High-Resolution Direct Push Sensing in Wetland Geoarchaeology—First Traces of Off-Site Construction Activities at the Fossa Carolina

: Wetland environments, with their excellent conservation conditions, provide geoarchaeological archives of past human activities. However, the subsurface soil is difﬁcult to access due to high groundwater tables, unstable sediments, and the high cost of excavation. In this study, we present a ground-based non- and minimal-invasive prospection concept adapted to the conditions of wetlands. We investigated the Fossa Carolina in South Germany, a canal that was intended in 792/793 AD by Charlemagne to bridge the Central European Watershed. Although the resulting Carolingian banks and the fairway with wooden revetments are very imposing, archaeological traces of off-site construction activities have not been identiﬁed hitherto. Based on a geophysically surveyed intensive linear magnetic anomaly parallel to the Carolingian canal, we aimed to prove potential off-site traces of Carolingian construction activities. In this context, we built up a high-resolution cross-section using highly depth-accurate direct push sensing and ground-truthing. Our results showed the exact geometry of the canal and the former banks. Thus, the magnetic mass anomaly could be clearly located between the buried organic-rich topsoil and the Carolingian banks. The thermoluminescence dating showed that the position of the magnetic mass anomaly reﬂected Carolingian activities during the construction phases, speciﬁcally due to heat exposure. Moreover, we found hints of the groundwater supply to the 5-metre wide navigable fairway.


Challenging Issues in Geoarchaeological Wetland Exploration
Wetlands provide important geoarchaeological archives about past landscape development, e.g., [1][2][3] as well as buried infrastructure and settlements, e.g., [4][5][6]. However, unfavourable properties like unstable sediments and a high groundwater table lead to high costs for direct access by excavations [7][8][9]. The post-excavation conservation of archaeological materials is expensive and complicated as well [10,11]. Thus, there is a demand for remote, non-invasive, and minimal-invasive exploration approaches for cross-scale, multi-method, and systematic surveys in wetland geoarchaeology in order to investigate, but also to monitor, vulnerable archaeological features, e.g., pile dwellings.
In addition, non-invasive geophysical surveys can be conducted from large to small spatial scales [9] in a cost-saving way in a wide range of combinations. They offer different sensitivities depending on physical material characteristics, e.g., suspicious values of the magnetic field strength of fire-affected sediments or striking values of electrical conductivity of saturated ditch fillings, e.g., [12,13]. However, they need sufficient contrast of the material characteristics in the subsurface soil [14,15] and ideally, ground truth data for reliable interpretations [16,17]. In general, for effective conduction and credible results of non-and minimal-invasive exploration methods, a well thought-out field strategy with the on-site decision [18] as well as a combination of carefully selected parameters with suitable sensitivities for the targeted features are necessary [14,19,20].
In terms of ground-truthing, minimally invasive vibracoring provides important point-by-point insight into the sediments and sampling opportunities, e.g., [21,22], but can record depths inaccurately, especially in wetlands caused by high compaction rates of the sediments [23]. Another method that has been applied recently as an alternative to ground truth [17,24] and for geoarchaeological exploration per se on decametre to sub-metre scale [16,25,26] is minimal-invasive and depth-accurate direct push sensing [27,28]. The colour logging tool (e.g., for peat detection [26,29]) and electrical conductivity logging (e.g., for sediment unit surveys [24,25]) have proven successful in geoarchaeological studies.

The Fossa Carolina-Previous Geoarchaeological Findings
The Fossa Carolina is an Early Medieval canal in southern Germany located in the foothills of the Southern Franconian Alb (Figure 1). The canal was designed in 792 AD and built in 792/793 AD by Charlemagne, linking the Rhine and Danube river systems and thus bridging the Central European Watershed, which is confirmed by dendrochronological dating [4,30] and written sources [30][31][32]. Small altitude differences (approximately 6 m) and the short distance between the Altmühl and Swabian Rezat rivers (approximately 2 km as the crow flies) provided favourable conditions for this project [23,33,34].
The canal is approximately 3 km in length and has a conspicuous s shape (Figure 1c) in order to minimise the work needed to complete the respective section [35]. According to dendroarchaeological findings, it is assumed that the work progressed from north to south towards the summit [36]. However, the southernmost part of the canal was never finished, as no archaeological remains were detected in the Altmühl floodplain [37]. Written sources explain this abandonment as being due to unstable trench edges and strong rainfalls [31]. Subsequently, the trenched areas were used as ponds to some extent until they were completely silted up [23,38].
The construction of the summit canal was designed as a stepped chain of ponds with approximately 2.5-5 m fairway width [4,29,30,39] and a northward shifted summit [36,39]. Large oak timbers were used for the stabilisation of the canal edges. The canal can be divided into different sections (Figure 1c). The central section and the west-east section are still marked by impressive banks from Carolingian excavation works. After a bend to the north, the banks become more and more shallow in the northern section, and the course of the canal is hardly to not at all visible in the northernmost north-eastern section [39,40].  [41], OSM-Open Street Map, Zielhofer et al. [39], LiDAR data are provided by the Bavarian Land Surveying Office).

Linear Magnetic Anomaly at the Fossa Carolina
A large-scale superconducting quantum interference device (SQUID) magnetic survey [42] revealed a conspicuous anomaly parallel to the canal course in the northern and southernmost north-eastern section ( Figure 2). Especially noticeable are the straight course and the interruptions (Figure 2e). In the grid view (range ± 10 nT/m), the anomaly is usually approximately between 7 and 8 m wide. The segment lengths vary between approximately 30 and 80 m. In situ volume magnetic susceptibility (κ) measurements at the anomaly's position (core QP2 position in Figure 2b, [43]) revealed the highest values of approx. 45 000 × 10 −6 SI at depths between 135 and 155 cm, where the magnetic mass anomaly is characterised by black, reddish-brown, and red colours. Further rock-magnetic laboratory analysis indicated a titanomagnetite/magnetite layer, which required heating over 700 °C [43]. However, to date, there is no reliable stratigraphic positioning of the anomaly, no age control, and therefore no clear knowledge about its possible origin.  [39], LiDAR data are provided by the Bavarian Land Surveying Office).

Linear Magnetic Anomaly at the Fossa Carolina
A large-scale superconducting quantum interference device (SQUID) magnetic survey [42] revealed a conspicuous anomaly parallel to the canal course in the northern and southernmost north-eastern section ( Figure 2). Especially noticeable are the straight course and the interruptions (Figure 2e). In the grid view (range ± 10 nT/m), the anomaly is usually approximately between 7 and 8 m wide. The segment lengths vary between approximately 30 and 80 m. In situ volume magnetic susceptibility (κ) measurements at the anomaly's position (core QP2 position in Figure 2b, [43]) revealed the highest values of approx. 45,000 × 10 −6 SI at depths between 135 and 155 cm, where the magnetic mass anomaly is characterised by black, reddish-brown, and red colours. Further rock-magnetic laboratory analysis indicated a titanomagnetite/magnetite layer, which required heating over 700 • C [43]. However, to date, there is no reliable stratigraphic positioning of the anomaly, no age control, and therefore no clear knowledge about its possible origin.

Aims of this Study
In this study, (i) we aimed to apply an advanced minimal-invasive, multi-method approach at the buried wetland site to explore the intensive magnetic anomaly. We combine direct push colour and electrical conductivity logging for a full 2D cross-section beyond the vertical and lateral borders of the Fossa Carolina and link the data with vibracoring ground truth data and subsequent laboratory analysis.
Then, we aimed to (ii) reconstruct the chronostratigraphy of the Carolingian canal fills and adjacent banks in order to determine the exact stratigraphic position of the magnetic mass anomaly outside the canal course. Doing so will provide chronological control for what is likely to be the first traces of Carolingian off-site construction activities. Moreover, we aimed to collect evidence concerning navigability and the potential water supply. The DEM provides indications of the location of the buried banks in the form of elevation that is slightly visible (b,c). The magnetic SQUID survey revealed an extensive, intensive magnetic anomaly (d,e,f) [42] parallel to the canal, which can be divided into segments (1-5) by its interruptions (e). (b-f) Locations of the preliminary vibracoring transect [39] and the archaeological excavation [4] are marked. (Database: LiDAR data are provided by the Bavarian Land Surveying Office).

Aims of this Study
In this study, (i) we aimed to apply an advanced minimal-invasive, multi-method approach at the buried wetland site to explore the intensive magnetic anomaly. We combine direct push colour and electrical conductivity logging for a full 2D cross-section beyond the vertical and lateral borders of the Fossa Carolina and link the data with vibracoring ground truth data and subsequent laboratory analysis.
Then, we aimed to (ii) reconstruct the chronostratigraphy of the Carolingian canal fills and adjacent banks in order to determine the exact stratigraphic position of the magnetic mass anomaly outside the canal course. Doing so will provide chronological control for what is likely to be the first traces of Carolingian off-site construction activities. Moreover, we aimed to collect evidence concerning navigability and the potential water supply.

Geographical Setting of the Fossa Carolina
The Fossa Carolina is located in the foothills of the Franconian Jura escarpment (Figure 3). The valley floors are characterised by clayey to sandy glacio-fluvial sediments from the Last Glacial. Holocene fen deposits (Rezat Fen), as well as Holocene flood deposits in the Altmühl and Rezat floodplains, have built up the present surface [33,44,45].
The geological composition and topography determine the hydrological setting. In the valley, Middle Jurassic and Miocene clayey sequences sealed the valley bottom. At the foothill positions, Middle and Upper Jurassic clayey sequences [46] correspond with spring horizons, e.g., the Swabian Rezat spring [33,44,45]. In the northern section of the  [42] parallel to the canal, which can be divided into segments (1-5) by its interruptions (e). (b-f) Locations of the preliminary vibracoring transect [39] and the archaeological excavation [4] are marked. (Database: LiDAR data are provided by the Bavarian Land Surveying Office).

Geographical Setting of the Fossa Carolina
The Fossa Carolina is located in the foothills of the Franconian Jura escarpment ( Figure 3). The valley floors are characterised by clayey to sandy glacio-fluvial sediments from the Last Glacial. Holocene fen deposits (Rezat Fen), as well as Holocene flood deposits in the Altmühl and Rezat floodplains, have built up the present surface [33,44,45].
The geological composition and topography determine the hydrological setting. In the valley, Middle Jurassic and Miocene clayey sequences sealed the valley bottom. At the foothill positions, Middle and Upper Jurassic clayey sequences [46] correspond with spring horizons, e.g., the Swabian Rezat spring [33,44,45]. In the northern section of the canal, the continuously increasing discharge of the Swabian Rezat river points to a substantial groundwater supply because tributaries are absent in this zone [47]. These conditions lead to a high groundwater table that is indicated by the presence of the Rezat Fen and buried organic-rich topsoil that is widespread [8,16,33]. canal, the continuously increasing discharge of the Swabian Rezat river points to a substantial groundwater supply because tributaries are absent in this zone [47]. These conditions lead to a high groundwater table that is indicated by the presence of the Rezat Fen and buried organic-rich topsoil that is widespread [8,16,33].

SQUID Magnetic Prospection and Depth Calculation of Intensive Magnetic Anomalies
For the large-scale magnetic prospection ( Figure 4), a motorised SQUID measuring instrument [48,49] was used. The high-resolution device recorded magnetic subsurface data with a configuration of 18 SQUID gradiometers and magnetometers. The corresponding position was localised synchronously by a differential GPS. Precisely georeferenced magnetograms of the entire canal area were created [42,50,51]. The recorded linear anomalies could not always be delineated sharply, for a variety of reasons, such as limited measurement possibilities (Figure 2e, segment 4) or a diffuse signal (Figure 2, segments 1 and 5). Segment 2 (Figure 2e), the segment that was investigated in our study in detail, is clearly recognisable. The subsurface distribution of the magnetic sources could be determined by the maxima of the magnetic information [52,53]. In this study, the un-gridded magnetic data of several SQUID sensors were used to calculate the depth of the magnetic sources from which the intense magnetic deviation originate. Calculation basics are the anomaly maxima and minima recorded by pairs of SQUID sensors [54] and a 1/r 3 amplitude dependence from the source-sensor distance r. After the evaluation of the SQUID magnetic data, we performed high-resolution direct push sensing and laboratory analysis from depth-corrected vibracoring samples. For age control, we used both thermoluminescence and radiocarbon dating ( Figure 4).

SQUID Magnetic Prospection and Depth Calculation of Intensive Magnetic Anomalies
For the large-scale magnetic prospection ( Figure 4), a motorised SQUID measuring instrument [48,49] was used. The high-resolution device recorded magnetic subsurface data with a configuration of 18 SQUID gradiometers and magnetometers. The corresponding position was localised synchronously by a differential GPS. Precisely georeferenced magnetograms of the entire canal area were created [42,50,51]. The recorded linear anomalies could not always be delineated sharply, for a variety of reasons, such as limited measurement possibilities (Figure 2e, segment 4) or a diffuse signal (Figure 2, segments 1 and 5). Segment 2 (Figure 2e), the segment that was investigated in our study in detail, is clearly recognisable. The subsurface distribution of the magnetic sources could be determined by the maxima of the magnetic information [52,53]. In this study, the un-gridded magnetic data of several SQUID sensors were used to calculate the depth of the magnetic sources from which the intense magnetic deviation originate. Calculation basics are the anomaly maxima and minima recorded by pairs of SQUID sensors [54] and a 1/r 3 amplitude dependence from the source-sensor distance r. After the evaluation of the SQUID magnetic data, we performed high-resolution direct push sensing and laboratory analysis from depth-corrected vibracoring samples. For age control, we used both thermoluminescence and radiocarbon dating ( Figure 4).

Direct Push Sensing
Direct push techniques provide a set of tools for minimal-invasive, depth-accurate in situ measurements by pushing steel rods equipped with different probes into the unconsolidated underground [27,55]. We used 38-millimetre steel rods at the colour logging tool (CLT) and the electrical conductivity probe (EC) that were driven by a Geoprobe 6610 DT caterpillar ( Figure 5). Based on on-site decisions, our sampling points were placed in distances of 0.25 to 2 m from each other along the cross-section. The depths were precisely logged by a potentiometer. For exact localisation, we defined each position using a Topcon HiPer II DGPS with an accuracy of 10 mm horizontally and 15 mm vertically.

Direct Push Sensing
Direct push techniques provide a set of tools for minimal-invasive, depth-accurate in situ measurements by pushing steel rods equipped with different probes into the unconsolidated underground [27,55]. We used 38-millimetre steel rods at the colour logging tool (CLT) and the electrical conductivity probe (EC) that were driven by a Geoprobe 6610 DT caterpillar ( Figure 5). Based on on-site decisions, our sampling points were placed in distances of 0.25 to 2 m from each other along the cross-section. The depths were precisely logged by a potentiometer. For exact localisation, we defined each position using a Topcon HiPer II DGPS with an accuracy of 10 mm horizontally and 15 mm vertically.

Electrical Conductivity Logging (EC)
The EC probe (SC-500; Keijr Engineering Inc., Geoprobe Systems, Salina, KS, USA) was used for the differentiation of sediment units. The probe is equipped with four electrodes in a Wenner array with a vertical alignment and a fixed distance of 2 cm. This enabled the analysis of layers with thicknesses ranging from 5 to 10 cm [59][60][61]. The probe measured the specific electrical resistivity, which is inversely expressed as the specific electrical conductivity in mS/m in fixed 1.5-centimetre steps [62,63].

Electrical Conductivity Logging (EC)
The EC probe (SC-500; Keijr Engineering Inc., Geoprobe Systems, Salina, KS, USA) was used for the differentiation of sediment units. The probe is equipped with four electrodes in a Wenner array with a vertical alignment and a fixed distance of 2 cm. This enabled the analysis of layers with thicknesses ranging from 5 to 10 cm [59][60][61]. The probe measured the specific electrical resistivity, which is inversely expressed as the specific electrical conductivity in mS/m in fixed 1.5-centimetre steps [62,63].
The values represent a sum parameter of the sediment properties of grain size, fluid saturation, chemical properties, and salinity of the pore fluid [63,64]. In saturated sediments, relatively high values usually indicate fine-grained sediments, such as clay, and lower values indicate coarser-grained sediments, such as silt, sand, and gravel [62,63].
We added spacing between the individual electrical conductivity logs from 0.5 to 3 m.

Vibra-Coring
We used vibracoring for the ground-truthing and sediment sample recovery. The open cores (60 mm diameter, 1 m length) were driven by an Atlas Copco Cobra Pro motorised hammer and localised with the Topcon HiPer II DGPS. In the field, we documented sediment properties, soil horizons, and reductive-oxidative properties using the KA5 soil mapping guide [65] and Munsell colours [66]. We sampled sediment sequences with an eye to stratigraphical changes as well as in equidistant 5-centimetre steps and corrected the sample depths by matching core sequences with depth-accurate direct push RGB colour logs ( Figure S1).

Geochemical and Rock Magnetic Sediment Analysis
We used elemental analysis by the vario EL cube (Elementar) to analyse the total content of sulphur, carbon, and nitrogen (CNS) to specify and divide the sediment stratigraphy. Furthermore, we used a Scheibler calcimeter (Eijkelkamp) to quantify inorganic carbon content. To calculate the organic carbon, we subtracted the inorganic carbon from the total carbon. We used laboratory measurements of volume-specific low-frequency magnetic susceptibility (κ, 0.465 kHz, 10 −6 SI) to detect magnetic anomalies within the recovered cores. Then, we filled plastic cubes with the sediment and used a Bartington MS3 equipped with an MS2B dual-frequency sensor [67,68].

Thermoluminescence (TL) Dating
Two samples were taken from the magnetic mass anomaly for TL dating because this is the most suitable luminescence method for heated material [69][70][71].
The samples were taken from an opaque core liner (core BK5) at depths of 135-140 cm (KG-TL1) and 150-155 cm (KG-TL2) and were subsequently prepared under subdued red light in the luminescence preparation room of the Institute of Geography at Leipzig University. Sample preparation included sieving with meshes with widths of 90 and 200 µm, treatments with 10% and 30% HCl for calcium carbonate destruction, and with 10% and 37% H 2 O 2 over 14 days to destroy organic matter. Subsequently, the fine grain fraction (4-11 µm) was separated according to Stoke's Law in Atterberg settling tubes.
In tandem with the luminescence samples, dose rate samples were taken from the core. The sample material from KG-TL1 was taken from its surroundings. Given that sample KG-TL2 was located near a stratigraphical border, one sample was taken from the lightcoloured material above (156-140 cm; DL-1) and another from the dark-coloured material below (170-158 cm; DL-2). Both equivalent dose (De) and dose rate determinations were carried out in the luminescence laboratory of the Curt-Engelhorn-Centre of Archaeometry (Mannheim, Germany).
The De was measured with blue thermoluminescence (multiple aliquots according to Aitken [72]) using a Risø TL-DA-20 device equipped with a combination of BG3 and BG39 (both 3-millimetre thick) filters and a 90 Sr/ 90 Y β-source (0.079 Gy/s). The a value was determined with a 241 Am α-source (0.116 Gy/s).
The sediment dose rates (contents of the radioactive elements U, Th, and K) were measured with low-level γ-spectrometry using a Canberra GCW4023 high-purity Germanium well detector. The water content of the samples was determined gravimetrically by weighing the samples before and after drying at 105 • C for 24 h. The cosmic dose rate was calculated according to Prescott and Hutton [73], assuming a density of the overlying sediments of 1.6 g/cm 3 .
Possible anomalous fading of the samples was checked by measuring five irradiated and five not irradiated aliquots of every sample and comparing those results with identical measurements of the same number of aliquots after a storage time of approximately 6 weeks.

Radiocarbon Dating
Seeds, charcoal, wood fragments, and organic soil bulk samples were radiocarbon dated. The samples with the MAMS laboratory code were analysed at the Curt-Engelhorn-Centre of Archaeometry (Mannheim, Germany) with an accelerator mass spectrometry (AMS) type MICADAS [74]. The sample with the KIA laboratory code was analysed at the

Age Control by Archaeological Findings
An adjacent archaeological excavation in 2013 (for location, see Figure 2) provided a high-resolution stratigraphy and age control [4] by archaeological finds like ceramics and horseshoe and the dendrochronological dating of oak timber, as well as by radiocarbon dating of short-lived botanical remains. We used these chronostratigraphic findings to refine our understanding of the canal-fill stratigraphy within our recently conducted direct push cross-section.

Chronostratigraphical Units of the Fossa Carolina Cross-Section
We generated a high-resolution 2D cross-section with the direct push colour and electrical conductivity logging as well as vibracoring ground-truthing, which shows different

Chronostratigraphical Units of the Fossa Carolina Cross-Section
We generated a high-resolution 2D cross-section with the direct push colour and electrical conductivity logging as well as vibracoring ground-truthing, which shows different subsurface units in a vertical and lateral direction (Figures 7 and 8 and Table S1 (Supplementary Materials)).  Laboratory results of analysis of the samples from core BK3 (a) and BK18 (b) compared with direct push sensing data. Within core BK 3, a magnetic mass anomaly is evidenced by the maximum of the magnetic susceptibility. Core BK18 is located at the centre of the canal. The core data were depth-corrected with reference to direct-push colour logs ( Figure  S1).
Unit II is characterised by greyish colours in the lower part up to approximately 415 m a.s.l., brownish colours in the upper part, and a dark brown to black layer at approximately 416.5 m a.s.l. It shows a horizontally bedded stratigraphy, with electrical conductivities generally higher (<60 mS/m) than those of Unit I. We drilled vibracoring BK3 at the position of the intensive magnetic anomaly or rather at the margin of the canal structure (Figure 7 and 8a). In Unit II, 416.03-416.57 m a.s.l., we recorded sandy to loamy sediments with several finer-grained layers. The rising values of low-frequency magnetic susceptibility, organic carbon content, and CNS are striking. The values of the Luminosity (L) show bright colours in the lower part and dark colours with minimal values at the transition to the overlying Unit III-a. The cross-section within Unit II shows a reductive-oxidative layering that could have been affected by the groundwater table. The increase in car- Figure 8. Laboratory results of analysis of the samples from core BK3 (a) and BK18 (b) compared with direct push sensing data. Within core BK 3, a magnetic mass anomaly is evidenced by the maximum of the magnetic susceptibility. Core BK18 is located at the centre of the canal. The core data were depth-corrected with reference to direct-push colour logs ( Figure S1).
The lowermost Unit I is characterised by brownish colours up to approximately 414 m a.s.l. and shows a horizontally bedded stratigraphy with relatively low electrical conductivities (<20 mS/m). According to core BK18 (412.06-412.96 m a.s.l., Figure 8b), predominantly sandy sediments appear. The CNSand inorganic carbon record is heterogeneous but relatively low and slightly enhanced values of the low-frequency magnetic susceptibility were measured. Furthermore, the unit is nearly plant fragment free. The record was interpreted as sandy valley fills of the Pleistocene or Early Holocene below the canal.
Unit II is characterised by greyish colours in the lower part up to approximately 415 m a.s.l., brownish colours in the upper part, and a dark brown to black layer at approximately 416.5 m a.s.l. It shows a horizontally bedded stratigraphy, with electrical conductivities generally higher (<60 mS/m) than those of Unit I. We drilled vibracoring BK3 at the position of the intensive magnetic anomaly or rather at the margin of the canal structure (Figures 7 and 8a). In Unit II, 416.03-416.57 m a.s.l., we recorded sandy to loamy sediments with several finer-grained layers. The rising values of low-frequency magnetic susceptibility, organic carbon content, and CNS are striking. The values of the Luminosity (L) show bright colours in the lower part and dark colours with minimal values at the transition to the overlying Unit III-a. The cross-section within Unit II shows a reductive-oxidative layering that could have been affected by the groundwater table. The increase in carbon content along with low Luminosity (L) values in the uppermost part indicates a buried topsoil layer. A striking increase in the red-green ratio (a), as well as in low-frequency magnetic susceptibilities, indicate an impact of heating in this zone (Figure 8a). Thus, this layer corresponds with the intensive magnetic anomaly that was detected during the SQUID survey (Figures 2 and 6). The modelled level of approximately 416.9 m a.s.l. is comparable to the elevation of the magnetic mass anomaly at approximately 416.6 m a.s.l. Radiocarbon dating ( Table 1) of an organic bulk sample from the fossil topsoil on the level of the magnetic mass anomaly provided a middle Holocene age of the soil formation. For chronological control of the magnetic mass anomaly, we chose to use thermoluminescence dating (TL). The TL dating of the magnetic mass anomaly or, respectively, the dating of the intensively heated sediment (part of Unit II and Unit III-a above) yielded a Carolingian age (Tables 2 and S2, Figure S2).  On top of the buried soil (above Unit II), a dark brownish sediment cover is presented (Unit III-a), which is directly adjacent to the trough structure (Unit III-d and III-e). In the core BK3, Unit III-a (416.57-417.40 m a.s.l.) presents coarser sediments with dark brown to dark yellowish-brown colours (Figure 8a) as well as hydromorphic oxidation and reduction features. CNS and the organic and inorganic carbon content are low. However, low-frequency magnetic susceptibility reaches its maximum at 416.49-416.79 m a.s.l., accompanied by the red-green ratio (a) maximum. The unit displays the sediments of the Carolingian excavation works that are preserved as buried banks; furthermore, the lower part of the unit was impacted by heat. The units are disrupted within the central section (10-24 m on the profile) of the NW-SE transect by inhomogeneous Unit III-b, which shows greyish and brownish colours below 412.9 m a.s.l. In the core BK 18, the Unit III-b (412.96-413.29 m a.s.l.) is characterised by the first appearance of plant fragments (described in the field), whereas the values of CNS and low-frequency magnetic susceptibility are low. The EC-values increase up to >50 mS/m. Along with a slightly fining upward trend, this could represent a first backfill process of bank material.
On the south-eastern side of Unit III-d, a wedge-shaped interruption (Unit III-c) is situated. This represents early backfill sediments as well, which were also radiocarbon dated to Carolingian times (Table 3). Unit III-d presents a trough-shaped sediment fill, which shows blackish colours up to 415 m a.s.l. as well as dark brown colours up to the top of Unit III-e. In core BK 18 (413. .09 m a.s.l.), Unit III-d consists of organic material with fine-grained clastic sediments that show a fining upward trend and colours between black and dark grey. Furthermore, organic carbon rises to its maximum, which positively correlates with the red-green ratio (a) and inversely with Luminosity (L). The EC values stagnate at approx. 60 mS/m. Further upwards, within Unit III-e (415.09-416.96 m a.sl. in BK18), the grain size becomes coarser, and the colours become brighter, ranging from black to dark yellowishbrown. Hydromorphic oxidation and reduction features appear. The organic carbon strongly decreases, and low-frequency magnetic susceptibility stagnates at a low level. Unit III-d and III-e represent the canal fill, featuring dark-coloured organic-rich sediments from ponding. The much higher EC values (approximately 60 mS/m) in the central part of the cross-section (Unit III-d and III-e) represent relatively fine-grained canal fills. For the chronology of the canal fills (Unit III-d to III-e), published radiocarbon ages were considered (Table 3), which were predominately age controls from the archaeological excavation located at a distance of 270 m to the north [4,30,77]. Accordingly, the peat deposits of Unit III-d and Unit III-e increased over the approximate 200 years after the canal's construction until approximately 1000 AD. Subsequently, fine-grained sediments were deposited by ponding during the High and Late Middle Ages. These filled up the canal to the level of the fossil topsoil.
The uppermost Unit IV shows brownish colours with suspiciously relatively low electrical conductivities (<30 mS/m). Core BK18 (416.96-417.96 m a.s.l.) and core BK3 (417. .03 m a.s.l.) reveal fine-grained loamy sediments. The values of CNS and lowfrequency magnetic susceptibility slightly increase. Unit IV (Tables 3 and 4) is ubiquitous in the landscape inside as well as outside of the canal, which becomes impressively apparent in the direct push colour transect (Figure 7b). It consists of both alluvial and colluvial sediments deposited since the Late Middle Age, according to archaeological age control. This indicates an intensively used landscape.

A Multi-Method Approach with High-Resolution Direct Push Sensing
Non-and minimal-invasive survey methods play an important role in the investigation of archaeological wetland sites when excavations are difficult or undesirable for conservation reasons. Sound approaches require sensitive and complementary methods for the recording of archaeological features at different spatial scales. In this context, we focus on the significance of the results of the SQUID magnetic survey and the high-resolution direct push sensing (Figure 9). This paper reports on a large-scale magnetic SQUID survey [51] undertaken to identify an archaeological structure (Figure 2), i.e., the large-scale intense magnetic anomaly described above. Further, depths of the magnetic mass anomalies were modelled ( Figures  6 and 9) in different segments, offering a third dimension to the data set. This approach has already provided valuable insights concerning circular ditch systems [78], the interpretation of buried walls [79], and potential hydro-engineering structures in the subsurface [16]. The results of magnetic depth modelling allowed the spatial upscaling by tracking the estimated elevation of anomalies in different segments and sections over longer distances ( Figure 6). However, this modelling has an uncertainty of several decimetres, so spatial coupling of the intensive magnetic anomaly segments over larger distances could only be approximated. Therefore, ground-truthing was required. Here, direct push sensing yielded high-resolution data concerning buried geoarchaeological structures along both horizontal and vertical axes. Direct push colour logging was utilised to determine the precise position of the magnetic mass anomaly within the stratigraphy of the valley fills and intercalated canal remains (Figure 7). This advanced the previous application of direct push colour logging as point-by-point-insight for the extension of sedimentological descriptions [25,58,80]. Additionally, direct push electrical conductivity sensing resulted in a highly depth-accurate dataset that noticeably improved the spatial significance of common, non-invasive ERT applications e.g., [22,37,81]. Thus, there is a high methodological potential for future combinations of depth-accurate direct push electrical conductivity logging as point-by-point ground-truthing with ERT data sets [17,24,82,83].
Finally, we showed that the selection of a complementary combination of methods (Figure 9) is the key to a reliable multi-method, non-to minimal-invasive approach in This paper reports on a large-scale magnetic SQUID survey [51] undertaken to identify an archaeological structure (Figure 2), i.e., the large-scale intense magnetic anomaly described above. Further, depths of the magnetic mass anomalies were modelled (Figures 6 and 9) in different segments, offering a third dimension to the data set. This approach has already provided valuable insights concerning circular ditch systems [78], the interpretation of buried walls [79], and potential hydro-engineering structures in the subsurface [16]. The results of magnetic depth modelling allowed the spatial upscaling by tracking the estimated elevation of anomalies in different segments and sections over longer distances ( Figure 6). However, this modelling has an uncertainty of several decimetres, so spatial coupling of the intensive magnetic anomaly segments over larger distances could only be approximated. Therefore, ground-truthing was required. Here, direct push sensing yielded high-resolution data concerning buried geoarchaeological structures along both horizontal and vertical axes. Direct push colour logging was utilised to determine the precise position of the magnetic mass anomaly within the stratigraphy of the valley fills and intercalated canal remains (Figure 7). This advanced the previous application of direct push colour logging as point-by-point-insight for the extension of sedimentological descriptions [25,58,80]. Additionally, direct push electrical conductivity sensing resulted in a highly depth-accurate dataset that noticeably improved the spatial significance of common, non-invasive ERT applications e.g., [22,37,81]. Thus, there is a high methodologi-cal potential for future combinations of depth-accurate direct push electrical conductivity logging as point-by-point ground-truthing with ERT data sets [17,24,82,83].
Finally, we showed that the selection of a complementary combination of methods ( Figure 9) is the key to a reliable multi-method, non-to minimal-invasive approach in wetland geoarchaeology.

Origin of the Magnetic Mass Anomaly
The SQUID magnetic survey showed an intensive anomaly running parallel to the canal (Figure 6) in the northern and north-eastern sections. Such linear structures in magnetic surveys usually indicate anthropogenic features, e.g., pathways, walls, or ditches [79,84,85]. However, natural linear structures, such as naturally filled gullies, are also known, especially in the vicinity of the Fossa Carolina [16]. The magnetic mass anomaly exhibits a reddish colouring within an intermediate stratigraphical position between the buried mid-Holocene soil (top of Unit II) and the covering Carolingian bank deposits (Unit III-a) (Figure 7b). The TL dating offers a Carolingian age of the heating process. In further segments of the intensive magnetic anomaly, its level was estimated as being close to the former surface of the buried soil ( Figure 6). Thus, based on the location and dating, its characterisation as a remnant of the canal construction site is very likely. Due to its stratigraphical position below the excavation material, its formation at a rather early point in the construction process-or at least before digging ended-seems probable.
Besides the red colouring, we know from the rock magnetic analysis [43] that the sediments contain titanomagnetite, which only forms in the presence of very high temperatures (700 • C). It is found primarily in magmatic rocks [86,87] or secondarily as a product of the weathering of basaltic rocks [88]. Neither exists in the area-see [44,45]-which leads to the assumption that it represents an anthropogenic feature.
Furthermore, the red colour might indicate fire exposure, which would also be supported by the rock magnetic analysis. A forest clearing fire at the beginning of the construction work would be a possibility, but these usually only generate temperatures of approximately 300-400 • C [87,88], too low for the formation of titanomagnetite [43]. Furthermore, the heating by fire decreases strongly from the surface to the depth into the sediments [89], which does not fit the red colouration and increased volume magnetic susceptibility over a depth range of approximately 20 cm, which might indicate deep and intense heating. Additionally, there is no evidence for charcoal in the recovered sediments, which could argue against there having been a fire at this place. Another mechanism could be an anthropogenic input of burnt material, e.g., to stabilise a parallel path for the construction site. This could explain the slight embedding (Figure 7b) of the sediments and their width of approximately 2 m (Figure 2), which is similar to the width of towpaths mentioned in other canal studies [90]. Additionally, remains of iron smelting and ceramics from the Early and High Middle Ages were found diffusely distributed during the inspection of adjacent areas [91]. These suggest the occurrence of activities in the near vicinity that might have produced heated sediments without any relation to the canal. However, the SQUID magnetic survey shows discontinuous segments that do not always run in alignment (Figure 2), which speaks against a path. Furthermore, neither ceramic nor iron was found in the recovered sediments of the magnetic mass anomaly.
Thus, we were able to firmly characterise the magnetic mass anomaly as an anthropogenic structure to the Carolingian canal, but we could not present a conclusive interpretation of the sediment of the magnetic mass anomaly.

New Hints about the Hydro-Engineering Concept and Navigability
Even after numerous studies, questions about the canal summit, water supply, and navigability have not been conclusively answered. The results of this study provide new information about the overall concept.
The direct push colour logs document the canal bottom as being at approximately 413.2 m a.s.l. This is significantly lower than the canal bottom as identified in the 2013 excavation to the north (414.05 m a.s.l. [4]) and significantly higher than the bottom in the west-east section (Figure 1) (approx. 412 m a.s.l. [29]), which is 550 m further southwest (as the crow flies). Hence, there is evidence that the summit of the canal must have been further to the north than previously documented by Zielhofer et al. [39].
The direct push colour logs also provide new insights into the hydro-engineering concept of the canal. The cross-section shows a significant colour change (Figure 7b, Unit II) from greyish colours (below approximately 415 m a.sl.) to yellowish-brown colours (above). This points to the groundwater-dependent reductive and oxidative zone [92] and thus at least to a (semi-stable) groundwater level. Furthermore, the buried organic-rich and boggy topsoil detected at approx. 416.5 m a.s.l. indicate at the very least a semi-terrestrial environment with long-term wet conditions at the Carolingian surface. Since the canal bottom (Unit III-b), at approximately 413.2 m a.s.l., is approximately 1.5-2 m lower than the reductive-oxidative boundary and 3-4 m lower than the organic-rich buried topsoil, an ample groundwater supply at the study site's canal section would be likely.
Furthermore, the direct push colour cross-section (Figure 7b) provides information on the canal geometry (Unit III-b, III-d and III-e). At the level of approximately 414 m a.s.l. (position AN_CLT_39), an edge appears on the south-eastern side, which points to a lateral or rather vertical boundary of the canal. This might point to oak timber revetments, which are well known from the excavations further north [4]. Using the top level of the timbers as the maximum water elevation (approx. 414 m a.s.l.), the canal has a depth of approximately 1 m and a width of approximately 5 m. This would have been suitable for Carolingian cargo scows [93,94] and therefore suggests navigability in this section. Furthermore, the latter dimension is in line with the computed canal width of approximately 5.2-5.3 m at the excavation site of 2013 [4].
Consequently, our results indicate a navigable section with potential for high groundwater supply that could have fed the summit zone of the Carolingian canal with water. This could be the answer to the lack of water supply structures described in studies [16].

Construction Site Remains at Early Medieval Buildings
For a large-scale construction site like the Fossa Carolina that involved hundreds or even thousands of workers, one would expect considerable remains of infrastructure and activities in the immediate area of the canal trench. The range of possible remains of the canal construction site covers several functional groups: traces of transportation and mobility infrastructure (e.g., wooden trackways, paved roads, and river crossings), traces of earthworks (e.g., pits and excavated earth), traces of landscape modifications (e.g., drainage ditches, channel engineering, and terrain levelling), traces of processing and craft activities (e.g., activity areas with finds and features associated with carpenters and metalworkers, huts and roofs to provide shelter, lumberyards for timber, and production waste such as wood chips), traces of lodging and supplies for the workforce (e.g., temporary cabins, tents, fireplaces, food waste, and food containers), and traces of surveying (e.g., postholes made by range rods).
The extent and provability of these functional groups are rather different. Due to the scale of the construction site and the wet and boggy conditions, considerable traces of transportation, especially wooden trackways for pedestrian and wheeled transport, seem rather likely. Such trackways, with a typical width of 2-4 m, were typically made of brushwood and hurdle, woven wooden panels, or more advanced timber constructions with transverse logs or planks, substructures, and fillings of sand or peat [95,96]. Early Medieval examples such as the eigth-century wooden trackway in the Klempsau bog (northern Germany), which is contemporary to the Fossa Carolina, feature a stratigraphy of up to 2 m composed of new sand and turf fillings and plank layers, which were placed on top of each other [95]. Archaeological investigation of the 8th century Kanhave canal (Denmark) and the Roman Fossa Corbulonis (Netherlands) have discovered (tow)paths made of artificially filled turf and sods that ran parallel to the canal trench [90,97]. Several other Roman canals feature similar (tow)paths made of wood, earth, and field stones [98,99].
In the immediate area of the Fossa Carolina, one would expect significant traces of processing and craft activities as well as of lodging and supply of the workforce. While the internal organisation of the construction site and workforce are unknown, there are different possibilities. On the one hand, it can be assumed that the village of Graben, located on the south side of the central section (Figure 1), sits on the site of a former construction camp and thus covers its traces. Diffusely distributed Early Medieval iron melting remains and ceramic sherds in the near west-east section could support this supposition [91]. Additionally, the SQUID magnetic survey showed several other anomalies in the canal environment [51] that could indicate traces; these have not yet been investigated. On the other hand, it is possible that tool maintenance, accommodation, and provisioning did not take place in close proximity to the construction site but in surrounding Early Medieval settlements, such as Dettenheim and Weißenburg. These places must have had road communication, storage facilities, and shelter, as well as workshops, such as smitheries [91,[100][101][102]. Therefore, it is unsurprising that there are very few archaeological records in this area. Systematic aerial archaeology, fieldwalking, and shovel pit testing in meadows have not yielded noteworthy features or finds such as 8th-century ceramics, iron tools, personal belongings, or any other material remains [37,91,[103][104][105]. Furthermore, direct push sensing gives another explanation. The dated Carolingian surface and thus potential findings are covered by an approximately 1 metre-thick layer (Unit IV) of alluvial and colluvial sediments (Figure 7). However, remains of half-finished timbers and diverse wood waste at the bottom of the excavated canal trench point to the fact that at least some portion of the work took place in the trench itself. This might contribute to a lack of surface finds in the surrounding areas [30]. Furthermore, the formation of the archaeological record of the construction site remains is rather complex and the poor preservation of organic objects under aerobic near-surface conditions, as well as their limited recognisability due to colluvial coverage, have to be taken into account [106][107][108].
As we are almost exclusively talking about woodworking and earthwork at the Fossa Carolina, the range of possible archaeological remains outside the construction pit is rather limited. Like the Fossa Carolina and in sharp contrast to sites with evidence of stonemasonry, lime burning, or glassmaking [109][110][111][112][113][114], other large-scale earthworks of the Early Middle Ages such as the Offa's Dyke, the Danewerk, and the Kanhave canal have yielded a very limited range of finds and features that may be interpreted as construction site remains e.g., [115][116][117].

Conclusions
A ground-based, non-invasive, magnetic SQUID survey revealed an intensive and linear magnetic anomaly in the northern section of the Fossa Carolina that cannot be clearly associated with the canal to which the anomaly runs parallel. (i) This paper documents the approach that was used and the specific results that support this contention. First, a minimal-invasive and highly spatial resolved direct push cross-section was conducted using dense colour and electrical conductivity logs with spacings up to 0.25 m. The direct push data were combined with depth-corrected cores, dating, and laboratory analysis. This methodology fills the gap between selective coring activities and large-scale non-invasive geophysical prospection techniques in wetlands. (ii) The magnetic mass anomaly was found to lie directly above the Carolingian surface and below the Carolingian bank deposits. There was evidence that the sediments of the structures had been heated to an extremely high temperature, and TL dating provided a compatible Carolingian age of 811 ± 250 AD. Thus, our study identified verifiable traces of construction activities outside the canal trench itself. For the spatially upscaled interpretation, depth modelling of the anomaly from SQUID data indicates its stratigraphic positioning close to the Carolingian surface level in the northern and north-eastern sections of the canal. This underpins its clear association with the context of the Carolingian canal. Furthermore, we present new indications for a sophisticated hydro-engineering concept of the canal. The direct push data suggest a geometry with a width of approximately 5 m and a depth of approximately 1 m for the waterway and gives hints for an ample groundwater supply in the summit zone of the canal course. We have determined the bottom level of the canal at approximately 413.2 m a.s.l. This is almost 1 m deeper than in the excavated section further north and therefore indicates that the summit of the canal was further north than previously recorded in former studies. Finally, this paper describes a multi-method approach that represents a promising advancement of ground-based non-and minimal-invasive prospection techniques for geoarchaeological investigation and monitoring in wetland environments.