This article followed all ethical standards for research without direct contact with human or animal subjects.

At SGR1, the unsaturated zone is approximately 8 m deep, with variable moisture contents expressed because of the presence of low resistivity regions (3 Ω.m – 75 Ω.m) (Figure 2, Table 1). Thus, the groundwater level was expected to be at a depth of approximately 8 m – 10 m across the profile. Deep weathering is noted in the riparian zone, which then decreases in depth towards the crest because of the presence of high resistivity material (1875 Ω.m – 5484 Ω.m) at shallow depths. The alignment of resistivity 219 Ω.m – 641 Ω.m at 20 m – 25 m depth across the hillslope indicates an interface zone between the weathered material and the fractured hard rock.

At SGR2, the riparian zone shows deep weathering, which decreases upslope because of the higher resistivity values (1875 Ω.m – 5484 Ω.m) at a shallow depth towards the crest. However, shallower weathering was revealed during the drilling at the riparian boreholes as compared to the crests with depths of approximately 29 m and 43 m, respectively.

At SGR3, an unsaturated weathered zone exists at a depth of 8 m – 12 m. At the riparian zone, high resistivity material (1875 Ω.m – 5484 Ω.m) occurs at 25 m, 26 m at the midslope and approximately 35 m at the crest. A significant sodic site (duplex fine-textured soils) is present at the midslope, which explained the low resistivity values (3 Ω.m – 75 Ω.m) from there towards the crest. Drilling exposed the weathering profile to be at 25 m, 26 m and 38 m at the riparian zone, midslope and crests, respectively. The riparian boreholes had a very low T of 4.80E-03 m²/day; therefore, low permeability conditions exist in both shallow and deep groundwater systems. The midslope and crest positions had T values of 3.3 m²/day – 4.7 m²/day and 9 m²/day, respectively.

Groundwater responses along the three transects are shown in relation to the daily rainfall totals where three key rainfall sequences were identified through the API⁷ (Figure 3). This figure should be interpreted together with the data presented in Table 2, summarising the change in groundwater head during these periods.

At SGR1, the 17 m crest borehole (SGR1_C) had more than 20 m difference in groundwater head following heavy rains. Meanwhile, the 103 m borehole, drilled into the hard rock, had a lag of two months in response to rainfall sequence 1 (see Table 2) of 5 mm/day – 31 mm/day during two weeks in September 2012, with an increase in groundwater head over time (Δh/Δt) of 0.08 m. This increased by 0.09 m during the moderate second rainfall sequence of 5 mm/day – 35 mm/day over three weeks. Δh/Δt increased by 0.32 m during the December 2012–August 2013 period, whilst the Δh/Δt increased significantly following the intense rainfall of December 2012–February 2013. The riparian 61 m borehole (SGR1_R) also responds to the major rainfall sequences, similar to the 103 m crest borehole, the only difference being that the borehole in the riparian zone has a slightly higher hydraulic head. A subtle positive gradient occurs from the riparian zone towards the crest.

At SGR2_C, both boreholes track a similar water level depth. During the second rainfall sequence, the 55 m borehole Δh raises by 0.20 m over two months. The modest sequence of rainfall events (10 mm/day – 35 mm/day over three weeks) during October 2012 triggered the Δh/Δt to increase by 4.38 m and returned a month later to a static deep groundwater level. However, a significant rise of 7.39 m arose as a result of the high-intensity rainfall sequence 3 (including over 200 mm during six days in January 2013) and regressed approximately two months later to the static deep level in the hard rock (cautionary note: these abrupt changes in water level could be attributed to a borehole construction error, with conduits). At SGR2_R, groundwater in the riparian zone also showed that both boreholes track each other expressing similar water levels. Rainfall sequence 2 (5 mm/day – 35 mm/day) allowed a Δh/Δt increase by 0.12 m and 0.14 m for the 49 m and 28 m boreholes, respectively. Meanwhile, both deep and shallow levels follow a more gradual gradient during sequence 3. The shallow 28 m borehole Δh/Δt increased by 0.35 m because of the intense rain of January 2013, with a two-week response time. The close proximity of these boreholes to the stream network and the rapid rise and fall of groundwater head in the shallow riparian zone indicates a through-flow system. There is also a moderate hydraulic gradient away from the stream away from SGR2_R, as expressed by the lower groundwater level in the crest position at SGR2_C.

At SGR3_C, both boreholes follow similar trends. There is a relatively steep gradient during September 2012 because of equilibration of the borehole following its installation, thereby not representative of ambient groundwater conditions. During September 2012, the Δh/Δt rose by 0.16 m and 0.06 m for the 55 m and 34 m boreholes, respectively, with a Δh/Δt increase of 0.07 m and 0.08 m during rainfall sequence 2. Then, the 55 m and 34 m boreholes’ Δh/Δt elevated again by 0.28 m and 0.29 m, respectively, where an increased gradient is observed after the high-intensity January 2013 events. This continues for two months after which it follows a more gradual gradient. At the midslope, the 49 m borehole has a higher elevation than the 23 m and 26 m boreholes and therefore a pressure gradient exists towards the shallower boreholes. A 0.31 m, 0.28 m and 0.23 m change in Δh/Δt for the 55 m, 26 m and 23 m boreholes, respectively, was observed between December 2012 and August 2013, signifying a high recharge period. At SGR3_R, the 43 m borehole has a groundwater-level pressure greater than the 20 m borehole, indicative of a hydraulic gradient away from the stream. The Δh/Δt increased by 0.39 m and 0.41 m for the 43 m and 20 m boreholes, respectively, with a significant change in gradient during the third rainfall sequence.

Streamflows (Figure 4) were very intermittent and directly linked to rainfall events. Runoff coefficients decreased at each increment in catchment order for low-intensity sequences with API⁷’s below 50 mm. The first-order stream (SGR1) had observable base flow, having the longest flow duration, albeit with the lowest peak flow; this would result from gains from groundwater and/or hillslope contributions. The third order (SGR3) meanwhile had the greatest peak flows but the lowest flow duration, and therefore high transmission losses and close to zero base flow contributions.

Figure 5 shows the d¹⁸O end-members for groundwater and surface water from the beginning of September 2012 until the end of January 2013. There is a clear distinction from September through December 2012, with more depleted values for groundwater (low delta values) than the surface waters. This likely results from enriched precipitation arising from convective air masses early in the rain season, whereas in the large events of January 2013, the surface waters also show a depleted signature, indicative of either mixing with pre-event groundwater and/or depleted rainfall arising from frontal ocean-derived weather systems.

For the sake of brevity, only the soil moisture responses for SGR3 are presented here. The reader is also referred to detailed work on the pedotransfer functions of these soils in Van Tol et al. (2020). Nevertheless, Table 3 summarises the matric potentials for all catchment orders during the monitoring period, with a distinction for the percentage of time that a soil was close to saturation (φ = <1000 mm) or in a drying phase (φ = >1000 mm to < 10 000 mm). SG1R_C, SGR2_M and SGR3_M were the regions where soils were closest to saturation most frequently, whilst the remaining soils typically were not close to saturation during the monitoring period.

At SGR3_C, high K_unsat (Table 4) was determined for the sandy loams at the soil surface (26 mm/h – 76 mm/h at φ = 5mm). Meanwhile at SGR3_M, low K_unsat was determined at low matric potential (7 mm/h – 20 mm/h at φ = 5mm); this of course was similarly expressed with K_sat being the lowest for the entire hillslope (2 mm/h – 4 mm/h). Furthermore, a seep line exists between the SGR3_C and SGR3_M because of textural discontinuities between the two catena units. Riparian soils at SGR3_R show modest K_unsat and K_sat compared to the rest of the catena elements, in particular K_sat decreases with depth indicative of slow recharge potential. Figure 6 shows the soil matric responses for the reported summer rainfall period. The crest shows that low matric potential dominates at 100 mm and 250 mm but with relatively rapid wetting–drying cycles for most precipitation events (unfortunately the 350 mm sensor was faulty and data were omitted). Water retention in the profile at low matric potential occurred ~40% of the time resulting from the heavy rains (rainfall sequence 3). This implies that water is not retained at the crests and that lateral contributions to the lower slope occur.

The upper-midslope at SGR3_M1 typically had low matric potentials for shallow soils between 100 mm and 300 mm, whilst the deep layers (800 mm) show a considerable lag, of a month or more, before they respond to significant wetting events. These deep soils only reach low matric potentials in response to the large January and March 2013 events; 90% of the time, these deep soils remain relatively dry (matric potential >1000 mm). It is notable that a shallow soil piezometer installed at this location remained dry after the significant rain event in January 2013. However, the lower-midslope, a response to the early rains (September 2012), is expressed throughout the profile to a depth of 400 mm. The shallow soils show a rapid drying cycle compared to the lower horizons. Interestingly, the intermediate soils (depth = 300 mm) retained moisture for much longer, whereas the deep and shallow soils dried out relatively quickly in comparison.

At SGR3_R, the entire profile responds up to a depth of 400 mm to most rainfall events with similar wetting and drying cycles. The observations confirm a well-drained but relatively dry profile as matric potential is > 1000 mm at least 90% of the time.

Cumulative aET curves per catena unit, derived for the 2012–2013 hydrological year, are shown in Figure 7. Distinct water uses are apparent between the three elements of the catena at SGR1, with the greatest demand at SGR1_R (809 mm), whereas the crest and midslope had 765 mm and 784 mm, respectively. During the early rain season, between October and November, there was no real difference between the midslope and the crest region. Meanwhile, at SGR2, no significant variation between catena elements was revealed until mid-December. Then, the crest and riparian zones showed similar water use with 782.5 mm and 779 mm, respectively, and midslope the least (755 mm). Overall, there was no significant differences in aET across catena elements at SGR3, with the highest water use at the crest (767 mm), followed by the riparian (765 mm) and then the midslope (762 mm).

The summary of water balances emanating from the HYDRUS 1D simulations is shown in Table 5. At SGR1, an increase in storage of 48 mm was simulated at the crest, whereas the midslope and riparian zones were shown to be losing water. Interestingly, this was not expected as the crest has shallow soils; it can, however, be explained because of loss through modelled evapotranspiration (ET) at the midslope (470 mm) and riparian (546 mm) zones, as compared to the crest (294 mm). Whilst the crest has a significant free drainage (FD), at the hillslope scale more water is being lost through ET, indicative of large extraction by the savanna vegetation when water is not limited.

The high FD at the crest was not expected here because of the existing E horizon, which according to Van Tol et al. (2015) should indicate perched water table conditions and thus restricted flow in some areas of the lithocutanic B. This should result in more lateral flow and less FD. Most soils in the model domain only respond to high-intensity events of approximately >100 mm/day. This implies that seasons with low-to-medium intensity events will result in low or insignificant amounts of FD, and thereby low potential recharge to groundwater, especially in the lower elements of this hillslope.

At SGR2, the crest showed the most significant storage change (45 mm) followed by the midslope, whereas the riparian zone will lose water through ET and FD. Furthermore, the riparian zone of the second order is the only modelled catena element that yielded surface runoff (R). It might be recalled that shallow interflow is expected here because of flow restriction of the B horizon in these duplex soils. Runoff is therefore induced from infiltration (saturation) excess. Meanwhile, ET at the second-order hillslope was the greatest on the midslope, with evaporation from the soil surface being the dominant mechanism rather than through transpiration. However, crest and riparian zones relinquish most water through transpiration.

SGR3 catena shows the greatest storage change at the riparian zone and the crest, whilst the midslope has negative storage induced through ET losses as a result of direct soil evaporation. It also has the greatest storage (combined) overall compared to the lower order hillslopes. Transpiration was the dominant mechanism, at the crest and the riparian zones, because of the increased density of woody and herbaceous plants. Free drainage is greatest at the riparian zone.

At SGR1, the large difference in groundwater head between the shallow and deeper boreholes at the crest implies the presence of a perched water table above an aquitard in the shallow weathered zone. This is supported by the presence of a low resistivity material between the high resistance (deeper hard rock) and shallow (unconsolidated) materials near the hillslope surface. Meanwhile, water level trends are indicative of a direct piston recharge mechanism following a sequence of intense rainfall events causing water levels to rise. Although a perched groundwater level was not observed at SGR2, the rapid rise and fall of groundwater head in the shallow crest boreholes infer a direct piston recharge mechanism, whilst in the riparian zone, a rapid response and recession implies a direct preferential recharge mechanism providing a slight hydraulic gradient away from the stream itself, through the hard rock. Streamflow data support this theory as flows at SGR2 are peakflow dominated with very little baseflow observed. Meanwhile, the coalescing of streamflow and groundwater stable isotope end-members following the significant rains of January 2013, affirm this. As revealed through the ERT data and the low hydraulic gradient from the crest to riparian boreholes, a comparatively flat groundwater level across the profile was identified across the SGR3 hillslope because of the evenness of low resistivity values; meanwhile, the relatively high transmissivity at the midslope to the crest indicates permeable material in shallow (unconsolidated weathered) and deep (fractured hard rock) groundwater flow systems. The gradual two-month response in deeper groundwater levels across all three orders also points towards a regionally connected rather than local groundwater flow system (i.e. larger groundwater catchment area).

In terms of soil hydraulic characteristics, lateral subsurface interflow governs the hillslope processes at the crest towards the stream of SGR3 because of distinct K_sat differences, being high in the A horizon when compared to the underlying horizons. The midslope generally has the lowest K for the entire hillslope, this being attributed to the clay loams. These soils are therefore responsive and have overland flow features because of infiltration excess (shallow responsive), typical during exceptional events. The riparian zone displayed rapid drying cycles, indicative of high evapotranspiration and/or rapid vertical drainage, which is concomitant with the high hydraulic conductivity (K) of the soils there. Based on these characteristics, hydrological connectivity between catena elements on this hillslope is highly temporal, being intensity driven with the riparian soils being hydraulically disconnected from the third-order drainage network.

The 1-D modelling provided an indicative water balance at the sub-hillslope level (i.e. by focussing at the scale of the catena unit), stimulating further refinement of the site conceptual model, based on the following narratives: the SGR1 hillslope is dominated by interflows, but most water leaves the soil domain as FD, thereby leading to potential groundwater recharge. This was particularly so during the early rains of October 2012. However, the water balance suggests first-order hillslope connectivity with the adjacent first-order stream through lateral flow (expressed by its low FD); the SGR2 hillslope is also dominated by interflow processes, but as the midslope did not generate overland flow even after high-intensity events, this implies subsurface interflow (at the A/B horizon interface). SGR3 crest comprises interflow soils, midslope with responsive soils, whilst high FD at the riparian zone represents a groundwater recharge zone. Comparing the three hillslopes shows that significant water is removed through ET, particularly so at SGR3. The modelled responses show that there is FD only after high-intensity events. Therefore, a threshold rainfall intensity is required for certain responses such as FD and connectivity between hillslope and stream components. Overall, interflow is the dominant flowpath on these granitic hillslopes.

Although it must be recognised that this study has focussed solely on one small region of the southern granite area of the KNP (Janecke et al. 2020), and in the absence of further hydrological studies in other representative areas, one can only use this to infer an understanding of the hydrological processes in the region more broadly. In summary, therefore, outcomes of the hydrometric observations in the Southern Granite Supersite of the KNP have demonstrated, through observed streamflow, that groundwater hydrodynamics and hydrochemistry are the distinct process zones across the catchment orders. The most striking process being the apparent first-order gains to streamflow arising from hillslope contributions through interflow and a possible perched groundwater wedge, this being because of the strong gradient from shallow groundwater at the crest towards the stream. This contrasts with significant losses observed at the riparian zone of the third-order hillslope. At the scale of the first order, the soils of the valley bottom and foot-slope region cause saturation excess flows (responsive soils), leading to localised hillslope connectivity with the stream at the surface. Overall, midslope responses generally indicate that interflow dominates, this being most apparent at the third order. The variability observed in soil hydrology augments the rationale of Vaughn et al. (2015) that soil texture and resultant hydrology explained variations in woody structure in these sites. This heterogeneity has also been highlighted by Van Tol et al. (2015) as a limitation when developing hydrological models at fine scales using remote methods, notably the omission of wetland responsive soils in the valley bottom at the first order. Similar process responses observed on the first-order granite catchments have been encountered elsewhere in the lowveld region on a similar geological template, albeit under higher rainfall, where perched groundwater contributions to a valley-bottom wetland also led to sustained streamflows at that scale (Riddell et al. 2013). Observations suggested that the third-order stream losses were primarily evapotranspiration driven, with streamflow losses as a result of diffuse flow through the alluvial sediments and thence returned to the atmosphere through the riparian vegetation. Interestingly, the first-order catchment appears to have a greater ET overall as compared to the higher orders, likely explained by accessible vadose zone storage and by augmented flow from the perched seasonal shallow groundwater to the valley bottom. Because of low T-values, the inclining trend of groundwater levels during and following the wet season indicate that the hillslopes act as groundwater storage areas, which in turn sustains a hydraulic head contributing to the regional groundwater flow during the dry season, and thereby providing some baseflow to the perennial rivers of the KNP.

This study has revealed certain important hydro-geological and hydro-ecological interactions relevant also for areas adjacent to the KNP under altered land uses, for instance, the important role that these low-order systems have in catchment processes, notably that streambeds are preferential recharge points. Similar findings of stream channel recharge have been discussed elsewhere in ephemeral river hydro-ecology (Larned et al. 2010). These ephemeral catchments are critical for regional groundwater recharge processes, and therefore should provide resilience to groundwater management (Kolusu et al. 2019), now threatened because of the growing anthropogenic pressures such as sand extraction in similar drainages outside of the protected area. The low transmissivity of the southern granites and the steady rise in the water level of the weathered and hard rock aquifers could be a geo-hydraulic boundary, characteristic of these ephemeral landscapes. Du Toit and Verster (2017, unpublished) made similar observations at Matlari in the Biyamiti sub-catchment of the Crocodile basin within KNP. It is therefore likely that hydraulic conditions at these watershed areas create a hydraulic gradient through the dry season to lower altitude (high-order) catchments. Implicit in recent studies is that that the perennial rivers of the lowveld at their lower reaches are maintained by regional groundwater contributions (e.g. Gokool et al. 2019; Petersen 2012; Saraiva-Okello et al. 2018); this information becomes important for catchment planning purposes through protection of groundwater recharge zones.

The trade-off between recharge and actual evapotranspiration suggests that the groundwater recharge process only follows a threshold ‘episodic’ rainfall sequence, as noted by Van Wyk, Van Tonder and Vermeulen (2011), and in the case of KNP this is typically in the region of 100 mm/24 h. Should rainfall of this intensity not occur, then direct recharge to the aquifer will not take place and the water is otherwise evaporated from the soil matrix. Although general climate forecasts for the region show that the mean annual precipitation will remain the same even though it is likely to arrive in more intense events (e.g. Cuthbert et al. 2019), this in turn could be beneficial as a groundwater recharge mechanism. Meanwhile, this effect may be limited dependent on whether vegetation density increases as a result and the thickness of the vadose zone, as has been shown in the cerrado (Oliveira et al. 2017). This certainly needs to be considered given the significant interannual variability of tree–grass biomass, influenced through wet and dry years, identified as a key determinant of total evaporation at landscape scale from the Skukuza Flux Tower site within the same region of the KNP (Andreu et al. 2019; Williams et al. 2009). Alternatively, an increase in the rainfall intensity could lead to an increase in woody cover and hence aET; this has already been identified as a possible future scenario in other savanna systems (Tietjen 2016).