Hydrogeology and geochemistry of a tectonically controlled, deep-seated and semi-fossil aquifer in the Zambezi Region (Namibia)

A digital elevation model was applied to identify characteristic landforms including dune fields, river fans and terraces, and faults and stream lineaments in the project area. The model uses high-resolution topographic global elevation data generated from NASA’s Shuttle Radar Topography Mission (SRTM). The raster data were downloaded from the U.S. Geological Survey’s EarthExplorer website (USGS 2015) on 22 January 2015, and have a resolution of 1 arcsecond, or about 30 m × 30 m. In addition, a true-color image dated 8 August 2017, delivered by the European Space Agency’s Sentinel-2 satellite mission, was also obtained from the USGS EarthExplorer website (USGS 2017). The RGB (red, green, blue) composite image was created from bands 4, 3, and 2 and has a resolution of 10 m × 10 m.

During the years 2003/2004, about 57 transient electromagnetic (TEM) soundings were collected in the ZR. In 2004/2005, the survey was supplemented by an additional 17 soundings in order to extend survey lines toward the north and west (Fielitz et al. 2004). Poseidon Geophysics (Pty) Ltd., Gaborone, collected these soundings as part of the Namibian-German Technical cooperation Project “Investigation of Groundwater Resources and Airborne Geophysical Investigation of Selected Mineral Targets in Namibia”.

The TEM soundings were collected using a GDP-32 receiver and a TEM/3 receiver coil (single-channel magnetic antenna) with an effective coil area of 10,000 m². In-loop configuration was applied. The transmitter coil was 200 m × 200 m, single-turn, and a GGT-10 transmitter was used. All equipment was supplied by Zonge International (Zonge 2018). In order to achieve investigation depths of some 200 m, the measurements were performed applying 20 A for 1 Hz, 15 A for 4 Hz, and 5 A for 16 Hz—the lower the frequency, the longer the transient and the deeper the investigation depth. The measurements were stacked and averaged, 64 cycles for 1 Hz, 256 cycles for 4 Hz, and 512 cycles for 16 Hz. Standard processing software by Zonge (SHRED, TEMAVG) was used for stacking, averaging and visualization of the data. For data inversion, the TEMIX Plus software program (Interpex Limited, Interpex 2016) was used, and a 1-D inversion was applied.

The soundings were positioned along 16 profiles, partly crossing each other, and with irregular spacing between the soundings. The location of the soundings was aimed at covering the supposed area of the deep freshwater aquifer including its borders. The data quality was very satisfactory. In most cases, it was possible to collect transients up to 20 ms and longer, with low noise. At some locations, the transients collected during the first and the supplementary campaign differed. The detected resistivity shift could not be attributed to technical causes and was considered a reflection of the geological heterogeneity close to the northern graben structure, where most of the supplementary measurements were located.

Groundwater contours for the UKA are based on water table data obtained from the database of the Namibian Department of Water Affairs and Forestry (DWAF). The water table measurements were converted into piezometric heads using elevations obtained from SRTM imagery (1 arcsec or approx. 30 m × 30 m, see Chapter 2.2). The dataset comprised about 750 values for water tables that were taken between 1970 and 2015 during different seasons, i.e. rainy or dry season. Variogram analysis and universal kriging with linear drift was applied for regionalization of water level data and mapping of groundwater contours. In order to construct groundwater contours for the LKA, 16 water table measurements were taken during the end of the dry season from 10 to 23 November 2015. Another nine older water table measurements were added from the database. Contours had to be constructed manually, as kriging or another interpolation method proved unsuccessful due to the limited number and quality of data.

In addition, results from pumping test analyses were statistically examined. Previous pumping test results were reported by Margane and Bäumle (2004) and Margane and Beukes (2007). Additional test data from the DWAF archives were analyzed. Finally, the initial, pumped and residual water levels were measured manually and continuously recorded with pressure probes during the sampling campaign conducted in November 2015. The results allowed for rudimentary test pumping analysis. In total, 42 single-well pumping tests from 35 boreholes were able to be analyzed: 20 for the UKA and 22 for the LKA. Pumping duration ranged from 1 to 72 h, and pumping rates also varied widely, from 0.2 to 58 l/s.

Groundwater samples from 22 boreholes were obtained during the field campaign carried out in November 2015. The bulk of groundwater samples were taken along a north–south-directed transect at around 23°45′E (Fig. 3). The boreholes are mostly used for providing drinking water or water for stock in cattle camps, and are installed with a submersible pump or in lesser cases with a hand pump. Five of the sampled boreholes are used as monitoring wells by DWAF and are thus permanently equipped with data loggers. Except for the monitoring boreholes, completion reports containing a detailed lithological description were not always available. It was hence difficult to determine with certainty the position of the confining unit separating the LKA from the UKA. It was deduced from this that boreholes with depths exceeding 130 m could potentially tap the LKA. Fifteen of the 21 sampled boreholes fulfilled this depth criteria (maximum depth 250 m). It should be noted that for boreholes 037223, 200501, 200503, 200504 and 200620, the position of the screens is unknown, and that for boreholes 202147, 202148 and possibly 201624, the upper parts of the screens are located within the UKA (Appendix Table 2). During this and two previous campaigns, a number of additional samples were taken from boreholes of the UKA and the Kwando River, as well as from a PVC rainfall collector that was set up at Camp Kwando Lodge (23°19′E, 18°03′S, 963 m a.s.l.), about 25 km south of Kongola Bridge, for purposes of comparison.

Physicochemical parameters were measured using commercial portable meters (WTW/Xylem Group) as part of through-flow cells to avoid any contact of groundwater with atmospheric oxygen and carbon dioxide. In situ measurements included temperature (T), electrical conductivity (EC), pH, dissolved oxygen (DO) and oxidation-reduction potential (Eh)(measured against an Ag/AgCl electrode and corrected with respect to standard hydrogen electrode). The pump rate was measured volumetrically using a 20-l container and a stopwatch. Alkalinity and acidity were determined onsite by potentiometric titration on 100-ml unfiltered samples using a field burette. The normality of the titrant solutions (HCl, NaOH) was 0.047 N. The titration was considered important, as in previous campaigns it was observed that pH measured in situ was usually higher (more alkaline) than in samples in the laboratory. The lowering of the pH was explained by dissolution of atmospheric CO₂ in the sample water. A fixed-endpoint method (to pH 4.3), and USGS Inflection Point Titration (IPT) and GRAN function plot methods were applied for analysis (Rounds 2012). The data were entered in a customized spreadsheet and preliminarily analyzed on a laptop in the field. The samples were taken once all in situ parameters except Eh had completely stabilized, which took between 60 and 120 min. Eh values dropped continuously from the start of pumping, but toward the end of pumping, the fluctuations gradually diminished. The submersible pumps yielded pump rates in the range of 0.8 to 1.4 l/s, which resulted in a total abstracted volume of water of between 2 and 8.6 m³. Much smaller amounts, however, were extracted from boreholes with hand pumps. The samples were continuously kept at low temperature using cooling boxes and frozen cooling elements.

Inorganic components and the stable hydrogen and oxygen isotopes were analyzed by the BGR laboratory in Hannover, Germany. Carbon isotopes and chlorine-36 were analyzed by the external accredited laboratory, Hydroisotop GmbH, in Schweitenkirchen, Germany. The applied analysis methods are summarized in Table 1. The isotopic composition of hydrogen (δ²H) and oxygen (δ¹⁸O) was determined using a Picarro cavity ring-down spectrometer (CRDS) equipped with a CTC autosampler. The measurements were normalized to the Vienna Standard Mean Ocean Water/Standard Light Atmospheric Precipitation (VSMOW/SLAP) scale, where values of 0 and −428 ‰ for δ²H and 0 and −55.5 ‰ for δ¹⁸O were assigned to VSMOW and SLAP, respectively. The external reproducibility (standard deviation of a control standard during all runs) was better than ±1.0 ‰ for δ²H and ±0.1 ‰ for δ¹⁸O. The results for δ¹⁸O and δ²H are expressed in the conventional notation as relative deviations in per mil (‰ or 10⁻³) from the VSMOW according to the following relationship (Coplen 2011):where R_sample is the isotopic ratio of the sample (²H/¹H or ¹⁸O/¹⁶O) and R_VSMOW is the isotopic ratio of the standard.

The deuterium excess D_ex is calculated from the isotopic composition of δ²H and δ¹⁸O as follows:

The D_ex is about 10 ‰ for rain that plots near the Global Meteoric Water Line (GMWL), but is much lower for surface water and groundwater that have lost some volume to evaporation. The D_ex thus serves as a valuable tool for tracing the origin of water.

Carbon-13 (¹³C) content was determined using isotope ratio mass spectrometry (IRMS). The ¹³C abundance is reported as the deviation in per mil (‰ or 10⁻³) of the ratio ¹³C/¹²C in the sample relative to the ¹³C/¹²C ratio of the Vienna Pee Dee Belemnite (VPDB) standard, i.e. as δ¹³C ‰ VPDB. Measurement reproducibility of duplicates was better than ±0.3 ‰. Carbon-14 (¹⁴C) was analyzed using accelerator mass spectrometry (AMS), and ¹⁴C activity is given in percent modern carbon (%mC). ¹⁴C content expressed in %mC represents the deviation from ¹⁴C activity in the undisturbed atmospheric equilibrium that equals 100 %mC or 0.226 Bq/g of carbon. Chlorine-36 (³⁶Cl) is measured as ³⁶Cl/Cl ratios and is also determined by AMS. Chloride needs to be extracted from acidified solutions by adding excess AgNO₃, resulting in the precipitation of AgCl. Samples with chloride content of less than about 20 mg/L are considered critical, as the mass of precipitated chloride may be insufficient for determining the radioactive isotope ³⁶Cl. Since sulfur concentrations are generally high in the study area, and because the sulfur-36 isotope interferes with ³⁶Cl during AMS analysis, sulfur has to be removed from the samples in an intermediate step. This is achieved by precipitation as BaSO₄ (with Ba(NO₃)₂) and subsequent removal by centrifugation. The ³⁶Cl concentrations are expressed as 10⁻¹⁵ atoms of ³⁶Cl per total atoms of Cl.

Radioactive tracers can be applied to determine the apparent age of groundwater. Radioactive tracers used for the longer timescales are ¹⁴C and ³⁶Cl. Both are cosmogenic nuclides, naturally produced in the upper atmosphere from cosmic radiation. For radioactive tracers with known or constant input function, an apparent age can be derived from concentrations using the equation of radioactive decay. Radiocarbon has a half-life of 5730 years, which allows the determination of timescales between some thousand and approximately 35,000 years. ³⁶Cl has a half-life of 301,000 years and is suitable for determining ages on the order of 10⁵ to 10⁶ years (e.g. Kalin 1999; Phillips 1999; Mook 2000).

Radioactive decay is described by the exponential function:

where

The apparent groundwater age of a sample is calculated as

Phillips (2013) described an easy mass balance equation for the absolute ³⁶Cl concentration in an aquifer. The model incorporates radioactive decay of the recharged ³⁶Cl, underground production in the aquifer toward a secular equilibrium ratio of ³⁶Cl /Cl and an additional source of chloride, such as diffusion from an adjacent aquitard, which may have another ³⁶Cl /Cl ratio in secular equilibrium. The Phillips model expressed for the tracer concentration \( {\mathrm{A}}_{36_{\mathrm{Cl}}} \) as given in Suckow et al. (2016) reads as follows:

with

where

If R_se2 = R_se1, the equation simplifies and reads:and the apparent groundwater age is calculated as (Bentley et al. 1986; Phillips et al. 1986)

Furthermore, if chloride is added from dissolution of halite, then R_se = 0, and consequently (Phillips 2013):

In this case, the equation for apparent groundwater age reads:where Cl and Cl_re represent the measured chloride concentration and the chloride concentration of recharge in mg/L, respectively.

The high-resolution DEM and true-color image proved suitable for detecting landforms and tectonic regimes in the study area. The Linyanti and Chobe faults are easily detectable in the DEM. The Linyanti fault line located south of the Linyanti River and the Magweqgana Stream is expressed in vertical displacements on the surface of about 4–8 m. Paleolakes likely existed several times in the down-faulted areas north of the Linyanti and Chobe faults. Lake Liambezi (Fig. 2), which episodically fills up after extreme flooding, can be regarded as a remnant of these paleolakes. The Sibbinda Fault to the north marks the boundary of a zone of gradually rising elevation and is considered a blind fault. The elevation south of the fault line is about 970 m above mean sea level (a.s.l.), whereas the plateau north of the fault reaches values well above 1000 m a.s.l. (Fig. 2).

The Kwando and Zambezi rivers have deposited fans into the ORZ similar to but smaller than the Okavango Delta to the southwest (Fig. 4). These fans were also recently discovered from digital elevation data by Burke and Wilkinson (2016). The Kwando Fan and parts of the Zambezi Fan are essentially inactive, because its formative rivers have incised into the relict fan deposits. The fans therefore form the upper part of the older alluvium. Figure 4 also illustrates the strong tectonic control on the drainage system in the study area. Apart from the obvious elbow-shaped river diversion near the Linyanti Fault, the course of the Kwando River is also deflected south of the Matabele-Mulonga Plains and north and south of Kongola. The Matabele-Mulonga Plains, in fact, are believed to represent an abandoned arm of the Kwando River, which hence initially joined the Zambezi in the area of the Barotse-Bulozi Floodplains (Moore et al. 2013). The near-parallel course of the Kwando and Upper Zambezi rivers, as well as the course of the eastern tributaries to the Zambezi such as the Lumbe and Njoko rivers, is also most likely an expression of the regional tectonic setting. The Ngonye Falls near Sioma provide evidence that the section of the Zambezi River downstream of the falls is a relatively young valley. The rapids are a clear sign of rejuvenation of an otherwise very mature longitudinal river profile. The Zambezi obviously once followed a route to the west of its modern course crossing the Siloana Floodplain, as indicated by the arrows in Fig. 4. Dune fields in the Northern Kalahari generally developed to the west of large river valleys which supplied the sands for dune formation. To the west of the Kwando and Okavango rivers, linear dune fields are well preserved. The fact that the Siloana Floodplain—in contrast—is characterized by largely degraded dunes and areas without remnant dunes is another indication that the area has undergone extensive fluvial reworking.

The supplementary TEM measurements revealed the boundaries of the alluvial aquifer system toward the east and north (Fig. 5). The basalt appears at sounding 4_16 at profile 9, at sounding 1_200 at profile 1 and at sounding 3_50 at profile 3. All soundings north/west and north along these profiles show the indication of high-resistive basalt (Fig. 5c). The TEM inversion results show a decrease in resistivity toward the alluvial basin. Along profile 9, the basalt dips toward the southeast, from about 20 m depth at sounding 4_08 to about 100 m depth at sounding 4_16. The alluvial filling, with its relatively low resistivity of some 20 Ωm, is clearly seen from sounding 4_20 toward the southeast. From sounding 4_08 to sounding 6_5, the resistivity of the basalt layer drops from about 1000 Ωm at 4_08 to some 50 Ωm at 6_5.

At the southern end of the TEM profiles, no indication of higher resistivity at depth is detected. In Fig. 5c, the inversion result of the most southern sounding of the whole campaign (sounding 1_30) shows very low resistivity at depth (about 5 Ωm). The shape of the transient gives no hint of resistive layers at depth. However, owing to the low resistivity of the layers at this site (below about 120 m depth), the investigation depth of this sounding is limited. Although the figure displays the vertical profile to a depth of some 350 m, such investigation depth cannot be reached by the measurement layout as described above given the low resistivity of the alluvial layers. From theoretical calculation it can be concluded that a resistive layer deeper than about 280 m would remain unnoticed by the TEM sounding at this point.

Figure 6 shows the piezometric surfaces of groundwater in the UKA and LKA. For the UKA, only minor manual corrections of automatically generated contour lines—mainly in areas with low measurement density—were required. Groundwater of the UKA flows in a north–south direction in the eastern parts of the ZR and in a west–east direction near the Kwando River floodplains (Fig. 6a). The latter is an indication that losing-stream conditions occur in the floodplains of the Kwando River.

The LKA appears to be confined throughout the entire area. Based on water level measurements taken in November 2015, the piezometric head in the LKA near borehole no. 041004 (Fig. 3) is over 14 m higher than in adjacent boreholes of the UKA. To the south, the measured head difference at the borehole group at site no. 200254 for the same period amounted to about 4 m. The piezometric head difference hence remains positive but gradually decreases toward the south. However, reliable groundwater contours could not be established until now. This is due to the fairly small number of boreholes, and hence limited data, and the prevailing low hydraulic gradients. In addition, most boreholes penetrate the aquifer only partially, and individual screen depths vary considerably. It is also probable that some boreholes are connected to the UKA. Despite these limitations, a general north to south flow direction could be established from hydraulic head measurements taken during November 2015 (Fig. 6a). The hydraulic gradient in this direction is estimated at 0.08 ‰.

The variable sedimentological environment, with phases of fluvial, fluvio-deltaic and lacustrine conditions, is not only responsible for the high lithological variability, but also explains the variations in hydraulic transmissivity. The box-plots shown in Fig. 6c represent the results from the available 42 single-well pumping tests. The LKA is characterized by higher transmissivity with a lower dispersion compared to the UKA. The median values of transmissivity amount to 59 m²/d for the LKA compared to 20 m²/d for the UKA.

The general chemical groundwater composition exhibits strong heterogeneity, as shown in the percentile plots of Fig. 7. Due to the effects of evapoconcentration and dissolution of salts, groundwater often becomes brackish. Total dissolved solid (TDS) content of up to 600 mg/L can largely be explained by the dissolution of calcite and magnesite. Weathering of silicates also plays a role; silica content, in fact, is exceptionally high, sometimes exceeding 100 mg/L SiO₂, due to the relatively high solubility of quartz at the prevailing groundwater temperatures of >25 °C and as a result of very active CO₂ production in subtropical soils. TDS values of above 600 mg/L indicate a growing influence of evapoconcentration and/or dissolution of halite and gypsum, leading to the dominance of Na, Cl and SO₄ content in groundwater (Plata Bedmar 1999). Groundwater within or near the Kwando floodplains is generally fresh, with low TDS (median of 444 mg/L) and very low sulfate and chloride (not included in Fig. 7) content. Groundwater of this section is usually of the calcium-magnesium-bicarbonate type. Other areas show overall increased levels of sodium, chloride, sulfate and total mineral content. Boreholes in the older alluvium and in the Linyanti-Chobe swamps show extreme variations; brackish groundwater with high sulfate content is common for both areas. By contrast, the LKA shows a homogeneous composition with fair water quality (Fig. 7). Median values for TDS, sulfate and chloride are 959, 211 and 126 mg/L, respectively.

Chemical reactions such as dissolution, precipitation, ion exchange and redox reactions control and may alter the chemical composition of groundwater along the flow path. The presumed flow direction within the LKA is from the Namibian border with Zambia in the north to the border with Botswana in the south. The total flow length is approximately 80 km. The general evolution of groundwater chemistry leads from a Ca-Mg-HCO₃ to an Na-(HCO₃ + Cl + SO₄) type, as shown in the piper diagram (Fig. 8). Along the groundwater flow direction, a notable increase in TDS, chloride, sulfate and sodium content, as well as alkalinity, can be observed (Fig. 9). In contrast, calcium and magnesium content are overall very low, with values between 2 and 4 mg/L. Selected tabulated chemical analysis results obtained during this study are given in Appendix Table 2. Saturation indices calculated by the PHREEQC model (Parkhurst and Appelo 1999) show that the groundwater of the LKA is generally saturated with respect to calcite, magnesite and amorphous silica, and undersaturated with respect to all other major salts including gypsum, fluorspar and, obviously, halite. TDS values exceed 1500 mg/L in the southernmost areas near the Linyanti River. The correlation with flow distance is fairly strong, as shown by the Pearson product-moment correlation coefficient, r, of 0.88. Chloride concentrations increase from below 100 mg/L to over 200 mg/L, with a median of 126 mg/L. The correlation is not as strong as for TDS (r = 0.62) due to a significant number of outliers. Note that borehole no. 037723, whose chemistry does not seem to fit into the regional picture, is located well outside the boundaries of the known extent of the LKA (Fig. 3). Sulfate content also increases steadily (r = 0.67) and reaches very high levels of above 300 mg/L. The SO₄/Cl ratio (meq:meq) is on the order of 0.8–1.5, which is about 7.5–15 times that of the marine ratio of 0.104, and shows no clear trend in the N–S direction. Four rainfall samples collected at the beginning of the rainfall seasons in November 2014 and 2015, respectively, show high variations in the SO₄/Cl ratio of between 0.5 and 5.0. Unfortunately, additional data, especially during the peak of the rainy season, is not available. Although deposition of ions in bulk precipitation (precipitation + dry fallout) in windy, semi-arid environments might be higher than in humid and coastal areas, evapoconcentration of rainwater cannot explain the increase in chloride and sulfate along the flow path under confined conditions. Hence, the increase in chloride and sulfate is most likely predominantly caused by dissolution of halite and gypsum. The Br/Cl ratio can be indicative of halite dissolution. Whilst moderate evapoconcentration causes only minor changes in the Cl/Br ratio, halite dissolution may increase the value of molar Cl/Br to several thousands (Braitsch and Günter 1962). Molar Cl/Br in marine water is about 655 ± 4, and that of continental rainwater between 50 and 650 (Alcalá and Custodio 2004; Fontes and Matray 1993). Observed values in the LKA show no trend in the N–S direction and vary between 950 and 1400. They are hence indicative of moderate halite dissolution. Higher and critical (>1.5 mg/L) fluoride concentrations occur in the south of the study area. Dissolution of fluorspar facilitated by low calcium content could be one explanation; however, this would not explain the increase in fluoride toward the south, as groundwater is undersaturated with respect to fluorite throughout the area. It is therefore more likely that desorption from clay minerals—facilitated by high pH (>8.5) and competition with hydroxyl ions—is the major mechanism, as described by Jacks (2016). Sodium experiences a strong and almost uniform increase (r = 0.81) along the flow path, with sodium content increasing from about 50 to >600 mg/L. The Na/Cl ratio (meq:meq) remains fairly stable, with only a minor increase toward the south. The values, typically in the range of 3–4, show that halite is not the only source of sodium in the system. The increase in total dissolved inorganic carbon (TDIC) and, therefore, total alkalinity is also strongly correlated with flow distance (r = 0.89). Maximum values exceed 100 mg C/L. Alkalinity clearly increases by a factor of 6 to 8 at a rate of about 1.54 mg C/L per km flow distance. Furthermore, a striking correlation exists between the increase in alkalinity and in “sodium-excess”, Na-Cl, defined as the molar difference between sodium and chloride in a sample. Na-Cl represents sodium that originates from sources other than halite dissolution, such as silicate weathering and ion exchange processes. The ratio of Na-Cl vs. (alkalinity + SO₄) within the LKA is almost consistently ±1, as shown in Fig. 9.

The processes that can best explain the geochemical evolution of groundwater in the LKA are cation exchange processes that consume alkaline earth metals and produce Na. The continuous removal of Ca by cation exchange along the flow path drives the progressive dissolution of calcite along the flow path and contributes alkalinity to the groundwater. In addition, sulfate is added by dissolution of gypsum. The geochemical evolution can be described by the chemical reactions:and results in an Na-HCO₃-SO₄ water type. Due to the observed low redox potential, and rotten smell at some sites, there is a possibility that alkalinity is also produced by reduction of sulfate following the equation

However, considering the high SO₄ content in the LKA, this process is most likely of minor importance. The increase in alkalinity and the development of Na-HCO₃ water is a common phenomenon in deep-seated old groundwater, and was also reported in the Great Artesian Basin in Australia (Cresswell and Duesterberg 2012; Suckow et al. 2016), the Cretaceous sandstones of the Milk River Aquifer in Alberta, Canada (Phillips et al. 1995), and the Triassic–Cretaceous Guarani aquifer system in Brazil (Gastmans et al. 2010). Calcite is known to be abundant in the Kalahari sediments in the form of calcareous sandstone and calcrete. The prevalence of sodium as the main sorbate and the occurrence of gypsum are indicative of a more saline environment. In fact, the processes resemble that of freshening of a salty aquifer in coastal areas as described, for instance, by Appelo and Postma (2007) and van der Kemp et al. (2000). Freshening is also known to occur in aquifers within continental sedimentary basins such as the transboundary Guarani Aquifer system in South America (Houben et al. 2015; Sracek and Hirata 2002), which is composed of continental fluvial and aeolian sandstones. The Guarani Aquifer was deposited under arid conditions similar to the Northern Kalahari environment. Consequently, the system was originally brackish. It is assumed, however, that in the Guarani Aquifer, large amounts of salts may have already been flushed out after Cretaceous tectonic uplift. In contrast to the LKA, high sodium, chloride and sulfate concentrations in the center of the Guarani Basin possibly originate from mixing with water from the underlying, more saline unit or diffusion from incompletely flushed zones of lower permeability within the aquifer. What is most noteworthy about the downgradient chemical evolution observed in the LKA is that it confirms hydraulic continuity along a north–south-directed flow direction despite very low and uncertain hydraulic gradients, and provides evidence that the LKA is an active hydrogeological system that still forms part of the modern hydrological cycle.

Hydro-isotope analysis results are given in Appendix Table 3. Figure 10a shows the stable isotope composition of groundwater, surface water and rainfall samples. The bulk of the samples from the LKA show the lowest observed values of deuterium and oxygen-18 of between −56 and −62 ‰ and −8.7 and −8.1 ‰ VSMOW, respectively. These samples plot near the Local Meteoric Water Line (LMWL) determined by Beyer et al. (2016) for the Ohangwena Region in northern Namibia. Four exceptions occur, namely the samples from wells no. 201624, 202147, 202148 and 041002. Their position in the diagram is clearly below the LMWL. These four wells are located in the northern margins of the known extent of the LKA (Fig. 3). Of these wells, the LKA has been properly sealed off only in well no. 041002; the screen position at boreholes no. 201624, 202147 and 202148 is 120–132 m, 90–145 m and 85–129 m, respectively (Appendix Table 2), and suggests that groundwater pumped from these wells represents mixed water from the UKA and LKA, which can explain the uncharacteristic isotope signature. Together with wells of the UKA, these samples lie near an evaporation line (EL) and are isotopically enriched due to kinetic isotopic fractionation during non-equilibrium evaporation of water. This is in good agreement with stable isotope results from previous studies in the ZR, which are shown together with the new analyses in Fig. 10b. The EL determined from all available samples follows the relationship y = 5.47x − 12.4 (r = 0.99). Samples plotting along the EL are generally characterized by low D_ex (Eq. (2)) of between 5 and −10 ‰, whereas most samples of the LKA have values between 7.7 and 10.3 ‰; the LKA values are closer to the D_ex of the local and global meteoric water, which implies that the LKA is replenished by direct recharge and fast infiltration (e.g. during storm events) and that indirect recharge from rivers or open water areas is unlikely. The orange line in Fig. 10b represents a mixing line and indicates that mixing of different groundwater with various exposure to evaporation could also, to some extent, explain the stable isotope variations. Mixing of groundwater of different age and origin may be induced by sampling boreholes with long well-screen sections which are typical for semi-arid areas like the ZR. The number of surface water samples (Fig. 10a) is insufficient for a reliable interpretation. As is to be expected, the available data also plot on an evaporation line indicative of evaporative losses; the best fit to the surface water samples shown in Fig. 10a, however, appears to differ from the EL for groundwater.

The isotope content of modern precipitation differs significantly from the characteristic low isotope content obtained for the LKA. The weighted average isotope content at climate stations of the Global Network of Isotopes (GNIP) in southern Africa is added to Fig. 10a (GNIP 2017). The Online Isotopes in Precipitation Calculator (OPIC) predicts—based on interpolation of the isotopic composition of modern meteoric precipitation—values of deuterium and oxygen-18 of −28 ± 6 ‰ and −4.6 ± 0.8‰ VSMOW for the study area (Bowen 2017; Bowen and Revenaugh 2003). The much lower values for the majority of sites from the LKA (and a noteworthy number from the UKA) could be explained by the temperature effect on isotope fractionation, a change in seasonality (i.e. from summer- to winter-dominated) or a stronger influence of continentality that could be produced by a change in the circulation pattern. As there is no evidence for a fundamental change in the rain-bringing climate systems in the region, the variation in average temperature is the most likely explanation. The lower isotope concentrations hence indicate that groundwater was formed under cooler than modern climatic conditions, such as during the Pleistocene, when average annual temperatures in southern Africa were about 5 °C lower than today (Holmgren et al. 2003; Kulongoski et al. 2004). The amount effect, i.e. if the percentage of high-intensity rainfall events is increased, could also contribute to lowering the δ¹⁸O and δ²H values of rainfall.

The bulk of the LKA samples are characterized by remarkably high ¹³C content of between −6.6 and −4.4 ‰ VPDB and very low ¹⁴C activity below 6 %mC (Fig. 11a). Exceptions are again boreholes no. 201624, 202147 and 202148, which are assumed to partially tap the UKA, with δ¹³C < −11 ‰ VPDB and ¹⁴C concentrations between 30 and 42 %mC. Borehole no. 037723, which is 164 m deep but located to the northeast of the proven extent of the LKA (Fig. 3), also shows both lower ¹³C and higher ¹⁴C concentrations. Borehole no. 041002 has ¹³C content typical of the LKA but much higher ¹⁴C activity.

In an open system, the carbon-13 content of dissolved inorganic carbon in groundwater, δ¹³C_DIC, depends on the ¹³C content of soil CO₂, \( \updelta {{}^{13}\mathrm{C}}_{\mathrm{C}{\mathrm{O}}_2\left(\mathrm{g}\right)} \), which is generated by the degradation of biomass, as well as pH and temperature. The pH determines the relative abundance of carbonate species CO_2(aq), H₂CO_3,\( \mathrm{HC}{\mathrm{O}}_3^{-} \)and\( \mathrm{C}{\mathrm{O}}_3^{2-} \) in the water, and temperature influences the fractionation enrichment factors (e.g. Clark 2015). Common trees and bushes (Acacia, Combretum, Sclerocarya) near the Tswaing Crater (25°25′S, 28°5′E) show organic δ¹³C_org of −30.7 to −25.5 ‰, whereas grasses and macrophytes (e.g. Cyperaceae) show values of −13.2 to −11.4 ‰ VPDB (Kristen et al. 2010). The lower ¹³C content of many grasses is due to the fact that they use C4 photosynthesis, whereas most savannah trees use C3 photosynthesis. C4 plants are able to minimize water loss in hot, dry environments due to improved photosynthetic rate and efficiency, and are therefore better adapted to drier climate and lower atmospheric CO₂ concentrations (e.g. Lara and Andreo 2011; Osborne and Sack 2012). Assuming that plants using C3 photosynthesis dominate, then \( \updelta {{}^{13}\mathrm{C}}_{\mathrm{C}{\mathrm{O}}_2\left(\mathrm{g}\right)} \) typically ranges from −23 to −20 ‰ VPDB (Clark 2015). At a temperature of 28 °C, the expected δ¹³C_DIC of groundwater in an open system is between −16.9 and −13.9 ‰ at pH 7, and −15.5 and −12.5 ‰ VPDB at pH 8. For a closed system, exchange with soil CO₂ is inhibited, and carbonate dissolution will reduce the initial ¹³C content by half according to the stoichiometry of the carbonate dissolution. ¹³C concentrations in calcretes, δ¹³C_calcrete, worldwide range from −14 to 4 ‰, with a maximum of −6 to −4 ‰VPDB (Talma and Netterberg 1983). Using −24 to −21 ‰ for \( \updelta {{}^{13}\mathrm{C}}_{\mathrm{C}{\mathrm{O}}_2\left(\mathrm{aq}\right)} \) and −4 ‰ for solid freshwater carbonate, the resulting δ¹³C_DIC of groundwater is −14 to −12.5 ‰ VPDB, and hence in a similar range as that of an open system.

Samples from the UKA and the few outliers of the LKA samples have ¹³C concentration similar to the values to be expected for open and closed systems. The majority of the samples from the LKA, however, have much higher values. One reason for this could be higher than presumed values of δ¹³C_calcrete or a different CO₂ source. Unfortunately, sedimentary analyses of calcretes are not available in the study area. According to Talma and Netterberg (1983), soil \( \updelta {{}^{13}\mathrm{C}}_{\mathrm{C}{\mathrm{O}}_2\left(\mathrm{g}\right)} \)of −18 to −7 ‰ VPDB was measured in grassland-dominated areas of southern Africa. An abundance of C4 plants thus could be responsible for higher initial soil \( \updelta {{}^{13}\mathrm{C}}_{\mathrm{C}{\mathrm{O}}_2\left(\mathrm{g}\right)} \). Assuming soil \( \updelta {{}^{13}\mathrm{C}}_{\mathrm{C}{\mathrm{O}}_2\left(\mathrm{g}\right)} \) to be about −12 to −10 ‰ during the time of recharge of groundwater, this results in δ¹³C_DIC of groundwater of −8.5 to −7.5 ‰ for closed-system conditions. For the LKA, this could mean that grassland may have prevailed during recharge. Recharge under grass-covered areas is usually higher than under densely wooded grassland, even if mean rainfall amounts are reduced due to the absence of deep-rooting trees. This observation is particularly pronounced if storm events with high rainfall intensity prevail (e.g. Kim and Jackson 2012).

Another, more likely, explanation assumes that the groundwater of the LKA is very old and the δ¹³C_DIC concentrations are dominated by the carbon isotope characteristics of rock. This is because the observed progressive dissolution of carbonate associated with an increase in alkalinity and DIC steadily dilutes the original δ¹³C_DIC (and ¹⁴C) content. Furthermore, due to the long residence time, solute–matrix exchange occurs and groundwater DIC will approach isotopic equilibrium with carbonate rocks (e.g. Han et al. 2014). Assuming pH > 8, temperature of 28 °C and δ¹³C_calcrete of −4 ‰, groundwater DIC at isotopic equilibrium with the rock matrix is enriched by about 1 ‰ compared to calcretes. The resulting value of −5 ‰ VPDB is in good agreement with measured δ¹³C_DIC values. Similar effects have been observed in central parts of the Guarani Aquifer system, where dissolution of calcareous cement and isotopic exchange resulted in increased ¹³C isotope ratios of about −6 to −8 ‰ (Gastmans et al. 2010).

Dissolution of carbonates is obvious from the observed geochemical evolution and increase in TDIC down the flow path. It is therefore tempting to assume that the very low observed ¹⁴C concentrations are linked to the dissolution of dead carbon. ³⁶Cl/Cl ratios, however, are very low, suggesting that groundwater is much older than 50 ka and that ¹⁴C must have virtually fully decayed due to its shorter half-life. This becomes obvious from the binary plot in Fig. 11b that shows the observed ¹⁴C concentrations and ³⁶Cl/Cl ratios together with a theoretical isotope composition following simple exponential decay of the estimated respective initial concentrations. The plot can also be used to estimate the experimental detection limit for ¹⁴C concentrations during this investigation, as suggested by Suckow et al. (2016). The purple horizontal line in Fig. 11b suggests that concentrations below about 5–6 %mC are affected by contamination of the samples by atmospheric CO₂ during sampling and/or analysis. This error range is plausible, as exposure of the samples to air could not be fully avoided, and CO₂ uptake is facilitated by the alkaline pH of the samples (Aggarwal et al. 2014). Mixing between older groundwater, characterized by high ¹³C content, low ³⁶Cl/Cl ratios and ¹⁴C of zero, with younger water of the UKA could be responsible for the different isotope content of the group of outliers within the LKA. However, as the measured values do not follow a simple binary mixing line—as indicated by the orange lines in Fig. 11—a more complex scenario involving additional end-members of mixing is apparent.

The measured ³⁶Cl/Cl ratios R for samples of the LKA range from 7 × 10⁻¹⁵ to 50 × 10⁻¹⁵ (atoms:atoms), with a median of 20 × 10⁻¹⁵. The values for the three boreholes of the UKA and of the deep borehole 037723 located to the northeast of the LKA show significantly higher ³⁶Cl/Cl ratios (Fig. 11b). For three boreholes, namely 034611, 035060 and 200248, R could not be determined because the chloride concentrations of the samples were too low (i.e. <3 mg/L). The challenge in interpreting the ³⁶Cl data on the one hand is in estimating the initial ³⁶Cl/Cl ratios, R_re, as well as the secular equilibrium ³⁶Cl/Cl ratios in the aquifer and of a potential additional source of Cl, R_se. On the other hand, values of Cl vary considerably in the samples, from below 10 to 279 mg/L, which means that the origin of the Cl added after recharge needs to be clarified. The two highest measured values of R are 161 and 176 × 10⁻¹⁵, which set a lower constraint to R_re. The uncorrected ¹⁴C ages of these samples are 3651 and 9655 years, which are too low to have a significant impact on R_re. In published studies applying the ³⁶Cl dating technique (still a rather modest number), estimates of R_re are mostly between 100 × 10⁻¹⁵ and 500 × 10⁻¹⁵. A value of 200 × 10⁻¹⁵thus appears suitable for this study. R_se is estimated at 5 × 10⁻¹⁵; this value is also similar to published data. Measured R values for the LKA are very low, and result in uncorrected apparent ages, i.e. assuming simple decay of the initial recharge concentration (Eq. (4)), of between 626 and 1433 ka. The addition of chloride could in principle be explained by variable evaporation rates in the vadose zone of the recharged water, or by addition of chloride along the flow path by dissolution of halite, mixing with brine/connate water or diffusion from adjacent aquitards. An evaporated recharge source yields a reasonable fit in a plot of R vs. the ³⁶Cl concentration, A, as shown in Fig. 11c. In this example, the average chloride concentration of LKA samples, amounting to about 130 mg/L, was chosen as input. The identified geochemical evolution down the flow path described above and the high molar Cl/Br ratios between 950 and 1400 of groundwater from the LKA, however, suggest a different source. The average linear groundwater velocity determined by Darcy’s law is 0.15 m/a applying estimates for hydraulic conductivity of 1 m/d, porosity of 0.2 and hydraulic gradient of 0.08 ‰. A value for the relative flux of chloride F of 5.9 × 10⁹ atoms/L/a (Eq. (6)), corresponding to an addition of 3.5 × 10⁻⁴ mg/L/a of Cl, leads to an increase from an initial chloride content of 20 mg/L to about 210 mg/L at a rate of 2.3 mg/L/km over the flow distance of 80 km (Fig. 12a). Two scenarios are investigated: one in which the additional Cl is added by halite dissolution (i.e. R_se = ,0) and a second in which the chloride is added by mixing with connate water or diffusion from an aquitard (R_se ≠ 0). In the first scenario, \( {\mathrm{A}}_{36_{\mathrm{Cl}}} \) is unaffected but R is “diluted” by the addition of dead chlorine (Eqs. (9) and (10)). In the second case, underground production, such as by diffusion from an adjacent aquitard, adds ³⁶Cl, and with increasing time, R in the aquifer will approach secular equilibrium R_se (Eqs. (7) and (8)). In Fig 12, the piston-flow model (PFM) refers to simple radioactive decay of recharged ³⁶Cl in an advective transport model, ADH refers to the first scenario involving added dissolved halite, and ASC to the second scenario with an added source of ³⁶Cl by diffusion, etc. Figure 12 shows the variation in Cl, R, \( {\mathrm{A}}_{36_{\mathrm{Cl}}} \) and apparent groundwater age along the presumed N–S-directed flow direction from Zambia across the study area. For the given value of R_se, the PFM only fits to measured\( {\mathrm{A}}_{36_{\mathrm{Cl}}} \), but not R. The ADH and ASC models, in contrast, produce similar curves and appear to be suited for simulation of both R and \( {A}_{36_{\mathrm{Cl}}} \). The measured ³⁶Cl/Cl ratios do not decrease significantly south of about 40 km, and instead appear to approach a possible secular equilibrium (Fig. 12b). The ASC is somewhat favored over the ADH, as it is the only model in which R does not approach zero, and \( {\mathrm{A}}_{36_{\mathrm{Cl}}} \) can rise with increasing time and travel distance due to underground production of ³⁶Cl (Fig. 12c). Figure 12d shows the apparent age of the samples calculated for the PFM according to Eq. (4) and the ASC model according to Eq. (8). Both show a general increase in groundwater age with travel distance. The ASC produces overall lower ages because aquifer production of ³⁶Cl is superimposed on radioactive decay of recharged water. Nevertheless, ages also substantially exceed 100 ka for this model. Due to the high uncertainty on both the hydraulic and isotope input parameters, however, absolute groundwater ages cannot be determined with sufficient accuracy. At this stage, the simulations therefore are aimed at providing a better understanding of the flow and transport processes, confirming hydraulic continuity and determining coarse residence time estimates. It is quite plausible that diffusion becomes a dominant process, considering the small advective flow velocities and associated large travel times. The scatter in the data, however, implies that the overall picture is more complex and that several processes, i.e. local differences in recharge concentrations, the amount of dissolution of halite, mixing and diffusion, occur simultaneously at various degrees.

The position of the two normal Linyanti and Sibbinda faults, together with (relatively sparse) drilling information and geophysical data, reveals that the position of the top of the Karoo basalts varies significantly and abruptly, representing the tectonic structures of a rift zone. Chemical, isotope and hydraulic information indicates that the LKA represents a continuous, extremely old (i.e. >100 ka), but still active hydraulic system within this section of the ORZ. Figure 13 illustrates a conceptual model of how the aquifer system in the ZR might have developed. Accordingly, the LKA originally represents a fluvio-deltaic sedimentary sequence that was deposited prior to rifting by discharge from the northern Kwando and Zambezi rivers. With the onset of rifting associated with the formation of the ORZ—presumably from the Early to Middle Pleistocene—the LKA was tectonically displaced. Rifting initially occurred along the southern Chobe Fault. This led to the formation of various large paleolake stages in the Makgadikgadi Basin. The paleolake desiccated step by step due to reduction in inflow from its northern tributaries (Moore et al. 2013). Intermittent lacustrine phases probably occurred, especially in the south of the study area, thus forming clay-rich confining layers. Repeated desiccation of sections of the paleolakes led to the accumulation of evaporates within the sedimentary sequence. This explains the presence of halite and gypsum and the predominance of sodium at the cation exchange sites. With progressive extension of the graben in the N–S direction and the establishment of the Linyanti and Sibbinda faults, the LKA was further tectonically lowered and covered by younger, overall loamier deposits that form the UKA. To the north of the Sibbinda Fault, the more permeable sands of the LKA are still preserved. This area is characterized by deep-rooting Kalahari woodland. Today, a horst-graben structure is fully developed, representing more or less the final stage of tectonic evolution. The hydrochemical and isotope results provide evidence that the N–S-directed flow is still sustained, which results in a gradual freshening of a more brackish or salty groundwater system. The LKA was probably formed during the Pleistocene and recharged under colder conditions than today, with reduced evaporation. The vegetation was possibly dominated by C4 grasses or macrophytes. Modern recharge of the LKA most likely occurs along the northern graben shoulder via preferential flow through fractures and faults. In a similar way, upward-directed groundwater discharges into the Linyanti Swamps via leakage along the Linyanti Fault. Due to the low gradients and the overall reduced direct recharge under modern climate conditions, however, these processes take place very slowly.