1 Introduction

Reflection seismics is a widely used geophysical exploration tool, especially for hydrocarbon and mineral exploration (Yilmaz 2001; Malehmir et al. 2012a; Manzi et al. 2012a), and mine safety and planning (Malehmir et al. 2012b; Manzi et al. 2012b). Similarly, the technique has also been successfully used in engineering, geohydrological and environmental applications to delineate the depth to the competent layer or aquifer horizon, map contaminant plumes and features such as caves, voids, faults, mine stopes and tunnels (Steeples and Miller 1988; Miller and Steeples 1991; Leucci 2004; Redhaounia et al. 2016; Isiaka et al. 2019; Onyebueke et al. 2018). The importance of imaging the near-subsurface in detail has stimulated many advances in technology to acquire data with higher frequencies (> 100 Hz) and lower background noise (for engineering, environmental and geotechnical applications) (Steeples and Miller 1988; Miller and Steeples 1991; Brodic et al. 2018). Integration of seismics with other geophysical methods such as geoelectric, gravimetric, GPR, magnetics, and magnetotellurics that respond differently to various subsurface physical properties provide better constrained models of the subsurface (Leucci and De Giorgi 2005; Gómez-Ortiz and Martín-Crespo 2012; Cooper 2014; Martinez-Moreno et al. 2014; Onyebueke 2014; Festa et al. 2016; Malehmir et al. 2016).

In this study, we integrate high-resolution reflection/refraction seismic and electrical resistivity methods to investigate and image the near-subsurface architecture (mined-out zones, tunnels and fluid-migration pathways) in a shallow (< 100 m) historic gold mine located near Krugersdorp, South Africa, as a pilot project for future detailed site characterisation. Our results can also be useful to define the survey strategies for similar studies in karst and shallow mining environments. The study area is located within the Witwatersrand Supergroup (WS) of the Witwatersrand Basin (WB), which is known to have produced 98% of the South African gold and 45–50% of gold in the world (Handley 2004; Anhaeusser 2012).

The impact of these intense mining activities on the environment has left some underground voids and fissures, which are only partially remediated by backfilling and packing. Consequently, progressive dewatering of these structures could eventually lead to the formation of sinkholes and subsidence on the surface, which could pose a direct risk to infrastructural facilities (such as buildings, roads, and farmlands) and in general impede future urban development in the area (Wolmarans 1996). Thousands of mining related seismic events, sinkholes and subsidence have occurred in the past 60 years in the Gauteng Province of South Africa, centred on the mining district and dolomitic formations of the Transvaal Supergroup, with serious safety and socioeconomic consequences (Buttrick et al. 2011). Sinkholes and subsidence occur in dolostone formations because of the progressive water ingress, which created secondary porosity through dolomitic dissolution and recrystallization, weathering and erosion. The processes can be enhanced by the presence of carbon dioxide in groundwater (mainly from percolation through the soil, but partly derived from the atmosphere) (Whitaker and Xiao 2010).

In addition to the formation of unstable subsurface voids, mine activities also produce considerable environmental pollution. Mine waste in the Witwatersrand Basin contains abundant pyrite, which oxidises to form insoluble ferric oxide and sulphate (Brink 1979) (Fig. 1); the latter further decomposes into sulfuric acid in the presence of oxygenated water (e.g., groundwater/rain water), creating acid mine drainage. Acid mine drainage permeates through any available migration path and can transport toxic elements such as uranium, lead, copper, cadmium, mercury, and zinc, all of which are available in the mine heaps, into underground water that may be used for domestic and industrial purposes. The sulfuric acid can also erode cement and concrete and result in accelerated wear and damage to infrastructures (Brink 1979).

Fig. 1
figure 1

a Satellite image and photographs showing residential encroachment towards mining area. b and c Show the environmental degradation caused by mining activities in the area

Full size image

Hence, the inherent risk associated with shallow underground mining activities has encouraged the use of integrated geophysical methods to map potentially unstable subsurface structures (Leucci and De Giorgi 2005; Gelis et al. 2010; Gómez-Ortiz and Martín-Crespo 2012; Martinez-Moreno et al. 2014; Redhaounia et al. 2016). In the past, South African scientists have attempted to develop warning devices and surface geophysical measurements that could disclose the potential unstable subsurface conditions in time to avoid such rapid and catastrophic events (Brink 1979; Breytenbach and Bosch 2011; Kleywegt 2015). Since modern urban and rural developments are fast encroaching on previously mined areas, adequate use of different geophysical methods to obtain near-subsurface information in these areas is crucial (Fig. 1).

2 Site Location, Geology and Borehole Information

The study area is located within the Witwatersrand Basin, which lies near the centre of the Kaapvaal Craton and has an extent of ~ 350 km NE-SW and 200 km NW–SE (Corner et al. 1990; Armstrong et al. 1991) (Fig. 2). The Witwatersrand Basin is approximately 84,000 km2 in the area and about 7 km deep. The basin is filled with terrigenous and volcanic sequences (Fig. 2).

Fig. 2
figure 2

(modified from Dankert and Hein 2010)

a Geology of the Witwatersrand Basin with the location of the study area. b Geological cross-section across the study area indicating the dipping conglomerate reefs of the Central Rand Group. c An outline of the South Africa map showing the position of the Witwatersrand Basin

Full size image

The Witwatersrand Basin formed over a period of ca. 800 Ma, because of sediment accumulation due to sagging, volcanism and subsequent rifting activities between ca. 3.1 and 2.3 Ga within the Kaapvaal Craton (De Wit et al. 1992; Tinker et al. 2002). The deformation process is said to be associated with the encroachment and eventual collision of the Kaapvaal Craton with the Zimbabwe Craton between ca. 2.84 and 2.72 Ga (Robb et al. 1991; Zegers et al. 1998). According to Hayward et al. (2005), heavy minerals such as gold and uranium are effectively concentrated in the coarse gravels of the large fan-delta complexes seen mostly at the apices of the fans. The progressive lithification of these alluvial deposits formed the conglomerate rocks (known as reefs) that bear the gold and uranium mineral deposits. The sources of the gold and uranium mineralization within the basin are still contentious. Drennan et al. (1999) identified metamorphic/hydrothermal fluid systems as a result of the widespread preservation of detrital uraninite and gold remobilization along major fluid conduits in the Witwatersrand Basin, while Frimmel and Hallbauer (1999) attributed the gold-mobilization fluids in the Witwatersrand reefs to a meteoric source. On the other hand, Kirk et al. (2002) suggested rapid crustal growth during the Kaapvaal Craton formation as an evidence for the significant metal flux from the mantle, which corresponds to the Middle Archaean gold mineralization event. The Witwatersrand Supergroup is unconformably overlain by the sub-ordinated volcanic and sedimentary deposits of the Ventersdorp (ca. 2.7 Ga), Transvaal (ca. 2.6 Ga) and Karoo (ca. 302–180 Ma) Supergroups (De Wit et al. 1992). Underlying the Witwatersrand Supergroup is the ca. 3.1 Ga Dominion Group (DG) and greater than 3.1 Ga Archaean granite-greenstone basements. The Au-bearing reefs of the Witwatersrand Supergroup (Main, Bird, Kimberley, Elsburg, Basal, Carbon Leader, Steyn, Vaal, B and Composite reefs), Ventersdorp [Ventersdorp Contact Reef (VCR)] and Transvaal Supergroups [the Black Reef (BR)] have been mined for more than 100 years till date.

The study area is situated on the Central Rand Group (CRG) rocks known as the upper zone of the Witwatersrand Supergroup. It is unconformably underlain by the West Rand Group (WRG), which is considered as the lower zone of the Witwatersrand Supergroup. The thickness of the CRG varies from ca. 2.8 km in the Vredefort area to ca. 1.5 km towards the East Rand area and ca. 650 m in the Evander area (NE Witwatersrand Basin) (Robb et al. 1991). The CRG predominantly consists of quartzites and conglomerates with minor shales. The group is further subdivided into the Johannesburg and Turffontein Subgroups. A larger part of the Witwatersrand Basin was buried by enormous Vredefort impact ejecta, preserving the CRG gold-bearing reefs over an estimated period of about 2 Ga (Therriault et al. 1997). The Witwatersrand Supergroup has undergone various stages of metamorphism, structural and thermal deformation. The metamorphism and structural deformation involved progressive loading of the Witwatersrand Basin by the Ventersdorp and Transvaal sequences under extensional and rifting tectonic environments (Gibson and Reimold 2000; De Waal et al. 2006). The thermal deformation involves the intrusion of ca. 2.05 Ga Bushveld Igneous Complex, while the ca. 2.023 Ga Vredefort impact crater caused the exposure of deep-sitting Witwatersrand Supergroup to be seen as an outcrop around the vicinity of the survey area (after massive erosion of the impact ejecta) and overturned strata around the centre (known as the Vredefort Dome) of the impact (Reimold and Gibson 1996; Gibson and Reimold 2001).

The shallow subsurface geology is composed of the Johannesburg Subgroup of the Central Rand Group. It is predominantly arenaceous and argillaceous with minor rudaceous material. The lithologies were deposited in a fluvio-deltaic environment (Robb and Meyer 1995). The payable Au-bearing reefs found in the Lancaster Gold Mine are part of the Kimberley Reef Package (KRP) of the Johannesburg Subgroup (JS) of the CRG. The KRP contains various conglomerate lenses of different economically Au grades that dip approximately 28°–32° towards the south and strike approximately East–West, seldom extending to more than 60 m depth in the area, stacked upon one another with intercalated quartzites and sandstones in between (Fig. 2 and Fig. 3). The Boulder Reef of the KRP is the only consistent layer in the area with moderate Au grades (2.34–5.08 g/t) and less than 300 mm thick. Approximately 95% of the Boulder Reef has been mined-out and replaced with backfill (packed waste rock, providing roof support for stopes). The KRP as a package is about 40–50 m thick in the area, with occasionally sericitic quartzites occurring below and above the package. The KRP is underlain by intercalated conglomerate bands (Fig. 4). The surface geology comprises friable, weathered conglomerates and sandstones/quartzites with minor rudaceous materials and disintegrated boulders. The surface materials are maroon in colour, though sometimes dark brown, with minor black and white matrices (Fig. 3).

Fig. 3
figure 3

a Dipping reefs exposed on the sidewall of an open cast pit, a mine waste dump is visible on the background, b Google Earth map indicating the survey profiles, borehole locations and possible tunnel positions in the research area. The index on top right of b is photograph showing a collapsed mine entrance

Full size image
Fig. 4
figure 4

The borehole information from the Lancaster Gold Mine used to correlate the lithology of the study area. Borehole locations are shown in Fig. 3b

Full size image

3 Physical Property Measurement and Synthetic Modelling

Table 1 lists the laboratory-measured bulk densities and P-wave seismic velocities for the rock samples (each averaged over 10 readings) from two boreholes close to line L01. The physical property values were used to construct synthetic seismograms and to constrain the (seismic reflection and refraction) starting model parameters. The synthetic seismogram computation assumed normal incidence waves, convolved with the first derivative of a Gaussian wavelet (minimum phase) with a dominant frequency (FD) of 100 Hz (Fig. 5). The analysis reveals that the interface between the weathered and fresh bedrock (reflection coefficient of 0.36–0.45) is anticipated to produce a noticeable reflection. The interface between the quarzitic conglomerate band (reef lenses) and underlying sandstone yielded a seismic reflectivity of about 6–13% (0.06 to 0.126). The physical properties of the backfill zone cannot be estimated from the core measurements since the boreholes were drilled before the mining took place decades ago. Moreover, we expect a low velocity region within the surrounding host rock. Probably, the velocity values should be close to the velocity of dry or water-saturated sand for the backfill zones and the velocity of air for the open voids. We also expect that the voids and backfilled zones may be water-saturated as it rained for several days prior to the data acquisition. In addition, we utilized integrated geophysical techniques such as refraction, resistivity and reflection seismic methods to constrain the data interpretation with respect to the hidden layer problem and ambiguities associated with geophysical results (Banerjee and Gupta 1975; Miller and Steeples, 1991; Steeples and Miller 1988; Steeples et al. 1997; Ivanov et al. 2005, 2006; Onyebueke et al. 2018).

Table 1 Physical property measurement of boreholes 1 and 2
Full size table
Fig. 5
figure 5

Borehole logs with the respective reflection coefficient and synthetic seismogram. Note that the multiple traces are repetitions of the zero-offset trace at the borehole position

Full size image

4 Geophysical Methods

4.1 Seismic Method

4.1.1 Survey Design and Acquisition

The seismic survey included three P-wave seismic profiles (L01, L02, and L03) acquired using the asymmetric split-spread geometry at a fixed geophone position (Fig. 6). A series of trial acquisition tests were conducted using different source and receiver offsets prior to the actual survey to fine-tune the acquisition parameters and evaluate the attenuation response of the ground. Line L01 was shot in the East–West direction parallel to the strike direction of the dipping conglomerate reefs and perpendicular to the mine-access tunnels, while L02 and L03 were shot approximately in a North–South direction parallel to the dip direction of the conglomerate reefs. Because the study area is in the mining environment, the locations of the lines were restricted along the road ways of the mine (L01 and L03) and on the covered open cast pit (L02) (Fig. 6). The area has significant topographical features; the elevation along line L01 is quite flat, while the elevation along L02 and L03 varies with a highest topographic difference of about 6 m.

Fig. 6
figure 6

Data acquisition setup showing some mine trenches alone the profiles. a and b are images taken along L01. c and d Are taken along L02. e and d Are taken along L03

Full size image

The equipment and parameters used in the seismic data acquisition are listed in Table 2. Four Geometric Geode Seismographs were connected, each having 24 vertical component 14 Hz geophones. The recording of the seismic data was carefully controlled to avoid cultural noise. Furthermore, the signal-to-noise ratio (S/N) was improved by stacking five vertical blows of a sledgehammer on an aluminium plate at every shot station (Fig. 6). The sampling interval and recording length were 0.25 ms and 250 ms, respectively. To avoid spatial aliasing and to obtain a high-resolution subsurface image, dense receiver spacings of 1.5–2.0 m and a shot spacing of 2.0 m were employed, yielding a common mid-point (CMP) subsurface spacing of 0.75–1.0 m and nominal fold of 76–96. The recording of environmental noise (borehole drilling and vehicular movement) during data acquisition was reduced by applying a low-cut and high-cut filter of 50 Hz and 800 Hz, respectively. The geophones and the energy source were firmly coupled to the ground (Fig. 6).

Table 2 Seismic acquisition parameters for line 1, 2 and 3
Full size table

4.1.2 Raw Data Analysis

The raw shot gathers are characterised by several seismic events: the direct (D), surface (S), refracted (r1 and r2) and reflected waves (R1 and R2) (Fig. 7). Traces were normalized with respect to the maximum amplitude of the entire shot gather. The data analysis shows that the near-source traces were dominated by high amplitude surface waves, while environmental noise dominated the far-offset traces due to rapid attenuation of the source signal. Although a 50 Hz low-cut filter was applied to suppress ground roll and other low-frequency cultural noise sources, a portion of the ground roll is still visible in the data, dominantly, within the bandwidth of 45–55 Hz. The linear moveout velocity of the direct wave varies approximately from 300 to 550 m/s. The sudden change observed between distances 10 and 25 m in the gradient of the first arrival travel-time curve is qualitatively interpreted as a velocity variation within the weathered and low-velocity layer, while the gradual change at 30–50 m is interpreted to be the contrast between the weathered layer and high velocity bedrock. The velocities of r1 and r2 are about 800–2000 m/s and 1400–3000 m/s, respectively. Moreover, the coherent events at 20–25 ms (R1) and 35–50 ms (R2) on the shot gathers (Fig. 7a–c) are suspected to be reflected energy from bedrock-overburden contact and wide-angle reflections emanating from the mined-out zones. The dominant frequency (FD) of the shot gathers varies between 60 and 70 Hz, which is likely due to dominant ground roll (Fig. 7d–f). The amplitude versus frequency spectra (Fig. 7d–f) indicate a broadband character with shallow reflection frequency above 100 Hz, while the normal moveout velocity of the reflected energy varies between 800 and 1000 m/s (R1) and 1200–1800 m/s (R2) (Fig. 7).

Fig. 7
figure 7

Raw shot gathers indicating different seismic events. a–c Show a single shot gather along L01, L02 and L03, respectively. d–f The amplitude versus frequency spectra of the respective shot gathers. Here, d is the direct wave, r1 is the first refracted wave, r2 is the second refracted wave, R1 is probably the first reflected wave, R2 is probably the second reflected wave, A is the air wave, S is the surface wave, M is a multiple reflection arrival of R2, and Mo is probably a monochromatic noise

Full size image

The seismic reflection method is governed by the Zoeppritz equation and acoustic impedance (the product of density and velocity, ρ and V, respectively) contrast that exist between two different geological units (ρ2V2 − ρ1V1)/(ρ2V2 + ρ1V1) (Sheriff and Geldart 1995; Yilmaz 2001). The Zoeppritz equation describes the partition of seismic wave energy at an interface of two lithological units and relates the amplitude of an incident P-wave to the reflected and refracted P- and S-wave at a given plane boundary of an incident angle (Sheriff and Geldart 1995). The ability of a lithological boundary to produce significant reflection event depends entirely on the reflection coefficient (Rc) between two geological units, which is governed by the acoustic impedance contrast between these units. A geological interface can only generate a reflection event if the reflection coefficient at the boundary is at least 6% (0.06) (Sheriff and Geldart 1995; Yilmaz 2001; Narayan 2012).

Based on an approximated average velocity (Va) of about 900 m/s (for R1) and 1400 m/s (for R2) with average FD of 100 Hz, the average horizontal resolution (the radius of the Fresnel zone (Fz) = 0.5Va (To/FD)1 /2) for the P-wave reflectors at 20 ms and 40 ms To are estimated at 6 m and 14 m, respectively. Since the CMP sampling was 0.75–1 m, based on the deployed acquisition parameter (shot and receiver spacing, see Table 2), the horizontal sampling will be about 8–14 points per Fresnel zone above 40 ms two-way time. The vertical resolution of the data can be estimated using 1/4 wavelength criterion (Widess 1973). Taking the average P-wave velocity of R1 and R2, we obtain an approximated dominant wavelength (λD = Va/FD) of 9 m and 14 m, respectively, which indicates that the best possible vertical resolution is about 2–4 m. However, thinner features can still be detected in the seismic section provided there is high acoustic impedance contrast (Narayan 2012; Manzi et al. 2014).

4.1.3 Data Processing and Results

4.1.3.1 Refraction Seismic

The basic principle in the refraction seismic method depends on the analysis of the first-arrival compressional head wave in the seismic shot gathers. In this study, we use an advanced seismic refraction analysis technique in ReflexW software (Sandmeier 1998) to analyse and generate the refraction tomogram (RT). The technique incorporates the simultaneous iterative reconstruction technique (SIRT) algorithm. The SIRT algorithm automatically adapts synthetic travel time data (the calculated travel time) to the observed data sets (the picked travel time). However, the information obtained from the acquisition geometry, the calculated apparent velocity from the raw data and physical property lab measurement of borehole core provided the initial input model for the final tomographic inversion.

The spatial limit of the starting model is 0–50 m for the vertical extent with a 0.5 m bin size. The minimum and maximum velocity values are set at 400 and 6500 m/s, respectively. The inversion was based on an iteration model (obtained after 50 iterations) with a defined data variance of 0.01 (misfit), permissible threshold of 0.001 and 50% maximum defined change of the input model. The final selected tomography models show velocity variations in the near-subsurface along all the three survey lines (Fig. 8). Qualitatively, the tomograms reveal that the velocity of the top layer (blue) is about 400–800 m/s at 0–10 m depth. The velocity of the second layer (green, the transition zone) is about 1200–2500 m/s at about 8–20 m depth, while the velocity of the Archaean sandstone bedrock typically varies from 3500–5500 m/s and can reach up to 6500 m/s (red to magenta).

Fig. 8
figure 8

Refraction tomography results. a–c Represent velocity model along L01, L02 and L03, respectively. Note that, the contour lines represent velocity isoline of different velocity values shown in different colour band, and LVL is low velocity layer, while BV is bedrock velocity

Full size image
4.1.3.2 Reflection Seismic

The adopted processing workflow (using ReflexW software package) is listed in Table 3. These include: geometry setup; trace edits and normalizations; frequency and velocity bandpass filtering; deconvolutions before stacking; elevation, weathering and automatic residual statics; standard CMP analyses; careful refraction muting; stacking; migration; time-to-depth conversion; deconvolutions and time-varying bandpass filter after migration.

Table 3 Reflection seismic data processing sequence and parameters
Full size table

During geometry setup, all traces and shots were placed in their correct positions. Traces were normalized with respect to the maximum amplitude of the entire shot gather. High-amplitude noise and dead traces were removed. Traces were normalized to correct for spherical divergence effects. However, near-surface source-generated noise still affected the shot gathers (Fig. 9a–c). Therefore, we carefully separated direct and refracted P-waves, and surface waves from shallow reflections. Such phase recognition and separation are difficult tasks in shallow 2D seismic surveys (Steeples et al. 1997; Spitzer et al. 2003). Spiking-deconvolution of 40 ms filter length with 5% white noise provided the best results in improving the vertical resolution and compressing the signal wavelet of the seismic data; it also enhanced the frequency bandwidth (Fig. 9d, e and f) (Cary 2006). The data were subjected to a 3rd order Butterworth bandpass frequency filter (95–250 Hz) to eliminate the dominant surface wave (ground roll) (Fig. 9d–f). The remaining steeply dipping events (such as the airwave) were further attenuated by applying 450 m/s low cut velocity filter (FK filter). Careful top muting was applied to remove direct and refracted arrivals before sorting the filtered data into CMP gathers for velocity analysis. Automatic gain control (AGC, 50 ms window length) was applied to correct for energy loss at depth with a subsequent 100 ms time cut (Fig. 9g–i).

Fig. 9
figure 9

Reflection processing steps along L01, L02 and L03. a–c Illustrate trace editing and amplitude scale, respectively. d–f Illustrate deconvolution and bandpass frequency filtering. g–i Illustrate velocity bandpass filtering, top muting and time cut. j–l The frequency spectrum obtained after subsequent filtering along the respective lines. Note, the arrow SA points to the surface wave moveout, R1 the first probable reflector and R2 the second probable reflector

Full size image

Velocity analysis was carried out on CMP gathers using interactive semblance and constant velocity stack analysis (Yilmaz 1987, 2001). The optimal final velocity was based on the best alignment of normal-moveout (NMO) corrected velocity and quality of the stacked section with 45% automatic stretch mute. The final stacking velocity was used for NMO correction and these gathers were stacked together at the zero-offset to generate the final stacked seismic section after an automatic residual static correction (Fig. 10a, c, e). The random noise was suppressed with a time-varying bandpass filter (90–250 Hz) before and after migration.

Fig. 10
figure 10

Stacked reflection seismic along L01, L02 and L03. a, c, and e The final stacked time sections. b, d, and f Time-depth migrated sections after post-stack processing and static corrections. R1 and R2 are the first and second reflectors, respectively

Full size image

Elevation and weathering statics were carefully applied to correct for the effect of topographic variations and near-surface velocity heterogeneity. Static correction proved to be important due to variable overburden thickness and high-velocity contrast at the base of the weathered layer with bedrock P-wave velocity reaching up to 3500–5500 m/s. Post-stack Kirchhoff time migration was applied to collapse the diffracted energy and correctly position the dipping events (Gray et al. 2001). The stacked time-migrated seismic sections shown in Fig. 10b, d, f were converted to depth sections using the stacking velocities, which were guided by borehole constraints and laboratory physical property measurements.

The horizontal axis on the processed seismic sections (Fig. 10) shows distance along the surface and the vertical axis is the two-way travel-time. The datum level for the seismic section along lines L02 and L03 varies between 1725 and 1730 m above sea level while the surface topography varies with a maximum range of 4–5 m. The topography along line L01 was approximately flat with less than 1 m variation in elevation and the datum level was 1730 m above sea level. The measured P-wave velocity of the weathered layer and upper Archaean sandstone bedrock correlated with our CMP velocity analysis. These velocities range from 800 to 1200 m/s within the weathered layer to 1400–3500 m/s beneath the bedrock-overburden contact. The seismic sections show several reflections. A strong reflection is observed at the two-way time of about 15–20 ms between depth ranges of 14 and 25 m (Fig. 10a, b). A gently dipping coherent reflection was also observed between 20–60 ms two-way time along line L02 and L03 with intermittent reflections between 50 and 100 ms two-way time (Fig. 10c, e). In line L01, a semi-continuous coherent reflection was observed at 40–50 ms two-way time at depth of 50–65 m. Intermittent strong reflections were also noted at depths of 75, 100 and 130 m along line L01.

4.2 Electrical Resistivity Method

4.2.1 Data Acquisition

Electrical resistivity data were acquired using the ARES Resistivity meter instrument with 48 planted steel electrodes connected to coaxial electrical cables (Fig. 6). Three electrical resistivity profiles were collected along the seismic profiles (Fig. 3b). The electrical resistivity method measures the apparent resistivity of the subsurface by injecting a direct electrical current into the ground using two electrodes, and measures the potential difference using another two electrodes at various chosen spacing and distances from the current source (Keller and Frischknecht 1996; Dahlin and Zhou 2004). For optimum horizontal and vertical resolutions, Schlumberger and dipole–dipole arrays were chosen with electrode spacing of 5 m (L01 and L03) and 3 m in line L02 (Dahlin and Zhou 2004). As a rule of thumb, the maximum depth of penetration is about the total spread-length divided by three–five depending on the ground conductivity (Keller and Frischknecht 1996; Loke and Barker 1996; Loke et al. 2013).

4.2.2 Data Processing and Results

The measured apparent resistivity data were pre-processed to remove outliers before running the inversions. Data were inverted by means of 2D inverse modelling software (RES2DIV) that uses the Loke and Barker inversion method. The software applies a quasi-Newton technique to reduce the numerical calculations (Loke and Barker 1996) and ultimately produces a 2D resistivity model that satisfies the measured data in the form of a pseudo-section. This method is highly suitable where both strong lateral resistivity and depth variations occur, such as in karstic and mine areas (Loke and Barker 1996; Leucci 2004; Dahlin and Zhou 2004; Martinez-Moreno et al. 2014; Festa et al. 2016) (Fig. 11). Moderate horizontal and minimum vertical flatness procedures were chosen since the inversion procedure is smoothness-restricted (Loke and Barker 1996; Loke et al. 2013). The goodness of the fit is expressed in terms of the relative root mean square (RMS) error. The resistivity tomography models (Fig. 11) were produced after 10 iterations with an RMS error of 2.3, 1.9 and 2.9 in lines L01, L02 and L03, respectively. A maximum investigation depth of about 49, 24 and 52 m along L01, L02 and L03, respectively, were achieved with maximum offset distance of 235 m at 5 m electrode spacing for lines L01 and L03 and 141 m at 3 m electrode spacing along line L02.

Fig. 11
figure 11

Colour-coded electrical resistivity tomography results along L01, L02 and L03. a, c and e The dipole–dipole array measurements in each respective line. b, d and f The Schlumberger array measurements along each respective line

Full size image

5 Interpretation

The interpreted reflection seismic sections (RSS), refraction tomogram (RT) and electrical resistivity tomogram (ERT) are shown in Figs. 12, 13, 14a–c respectively. L02 and L03, which trend NW–SE in the investigated area, are approximately parallel to each other and perpendicular to the line L01 (trending E–W). The aim was to obtain the subsurface image of the mined area along the survey profiles. The geophysical datasets were constrained with borehole logs and experimentally-determined physical properties of borehole cores. The seismic sections are characterised by significant reflection events across the survey profiles due to the acoustic impedance contrasts at the lithological interfaces. Geological structures such as faults, joints and fractures were observed on the seismic sections (Figs. 12a, 13a, 14a) and correlated with the refraction and resistivity tomograms (Figs. 12, 13, 14b, c). The refraction and resistivity tomograms exhibit different colour bands which were carefully constrained and interpreted as different material responses. The dipole–dipole array of the resistivity tomogram was preferably used to compare with the refraction tomogram and reflection seismic section, as its better resolves lateral contrasts and isolated bodies (cavities and tunnels) (Dahlin and Zhou 2004) as shown in Figs. 12, 13 and 14c.

Fig. 12
figure 12

a Interpreted reflection seismic section along L01 showing the delineated interfaces; b interpreted refraction tomogram superimposed on the stretched-scale reflection seismic section, indicating some structural boundaries; c interpreted electrical resistivity tomogram superimposed on the stretched-scale reflection seismic section indicating the mined-out zones. Note, R1 and R2 are the first and second reflectors, respectively, CB conglomerate band reflector

Full size image
Fig. 13
figure 13

a Interpreted reflection seismic section along L02 showing the delineated interfaces; b Interpreted refraction tomogram superimposed on the stretched-scale reflection seismic section, indicating some structural boundaries; c Interpreted electrical resistivity tomogram superimposed on the stretched-scale reflection seismic section, indicating the mined-out zones. Note, R1 and R2 are the first and second reflectors, respectively, CB indicates the conglomerate band reflector, UM unmined surface block

Full size image
Fig. 14
figure 14

a Interpreted reflection seismic section along L03 showing the delineated interfaces. b Interpreted refraction tomogram superimposed on the stretched-scale reflection seismic section, indicating some structural boundaries. c Interpreted electrical resistivity tomogram superimposed on the stretched-scale reflection seismic section, indicating the mine-out zones. Note, R1 and R2 are the first and second reflectors, respectively, CB indicates the conglomerate band reflector, SS slump structure

Full size image

5.1 2D Reflection Seismics

Based on borehole data (BH1, BH2, BH3 and BH4) (Fig. 4) and synthetic seismograms (Fig. 5), the strong reflection (R1) observed at 8–20 m depth was interpreted as the boundary between the weathered layer and the bedrock, which responds distinctively to different degrees of weathering and erosion. The reflection (R2), which dips at 26°–30° on L02 and L03 (Figs. 13a, 14a) and strikes horizontally on profile L01 (Fig. 12a) is considered to emanates from the interface between the stope and the host Archaean sandstone. The intermittent and diffuse reflections (CB) at about 40 m depth most likely originate from the intercalated conglomerate bands.

The discontinuity in the amplitude of the reflection (R2) is ascribed to voids or the presence of mine pillars, or as a result of velocity artefact due to high velocity changes in the near-subsurface. Reflection truncations were observed between distances 35–140 m along L03 and slump structure (SS) at distance 50–145 m (Fig. 14a). This chaotic reflection exhibits a discontinuous amplitude reflection character, which could also be due to the presence of mine pillars and tunnels cutting across the profile. The old mine maps also indicated a tunnel between distances 70–140 m.

5.2 Refraction and Resistivity Tomography

A low velocity (400–1900 m/s) and resistivity region (150–800 Ωm) between distances 50–110 m at 12–30 m depth are observed along L01 (Fig. 12b, c). Since this region is low in both resistivity and velocity, it is likely a water-saturated zone. Hence, we interpret this region to be a backfilled and water-saturated region and a potential zone for subsurface erosion during water table fluctuation. The low resistivity values (200–1500 Ωm) towards the west between distances 115–130 m at 13–18 m depths along profile L01 are likely water ingress pathways from the pit and trenches seen on the surface, which is a favourable location for subsurface erosion. The high resistivity (1500–8000 Ωm) material at a depth of about 28 m below the ground is comparable with the strong reflection (R2) at approximately 30–32 m in depth and unresolved depth in refraction tomography (RT) (due to poor vertical penetration and resolution).

Comparably, on L02, the high resistivity regions (UM) between distance 80 and 130 correspond with the high-velocity zone (UM) observed on the refraction tomogram between distances 70–110 m and 130–143, respectively, at 6–18 m deep (Fig. 13b, c). This area is interpreted as an unmined surface block that is currently under planning for an open cast pit mining due to its variable Au grade reefs near the surface. The lower resistivity and velocity values at distance 0–70 m coincide with the loose material used to cover the mined open pit in the area. The low resistivity and velocity values across distances 35–55 m are likely due to water ingress from water-filled trenches located approximately at the 45 m position along the profile. This region may be a fluid migration path into the mined-out regions and therefore maybe subjected to subsurface erosion.

In addition, RT and ERT are comparable in resolving probably a saturated filled mine gallery/weak zone and cavity between surface distances of 30–110 m and 80–150 m, respectively on L03 (Fig. 14b, c). The weak and saturated-filled mined-out zone revealed velocity variations between 400 and 2000 m/s, and resistivity values between 100–1500 Ωm. The cavity/tunnel region reveals anomalously high resistivity values of ~ 11,000 Ωm, with a low P-wave velocity of about 400–2500 m/s. This high resistivity and low velocity zone correspond at the same position of the slump structure observed on the reflection seismic section (Fig. 14a), which could also produce the chaotic reflections noticed in the region. Hence, this region could be a potential hotspot for subsidence or sinkhole collapse. The low resistivity (80–500 Ωm) zone at distance 40–90 m with 20 m maximum depth in ERT section is probably a result of water inundation seen on the surface along the profile, which could also act as a fluid migration path into the underground mined gallery. Furthermore, the low-resistivity values (280–900 Ωm) observed at distances 120–170 m on ERT and RT section correspond with a low-velocity zone on the refraction tomogram at the same position. This area could be another potential fluid migration path into the stopes because of water ingress from the surface and could further enhance the erosion of cavities. From the ERT and refraction tomography results, we infer the velocity of the bedrock to be about 3500–5800 m/s and the resistivity of about 2500–5000 Ωm.

6 Discussion

In summary, by integrating and cross-referencing the reflection seismic with the refraction tomography (RT) and electrical resistivity tomography (ERT) sections, it is possible to map the interfaces of subsurface structures such as cavities and mined-out zones. The ambiguities in the interpretation of seismic refraction (voids filled with air or other materials) were detected and resolved using integrated geophysical methods. Hence, it is advantageous to integrate different but complementary and highly task-appropriate geophysical techniques. The ERT sections provide the best constraint on water-saturation, cavity and water pathways, while the RT sections provide the best constraint on the subsurface material competence. Both the RT and ERT sections provide greater resolution of the near-surface above 30 m compared to that of the reflection seismic sections and allowed the detection and estimation of the geometry of isolated fissures.

Complementary to RT and ERT, the reflection seismic sections highlight possible material interfaces and provide deeper investigation depth. The borehole logs provide ground-truth evidence and correlation between different geophysical responses on the seismic section (Fig. 15). The synthetic seismogram correlates very well with the borehole lithology and further constrains the final model (Fig. 16). Figure 16 provides a 3D picture of near-surface using the reflection seismic characteristics of compressional wave arrival to map the subsurface architecture such as the mined-out zone–surrounding rock interface and bedrock–overburden contact. The slump structure (SS) encountered across distances 30–120 m at ~ 90–110 m depth, dipping towards the southeast could be attributed to variations in the depositional structure or post-depositional deformation (Fig. 14a). This feature could not be resolved by RT and ERT due to their shallow depth resolving capabilities. However, the refraction and electrical resistivity tomograms reveal apparently a saturated or weathered material at depths of about 5–20 m with velocity and resistivity values of about 400–2500 m/s and 100–2800 Ωm, respectively. The Archaean Central Rand Group (CRG) sandstone/quartzite velocity and resistivity may be as high as 6500 m/s and 5000 Ωm, respectively, as per the RT and ERT sections.

Fig. 15
figure 15

Borehole logs correlated and superimposed on the interpreted reflection seismic section indicating the mined-out Boulder Reef and material interfaces. Note that, R1 and R2 are the first and second reflectors, respectively, and CB conglomerate band reflector. Note the jointing, fracturing and layering pattern of the Central Rand Group sandstone on the exposed sidewall of an open cast pit on the photograph. Borehole and synthetic seismogram superimposed on the seismic section (oriented west–east) indicate the bedrock-overburden contact and mined-out regions. The blue and white dotted lines indicate linear fractures/jointed pattern

Full size image
Fig. 16
figure 16

A 3D correlation of the reflection seismic sections along L01, L02 and L03. The photographs indicate the profile directions, the dipping and strike direction of the layers. KPR Kimberly Reef Package and UM unmined region

Full size image

7 Conclusion and Recommendation

Surface reflection/refraction seismic and electrical resistivity techniques were used to study the subsurface architectures in a historic shallow gold mine. The large material differences between the top weathered layer with that of the Witwatersrand sandstone, and the country rock with the stope resulted in high-amplitude coherent primary reflections induced by an active source (sledgehammer). The unwanted noise was removed using a standard reflection seismic processing procedure to generate reliable seismic sections. These sections were jointly interpreted with the results of other geophysical methods (seismic refraction and electrical resistivity methods) and borehole logs around the vicinity of the study area. Physical property measurements (velocity and density) were conducted on the borehole cores. These values were used to generate a synthetic seismogram, which was used to constrain the final model.

The refraction tomography (RT) results were compared with the electrical resistivity tomography (ERT) sections, which were acquired along the same profile. Their joint interpretation provides enhanced delineation of the shallow mined-out areas, the bedrock-overburden contact, cavity, weak zones and micro-geological structures. Moreover, the reflection seismic method supports the interpreted bedrock-overburden contact and the mined-out interface with the Archaean sandstone and quartzite rocks of the Central Rand Group. Hence, the southward dipping coherent reflection (R2) are interpreted as the stopes created to mine the high Au-grade conglomerate Boulder Reef. The interrupted coherent reflections (CB) are interpreted to be the intermittent conglomerate bands.

Finally, the exact geometry (widths and thicknesses) of the subsurface structures could not be resolved due to the relatively low-frequency and low-energy source, coupled with the limited imaging capability of 2D data. However, the approximate geometries of the mined-out zones, cavities and the bedrock-overburden contact, as well as some structural boundaries in the area were successfully obtained. The complementary RT and ERT sections provided comparably better subsurface images and resolutions than the reflection seismic sections above 30 m in depth. Nonetheless, a 3D acquisition technique with a high-energy source and denser receiver and shot spacing should be able to resolve the thicknesses of the mined-out regions adequately and provide further knowledge of the subsurface architecture in the area. Therefore, the integration of the reflection seismic, refraction and electrical resistivity methods were successful in mapping potentially-hazardous shallow subsurface structures in the study area and is a viable basis for other similar subsurface characterization studies.