Improved imaging of ground deformation and brine seepage around 2 abandoned flooded salt mines by joint inversion of multiphysics data

Abstract


INTRODUCTION 1.1 Problem Definition
Abandoned mine workings often raise serious issues in relation to planning, development and environmental protection.Subsidence due to large scale salt mining and brine extraction is well documented in some areas of the UK and other parts of the world (Cooper, 2002;Bell et al., 2005;Kulessa et al., 2004;Carrier et al., 2021).The salt-mining region of Carrickfergus in County Antrim lies 15 km north-east of Belfast in Northern Ireland (see inset map in Figure 1).
The abandoned salt mines in this region are thought to pose a significant risk to public safety.In 1990 a programme of monitoring was initiated after the unexpected collapse of the Tennant mine in the area.The programme focused on six abandoned salt mines, of which Maiden Mount and French Park mines (Figure 1) were considered to be of significant concern (Atkins, 2005).Both mines have caused considerable disruption to surface activities (e.g., road closure) and are thought to be central to the development of brine seepages in the early 2000s, degrading arable land near to the French Park mine as well as threatening encroaching infrastructure as the town of Carrickfergus is expanding uphill from Belfast Lough (Kulessa et al., 2004).The first surface brine seepage appeared at the south-west end of French Park in 2001 (s1 in Figure 2) shortly before the sinkhole collapse on 19 August 2001 in the Maiden Mount mine (CH in Figure 2).
Both mines are currently monitored by a number of techniques including changes in the water level in a number of boreholes and measurement of ground movement in the immediate area (e.g., Cigna et al., 2017;Cooper, 2020).However, these techniques give little prior warning of ground collapse or subsidence and do not provide extensive information with regard to the subsurface structure, hydrogeology or presence and seepage of contaminants.
A major factor leading to instability of these mine workings was considered to be seepage of freshwater into abandoned mines via mine shafts and the engineered brining operations, resulting in the erosion of the mine pillars and roof strata and leading to a breakdown of the overlying strata and surface subsidence (Atkins, 2005;Arup and Partners 1992).How to assure that this proposed mechanism of surface deformation is correct and how to remotely predict zones in the near-surface that are more likely to experience subsidence and/or seepage-related hazards still remain difficult propositions.As the Maiden Mount site is located uphill of French Park, it has been suggested that the hydraulic processes occurring at Maiden Mount may influence those at French Park.Moreover, there were significant mine spoil heaps on the surface and infiltrating water could generate saline leachate that can contaminate the near-surface, the extent of which is still unknown.The prospect of increased future rainfall and large flooding events in the northern hemisphere will exacerbate the problem even further.The Woodburn aqueduct ran across our study area (WA in Figure 1) as a conduit below surface in pre-1935 ordnance survey maps (Sheet 52, 1930) as part of the main public water supply infrastructure for Belfast city.The location of the newer seep observed in our region in 2004 (s2 in Figure 1) is geographically coincident with the trace of this aqueduct, but their possible connection is unknown.

Figure 1. Location and main features of our study area in Carrickfergus region of Northern
Ireland.Shown is the ordnance survey map (adapted from Ordnance Survey Map of Northern Ireland, Sheet 52, 1930, 1:10560, reproduced with Crown copyright permission 2019) and the locations of Maiden Mount, French Park and Duncrue mines relative to our area of geophysical investigation (red polygon).Orange polygon shows the area classified as "unstable mine" in recent maps (e.g., Donald, 2015).Blue-rimmed inset map shows the location of our study area in Carrickfergus, County Antrim in Northern Ireland (NI); RI, Republic of Ireland; WA, Woodburn aqueduct.
While geophysical techniques are central tools for the non-invasive investigation of such abandoned flooded mine areas, individual methods on their own provide non-unique models of subsurface property and fluid-type present but integrating them together with good structural linkage maximizes accuracy, minimizes uncertainty and leads to a more reliable model, necessary for making cost-effective decisions to mitigate risks at such sites.This is what we set out to demonstrate in this paper, but the special focus is on identifying the attendant surface deformation mechanism necessary for understanding geohazards evolution and their mitigation for such sites.Using legacy multi-physics data acquired in 2004, we explore whether state-ofthe-art joint inversion of seismic and electrical resistivity data, combined with electrical selfpotential (SP) mapping, can provide valuable indications of possible land stability hazards (Kulessa et al., 2007;Thompson et al., 2012Thompson et al., , 2017) ) and subsurface brine or leachate pools in the area of study (compared to individual resistivity and velocity models derived from using popular commercially available software and presented as supporting information S1).A quantitative integrated analysis of these multiphysics data will help to test the optimality of the past mapping of hazardous mined area (Kulessa et al., 2004;Donald, 2015).It will also help us to better understand the deformation mechanism in the near-surface in response to underground space availability created by the past mining and brining operations as well as explain any related processes such as the possible hydraulic link between brine seepages in the south and subsequent sinkhole collapse in the north of our study area (features CH, s1 and s2 in Figure 2).

Land Use Pattern at Study Site
The study site is underlain by Triassic rocks of the Mercia Mudstone Group (Griffith et al., 1983).It contains red marl with gypsum (Keuper Marl) in the upper 160-170 m above the halite deposits that were the target for Victorian brine-pumping extraction and are still mined to the north-east of the study area.Superficial deposits comprise of 5-25 metres of glacial till with heterogeneous infill in places and maximum till thicknesses decreasing from about 20 m at Maiden Mount to about 15 m at French Park (Griffith et al., 1983;Nicholson, 2014).The extraction of salt in the Carrickfergus region (Figure 1) began in the latter half of the 19 th century when salt of the Mercia Mudstone group was discovered during coal exploration (Cooper, 2002).
The French Park mine was sunk in 1870, after the neighbouring Duncrue mine was declared unsafe.Maiden Mount became operational around seven years later and was at the time the deepest mine in the UK.The dominant mining process in the area was the pillar and stall method, horizontal caverns supported by regularly spaced salt pillars (Cooper, 2002).In later years, flooding and uncontrolled brine extraction occurred in many of the mines for extended periods of time.This is likely to have dissolved some, if not all, of the supporting pillars, leaving the mines less stable and prone to collapse, increasing the likelihood of subsidence (Bell et al., 2005).Salt was extracted from the French Park Mine until around 1938 when the wall separating the mine from the adjacent Duncrue Mine collapsed, flooding French Park.Brining was initiated at this time and continued until the late 1950s.
A survey carried out in 1992 discovered that as a result of extensive brining, the mine cavity has over time, migrated towards the surface and all of the supporting pillars are thought to have collapsed (Arup and Partners, 1992).Due to this discovery, a nearby road was closed where it over-ran the southern end of the mine (see Figure 1) and has since been re-routed, showing the effect of such historic mining has on infrastructure.Testing of borehole water samples showed that the salt concentration of the water in the mine was approaching saturation and it was suggested that this could halt further dissolution (Arup and Partners, 1992); however, the prevention of further dissolution depends primarily on the input of fresh water to the system, and a 2005 report concluded that mine collapse within the next 15-20 years is likely (Atkins, 2005).
Maiden Mount mine is located around 500 metres to the north-west and upstream of French Park.The North and South shafts sunk in 1877 at the Maiden Mount mine were 900' (274 m) and 913' (278 m) deep respectively.Maiden Mount mine was taken over by Imperial Chemical Industries (ICI) in the 1950s and it was decided to change to brining (1953)(1954)(1955)(1956)(1957)(1958) with the original shaft in-filled; two 900' deep boreholes were sunk further north for this, one for flooding the mine caverns and the other for extracting brine.Maiden Mount and French Park mines eventually became connected due to brining operations at depth in later years (Cigna et al., 2017) and hence raising the possibility of major voids or galleries being present in the subsurface.On 19 August 2001, the Maiden Mount mine void space reached the surface and a water-filled cylindrical chimney (or crown hole) initially ~50 m wide and ~15 m deep opened up (Figure 2) above a ~100 m wide, ~30-40 m deep void, some 100 m above the main salt-bearing horizon of Triassic mudstones (Kulessa et al., 2004).The collapsed area was fenced off and regularly monitored while the possibility of in-filling the void and returning the land to agricultural use was being discussed (Atkins, 2005).Prior to any such remediation action, the stability of the land surrounding the crown collapse must be ascertained.Several geophysical measurements were made to support this quest (including the unpublished Multiphysics data adopted in the current study) and the collapsed area was subsequently backfilled (see Donald, 2015;Cigna et al., 2017, Cooper, 2020) but resulted in a much larger NW-SE trending surface depression (Figure 2d) that altered the local topography (see Cigna et al., 2017, Figure 8) suggesting that the underlying process remains poorly understood.here with two ground radar operators for scale (Kulessa et al., 2004).s2 ("new seep") was observed in 2004.(d) Google Earth satellite image emphasising the reactivation of the backfilled area of crown collapse taken on 19 August 2016 (Cigna et al., 2017).
Maiden Mount and French Park mines are also thought to play a pivotal role in the development of a number of brine seepages in the area (Figure 2), threatening its present-day agricultural value as well as encroaching infrastructure as the town of Carrickfergus is expanding uphill from Belfast Lough.The first seepage appeared at the south-west end of French Park in 2001, shortly before the Maiden Mount collapse.As the Maiden Mount site is located uphill of French Park, it has been suggested that the hydraulic processes occurring at Maiden Mount may influence those at French Park.Here, seepage through the former Marquis of Downshire mineshaft (MDM in Figure 3) and along associated historic excavation surfaces was previously hypothesized to facilitate brine flow from depth to the ground surface based on GPR imaging (Kulessa et al., 2004).A simple pipe and hardcore drainage system was implemented with a herringbone design to capture water seepages over a large area (Atkins, 2005)

Collocated Multiphysics Measurements
Electrical resistivity and seismic refraction depth sounding data were acquired along ten copositioned survey profiles in 2004 to characterise the subsurface structure and detect and delineate any brine seepage from the upstream mine area of Maiden Mount towards French Park mine downstream (Figure 1).The survey profiles were positioned to cover the agricultural area of interest between the two mines as comprehensively as possible within site access constraints, and each electrode or geophone position was measured in 3-D space using differential GPS.The electrical resistivity data were acquired using an IRIS Sycal Pro imaging system (Kulessa et al., 2007;Ruffell and Kulessa, 2009) with 36 stainless steel electrodes spaced at 5 m intervals along the survey profiles, so that each resistivity profile was 175 m long (Figure 3).The Wenner-α electrode configuration was used because it provides the best compromise between vertical and horizontal resolution and signal to noise ratio (Sharma, 1997).For each electrode quadrupole between four and eight repeat measurements were acquired and stacked, with a desired maximum standard deviation of 3% between them.Seismic refraction data were acquired with a Geometrics Geode based system with 24 40-Hz geophones, a 12-lb sledgehammer and an aluminium plate (Kulessa et al., 2007;Hiemstra et al., 2011).The geophones were placed at the same positions as the stainless-steel electrodes used for our resistivity surveys, between profile distances of 15 m and 130 m, with respective forward and reverse off-end shots every five metres to profile distances of -30 m and 175 m.Data acquisition during periods of low surface wind speed, with a minimum of three stacks per shot location, ensured that high-quality seismic refraction data were acquired.Electrical self-potential (SP) data were subsequently acquired along eight survey profiles to identify whether any subsurface brine is stationary at present, or actively flowing down the presumed hydraulic gradient.The SP data were acquired manually with PMS 9000 Pb-PbCl2 lead-lead-chloride non-polarising electrodes (Petiau, 2000) connected by single-conductor wire to a METRA HIT 22S multimeter, having 10 MΩ input impedance (Doherty et al., 2010;Thompson et al., 2012).At each stainless-steel electrode position used in the resistivity surveys along lines 1-7 and 10, surface vegetation was removed (and then replaced) to ensure favourable ground contact.This facilitated the acquisition of high-quality SP data between a roving PMS 9000 electrode placed sequentially in each cleared location, and a fixed PMS 9000 reference electrode effectively buried in topsoil upstream of the survey area (Figure 3).Standard closed-loop techniques of drift correction were used (Naudet et al., 2004;Doherty et al., 2010).The resulting processed SP data are presented in Figure 3, showing a change in potentials possibly associated with seepage from uphill (north) to downhill (south) direction (Bolève et al., 2009;Thompson et al., 2012)

Joint Inversion for Common Structure to Maximize Accuracy and Reduce Uncertainty
The cross-gradients joint inversion method originally developed for recovering structurally consistent models from electrical resistivity and seismic refraction data (Gallardo andMeju, 2003, 2004) will be adopted here.The structure-coupled inversion reconstructs those structurally similar images that match the measured electrical resistivity and seismic travel-time data as closely as possible, allowing for the fact that these data contain random noise whose distribution is Gaussian with zero mean and variance.Gallardo andMeju (2003, 2004) showed that joint inversion in this manner led to resistivity and velocity images with remarkable structural agreement, and as such were distinctly superior to those from more conventional separate 2-D inversions of the respective data sets (e.g., Meju et al., 2003), even in heterogeneous near-surface materials.By making the assumption that the different geophysical methods sample the same underlying geological structure (a common frame of reference), structural similarity is quantified and used to guide the joint inversion process.The linkage criterion is that the cross-products of the gradients of the dc resistivity property field and the seismic property field must be zero at a boundary.For a 2D profile oriented in the x direction with z direction pointing vertically downwards, if ms(x,z) denotes the vector field of the gradients of the seismic velocity model and mr(x,z) denotes that of dc resistivity model, then the cross-gradients function (Gallardo and Meju, 2002) is given by t(x,z) = mr(x,z) × ms(x,z) (1) and should tend zero at a significant structural boundary.The geological implication is that if a true boundary or transitional zone exists, it should be sensed by the resistivity and seismic models in the same or opposite direction.While this cross-gradients condition encourages structural conformity of the models, it also has some flexibility.First, the changes are not restricted in amplitude, which means that it allows each model to change according to their respective data requirements.Second, the condition does not force the models into a pre-defined common direction unless it is justified by the observed data; thus, the direction is determined by common agreement between the resistivity and seismic data sets.Finally, there is the possibility of allowing a boundary that only occurs in one of the models (resistivity or velocity) when either mr(x,z) or ms(x,z) vanishes somewhere in the model, which makes sense in cases where we have significant variation in only one of the physical property fields.
For joint dc resistivity and seismic refraction data inversion incorporating the cross-gradient constraint, the inverse problem is defined as (Gallardo and Meju, 2003): (2) Here, t(mr,ms) represents the cross-gradients function, dr and ds represent the resistivity and seismic data respectively, fr(mr) and fs(ms) are respectively the dc resistivity and seismic traveltime data computed by forward theory, Cdd is the diagonal matrix of covariances of the data (which are assumed to be uncorrelated), D is the matrix of the smoothness operator, s and r are weighting factors that define the level of smoothness required in the models, mRr and mRs are the a priori model parameters with covariance CRR (also assumed diagonal).The a priori model parameters and their covariance matrix are conveniently chosen to constrain the solution and reduce its variability for parts of the models not constrained by the data (for example areas not covered by seismic rays) and at the same time to limit the variability of the final models in zones where the certainty of the a priori model is high.Note that reliable a priori information can be incorporated into the objective function via the a priori model parameters and their covariance.
Indeed, follow-on applications and adaptations by others to a variety of geophysical data sets (e.g., Doetsch et al., 2010aDoetsch et al., , 2010b;;Moorkamp et al., 2011Moorkamp et al., , 2013;;Um et al., 2014;Carrier et al., 2021) showed the cross-product given by equation ( 1) to be remarkably robust in producing structurally-similar geophysical images.In the present case we can therefore expect to obtain images of structural anatomy of our study area, and any brine infiltration into it, that would be better than those obtained by separate inversion of our seismic refraction and electrical resistivity data.The interested reader is referred to these earlier publications and reviews by (Gallardo and Meju, 2011) and (Meju and Gallardo, 2016) for exhaustive details of this structure-coupled joint inversion scheme now widely used for quantitative integration of Multiphysics measurements.
We jointly inverted all the seismic and resistivity datasets to yield models that are structurally consistent and fit the observed data best in a statistical sense (normalised RMS error of 1) as in the representative example for line 2 shown in Figure 4.The finite-difference based computations employed a common numerical grid for the joint inversion process but with separate dc and seismic forward calculation grids optimally matched to the sensitivity requirements of each method (Gallardo andMeju, 2003, 2004).The results for all the ten survey lines are consequently presented from east to west following the seismic profiling direction adopted in our common grid joint inversion scheme.

Integrated Electro-Mechanical Systems Appraisal to Improve Decision Making
We appraise the reconstructed physical models along each survey line pixel-by-pixel for resistivity-velocity relationships or diagnostic patterns presented as a cross-plot (Gallardo andMeju, 2003, 2004;Meju et al., 2003) which is popular for reservoir characterization in the environmental and petroleum industries (Meju et al., 2003;Doetsch et al., 2010a;Moorkamp et al., 2013).For post-inversion extraction of robust features of these models, we adopt the geospectral imaging (RGB blending) approach (Gallardo et al., 2012).We blend RGB images of the resistivity and velocity models into the cross-plot for more robust trend deductions.For insights on the possible presence of subsoil contaminant plumes, material flow or cavities, we search for consistent patterns in the SP profiles and these resistivity-velocity images of the subsurface and their correlation with any features visible on the ground surface (brine seepages or mine waste spoils mapped in Figure 1).To avoid over-interpretation of the joint inversion models, we examine the seismic ray-paths and only the features crossed by sufficient ray paths are deemed to be justified by the field data thus enabling us to define the effective zone of investigation (EZI) from joint inversion.
For illustration, the resulting joint inversion models for line 2 are shown in Figure 5.Note the remarkable structural similarity of the main features present in the reconstructed models (Figure 5b and 5c).The geospectral image is also shown in Figure 5d to guide the interpretation of the subsurface.The seismic ray-paths are shown superimposed on the geospectral image for comparison; only those features crossed by sufficient ray paths are deemed justified by the field data and hence the yellow line in this image is our interpreted limit of the zone of effective investigation.We also computed a resistivity-velocity cross plot from the joint inversion models, shown on the righthand-side of the geospectral image in Figure 5d.The cross-plot enables the distinction between unconsolidated sediments and consolidated rocks or local basement as well as the pattern of fluid saturation (e.g., Gallardo and Meju, 2003;Meju et al., 2003) and relative direction of changes in total dissolved solids (TDS) or saltiness of the fluid saturating a given material (see red arrow in this cross-plot).In general, the geographical coincidence of low velocity and high resistivity (LVHR) could signify the presence of unsaturated non-confined loose materials, freshwater or voids; low velocity and low resistivity (LVLR) could signify confined brine-mud pools; and high velocity and moderate resistivity (HVMR) could signify consolidated marl with gypsum as annotated in Figure 5b, 5c and 5d.The collocated SP data (Figure 5a) enable us to further distinguish between competing scenarios in our integrated approach; low SP values would corroborate our inferred positions of unstable ground or material transport.We have rigorously applied the above workflow to get the interpretable results for the ten survey lines (summarised in Figure 6 and supplementary material S1).

Structurally-consistent Subsoil Models
The reconstructed distributions of electrical resistivity and seismic compressional wave velocity (Vp) along two survey lines selected for illustration are shown in Figure 6 6).Interestingly, the subsurface resistivity and velocity pattern in the un-mined northwest segment of our study area is different from that in the mined segments.
Notably on line 10, there is a marked change in the glacial till layer resistivity between the lowresistivity eastern half passing over the spoil heap from the main shaft at Maiden Mount mine and the western half which is of much higher resistivity (Figure 6).This could suggest that there is a significant contribution to this lowering of resistivity from leaching of mine tailings dumped at the ground surface by infiltrating water.salt would cause such high seismic velocities but is unlikely to exist at these shallow depths in the damp climate of Northern Ireland; marl with gypsum was encountered in mine shafts at that depth and the typical Vp values for marl rocks are higher than 2.8 km/s (Mari et al., 1999).
Understanding the full implication of the evinced structural geometry for the attendant subsurface processes will be explored later, noting that brine-mud spillage is visible on the surface on line 1 where the 5 m thick cover of unsaturated deposit is breached by this conductive marker unit (Figure 6).

Integrated Analysis of Inferred Water-Rock Interface Properties
In terms of the attendant hydrodynamic processes, the resistivity-velocity cross-plots and spectral signatures extracted from our joint inversion models can provide useful insights (and are presented for all ten lines in Figures 6 and S1).The geospectral images and crossplots for Lines 1 to 4 are similar and show the unsaturated part of the glacial till to be characterised by high resistivity (ρ) and low compressional wave velocity Vp (<1 km/s) while the saturated till and/or uppermost part of the clay with gypsum layer is marked by low resistivity and increasing Vp (Figures 6 and S1).Beneath them is a zone consisting of consolidated marl with gypsum marked by resistivity and velocity increasing with depth.The diagnostic boundary between the unsaturated and saturated zones is labelled point A (Figure 6).For lines 1 to 4, point A is characterised by Vp of 1.8-2.1 km/s and log ρ of 0.5-0.7 Ωm suggesting the possible presence of brine (a Vp of ~1.5 km/s will be expected for freshwater-saturated sediments (Gallardo and Meju, 2003;Meju et al., 2003).However, for lines 5 to 8 (Figure S1), there is a superposition of two resistivity-velocity trends with significant resistivity reversal at points A (Vp = ~2 km/s; log ρ = 0.65-1.1 Ωm) and B (Vp = 2.7-3.1 km/s; log ρ = 1.2 Ωm) with B being consistent with the typical Vp value of marl rocks (Mari et al., 1999).Incidentally, these are the survey lines that extend from the westerly unmined areas to the easterly mined areas with spoil heaps on the surface lending credence to our inference (discussed later) of potential infiltrating freshwater in the western part and leachate brine in the eastern part.Furthermore, it was noted that the parameters of point A are similar for all the line segments located within the mined areas (see Figures 6 and S1).The lowest resistivity at point A is seen on line 4 while the highest value is found upstream on line 9 (with A parameters of Vp = 2.35 km/s, log ρ = 1.2-1.25 Ωm) and the western part of line 8 (with A parameters of Vp = 1 km/s, log ρ = 1.4 Ωm).This is consistent with our inference of upstream partitioning of infiltrating water-types and could suggest a genetic link between north and south in which the groundwater's composition is evolving following the classic Chebotarev (1955) sequence (discussed in more detail below).It doing so, composition changes from relatively low TDS (synonymous with higher resistivity) in upstream northerly recharge area through Line 4 in the middle part where it is slowly moving or relatively stagnant (with lowest resistivity or highest TDS) to the southerly discharge area near Lines 1 and 2.

Appraisal of Different Results at Known Hazardous Locations
Seismic velocity imaging is known to help resolve structure while resistivity imaging identifies significant brine-filled materials or cavities in the subsurface.It is well-known that a multiphysics approach combining both methods using the cross-gradients algorithm leads to better models than possible from the individual methods (Gallardo andMeju, 2003, 2004).Line 1 crossed the zone of surface brine seepage in the east while Line 10 extended eastwards into the area of known collapse structures.Also, both lines cut across areas of known surface spoil heaps from mining according to historical ordnance survey maps.(e) tomographic inversion of seismic refraction data using SeisImager commercial software.The blue arrow marks the location where brine spilled onto the ground surface on line 1 (see Figure 2c).Black arrows and white stars are positions of features used for assessing the similarity or dissimilarity of the reconstructed images of the subsurface.
In Figure 7, we compare SP profiles, our models from cross-gradient joint inversion and those from separate inversion using other well-established software.In this figure can be seen features that were reliably sensed by electrical resistivity imaging but not satisfactorily by seismic velocity imaging on its own (white stars) and those imaged by seismic data but not reliably sensed by resistivity inversion on its own (black arrows) using the conventional separate inversion software.Notice that both sets of features were completely recovered and refined in the cross-gradients joint inversion resistivity and velocity models including the surface breakthrough or isolated feature at the sites of surface spillage or sinkhole collapse (black star).Further upstream along line 10, the individual resistivity inversion model shows a corrugated low resistivity layer in an otherwise relatively resistive half-space and confined to the eastern half of the transect (Figure 7d).This is captured in the joint seismic-resistivity inversion models but not The SP, seismic velocity and resistivity profiles for lines 5, 6 and 7 which cross the original position of this aqueduct are summarised in Figure 8.It would appear at first glance that the Woodburn aqueduct-related facility (also reported as the Water Commissioner's French Park conduit in old mine surveyor's maps) was a potential contributor of water (leakage or natural recharge?) to the subsoil since there is a somewhat consistent change in the resistivity structure at depth across that locality on lines 5, 6 and 7.This can be deduced along a zone offset ~10m to the southeast (downhill) of the main axis of the aqueduct in Figure 8 and coincident with the eastern boundary of the water conduit facility where ditches were apparent on old maps.
However, notice the corrugated nature of the cover layer on line 7.The cover layer becomes progressively more corrugated eastward on line 10 (see Figure 6).

Inferred Water-Associated Ground Deformation Mechanism and Attendant Processes
The joint inversion sections and coincident colour-coded SP anomalies are presented at their respective geographical positions in Figure 9 to permit an areal assessment of the results.The known extents of underground mining from past historical records and the zone of visible brine spillage are also shown in this figure for reference.The low resistivity zone on lines 1 to 7 extend from the west to the north-south running surface depression at the eastern edge of survey lines 1-7 while that on line 10 extends from the east towards this surface depression (see Figure 2a).The low resistivity zone is shallowest towards the area of known underground mining to the east and also underneath the area shown on the map to contain mine spoil heaps; this observation would suggest rainfall and/or surface water infiltration and leaching of mining waste as the possible sources (yellow and blue arrows in Figure 9) of the laterally-confined conductive anomaly.It is obvious in this figure that the conductive anomaly extends westward beyond the currently accepted limit of unstable mined ground (Donald, 2015).
When considered in the context of the SP anomaly map (Figure 3), the resistivity images could at first glance be taken to suggest the evolution of a low-resistivity brine plume from upstream (line 7 or even line 10) to downstream (Line 1) such that, especially in Lines 1 and 2, there evolves a connection between the plume and the ground surface where surface brine-mud leakage occurred in a farmer's field (Figure 2c).Earlier study (Kulessa et al., 2004) interpreted this connection as a former mineshaft based on GPR data.However, does this simplistic analysis explain all our observations at this site?The resistivity-velocity models show distinct shallowing depth-wise of steep thrust-like structures in the localised area with marked positive SP values on lines 1 to 3 (Figure 3).However, Figures 5 and 7 also show that the SP profile across the area of surface brine-mud seepage has a local negative signature superimposed on the areal positive anomalies.
Some other mechanism may be at work in the subsurface in this area where the mined and unmined areas evidently have contrasting velocity and resistivity characteristics., 2015).Purple lines demarcate the bands of extensional and contractual deformation while the wide arrows show our interpreted trend of geochemical evolution of groundwater (Chebotarev, 1955) from the northerly recharge area (orangefreshwater; blueincludes leachate from spoil heap) to the southerly discharge zones s1 and s2.
Based on the similar subsurface structure derived from data inversion, we opine that the resistivity and velocity models evoke a picture consistent with concealed gravitational gliding deformation or land sliding (Figure 10).Water ingress (from rainfall, surface ponds and widespread artificial injections; Figure 9 Our new inference of the water-rock-gravity interactions in the near-surface at this study site is summarised in the cartoon shown in Figure 10c.It is consistent with our interpretation of partitioned (extensional and contractional) ground deformation possibly driven by gravity gliding on destabilised gypsiferous marl -likely associated with salt withdrawal (Trude et al., 2012) or the collapse of salt pillars in the underground mines below the Keuper Marl (Arup and Partners, 1992) -and also the geochemical evolution of groundwater according to Chebotarev (1955) sequence.Our conceptual model posits that dissolution-induced fluid overpressure in clays ultimately causes gravitational gliding deformation (or landslides) in such environments.Obviously, our study is limited by the available 2D legacy data but a future detailed 3D acquisition of seismic and preferably controlled-source electromagnetic (audio-magnetotelluric) data can be used to test or refine our proposed mechanism of near-surface deformation as well as investigate the connections to deeper (up to 300 m deep) structures at this site that will be inaccessible to the conventional dc resistivity method.

CONCLUSION
Mine collapse and surface and subsurface contaminated discharges are two potential hazards that pose significant risks to urban areas, infrastructure, agriculture and heritage.Here, we use structurally-coupled cross-gradient inversion of seismic travel-time and electrical resistivity data combined with electrical self-potential (SP) data to characterise the upper 35 m of a former salt mining area in Carrickfergus, Northern Ireland, which is punctuated by a large upstream crown collapse.Our inverted cross-gradient data allow us to distinguish unsaturated from saturated till layers overlying the saturated gypsum-bearing marl member of the Triassic Mercia Mudstone Group.We were particularly able to delineate those glacial till and marl segments saturated with salt brine.Our SP data then confirm water ingress and salt brine seepage through the saturated till and marl layers down the hydraulic gradient into farmland and towards urban areas.Indeed, this seepage previously led to downstream surface spillage of salt brine contamination, which required remediation and therefore typifies the growing anthropogenic threat that such abandoned mines are posing.We can therefore conclude that state-of-the-science integrated geophysical imaging has the capacity to provide unique and spatially extensive information in support of identification and implementation of appropriate hazard mitigation measures at historic salt mines and their interconnections.Overall, we have shown that quantitative integration of Multiphysics data has the potential to maximize accuracy and reduce uncertainty in understanding the mechanism and processes underlying ground deformation and groundwater contamination in this former salt mining area.More importantly, they allowed us to determine the hitherto unknown gravity-driven deformation of the thick gypsum-bearing marl deposits and the geochemical zonation of groundwater in the area thus providing new understanding of the attendant deformation mechanism and processes in the near-surface which have implications for the future evolution of geohazards and their mitigation for such sites worldwide.

Figure 2 .
Figure 2. Geological hazards at the site of the collocated multimodal geophysical surveys.(a) but its effectiveness remained an open question.The surface brine leakage s1 ("old seep" in Figure 2) that occurred in 2001 just before the crown hole collapse on 19 August 2001 in Maiden Mount mine prompted the 2004 multimodality geophysical surveys described below. figure.

Figure 3 .
Figure 3. Map showing the distribution of coincident dc resistivity, seismic refraction and SP t(mr,ms) =0.

Figure 4 .
Figure 4. Example of fit between observed and modelled Dc resistivity data (top image) and

Figure 5 .
Figure 5. Example of resulting integrated structurally consistent models from crossgradient joint together with their associated geospectral image and resistivity-velocity cross-plot.(The results for all the ten lines are summarised in the supplementary material S1.)All the lines are presented from east to west following the seismic profiling direction used for the common numerical inversion grid.The computed responses for all the inversion models matched the observed seismic travel time and apparent resistivity data to within a normalised root-mean square (nRMS) error of 1 as in the example shown for line 2 in Figure 4.In Figure 6 (and S1), notice the remarkable structural consistency between the seismic velocity and Dc resistivity models for each line as expected for cross-gradient imaging.The glacial till and marl with gypsum in the top 40 m shows up well in these structurally consistent images.In most cases, the glacial cover is divisible into upper and bottom sections of different resistivities and velocities, suggesting different levels of fluid saturation (Figures 5 and

Figure 6 .
Figure 6.Representative examples of the result of 2D cross-gradient joint resistivity-velocity inversion and geospectral imaging for lines 1 and 10 where surface brine-mud seepage or subsidence occurred.For each line are shown the resistivity (top left), velocity (middle left) and RGB blended geospectral (bottom left) models; the associated interpretative resistivity-velocity cross-plot is shown in the right top corner.The numbers 1, 2 and 3 refer to lithologic or hydrologic facies (or clusters) with their boundary points represented as A or B. Symbol C in the velocity models are zones of significant irregularity that possibly suggest ground instability.The black arrow marks the location where brine spilled onto the ground surface on line 1 (old seep s1) and its inferred position on line 2.

Figure 7 .
Figure 7.Comparison of various geophysical results along two key lines where surface brine- For example, along line 1 the individual seismic tomographic inversion (Figure 7e) identifies a steep dyke-like feature at profile distance 40-70 m and minor doming of the boundary of the consolidated marl with gypsum layer at profile distance 145-160 m.A steep feature is imaged at the 40-70 m profile location but not the dome by the individual resistivity imaging (Figure 7d) and the depth to its top as well as the lateral extents are different.The joint resistivity-velocity inversion models (Figures 7b and 7c) consistently show two steep features at the 40-70 and 145-160 m profile distances.There are also two localised conductive bodies of circular cross-section seen only in the individual resistivity and joint resistivity-velocity models but not imaged by separate seismic inversion.The seismic velocity model from joint inversion benefited from the contribution from resistivity imaging and vice versa in the cross-gradients process.The SP anomaly along this line is relatively higher over the zone of steep HVMR feature than elsewhere but with depressed values at the location of known surface brine-mud spillage.
Figure 8. Correlation of coincidentally-located SP profile, cross-gradients velocity and

Figure 10 .
Figure 10.Structural interpretation of the cross-gradient velocity models for line 1 (a) and line