varying magnetic properties is present. In many cases, the

Integration of Prospection Data from Denmark

Fig. 10 Comparison of aerial (a), magnetic (b), and GPR (c) data interpretation

function of these pits remains unclear. One pit is situated at the eastern end of H2. It creates a strong positive magnetic anomaly, and in a depth of about 1–1.3 m, the GPR still shows the pit in roughly the same dimensions with an absorbing filling. In a depth of around 1.4–1.7 m, the backfill is strongly reflective and the anomaly vanishes after that. This could lead to an interpretation of the structure as a waste pit. When comparing the larger pits to the excavated well (W) from 2009, several agreements in size, shape, and positioning can be found—however, none of the above-discussed pits can unambiguously be interpreted as additional wells.

Filzwieser et al.

The overall benefits of a complementary approach in archaeological prospection are evident and have frequently been demonstrated (Scollar et al. 1990; Neubauer and Eder-Hinterleitner 1997; Powlesland et al. 1997; Doneus and Neubauer 1998; Gaffney and Gater 2003; Trinks et al. 2014). The application of various prospection methods has been adopted by a majority of archaeologists and their use is often rather a matter of funding than of methodological consideration. Also in this Vesterager case study, it was clear that none of the applied prospection methods were able to reveal the full range of archaeological structures present (Fig. 10). Although the final interpretation map draws from a combination of different prospection data, one can still not speak of a truly integrated approach to archaeological mapping as long as only individual data layers are interpreted separately. Simply changing the opacity/transparency of stacked data layers, a procedure that is commonly applied in GIS packages to approximately view two data sources at the same time, cannot be considered true image integration, since it lacks any form of intelligent data visualisation and combination. Since an integrative approach to interpreting and mapping archaeological prospection data thus mainly fails due to the lack of available software suited for this purpose, the LBI ArchPro has invested efforts into the development of tools that enable novel ways of visualising, animating, and fusing multiple data layers (e.g. ApMag, ApRadar, ArchaeoAnalyst, and TAIFU). Of relevance to this article is the first use of TAIFU, a toolbox that enables true data integration. Ideally, TAIFU should allow for the creation of a unified output image, which keeps all relevant features of the various input image modalities while also rendering certain features more distinct. Since image fusion has hardly been used in any type of archaeological prospection research (apart from pansharpening algorithms that are popular in satellite remote sensing), this article provides the first insights into the benefits—but also the risks—of relying on fused data sets for interpretative mapping of geophysical prospection data. The aim of the above-presented fusion of prospection data was for an improved representation of longhouse H2. In particular, the visibility of all the inner postholes (extracted from the radar data) with a simultaneous retention of the outer walls—shown more distinct in the magnetics—was one of the positive results. A great advantage of using image fusion for the here-presented prospection data sets should be seen in a better depiction of structures that would otherwise be less clearly visible in the individual data sets; the representation of postholes could particularly be improved throughout all GPR depth slices. In the case of H2, these were shown as absorbing features in the upper layers, while in the lower layers, they created strong reflective anomalies. A simple merger would have resulted in a deterioration of their recognisability. For a rather homogeneous data set with a larger number of corresponding structures, as encountered at various sites in Denmark (Filzwieser et al. 2017; Nau et al. 2015), this approach could prove to be highly effective. Another problem that could be overcome by image fusion was the oftenencountered fact that some posthole-related magnetic anomalies had a far larger spatial extent than the actual structures, while other postholes were only very faintly visible. By fusing the magnetic data visualisation with the GPR data, the exaggerated positive magnetic anomalies were reduced to more plausible extents, while the barely visible features became more prominent.

Integration of Prospection Data from Denmark

Although successful, a number of critical issues must be taken into account when dealing with fused imagery. As the process of fusion seeks a targeted, combined—and to a certain extent simplified—depiction of the archaeological remains, the fused output might come at the price of losing some information (the degree of which is dependent upon the method used). In order to evaluate and to minimise this loss, all image layers were loaded into ArcMap 10.2 and also interpreted separately. The interpretation of the fused image had the clear advantages that the extent of the structure could be observed much more explicitly and that the chance of attributing features to the building that belong to neighbouring structures was greatly reduced. At the same time, the exact local limitation and concentration of the relevant postholes and ditches was enhanced. Also, when compared with the individual interpretations of the magnetics and the GPR data separately, the fused image was more revealing than either of the others (Fig. 11). Yet, where the lack of information in one data set negated the other’s (i.e. for example where traces of the ditch were missing in the GPR data in contrast to the magnetics), the combined interpretation of all individual geophysical data images provides sometimes still more detailed information than the composite fused output (Fig. 11). However, it needs to be stressed that this result is just the first assessment of a possible image-fusion-based workflow. Since TAIFU offers a wealth of fusion algorithms, most of them with a multitude of adjustable parameters, it might well be that other methods and/or parameters still have the potential to overcome these limitations. One possible solution for this negation could be, for example, to visualise the data sets in perception-based colour schemes. TAIFU has several perception-based divergent colour schemes implemented. They are especially suited for data with a natural or artificial mid-point and symmetrically distributed values, such as magnetic data. In addition, also perceptually linear sequential colour schemes can provide good visualisations. However, given not only the possibilities but also the intricate issues of applying such colour schemes for data visualisation, their assessment will not be tackled here but be the topic of a future publication. In addition, further research should

Fig. 11 Interpretation map of house H2. Comparison between the interpretation of a fused GPR + magnetic image versus the separate individual interpretations

Filzwieser et al.

also establish if other data processing and visualisation routines are needed in order to create the input images. Even if it would turn out that no further improvement is possible, the positive effect of a more coherent and better readable presentation of the building clearly outweighs the mentioned disadvantages. However, what will likely not change with other fusion methods or visualisation parameters is the lost relationship between pixel values and the underlying geophysical phenomena. As an example, thermoremanent anomalies or dipoles are easily confused with contrasts in dielectric permittivity, responsible for features observed in GPR data, if one loses track of the underlying input source data. This can be shown at a strong negative anomaly in the magnetic data (most likely created by an iron object), which compromises the fused output image in the south-western corner of H2 (Fig. 8). After the fusion, it is no longer possible to safely decide whether this anomaly derives from the magnetic or the GPR data. The feature has furthermore already been altered by the initial greyscale inversion of one of the GPR images (Fig. 7). At this point, it thus seems that it would be essential to conduct a semi-automatic detection of iron objects, as well as a reduction of dipole anomalies through image filtering prior to the image fusion (although, again, this might depend on data, soil, and fusion conditions). But, even then, it will most likely always remain important to consult the original imagery for a proper understanding of the physical principles and soil conditions that gave rise to the amplitude values depicted by the pixels. In that respect, archaeological interpretative mapping of image fusion data seems very similar to interpreting highly processed airborne and spaceborne remote sensing imagery. Although the latter also have the potential to reveal previously unnoticed features, their archaeological interpretation also greatly benefits from accompanying normal imagery shot in the visible spectrum (Verhoeven 2009). Furthermore, one cannot assume that the positive results of one fusion method with respect to a specific feature will automatically translate to all other comparable structures within the survey area. Although individual fusion processes could be executed for all relevant features, future research should focus on routines that have the potential to enhance the majority of features in a data set. Only then will image fusion live up to the expectation of being a very powerful tool that enables a more comprehensive and standardised approach to identify and map archaeological remains in various large-area prospection data.

Conclusions The complementary use of high-resolution GPR and magnetometer measurements together with aerial imagery, and the analysis of the integrated archaeological interpretation, was able to provide a considerable increase of detailed insight into the Iron Age settlement at Vesterager. The results, in particular the discovery of further ring ditches and many larger pits as well as the more comprehensive depiction of the longhouses, underline the potential of this approach. The detailed GPR data was especially of great use for the documentation of smaller structures as well as the detection of successive settlement phases, while other prospection techniques were limited. At the same time, the prospection data was in good agreement with the excavation results. The initially stated aim of overcoming some limitations of the aerial photography by providing new

Integration of Prospection Data from Denmark

complementary data sets, and therefore deriving at a more meaningful interpretation of the settlement, can thus be regarded as achieved. The obtained data, however, was additionally complemented by various image fusion approaches through the use of TAIFU, which led to an enhanced data integration. The advantages of implementing this novel post-processing step into an already standardised interpretation process can not only be seen as a more comprehensive representation of individual anomalies caused by archaeological structures but also as a possible increase of efficiency by detecting these structures within entire archaeological landscapes through the use of large-area prospection data. Finally, it can be concluded that image fusion, in addition to a whole series of improvements, also entails a number of risks, which must be minimised and surpassed in future research. Therefore, the use of image fused outputs with a simultaneous comparison to conventional approaches is probably the most effective way of finding new and promising routines for data integration aiding the interpretation of archaeological prospection data. In the case of Vesterager, the interpretative mapping of aerial images and newly acquired, dissimilar geophysical data on the one hand, as well as some fused geophysical data products on the other, could be determined as an extremely useful method for the investigation of prehistoric settlements in western Denmark. Therefore, it would be highly desirable to systematically apply the approach presented here to other, comparable sites in Northern Europe. Acknowledgements The Ludwig Boltzmann Institute for Archaeological Prospection and Virtual Archaeology (archpro.lbg.ac.at) is based on the cooperation of the Ludwig Boltzmann Gesellschaft (A), the University of Vienna (A), the Vienna University of Technology (A), ZAMG, the Austrian Central Institute for Meteorology and Geodynamics (A), the Province of Lower Austria (A), Airborne Technologies (A), 7reasons (A), the Austrian Academy of Sciences (A), the Austrian Archaeological Institute (A), RGZM, the Roman-Germanic Central Museum Mainz (D), the National Historical Museums—Arkeologerna (S), the University of Birmingham (GB), Vestfold County Council (N), and NIKU—the Norwegian Institute for Cultural Heritage Research (N). Holstebro Museum is one of 27 archaeological museums in Denmark and home for the first bigger project on aerial archaeology and non-destructive archaeology in Denmark. Furthermore, we would like to thank all the farmers and landowners for permitting our work on their fields, as well as the anonymous reviewers for their critical comments and Kelly Gillikin for proofreading.

References Conyers, L. B. (2013). Ground-penetrating radar for archaeology. Lanham, California: AltaMira Press. Doneus, M., & Neubauer, W. (1998). 2D combination of prospection data. Archaeological Prospection, 5(February), 29–56. doi:10.1002/(sici)1099-0763(199803)5:1<29::aid-arp92>3.0.co;2-t. Draganits, E., Doneus, M., Gansum, T., Gustavsen, L., Nau, E., Tonning, C., et al. (2015). The late Nordic Iron Age and Viking Age royal burial site of Borre in Norway: ALS- and GPR-based landscape reconstruction and harbour location at an uplifting coastal area. Quaternary International, 367, 96–110. doi:10.1016/j. quaint.2014.04.045. Eriksen, P., & Henningsen, H. (1995). En gård fra vikingetiden nær Staby kirke. FRAM, 52–63. Eriksen, P., & Olesen, L. H. (2000). Vikingetiden. In K. Dalsgaard, P. Eriksen, J. V. Jensen, & J. R. Rømer (Eds.), Mellem hav og hede. Landskab og bebyggelse i Ulfborg herred indtil 1700 (pp. 144–173). Aarhus. Eriksen, P., & Olesen, L. H. (2002). Fortiden set fra himlen. Luftfotoarkæologi i Vestjylland. Holstebro: Holstebro Museum. Eriksen, P., Egeberg, T., Olesen, L. H., & Rostholm, H. (2009). Vikinger i Vest - Vikingetiden i Vestjylland. Højbjerg: Jysk Arkæologisk Selskab.

Filzwieser et al. Ethelberg, P., Hardt, N., Poulsen, B., & Sørensen, A. B. (2003). Det Sønderjyske Landbrugs Historie (2nd ed.). Haderslev. Filzwieser, R., Olesen, L. H., Neubauer, W., Trinks, I., Mauritsen, E. S., Schneidhofer, P., et al. (2017). Largescale geophysical archaeological prospection pilot study at Viking Age and Medieval sites in West Jutland. Denmark. Accepted for publication in Archaeological Prospection. doi:10.1002/arp.1576. Gaffney, C., & Gater, J. (2003). Revealing the buried past: geophysics for archaeologists. Tempus. GEUS. (1989). Geological survey of Denmark and Greeland, Digital geological map of Denmark 1:200000. http://www.geus.dk/UK/data-maps/Pages/j200-dk.aspx. Accessed 24 April 2017. Hatt, G. (1953). An early Roman Iron Age dwelling site in Holmsland, West Jutland. Acta Archaeologica, 24, 1–25. Hinterleitner, A., Neubauer, W., & Trinks, I. (2013). Removing the influence of vehicles used in motorized magnetic prospection systems. Proceedings of the 10th International Conference on Archaeological Prospection Vienna, May 29th – June 2nd 2013, 384–386. James, A. P., & Dasarathy, B. V. (2014). Medical image fusion: a survey of the state of the art. Information Fusion, 19, 4–19. doi:10.1016/j.inffus.2013.12.002. Larsson, L., Trinks, I., Söderberg, B., Gabler, M., Dell’Unto, N., Neubauer, W., & Ahlström, T. (2015). Interdisciplinary archaeological prospection, excavation and 3D documentation exemplified through the investigation of a burial at the Iron Age settlement site of Uppåkra in Sweden. Archaeological Prospection. doi: 10.1002/arp . Nau, E., Olesen, L. H., Schneidhofer, P., Gabler, M., Filzwieser, R., & Mauritsen, E. S. (2015). Large-scale high-resolution GPR and magnetic prospection in West Jutland, Denmark. Archaeologia Polona, 53, 485–488. Neubauer, W. (2001). Magnetische Prospektion in der Archäologie. Wien: Verlag der Österreichischen Akademie der Wissenschaften. Neubauer, W. (2004). GIS in archaeology: the interface between prospection and excavation. Archaeological Prospection, 11(3), 159–166. doi:10.1002/arp.231. Neubauer, W., & Eder-Hinterleitner, A. (1997). Resistivity and magnetics of the Roman town Carnuntum, Austria: an example of combined interpretation of prospection data. Archaeological Prospection, 4(4), 179–189. doi:10.1002/(SICI)1099-0763(199712)4:4<179::AID-ARP85>3.0.CO;2-U. Olesen, L. H. (2011). An aerial view of the past—aerial archaeology in Denmark. Remote sensing for archaeological heritage management: proceedings of the 11th EAC Heritage Management Symposium, Reykjavík, Iceland, 25–27 March 2010, EAC Occasi, 275–282. Olesen, L. H., & Mauritsen, E. S. (2015). Luftfotoarkæologi i Danmark. Holstebro: Holstebro Museum. Powlesland, D., Lyall, J., & Donoghue, D. (1997). Enhancing the record through remote sensing: the application and integration of multi-sensor, non-invasive remote sensing techniques for the enhancement of the Sites and Monuments Record. Heslerton Parish Project, N. Yorkshire, England. Internet Archaeology, (2). doi:10.11141/ia.2.4 Pregesbauer, M., Trinks, I., & Neubauer, W. (2013). Automatic classification of near surface magnetic anomalies—an object oriented approach. Proceedings of the 10th International Conference on Archaeological Prospection Vienna, May 29th – June 2nd 2013, 350–353. Sandici, V., Scherzer, D., Hinterleitner, A., Trinks, I., & Neubauer, W. (2013). An unified magnetic data acquisition software for motorized geophysical prospection. Proceedings of the 10th International Conference on Archaeological Prospection Vienna, May 29th – June 2nd 2013, 378–379. Schneidhofer, P., Nau, E., Hinterleitner, A., Lugmayr, A., Bill, J., Gansum, T., et al. (2016). Palaeoenvironmental analysis of large-scale, high-resolution GPR and magnetometry data sets: the Viking Age site of Gokstad in Norway. Archaeological and Anthropological Sciences. doi:10.1007 /s12520-015-0312-x. Schofield, A. J. (1989). Understanding early medieval pottery distributions: cautionary tales and their implications for future research. Antiquity, 63(240), 460–470. Scollar, I., Tabbagh, A., Hesse, A., & Herzog, I. (1990). Archaeological prospecting and remote sensing. Cambridge: Cambridge University Press. Sevara, C., Pregesbauer, M., Doneus, M., Verhoeven, G., & Trinks, I. (2016). Pixel versus object—a comparison of strategies for the semi-automated mapping of archaeological features using airborne laser scanning data. Journal of Archaeological Science: Reports, 5, 485–498. doi:10.1016/j. jasrep.2015.12.023. Trinks, I., Johansson, B., Gustafsson, J., Emilsson, J., Friborg, J., Gustafsson, C., et al. (2010). Efficient, largescale archaeological prospection using a true three-dimensional ground-penetrating radar array system. Archaeological Prospection, 17, 175–186. doi:10.1002/arp.

Integration of Prospection Data from Denmark Trinks, I., Neubauer, W., Doneus, M. (2012). Prospecting archaeological landscapes. In EuroMed 2012 International Conference on Cultural Heritage, Ioannides M. et al (eds), Lecture Notes in Computer Science (LNCS) 7616. Springer-Verlag: Berlin Heidelberg; 21–29. Trinks, I., Neubauer, W., Nau, E., Gabler, M., Wallner, M., Hinterleitner, A., Biwall, A., Doneus, M., & Pregesbauer, M. (2013). Archaeological Prospection of the Unesco World Cultural Hertitage Site BirkaHovgården. Proceedings of the 10th International Conference on Archaeological Prospection Vienna, May 29th – June 2nd 2013, 39–40. Trinks, I., Neubauer, W., & Hinterleitner, A. (2014). First high-resolution GPR and magnetic archaeological prospection at the Viking Age settlement of Birka in Sweden. Archaeological Prospection, 21(3), 157– 234. doi:10.1002/arp. Trinks, I., Neubauer, W., Doneus, M., Hinterleitner, A., Doneus, N., Verhoeven, G., et al. (2015). Interdisciplinary archaeological prospection at unprecented scale and resolution. The first five years of the LBI ArchPro Research Initiative 2010-2015. Archaeologia Polona, 53, 144–147. Verhoeven, G. (2009). Beyond conventional boundaries: new technologies, methodologies, and procedures for the benefit of aerial archaeological data acquisition and analysis. University of Gent. Verhoeven, G., Nowak, M., & Nowak, R. (2016). Pixel-level image fusion for archaeological interpretative mapping. In E. U. P. de València (Ed.), Proceedings of ARQUEOLÓGICA 2.0 – 8th International Congress on Archaeology, Computer Graphics, Cultural Heritage and Innovation. Advanced 3D documentation, modelling and reconstruction of cultural heritage objects, monuments and sites. ARQUEOLÓGICA 2.0, (pp. 404–407). Valencia. Webley, L. (2008). Iron Age households—structure and practice in Western Denmark, 500 BC - AD 200. Højbjerg. Zimmermann, W. H. (1991). Erntebergung in Rutenberg und Diemen aus archäologischer und volkskundlicher Sicht. Néprajzi Értesítö, 71–73, 71–104.