Large‐Scale Cenozoic Wind Erosion in the Puna Plateau: The Salina del Fraile Depression

Wind erosion is integral to the evolution of arid landscapes on Earth and Mars, but the nature of long‐term wind erosion of bedrock is poorly understood. Here we describe the Salina del Fraile (SdF) depression in the Puna Plateau of the Central Andes, NW Argentina, as a landform excavated by wind over several million years. New structural cross sections and a compilation of chronostratigraphic ages rule out the hypothesis that the depression was created by transtensional tectonics. Dated remnant lacustrine and alluvial deposits in the floor of the depression constrain the rate and timing of erosion. Late Oligocene‐Miocene compressional folding uplifted friable strata that were preferentially eroded, resulting in the high‐relief (900 m) depression. Up to 1.95 km and an average of 1.05 km of strata were eroded during the last 8.2 to 17 Ma, at rates of 0.06 to 0.23 mm/yr. These rates are similar to long‐term average wind erosion rates reported in other regions. Coarse‐grained eolian megaripples, yardangs, and elongated ridges indicate ongoing eolian abrasion and deflation, aided by salt weathering, of the floor of the depression. Megaripple migration across stony lag surfaces exposes fresh bedrock to continued erosion. The SdF also contains kilometer‐scale mesas and ridges that we interpret as erosional remnants. These landforms are similar to megayardangs and erosional topography identified on the lower flanks of Mount Sharp, Gale crater, Mars. In such hyperarid landscapes characterized by lithologic heterogeneities, high‐relief landforms can be generated and sustained by wind erosion, without significant fluvial or glacial incision.


Introduction
The degree to which eolian processes shape the landscape morphology of arid regions is a long-standing point of contention (Brock & Twidale, 2011;Goudie, 2012), albeit one renewed by the exploration of wind-dominated planetary surfaces and their terrestrial analogues (e.g., Xiao et al., 2017). Wind erosion may outpace fluvial incision of bedrock in hyperarid regions, and wind can enhance the erosion of favorably oriented bedrock canyons (Perkins et al., 2015). On a regional scale, eolian deflation and abrasion can remove hundreds of meters of strata, potentially accelerating tectonic deformation (Kapp et al., 2011). Regional eolian deflation influences Earth's climate through dust emission and often results in the development of widespread desert pavements, which may drive faster surface winds through a wind-albedo-wind feedback (Abell et al., 2020;Pullen et al., 2018). Such research shows how wind erosion, although outpaced by fluvial and other processes in more humid regions, plays an integral role in parts of Earth's arid Cenozoic orogenic belts. On Mars, wind-driven retreat of bedrock scarps is responsible for relatively young surface exposure ages in Gale crater, offering potential locations for sampling well-preserved complex molecules (Farley et al., 2014). Identifying regions of active eolian erosion on Mars is important for exploring geologic units that may preserve signs of ancient life (Chojnacki et al., 2018). Studying the effects of wind erosion is therefore critical for understanding how tectonic processes interact with Earth's climate, and how the surfaces of wind-dominated planets evolve over geologic time scales. Both of these efforts, however, are hampered by a relatively poor understanding of large-magnitude eolian landforms generated over many millions of years on Earth.
Here we show that the Salina del Fraile (SdF), a large, internally drained depression in the southern Puna Plateau, NW Argentina, is an erosional landform excavated by wind over a time period of at least 8.2 Ma. We present new structural measurements and cross sections that argue against tectonic lowering of the depression and instead imply up to 2 km of erosion where anticlinal folding exposed a thick package of friable, fine-grained strata. We describe remnants of alluvial surfaces within the depression to argue against significant fluvial incision during this time period. We also document eolian landforms over a wide range of spatial scales, both within and around the depression, to demonstrate the ubiquity of eolian erosion in this region. Finally, we use existing chronostratigraphic data to estimate the timing and rate of erosion and explore some implications of such long-term, large-scale wind erosion.

The Salina del Fraile
The SdF is an evaporite basin located to the west of the Salar de Antofalla in the southern Puna Plateau, NW Argentina ( Figure 1). The present climate of the region is hyperarid, with mean annual rainfall of onlỹ 14 mm (with a range of 0-70 mm) in the vicinity of the SdF (Bookhagen & Strecker, 2008). The National Aeronautics and Space Administration (NASA) Modern-Era Retrospective analysis for Research and Applications (MERRA) reanalysis data indicate that the region experiences a cold desert climate, with summertime daily temperature highs of 20-25°C and lows of −5-5°C, and wintertime highs of 0-10°C and lows of −10 to −15°C, giving the region a cold desert climate ( Figure S1 in the supporting information). Aridity in this region began as early as the Eocene (Quade et al., 2015) and intensified during the Middle to Late Miocene, when the major evaporite playas such as the Salar de Antofalla began forming (Alonso et al., 1991;Strecker et al., 2007). Arid climates, in addition to suppressing fluvial processes, enhance eolian deflation and abrasion due to sediment desiccation, salt weathering, and reduced vegetation cover (e.g., Goudie, 2018).
Eolian deflation commonly occurs on the margins of evaporite playas, such as the SdF, because salt accumulation rapidly disintegrates bedrock and alluvial deposits and provides fine-grained particles for removal by wind (Goudie & Day, 1980). Salt accumulation in closed depressions therefore tends to reinforce further eolian excavation in a positive feedback, especially if the host lithology is susceptible to salt weathering, such as would be for porous sandstones or fine-grained siliciclastics (Goudie & Wells, 1995). Eolian deflation aided by salt weathering can lead to the formation of large depressions, as has been suggested for the Qattara Depression in Egypt (Aref et al., 2002). Eolian processes also interact with fluvial and lacustrine deposits in arid playas because precipitation events result in fine-grained sediment being transported and deposited on surfaces prone to deflation (Goudie, 2018).
The SdF is situated within a continental-scale NNW-SSE eolian system, in a region characterized by erosional landforms such as yardangs, megayardangs, and periodic bedrock ridges ( Figure 1, and references therein). Significant eolian sand deposits are rare in the vicinity of the SdF but occur further downwind in the SE Puna, in the Fiambalá basin (Figure 1), and in a belt stretching several hundred kilometers further SE (Isla & Espinosa, 2017). Strong winds in the region are associated with synoptic polar fronts that typically occur during a windy season that lasts from May to October (Milana & Kröhling, 2017). Major dust storms have been imaged originating from evaporite playas on the Puna Plateau, including from the SdF (Goudie & Wells, 1995;Milana & Kröhling, 2017). Although the Puna Plateau is internally drained, sediment is removed from the southern Puna by wind, potentially providing a source of the Chaco-Pampean loess deposits in eastern Argentina (Milana & Kröhling, 2017) and dust in Eastern Antarctic ice cores (e.g., Gaiero, 2007).
The SdF-Antofalla region was tectonically deformed as it was incorporated into the Andean orogenic wedge~30-20 Ma, uplifting and deforming a Paleogene foreland basin system (Carrapa et al., 2005;Kraemer et al., 1999;Zhou et al., 2017). Cenozoic units outcropping in the SdF consist of the Paleogene Quiñoas and Chacras Formations, a coarsening-upward sequence of continental siliciclastics; the Neogene Potrero Grande Formation, a volcaniclastic unit with interbedded ashes and ignimbrites; and later volcanic flows and ignimbrites (Kraemer et al., 1999). The Quiñoas Formation unconformably overlies pre-Cenozoic rocks. In the study area, pre-Cenozoic rocks consist of Permian to Jurassic sedimentary and volcanic strata exposed in the northern part of the SdF depression (Voss, 2002). An angular unconformity separates the Potrero Grande Formation from underlying units, which are tilted to the SE.
The SdF is a rhomboidal topographic depression, 300 km 2 in area, with up to 900 m of topographic relief. Goudie and Wells (1995) hypothesized that the SdF (therein termed Potrerillos Depression) is analogous to a giant deflation pan, "the largest Altiplano basin of primarily aeolian origin" (pg. 22), but later research in the region diverged from this interpretation ( Figure 2). The depression's conspicuous planform shape, rectilinear slopes, and what appeared to be rotated fault blocks along the western edge (labeled "elongated ridge" in Figure 3), led some to interpret it as a pull-apart basin, with the basin floor lowered by major normal faults as a result of a left step in an inferred N-S trending left-lateral fault system (Reijs & McClay, 2003;Voss, 2002). Others mapped the depression as the core of an anticline with only minor faults, an interpretation that also satisfies chronostratigraphic constraints (Adelmann, 2001;Kraemer et al., 1999;Seggiaro et al., 2007). The structural interpretation of the SdF is critical, because a compressional origin requires the erosion of material in order to create a morphologic depression, while a pull-apart basin would not necessitate erosion ( Figure 2).

Materials and Methods
We collected measurements of bedding planes in the field using a Brunton Transit (n ¼ 40) and mapped the Cenozoic geology along two transects spanning the depression (Profiles A and B in Figure 3). Transects were oriented perpendicular to major structures, allowing us to test the two hypotheses for the formation of the depression (pull-apart basin or compressional folding). To supplement our structural data, we also digitized strikes and dips from Voss (2002, n ¼ 52) and Reijs and McClay (2003, n ¼ 39) within the SdF by georeferencing their maps in a GIS environment. Our field measurements closely agree with nearby data from these sources. Cross sections were constructed using the kink-band method; because a main objective is to estimate the magnitude of erosion, cross sections were drawn to minimize the amount of eroded material, within realistic geologic constraints.
Chronostratigraphic data were compiled from Adelmann (2001, n ¼ 29), Canavan et al. (2014, n ¼ 14), and Quade et al. (2015, n ¼ 11) and mapped when sample locations were given. For ages referenced to a stratigraphic column, we mapped the locations of the samples onto cross section B-B′ using their stratigraphic positions. These data consist of single grain Ar-Ar (n ¼ 16; biotite, plagioclase, and potassium feldspar), K-Ar (n ¼ 12; feldspar, biotite, and glass), and U-Pb (n ¼ 21, zircon) ages and a K-Ar whole rock age (n ¼ 1, andesite). Dated lithologies were largely volcanic tuff (n ¼ 38), followed by ignimbrite (n ¼ 5), andesite (n ¼ 2), and other volcanic lithologies (n ¼ 5). The data tabulated by Adelmann (2001, Appendix 1 therein) were published in part by Kraemer et al. (1999 , Table 1 therein), and the analytic methods were described by Voss (2002 , Table 1 therein). Ar-Ar age determinations were performed by the GEOMAR Forschungszentrum (Kiel, Germany) and K-Ar age determinations by GEOCHRON Laboratories, Krueger Enterprises, Inc. (Cambridge, Massachusetts). U-Pb ages of volcanic zircons were published by Canavan et al. (2014, Tables DR1 and DR2 therein) and Quade et al. (2015, Table 4 therein) and were determined by laser ablation multicollector inductively coupled plasma mass spectrometry (LA-MC-ICPMS) at the University of Arizona LaserChron Center. All ages compiled here are reported with 2σ uncertainty. Structural and chronostratigraphic data used here are available in the University of Toronto Dataverse (McMillan & Schoenbohm, 2020).
Predominant strong wind directions were inferred from landforms mapped using high-resolution aerial imagery obtained from Bing Maps (~30 cm per pixel; Figure S2). Landforms include yardangs, megaripples, dust streaks, and abraded bedrock surfaces. We assume prevailing wind directions are well represented by the long axes of yardangs that taper downwind (n ¼ 21); by directions normal to the average crestline orientation of asymmetric megaripples, with more gentle stoss slopes and steeper lee slopes (n ¼ 5); and by light-toned dust streaks extending downwind from evaporite or alluvial playas (n ¼ 4). Each landform yielded a single measurement of wind direction. Where fields of periodic yardangs or megaripples occur, the wind directions mapped in Figure 3 represent an average of all measurements for the field. Although aerial imagery alone did not permit us to unambiguously distinguish the stoss and lee slopes of megaripples, their orientations and associated wind directions were verified in the field. Abraded bedrock surfaces displaying elongated ridges and grooves may also provide information about wind directions; although we are less confident about these measurements (n ¼ 2), they parallel other landforms and are therefore shown as dashed arrows in Figure 3. Following Perkins et al. (2019), regional wind and temperature data were obtained from the NASA MERRA reanalysis data set (https:// disc.gsfc.nasa.gov/) for a period spanning the windy seasons (May to October) of the years 1982-1992. This decade, the earliest in the data set, was chosen to minimize any effects of anthropogenic climate change. Hourly wind speeds and downwind directions at a height of 2 m are plotted in the inset of Figure 3. Daily statistics for surface temperatures at a height of 2 m are plotted in Figure S1.
We mapped alluvial surfaces and terrace remnants in the depression using the ALOS World 3-D (1 arc sec) digital surface model (DSM) provided by JAXA and verified these interpretations with field observation. The depths of internally drained portions of the landscape were analyzed using TopoToolbox (Schwanghart & Scherler, 2014).

Structural Interpretation: Against Extensional Lowering of the Depression
Strata are deformed by a series of roughly N-S trending folds, with no basin-bounding normal faults ( Figure 3). The northern portion of the SdF consists of a roughly N-S trending, southward plunging anticline that exhumes Permian and Jurassic strata ( Figure 4). The fold axial plane dips to the east, and the Potrero Grande thickens rapidly, with dips decreasing up section on the western limb, consistent with syntectonic sedimentation (Figure 4). On the eastern limb of the fold, late Paleogene strata are truncated by the unconformity at the base of the Potrero Grande Formation. Chronostratigraphic data confirm that the flat interior of the depression here is composed of the lower member of the Quiñoas Formation, with K-Ar ages of 37.6 Ma of a volcanic tuff and 36.7 Ma on a reworked pumice clast (Adelmann, 2001). This  Voss (2002), and Reijs and McClay (2003). Dated ashes and their locations are from Adelmann (2001). Inset plots regional wind speeds and downwind directions from NASA MERRA hourly reanalysis data (https://disc.gsfc. nasa.gov/) during the windy seasons of 1982-1992, with winds >18 m/s shown as bold dots. Pink segment shows orientations of yardangs mapped in Figure 6c. interpretation is supported by U-Pb zircon ages from Canavan et al. (2014), who dated six volcanic tuffs outcropping in the eastern wall of the depression, 3.5 km north of Section A-A′, with ages ranging from 36.1 to 34.4 Ma.
The central portion of the SdF consists of a series of gentle anticlines separated by a doubly inward-plunging syncline ( Figure 5). Small kink-band folds mapped as monoclines are present on the western limb of the westernmost anticline, where strata are locally overturned. The Vega Chacras canyon on the western limb of this anticline is well studied, with a chronostratigraphic record comprised of 17 ages in a transect across what had been interpreted as rotated fault blocks (Adelmann, 2001;Canavan et al., 2014;Quade et al., 2015).  (2001), Canavan et al. (2014), and Quade et al. (2015). Strikes/dips measured in the field supplemented with those from Voss (2002) and Reijs and McClay (2003) as dark gray symbols. Stratal ages range from 18.0 Ma at the top of the canyon (K-Ar of volcanic glass) to 38.5 Ma near its base (U-Pb of volcanic zircon). These ages are consistent with uplifted, west dipping, Cenozoic strata on the western limb of an anticline ( Figure 5), rather than a late Miocene normal-fault growth sequence (Reijs & McClay, 2003). We found no field evidence for repetition of strata in this location, and stratal ages show no offsets across what had been interpreted as normal fault blocks ( Figure 5).
The relative age of strata in the floor of the depression is a key prediction of both hypotheses (Figure 2). Strata outcropping throughout the floor of the depression have been dated by multiple studies. Quade et al. (2015) obtained an age of 38.5 ± 0.7 Ma from an interbedded tuff near the base of Vega Chacras canyon, while Canavan et al. (2014) obtained ages of 35.7 ± 0.9 and 33.9 ± 0.9 Ma from an outcrop in the southern portion of the depression. Adelmann (2001) obtained ages of 37.6 ± 0.3 and 36.7 ± 0.3 Ma from bedrock exposures in the floor of the northern part of the depression. These studies indicate that the floor of the depression is composed of relatively old bedrock, the Lower Member of the Quiñoas Formation (late Eocene to early Oligocene). The lithology of this unit consists playa mudstones, evaporites, siltstones, and sandstones (Kraemer et al., 1999) up to 2 km thick in the SdF (Figure 4). Although the depression does contain thin Neogene terrace deposits (Figures 4 and 5, see below), these deposits rest in angular unconformity on older units and are not consistent with structural lowering of the depression. It is possible that smaller extensional structures are hidden beneath the modern evaporite playa, but exposures of the Quiñoas Formation would necessarily limit any such structures to a small area, and they would not be capable of explaining the origin of the depression.
These observations indicate a purely contractional structural setting for the SdF, rather than an extensional or transtensional one (Hypothesis 2, Figure 2). The eastern dip of fold axial planes is consistent with fault propagation folding, possibly due to blind, west vergent thrust faults associated with a major reverse fault system to the east (Adelmann, 2001;Seggiaro et al., 2007;Voss, 2002). The two main N-S trending anticlines in the NW and SE parts of the study area have a wavelength of~15 km, and their geometry mirrors the rhomboidal planform geometry of the depression (Figure 3). Compressional folding and subsequent erosion thus provide a compelling explanation for the geometry of the depression.

Little Evidence for Fluvial Incision
If the SdF depression is not formed by structural lowering of the basin floor, then it must have been carved by erosive processes. We rule out the possibility of glacial erosion in this hyper-arid region because no glacial or paleoglacial landforms are present in the Salar de Antofalla region (Haselton et al., 2002). We therefore examine the relative roles of fluvial and wind erosion. There is evidence of minor fluvial reworking within the depression. For example, several terraces cut into tilted Paleogene strata are present within the depression and capped by deposits 10-120 m thick (Figure 3). The terrace deposits (lacustrine and alluvial strata with gypsum-rich soil horizons) overlie tilted Paleogene bedrock in angular unconformity ( Figure 6a) and were deposited in an existing topographic low (Voss, 2002). The terraces have been partially eroded, and their remnants now stand in relief above the SdF floor, often as yardangs ( Figure S3b). Interbedded ashes in the terrace deposits dated to 2.6, 4.0, and 8.2 Ma (Adelmann, 2001) imply that significant erosion of the SdF occurred both before 8.2 Ma (to create the initial depression) and since 2.6 Ma (to erode the terraces and further deepen the depression). Erosional episodes were evidently punctuated by periods of net deposition or soil development.
The depression is not presently part of an integrated fluvial network. Making the highly conservative assumption that the depression was originally a fluvial valley and that wind erosion is only responsible for the portion of the internally drained topography provides an estimate of the minimum amount of wind erosion. The lowest point on the southern margin of the depression is blocked by an escarpment that extends >212 m above local base level and does not appear to have been uplifted structurally (Figure 7). Wind, which is capable of excavating internally drained depressions, is thus interpreted to be responsible for a minimum of 212 m of erosion (Figure 7). However, we find it unlikely that the depression was a fluvial valley in the past and argue for a larger magnitude of wind erosion. If the depression had been externally drained when the internal Neogene terraces were deposited, they would decrease in elevation to the south (i.e., downstream); in contrast, they decrease in elevation to the north, toward the center of the depression. Terrace deposit facies are also indicative of deposition in an internally drained basin in an arid climatic regime similar to the present. Finally, late Miocene-Pliocene units ouctropping on the eastern border of the Salar de Antofalla are sourced from, and thicken toward, the Sierra de Calalaste to the east, rather than the SdF to the west (Adelmann, 2001). It is therefore unlikely that the SdF was drained fluvially during the last 8.2 Ma (the oldest dated ash in a terrace deposit).
While we argue that wind was responsible for excavating the internally drained depression, we do not discount the importance of fluvial processes for redistributing sediment within the SdF. Fluvial canyons and gorges are incised into bedrock along the margins of the SdF, most notably at Vega Chacras, Potrerillo, and Potrero Grande ( Figure S2). These channels have headwaters in grassy marshes and are fed by groundwater seepage and possibly by overland flow during rare precipitation events. The drainage network around Potrero Grande, in particular, is responsible for significant alluvial and fluvial deposits in the NE part of the depression. In the southeastern corner of the depression, small rills and gullies incise gravelly surfaces capping bedrock terraces ( Figure S3b). Although fluvial erosion is capable of widening a closed depression, eolian entrainment of the resulting deposits is required to increase its depth, as rivers cannot erode below the base level set by their outlet. Fluvial processes are likely important for detaching sediment and transporting it to the floor of the depression, where clasts are vulnerable to salt weathering and particles can be entrained by wind. Eolian entrainment of fluvial deposits also provides a source of abrasive sand for eroding bedrock surfaces downwind. As is clear from the bedrock terraces described above, deflation of material from the basin has outpaced deposition, resulting in deepening of the depression over time. The lacustrine, alluvial, and fluvial deposits capping terraces indicate that excavation of the SdF was episodic and punctuated by periods of net deposition. These episodes were likely driven by changes in climate, as has been suggested for internal terraces in depression in the Negev Desert, Israel (Plakht et al., 2000).

Evidence in Favor of Wind Erosion
Rather than extensional tectonics or fluvial erosion, we argue that wind is primarily responsible for forming the depression. There is abundant evidence for both eolian abrasion and deflation within and around

Journal of Geophysical Research: Earth Surface
the SdF ( Figure S2; Goudie & Wells, 1995). Yardangs 10-100 m in length with~1-5 m of relief are carved into lower Quiñoas Formation where it dips gently parallel to the wind direction (Figure 6c). Yardangs are streamlined eolian erosional landforms whose size and shape are controlled by the bedding orientation and heterogeneity of host strata Pelletier et al., 2018) as well as lithology. Five kilometers south of the yardangs, the Quiñoas Formation dips perpendicular to wind direction and outcrops as rough, sharp-crested hogbacks and smooth elongated ridges with length/width aspect ratios up to~13 ( Figure S4c). These ridges occur over a larger area than do the yardangs and also indicate the action of eolian abrasion and deflation scouring the basin floor.
Eolian megaripples are locally present with wavelengths of 1-3 m and heights of 10-20 cm (Figures 8 and S4a). The origin of coarse-grained megaripples is debated along with their classification as migrating (Hugenholtz et al., 2015;Milana, 2009) or static bedforms (Bridges et al., 2015;de Silva, 2010;de Silva et al., 2013). Observations of the internal structures and morphology of megaripples are broadly consistent with their downwind migration driven by saltation-induced creep (Lämmel et al., 2018). Aerial imagery reveals extensive fields of megaripples that developed on subhorizontal surfaces capped by conglomeratic deposits (Figure 8b). The planforms of these megaripples show evidence of migration similar to that observed by Hugenholtz et al. (2015), with orientations indicating NNW wind directions, consistent with other eolian landforms ( Figure 3). The megaripples observed in the SdF thus give additional evidence of strong winds and the presence of abrasive, saltating particles that drive their downwind migration. Their downwind migration also exposes fresh bedrock (orange surfaces visible in Figure 8), providing a means for disrupting stony lag deposits that would otherwise protect bedrock against further erosion (Abell et al., 2020). Stony deposits are also disrupted by small gullies and rills that expose bedrock surfaces in the SE part of the depression ( Figure S3b).
The sides of the depression are prominent rectilinear bedrock slopes with nearly uniform~27°angles ( Figure S2). The slope angle does not appear to be related to bedrock strength, as no evidence for slope failures is present. The possibility that these slopes are major fault scarps can be ruled out by our structural interpretation. Major erosional rills or gullies in these slopes are rare, occurring only around localized seeps and springs, and the slopes are very smooth overall, but small linear gullies and associated debris aprons are present ( Figure S4b). Most of these slopes have not accumulated talus at the base, where bedrock is often exposed ( Figure S4d). Others appear to have shed a thin layer of debris into the depression (contributing the terrace deposits described above). Because the slopes drain into a closed depression, wind erosion is likely required to explain the lack of colluvial and alluvial debris that would accumulate at the base. The rectilinear profile indicates that the slope retreat is weathering-limited; that is, debris is removed faster than it can accumulate. Given the limited occurrence of major erosional features on the slopes, it is possible that some portions of them retreat mainly by eolian deflation aided by salt weathering, as proposed for rectilinear bedrock slopes in the Dry Valleys of Antarctica (Selby, 1971). Infrequent precipitation events may also be effective at transporting debris downslope.
Zones of active of salt weathering, previously noted by Goudie and Wells (1995), were observed around the SdF. Salt presently accumulates around the margins of modern pans and evaporites evidently accumulate on subsurface discontinuities such as unconformities (Figures S5 and S6). Salt crusts are visible in aerial imagery in many locations around the margin of the SdF (Figures S4b, S4c, and S5). We observed tafoni on the windward face of an ignimbrite outcrop located downwind of an evaporite pan NW of the SdF Figure 7. Internal drainage of the Salina del Fraile. The internal drainage depth approximates the minimum amount of wind (versus fluvial) erosion required to generate the observed topography. The transect follows the lowest-elevation path to the Salar de Antofalla. Internal drainage depth calculated using "fillsinks" function of TopoToolbox; transect calculated by routing flow using the "carve" preprocessor, which chooses the lowest-elevation path across topographic highs (Schwanghart & Scherler, 2014).

Journal of Geophysical Research: Earth Surface
( Figure S7). Given these observations, we infer that salt weathering in the SdF provides a means of breaking down Cenozoic bedrock into particles that can be more easily entrained by wind.
Near its western rim, the SdF hosts large, smooth, elongated ridges, 5-10 km in length and up to 600 m tall (Figure 3), elsewhere interpreted as fault scarps (Reijs & McClay, 2003). Cross Section B ( Figure 5), however, indicates that these ridges are erosional. They are not parallel to wind directions inferred from smaller-scale landforms, but their long axes generally parallel the strike of underlying strata. The ridges thus reflect the strong influence of bedding orientation and lithology superimposed on erosional processes in the SdF. Their dip slopes correspond to the upper Chacras Formation, a unit of relatively resistant cross-bedded sandstones and conglomerates that contrasts with overlying Potrero Grande volcaniclastics (Adelmann, 2001). We hypothesize that these resistant strata slowed the excavation of the depression, leading to elongated erosional remnants that parallel the strike of bedding planes; these topographic remnants were likely smoothed by weathering, colluvial transport, and eolian deflation.
Five kilometers southeast of the SdF, three~300 m tall mesas with smooth, linear slopes are present (Figures 3 and 6b). They are 5.5-10.5 km 2 in area with curvilinear planforms 2.5-6 km long that taper slightly to the SE. The tops of the mesas are subhorizontal and characterized by heavily abraded bedrock surfaces ( Figure S3d). Although the mesas resemble alluvial terraces in some respects, no evidence of fluvial channels or paleochannels is present. Instead, they correlate with a regional, subhorizontal, Late Miocene-Pliocene surface representing the top of the Potrero Grande Formation and/or Late Miocene volcanic flows in this area (Richards et al., 2006) and are therefore composed of Potrero Grande volcaniclastics.
Although it is difficult to imagine how fluvial or tectonic processes could produce these mesas, our wind erosion hypothesis provides a potential explanation. The mesas are larger and smoother than megayardangs described in indurated ignimbrites nearby on the Puna Plateau and in Northern Chile Goudie, 2007) but are similar in planform shape to megayardangs in Crommelin crater, Mars, that developed in canyon networks reworked by wind erosion (Perkins et al., 2015, and Figure 1b therein). As evidenced by the lack of boulder-sized colluvium, the lithologic composition of the smooth mesas is more friable than well-indurated ignimbrite and not capable of supporting the near-vertical slopes characteristic of some megayardangs . The volcaniclastic sediments may also be more prone to disintegration by weathering and deflation. Like terrestrial megayardangs, the smooth mesas most likely result from a combination of processes, including colluvial and alluvial diffusion, but their overall forms are indicative of eolian streamlining and deflation. The base of the smooth mesas corresponds to the contact between Potrero Grande Formation volcaniclastics and Chacras Formation conglomerates, an erodibility barrier, while the top of the mesas correlates to a regional subhorizontal surface composed of thin volcanic flows resting on top of the Potrero Grande Formation (Richards et al., 2006;Voss, 2002), which could act as a caprock. Mesa height (~300 m) is roughly equivalent to the thickness of the Potrero Grande Formation. The mesas are separated by smooth troughs elongated in the direction of the dominant wind direction. The troughs are floored by abraded bedrock surfaces, mostly composed of Chacras conglomerates, with meter-scale erosional and fault scarps ( Figure S3d). The planform shape of the mesas is elongated and tapers slightly in the downwind direction, and their windward slopes appear to be smoothed and streamlined. The mesas are morphologically similar to streamlined hills and islands described in megaflood landscapes (e.g., Baker, 2009). Given the lack of evidence for megafloods in the study area, we suggest that the mesas formed by eolian downcutting into an erodible layer (Potrero Grande volcaniclastics) underlying a resistant caprock, a model similar to one proposed for the interior of Gale crater, Mars (Day & Kocurek, 2016).
The mesas are deformed by a swarm of penetrative normal faults with 400-800 m spacing that display smaller scarps in the troughs separating the mesas than on their top surfaces ( Figure S3d). Twenty kilometers to the southwest, this fault swarm deforms volcanic flows dated to 3.61 Ma (Adelmann, 2001), indicating that a significant portion of the extensional deformation was accommodated by faulting after 3.6 Ma. This timeframe agrees with the one we infer for wind erosion, suggesting that the smaller scarps observed in the troughs between the mesas is due to preferential erosion of the troughs during active faulting.
Predominant wind directions inferred from the long axes of yardangs, megaripple crests, and other landforms match closely with the downwind direction of strong winds obtained from the NASA MERRA reanalysis data (Figure 3, inset). Although the hourly MERRA data do not capture extreme wind gusts, which likely exceed 100 m/s in the southern Puna (Milana, 2009), they support the notion that these features result from strong, unidirectional, relatively infrequent winds. The data also suggest that the wind regime in the region has been stable throughout the late Quaternary and similar to the present, with strong winds directed toward the SE.

Timing and Rate of Erosion
Structural cross sections, drawn conservatively to minimize the amount of erosion, show that up to 1,170 and 1,950 m of strata were removed from the northern and central SdF, respectively (Figures 4 and 5). Existing chronostratigraphic ages provide constraints on the timing and rate of this erosion (Figure 9a). The initiation of erosion is bracketed by the top unit exposed in the depression, the Potrero Grande Formation, 17.0 Ma (Adelmann, 2001, K-Ar of biotite), and by the oldest volcanic ash deposited within the floor of the depression, 8.2 Ma (Voss, 2002, Ar-Ar of biotite). The depression therefore must have already been partially excavated by 8.2 Ma. Streamlining and erosion of terraces containing 2.6 and 4.0 Ma ashes in the floor of the depression (Voss, 2002, Ar-Ar of plagioclase) also indicates significant erosion since the Pliocene. Yardangs and megaripples, active salt weathering zones, a lack of major sedimentary fill, rectilinear bedrock slopes, and imaged dust storms all indicate that erosion is currently active in the region (Goudie & Wells, 1995;Milana & Kröhling, 2017).
Given that erosion began between 17 and 8.2 Ma and is presently ongoing, we estimate erosion rates using the structural cross sections presented in section 3.1. A maximum thickness of eroded material of 1.95 km measured from section B (Figure 5), yields maximum erosion rates of 0.11-0.23 mm/yr in the central part of the depression. An average of 1.05 km of strata have been eroded from Cross Sections A and B (Figures 4 and 5). Using this as a spatial average over the area of the basin floor, wind has removed 315 km 3 of material from the depression since 17-8.2 Ma, averaging 0.062-0.13 mm/yr. Fluvial drainage networks ( Figure S1) presently cover approximately 10% of the surface in the SdF. These erosion rates may overestimate the contribution of wind erosion of bedrock by up to 10%, but any fluvial sediment ultimately requires wind to be removed from the depression. While these estimates are long-term (~10 7 yr), they fall within the range of rates reported elsewhere for wind erosion over time scales of millions of years ( Figure 9b) and are 2-10 times lower than those calculated from cross sections in the Qaidam basin, China, involving wind erosion of folded Pliocene strata (Kapp et al., 2011;Rohrmann et al., 2013). The

10.1029/2020JF005682
Journal of Geophysical Research: Earth Surface spatially averaged erosion rate of~0.06 mm/yr estimated here is very similar to the average erosion rate estimated for the Hami depression in northern China (Figure 9b), which involved deflation of Neogene strata containing conglomeratic and sandstone layers (Pullen et al., 2018).

Implications
The erosional origin of the SdF depression has implications for the late Cenozoic tectonics of the Puna Plateau. Previous researchers suggested a Pliocene phase of orogen-parallel sinistral strike-slip kinematics based on the morphology of the depression, with up to 7.7 km of offset on a purported SdF fault system (Reijs & McClay, 2003). Such an interpretation, however, challenges measurements of nearby fault kinematics that imply orogen-parallel dextral deformation during this time (Allmendinger et al., 1989;Kraemer et al., 1999;Zhou et al., 2015). Our inference that the depression is erosional obviates the need to invoke orogen-parallel sinistral tectonics and implies that Pliocene to Recent extensional deformation was accommodated by the normal faults apparent in Figure 3, which may be splays off the dextral Acazoque fault system that terminates near the Salar de Antofalla (Kraemer et al., 1999).
The initiation of wind erosion between 17 and 8.2 Ma is broadly consistent with a documented phase of aridification on the Puna Plateau partially driven by the uplift of ranges east of the study area (Strecker et al., 2007), such as the Sierra Laguna Blanca (exhumed 15-10 Ma; Zhou et al., 2017). Late Miocene aridification is also partially attributed to global cooling forced by a decline in atmospheric CO 2 (Herbert et al., 2016). Global aridification and cooling during the Late Miocene is reflected in ocean temperature proxies and in various moisture and temperature proxies compiled throughout Eurasia (Miao et al., 2012). Large-scale eolian deflation and excavation of topographic depressions, such as the SdF and the Hami basin in central Asia (Pullen et al., 2018) may represent a topographic signal of this episode of global climate

10.1029/2020JF005682
Journal of Geophysical Research: Earth Surface change. Alternatively, existing data also permit the interpretation that erosion began as soon as the package of friable strata was tectonically deformed and exposed at~17 Ma.
The idea that high-relief topography can be generated and sustained by eolian processes, with magnitudes of deflation up to 2 km, has implications for similar landforms on Earth and other planetary surfaces. The Central Andes hosts several depressions 10-50 km in width that, while little studied, may have been partially formed by wind erosion (Figure 1). Previous research indicates that the Purulla Basin (Figure 1, Location B) may have experienced an average deflation of roughly~4 m during the Holocene, approximately 0.3 mm/yr (Milana & Kröhling, 2017). Although more work should be done to quantify erosion rates in wind-dominated areas of the Puna, the SdF suggests that the magnitude of eolian erosion throughout the Cenozoic may be significant in these internally drained regions lacking major fluvial incision.
Many Martian landforms are thought to be formed by wind erosion over long timescales. Mounds of sedimentary strata located in the centers of impact craters preserve records of past climates and are important exploration targets (Anderson & Bell, 2010). A substantial amount of wind erosion (up to a few kilometers) is necessary to form the present morphology of many central mounds, whether they are remnants of crater-filling material (Bennett & Bell, 2016;Steele et al., 2018), or they formed in-place (Kite et al., 2013). The lower flank of Mount Sharp, the central mound of Gale crater, consists of erosional buttes, mesas, and canyons that strongly resemble the smoothed mesas and elongated ridges described above (Figure 10).
Mudstones on the floor of Gale crater have exposure ages of only 78 ± 30 Ma, likely due to wind-driven bedrock scarp retreat (Farley et al., 2014). This relatively recent exposure increases the likelihood that complex organic molecules are preserved in near-surface materials. Identifying regions of rapid wind erosion is therefore critical for future missions exploring potentially habitable environments on Mars (Chojnacki et al., 2018). Like Gale and other Martian craters including Jezero, the proposed landing site of the Perseverance rover, portions of the interior of the SdF are characterized by ongoing eolian erosion of lacustrine and alluvial strata that once partially filled the depression (Chojnacki et al., 2018;Palucis et al., 2016). The SdF thus represents a terrestrial analogue to Martian landforms that are the target of exploratory missions.

Conclusions
Structural and chronostratigraphic data show that the SdF depression was excavated by erosion of up to 2 km of friable, Paleogene strata exposed in compressional folds. Wind, rather than water or ice, was the dominant agent of erosion during at least the past 8.2 Ma. This inference is based on depositional ages of alluvial and lacustrine deposits contained within the depression, the prevalence of erosional eolian landforms within and around the depression, the lack of integrated fluvial networks, and the regional geomorphic context, which is presently dominated by wind erosion.
Inferred erosion rates of 0.06-0.23 mm/yr are typical of arid regions on Earth, but the large magnitude of erosion reflects the long time period (>8.2 Ma) during which eolian deflation took place. The SdF documents a Late Miocene to Quaternary source of eolian sand and dust in the southern Puna Plateau and exhibits some of the largest eolian landforms on Earth. In addition to the SdF depression itself, these include kilometer-scale elongated ridges and smoothed mesas resembling megayardangs and eolian landforms on Mars. Due to lithologic variation and tectonic folding, which produce erodibility contrasts, these high-relief landforms can be sustained without significant fluvial or glacial incision. Although the SdF is relatively well studied, similar internally drained depressions in the Central Andes and other regions may also have been partially formed by wind erosion. Study of these landforms may yield insights into the evolution of Earth's arid regions on timescales of 10 6 -10 7 yr, including how tectonic deformation and climate interact in such landscapes.