BRAIDED-RIVER ARCHITECTURE OF THE TRIASSIC SWARTBERG MEMBER, KATBERG FORMATION, SOUTH AFRICA: ASSESSING AGE, FLUVIAL STYLE, AND PALEOCLIMATE AFTER THE END-PERMIAN EXTINCTION

The end of the Permian marked the most severe crisis in the Phanerozoic history of life, when more than 90% of marine species became extinct during a short-lived biosphere collapse (Benton and Newell 2014; Zhu et al. 2019). On land, the extinction period was marked by a profound loss of vertebrate families (Benton and Newell 2014), a decrease in invertebrate bioturbation (Buatois et al. 2021), and a decline in plant species richness (Looy et al. 2001; McElwain and Punyasena 2007). An enhanced flux of sediment and acidified soil material has also been reported (Algeo and Twitchett 2010; Sephton et al. 2015). The currently favored killing model highlights the Siberian and Emeishan large igneous provinces, with the main phase of extinction possibly triggered by the emplacement of extensive sills (Bond and Grasby 2017; Burgess et al. 2017; MacLeod et al. 2017). Multiproxy datasets based on geochemistry, sedimentology, and paleontology document the destabilization of the carbon cycle and Earth system in the marine and terrestrial spheres. This has been attributed to a combination of warming, release of greenhouse gases, anoxia, acidification of land and oceans, ozone depletion, and toxic-metal poisoning (Algeo et al. 2011).

For the terrestrial realm, changes in fluvial style across the extinction level provide important evidence for ecosystem alteration (e.g., Newell et al. 1999; Arche and López-Gómez 2005; Davies et al. 2010; Midwinter et al. 2017; Zhu et al. 2019). In the Karoo Basin of South Africa (Fig. 1A), Permian–Triassic outcrop belts with abundant vertebrate fossils have played a central role in evaluating trends in fluvial style and paleoclimate (Smith 1995; Abdala et al. 2006; Viglietti et al. 2016; Smith et al. 2020; Retallack 2021) (Table 1). A key post-extinction stratigraphic unit is the Triassic Katberg Formation (Fig. 1B), characterized by its relatively arenaceous character when compared with the adjacent Balfour and Burgersdorp formations (Groenewald 1996). Neveling (2004) recognized an upward-thickening and -coarsening trend in the formation, culminating in stacked cliff-forming sandstones in the middle to upper part. He proposed the term Swartberg member for this sandstone-rich interval, based on the cliff that caps Swartberg Koppie (hill) on the Free State farms of Bethel, Heldenmoed, and Donald 207 (Fig. 1C; Facies E of Smith and Botha-Brink 2014). Strata below the Swartberg member are referred to informally as the lower Katberg Formation, and the thickness of sandstone units and average grainsize decrease above the Swartberg member. Due to its thickness and prominence in outcrops, the Swartberg member has been considered a useful lithostratigraphic marker to correlate localities across the basin (Botha et al. 2020).

The sandbed rivers that laid down the Katberg Formation have been linked to sediment transport under a relatively warm and dry climate (Stavrakis 1980; Hiller and Stavrakis 1984; Smith and Botha-Brink 2014). Several authors have attributed this fluvial style to aridification, vegetation dieback, and enhanced landscape erosion associated with the end-Permian crisis (Ward et al. 2000, 2005; Viglietti et al. 2018; Botha et al. 2020). These and other features of the formation have been extrapolated worldwide as characteristic of fluvial regimes associated with the extinction (Zhu et al. 2019). However, a rapid and unidirectional drying trend in the Karoo Basin has been disputed (Neveling et al. 2016a, 2016b; Li et al. 2017; Gastaldo et al. 2014, 2019a, 2019b, 2020a) and studies elsewhere have documented little change in paleoenvironment across the extinction level (Fielding et al. 2019, 2021; Ji et al. 2023).

Two problems exist with using the Katberg Formation to infer ecosystem dynamics. The first problem is the paucity of accessible outcrop and a consequent lack of detailed architectural analysis beyond the proximal outcrop belt in the Eastern Cape and northeast Free State provinces (Groenewald 1996). The formation crops out in low escarpments with exposures in gullies (dongas) and on largely inaccessible mountain tops. Consequently, the fluvial systems have been characterized mainly in roadcuts where the strata are limited in thickness and extent, yielding measured sections but few extensive lateral profiles (Neveling 2004; Gastaldo et al. 2021). Studies to date have broadly characterized the Katberg Formation as composed of thick, linear and multi-storied channel bodies (Groenewald 1996; Neveling 2004) separated by mudrock intervals. The latter yield evidence of landscape stasis (Gastaldo and Demko 2011) as evidenced by paleosols (Gastaldo and Rolerson 2008; Pace et al. 2009) and the preservation of vertebrate burrows (Bordy et al. 2010; Bordy and Krummeck 2016).

To provide a more comprehensive assessment of fluvial architecture and, by inference, paleoclimate for the Swartberg member at Swartberg Koppie, we employ an unmanned aerial vehicle (UAV, or drone) to generate 2-D photomosaics and 3-D models for the largely inaccessible cliff. This locality has yielded some 80% of the vertebrate fossils used to interpret a terrestrial end-Permian extinction event (Gastaldo et al. 2019a) and is one of eight classic localities where a changeover in fluvial style has been reported (Ward et al. 2000, 2005).

The second issue concerns the age of the Katberg Formation. Several workers (Neveling 2004; Botha and Smith 2006; Smith and Botha-Brink 2014, their Fig. 1) assigned the formation at Swartberg Koppie to the Induan stage of the Early Triassic (Fig. 1D). This assessment was based on the presence of vertebrates placed in the Lystrosaurus declivis Assemblage Zone (AZ), which has traditionally been considered to represent the post-extinction biosphere (but see Gastaldo et al. 2021). To date, geochronometric constraints are lacking, although Gastaldo et al. (2018) reported that the unit lies in a normal-polarity magnetozone. We use laser ablation inductively coupled plasma mass spectrometry (U-Pb LA-ICPMS; Davis and Rochín-Bañaga 2021) to date calcite cements of nodules from paleosols and pedogenic nodule conglomerates (Pace et al. 2009) in the hillslope more than 70 m below the base of the Swartberg member.

The Karoo Basin developed as a foreland basin associated with the Cape Fold Belt (Hälbich 1992; Gresse et al. 1992; Catuneanu et al. 1998; Hancox et al. 2002; Johnson et al. 2006). The fold-and-thrust belt borders the southern and western coast of South Africa and forms part of an extensive orogenic belt from Argentina to Antarctica, linked to subduction of the paleo-Pacific Plate. Extending north from the Cape Fold Belt onto the Kaapvaal Craton, the Karoo Basin has an area of 700,000 km² and contains some 12.5 km of Carboniferous to Jurassic strata of the Karoo Supergroup (Fig. 1B), the greater part of which accumulated in a continental interior. The basin was situated at about 50° S by the end of the Triassic (Seton et al. 2012). Carboniferous to Early Permian glacial deposits of the Dwyka Group (Isbell et al. 2008) were followed by marine turbidites (Ecca Group: Sixsmith et al. 2004; Cairncross et al. 2005), deltaic deposits (Waterford Formation of the Ecca Group: Rubidge et al. 2012), and fluvial endorheic strata of the Beaufort Group (Gulliford et al. 2017). Beaufort fluvial styles in the eastern part of the basin (east of 24°) include Middle Permian meandering systems of the Balfour Formation (Smith 1987, 1995), a fluvial style that is considered to have dominated the landscape until latest Permian time (Barberskrans Member). Thereafter, the Elandsberg and Palingkloof members of the Balfour Formation were reported to mark a changeover to a low-sinuosity system (Smith 1995; Ward et al. 2000, 2005; Smith and Botha-Brink 2014; Viglietti et al. 2018), passing up into braided systems of the Katberg Formation of the Tarkastad Subgroup. The younger Triassic succession includes the high-sinuosity fluvial system of the Burgersdorp Formation (Hancox 1998) and the braided-fluvial Molteno Formation of the Stormberg Group (Turner 1983).

In the southwestern part of the basin near Beaufort West, the Lower Beaufort Group (upper part of the Abrahamskraal Formation and lower part of the Teekloof Formation) have been attributed to a range of high- and low-sinuosity fluvial systems, with no single style predominant but with an upward transition to higher-sinuosity systems (Wilson et al. 2014; Gulliford et al. 2014; Bordy and Paiva 2021). The paleoclimate has been considered relatively arid for much of the Permian succession, based on evidence from sedimentary structures, paleosols, and assessments of fluvial style (Table 1).

The Katberg Formation is about 1000 m thick in coastal exposures at East London, with coarser pebbly sandstone. The rock unit thins and fines northward to a few hundreds and then tens of meters over a distance of several hundred kilometers, consisting largely of fine- to very fine-grained sandstone in the Bethulie area (Neveling 2004; Gastaldo et al. 2019a). The base of the lower member is defined at a sharp increase in the ratio of sandstone to mudstone above the Palingkloof Member (Groenewald 1996), thickening upwards into the thick sandstones of the Swartberg member above. Sandstone-dominated bodies at the level of the Swartberg member can be traced for kilometers in cliff faces along the strike of mountain ridges. This exposure pattern results from the resistance of the sandstone bodies, with the adjacent valleys marking fine-grained, interfluvial units (now eroded) that were formerly laterally equivalent at ridge-top level. Exposures along cliff faces and in valley floors represent variously oriented longitudinal and transverse profiles of the ancient river systems (Sasajima et al. 2015; Gastaldo et al. 2019b). Jurassic dolerite intrusions cut the outcrops locally, associated with continental breakup and the emplacement of the Karoo Large Igneous Province.

Depositional packages and unconformities in the Karoo Basin have been linked to episodes of thrust loading and relative quiescence in the Cape Fold Belt (Turner 1983; Hiller and Stavrakis 1984; Catuneanu et al. 1998; Catuneanu and Elango 2001; Viglietti et al. 2017). Earlier tectonic analysis suggested that hinterland deformation and flexural subsidence was active from late Carboniferous to at least the end of the Triassic (Hälbich 1992; Gresse et al. 1992; Bordy et al. 2004; Catuneanu et al. 2005). Recent research using ⁴⁰Ar/³⁹Ar muscovite plateau ages has indicated a major phase of deformation between about 253 and 250 Ma, across the Permian–Triassic boundary and possibly overprinting earlier events (Blewett and Phillips 2016; Hansma et al. 2016; Blewett et al. 2019). However, the precise correlation of tectonic events in the Cape Fold Belt with rock units in the Karoo Basin remains uncertain (Bordy and Paiva 2021).

Much of the Permian and Triassic succession was probably deposited on large distributive fluvial systems (DFS, or megafans; Weissmann et al. 2010) that built northwards from the Cape Fold Belt, generating northward and locally radiating paleoflow patterns (Turner 1983; Hiller and Stavrakis 1984; Gulliford et al. 2014; Wilson et al. 2014; Bordy and Paiva 2021). Consistent with a DFS model, Gulliford et al. (2014) noted similar paleocurrents in early frontal splays (precursor drainage) and younger channel belts (main prograding system) in the Abrahamskraal and Teekloof formations. The Katberg Formation has been attributed to a pulse of thrusting and the progradation of a clastic wedge with a large sandy braided system (Smith 1995; Catuneanu et al. 1998, 2005), although Ward et al. (2000) later suggested a predominantly climatic interpretation for the sandy fluvial systems. For megafans in general and those of the Karoo Basin in particular, resolving the effects of tectonics and climate has proved challenging (Bordy and Paiva 2021).

The Swartberg Member is named for an exposure capping an escarpment on Bethel 763 Farm, Free State Province, between the intersection of the R207 and the Caledon River, some 30 km northeast of Bethulie (S 30.419256°, E 026.259594°; Fig. 1C). Exposure is restricted to cliff faces with nearly horizontal strata along Swartberg Koppie, which rises > 200 m from the adjacent valley to an elevation of ∼ 1565 m. The cliff comprises two long, steep faces that are broadly at right angles and curve into each other at an outcrop corner. The South–North (S–N) face is about 1 km long, with 492 m of nearly continuous cliff exposure at its northern end (Fig. 2). The East–West (E–W) face is 675 m long (Fig. 3), although only ∼ 100 m of the face that adjoins the S–N face is well exposed. Areas of overlap between photomosaics of the two faces are marked on the composite drone images of Figures 2 and 3, which show the locations of photos and interpreted panels. The outcrop ends abruptly at the top of Swartberg Koppie.

Despite playing a central role in an end-Permian terrestrial-extinction model, architectural features at Swartberg Koppie have not been described in detail. This is largely due to the sheer nature of the outcrop, which is accessible for only 3 m at the base. Dolerite sills and dikes a few meters wide run obliquely and discontinuously through the S–N face (but not the E–W face). Strata are locally displaced but can be correlated across the intrusion, and some areas appear to have been thermally affected.

Two UAVs were deployed to obtain orthogonal images of the cliff, a DJI Phantom 3 Pro and a DJI Mavik Pro. Waypoint paths were set before each flight, with geo-referenced images taken every two seconds to ensure > 80% overlap. Images were captured in JPEG format to manage the large amount of field data, with preliminary models constructed on site between successive flights to ensure that sufficient images had been acquired. The GPS software tagged each image with metadata on longitude, latitude, and relative altitude above ground level, for use in building 2-D and 3-D outcrop models. Altitude data in the metadata set vary because they are referenced from the take-off location and are calculated using the UAV’s barometer and GPS system.

Digital images were imported into Agisoft Metashape Pro (http://www.agisoft.com/; Supplemental Materials S2, S3), generating 2-D photomosaics and 3-D virtual outcrop models through structure-from-motion. The workflow included initial generation of a sparse point cloud (Stonesifer et al. 2018), followed by a dense point cloud, construction of meshes, and eventual texturing in RGB space. The final model textures used four JPEG tiles with 8192 × 8192 pixels, generating sufficient resolution to recognize key elements of stratal geometry, which can be further viewed from different perspectives by zooming and rotating the 3-D model.

Metashape models were exported to interactive PDF files (Supplemental Material S2) that were imported into virtual-reality tools (Stonesifer et al. 2018). Orthogonal views of the two cliff faces were captured with the highest resolution available. A 10-m scale bar was added to each model using the built-in measurement tool in Metashape, and measurements presented in the results section originate from orthogonal screen shots that use these scale bars. Architectural analysis is based primarily on the S–N face using the digital model, supplemented by close-up images from the original flight. Due, in part, to insufficient image overlap, the stratal tracing along the E–W face only partly reflects the original outcrop, and our evaluation of this area has relied on the original drone images and photomosaics. Compared to 2-D photo-stitching software that may require considerable manual effort, the automated system of Agisoft Metashape Pro overcomes the challenge in identifying accurate control points caused by low color contrast in the Karoo setting.

We use a conceptual Landscape Systems approach in characterizing the Swartberg member, based on reconstructions of the alluvial landscape as it might have appeared in a bird’s-eye view (Gibling et al., in press; see also Ghinassi and Ielpi 2017; Nesbit et al. 2018). Using detailed tracings of the cliff face, we analyze facies and facies successions, geomorphic units, and fluvial architecture, inferring the paleoflow direction in this part of the basin and estimating paleodischarge. The interpretations draw on studies of modern sandy and gravelly braided rivers for which subsurface profiles are available from ground-penetrating radar (GPR), using radar facies to link surface morphology to preserved channel-bar complexes. Fine-scale sedimentological features are typically below the resolution of the UAV images, limiting an assessment of depositional processes, but the orthogonal photomosaics and 3-D models provide an exceptional opportunity to establish larger-scale architectural features.

Analysis centered around the manual tracing of stratal surfaces on the S–N and E–W 2-D photomosaics, carried out at high resolution using CorelDRAW 2019^© and yielding an estimated accuracy of 0.1 m. All dimensions recorded in the text are apparent, measured on the photomosaics. Special attention was given to the tracing of irregular and discontinuous surfaces, without assuming planar geometry or continuity across the outcrop. Examination of the surfaces focused on their erosional or non-erosional nature, truncation of older by younger surfaces, surface style (sub-planar to wavy, concave- or convex-up), and apparent inclination. Varied line weights were later applied to the surfaces to denote their relative importance in the alluvial architecture, as judged by their extent, prominence of erosion, and the lithological contrast across them (e.g., sandstone on mudstone remnants). Although no surface hierarchy was assumed a priori, the surfaces were later interpreted in relation to the sediment bodies that they bound (e.g., channel bases, bar tops, increments of bar accretion). Unexposed, poorly exposed, and vegetated areas were mapped to indicate areas of uncertainty, and major fractures and fallen blocks were noted to ensure that structural artifacts were not misidentified as sedimentary features.

A standard set of facies (recurring sediment types) and facies successions were identified in the basal, accessible few meters of the Swartberg member outcrop (–30.419108°, 26.259319°), as well as from bedset appearance at high resolution on the drone images. Facies codes are based on, or modified from, Miall (1978). Resolution was inadequate to note bioturbation, small fossils, and subtle sedimentary features. Subsequently, we identified what we term remote outcrop facies (ROFs) in the inaccessible cliff face. These facies are analogous in some respects to GPR radar facies for modern rivers, which are identified on the basis of reflector amplitude, frequency, and continuity, generating mappable 3-D units with reflections that differ in character from those of adjacent units (Huggenberger 1993). Remote outcrop facies are characterized by their sedimentary structures (where known), the style and continuity of stratal surfaces, the apparent inclination of the strata, and the resistant or recessive nature of the outcrop. Their identification was hampered where rock falls had left sheer, unweathered surfaces with cryptic features. Modern GPR studies are usually conducted across known surface morphology, testing stratification along and across flow directions to assess accretion directions (upflow, downflow, lateral). However, as paleocurrent measurements could not be collected for the inaccessible cliff, stratification styles could not be assessed along pre-determined directions.

For seven rectangular 2-D panels in the S–N face, we estimated the areal proportions of the remote outcrop facies. Facies areas were outlined and color-coded in CorelDraw and then exported as JPEG images. We then calculated the relative pixel fractions in the panels using an R script. The panels represent ∼ 31% of the channel-belt outcrop, excluding dikes. Much of the remaining outcrop area was unsuitable for analysis due to small or tapering exposures, partial vegetation cover, and distortion around dikes and sills; we estimate that our analysis covers about half of the suitable exposure. Several studies in modern rivers have employed a similar, quantitative analysis of radar facies (e.g., Horn et al. 2012a).

The remote outcrop facies were attributed to one or more geomorphic units, comparable to architectural elements (Allen 1983; Miall 1996), macroscale forms (Bridge 1993), and geomorphic elements (Horn et al. 2012a). The geomorphic units were assigned lower-case codes. In modern river systems, geomorphic units are defined as mappable landforms with distinctive geometry, sedimentary features, and bounding surfaces (Fryirs and Brierley 2018), and they constitute the “building blocks” of alluvial systems, fitting together to generate a reach-scale sediment body (Wheaton et al. 2015; Fryirs and Brierley 2018). The identification of units is conditioned by five characteristics: vertical position or formative flow stage relative to the channel bed; planform; unit type (e.g., bars, depressions); type and density of associated vegetation; and bed-material texture and other sedimentological features. Most of these aspects must be inferred for the present study, commonly leading to multiple interpretations. Terminology for architectural components is based on Miall (1996), Holbrook (2001), Gibling (2006), and Gibling et al. (in press). Broad criteria for identifying fluvial styles are set out in Table 2, with selected references for modern analogues.

Gastaldo et al. (2019a, their Fig. 5) developed a stratigraphic framework across the Bethel and Heldenmoed farms, with eight measured sections of various thickness and paleomagnetic data available for three sections. The sections were correlated using prominent sandstone benches exposed along the S–N and E–W mountainsides in the lower part of Swartberg Koppie, waypointed every 40 to 80 m depending on physical conditions (WSG84 standard; Supplemental Material S1). Two additional sections were measured in 3/2019 and 8/2022 to determine the thickness of the succession down to the base of the Swartberg member (Supplemental Material S1). Measurements were made with a Jacob’s staff equipped with an Abney level (see methods in Gastaldo et al. 2021). During both field seasons, representative calcite-cemented pedogenic nodules from available paleosols and intraformational conglomerate were collected.

Calcite-cemented pedogenic nodules were sectioned and polished to generate a flat surface. Individual nodule surfaces were analyzed for U-Pb isotopic composition at the University of Toronto using an Agilent 7900 ICPMS and an NWR193 excimer laser system. U-Pb data (²³⁸U, ²⁰⁶Pb, and ²⁰⁷Pb) were collected using scan-lines over individual nodules, resulting in hundreds of cycle data per line following the methodology of Davis and Rochín-Bañaga (2021). Analytical data are available in Supplemental Materials S4 to S8.

Each scan-line analysis, 4–5 mm long, was pre-ablated at a fast scan rate, 200 μm/s, and high frequency, 20 Hz, using a larger diameter beam to remove surface contamination. U-Pb line analyses were conducted with a laser frequency of 193 nm, fluence of ∼ 4.5 J/cm², diameter beam of 120 µm, and frequency of 10 Hz at a rate of 15 μm/s. Baselines were accumulated on each scan-line for 20 seconds before opening the laser. LA-ICPMS parameters for U-Pb analyses are found in Supplemental Material S7.

Two standards, the Glass standard NIST612 and Duff Brown Tank, were used to correct for mass-and-oxide elemental fractionation bias. The uncertainty of our ²³⁸U–²⁰⁶Pb ages includes measurement errors in mass signals from the sample, NIST612, and Duff Brown Tank calcite-standard measurements. The Duff Brown Tank standard was used as a matrix-matched reference. The age of the Duff Brown Tank was measured here at 64.1 ± 1.4 Ma (2s; MSWD = 0.87) which agrees with its true age of 64.04 ± 0.67 Ma by isotope dilution (Hill et al. 2016). During each analytical session, another calcite material with a known age, determined by isotope dilution (Walnut Canyon standard of 254.4 ± 6.4 Ma; Roberts et al. 2017), was measured as a secondary standard to check on accuracy; the Walnut Canyon standard gave an age of 255 ± 3 Ma (2s; MSWD = 2.3).

U-Pb regression was carried out using a Bayesian statistical program (UtilChron) that regresses mass-count data in a 3-D signal space (Davis and Rochín-Bañaga 2021). Calcite minerals usually contain significant common Pb, as well as radiogenic Pb. Hence, age determinations are based on defining a mixing line in isotope-ratio space. Intersection with the concordia curve (showing radioactive equilibrium at a given age) provides the age of the radiogenic component, and its Y (²⁰⁷Pb/²⁰⁶Pb) axis intercept on the Tera-Wasserburg plot gives the isotopic composition of the initial common Pb component (denoted PbC). Mixing lines are defined using a data set, or multiple data sets, from samples with similar age and PbC. Data were processed as single or small-number average measurement cycles to exploit the maximum U over common Pb heterogenicity in the samples, giving the maximum spread along the mixing line. U-Pb data are plotted and regressed using Tera-Wasserburg (inverse) concordia diagrams, where the age represents the intersection of the best-fit line with the concordia curve and PbC is given by its Y axis intercept. Data sets consist of hundreds of measurements; as such, error ellipses are not shown but are taken into account in the calculations.

The Swartberg member is 45 m thick (Fig. 4A) and contains six large-scale units that can be identified in both S–N and E–W faces (Figs. 2B, 3B). A basal mudstone-dominated unit with lensoid sandstone bodies is only partially exposed, overlying poorly exposed reddish gray and greenish gray mudstone in the slopes below (Supplemental Material S1). Above, three thicker units consist predominantly of sandstone with minor fine-grained lenses, separated by two additional mudstone-dominated intervals. The upper surface of the topmost thick sandstone unit forms the top of the outcrop at ∼ 1565 m. The Swartberg member may attain a greater overall thickness, but vegetation obscures the extent of the resistant benches in Google Earth images of the southwest side of the koppie, which was inaccessible during field work.

The three mudstone-rich intervals (Fine Intervals A, B, and C) comprise small fluvial channel bodies and green-gray laminated fines, interpreted as fluvio-lacustrine deposits. The three sandstone-dominated units (A to C) are interpreted as channel belts, occupying shallow valleys, and defined in preserved strata as the amalgamated deposits of a channel system that developed through time. The channel belts are interpreted as braided-river systems based on cross sections with inferred braid bars and adjacent channels (Bridge 2003). For modern braided rivers, Lunt and Bridge (2004) defined the channel belt as the extent of flow during bankfull flood events (the “active channel belt” of Horn et al. 2012a). This definition covers active and partially abandoned channels, as well as simple and compound bars and bar assemblages.

The channel belts each comprise three to seven channel bodies that are vertically stacked or locally inset, underlain by prominent erosional surfaces. Channel bodies are also referred to as stories (Friend et al. 1979). The term channel fill is used here to denote fluvial-channel sediments in general and is not restricted to the filling of a discrete, unmodified channel cut. The channel bodies contain numerous bars, defined as within-channel depositional features with a length proportional to the width of the adjacent channel and a height comparable to the bankfull channel depth (Lunt and Bridge 2004).

Although field measurements were not possible, several lines of evidence collectively indicate a general northward paleoflow direction along the S–N face (roughly longitudinal section), emerging in the E–W face (roughly transverse section). Channel bodies are mainly planar and continuous in the S–N face but are more lensoid in the E–W face (Figs. 2, 3). Inclined bar-accretion surfaces show apparent northward dips in 29 instances and apparent southward dips in three instances, in accord with the predominantly downstream dip of accretion surfaces in many GPR profiles of sandbed rivers (e.g., Calamus River: Bridge et al. 1998). Where trough cross-sets can be discerned, apparent dips are northward. For the outcrop corner (Figs. 4B, C), rotating the 3-D model to examine the strata from various angles confirmed the north–south orientation of linear, scour-based sediment bodies and the northward dip of cross-beds and larger-scale accretion sets. Additionally, as discussed below, large-scale bedsets in Channel Bodies 9 to 11 show northward depositional dips in the S–N face, and channel belts and bodies taper out westward in the E–W face, indicative of orientations along- and across-paleoflow, respectively.

Ten sedimentary facies were identified at the outcrop base and in drone images (Table 3; Figs. 5, 6). A similar range of facies was identified in other studies of the Katberg Formation (Stavrakis 1980; Groenewald 1996; Neveling 2004) and other formations in the Karoo Basin (e.g., Gulliford et al. 2014; Wilson et al. 2014; Bordy and Paiva 2021). The coarser beds are fine- to very fine-grained sand, with admixtures of silt but relatively little coarser-grained sediment.

Thin units with pebble-size mudstone intraclasts are identified as gravel lag (Gl), overlying prominent erosional surfaces in the lower part of Channel Body A (Figs. 5A, C, 6A). Trough cross-bedded sandstone (St) is characterized by grouped trough forms that range in relief from 0.2 to 0.9 m, averaging 0.4 m (n = 25); individual cross-beds were difficult to resolve in the troughs, but the strata are interpreted as cosets of trough cross-beds. Cross-sets of smaller size were difficult to resolve. Trough cross-beds are typically preserved to about one third of the original dune height (Leclair and Bridge 2001), and many of the original dunes may have been > 1 m high. Planar cross-bedded sandstone (Sp) comprises isolated or grouped cross sets that rest on planar surfaces. Individual cross-sets are up to 2.2 m thick, average 0.9 m (n = 21), with apparent downdip extents of several meters (Fig. 6B) to 50 m. Ripple cross-laminated sandstone (Sr) was difficult to identify and may be more abundant than our data implies.

Plane-bedded sandstone (Sh) comprises sandstone bedsets with even lamination (Fig. 5D). Scour-and-mound sandstone (Sac; Walker and Holbrook 2022) is present in the topmost beds of Channel Belt C, where a series of sandstone mounds, with form-concordant internal stratification and quasi-regular spacing, is locally associated with facies Sh (Fig. 6B). The relief of the mounds ranges from 0.6 to 1.0 m, averaging 0.9 m (n = 11). These features are provisionally interpreted as high-flow-strength bedforms, probably antidunes, formed under shallow, supercritical flows.

Massive sandstone (Sm) lacks obvious stratification. Most occurrences are interbedded with mudstone and comprise discrete layers with abrupt bases and tops, although some fine upward from sand to mud. Thickening/thinning rhythms are present in places (Fig. 5B). A thicker bed of facies Sm occupies a shallow scour at the base of the cliff (Fig. 5A). Scour-fill sandstone (Ss) overlies solitary, concave-up erosional surfaces of decimeter scale (Fig. 6A–C); their solitary nature distinguishes them from grouped trough forms of facies St.

Laminated mudstone (Fl) is commonly interstratified with massive sandstone and is well exposed at the outcrop base (Fig. 5A, B). Massive mudstone (Fm) is present locally as units > 1 m thick, commonly within scour hollows.

Meter-scale facies successions in the basal strata commence with erosional surfaces overlain by facies Gl, St, and Sh with facies Sr locally, capped abruptly by lensoid dm-scale bedsets of facies Fl, Fm, and Sm (Fig. 5C). They are interpreted as shallowing-upward successions in small channels.

Seven remote outcrop facies (ROF, Table 4) are illustrated at a comparable scale in Figure 7, provisionally linked to geomorphic units (Figs. 8–11) through the identification of key surfaces (Table 5). Dips and horizontal distances reported below are apparent and are not corrected for paleoflow. Thickness data for some ROF are illustrated in Figure 12A.

ROF 1 is present in all the channel belts as horizontal to slightly inclined bedsets up to 5 m thick that extend for tens of meters (Fig. 11A). Bed surfaces are variably planar, undular, trough and wavy, and the strata comprise trough cross-stratification, planar stratification, and ripple cross-lamination (St, Sh, Sr). Scour-and-mound stratification (Sac) is present in ROF 1 at the top of Channel Belt C.

The facies is attributed to sand sheets (ss), generated from 3-D dunes with contributions from other bedforms.

ROF 2 comprises flat-lying to gently inclined bedsets, up to 3.2 m thick, that onlap and overlap an underlying erosional surface, and are prominent in all channel belts (Fig. 11B). The surfaces range from strongly concave-up to gentle scours to flat-lying surfaces with scooped segments. Bedforms are difficult to identify but trough cross-stratification is visible locally.

The main subset (2A) is attributed to channel scours (cs) that represent primary or secondary channels or confluence scours where braid channels join. A minor subset (2B) is incised into the tops of mounded macroforms and is attributed to bartop channels or bartop hollows (btc).

ROF 3 is characterized by dipping bedsets with up to 3.2 m of relief on individual surfaces and apparent dips of up to 15° (Fig. 9A–C). The surfaces appear irregular in cross section with some flat-lying segments, and they have modest apparent extents of up to 10 m, passing laterally into sand sheets (Fig. 9A, C).

The main subset (3A) is prominent in Channel Belt A and is attributed to bank-attached bars that may include point bars (pb) in high-sinuosity channels and alternate bars (ab) or side bars (sb) in low-sinuosity channels. Deciding between these options would require a fuller 3-D or planform exposure to explore channel sinuosity and bar disposition. Flat-lying segments are attributed to benches (bc) on the bar surfaces, developed at levels below the tops of the bordering scours. A minor subset (3B) comprises thin bedsets with steeper dips of up to 25°, mantling the upper parts of some scoured margins and passing downdip into facies 3A. The bedsets are attributed to accretionary banks (ab). The bar, bench, and bank deposits record deposition in stably positioned channels.

ROF 4 comprises solitary or stacked planar cross-sets (facies Sp) up to 1.8 m thick, averaging 0.9 m, and is present in all channel belts. The cross-sets have apparent dips of up to 21° and apparent extents of up to 10 m (Fig. 8A, B), and they are considerably thicker than associated trough cross-beds (facies St). Most of the cross-sets rest on planar surfaces, but a few overlie scours. An especially prominent example at the top of Channel Belt C (Fig. 11A) shows a set of inclined strata 2.2 m thick and 50 m in apparent extent, with reactivation surfaces on the upper parts of some foresets.

The facies is interpreted as the slip-face component of bars, probably mainly unit bars (ub), which are widespread in modern sandbed rivers.

Mounded macroforms of ROF 5 are prominent in all the channel belts and have no apparent connection with channel margins (e.g., Figs. 8B, 10A). The mounds rest on flat-lying surfaces or are built over scours, aggrading across the scour margins, and they are probably linear with an along-stream orientation (Fig. 4C). Mound stratification is flat-lying or mounded in concordance with the top surface, and some contain inclined sets with dips of up to 10°. The mounds average 1.7 m thick (73 examples) but are up to 2.8 m thick, with apparent extents of up to 10 m. In one case, a planar cross-set (ROF 4) stalled on the upflow side of a mound, generating a composite form 2.2 m thick (Fig. 7). In another case, a mound built over the top of a planar cross-set (Fig. 7).

ROF 5 is interpreted as the mounded cores of mid-channel bars (mcb), probably a component of compound bars. Some mounds may be remnants of sediment sheets that were scoured on two sides, but many modern river bars experience erosion along their margins.

ROF 6 was identified in all three channel belts as mounded or flat-topped accumulations associated with features interpreted as upright trees or log stacks (Figs. 10B, 11B–E). The vegetation elements range from 0.4 to 2.5 m high (average 0.4 m) and taper upwards. Although at the limit of resolution in the drone images, one well exposed upright tree contains sediment layers that built up in a hollowed trunk (Fig. 11E). Sediment accumulations up to 2.5 m thick and 4 m in apparent extent built up over scours that bordered or formed above the vegetation elements. In some instances, inclined strata are present on one side of the vegetation element or dip inwards on both sides (centroclinal cross-stratification: Underwood and Lambert 1974). The vegetation elements are mainly preserved on flat-lying or inclined parts of prominent erosion surfaces, some of which mark the bases of channel bodies that may have been active or abandoned at the time of vegetation growth or log buildup.

ROF 6 is interpreted as representing forced mid-channel bars that typically overlie forced mid-channel pools (fmb, fmp), with hydraulic forcing around the obstacles. Similar features have been documented as vegetation-induced sedimentary structures or VISSs (Rygel et al. 2004; Ielpi et al. 2015). A few sheets to lenses of facies Fm locally show V-shaped features 1.0 m deep (Fig. 11D) that are inferred to mark the former positions of trees lost to windthrow or decay. The bedsets are interpreted as the remnants of floodplains (fp), possibly denoting vegetated islands in the channel belts.

ROF 7 deposits are flat-lying and gently inclined sheets and lenses of gray strata in all the channel belts (e.g., Fig. 8B). Exposures at the base of Channel Belt A and fine, pale layers visible in drone images indicate that most fills comprise thin beds of massive sandstone (Sm), along with drab-colored laminated and massive mudstone (Fl, Fm). Contorted laminae in some fills may denote small slumps, and one prominent slump block was identified (Fig. 3). Most occurrences in the channel belts (subset 7A) occupy discrete hollows and are up to 2.5 m thick in the S–N face, averaging 0.8 m (66 examples). In some cases, extents of tens of meters generate an elongate appearance in the broadly flow-parallel S–N face.

ROF 7 is interpreted as abandoned channel fills (abc) of primary and secondary channels. The low average thickness of the occurrences indicates that the sediment filled small channels or larger channels that had been substantially filled with sand. In the three fine intervals, extensive sheets up to 3 m thick (subset 7B) are interpreted as the deposits of floodplain lakes (fl; e.g., Figs. 8A, 10A, C).

As shown in Figure 12A, the depth of channel scours (ROF 2A) scales broadly with the thickness of strata interpreted as bank-attached bars, components of mid-channel bars, and forced mid-channel bars (ROFs 3–6). Preserved planar cross-sets of ROF 4 are much thinner on average. There is little indication of systematic change in the scale of the inferred geomorphic units between the three channel belts.

Sand sheets (ROF 1) and channel scours (ROF 2) average 48% and 28% of the outcrop panels, respectively (Table 6). The boundaries of the two types are relatively diffuse, in contrast with other types and, together, they average 76% as an interconnected geomorphic assemblage (see Fig. 11A, B for good examples). Bank-attached and related geomorphic units (ROF 3) are relatively uncommon, averaging only 1% but forming 7% in one panel (Fig. 8B). Several prominent occurrences are present in the southern, less well exposed part of Channel Belt A, which was not amenable to analysis. The slipface components of unit bars (ROF 4) average 2% but form 6% in one panel with stacked sets (Fig. 8A). Mounded bar cores (ROF 5) average 11% but form 18% in one panel. Bars and pool fills associated with vegetation (ROF 6) average 1% but form 3% and 5% in two panels. Collectively, the bar components represented by ROFs 3–6 average 15%, with 22% in one panel. Abandoned-channel fills (ROF 7A) average 8%.

Oriented broadly along paleoflow, the S–N face is suitable for identifying downstream-accreting components of macroforms but is less suitable for identifying obliquely or laterally accreted components that may include bank-attached bars. The latter, however, are not prominent in the broadly across-flow E–W face (Fig. 3). The proportion of abandoned channel fills is underestimated in the along-flow section, which does not necessarily show the maximum thickness of the fills. For Channel Belt A, abandoned-channel fills comprise 8% of each of the two panels in the S–N section but constitute 16% in the equivalent section of the E–W face, where individual occurrences are thicker (Fig. 3).

Table 6 demonstrates that bar components (ROFs 3–6) are more abundant in Channel Belts A and B than in Channel Belt C. The latter channel belt has a relatively higher proportion of sand sheets, generating a planar appearance, and a lower proportion of channel scours (Figs. 11, 12B). Forced mid-channel bars and pools form low proportions in the upper two channel belts but are also present outside the analyzed panels in Channel Belt A (Fig. 11C–E). The distinction of sand sheets (ROF 1) and channel scours (ROF 2) may be problematic because: 1) many flat-lying sand sheets have scoured bases, 2) channel scours with concave-up erosional forms are best identified in transverse sections, and 3) channel-scour fills commonly grade up into unconfined, sheet-like bedsets.

Channel Belts A–C are 16.5 m, 9.5 m, and 9 m thick, respectively (Figs. 2, 3), and they are associated with a suite of stratigraphic surfaces (Table 5). Channel-belt bases are relatively planar along the S–N face, with a few meters of relief. Channel Belt A thins southward, and Channel Belts B and C are amalgamated in the central part of the face, cutting out Fine Interval C. Channel-belt tops for Channel Belts A and B appear scalloped with small channel cuts, grading up locally into the overlying finer intervals. In contrast, Channel Belt C has a sharp, planar upper surface all along the S–N face.

Sixteen channel bodies or stories were identified in the multistory channel belts in the S–N face, resting on channel bases that form prominent, extensive erosional surfaces. Although a few channel bodies have relatively uniform thickness all along the S–N face (e.g., 14 and 16), most are cut out below overlying channel bodies (e.g., 5 and 15). Some rest on erosion surfaces that lose their prominence laterally, requiring channel bodies to be grouped (e.g., 9–11). The channel bodies contain numerous smaller channel scours, and it was a matter of judgment which to number in the spectrum of thickness and extent. The most extensive channel bodies are inferred to represent channel belts and primary channels, with component channel scours representing smaller braid channels. Channel-body thickness shows little upward change in the channel belts. An exception is channel body 7, which is a composite of thin channel fills along the top of Channel Belt A, representing a suite of channels emplaced in the final stage of the channel belt or cut after avulsion.

The numbered channel bodies average 3.6 m in maximum thickness, ranging from 1.8 to 6.5 m (Fig. 12A). Some sandstone bedsets lack major erosional surfaces and pass upward into fine-grained abandoned-channel fills that are truncated below the next prominent erosional surface. These large-scale facies successions are slightly thinner, averaging 2.9 m in thickness and ranging from 1.0 to 5.1 m (n = 50, Fig. 12A).

Although the true widths of the channel bodies could not be established, most of those recognized on the S–N face could be traced round the outcrop corner into the E–W face, where additional channel bodies were identified and numbered in accord with their relative elevation (e.g., 11A–D, Fig. 13A, B). The apparent widths of channel bodies that have both margins visible in Figure 13B range from 15 to 205 m, with an average of 82 m (n = 19).

The fluvio-lacustrine strata of the three fine intervals (9.5 m, 6.5 m, and 3.5 m thick, respectively) extend all along the S–N face, except where Fine Interval C is cut out by channel-belt amalgamation. Bedsets of facies Sm and Fl up to 4 m thick (Fig. 5A, B) are associated with lensoid channel bodies that rest on concave-up to sub-planar bases, locally crosscutting and amalgamated in places with the overlying channel belts (Fig. 2). These channel bodies contain flat-lying to gently concave-up sandstone layers attributed to sand sheets, channel scours, and unit bars and the cores of mid-channel bars. Their fills aggraded upward and inward from the channel margins, in some cases leaving an abandoned channel that was later filled and overtopped with fines (Figs. 8A, 10A, C). Lensoid channel bodies in Fine Interval A average 4.3 m thick (3.5 to 5 m range, Fig. 12A), and eight channel-bodies in Fine Intervals B and C average 2.5 m thick (1.8 to 3.8 m range). These values are less than the overall thickness of channel bodies and facies successions in the channel belts, but the aggregate thickness is difficult to determine where channels in the fine intervals were overtopped by fines.

Ten numbered events in channel-belt evolution were recognized in the transverse cross section of Figure 13B. During initial emplacement of the Swartberg rivers, Channel Belt A incised to its deepest level near the outcrop corner (event 1), with the base rising westward by 20 m over 235 m. The abundance of facies Gl with reworked mud fragments in the basal strata indicates considerable erosion into the underlying fluvio-lacustrine strata during avulsion (reworked mudstone fragments are not apparent in drone images of the higher channel belts). Channel bodies 1 to 7b progressively onlapped this basal surface, with renewed incision of channel body 3a and a broadening of individual channel bodies (events 2, 3) before abandonment (event 4).

Channel Belt B reoccupied the area (event 5) with the point of deepest incision ∼ 100 m west of the deepest point of Channel Belt A. The base rises westward by 10 m over 100 m. Channel bodies 7c to 11d progressively onlapped the basal surface (event 6), broadening the channel belt before abandonment (event 7). Channel Belt C has a point of deepest incision close to that of Channel Belt A (event 8), broadening with several extensive channel bodies and less systematic basal onlap (event 9), before abandonment (event 10). The maximum relief on the channel-belt base is 6 m. Successive channel belts were thinner and less extensive.

The numerous channel bodies in the multistory channel belts imply the repeated abandonment and relocation of primary channels in the belt. The avulsion that ended sediment accumulation in each channel belt probably resulted from aggradation in the initial incision and later broader conduit until the conduit was unable to contain the largest discharges. There is no indication that extrinsic factors, such as tectonic or climatic events, controlled channel formation, filling, and avulsion, although that cannot be ruled out.

Strong and Paola (2008) distinguished topographic valleys, which existed at near-maximum depth during initial incision, from stratigraphic valleys, for which the bounding surface evolved when later channels cut into aggrading floodplain sediment. In the latter case, the preserved valley form does not correspond to a discrete valley at any point in time. Channel Belt A, with 20 m of relief from its deepest eastern point to its westward termination, probably occupied a shallow topographic valley underlain by a master surface because abundant mud fragments in the basal strata are indicative of strong local incision. Furthermore, the thickness of channel body 1 is only 4.5 m, which would imply the considerable accretion of some 15 m of floodplain sediment during the life of the channel belt, if the channels and floodplains had aggraded together. Shallow topographic valleys also probably best describe the higher, thinner channel belts with their zones of deeper incision.

Channel clustering also may have influenced the architecture. Channel bodies may show clustering due to repeated local avulsions and the tendency for channels to occupy depressed areas generated by subsidence or differential compaction (Hajek et al. 2010; Hofmann et al. 2011). In the present study, channel bodies 3a and 14c are more deeply incised than the adjoining channel bodies. These channel bodies may represent separate avulsions in the fluvial zone, with channel fill later incorporated into the channel belts when overlying, more extensive channel bodies (6 and 16b) cut across them. Additionally, channel body 3a appears to be capped by floodplain deposits at its left-hand margin (Fig. 13B), implying a gap in time before the rest of Channel Belt A was formed. Examination of the adjoining floodplain deposits might indicate the relative importance of valley formation and channel clustering because: 1) mature paleosols adjacent to channel-belt tops would imply limited overbank flow and prolonged soil formation adjacent to a valley cut, whereas 2) crevasse splays extending from channel bodies into the floodplain deposits would imply the presence of alluvial channels with active floodplains. However, exposure of fine-grained strata is minimal in the cliff face, where vegetation and debris cover the outcrop. An unsuccessful attempt was made to trace finer beds laterally from the accessible basal strata, and traceable beds were not identified in the drone images.

The deepest incision in all three channel belts is close to the cliff corner. This implies that the S–N cliff line approximates the thickest original development of the Swartberg Koppie fluvial channel belts. These channel belts thinned westward (Fig. 13B) and probably also eastward, where fine-grained floodplain and fluvio-lacustrine deposits would have been eroded to yield the modern escarpment and valley. The southward rise of the basal surface of Channel Belt A (Fig. 2) indicates the presence of a valley margin in that area and possibly an oblique orientation of this part of the channel belt relative to the cliff line.

Channel Belt A appears to be the widest of the three, in accord with its greater thickness. Its width is at least 235 m in a broadly transverse section (Fig. 13B). With deepest incision at or near the outcrop corner, an original channel-belt width of twice this amount (∼ 500 m) is likely. Channel Belt B thins both east and west from its area of greatest thickness, and its width may not have greatly exceeded 200 m. Channel Belt C has a width of at least 170 m and may originally have been 200–250 m wide.

Channel Bodies 9 (3.9 m) and 10 (2.7 m) in Channel Belt B rest on inclined erosional surfaces that show an apparent northward dip, as do their overlying bedsets (Figs. 2, 13C, D). Apparent dips are typically 3–5° but locally as high as 9° (minimum estimates of the true dip). The steeper bedsets in channel body 10 extend for ∼ 100 m along the outcrop face, and the apparent overall extent of inclined strata is ∼ 150 m. In contrast, bedsets in the overlying channel body 11 are flat-lying, as are those in Channel Belts A and C.

Such steep depositional dips are unusual for a trunk channel. The base of Channel Belt B appears flat-lying, and the steep dips are not readily explained as cascades or rapids at a knickpoint in the river floor. However, similarly steep dips have been documented in small alluvial fans, such as those of Death Valley (Blair 1999), which have surface slopes of up to 5° and bedsets that dip downfan at 5°. In a Pennsylvanian valley fill in Kansas, Feldman et al. (1995) mapped well defined tributaries with benches on limestone substrates that imply rapids or waterfalls dropping steeply to the trunk channel. A steep tributary system provides a possible explanation for the inclined bedsets of Channel Belt B, with the northward dips reflecting an angled confluence into the trunk channel. It is not clear whether such a tributary entered from the west or the east.

In the 200–250 m region of Figure 2, relatively thick fluvio-lacustrine beds of Fine Interval C lie to the south (upflow) of amalgamated sandstones of Channel Belts B and C. A lensoid channel body (part of channel body 9) underlies the lacustrine beds, which are bounded to the north by channel body 10 and are cut by channel body 11 of Channel Belt B. Tracing was difficult in this part of the profile, which is also affected by a dolerite intrusion. However, sand deposited from tributary systems (possibly represented by the three channel bodies) may have impounded small lakes in the tributary or trunk channel. In upland valleys of western Canada, steep tributary systems with high loads build out into the trunk channels, locally overloading the system and impounding the flow in wetlands (Makaske et al. 2017).

The maximum (story) thickness, corrected for compaction, for the thicker channel bodies is considered representative of the braided-channel belt at its largest dimensions. The thinner channel bodies may represent smaller channel belts, tributary channels, or local braid channels. The maximum thickness serves as a proxy for maximum thalweg depth (d_max), which probably represents the deepest scour at braid-channel confluences (Long 2021). Although most channel-body thicknesses are minimum estimates measured between major erosional surfaces, inspection showed that the topmost, complete stories in each channel belt are similar in thickness to those below. Facies successions that include mudstone are on average thinner than maximum story thicknesses (Fig. 12A), although decompaction of the mudstone would generate a better match.

The maximum thalweg depth and the calculated bankfull width yield an approximation of the active-channel dimensions. The maximum thalweg depth, corrected for compaction, was calculated as 2.0 to 7.5 m for the 16 channel bodies, averaging 3.9 m (Table 7). The average bankfull depth was calculated at 1.1 to 3.8 m, averaging 2.1 m. These results exceed slightly, but accord broadly with, the measured height of bank-attached bars (average 2.0 m) and mid-channel bar components (average 1.6 m) in Figure 12A, not corrected for compaction. The bankfull width was calculated at 37 to 172 m, averaging 84 m. Although true channel width could not be determined, the estimates are broadly in accord with the average apparent width of 82 m in the E–W face.

Bankfull discharge averaged 99 m³s⁻¹, and drainage basin area averaged 1370 km². These results are considered provisional due to uncertainties in channel-body and belt geometry for multichannel, braided systems (Long 2021).

Four fluvial styles are represented in parts of the Swartberg member outcrop (Fig. 14). Style 1 (Channel-Bar Complex) is the predominant style of Channel Belts A and B, comprising sand sheets, channel scours, components of mid-channel and unit bars, bank-attached bars, and abandoned-channel fills. The abundance of trough cross-beds in the sand sheets, and the prominence of mid-channel bar components, are indicative of perennial flow conditions.

Style 2 (Low-Sinuosity Complex) is notable for the prominence of bank-attached bars with benches and associated accretionary banks, passing laterally into sand sheets. The style is best developed in discrete channel cuts in the southern part of the S–N face. Here, Channel Belt A thins and the bars developed against steeper banks with more resistant substrates of the underlying Fine Interval A (Fig. 2). The limited extent of inclined bedsets is evidence to support an interpretation of alternate or side bars rather than point bars, which typically show sustained migration over longer distances. Style 2 is likely indicative of a low-sinuosity channel form, although planform outcrops are not available for confirmation.

Style 3 (Sand Sheet and Scour Complex) comprises sand sheets composed of flat-lying bedsets with trough cross-beds, along with channel scours, a small proportion of mid-channel bar components, and abandoned-channel fills. The style characterizes Channel Belt C, which is notable for its planar form and even upper surface with plane beds (Sh) and mounds (Sac). The channel belt is capped by a prominent unit bar that accreted over an apparent distance of 50 m.

Style 4 (Fluvio-Lacustrine Complex) represents the three fine intervals, where stratified mudstone and sandstone were laid down in shallow standing water bodies (lakes). Thickening/thinning rhythms (Fig. 5B) suggest the advance and retreat of supply from point sources. Bioturbation was not observed where the stratified deposits could be examined, for reasons that are not clear, but lake deposits elsewhere also show a scarcity of bioturbation (e.g., Simon et al. 2016). Associated lensoid channel bodies with sand sheets, channel scours, and a few mid-channel and unit bars are interpreted as distributary channels a few meters deep (Fig. 12A).

Fine Interval A comprises nearly 10 m of strata and is well exposed at the outcrop corner, overlying reddish-gray mudstone. This interval represents a prolonged lake phase with Style 4 channel bodies, before the avulsion into the area of a major fluvial system. There is no indication from the outcrop belt as to why a lake formed at this level and what the containing dam may have been. Fine Intervals B and C are thinner (6.5 and 3.5 m, respectively), with similar but thinner Style 4 channel bodies, representing standing water in shallow trunk and tributary valleys following avulsion of the channel belt. Wetlands and lakes are prominent in many modern valleys, including systems such as Mulwaree Ponds of Australia (Williams and Fryirs 2021), where reduced river flow has led to chains of lakes, some of them occupying former thalweg positions. The tops of the channel belts are uneven, and there is plenty of scope for local blockage by trunk and tributary sand.

Analytical results of U-Pb LA ICPMS dating of calcite-cemented pedogenic nodules are presented in Table 8. Sample 120822-7 is an in situ paleosol nodule collected in the hillslope outcrop ∼ 94 m below the base of the Swartberg member (Fig. 15; S–30.41836°, E 026.26051°; Supplemental Material S1). The sample gives an age estimate of 243 ± 2 Ma (2σ; MSWD = 1.8) indicating that crystallization occurred in the late Anisian or early Ladinian (Fig. 1D). Sample 120822-8 (S 30.41847°, E 026.26016°) was collected from a reworked intraformational conglomerate, representing the recalcitrant remains of a “ghost paleosol” (Gastaldo et al. 2020b), overlying the previous sample site by ∼ 15 m (∼ 80 m below the Swartberg base). This sample gives an age estimate of 241 ± 3 Ma (2σ; MSWD = 1.0) indicating that the paleosol in which the nodule precipitated was early Ladinian or late Anisian.

One additional sample, 120822-4 (S 30.414654°, E 026.261167°) collected lower in the section contained high common Pb and was insufficiently radiogenic to yield a precise age date. This is reflected in the clustering of data points towards the Y axis (common Pb component), indicating that the system was reset (Supplemental Material S1). Hence, the age of this sample cannot be determined with a high degree of accuracy, resulting in an uncertainty of 10 Ma (247 ± 10 Ma, MSWD = 1.5).

Sandy braided rivers of the western North American plains provide partial analogues for the Swartberg Koppie outcrop. These rivers include the Platte (Smith 1970, 1971, 1972; Blodgett and Stanley 1980; Crowley 1983; Horn et al. 2012a, 2012b), Niobrara (Skelly et al. 2003), Calamus (Bridge et al. 1986, 1998), and Wisconsin (Mumpy et al. 2007) of USA, and the South Saskatchewan of Canada (Sambrook Smith et al. 2006; Parker et al. 2013). The grain size of these rivers is typically greater than that in the Swartberg member. Large rivers elsewhere, such as the Brahmaputra and the Paraná, also provide partial analogues (Best et al. 2003; Sambrook Smith et al. 2009). Lunt and Bridge (2004) noted that a generalized braided-fluvial model applies to sandy and gravelly systems, and analogues may also be found in the gravelly Sagavanirktok River of Alaska (Lunt and Bridge 2004) and the Squamish, Fraser, and Kicking Horse rivers of Canada (Wooldridge and Hickin 2005; Cyples et al. 2020). Most of these rivers exhibit anthropogenic effects, including dams and water takeout, requiring caution in their application to fluvial systems in deep time.

Fluvial parameters for the Swartberg Koppie outcrop match the scale of braided channels on the North American plains. For example, the South Saskatchewan has a channel belt ∼ 600 m wide with the main channels 50–150 m wide and 2–5 m deep, with maximum thalweg depths typically ∼ 3 m below the tops of the adjacent compound bars (Sambrook Smith et al. 2006; Parker et al. 2013). The Niobrara has a channel belt 90–330 m wide (Skelly et al. 2003), and vegetated banks of the Platte are up to 2 m high (Smith 1971). In all these rivers, the channel belts comprise channel-bar complexes with primary and secondary channels, mid-channel unit bars, amalgamated compound bars, and bank-attached bars. Vegetated islands with trees and woody debris are prominent. Swartberg fluvial styles 1–3 have good analogues in many of these river systems. Near Outlook, Saskatchewan, Parker et al. (2013) noted that bars were generally overtopped when flows exceeded 230 m³s⁻¹, in accord with the maximum bankfull discharge calculated for the Swartberg member channel systems.

Reflector patterns similar to ROF 1, with flat-lying strata and abundant cross-stratification, are widely represented in GPR profiles, especially in along-flow traverses (Skelly et al. 2003). The facies is prominent in primary and secondary channels where sinuous-crested dunes are abundant (Sambrook Smith et al. 2006; the “open-water channels” of Horn et al. 2012a, 2012b), as well as within and capping mid-channel compound and unit bars (Mumpy et al. 2007; Parker et al. 2013). Radar facies with these features account for 40–83% of the profile areas (Wooldridge and Hickin 2005; Sambrook Smith et al. 2009; Parker et al. 2013; Cyples et al. 2020). Flat-lying beds with transcritical plane beds are prominent in ephemeral rivers with strongly seasonal flow, including the Ghash River of Sudan (Abdullatif 1989) and the Luni River of India (Carling and Leclair 2019), but the prominence of trough cross-beds in the Swartberg outcrop is indicative of more sustained flow conditions. Many analogues exist for the channel scours of ROF 2. For example, GPR profiles in the Platte River, Nebraska, imaged the fills of secondary channels 1–3 m deep and 10–20 m wide, as well as smaller bartop channels (Horn et al. 2012a).

The macroforms represented in ROFs 3–6 also can be matched in many GPR profiles. The bank-attached bars of ROF 3 resemble those of the low-sinuosity (< 1.3) channel of the Calamus River, where alternate bars and point bars form single or double rows depending on channel width (Bridge et al. 1986). The side bars of the Kicking Horse River are also possible analogues, attached to an accretionary bank and accreting downstream (Cyples et al. 2020). Bank-attached bars with benches and accretionary banks are prominent in strongly seasonal rivers of Australia and Canada (Brooks 2003; Page et al. 2003).

High-angle inclined bedsets, similar to those of ROF 4, form components of unit bars in many GPR profiles (transverse bars of Smith 1971, 1972; Horn et al. 2012a), with simple forms that remain relatively unmodified during bar migration (Smith 1971). The bars are typically lobate with straight-to-arcuate avalanche faces and a predominant down-current migration direction, with subsidiary lateral to oblique accretion. They generate tabular sediment bodies that are typically 1–2 m thick, and they commonly build into channel thalwegs and confluence scours (Sambrook Smith et al. 2019). The inclined bedsets dip steeply (15–36°: Smith 1972); the relatively low dips of 8–21° recorded in the Swartberg outcrop may indicate that the cliff cuts obliquely through the macroforms.

The mounded macroforms of ROF 5 have analogues in the convex-up bar cores imaged in across-flow radar profiles of the Niobrara River, Nebraska, developed over scours and with flanking clinoforms traceable for up to 10 m (Skelly et al. 2003). Similar features have also been noted in the Calamus and South Saskatchewan rivers (Bridge et al. 1998; Parker et al. 2013). The mounded macroforms at Swartberg Koppie are interpreted as components of compound bars, generated as assemblages of unit bars that evolved through erosional and depositional events (Lunt and Bridge 2004). In the Sagavanirktok River, Alaska, Lunt and Bridge (2004) documented compound bars composed of 3–7 sets of large-scale inclined strata, each formed by a unit bar.

Hyperbolic GPR reflections in rivers with abundant woody material have been attributed to buried logs (Mumpy et al. 2007; Cyples et al. 2020), suggesting parallels with ROF 6. The reflections are probably associated with forced mid-channel bars and pools, but these are difficult to image with GPR. Abandoned-channel fills (ROF 7A) are not widely documented in subsurface studies, but partially to completely abandoned channels are present in the Tagliamento River of Italy, especially at channel-belt margins (Tockner et al. 2003). The vegetated intervals of sandstone and mudstone illustrated in Figure 11D may represent vegetated islands.

Many studies have noted a contrast between surface morphology, where macroforms with avalanche faces are generally prominent, and the subsurface record where mesoform cross-beds, planar strata, and channel scours are prominent (Horn et al. 2012a; Parker et al. 2013). GPR profiles through bars are typically dominated by dunes, with a modest proportion (15–20%) of inclined reflector sets (Wooldridge and Hickin 2005; Sambrook Smith et al. 2009; Cyples et al. 2020). As discharge falls, bar tops and margins developed at high-flow stage are truncated, and slipface deposits are replaced with smaller dune sets and channel scours. Rhythms of macroform growth and reworking may operate over periods of only a few years.

In the South Saskatchewan River, Canada, Parker et al. (2013) noted some “textbook examples” of compound bars in GPR profiles, but typically only remnants were preserved, making diagnosis difficult. In one good example, an initial convex-up unit bar, ∼ 50 m wide, developed into a large compound bar through the accretion of thalweg, bar surface, and avalanche-face deposits, the latter forming only a small proportion of the GPR panel (Fig. 16A, with 4.5× vertical exaggeration).

With a S–N face ∼ 500 m long, the Swartberg Koppie exposure is long enough to image compound bars comparable in scale to those of modern rivers, which are commonly tens to several hundred meters in extent. However, the high proportion of flat-lying bedsets and channel scours (ROFs 1 and 2) and the modest proportion of macroforms (ROFs 3–6) are indicative of substantial reworking, as in the modern analogues. An inferred compound bar with a thickness of ∼ 4 m and an apparent length > 75 m is illustrated in Figure 16B, with 1.5× vertical exaggeration and four events highlighted. Resting on a scoured surface cut into an abandoned channel fill (abc), the mounded core of a mid-channel bar (mcb) occupies the 30–70 m position (event 1), with inclined accretion sets that show an apparent northward dip (Fig. 8B). The mound is cut by a shallow bartop channel (btc) and passes southward into a sand sheet (ss), overlain by a second mound (event 2). A shallow scour occupies the hollow between the two mounds and contains a small bank-attached bar that built from the flank of the initial mound, passing upward into sand sheets (event 3). Finally, sand sheets built up and flattened the topography over the mounds (event 4). Minor hollows on the mounded topography filled with mud. Joeckel et al. (2014) noted the amalgamation of architectural elements into compound bars in the Miocene Ogallala Group of the U.S. Great Plains.

Fine-scale sedimentary features such as desiccation cracks, rain prints, and evaporite pseudomorphs have been widely used for paleoclimate reconstruction in the Beaufort Group (e.g., Stear 1978; Smith 1980), but they are below the resolution of the UAV images for the Swartberg member. However, the predominance of trough cross-beds in the sand sheets of all three channel belts implies perennial subcritical flow, allowing systematic bedform migration. Local plane-laminated and scour-and-mound sandstones (facies Sh and Sac) imply periods of transcritical and supercritical flow in shallow water, probably associated with episodic discharge (Wang and Plink-Björklund 2020). Facies successions that pass upward from subcritical to transcritical bedforms (Fig. 5A) are probably linked to shallowing upward in flow conduits. Channel scours (facies Ss and larger scour fills) indicate frequent erosive events. The prominence of bedsets interpreted as mid-channel bars supports an interpretation of sustained-flow periods close to bankfull level. The flat-topped bar at the top of Channel Belt C (Fig. 7), with an apparent accretion distance of 50 m and foreset erosion and reactivation, indicates seasonal variation in discharge with alternate high-stage accretion and low-stage erosion.

These observations collectively imply a humid, seasonal paleoclimate, under which rivers experienced sustained, perennial discharge with seasonal variation. The fluvial style accords with patterns of intermediate discharge variation (Fielding et al. 2018). Although the discharge pattern may reflect the paleoclimate in the Cape Fold Belt source area, water was in good supply on the landscape, as indicated by standing water in lakes in fine intervals, in blocked trunk and tributary channels, and in abandoned channels. The wide range of calculated discharge for the 16 channel bodies may imply variation in channel-belt discharge but more probably reflects scatter in the thickness of the sampled channel fills. Turner (1983) used similar reasoning to infer a humid paleoclimate for the Upper Triassic Molteno Formation.

Vegetation, including upright trees, is preserved in all the channel belts. Its presence is inferred from vegetation-induced sedimentary structures (VISSs) (Fig. 11C–E), associated with forced mid-channel bars and pools and floodplain remnants. Occurrences in the basal deposits of some channel bodies implies seasonal dryness or the growth of vegetation in abandoned channels. These structures have particular utility for evaluating the Permian and Triassic vegetation record of the Karoo Basin, where organic matter is commonly poorly preserved due to taphonomic factors (Gastaldo et al. 2005). Plant fossils are typically below the resolution of the drone images and may be more abundant than noted, although megafloral elements are preserved in less than 1% of these rocks (Gastaldo and Bamford 2023). Riparian vegetation is prominent in modern dryland areas, and the identification of VISSs does not necessarily imply a humid climate, although it may support other lines of evidence.

Advances in obtaining U-Pb age estimates from calcite cements, including those from pedogenic nodules (Palafox et al. 2022; Rochín-Bañaga et al. 2023), provide a means to test hypotheses about the age and continuity of continental successions in the absence of volcaniclastic deposits. Currently, the Swartberg member at Bethulie is dated as Induan (e.g., Smith et al. 2022), based on the presence of an interval in which Lystrosaurus skeletons predominate (Viglietti et al. 2013). The genus has been interpreted as a disaster taxon (Botha-Brink et al. 2016; but see Modesto 2020) and is believed to represent post-extinction ecosystems of the very earliest Triassic.

Smith et al. (2022, their Fig. 2) reported a stratigraphic distance of less than 50 m between their vertebrate-defined End Permian Mass Extinction interval (EPME, the three-phased extinction model of Smith and Botha-Brink 2014, and others) and the Lystrosaurus-rich interval described as located less than 20 m below the base of the Swartberg member. Hence, these authors show a stratigraphic thickness between the EPME and the base of the Swartberg member of about 70 m, corresponding broadly to the lower part of the Katberg Formation. However, in previously published measured sections, Botha and Smith (2006) placed the base of the Swartberg member at 162 m above the Permian–Triassic boundary as defined by vertebrates. Consequently, there is significant ambiguity about: 1) where the base of the Swartberg member lies relative to the vertebrate-defined datum, and 2) the thickness of strata exposed on Swartberg Koppie, especially for the lower Katberg Formation. Our measured sections (Supplemental Material S1) indicate that the base of the Swartberg member lies ∼ 210 m above the vertebrate-defined datum of Botha and Smith (2006), and hence we document an additional 50 m of section in this classic locality. Regardless of boundary issues, an Early Triassic age is currently assigned to the lower member of the Katberg Formation (Smith et al. 2022).

The ²³⁸U-²⁰⁶Pb dates of 243 ± 2 Ma (late Anisian) and 241 ± 3 Ma (early Ladinian), obtained from two samples of calcite-cemented pedogenic nodules, were collected from exposures ∼ 15 m apart in the Katberg Formation below the base of the Swartberg member (Fig. 15). They indicate that pedogenic nodule cementation centered around the late Anisian (Middle Triassic), implying a Middle Triassic or younger age for the Swartberg member. These results correspond to data reported from the same stratigraphic interval exposed at the farm Nooitgedacht (Rochín-Bañaga et al. 2023), located 33 km to the west-northwest of the study area. Fossils of the reported Lystrosaurus bone bed, attributed to the post-EPME recovery ecosystem at ∼ 251 Ma (Smith and Botha 2005; Botha and Smith 2006; Smith and Botha-Brink 2014; Smith et al. 2022), are located only a few tens of meters below the base of the Swartberg member. Thus, they are likely also late Middle Triassic in age.

The results indicate that the Katberg Formation is a highly condensed stratigraphic unit, in which erosional surfaces and, probably, mature paleosols incorporate considerable time. These observations have implications for the continuity of measured sections, interpretation of fluvial style, and inferred unidirectional trends, as explored schematically in Figure 17. For the Swartberg member, the three channel belts taper out westward within shallow valleys, which have 6 m to 20 m of basal relief and are deeply incised at their lowest points and locally elsewhere. In the succession, four fluvial styles are closely associated in a multi-style system, with low-sinuosity complexes prominent in thinner, marginal areas. Sheetlike sandstones in the Karoo Basin successions have commonly been attributed to low-sinuosity, ephemeral rivers (e.g., Smith 1995; Smith et al. 2022), but similar deposits (Style 3) form components of all three channel belts, especially the topmost channel belt (Fig. 17). The barforms may include downstream, lateral and oblique components of accretion, and in the absence of paleocurrent data, they could be attributed variously to high- or low-sinuosity systems, especially as many mid-channel bars feature a strong component of lateral accretion (Bristow 1987).

For measured sections in the lower member at Carlton Heights (as illustrated in Fig. 17), Pace et al. (2009) documented suites of narrow fixed channels with basal relief of up to 3 m, pinching out over distances of tens of meters, as well as lenticular sheetflood deposits up to 2 m thick. Channel-lag accumulations of carbonate paleosol fragments (also documented at Bethulie: Smith 1995; Smith et al. 2022), as well as pedogenic mud aggregates (Gastaldo et al. 2013), indicate alternate aggradation and degradation of the landscape, implying the presence of significant diastems. Paleosol characteristics show evidence of fluctuating paleoclimate, with overprinting of older by younger paleosols.

Depending on where sections were measured in the transition from valley axis to valley margin to interfluve (Fig. 17), assessments of stratigraphic continuity, thickness, and fluvial style would be radically different. Similar Quaternary rhythms of aggradation and degradation have been documented for channel and floodplain deposits in the Himalayan Foreland Basin of India, with evidence of high-frequency climatic fluctuation, reworked carbonate and mud clasts linked to floodplain gully erosion, and numerous diastems (Gibling et al. 2005, 2011; Tandon et al. 2006).

Similarly, fluvial systems of the Abrahamskraal and Teekloof formations display a range of styles in complex channel bodies, with a close association of mid-channel and bank-attached bars (downstream and lateral accretion) and no single style dominant (Gulliford et al. 2014; Wilson et al. 2014; see also Bordy and Paiva 2021). Wilson et al. (2014) noted also that incision into semi-consolidated alluvium (including indurated paleosols) influenced fluvial style by confining the channels, disconnecting them from the adjoining floodplain so that they cease to be truly alluvial channels. Gastaldo et al. (2018, 2019a, 2021) found no consistent change in sandstone body or bedset thickness in the Elandsberg and Palingkloof members, beyond local thinning of fluvial sheets or ribbons, closely linked to outcrop orientation with respect to paleoflow. Trough cross-beds were common throughout, a similar range of primary sedimentary structures was present, and intraformational conglomerate was prominent in channel lags throughout the successions, in the form of pedogenic nodules, mudclasts, and bone. Gastaldo et al. (2021) concluded that fluvial style showed little systematic change through the latest Permian and earliest Triassic.

Mudclasts (facies Gl) are prominent at the base of the Swartberg member (Fig. 17) where the avulsing fluvial system incised deeply into floodplain and lacustrine deposits, as well as in the lower member (Pace et al. 2009). Reworked biscuit-shaped wisps and nearly spherical particles of mudstone are widespread in the upper Balfour and Katberg formations (Gastaldo et al. 2015, 2018, 2021; Viglietti et al. 2017), where they have been attributed variously to erosion of short-lived mud drapes in channels and shallow ponds and to erosion of thicker accumulations in abandoned-channel fills and floodplain deposits (e.g., Stear 1978; Hiller and Stavrakis 1980; Stavrakis 1980; Smith 1995; Wilson et al. 2014). Retallack et al. (2003) attributed mudclasts with soil microfabrics in the Palingkloof Member to a global soil-stripping crisis linked to vegetation dieback and landscape destabilization. The inferred first appearance of such clasts, reported to be restricted to the transitional interval, was used as a diagnostic criterion for the onset of aridification (Smith and Ward 2001; Ward et al. 2005; Smith and Botha-Brink 2014, p. 107). However, in view of the abundance of such clasts in fluvial deposits of the Karoo Basin, the Palingkloof occurrences are unlikely to be unique, especially in the absence of a comparable technical analysis of mudclasts at other levels.

The Katberg Formation has played, and continues to play, an important role in the interpretation of continental landscapes following the end-Permian ecosystem crisis in South Africa and worldwide. The formation has been interpreted as part of a northward-tapering megafan or DFS (Groenewald 1996), resting on an unconformity with the Balfour Formation (Catuneanu and Elango 2001; Viglietti et al. 2017). A coarsening-upward succession culminates in the Swartberg member, with subsequent upward fining through the Burgersdorp Formation, with a prominent nonconformity separating the two formations in distal areas (Neveling 2004). A northward paleoflow assessment from our study is in accord with deposition on a megafan building from the Cape Fold Belt (Hiller and Stavrakis 1980; Stavrakis 1980; Neveling 2004). Similarly, the Abrahamskraal and Teekloof formations in the western part of the basin (Fig. 1B) are attributed to deposition on megafans (Gulliford et al. 2014; Bordy and Paiva 2021), and the Molteno Formation was deposited as a clastic wedge in Late Triassic time (Turner 1983; Hancox 1998).

Although more research is needed to explore a megafan model for the Katberg Formation, such a depositional setting needs to be considered when inferring secular change in fluvial style and paleoclimate. Megafans may extend for hundreds of kilometers from mountain exit points (Weissmann et al. 2010), and they display proximal-to-distal trends in fluvial style, as well as progradation and retrogradation due to thrust loading and relaxation in the hinterland (Owen et al. 2017). Fluvial style is highly variable across megafans due to avulsion of major and minor drainage systems (e.g., Sinha et al. 2014; Chesley and Leier 2018), the formation of new fan lobes laterally and downfan (Abrahami et al. 2018), and superimposed high-frequency climatic rhythms that affect channel and floodplain equilibrium profiles (Tandon et al. 2006). Owen et al. (2017) stressed the inherent inhomogeneity and complex surface mosaic of fluvial sequences in DFS successions.

The end-Permian extinction was associated with major changes to vegetation (Looy et al. 2001), with evidence for increased sediment supply in continental ecosystems (Algeo and Twitchett 2010). However, in a DFS context, the abrupt appearance of the Swartberg member is most simply explained by the avulsion into the Bethulie area of a large sandbed river, on the scale of the Platte or the South Saskatchewan. Major river avulsions are well documented for megafans in the Himalayan Foreland Basin, as in the 2008 Kosi River avulsion (Sinha et al. 2014) and the relocation of the Sutlej and Yamuna rivers on their respective fans (Singh et al. 2017). Although climatic changes are likely to have accompanied the relocation of the Swartberg river systems, as with the Sutlej and Yamuna systems (Singh et al. 2017), the tectonically active Cape Fold Belt and Karoo foreland basin provides an obvious explanation for this major change in the fluvial system.

The importance of the Karoo Basin of South Africa for interpreting end-Permian extinction in continental settings is not in doubt. However, much ambiguity remains about how the landscape evolved, especially as regards the style of fluvial systems, paleoclimate and relative aridity, and landscape stability and sediment flux. Our study assesses the fluvial style and paleoclimate for a key stratigraphic unit, the Swartberg member of the Katberg Formation, previously assigned to the Induan (Early Triassic).

Our approach is to use UAV (drone) images to identify what we term remote outcrop facies in the inaccessible cliff face, leading to interpretations of geomorphic units. This approach draws on GPR studies of modern rivers, in which the identification of radar facies links the preserved, subsurface strata with the geomorphic units at surface. In this way, we have been able to make a direct comparison between the fluvial style of the Swartberg member and possible modern analogues. The Swartberg member exhibits a predominance of planar-bedded sand sheets and channel-scour fills, as well as a range of mid-channel and, less commonly, bank-attached bars, inviting comparison in scale and geomorphic style with braided rivers in western North America, for which an abundance of GPR data are available. The modest proportion of identifiable barforms with slipfaces in the Swartberg member accords with observations of these modern braided rivers, in which bars accrete at high-stage flow but are reworked during falling and low-stage flow. Observations of dipping accretion surfaces and cross-beds in the cliff sections, aided by rotation of the 3-D model, established a broadly northward paleoflow direction, in accord with regional assessments and deposition on megafans emerging from the Cape Fold Belt, which is known to have been active at least into the Early Triassic.

We were unable to collect fine-scale evidence for paleoclimate from sedimentary structures, hindering comparisons with earlier research (Table 1). However, the predominance of sand sheets with trough cross-beds and the presence of barforms accords with deposition in perennially flowing rivers, and the presence of bar reactivation surfaces and numerous scours suggests strongly seasonal discharge. We infer a relatively humid but seasonal paleoclimate for the Swartberg member. Although river discharge patterns may reflect the paleoclimate of the source area in the Cape Fold Belt, standing water was prominent in lakes and abandoned channels in the Swartberg channel belt. The identification of vegetation-induced sedimentary structures (VISSs) in the channel belt indicates the presence of vegetation, even though macrofloral remains and palynomorphs are scarce (Gastaldo et al. 2005).

Our Middle Triassic geochronometric dates for pedogenic carbonate nodules in strata underlying the Swartberg member, more than seven million years after the Permian–Triassic boundary, requires substantial revision to the age model for the succession. The dates imply that major, currently unidentified gaps in time are present, probably associated with major breaks in the distal part of the basin linked to megafan tectonics, smaller-scale erosional surfaces in the succession, and periods of landscape stasis associated with mature paleosols. Similarly disjunct successions are typical of Quaternary dryland river plains and megafans, where tectonic effects, high-frequency climate rhythms, and autogenic channel relocation cause episodes of aggradation and degradation.

In view of the revised age model and probable time gaps in the succession, we suggest revisiting the inferred unidirectional trend of fluvial style and aridification across the Permian–Triassic boundary in the Karoo Basin. As illustrated by the studies of Fielding et al. (2019, 2021) in Australia and Ji et al. (2023) in China, the end-Permian extinction in terrestrial settings was not necessarily accompanied by changes in fluvial style and climate.

Supplemental materials are available from the SEPM Data Archive: https://www.sepm.org/supplemental-materials.

The authors thank the following colleagues who either assisted during field work in South Africa and/or were instrumental in instructing and facilitating the use of Metashape for the current study: Timothy Stonesifer, Colby College; John W. Geissman, UT-Dallas; Sandra L. Kamo, University of Toronto; and Lucas Joel, University of California–Riverside. We are grateful to Tobi Payenberg, Brian Willis, and an anonymous reviewer for comments that greatly improved the manuscript, and to Alessandro Ielpi for a thoughtful review of an earlier draft. We acknowledge the hospitality and assistance of Tony Hocking and cordial staff of the Royal Hotel Bethulie. Student participation was supported by the Selover Family Student-Research Endowment and Barrett T. Dixon Geology Research-and-Internship Fund for undergraduate experiences in the Department of Geology, Colby College. Research was supported by the Council for Geoscience; National Science Foundation EAR 0749895, 0934077, 1123570, and 1624302; a Fulbright Award to RAG at the Geology Department, Rhodes University; and faculty start-up funding to JWG from the University of Texas at Dallas.