Improving quantitative ventifact analysis for climate investigations using the Dyngjusandur sandsheet in Iceland as a planetary analogue Open Access

Ventifact formation is a function of original rock shape and texture, duration of exposure, orientation relative to wind direction and time of emplacement. Research by Bridges et al. (1999) indicated that original rock texture did not play a significant role in ventifact feature development based on observations of terrestrial ventifacts and ventifacts found at the Mars Pathfinder site.

Following the terminology of Knight (2008), we refer to macroscale ventifact features (i.e. less than a few decimetres in diameter) as facets, characterized by the number and orientation of faces developed on the rock surface (Fig. 2c). Facets form from material being abraded away on the windward (exposed) side of the clast (Knight 2008) through grain saltation, which Laity and Bridges (2009) showed is strongly controlled by the angle of the ventifact face with respect to wind direction. Wind tunnel tests show that steep (c. 90°) facets were abraded at a much higher rate than lower angle (c. 30°) facets, as the steeper face absorbs more energy than shallowly dipping surfaces (Bridges et al. 2005).

Mesoscale features (i.e. less than tens of centimetres in diameter) that form on the windward surface of a ventifact by impacting abradant vary spatially depending on the original shape and orientation of clasts (Knight 2008). Pits occur when saltating grains exploit existing weaknesses within the rock to dislodge individual grains or crystals from the facet surface (Fig. 2b). These form perpendicular to wind direction and are typically small (<1 cm), extending into the rock face at right angles (Knight and Burningham 2003). Over time, a pit can become elongated and form a groove, oriented parallel to the direction of wind flow, and closely spaced grooves can merge to create a corrugated texture (Fig. 2b and d). Flutes form on the windward side of ventifacts and are asymmetric in long profile, with conical shapes oriented in the downwind direction (Greeley and Iversen 1987). Flutes have more resistant material at the upwind tip, such as a crystal or lithic fragment, that erodes more slowly and protrudes from the surrounding matrix, creating a protective shadow downwind and resulting in a cone-shaped projection above the surrounding lithic surface (Favaro et al. 2017).

Terrestrial ventifact formation mechanisms can inform our understanding of the origin of ventifacts found on other planetary bodies, where ancient climate and wind regime conditions are poorly constrained. The Mariner 9 orbiter (1971) and Viking lander (1976) missions to Mars identified wind as an important driver of geological processes and returned images of rocks with etching and grooves that aligned with the prevailing northern wind (McCauley et al. 1980). During the Mars Pathfinder (1996) mission, about half the rocks at the Ares Vallis landing site appeared to be ventifacts exhibiting evidence of aeolian abrasion with similar features to terrestrial counterparts (faceted edges, pits and flutes with orientations aligned with prevailing wind direction) (Golombek et al. 1999).

Wind-related features, including ventifacts, were also observed by the Mars Exploration Rover (MER-A), named Spirit. The Spirit rover captured images that led to the identification of 1520 orientation features (e.g. bedforms and ventifacts) (Fig. 1b) that indicated primary winds from the NNW, which agreed with the interpretations made from orbiter images and atmospheric models (Greeley et al. 2006). Cuttings from the MER-A Rock Abrasion Tool (RAT) indicated actively blowing winds from the NW, which was consistent with ventifact orientation and inferred wind direction from dust devil tracks seen in orbital images (Greeley et al. 2004, 2006).

The Mars Science Laboratory (MSL) Curiosity's traverse from Bradbury Landing to Rocknest identified 50 rocks as ventifacts within Gale Crater (Bridges et al. 2014). The specific location, size and orientation of ventifacts was computed using Navcam stereo mosaics in the Mars Science Laboratory Interface (MSLICE) software (Crockett et al. 2011). Orientation of facets and broad-scale features were determined for 48 of the 50 ventifacts with an average of three measurements per ventifact (Bridges et al. 2014). Ventifact features indicate bimodal wind directions with a long-lived NW trend and an older SW trend (Bridges et al. 2014).

Ventifacts were also identified by the Mars 2020 Perseverance rover on the floor of Jezero Crater (Farley et al. 2020). Images acquired by the onboard Navcam (Maki et al. 2020; Maki 2021) and Mastcam-Z (Hayes et al. 2021; Paar et al. 2023) instruments were used to produce 3D digital ordered point cloud (OPC) datasets using the Pro3D software package (Barnes et al. 2018). Findings showed that ventifact abrasion features recorded an ancient wind direction that differs significantly from measurements of current and recent wind directions, indicating a change in climate regime (Herkenhoff et al. 2023).

There are many potential candidates for abradant material, including high-powered winds with entrained snow, ice (Whitney 1978), dust (Dietrich 1977) and/or sand (Cutts and Smith 1973; Edgett and Christensen 1991). Numerous studies have investigated which materials are capable of abrading rock, and under what conditions, and results indicate that sand, which is ubiquitous on Earth as well as on Mars, is the only medium with adequate mass to physically erode rock (Laity and Bridges 2009).

To understand the effectiveness of sand as an abrasive agent, Laity and Bridges (2009) conducted a high-speed video impact study that used SEM to analyse surface texture of abraded material within a wind tunnel. Findings showed that although sand may have a lower velocity compared with dust, it has 1000 times more mass, making sand a more effective abradant. They conclude that sand is the principal, and probably only, source of abradant on Mars and showed that sand abrasion under unidirectional flow will erode and shape only the windward face of the rock, preserving direct environmental records of long-term palaeo wind direction.

Regional volcanic activity in the Askja/Holuhraun region has emplaced blocky basaltic lava flow-fields and subglacial hyaloclastites, as well as significant amounts of tephra, including ash and pumice (Sparks et al. 1981). The Askja caldera is part of the Dyngjufjöll fissure system located on the Northern Volcanic Zone (NVZ) of Iceland (Fig. 3a). The volcanic system has been active throughout the Pleistocene (Brown et al. 1991) and the most recent Askja eruption in 1961 emplaced a lava flow-field covering an area of 11 km² (Thorarinsson and Sigvaldason 1962; Blasizzo et al. 2022). The Holuhraun eruption site, located c. 18 km to the SE of the ventifact field site that is the focus of this study, produced lava flows that covered 83.83 km² (Voigt et al. 2021) (Fig. 3a).

The Dyngjusandur sandsheet comprises a continuous area of 270 km² and consists of large swaths of wind-deposited black sand between c. 10 cm and c. 10 m thick (Mountney and Russell 2004) (Figs 2a and 3). The striking dark colour of the sandsheet is a result of the abundance of mafic-sourced glass, minerals and rock fragments, in contrast to other terrestrial sands, which are usually much lighter, owing to quartz-, calcium carbonate- or dolomite-dominated compositions (Mangold et al. 2011). The sands are sourced from basaltic protoliths and composed of fragments of glass, minerals including olivine, pyroxene and plagioclase, and lithic fragments of microcrystalline groundmass of lava and hyaloclastite (Mangold et al. 2011; Sara 2017). Volcanic glass grains account for 80–90% and basaltic minerals (olivine, pyroxene, plagioclase) account for the remaining 10–20%. Grains are dominantly <0.5 mm in diameter, and sphericity is largely homogeneous with values ranging between 0.71 and 0.88 (Sara 2017). Major and trace element compositions of Dyngjusandur sand are largely homogeneous with MgO and Fe₂O₃ contents of 6–9 wt% and 11–15 wt%, respectively (Sara 2017). Determining sediment transport threshold velocities requires quantitative modelling of transport regime and ambient conditions (Tholen et al. 2022), which is complex in even simple environments (Pähtz et al. 2020). Two nearby studies, at Holsfjöll (80 km to the NW) and Geitasandur (35 km to the west) show threshold velocities of 120–670 kg m⁻¹ a⁻¹ with mobilization heights of >60 cm (Arnalds et al. 2012).

The sandsheet divides into three distinct regions based on the distribution of aeolian landforms (Mountney and Russell 2004) (Fig. 3a): (1) the upwind portion of the sandsheet is a 70 km² area that is adjacent to the northern margin of the Dyngjujökull outlet glacier and is heavily influenced by glacial meltwater; here, fluvially transported material ranges from fine-grained, plane-bedded sands to pebble- and cobble-sized deposits, as well as ventifacted boulders; (2) an accumulation zone in the central portion of the sandsheet is a roughly 100 km² area; accumulation of aeolian sand can reach 10 m in thicknesses and multiple scales of aeolian ripples are present ranging from 1 mm to 15 cm in height (Sara 2017); (3) a 100 km² downwind portion to the north and NE of Askja has localized accumulation in topographic lows across basaltic lava fields. Extensive meteorological data ranging from 1969 to 1999 from the Brú á Jökuldal weather station located c. 45 km to the NE of Askja officially classify the area as semiarid, based on classification of Meigs (1953), as receiving between 250 and 500 mm of annual precipitation, and show a mean annual windspeed increase from 2.6 to 4.2 m s⁻¹ over the period of 1962–1999 (Mountney and Russell 2004). Characterization of aeolian and related landforms (i.e. wind ripples, plane beds, grain flow strata and deflation surfaces) by Mountney and Russell (2004) was used to map the direction of wind flow moving from the SW to the NE (Fig. 3). Extending the record from 1998 to the present indicates that the region continues to be semi-arid, experiences a mean temperature of 1.87°C, and windspeed has continued to increase with time to a current average windspeed of 4.73 m s⁻¹ (Fig. 4). Windspeed data for 2000–2020 from the Upptyppingar weather station located at the distal end of the sandsheet, 20 km NE of the ventifact field (Fig. 3a), provide a mean windspeed of 5.5 m s⁻¹ with wind gusts reaching a maximum velocity of 50.3 m s⁻¹ (Icelandic Meteorological Office 2024).

Owing to the periglacial nature of the sandsheet, sediment supply is highest in summer months when glacial ablation is at its maximum. Downstream transmission losses of water owing to percolation result in accumulation of sand along the glacial margin that serves as the main supplier of sediment to the upwind region. Sediment supply during cold weather months is hindered by the development of surface frost, which acts to bind sediment, effectively halting transport (Koster 1988). Weathering of exposed basaltic lava, as well as hyaloclastite and scoria in outcrops surrounding and underlying the region, also contributes to the overall sediment budget (Baratoux et al. 2011).

The 1.0 × 0.5 km ventifact field that is the focus of this study is located within the central accumulation zone of the sandsheet (Fig. 3), where the estimated 1625 ventifacts were probably deposited by glacial outburst flooding. Tephrochronology of ash layers from the Askja region indicates that the area became ice-free at around 10.3 ka, and the exposed lavas in the vicinity of the ventifact field are younger than 7.2 ka, based on a superimposition relationship with the Helka-5 tephra layer (Hartley et al. 2016), which constrains the maximum age of the ventifact field. Based on an estimate of the amount of flood energy necessary to move large boulders, Mountney and Russell (2004) suggested that the ventifacts found at the Dyngjusandur sandsheet have developed over periods of 10²–10³ years.

We conducted a survey of ventifacts within a 1.0 × 0.5 km (Fig. 3b) area. Each ventifact's geographical coordinates were identified with a Garmin eTrex 22x handheld Global Positioning System (GPS) unit, with an accuracy of ±5 m, in open sky settings (Wing et al. 2005). To increase the fidelity of the survey, we also performed an aerial photogrammetric survey. High-resolution images (20 megapixels) were collected using a DJI Mavic Pro 2 Unoccupied Aircraft System (UAS) flown at an altitude of 75 m above ground level (AGL). For extended coverage, we flew a larger area at 100 m AGL. The coordinates of four evenly and homogeneously distributed ground control points (GCPs), each consisting of a black and white reflective target (1 m²) placed on the ground in the field, were determined using a Trimble R10 survey-grade Global Navigation Satellite System (GNSS) receiver and base station, allowing differential GPS corrections using a real-time kinematic (RTK) workflow (Scheidt and Hamilton 2021). Postprocessing of the images was performed using Agisoft Metashape™ Professional (version 2.1.1), and the GCPs were used to control the geographical positioning of the resulting orthomosaic image of the ventifact field. The average root mean square error (RMSE) between GNSS and image targets was 0.02 m. The resulting orthomosaic image of the ventifact field improved constraints for location and orientations for an estimated total of 1625 ventifacts. In situ measurements were performed using a handheld Brunton compass on 40 targeted ventifacts selected as study samples distributed across the ventifact field. Ventifacts were selected using a uniform selection process to minimize any bias in selecting individual ventifacts for study. The field was divided into 10 m² sections using a grid system overlain on top of a Google Earth image to ensure even coverage over the entire area. Ventifacts were selected for analysis by identifying the centre of each 10 m² section and choosing the closest ventifact to that position. Prior to making measurements, each sample was coded, its location was recorded using GNSS measurements, and it was photographed with a scale ruler, sketched and described. Descriptions noted specific shape, lithology and large planar features, as well as the presence, abundance and size of vesicles, phenocrysts and ridges. Strike and dip planar Brunton measurements were collected on windward ventifact faces (Lisle 2003) assessing orientation of macroscale abrasion relative to wind direction (Laity and Bridges 2009). The number of planar measurements made was largely dictated by the ventifact size and shape, and ranged from two to 10 measurements, with an average of 6.4 planar measurements for each of the 40 ventifacts (n = 500 total). In instances where increased wind abrasion created curved surface features, multiple measurements were made to encompass the more complex features. Measurements were made across the entire windward surface of the rock face in 10–20 cm increments, with trend and plunge orientation data collected along the orientation of mesoscale troughs and ridges, when present. The number of measurements corresponded to the abundance of mesoscale ventifact features, which was lithology dependent, and ranged from a high of 20 per ventifact for vesiculated and phenocryst-abundant lithologies to a low of four per ventifact for massive lithologies. To make accurate measurements of millimetre- to centimetre-scale vesicle and ridge features, it was necessary to adapt existing methods used to measure larger features following the procedures of Knight (2008). A sharpened 10 cm long rod was used to project an equivalent lineation extending the corresponding trough or ridge, allowing for the trend and plunge of the feature to be recorded more accurately.

From the 40 ventifacts measuring using in situ techniques, a subset of 20 ventifacts were imaged at close range in high resolution for 3D photogrammetric analysis. These were selected based on the presence of measurable macro- and mesoscale features, as well as being representative of the four lithological types present within the ventifact field (see ‘Ventifact lithologies’). Images were acquired using a Nikon D300 Digital Single-Lens Reflex (DSLR) camera with a 12.3 megapixel Complementary Metal Oxide Semiconductor (CMOS) sensor and a fixed focal length 50 mm lens. Three fiducial ground targets were placed at the base of the ventifact to provide quantitative metric scaling with known dimensions. These targets were 10 × 10 cm and have sub-markings showing 5 × 5 cm and 1 × 1 cm (Fig. 2). Targets were placed close to the ventifact base and distributed to cover c. 270° around the circumference of the ventifact to have more than one target visible in as many photogrammetric images as possible. To provide scale information for later scaling and orientation, distance between the centres of each of the three fiducials was recorded, the orientation of each of the cards in relation to the others was measured, and the trend and plunge of one edge of each card as it was oriented resting on the sand surface was also measured. Digital photographs were acquired at 5° increments in an array surrounding the ventifact (Fig. 5) with care taken to include at least two of the ground targets in each image ensuring 60–0% overlap between consecutive images to facilitate later alignment. Ventifacts greater than c. 30 cm high had two concentric rings of images, to best capture the details of the base and top, and ventifacts higher than c. 50 cm had three concentric rings of images (Fig. 5). Individual ventifacts had from 25 to 55 images each, depending on size. The entire photogrammetric imaging process, including measurement of fiducial ground targets, required from 15 to 20 min per ventifact.

Images of each ventifact were processed using Agisoft Metashape™ Standard (version 1.5.4) to generate spatially accurate 3D models that serve as an entirely digital photogrammetric dataset. Camera calibrations were performed using built-in calibration parameters and the Align Photos function was used to calculate camera position and generate a sparse point cloud to produce a low-resolution model of each ventifact. A dense point cloud was then generated using the Build Dense Cloud function, with quality settings set to High, and noise filtering applied to enhance accuracy. A detailed 3D mesh was then constructed from the dense point cloud using the Build Mesh function. Texture was then mapped onto the ventifact models using the Build Texture function, using mosaic blending modes to produce the most accurate, high-resolution, photorealistic surface details. Finally, the completed 3D modelled ventifacts were exported in PLY format for further analysis and visualization. Ventifact models were then uploaded to CloudCompare (version 2.11 alpha) to be scaled, properly oriented and measured (Fig. 6). The orientation of ground fiducial targets for each ventifact was utilized to orient the x–y–z axes of each ventifact in 3D space relative to its real-world orientation and was scaled by applying transformations based on the known size of the ground targets and their respective spacing. Measurements of ventifact features were undertaken using the lineation tool within CloudCompare, allowing for measurements to be made across the entire surface of the rock face (Fig. 6), referred to as ‘synthetic’. A total of 1145 planar measurements were made with an average of 57 measurements per synthetic ventifact. All data associated with this study have been archived at the US Geological Survey (Hudziak et al. 2024) and are accessible via https://doi.org/10.5066/P1SXIRVQ.

We use directional or circular data rose diagrams to compare and visualize these two linked but unique datasets (in situ and photogrammetric). Because these data represent the angle of ventifact feature orientation, traditional linear statistical analysis is not applicable owing to the inherent differences between a circle and a straight line (Hassan et al. 2009) (Fig. 7). Ventifact orientation features were analysed within the Oriana circular statistics program (Kovach 2011) to assess whole-field orientation relationships as well as to make more detailed comparisons between the two datasets. This method of statistical analysis outputs mean vector values, circular variance and circular standard deviation as well as 95 and 99% confidence intervals for all values (Table 1).

The Watson–Williams F-test (hereinafter referred to as F-test) was used to compare mean feature orientation values of the 500 in situ measurements v. the 1145 photogrammetric measurements to assess their significance and potential correlation. The F-test assumes that the samples are independent with a von Mises distribution (data are normally distributed around a unit circle) and n > 2. A P-value of <0.05 from the F-test indicates a rejection of the null hypothesis, indicating that the means are significantly different. If the calculated P-value is >0.05, then the null hypothesis is not rejected, indicating that the means of the two compared datasets have significant correlation and a value for the samples tested will be reported as an estimate of the overall population mean. Circular variance and circular standard deviation are equivalent to linear variance and linear standard deviation but require different steps to calculate. Circular variance is calculated using length of mean vector using the formula V = 1 − r (where r is mean vector length), which then allows for circular standard deviation to be calculated using $σ$ = [−2 ln(r)]^1/2. The resulting value for $σ$⁠, which is given in radians, is then converted to degrees by multiplying by 180/$π$ (Kovach 2011).

The ventifact field contains an estimated 1625 ventifacts and is situated on a flat, vegetation-free area (Figs 2 and 3). The ventifacts are evenly dispersed over an elongate area of 1.0 × 0.5 km, oriented east–west (Fig. 3b), with an average of 13 ventifacts per 20 m². Size of individual clasts varies widely from large, partially exposed to fully exposed boulders with long axes up to 3 m, to smaller clasts with 0.2–0.3 m long axes (Fig. 2). Cobble-sized clasts that displayed signs of ventifaction are present; however, determining original orientation of small clasts is unreliable because their position can be shifted by strong winds and other geomorphological processes (Hoare et al. 2002; Bridges et al. 2014), and, as such, all ventifacts that were selected for detailed analysis were greater than 0.2 m along the longest axis.

The ventifact field is composed primarily of basaltic clasts with a minority (≲5%) of hyaloclastite clasts. The 40 studied ventifacts can be grouped into four distinct lithological types based on crystallinity, vesicularity and texture (Fig. 2):

1. Vesicular basalt ventifacts (55%, n = 22) are grey in colour with a cryptocrystalline groundmass and exhibit vesicles ranging in size from <1 to 50 mm (Fig. 2b). Vesicle distribution ranges from homogeneous to bands of vesicles that range from 10 to 30 cm thick separated by massive texture. Ventifaction of the rock preferentially erodes heavily vesiculated areas and areas surrounding larger vesicles, resulting in the formation of distinct pits and grooves oriented parallel to prevailing wind direction.

2. Vesiculated phenocryst-bearing basalts (30%, n = 12) are fine-grained, porphyritic clasts with plagioclase phenocrysts that range in size from <1 mm to 2 cm (Fig. 2d). Phenocrysts constitute c. 7–10% of the rock by volume and tend to be uniformly distributed within the grey fine-grained groundmass.

3. Massive aphanitic basaltic clasts (12.5%, n = 5) lack surface textures (e.g. vesicles or phenocrysts), and although they dominantly have one facet, c. 15% have more complex multiple facets. They do not form mesoscale ventifact features such as pits or grooves (Fig. 2c). Wind-driven abrasion, in this instance, produced only macro-scale features, typically with a single significantly abraded windward facet and a non-abraded downwind facet, with some ventifacts having multiple facets.

4. Hyaloclastite ventifacts (2.5%, n = 1) are the least common lithology within the ventifact field. The materials are volcanoclastic rocks consisting of volcanic glass that typically forms by nonexplosive shattering (i.e. thermal granulation) of lava that interacts with water during an eruption (Sigurdsson et al. 1999) or by explosive fragmentation via analogues to laboratory molten fuel–coolant interaction (Jones et al. 2022, and references therein) (Fig. 2e). Hyaloclastite glass quickly undergoes alteration to become palagonitized hyaloclastite, which is sometimes referred to as paloginite (Stroncik and Schmincke 2001). Palagonitized hyaloclastite ventifacts in our study area have a tan to brown glassy fine-grained matrix that supports blocky crystalline to microcrystalline basaltic lithic fragments ranging in size from <1 to c. 50 mm. Ventifaction of hyaloclastites preferentially affects the more easily eroded matrix, resulting in matrix-supported clasts protruding from the surface of the ventifact in an upwind direction by up to 2 cm beyond the surrounding rock (Fig. 2e). There is also evidence of ventifaction on the more resistant basaltic clasts; however, the matrix erodes significantly faster, and clasts appear to fall off the eroding face of the ventifact before they develop distinct ventifaction textures.

A total of 500 mesoscale orientation measurements were made, with an average of 13 measurements on each ventifact. The number of measurements of mesoscale features (e.g. pits and grooves) per ventifact varied depending on lithology (e.g. massive v. vesiculated) with a minimum of four features in sample R_13 (massive) and a maximum of 20 features in sample R_4 (vesiculated). The mean orientation was 250° with a 95% confidence interval of 248–252° and a 99% confidence interval of 247–253° and a median value of 253° (Supplementary material Fig. A1). Distribution was unimodal and non-uniform with a circular standard deviation of the 500-measurement dataset of 24.5° (Table 1) (Fig. 8a).

The synthetic dataset consists of orientation measurements of mesoscale features from photogrammetric 3D models of 20 ventifacts, which were also measured with in situ techniques: nine vesiculated basalt, seven vesiculated and plagioclase phenocrystic basalt, three massive basalt and one hyaloclastite (Fig. 3b). Owing to the enhanced ability to recognize features and make measurements in a virtual 3D space of the scaled and oriented synthetic ventifacts relative to in situ methods, it was possible to collect a substantially greater number of orientations. In total, 1145 individual features were measured in the synthetic dataset with an average of 57 measurements per ventifact. The number of measurements per ventifact varied between 17 in sample R_33 (hyaloclastite) and a maximum of 94 in sample R_8 (vesiculated + phenocrystic). Measurements of 3D generated photogrammetric ventifacts features displayed similar characteristics to measurements made on the in situ counterparts, with unimodal and non-uniform distribution. Orientation data of the synthetic ventifacts had a grand mean of 252° with a 95% confidence interval of 240–265° and a 99% confidence interval of 236–269°, and a median value of 250°. Circular deviation of the synthetic dataset was 31° (Table 1). On average, the synthetic photogrammetric method generated 5.16 times as many measurements as the in situ method on the ventifacts measured with both techniques.

To assess the statistical correlation between the in situ and synthetically generated datasets, the 500 in situ orientation measurements from 40 ventifacts were compared with the 1145 synthetic measurements from the 20 ventifact synthetic subset (Fig. 7). Results of a F-test revealed that the datasets are statistically correlated, with a P-value of 0.314 (P-value >0.05 indicative of significant correlation) and a calculated estimated mean of 251° (Table 2). To further explore the correlation between the two datasets, the 20 synthetically measured rocks (n = 1145) were then compared with the corresponding 20 ventifacts that were measured with in situ methods (n = 222). Statistical results of this one-to-one comparison revealed a higher level of correlation, with a P-value of 0.451 and an estimated mean of 252° (Table 2), indicating significant correlation between real world (in situ) measurements and measurements made in an entirely synthetic environment on the same suite of ventifacts.

Analysis of in situ measurements by lithological type showed a lack of statistical correlation among measurements of different ventifact rock types (P-value of 0.026), indicating that rock type does have a measurable impact on the formation of erosional features. However, there is significant correlation between vesiculated basalt and vesiculated and plagioclase-phyric basalt (P-value of 0.569) probably owing to similarities in texture. Rock type also exerted control on feature orientation in the synthetic dataset, with a lack of statistical correlation across the four lithological types (P-value of <1 × 10^–12). As in in situ results, after removing massive and hyaloclastite samples, vesiculated clasts and clasts that are both vesiculated and phenocrystic do show a statistically significant correlation (P-value of 0.560). When comparing in situ and synthetic measurements based on lithological type (i.e. in situ vesiculated v. synthetic vesiculated), a statistically significant relationship was found for each lithology excluding massive clasts (Table 2). Massive clasts had significantly fewer meso- and micro-scale features, with 31 measurements made on five in situ ventifacts and 132 measurements made on three synthetic ventifacts. The limited number of in situ and synthetic measurements made resulted in a lack of a statistically significant relationship of the massive lithology in general. Results indicate that neither the in situ nor synthetic dataset was in statistical agreement, nor was there correlation between the same ventifacts when measured in situ v. synthetically (e.g. R_24 has a mean in situ orientation 260° and mean synthetic orientation 294°). These results indicate the control that the number of measurable mesoscale ventifact features has on both methods of data acquisition.

Results from statistical analysis of in situ and synthetic measurements from a ventifact at the most windward portion (R_36) relative to a ventifact located at a downwind position (R_21) as well as a ventifact located in the interior of the ventifact field (R_16) all indicate P-values >0.05 and both have a mean of 245°. This result indicates that the orientation at which ventifact features develop is not dictated by any specific location within the field while also showing a low variability of ventifact features across the span of the field.

Individual ventifacts located at the most upwind portion of the field, centre of the field and most downwind portion (Fig. 3b) were compared with whole-field mean orientation values to assess the ability of a single ventifact to record mean orientation values across the entire field. Findings showed that measurements made on the most prominent facet of each rock were not statistically correlated to either the in situ or synthetic calculated means for the entire ventifact field (Fig. 7f, i and s). As such, measuring the most prominent facet of a rock clast is not an adequate representation of true ventifact feature orientation, necessitating the need to make multiple measurements per sample.

There is a high level of correlation, but there are non-identical results, between the full final in situ and photogrammetrically derived datasets (Table 1). Comparing mean ventifact feature orientation of gross in situ measurements from 40 ventifacts (n = 500) with photogrammetric measurements of the 20 ventifact subset (n = 1145) resulted in a mean in situ orientation of 250° with a circular standard deviation of 24.5° v. a mean photogrammetric orientation of 252° with a circular standard deviation of 31° (Table 1; Fig. 8). Although the mean values between the two datasets differ by only a few degrees, this suggests that there is greater overall variability in the photogrammetric dataset, which is probably a function of the ability to measure more subtle surficial features, in turn capturing a larger range of ventifact textural orientations. Acquisition of photogrammetric data was additionally more efficient, requiring a maximum of 20 min per ventifact, v. in situ field measurements, which took 30–55 min, effectively reducing field data collection time by a factor of 0.36–0.66.

The difference in number of measurements between the in situ (n = 222) and synthetic (n = 1145) datasets represents an increase of 5.16 times more photogrammetric measurements per ventifact relative to in situ techniques. Both the in situ and synthetic datasets were highly correlated with a P-value of 0.314 and had an estimated mean of 251° (Table 2). To determine if the significantly larger data volume in the photogrammetric dataset was affecting the calculation of the estimated mean, mean feature orientation of the 20 in situ ventifacts was compared with mean feature orientation of the corresponding 20 photogrammetrically measured ventifacts, to compare one measurement per ventifact and have an equal number of samples in the two datasets (Fig. 9). This produced a P-value of 0.451 (Table 2), which agrees with the full dataset. The in situ dataset shows a main peak at 250° with a spread of 125–325°, and the photogrammetric dataset shows a main peak at 252° with a much larger spread of 25–355° (Table 1). The 2° variation between the mean values, paired with the statistical correlation between the two methods, implies a high level of correlation.

The ventifact feature orientations showed a higher level of detail, and a higher level of variability, in the photogrammetric dataset compared with the in situ dataset. Although this is not a surprising result, considering differences in data collection between the two methods, these findings provide a one-to-one, detailed comparison highlighting the advantages of collecting photogrammetric data. Owing to the need for the observer to physically orient their body and the compass to the surface of the ventifact to record an accurate in situ measurement, small features or those in difficult to reach areas such as at the base of the ventifact were less easily measured. Additionally, time and weather limitations during fieldwork constrained the number of measurements that could be collected on individual ventifacts. Photogrammetric methods benefit from the ability to make measurements in a controlled environment not affected by time or physical constraints. The ability to manipulate and orient a ventifact in 3D space facilitates measurements that would not be obvious or even feasible in the field; for example, the mesoscale features located at the nexus between the base of the ventifact and the sandsheet (Fig. 6). Photogrammetric methods afforded the ability to manipulate ventifact orientation to effectively bring the observer's ‘eyes’ and measurement tools directly to the surface of the rock, which increases the ability to identify and measure mesoscale features that were potentially missed during in situ analysis. Additionally, Agisoft Metashape™ visualization allows for the control of the lighting and contrast of the synthetic rendering, which highlights subtle surficial features.

To investigate variability of ventifact feature formation, three single vesiculated clasts with similar ventifact features distributed across the 1.5 km extent of the field were selected to quantify how individual ventifacts may record variability, and if features of single ventifacts can be diagnostic of the larger area in which they reside (Fig. 3b). At the windward extent, R_36 had 14 feature measurements with a mean of 249°. Ventifact R_16 in the interior of the field had a mean feature orientation of 239° based on 14 measurements and R_21 at the downwind end had a mean of 251° based on six measurements (Table 1). Although individual means differ, these three ventifacts have a statistically significant combined mean of 245°, but they are not representative of the whole field, which has a mean in situ value of 250°. Photogrammetric analysis of the three ventifacts agreed with in situ findings that mean orientation is variable on a case-by-case basis, and R_31 (235°), R_16 (236°) and R_21 (272°) differ greatly from the gross mean calculated from the entire 20 ventifact photogrammetric dataset of 252°. There is also a lack of strong correlation between in situ and photogrammetric measurements for the same ventifacts. Variability between in situ and photogrammetric mean values is greatest in R_21, with in situ data recording an orientation 21° different from the photogrammetric mean. Watson–Williams F-test results comparing the in situ with photogrammetric datasets for R_36 and R_21 exclude a statistically viable mean (Table 2), but R_16 results for both datasets yielded an estimated mean of 236°, which does not correspond to either the gross in situ mean (250°) or gross photogrammetric mean (252°) (Tables 1 and 2).

Four distinct lithologies provide the opportunity to quantify the impact of rock type on the development of ventifact features (Fig. 2). Previous work by Bridges et al. (1999) suggested that lithology does not affect ventifact feature formation, but with our dataset this is a testable hypothesis.

Comparison of the 40 ventifacts that were measured in situ shows that development of ventifact features does vary according to clast lithology. Mean feature orientation of vesiculated clasts, vesiculated + phenocrystic clasts and hyaloclastite differs from the gross in situ mean of 250° only slightly, but mean feature orientation of massive clasts differs by 22° (Table 1). The variation seen within the massive clasts compared with the rest of the lithologies (vesiculated, vesiculated + phenocrystic and hyaloclastite) indicates that ventifact lithology does affect the way in which ventifact features form. Clasts with more primary textures both facilitate and enhance development of ventifact features identified and measured with in situ and photogrammetric methods. These findings bring into question interpretations from Mars Pathfinder studies that suggest that lithology is not believed to play a significant role in ventifact characteristics (Bridges et al. 1999), and also indicate that analysing lithologies with existing textural variations may lead to the acquisition of more measurements and ultimately produce the most representative dataset.

Subdivision of the 20 photogrammetric ventifacts by lithology showed that all four types of clasts differed from the gross photogrammetric calculated mean of 252°. F-Test indicates that only vesiculated clasts and clasts that are vesiculated and phenocrystic are statistically similar with a P-value of 0.560 and an estimated mean of 248° (Table 2), which is 4° from the gross photogrammetric mean of 252° (Table 1). Photogrammetric analysis appears to enhance the divergence of measurements among lithologies, possibly related to the ability to measure more subtle features than field-based measurements, indicating that it is necessary to analyse all lithologies present to accurately characterize the field.

There are consistent trends within each lithological type that may help explain how rock type is affecting the measured features. Orientation of features on vesiculated clasts is similar to that observed on vesiculated and phenocrystic clasts, possibly related to the similarity in vesicle texture and the existence of surficial flaws on the rock surface that wind-driven particles can preferentially erode. F-Test results comparing in situ measurements of vesiculated and vesiculated + phenocrystic ventifacts with photogrammetric measurements of the same two lithologies shows that the four datasets are statistically similar with a P-value of 0.779 and an estimated mean of 248° (Table 2). Vesiculated clasts and clasts that are vesiculated and phenocrystic make up the majority of the in situ (85%, n = 34) and photogrammetric (80%, n = 16) datasets, and results show that in situ and photogrammetric methods are statistically similar for the majority of the measured ventifacts from both the groups.

Mean feature orientation of massive clasts differs from the gross mean feature orientation from the in situ and photogrammetric datasets by 12° (in situ n = 41) and 22° (photogrammetric n = 132). F-Test results from comparison of in situ and photogrammetric data from massive clasts show that the two are not statistically correlated with a P-value of 0.014 and as a result an estimated mean cannot be calculated (Table 2). Analysis of measurement distribution for massive clasts shows that the in situ data have a singular main peak between 235 and 290° with anomalous measurements at 110, 320 and 345°. The photogrammetric data have two peaks at 260 and 290° and do not show any similar outlying values as seen in the in situ data (Fig. 10g and k). Because these individual measurements were not observed in the photogrammetric data, which had 3.22 times as many measurements, these measurements may represent errors in the field, and photogrammetric methods could provide a valuable check on field data collection.

Hyaloclastite measurements come from a single clast and make up a small proportion of the total ventifact dataset. The photogrammetric measurements may be more representative of the true mean feature orientation owing to the difficulty in identifying and measuring hyaloclastite ventifact features in the field (Fig. 2). Photogrammetric analysis of the hyaloclastite sample allowed for the identification and measurement of subtle mesoscale features that encompass a wider range of orientations from 204 to 270° relative to the in situ analysis, which ranged from 221 to 264° (Table 1; Fig. 10h and l). F-Test results show that the in situ and photogrammetric values are statistically correlated with a P-value of 0.213 and an estimated mean of 238° (Table 2). This value differs significantly from the gross estimated in situ (252°) and photogrammetric means (250°; Table 1) and suggests that analysis of a single hyaloclastite sample can be highly variable.

Results from this study indicate that mean ventifact feature orientation is found to be influenced by lithology. As such, datasets should encompass all lithologies present to obtain the most complete representation of an entire field, and less resistant lithologies may record shorter duration changes in wind direction, which may not have been significant enough to affect more resistant clasts. Additionally, owing to the increased weathering rate of hyaloclastite clasts relative to basaltic clasts, they may not record sharp and distinct ventifact features or may have eroded away entirely, which could be a reason for the low number of hyaloclastite clasts present within the ventifact field. Vesiculated basaltic clasts, and basaltic clasts that are vesiculated and phenocrystic, displayed similar mean feature orientations owing to the existence of erosional features that facilitate erosion and a resistant groundmass that preserves sharp erosional textures. The presence of surficial flaws in this instance facilitates aeolian-driven ventifaction by exploiting pre-existing weaknesses or crystalline/mineralogical variability across the surface of the rock (Knight 2008). Clasts that display surficial flaws in the form of vesicles and/or phenocrysts may develop finer-scale features more readily owing to presence of existing inhomogeneities that facilitate aeolian weathering as compared with massive clasts. Massive clasts that lack surficial flaws appear to be the most resistant to ventifaction and therefore may represent a more time-averaged record of the field as a whole.

The study area was bisected into an upwind section with 19 ventifacts and a downwind section with the remaining 21 ventifacts, following the methods of Mountney and Russell (2004) (Fig. 3, Table 1). The upwind portion of the field had 127 measurements with mean feature orientation of 252° and a range from 180 to 330° with a single facet measurement at 115°, whereas the downwind area had 127 measurements with a mean of 252° and range of 165–300° with 46% of measurements greater than the mean. Despite the variance in data between the two halves, F-tests indicate that both sections are statistically correlated. Photogrammetric data from the same subsets had eight ventifacts with 493 measurements in the upwind area and 12 ventifacts with 626 measurements. The means of the upwind and downwind sections are 253 and 251° respectively, with F-tests showing that both sections are also statistically correlated. Both sets of results indicate that the upwind and downwind portion of the field are recording subtly different mean feature orientations, but the variance across the two halves within each dataset as well as between the two datasets is small enough that they are all statistically correlated. It is interesting to note that despite the statistical similarities, the data are not identical and, for example, roughly 7% of the photogrammetric measurements in the downwind dataset are not accounted for in the in situ record. These measurements probably represent those that were not feasible to measure in situ with a hand compass owing to position on the underside of a sloped face, showcasing that the photogrammetric data are capturing additional features and increased variability that was not recorded while in the field.

A further division into three subsets, upwind, core (or centre) and downwind, termed Zones 1, 2 and 3, respectively (Fig. 3b), had 13 ventifacts with 158 measurements (Zone 1), 14 ventifacts with 183 measurements (Zone 2), and 14 ventifacts with 158 measurements. All three zones are statistically correlated with a mean of 250°. Although the three zones are statistically similar enough to meet the threshold for calculating a combined mean, there is a 6˙° variation between Zone 1 and Zone 2, a 4° variation between Zone 1 and Zone 3 and a 2° variation between Zone 2 and Zone 3, which are separated by no more than 500 m. Photogrammetric data from the three sections show a higher level of variability among the sections. Zone 1 (seven ventifacts, 377 measurements) has a mean value of 253°, Zone 2 (five ventifacts, 305 measurements) has a mean of 239°, and Zone 3 (eight ventifacts, 382 measurements) has a mean of 260°. F-Tests show that none of these zones are statistically correlated. Zone 1 in situ and photogrammetric data are the only sets that are statistically correlated among the two datasets; however, there are overall trends that are consistent. Mean feature orientation of both datasets decreases from Zone 1 to Zone 2, and then increases from Zone 2 to Zone 3 (Fig. 3b). The scale of the observed decrease from Zone 1 to Zone 2 (6° for in situ, 14° for photogrammetric) and the observed increase from Zone 2 to Zone 3 (2° for in situ, 21° for photogrammetric) is consistently greater in the photogrammetric dataset, probably owing to the larger number of measurements capturing more subtle variations in texture (Fig. 8).

The ventifact field located on the Dyngjusandur sandsheet of the Icelandic Highlands has been used to quantify the robustness of in situ and photogrammetric datasets to accurately record and identify ventifact features for climate interpretation. Macro- and mesoscale feature measurements were made on 40 ventifacts in the field and 20 ventifacts were also photographed in the field and reproduced as 3D models and measured using photogrammetric techniques. Our key finds are as follows.

1. Inferred palaeowind directions from in situ and photogrammetric datasets are comparable within 2° ± 2° at 1$σ$ (n = 20). However, photogrammetry significantly reduces field time by a factor of 0.36–0.66 (20 min per ventifact compared with 30–55 min for in situ measurements) and provides 5.16 times more measurements per ventifact. Although in situ measurements offer valuable ground truth for comparison with digital analyses, given the efficiency and depth of photogrammetric methods, we suggest prioritizing photogrammetric image acquisition if field time or access are limited.

2. Lithology governs ventifact feature development, erosional timescales and the details of feature preservation. Quantifying and optimizing lithological variety are critical to fully capture a true overview of ventifact textural variability.

3. Ventifact analysis on Mars would benefit from targeting future photogrammetric data collection via UAS (Unmanned Aircraft Systems) to capitalize on enhanced mobility and potential for more rapid image acquisition.

4. Photogrammetry, using either ground-based (e.g. rover- or astronaut-acquired) images or rover/UAS technology, allows a wide range of geomorphological investigations. This study highlights the value of 3D reconstructions for revisiting virtual outcrops post-field deployment to perform detailed measurements that would otherwise not have been possible in real time.

We thank the Vatnajökull National Park Service (Vatnajökulsþjóðgarður) for providing permission to work in the Dyngjusandur region and the Icelandic Institute of Natural History (Náttúruminjasafn Íslands) for permission to export samples.

SXH: conceptualization (lead), data curation (lead), formal analysis (lead), investigation (lead), methodology (lead), software (lead), validation (lead), visualization (lead), writing – original draft (lead), writing – review & editing (lead); IU: conceptualization (supporting), data curation (supporting), funding acquisition (lead), investigation (supporting), methodology (supporting), supervision (lead), writing – original draft (supporting), writing – review & editing (supporting); DP: writing – original draft (supporting), writing – review & editing (supporting); PW: data curation (supporting), methodology (supporting), software (supporting), writing – original draft (supporting), writing – review & editing (supporting); SS: conceptualization (supporting), data curation (supporting), formal analysis (supporting), investigation (supporting), methodology (supporting), software (supporting), visualization (supporting), writing – review & editing (supporting); CWH: data curation (supporting), writing – review & editing (supporting).

Fieldwork in 2019 was organized by NASA's Goddard Instrument Field Team (GIFT). This work was supported by funding from the NASA Internal Science Funding Model (ISFM), GIFT, the NASA Post-Doctoral Program, and supported by NASA under award 80GSFC17M0002. C.W.H. also acknowledges funding support from NASA Planetary Science and Technology Through Analog Research (PSTAR) Grant 80NSSC21K0011. Work by S.S. and P.W. is supported by NASA under award number 80GSFC21M0002. I.U. acknowledges funding support from NASA Solar System Workings grant 14-SSW14_2-0101 and the Royal Society New Zealand Catalyst Seeding grant 3726349.

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

This data release contains imagery and a digital terrain model from a UAS image data collection campaign over a section of the Dyngjusandur sandsheet, Iceland, situated north of the Holuhraun lava flow margin and south of Askja and Dyngjuvatn. The UAS equipment was operated by personnel from the University of Maryland and the University of Iowa, and data were utilized for academic research and provided to NASA and the US Geological Survey for archiving under https://doi.org/10.5066/P1SXIRVQ.