3D Geological Modelling of a Complex Buried-Valley Network Delineated from Borehole and AEM Data

Journal of Applied Geophysics 122 (2015) 94–102

Contents lists available at ScienceDirect

Journal of Applied Geophysics

journal homepage: www.elsevier.com/locate/jappgeo

3D geological modelling of a complex buried-valley network delineated from borehole and AEM data

A.-S. Høyer a,⁎, F. Jørgensen a,P.B.E.Sandersena, A. Viezzoli b,I.Møllera a Geological Survey of Denmark and Greenland, GEUS, Højbjerg, Denmark b Aarhus Geophysics Aps, Aarhus C, Denmark article info abstract

Article history: Buried tunnel valleys are common features in formerly glaciated areas, and where present, they are very impor- Received 30 March 2015 tant for the groundwater recharge and ﬂow. Delineation of the structures and modelling of the inﬁll is therefore Received in revised form 22 June 2015 very important in relation to groundwater mapping. Typically, borehole information is too sparse to enable a de- Accepted 2 September 2015 tailed delineation of the structures, whereas densely covering airborne electromagnetic data have proven to be Available online 8 September 2015 very useful for this. In the last decades, the mapping approach has been studied carefully, but the 3D modelling

Keywords: of the valley structures has not been described to the same degree yet. In this study, we create a 3D geological 3D geological modelling model of an area that is characterised by a complex network of buried valleys mapped with a spatially dense air- Buried valleys borne electromagnetic survey. Due to the comprehensive dataset, the modelling requires formulation of an ad- Geological interpretation of AEM data vanced strategy. This contains a number of steps, where the AEM-derived resistivity data are initially Manual voxel modelling interpreted based on the geological background knowledge to identify the buried valleys and build a conceptual Modelling strategy geological model. Secondly, the age relationships between the valleys are established from the valley orientations and their internal cross-cut relationships. Thirdly, the deep erosional surfaces are modelled. Subsequently, the interpreted age relationships are utilised to trim the valley floor surfaces, such that younger valleys cut older. Finally, a voxel model is built and populated with lithofacies and stratigraphical units. The model is constructed as a combined layer-based and voxel model in order to map both the overall structures as well as the lithological variations within the 3D model domain. The final model contains 20 buried valleys that show a complex cross-cut setting that indicate the presence of at least eight valley generations. Most of the valley infills show lithological variations, and the final voxel model thus contains 42 different geological units. © 2015 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/).

1. Introduction groundwater flow paths and may result in potentially scattered recharge areas with greater risk of downward transport of contaminants. Buried and open tunnel valleys are common features in former The buried valleys in Denmark are typically between 50–350 m deep glaciated regions (Kehew et al., 2012; van der Vegt et al., 2012). The bur- and 0.5–4 km wide (Jørgensen and Sandersen, 2006). The individual ied valleys are often a key factor for groundwater recharge and flow and valleys show characteristic shapes, such as undulating thalwegs and in areas with buried valleys, delineation of the valleys and their infill is abrupt terminations. The valleys were formed by erosion of the substra- therefore important in relation to groundwater modelling (Andersen ta by highly pressurised meltwater flow underneath the ice sheet et al., 2013; Sandersen and Jørgensen, 2003). In places where the buried (Jørgensen and Sandersen, 2006). Generally, in areas with impermeable valleys are eroded into an impermeable substratum, the coarser infill of substrata, the subglacial meltwater cannot drain as groundwater and the valleys typically constitute important aquifers that tend to be deep channels were consequently eroded deeply into the underlying seated and well protected. However, buried valleys can also be filled sediments (Sandersen and Jørgensen, 2012). The same process has with fine-grained deposits with low hydraulic conductivity, in which been observed in other places, as for instance, in Poland (Piotrowski case they can act as barriers for the groundwater flow. The buried val- et al., 2009). When tunnel valleys have been eroded and subsequently leys have typically been influenced by repeated erosional incisions filled with coarser sediments, the hydraulic properties of the substrata and depositions, and the resulting valley infills are therefore often were changed and the tunnel valleys eroded by the subglacial meltwa- highly heterogeneous in nature. This heterogeneity affects the ter from one ice advance tend to be re-used by the meltwater of following ice advances (Jørgensen and Sandersen, 2006; Sandersen and Jørgensen, 2012). The result is complex networks of buried valleys fi ⁎ Corresponding author: with very heterogeneous in ll caused by variations in both sedimenta- E-mail address: [email protected] (A.-S. Høyer). tion history and history of erosional events. Especially the formation of

http://dx.doi.org/10.1016/j.jappgeo.2015.09.004 0926-9851/© 2015 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/). A.-S. Høyer et al. / Journal of Applied Geophysics 122 (2015) 94–102 95 cross-cutting valleys adds a high degree of complexity to the entire ar- Quaternary deposits are mainly of glacial origin and composed of clay chitecture of the valley systems. till and fine to coarse-grained meltwater deposits. Meltwater sand and In order to map the buried valleys, dense data coverage is essential. gravel within the buried valley structures constitute important aquifers Typically, the borehole density is relatively small and especially deep used for regional drinking water supply. boreholes are sparse. Often, borehole data are therefore not sufficient by themselves (Jørgensen and Sandersen, 2009a). A number of studies 3. Data demonstrate that the spatially dense resistivity information from airborne electromagnetic (AEM) data can improve the mapping of buried 3.1. SkyTEM survey valleys (Bosch et al., 2009; Jørgensen et al., 2012; Oldenborger et al., 2013) and that buried valleys can be successfully mapped when com- The airborne transient electromagnetic data (Fig. 1)wereacquired bined with stratigraphic and lithological information from boreholes. with the SkyTEM system (Sørensen and Auken, 2004). In this survey, While 2D mapping of buried valleys have been performed and de- the early-time data have been corrected for the primary field response, scribed from several studies (e.g. Gabriel et al., 2003; Jørgensen and using a method comparable to the one in Schamper et al. (2014).This Sandersen, 2009a,b), 3D modelling of the valleys has not yet been re- allowed unbiased data to be obtained from as early as 7 μs after begin- ported in literature to the same degree. The handling and interpretation ning of ramp down (equivalent to 3 μs from end of ramp) for the of the very large data sets from areas with complex buried valley net- super low moment. The last gate was centred at 9 ms. The very early works require formulation of an advanced modelling strategy. Thus, times carry information about thin layers, in the order of a few meters, the cross-cutting nature of the valleys, their abrupt terminations, and in the upper c. 20 m, similar to the performance achieved previously generally heterogeneous infill facies make the valleys difficult to with other inversion techniques (Schamper et al. (2014)). At the same model with a traditional layered modelling approach, and more ad- time, layers in more than 250 m depth can be mapped, although with vanced methods therefore have to be used. both decreasing lateral and vertical resolution and highly dependent In the current study, we operate in an area that is characterised by a on the resistivity structure. The survey covers 333 line km with a line complex network of buried valleys. The geological modelling is based on spacing of 100 m. Due to the dense infrastructure in the area, many of densely spaced AEM data and borehole information available prior to the soundings suffered from couplings to man-made conductors and this study. Here, we present a 3D modelling strategy with which we were removed during the data processing, and prior to the inversion. have produced a detailed 3D geological model where a large number Failure to eliminate these data would have inevitably resulted in arte- of cross-cutting buried valleys are resolved and included as discrete facts in the resistivity models recovered (Viezzoli et al., 2013). We esti- elements. mate that approximately 30% of the acquired data had to be culled. Consequently, about 9500 soundings were used for the inversion. The 2. Study area inversion was based on the spatially constrained (SCI) quasi-3D approach: a local, 1D exact forward response (Auken et al., 2014), with The study area covers 45 km2 and is located in the central part of model parameters spatially constrained in 3D (Viezzoli et al., 2008). Denmark (Fig. 1). This part of Denmark has been covered by numerous The inversion was performed with 25 layers that provided smooth tran- ice sheets from different directions during the Pleistocene (Houmark- sitions in the resulting models. The resistivity models were then further Nielsen, 2004). Within the study area, the terrain surface varies from interpolated into horizontal grids and stacked, resulting in a 3D resistiv- sea level to almost 100 m above sea level. ity grid. Based on earlier mapping projects (e.g. Jørgensen and Sandersen, 2006), the geology in the uppermost 200 m can be divided into two 3.2. Boreholes major geological elements: i) thick deposits of hemipelagic Palaeogene clays and marls that generally occur at shallow depths and ii) deeply Borehole information from 435 boreholes (Fig. 1) was extracted (up to 130 m) incised tunnel valleys arranged in a complex network. from the Danish national borehole database, Jupiter (Møller et al., In most of the area, Paleogene clays and marls directly underlie the 2009). Most of the boreholes are shallow, with 63% of the boreholes Quaternary deposits, but sandy Miocene sediments occur locally. being no more than 35 m deep and with only 3% of the boreholes deeper Most of the buried valleys are incised into the Paleogene deposits, than 100 m. The majority of the boreholes have been drilled for water which accordingly show a considerable variation in surface relief. The abstraction, and the deep boreholes are therefore typically clustered in

Fig. 1. A) Location of the study area and B) location of the data. Black dots in (B) mark positions of SkyTEM soundings remaining after processing. Soundings are so closely spaced that they appear as black lines. Boreholes within the study area are shown with colours according to their depths. 96 A.-S. Høyer et al. / Journal of Applied Geophysics 122 (2015) 94–102 areas around waterworks. The boreholes vary considerably in quality due to e.g. age and drilling method. The impact from these factors is evaluated during the geological modelling process.

4. Mapping and modelling strategy

When modelling in areas with complex geology, we typically operate with three main phases: performing a basic geological interpretation and setting up a general conceptual model, performing the layer modelling, and finally, performing the voxel modelling. When modelling buried valleys, it is necessary to follow a dedicated modelling strategy in order to address the highly complex geology. This strategy encom- passes five main steps: i) constructing a conceptual geological model of the buried valleys, ii) determining the age relationships between the valleys, iii) layer modelling of the valley floors and sides, iv) trimming of the valley floor surfaces in accordance with the age relationships, and v) building a voxel model and populate it with lithofacies and stratigraphic units

4.1. Basic geological interpretation and conceptual model

The resistivity models recovered from the inversion of the SkyTEM data were thoroughly interpreted together with borehole data and the mapping was carried out by following the guidelines described in Jørgensen et al. (2003). Apart from being visualised as a 3D grid, the resistivity data were jointly displayed in vertical sections and horizontal slices (Fig. 2A). The buried tunnel valleys were detected by the significant resistivity contrasts between the infill and the incised sur- roundings. Particularly good results were achieved where the valleys were incised into the low-resistive substratum, because the valleys are filled with deposits characterised by higher resistivities. To identify the shapes (floor and walls) of these valleys, a surface map of the deepest conductive layer was extracted (Fig. 2B). Other valleys that are cut into sandy sediments were possible to identify and delineate if they are filled with low-resistive sediments. Some valleys with sandy fill and without any significant contrast to the substratum were some- what imprecisely mapped and some valleys may remain undetected. The valleys were in general seen as elongate, often irregular, anomalies in horizontal slices through the resistivity data, either as high-resistive bodies in low-resistive environments or as low-resistive bodies in high-resistive environments. A basic, conceptual understanding of the area's geological history was developed and used as a guideline during the subsequent construction of the 3D geological model (Fig. 3). All Fig. 2. Initial interpretations of the SkyTEM resistivity data. A) Horizontal slice through the 3D resistivity volume at the elevation = 5 m above sea level. B) Depth to the top of the the mapped valleys are shown in map view in Fig. 2C. deepest conductive layer the Paleogene clay. C) Map of the interpreted buried valleys. The grey shadings show the extent of the valleys, whereas the lines mark the thalwegs 4.2. Layer modelling for the individual valleys. Valley numbers are written on the figure.

The geological modelling was made in the software, Geoscene3D this level. However, due to later erosional events, the buried valleys (I-GIS, 2014). The SkyTEM resistivity data were displayed both as 1D are not possible to trace in the present-day terrain. A thin sheet of glacial models and as an interpolated 3D resistivity grid. During the basic sediments (less than 2–4 m) seems to exist above most of the buried geological interpretation, the resistivity grid was sliced horizontally at valleys that reach the surface, but this thin sheet does not clearly different levels and a surface map of the deepest conductive layer was occur in the data due to limitations in resolution capability of the TEM studied in order to find the exact extent and position of the valley struc- method and degree of detail in the borehole descriptions (Schamper tures. The modelling was mainly performed along vertical sections, et al., 2014). where lithological information from boreholes were shown together During modelling of the valley floors, the age relationships between with SkyTEM resistivity data within a predefined buffer zone, typically the individual valleys were determined. This was done based on the 100 m (Fig. 4A,B). The profiles were initially drawn perpendicular to structural and compositional characteristics of the infill, cross-cut rela- the valleys and gradually moved 100 m along the valleys. They were tionships, orientation of the valleys, and on the background geological continuously checked against profiles drawn along the thalwegs of the knowledge from the area. The age relationship was imperative during valleys as mapped during the basic interpretation (Fig. 2C). the subsequent trimming (cutting) of the valley surfaces. During the The first step was to model the surface of the valley floors and walls. trimming, the surfaces of older valleys surfaces were cut by valley The appearance of the resistivity structures and the lithological informa- surfaces of younger valleys. This is for instance illustrated in Fig. 4B, tion from the boreholes were used to place the interpretation points. where valley 5 cuts down into valley 2. The final, trimmed valley These points were subsequently gridded using kriging to obtain the surfaces are shown in Fig. 5. floor of each valley. Most of the buried valley structures reach the ter- The final step of the layer modelling was to construct surfaces of the rain surface, and the gridded valley surfaces were accordingly cut at top of the Paleogene clay and the Miocene sand. As for the valley A.-S. Høyer et al. / Journal of Applied Geophysics 122 (2015) 94–102 97

Fig. 3. Profile sketch illustrating the conceptual model of the study area. surfaces, this was done using interpretation points and subsequent (Fig. 4A,B) that the valleys are filled with both sandy (high resistivities) kriging interpolation. In areas with incised buried valleys, the bottoms and clayey deposits (low resistivities). For instance, valley nos. 2, 11, of the valleys also constitute the top of the Paleogene. The Paleogene and 14 (Fig. 4B) are filled with deposits with low to very low resistivities clay shows very low resistivities and is easy to recognise in the models corresponding to clay, whereas nos. 1, 5, and 12 are filled with resistive, obtained from SkyTEM (Fig. 4B). The Miocene is on the other hand sandy deposits. The resistive layers (N50–60 ohmm) are found to be impossible to discern in the resistivity models, because these sandy coarse-grained glaciofluvial deposits, whereas the low-resistive layers sediments do not show resistivity contrasts against the Quaternary correspond to glacial clay, either clay till or glaciolacustrine clay. Most sandy sediments. Modelling of the Miocene unit is therefore solely valleys are filled with varying sediments both vertically (e.g. no 1 in based on borehole information. Fig. 4B) and laterally. The very low resistivities (b12 ohmm) outside the valleys correspond to Palaeogene clay. 4.3. Voxel modelling A total of 20 buried valleys were integrated in the layer model. The dominant orientation, depth, width, and interpreted generation of the The voxel model covers the entire model area from 180 m below sea valleys are presented in Table 1. When shown in 3D (Fig. 5), it is seen level to the ground level at 0–100 m above sea level. The voxel grid how the valley surfaces cut into one another resulting in a complex discretisation is 50 m laterally and 5 m vertically. These dimensions buried-valley network. Most of the valleys show a preferred orientation were chosen to achieve a proper resolution of the mapped valley from SW-NE (Table 1, Fig. 2C). The depths of the valleys range from a structures. few tens of metres to 170 m, and the width generally falls within the The property units of the voxel model were selected on the basis of range of 400–1100 m (Table 1). The lengths of the valleys are difficult the basic geological interpretation and on the encountered lithofacies to estimate, since most of the valleys are expected to continue outside in the boreholes. The SkyTEM resistivity information was interpreted the modelled area. The shapes of the valleys are clearly irregular with into lithologies by considering the typical formation resistivities, know- hollows and thresholds along their thalwegs (Fig. 2B and 5). As noted ing that the natrium chloride content of the groundwater in the area is above, most valley surfaces have been trimmed at ground level indicat- low (Søndergaard et al., 2004). The Pre-Quaternary sediments were ing that this constitutes a major erosional boundary. grouped into Paleogene clay and Miocene sand. The Quaternary was According to the analysis of the age relationships, the buried valleys grouped into deposits occurring outside the valleys (sand till, clay till, fall within at least eight different generations (Fig. 6). Different valley and meltwater sand) and valley infill deposits (meltwater sand, generations may have been formed during the same glaciation by differ- meltwater clay, and clay till). The units were thus defined both on ent ice advances (Jørgensen and Sandersen, 2006), but since we see a basis of lithology and valley stratigraphy, which resulted in a total of number of generations with very different orientations it is likely that 42 geological units. they were formed during more than just one glaciation. This is in The voxel modelling was performed using a manual approach, accordance with (Jørgensen and Sandersen, 2006, 2009b), who argue where the geological units were interpreted without any automated that the valleys in the area are of Weichselian to Elsterian age, perhaps routines. This was done by using a set of specific tools in GeoScene3D, even pre-Elsterian. The number of valleys belonging to each generation by which it is possible to select and populate the voxels in different varies. Some of the valleys (e.g. nos. 3 and 8, Fig. 2Cand6)showabrupt ways (Jørgensen et al., 2013). In the current work, we have primarily bends of up to 90 degrees over short distances, possibly because the val- used a tool that enables the selection of voxels within volumes delineat- leys tend to form where the substratum is most easily eroded (Janszen ed by surfaces, and a tool where voxels were selected within digitised et al., 2012; Sandersen and Jørgensen, 2012), which in this case is the polygons on cross-sections. Using these tools, major volumetric bodies older coarse-grained valley infill instead of the less erodible Paleogene were populated by picking groups of voxels that were constrained by clay deposits. the initially modelled surfaces, whereas minor bodies were manually The 42 different geological units in the 3D voxel model can be digitised. thematised with regard to either valley stratigraphy (Fig. 4C, 7)or lithofacies (Fig. 4D, 8). An overview of the voxel model results 5. Results is shown in Figs. 7 and 8 as a fence diagram with a number of vertical N-S sections through the final voxel model. The base of the model is The basic interpretation of the data resulted in a conceptual model constituted by Paleogene clay in the entire model area. Above this, a for the area with several buried tunnel valleys cutting down into the thin layer of Miocene sand is present in the central part. This is followed Palaeogene clay and into one another, giving a “cut-and-fill” architec- by the heterogeneous Quaternary which covers the entire area. The gla- ture in vertical section view (Fig. 3). Many of the valleys are deep, but cial deposits are grouped into valley fill and deposits that occur outside also shallow valleys occur. The resistivity distribution and borehole in- the buried valleys (Fig. 8). Outside the valleys, the deposits are mainly formation indicate in both map view (Fig. 2A) and section view composed of clay till, but with frequent occurrences of meltwater 98 A.-S. Høyer et al. / Journal of Applied Geophysics 122 (2015) 94–102

Fig. 4. Profile along one of the deep buried valleys (valley 1, see Fig. 2C). The figure shows A) SkyTEM soundings along the profile (buffer 100 m), b) interpolated 3D SkyTEM resistivity grid with modelled valley surfaces on top, C) 3D geological model, and D) simplified, lithofacies model. The boreholes within 100 m distance of the profile are shown on a–d. The colour-coding for these is the same as the colour-coding for the lithofacies model in (D). Black bars next to the boreholes show the screens. Vertical exaggeration = 5.4×.

sand inclusions. Each valley is typically filled with a predominant lithol- complex, provide lots of information, and are difficult and very time ogy with minor occurrences of other lithofacies (Fig. 8). Slightly more consuming to understand and interpret. Proper geological modelling than half of the valleys contain meltwater sand as the main lithofacies, of the buried-valley networks using large-scale AEM data therefore re- whereas the rest primarily contain clay till. Only one valley (no. 2) has quires a clear strategy with a well-defined workflow. Buried valleys meltwater clay as the predominant lithofacies. are difficult to model with a traditional layer-based approach due to a number of factors. Firstly, the cross-cutting nature of the valleys 6. Discussion makes them complicated to interpret and model with respect to the age relationships of the valley generations. Secondly, the abrupt termi- Mapping and modelling complex geological environments like nations and sudden sharp bends of the valley structures make them dif- buried-valley networks require dense data coverage, which can be ob- ficult to follow and map. Finally, as most valleys are filled with rapidly tained through airborne electromagnetic data (Jørgensen and shifting material, several internal layers within the valley fill would be Sandersen, 2009a; Sapia et al., 2014). Such data are typically very required to model the actual geological setting. To be able to include A.-S. Høyer et al. / Journal of Applied Geophysics 122 (2015) 94–102 99

Fig. 5. 3D view of the modelled valley surfaces in the study area, seen from a western direction. Three slices through the model are shown. The different colours represent the different valley generations (for legend see Fig. 7). Vertical exaggeration of the 3D view = 3×.Vertical exaggeration of the proﬁles = 5.4×. these strong internal variations the manual voxel modelling methodol- with the objective to create a faster routine and avoid the subjective ogy proved to be advantageous for modelling the valley ﬁll. element inherent in the interpretations (e.g. Foged et al. (2014) and Recently, research has focused on developing automatic or semi- Gunnink and Siemon (2015)). Høyer et al. (2015) made a comparison automatic routines for hydrostratigraphical modelling of AEM data of the cognitive, manual approach and two different automatic modelling methods. Here, the manual model was the only one that could provide stratigraphical information, but all the model approaches re- Table 1 sulted in highly comparable results if translated into binary sand-clay Valley interpretation data. The table informs about the dominant orientation, the maxi- models. It must be noted though that in Høyer et al. (2015),the mum depth and width, and the interpreted generation of the valleys. geological setting was relatively homogeneous compared to the buried Valley Dominant Max depth Max width Valley number orientation (m) (m) generation

1 SW-NE 140 1000 2 2 SE-NW 120 700 3 3 SW-NE 100 700 5 4 SW-NE 105 700 4 5 SE-NW 50 700 5 6 W-E 100 600 7 7 S-N 60 500 6 8 SW-NE 90 700 7 9 SW-NE 40 400 8 10 SW-NE 140 700 4 11 S-N 40 400 7 12 SW-NE 30 800 6 13 SW-NE 170 1100 6 14 W-E 40 400 8 15 SW-NE 110 600 7 16 SE-NW 90 700 7 17 SW-NE 120 700 8 18 S-N 50 400 6 19 S-N 95 600 6 Fig. 6. Relative age of different valley generations. Oldest valleys are shown with darkest 20 SE-NW 60 500 1 colours. The valleys have been divided into 8 different generations. 100 A.-S. Høyer et al. / Journal of Applied Geophysics 122 (2015) 94–102

Fig. 7. Vertical N-S slices through the final voxel model. The infill of each valley is shown with its own specific colour. Vertical exaggeration = 3×.

valley-network in the current study, where the stratigraphical informa- why manual interpretation is preferred when the product is a geological tion is vital for the model output. Since the stratigraphical information is model with stratigraphic content. unobtainable using automatic methods, we therefore expect more sig- In this study, the deepest valley structures were easily delineated by nificant differences when operating in this type of environment. The investigating the depth to the deepest conductive layer, which corre- geological knowledge required for making the geological model in- sponds to the boundary between the Paleogene clay and the Quaternary cludes subjects like regional geology and the glacial history of the deposits. Modelling of the deep valley structures is therefore associated area, without which it would be impossible to infer the different valley with only little uncertainty and would also be outlined using a proper generations. Another important input is knowledge regarding glacial automatic modelling method. The shallow valleys, on the other hand, processes and the formation of the buried valleys, which is essential were considerably more difficult to interpret and model and could for delineating realistic shapes and extents of the structures. This is es- only be delineated by using the geological knowledge of the area. The pecially crucial when modelling the shallow valleys that are incised lower resistivity contrasts between the shallow valleys and their sur- into other glacial deposits and therefore only show low resistivity con- roundings will induce model uncertainties. In this study, we have trasts. Also the geophysical knowledge regarding resolution capabilities been able to map and model several valleys, but even more valleys is vital when interpreting the AEM information. Generally, the vertical may be present in the area. First of all, buried valleys that are too and the horizontal resolution diminish drastically with depth, such small to be resolved in the data may be found, and secondly, buried val- that deep geological structures have to be considerably larger in order ley structures can be unresolved in both resistivity data and borehole to be resolved than structures located at shallower depths. Resistivity data, if the sediments lack resistivity and lithology contrasts. Modelling variations within resistive deposits such as meltwater sand or gravel these valleys require other data types such as reflection seismic data, by aremoredifficult to identify using AEM compared to variations in con- which it is possible to resolve the erosional boundaries of the valleys ductive deposits, due to the inductive nature of the methodology (Jørgensen et al., 2003). (Fitterman and Stewart, 1986, Schamper et al., 2014). Another impor- The lithological information delivered by the geological model is tant aspect is the appearance of sloping structures in the geophysical crucial, when the model is going to be used for groundwater modelling. sections. These tend to be underestimated in the inverted resistivity sec- While some of the sand-filled valleys constitute important aquifers, the tions if the data are inverted using a 1D inversion approach. This effect is clay-filled valleys might act as barriers for groundwater flow. An exam- most pronounced for the steepest slopes, which also can cause 2D/3D ple of this is seen in the western part of the area, where a clay-filled bur- effects (Danielsen et al., 2003). Most of these considerations are impos- ied valley (no. 11) cuts through a valley filled with sandy deposits sible to take into account in automatic modelling approaches, which is (valley no. 1, Fig. 4). The narrowing of the aquifer is expected to have A.-S. Høyer et al. / Journal of Applied Geophysics 122 (2015) 94–102 101

Fig. 8. Vertical N-S slices through the final lithofacies voxel model. Vertical exaggeration = 3×. asignificant influence on the resulting transmissivity, such that the contains 42 different geological units. A high number of units have water flow is diminished across the cross-cutting valley (no. 11). Also, been chosen, because the Quaternary deposits are grouped into deposits the stratigraphic information can be vital with regard to groundwater occurring outside and inside the individual valleys, and most of the val- modelling, and this information is only obtainable using the manual leys comprise more than one lithology. This grouping was made, since modelling approach. A certain sediment type deposited under different the hydraulic conductivities of the different units are influenced by conditions will typically not show the same sorting degree or grain size their depositional characteristics. The model contains 20 buried valleys distribution. Consequently, the hydraulic conductivity of a deposit like that show a complex cross-cut relationship, which suggests at least meltwater sand is expected to differ, when deposited under different eight different valley generations. Some of the valleys constitute impor- valley infill events. Furthermore, the depositional history in the individ- tant aquifers, but about half of the valleys contain clayey deposits that in ual valleys can hold information regarding anisotropy of the valley infill, some cases may act as barriers for the groundwater flow. Consequently, and thereby anisotropy for the resulting hydraulic conductivities. The the modelling approach developed in this study resulted in a highly de- detailed information contained in the geological model is consequently tailed model of both lithologies and stratigraphy. Apart from being ap- highly valuable, when transferring the model into a hydrogeological plicable in buried valley settings, the same methodology may be model for further groundwater modelling. applicable in other complex geological environments, where the conceptual geological understanding is crucial for the model output. 7. Conclusion Acknowledgements In this study, we have formulated and followed a strategy for 3D geological modelling of a geological environment characterised by a This study was supported by HyGEM, Integrating geophysics, network of buried valleys. The approach is based on 5 steps: i) The geology, and hydrology for improved groundwater and environmental AEM-derived resistivity models are examined to identify the buried val- management, Project no. 11-116763. The funding for HyGEM is leys in the area, and the resistivity data and the geological background provided by The Danish Council for Strategic Research. We are grateful knowledge are utilised to build a conceptual geological model. ii) The that we have had the opportunity to use the SkyTEM data, which internal age relationship between the valleys is determined from their were collected by SkyTEM aps. Thanks are also due to Bjarke Roth, direction and cross-cut relationships. iii) The valley floors and sides who conducted the processing and inversion of the SkyTEM data. Also, are modelled using a layer modelling approach. iv) The internal age re- Anne Mette Nielsen is thanked for contributing to the geological model- lationship is used to trim the valley floor surfaces. v) Finally, a voxel ling. Finally, we would like to thank the two anonymous reviewers and model is built and populated with lithofacies and stratigraphical units. Holger Kessler, who contributed with helpful comments and sugges- The result is a 3D geological voxel model that, for this case study, tions that improved the paper. 102 A.-S. Høyer et al. / Journal of Applied Geophysics 122 (2015) 94–102

References Jørgensen, F., Scheer, W., Thomsen, S., Sonnenborg, T.O., Hinsby, K., Wiederhold, H., Schamper, C., Burschil, T., Roth, B., Kirsch, R., Auken, E., 2012. Transboundary geo- Andersen, T.S., Poulsen, S.E., Christensen, S., Jørgensen, F., 2013. A synthetic study of physical mapping of geologial elements and salinity distribution critical for the as- geophysics-based modelling of groundwater flow in catchments with a buried valley. sessment of future sea water intrusion in response to sea level rise. Hydrol. Earth – Hydrogeol. J. 21 (2), 491–503. Syst. Sci. 16 (7), 1845 1862. Auken, E., Christiansen, A.V., Kirkegaard, C., Fiandaca, G., Schamper, C., Behroozmand, A.A., Jørgensen, F., Møller, R.R., Nebel, L., Jensen, N.-P., Christiansen, A.V., Sandersen, P.B.E., Binley, A., Nielsen, E., Effersø, F., Christensen, N.B., Sørensen, K., Foged, N., Vignoli, G., 2013. A method for cognitive 3D geological voxel modelling of AEM data. Bull. Eng. – 2014. An overview of a highly versatile forward and stable inverse algorithm for Geol. Environ. 72 (3), 421 432. airborne, ground-based and borehole electromagnetic and electric data. Explor. Kehew, A.E., Piotrowski, J.A., Jørgensen, F., 2012. Tunnel valleys concepts and controver- – – Geophys. 1–13. sies a review. Earth-Sci. Rev. 113, 33 58. Bosch, J.H.A., Bakker, M.A.J., Gunnink, J.L., Paap, B.F., 2009. Airborne electromagnetic Møller, I., Søndergaard, V.H., Jørgensen, F., Auken, E., Christiansen, A.V., 2009. Integrated measurements as basis for a 3D geological model of an Elsterian incision. Z. Dtsch. management and utilisation of hydrogeophysical data on a national scale. Near Surf – Ges. Geowiss. 160 (3), 249–258. Geophys. 7, 647 659. Danielsen, J.E., Auken, E., Jørgensen, F., Søndergaard, V., Sørensen, K.I., 2003. The applica- Oldenborger, G.A., Pugin, A.J.-M., Pullan, S.E., 2013. Airborne time-domain electromag- fl tion of the transient electromagnetic method in hydrogeophysical surveys. J. Appl. netics, electrical resistivity and seismic re ection for regional three-dimensional Geophys. 53 (4), 181–198. mapping and characterization of the Spiritwood Valley Aquifer, Manitoba, Canada. – Fitterman, D.V., Stewart, M.T., 1986. Transient electromagnetic sounding for groundwater. Near Surf Geophys. 11 (1), 63 74. Geophysics 51, 995–1005. Piotrowski, J.A., Hermanowski, P., Piechota, A.M., 2009. Meltwater discharge through the Foged, N., Marker, P.A., Christansen, A.V., Bauer-Gottwein, P., Jørgensen, F., Høyer, A.-S., subglacial bed and its land-forming consequences from numerical experiments in the – Auken, E., 2014. Large-scale 3-D modeling by integration of resistivity models and Polish lowland during the last glaciation. Earth Surf. Proc. Land 34, 481 492. borehole data through inversion. Hydrol. Earth Syst. Sci. 18, 4349–4362. Sandersen, P.B.E., Jørgensen, F., 2003. Buried Quaternary valleys in western Denmark-oc- Gabriel, G., Kirsch, R., Siemon, B., Wiederhold, H., 2003. Geophysical investigation of currence and inferred implications for groundwater resources and vulnerability. – buried Pleistocene subglacial valleys in Northern Germany. J. Appl. Geophys. 53 (4), J. Appl. Geophys. 53 (4), 229 248. 159–180. Sandersen, P.B.E., Jørgensen, F., 2012. Substratum control on tunnel-valley formation in Gunnink, J.L., Siemon, B., 2015. Applying airborne electromagnetics in 3D stochastic Denmark. In: Huuse, M., Redfern, J., Le Heron, D.P., Dixon, R.J., Moscariello, A., Craig, geohydrological modelling for determining groundwater protection. Near Surf J. (Eds.), Glaciogenic Reservoirs and Hydrocarbon Systems. Geol. Soc. Lond., Spec. – Geophys. 13, 45–60. Publ. 368, pp. 145 157. Houmark-Nielsen, M., 2004. The pleistocene of Denmark: a review of stratigraphy and Sapia, V., Viezzoli, A., Jørgensen, F., Oldenborger, G.A., Marchetti, M., 2014. The impact on glaciation history. In: Ehlers, J., Gibbard, P.L. (Eds.), Quaternary glaciations - extent geological and hydrogeological mapping results of moving from ground to airborne – and chronology. Elsevier, Amsterdam, pp. 35–46. TEM. J. Environ. Eng. Geophys. 19 (1), 53 66. Høyer, A.-S., Jørgensen, F., Foged, N., He, X., Christiansen, A.V., 2015. Three-dimensional Schamper, C., Jørgensen, F., Auken, E., Effersø, F., 2014. Assessment of near-surface map- – geological modelling of AEM data – a comparison of three methods. J. Appl. Geophys. ping capabilities by airborne transient electromagnetic data an extensive compari- – 115, 65–78. son to conventional borehole data. Geophysics 79 (4), B187 B199. I-GIS, 2014. www.i-gis.dk, Geoscene3D (Aarhus, Denmark). Søndergaard, V., Thomsen, R., Dyrsø, O., Nyholm, T., Fuglsang, E., Thorling, L., Misser, P.V., Janszen, A., Spaak, M., Moscariello, A., 2012. Effects of the substratum on the formation of Hansen, B., 2004. Redegørelse for grundvandsressourcerne i ÅrhusNord-området. – glacial tunnel valleys: an example from the Middle Pleistocene of the southern North Vandforsyning Delrapport I, Århus Amt, report in Danish. Sea Basin. Boreas 41, 629–643. Sørensen, K.I., Auken, E., 2004. SkyTEM - a new high-resolution helicopter transient – Jørgensen, F., Sandersen, P.B.E., 2006. Buried and open tunnel valleys in Denmark-erosion electromagnetic system. Explor Geophys. 35 (3), 191 199. beneath multiple ice sheets. Quat. Sci. Rev. 25 (11–12), 1339–1363. van der Vegt, P., Janszen, A., Moscariello, A., 2012. Tunnel valleys: current knowledge and Jørgensen, F., Sandersen, P.B.E., 2009a. Buried valley mapping in Denmark: evaluating future perspectives. In: Huuse, M., Redfern, J., Le Heron, D.P., Dixon, R.J., Moscariello, mapping method constraints and the importance of data density (Kartierung A., Craig, J. (Eds.), Glaciogenic Reservoirs and Hydrocarbon Systems. Geol. Soc. Lond., eiszeitlicher rinnen in Dänemark: bewertung methodischer grenzen und bedeutung Spec. Publ. 368. der datendichte). Z. Dtsch. Ges. Geowiss. 160 (3), 211–223. Viezzoli, A., Christiansen, A.V., Auken, E., Sørensen, K., 2008. Quasi-3D modeling of air- – Jørgensen, F., Sandersen, P., 2009b. Kortlægning af begravede dale i Danmark. Opdatering borne TEM data by spatially constrained inversion. Geophysics 73 (3), F105 F113. 2007-2009. GEUS Special issue 978-87-7871-259-2 375 p. In Danish. Viezzoli, A., Jørgensen, F., Sørensen, C., 2013. Flawed processing of airborne EM data af- – Jørgensen, F., Lykke-Andersen, H., Sandersen, P.B.E., Auken, E., Nørmark, E., 2003. Geo- fecting hydrogeological interpretation. Groundwater 51, 191 202. physical investigations of buried Quaternary valleys in Denmark: an integrated appli- cation of transient electromagnetic soundings, reflection seismic surveys and exploratory drillings. J. Appl. Geophys. 53 (4), 215–228.