More than one-third of the Australian mainland is occupied by desert linear dunefields, which are distributed in an anticlockwise whorl about the centre of the continent (Jennings, 1968; Hesse, 2010). Linear dunefields in central Australia experience the greatest variability in orientation compared with those on the arid periphery (Hesse, 2010). Other landforms including playa lakes, uplands, stony desert and floodplain substrates, and active, ephemeral and abandoned fluvial channels also occur in the desert (Fitzsimmons, 2007).

We focus on a small (c. 8450 km²) dunefield located on the southwest margins of Kati Thanda (Lake Eyre) in central Australia (in the north of the state of South Australia) (Figure 1). Our research area was selected because of its high degree of morphologic variability within a relatively small area, as well as the presence of bedrock uplands, chains of playas and fluvial channels. This setting facilitates an investigation of the interaction between dune morphologic parameters and other landforms.

Kati Thanda lies within the most arid region of Australia. It experiences 150 mm of rainfall on average per year, with substantial inter-annual variability (Simon-Coinçon et al., 1996) and no clear seasonal dominance of precipitation (Magee et al., 2004). Kati Thanda (Lake Eyre) is the terminal playa lake of an endorheic basin which occupies approximately 15% of the Australian mainland (May et al., 2022). Kati Thanda (Lake Eyre) overlies the Mesozoic Eromanga basin (Krapf et al., 2022); arid conditions have persisted in the basin since the Pliocene (Fujioka et al., 2005). The Kati Thanda dunefield which forms our focus is bordered to the north by the Davenport Range, which rises 140 m above the dunes and comprises Neoproterozoic sediments and intrusive mafic rocks (Cowley et al., 2020). The Stuart Range lies to the west and south of the dunefield and consists of deformed lower Cretaceous Eromanga Basin sediments (Drexel et al., 1995). Kati Thanda, with its generally dry lake bed of clays and salt crust, lies to the east and northeast of the dunefield. Clayey through sandy Eromanga Basin rocks occasionally crop out along the playa shoreline (Drexel et al., 1995; Cowley et al., 2020) and are in many places cemented by silcrete and ferricrete (Krapf et al., 2018; Wakelin-King, 2022). We assume the linear dunefields to have formed from the early Pleistocene onward (Fujioka et al., 2009), periodically activating during phases of intensified aridity and increased sediment supply throughout the Pleistocene and into the Holocene, as has been found to be the case elsewhere within the Lake Eyre Basin (Hollands et al., 2006; Fitzsimmons, 2007; Fitzsimmons et al., 2013).

The dunefield itself is sparsely vegetated and the dunes appear to be largely inactive (Wopfner and Twidale, 1988; Hesse, 2010) or in a state of partial activity, with bedforms retaining a stable morphology (Hesse and Simpson, 2006). Overall, the linear dunes of the study area shift from a predominantly WSW-ENE orientation in the southwest to a more N-S direction in the far north and east (Figure 1) and lie on the southeastern edge of the anticlockwise linear dune whorl observed across the Australian continent by previous workers (Wasson et al., 1988). Linear dunes are postulated to broadly result from aeolian transport parallel to the resultant wind vector at the time of formation (Tsoar et al., 2004; Pye and Tsoar, 2008).

The elevation of the dunefield ranges from 110 m asl on its western margins to −10 m at the shoreline of Kati Thanda, along a 120 km transect. We observe a range of non-aeolian landforms within the study dunefield. These include several small playas, including chains of depressions which align north-south and parallel Kati Thanda, and may represent abandoned fluvial depressions or lake floor (Wakelin-King, 2022). Several ephemeral creeks drain eastward into the lake (Armon et al., 2020; May and Worthy, 2022), and we observe a number of local drainage systems are associated with individual small playas (Hesse, 2010). Several fossil and active mound springs associated with discharge from the underlying Great Artesian Basin (Boucaut et al., 1986; Fitzsimmons, 2005) occur within the dunefield.

We selected three smaller case study areas within the dunefield in order to maximize the efficiency of data acquisition (Figure 1). These are described as the centre north (CN), northeast (NE) and southwest (SW) case study areas (Sections 2.1.1–2.1.3, respectively). Other areas (north, west, center-south and southeast), which were not investigated in as much detail but discussed in part below, are delineated in Figure 1.

Our study makes use of several data processing methods (for a detailed overview, refer to Supplementary Material). We used satellite remote sensing data to produce terrain information for our assessment of topographic features. We then extracted dunefield morphologic data using semi-automated GIS in the form of Geographic Object-Based Image Analysis (GEOBIA; Section 2.3).

Since our ultimate aim was to use this workflow (Figure 2) to investigate interactions between linear dunes and other landscape types, we first characterized the regional terrain using remote sensing-derived digital elevation models (DEMs). The challenge for our chosen study area is its relatively subdued height variation (c. 120 vertical meters; see Section 2.1); a particularly high-resolution DEM was therefore necessary to detect the subtle changes in elevation. Following visual inspection and literature review of seven open-source DEMs (Uuemaa et al., 2020), we selected the AW3D30 Digital Surface Model (DSM), using the Japan Aerospace Exploration Agency (JAXA) dataset and remote-sensing data derived from the Panchromatic Remote-sensing Instrument for Stereo Mapping (PRISM) aboard the Advanced Land Observing Satellite (ALOS). AW3D30 is a freely available downscaled version of a higher resolution commercial photogrammetric DSM and exhibits high robustness and vertical accuracy compared to its competitors (Courty et al., 2019; Uuemaa et al., 2020), Furthermore, its permissive license fits in our general framework using open software and data. These merits outweigh the advantages of genuine DTMs (with vegetation, human made structures, etc.), that play little role for our chosen region since there is minimal human impact and neglectable vegetation cover on the dunefield. We worked with six AW3D30 tiles (each one degree of latitude and longitude in area) to produce a DSM with pixel resolution of c. 27.5 m (Longley, 2005) (see Supplementary Figure S15).

Data pre-processing was undertaken using the free, open-source software SAGA GIS Version 7.8.2 (System for Automated Geoscientific Analysis; Conrad et al., 2015). SAGA GIS is compatible with the freeware QGIS (QGIS-Development-Team, 2022) which was used to classify dune morphological categories and to undertake the mapping. We merged the six data tiles into a single grid using the SAGA GIS mosaic tool, resampling using the nearest neighbor option. Following alignment of the input data, we projected the dataset into the Universal Transverse Mercator (UTM) 53-south grid (UTM zone 53 J), using EPSG code 32753 and the UTM projection tool in SAGA GIS, again using Nearest Neighbour resampling. We improved processing efficiency by reducing data frame size, clipping the input grid using a polygon shapefile containing the study area outline in QGIS.

We undertook our dunefield mapping and morphologic analysis using a GEOBIA approach. GEOBIA refers to a category of digital remote-sensing research which investigates and delineates entire geographic entities (Blaschke et al., 2014; Chen et al., 2018; Kazemi Garajeh et al., 2022). Alternatively, to classical pixel-based approaches, classification is performed on segments comprising multiple pixels, which allows to integrate object- and shape parameters into the classification workflow. Combined with digital elevation data (Section 2.2), we can analyze both topography and dune morphologic variability for the studied area.

The overall dominant dune orientation for the dunefield–some 75% of all dune bedforms - lies along a WSW-ENE axis. Variation within individual case study areas, as well as for the entire region, is illustrated in Figure 4. Table 2 summarizes the main dune orientation statistics.

The northernmost case study area comprises dunes predominantly oriented SSW-NNE and is distinct from the rest of the region (Supplementary Figure S2A). The divergence in orientation of the northern area compared with the rest of the dunefield is visible in a comparison of the dune roses and in the small proportion of NNE-oriented dunes within the overall dune rose (Figure 4A). It is important to note here the prominent north-south escarpment of the Davenport Range on the eastern margins of this area, which rises 140 m above the main dunefield and potentially represents a topographic barrier to longitudinal dune migration.

The CN case study area, immediately to the south of the north area, is dominated by W-E oriented dunes. Nevertheless, some variability in orientation is observed within this sub-region: the northern and southern parts of the case study area exhibit more WSW-ENE axes. The variation within this area is visible in the wider distribution of its dune rose (Figure 4B). The CN case study area also hosts a larger proportion of non-linear dunes and poorly defined hummocky bedforms than the others investigated. This may contribute to systematic error in the dune orientation analyses through the generation of unrealistic, oversized polygons in the classification process, which were then classified as single dunes with single orientations that do not reflect the true morphology. This error was observed in several cases, particularly in the western part of the CN case study area. The result is likely to be an underestimation of the proportion of WSW-ENE oriented dunes. We also observe a north-south oriented depression (in the form of several small playas) in the western part of the CN case study area, and an SSE-NNW oriented ridge through the centre of the area which rises up to 40 vertical meters above the depressions in the west (Figure 5B). The Davenport Range rises up to 230 vertical meters above the CN part of the dunefield on its northeastern margins. We will discuss the potential influence of these topographic anomalies on dune orientation in Section 4.2.

The NE case study area hosts the most of linear dunes, with a dominant SW-NE orientation and a secondary WSW-ENE axis (Figure 4C). Two narrow north-south ridges overlain by dunes rise c. 20 vertical meters above the surrounding underlying relief and straddle a small playa in the south-western part of the NE case study area (Figure 3B). North-eastwards, the underlying topography slopes gently down to the Kati Thanda shoreline. This case study area appears to be the most suited to our workflow due to the morphology, size, spacing and length of the dunes.

The SW case study area yields the least variability in dune orientation of the subregions analyzed. Almost 75% of the dunes in this area are oriented WSW-ENE (Figure 4D). However, a number of hummocky dunes in this area generate small polygons which do not follow the main dune orientation, so yielding a wider distribution of orientations in the resulting dune rose than the NE case study area but one less variable than observed in the CN case study area. The central and northeastern parts of this subregion host clearly linear dune forms, with less well defined, hummocky forms more common in the western quadrant (Figure 3C).

The spatial distribution and density of linear dune junctions may relate to aeolian transport and sediment supply (Wasson and Hyde, 1983; Wasson et al., 1988), as well as interaction with underlying topography and non-aeolian landforms (Fitzsimmons, 2007). We concede that our study focusses on a region of vegetated linear dunes, and therefore cannot comment on whether dune junctions such as these also occur in unvegetated regions; nevertheless, the tools for mapping are applicable anywhere. In order to investigate these possible influences on dune planform, we present the data both as cross-sections of dune junction density, overlain on the topographic profiles along the mean longitudinal axis for each case study area (Figure 5), and as heatmaps of junction density overlain on the satellite imagery (Figure 6).

The results of our Nearest Neighbor Analysis show, that all study areas have index values below NNI < 1 and therefore the observed mean distances are shorter than expected random distances. This indicates clustering of junctions. The z-values are lower than a significance threshold of z < −2.58 (p < 0.01), highlighting that the pattern of junctions differs significantly from a random distribution (Table 3).

The NE case study area exhibits the lowest density of dune junctions. We mapped 105 junctions across the 973.7 km² area, resulting in a mean density of 0.107 junctions per km². Two peaks in dune junction frequency can be observed along our reprojected transect, occurring in the west at c. 10 km and in the east at c. 45 km (Figure 6C). The western quadrant peak in junction frequency corresponds to a topographic high, whereas the eastern peak occurs in a low elevation, low relief area. The kernel density heatmaps indicate a cluster of dune junctions in the northeastern corner of the subregion (Figure 6B).

The CN case study area contains an intermediate density of dune junctions; 242 junctions were mapped onto an area of 1,219.3 km², resulting in a mean density of 0.198 junctions per km². The highest concentration of dune junctions occurs at c. 15 km along the reprojection line (Figures 5A, 6A), which corresponds to the west-facing mid-slope of the underlying topography. We observe an additional, minor cluster of junctions at c. 44 km on the eastern margins of the CN case study area (Figures 5A, 6A). There is no clear correlation at this location with the underlying topography just to the west of the topographic ridge, which is likely to correspond to the 15 km cluster observed in our reprojected transects (Figure 5A). Another cluster is located on the eastern margins of the CN case study area—correlating with the 44 km peak—in the vicinity of the transition between dunes and the Davenport Range to the east.

The highest density of dune junctions can be observed in the SW case study area. In this area, we mapped 246 junctions over an area of 833 km², resulting in a mean density of 0.295 junctions per square kilometer. The SW case study area is underlain by more variable topography than the other two case study areas; three topographic highs and three topographic lows occur along the 45 km reprojected transect (Figure 5F). The dune junction density is highest at c. 20 km along the transect (Figure 5E), which corresponds to a zone of higher elevation. Our kernel density heatmap highlights a cluster of junctions on the southern margins of the SW case study area (Figure 6C), adjacent the Wattiwarriganna Creek bed, only when the radius size is increased to values where it is questionable whether a resulting cluster would be meaningful.

The spacing distances between linear dunes may relate to sediment availability, underlying substrate material (which may or may not be associated with sediment availability), vegetation cover, wind capacity, drift potential and interactions between dune bedforms and underlying topography or proximal non-aeolian landforms (Wasson et al., 1988; Hesse and Simpson, 2006; Hollands et al., 2006; Fitzsimmons, 2007; Fitzsimmons et al., 2020). Although the measurement of dune spacing was challenging due to the strong dependence on the assumptions described in Section 2.3.4, we nevertheless observe geographical variability in dune spacing characteristics. The results of our dune spacing measurements for the three case study areas are summarized in Table 4.

We present the distribution of dune spacing for the respective case study areas as box plots and a raincloud plot in Figure 7. We observe a number of differences in spacing characteristics between the case study areas. Both the mean and median dune spacing values for the CN case study area are intermediate (470 m and 421 m, respectively), but with a high standard deviation indicating variable spacing across the area. The CN case study area also hosts dune pairs with the widest spacing observed in the dunefield (2,447 m). By contrast, the NE area contains the largest mean distance between dune pairs (609 m), although this area has the largest standard deviation of the three areas studied (269 m). The SW case study area hosts the shortest mean spacing (425 m) and smallest standard deviation (165 m); the largest number of short inter-dune distances are found in this area. Overall, however, the mean spacing and their standard deviations are similar between the CN and SW case study areas; the main difference in spacing characteristics for these two areas lie in the occurrence of dune pairs spaced very far apart in the CN area. Most dunes are spaced between 300 and 500 m apart across all three case study areas and are distributed normally around this peak when a logarithmic scale is used for the spacing distance.

We consider the workflow developed in this study to be effective for mapping dunefields and extracting morphological data on linear dune bedforms. Nevertheless, some improvements could be made to increase efficiency and improve accuracy. Here we assess our workflow and suggest possible changes for future studies.

Given our emphasis on an open-source approach, we consider the primary datasets used and their preprocessing to represent the optimal compromise between resolution and accessibility. The vertical precision of our primary data represents a limiting factor to our workflow, particularly given the low relief of the study region. Although Light Imaging Detection And Ranging (LIDAR) data would improve the resolution of our spatial analysis (Florinsky, 2016; Solazzo et al., 2018; Dong et al., 2021), there are as yet no free-to-use datasets covering our study area. Furthermore, LIDAR processing is computationally intensive and increases the costs for hardware and runtime for researchers. The DSM used in our study, with 27.5 m spatial resolution, represents the best possible compromise for our open-source approach. The UTM coordinate system is a convenient choice for mapping at a regional scale with study areas less than ∼1,000 km wide, as it combines high horizontal accuracy with metric coordinates and global applicability. Visual inspection of our output data confirmed that our GEOBIA approach based on the segmentation of a low-pass filtered DSM using thresholds and areal object parameters produces the best possible result. The accuracy of dune polygons is most dependent on the classification threshold values, since these determine the resultant dune maps and morphologic data. We argue that the values used in this study are optimal for this region in order to find the balance between reducing the data sufficiently to easily identify and remove irrelevant features, while not losing too many pixels which carry dune-related information. The low pass filter is scale-dependent and varies with the magnitude of dune size and kernel, so that these values are specific to the dunefield studied and should be calibrated to local conditions for future studies in other regions.

Our measurements of dune orientation were strongly dependent on the size and accuracy of the segment polygons, which were generated using the threshold values determined during preprocessing. The classification assumptions described in Section 2.3.1 resulted in the production of a number of very large polygons with multiple longitudinal axes, particularly in the CN case study area. Such polygons could only be rectified into individual dune forms by time-consuming manual division into smaller polygons, or by accepting lower accuracy in the analysis of orientation statistics. Again, the balance must be struck between efficiency and accuracy given the scale of the area to be mapped. Visual inspection of our results indicates an acceptable degree of accuracy, although this should always be assessed in the context of the research question and scale. Our approach to derive the angles of polygon longitudinal axes using the available QGIS script was effective, although the implementation of the data in R could be improved with further script development.

One disadvantage of our proposed workflow is the substantial effort involved in generating dune junction statistics. Although the use of chordal axis intersections produces an accurate dataset of dune junctions, it is debatable whether this is the right approach for this particular dunefield given its subdued topography and unquantifiable sediment supply without ground truthing. The low relief of the southwest Kati Thanda (Lake Eyre) dunefield prevents an effective classification of dune height thresholds and requires intensive manual correction of dune polygons both to run the CAT and to map junctions according to chordal axis intersections. The “blurring” effect of low relief dunes results in artificially large polygons which affects the CAT triangulation process. In the context of our study region, this process could really only be improved by developing better reclassification thresholds or more manual correction of the output data. However, our approach would be effective in regions with greater topographic relief. Finally, we assess the reprojection of the density plots to be an effective means of obtaining information relating junction densities with underlying topography, and by proxy, interactions between dunes and other landforms.

We argue that our approach to measuring dune spacing could be improved. Our method not only requires substantial manual effort, but also has a strong dependence on the location of the measurement lines relative to the dune central axes. This can result in systematic error, although we concede that over the scale of the dunefield studied, the overall errors may not be so large as to compromise accuracy. Interdune distance measurements may be improved using the “Hausdorff distance”, which measures the mean distance between two polygons (Alt et al., 2003) and can be undertaken using the QGIS field calculator. Developing a semi-automated method for measuring the “Hausdorff distance” would however require substantial effort to implement.

Overall, we argue that open-source GIS, such as SAGA and QGIS, can be effectively used in a workflow to extract meaningful information from spatial terrain data. The precision of our output is sufficiently high to extract dune-related data despite the significant challenge of low topographic relief; for other dunefields with higher dune bedforms and greater topographic relief, the workflow developed here would be even more effective. The widespread adoption of our open-source workflow will hopefully make meaningful assessments of dunefield morphologic variability as a framework for larger questions relating to dunefield expansion more accessible to our colleagues with fewer financial resources.

The orientation data for the SW Kati Thanda (Lake Eyre) dunefield provides us with an opportunity to investigate the degree to which linear dune orientations represent a response to factors additional to wind regimes, such as topographic and geomorphic barriers. We recognize that dunefields are composite landscapes comprising individual landforms which may not have been active all at the same time (Hesse and Simpson, 2006; Fitzsimmons, 2007; Fitzsimmons et al., 2013) and our interpretation is thereby limited, since we do not have the temporal perspective. Compared with linear dunefields elsewhere across Australia (Jennings, 1968; Hesse, 2010), the SW Kati Thanda (Lake Eyre) dunefield exhibits substantial variability in dune orientation across a relatively small area (Figure 3). Fundamental to our interpretations here is the assumption that dune orientations represent formation parallel to the resultant sand-shifting wind vector at the time of formation (Tsoar, 1989). Based on a comparison with less topographically diverse adjacent dunefields (the Victoria to the west and Tirari to the east (Hesse, 2010), we can expect our study region to translate to a zone of transition from westerly winds in the southern part to winds with a more south-westerly component in the northern part. We observe a substantial swing in linear dune orientation in the northern half of our area in the proximity of the Davenport Range which rises 140 vertical meters above the dunefields, where we might expect a SW-NE (bearing 45°) dune orientation if there were no topographic barriers. In fact, we observe a mean linear dune orientation of 46° in the northernmost part, which diverges only 1° from the hypothesized 45° bearing. Dunes in this northernmost area occur in a c. 20 km wide plain between the Davenport Range to the east and the Stuart Range escarpment to the west of roughly the same height. We propose this divergence from the hypothesized non-topographically influenced dune orientation to be a response of the dunes to surrounding topography, due to the deflection of near-surface winds by the uplands northwards along the axis of the main escarpments (Fitzsimmons et al., 2020). We also observe a shift in mean dune orientation in the CN case study area which lies immediately southwest of the Davenport Range. In the case of CN, it is not the entire dunefield which is deflected; rather there is a net orientation change in the dunes of 22.5°, from a mean WNW-ESE orientation shifting to W-E. This transition starts 27 km west (and therefore upwind) of the Davenport Range. The dunes terminate on the windward side of an ephemeral channel at the base of the escarpment. We therefore hypothesize that the orientation of linear dunes, as an approximate proxy for near-surface sand-shifting winds, are affected by topographic uplands through the deflection of air masses and thereby aeolian transport along an escarpment. This effect would appear to initiate some distance upwind (in this case 27 km) from the topographic barrier. That distance may be proportional to the height (and possibly slope angle) of the topographic barrier, but is beyond the scope of this study to define more precisely. Shifts in dune orientation relating to nearby topographic highs have previously been subjectively observed in Australia (Hollands et al., 2006) and Central Asia (Fitzsimmons et al., 2020); the data from our study provides the first opportunity to investigate this phenomenon in a semi-quantitative way. Although there are differences in vertical relief of up to 40 m across the dunefield, we do not observe any substantial, sudden shifts in orientation on the order of what we observe near the Davenport Range. We also observe no substantial change in dune orientation relating to the occurrence of ephemeral creeks or playas, which form gentle depressions in the landscape and another potential barrier to aeolian transport (Wopfner and Twidale, 1988). We therefore infer that the topography beneath the dunefield is insufficiently high or steep to deflect the near-surface winds and alter the dune orientation; rather the more gradual change in orientation is likely to relate to the synoptic-scale changes in wind regimes which are believed to be responsible for the continental-scale dune whorl (Wasson and Hyde, 1983; Hesse, 2010).

Our analysis of the distribution and frequency of linear dune junctions across the dunefield was undertaken with the aim of investigating possible interactions with underlying topography, non-aeolian landform types, and by extension, local sediment supply. Until recently, dune junctions have been largely overlooked as a source of information about large-scale dunefield formation, in part due to technical difficulties of quantifying these features over large scales [e.g., (Fitzsimmons, 2007)]. Recent work using deep-learning approaches produced data in the form of junctions per kilometer of dune crest in the Simpson, Strzelecki and Mallee dunefields in Australia (Shumack et al., 2020). Unfortunately, however, our workflow prevented directly comparable data and therefore we cannot put our own results into the wider perspective of dunefields across the continent. Nevertheless, our dataset highlights some potential external influences on dune morphology.

The increase in dune junction density at c. 15 km along our longitudinal transect for the CN case study area coincides with a topographic ridge which runs perpendicular to the main dune axis. Visual inspection of the dunes in this locality shows that the dunes are also less linear and more sinuous. The reduction in linearity and increase in dune junctions may have several explanations: 1) there is less available sediment on the ridge slope, which has a more shallow sediment cover, so requiring the bedforms to merge with one another in a form of sediment economization; 2) the ridge slope would require higher kinetic energy to transport sand uphill, resulting in further economization of sediment on the slope; 3) the presence of the Sturt and Davenport Ranges on either side of the CN area may create more turbulent, multi-directional wind flow at this point which may redirect the aeolian bedforms and create junctions. We suggest the latter case to be the least likely, since there is no reason to assume that the near-surface wind regime would become more turbulent at the 15 km point than anywhere else between the two ranges. It is feasible that an interplay between the first two hypotheses explains the higher density of dune junctions at this locality. At this stage, however, it is difficult to prove these theories due to the limitations in the resolution of the underlying DSM to fully characterize the terrain on the ridge, and in the lack of freely available geophysical data providing information about sediment cover thickness on the ridge. Comparing the junction density results with dune spacing at this locality–another parameter which may reflect sediment supply (Wasson and Hyde, 1983; Fitzsimmons, 2007; Lancaster, 2009)—may help us interrogate the mechanisms involved. It is interesting to note a reduction in junction frequency on the ridge peak; if we invoke the first two arguments above, then perhaps the dunes by this stage have exhausted sediment supply to the point where no further merging is possible. In any case, these arguments indicate a degree of longitudinal migration of the linear dune forms; perhaps not as elongated as originally proposed by (Twidale, 1968; Twidale, 1972) but more than the local sand supply model proposed by (King, 1960). An additional cluster of junctions occurs on the northeastern margins of the CN case study area, at the juxtaposition with the Davenport Range and with an ephemeral peak at the base of the escarpment. The merging dune forms at this point provide additional support for the longitudinal migration model, since they appear to be merging in response to a geomorphic (the creek) or topographic (the escarpment) barrier impeding further migration.

In other localities, peaks in dune junction density coincide with non-aeolian landforms such as ephemeral creeks and chains of playas. Since these represent topographic lows, but also possible sources of sediment, we must examine the context in order to hypothesize which of the two possible influences are the greater. We observe increased junction frequency to the west of William Creek in the NE case study area, on the northeastern margins of the same where the Douglas Creek cuts through the dunefield, and in the central section where a chain of playa depressions underlies the dunefield (Krapf et al., 2018). In the SW case study area we observe more frequent dune junctions on the western bank of the Wattiwarriganna Creek. In these cases, the increase in dune junctions occurs upwind of the respective geomorphic features (Bowler, 1976), and may reflect a relative decrease in sediment supply as the underlying substrate provides less available sand for transport in a downwind direction (Goudie and Middleton, 2006; Fitzsimmons, 2007).

There is an overall lack of dune spacing data globally with which to assess our dune spacing results; what data do exist, are often in different forms (Ewing and Kocurek, 2010; Fitzsimmons, 2007; others) and do prevent direct comparison between dunefields. Previous work nevertheless suggests that dune spacing is closely related to local sediment supply (Fitzsimmons, 2007); others), whether through active sediment input from fluvial channels or a lack of available sand-sized material for entrainment over stony pavements and clay-rich playas. In this study, we observe generally more closely spaced dunes in the SW and CN case study areas than in the NE. Visual inspection of underlying geomorphic features, which may provide information about sediment availability, suggests that sediment supply may have driven the larger interdune distances in the NE, since this area overlies both stony pavement and a larger proportion of area covered by playas than observed in the other two areas. In addition, the Wattiwarriganna Creek cuts through the SW and CN areas, and a number of probable fossil abandoned creeks underlie these areas. These features are likely to have provided sediment supply, whether by active input downwind of active channel banks or the presence of unconsolidated sandy substrate. As such, there is a range of mechanisms by which the aeolian dunes may interact with fluvial systems, and which have an influence on their morphological characteristics.

Dune junctions and inter-dune spacing are inversely proportional in all three case study areas. Area SW shows the highest number of junctions and the lowest mean inter-dune distances. NE shows the lowest counts in junctions and the highest mean inter-dune distance. CE lies in both categories in the middle. Tests on statistical significance or correlations between both categories have not been made. This would be a subject of interest for future research on a potential interaction between both categories. The observed non-normal distribution (all α 5% significance) of the values indicates either a random mechanism of inter-dune distancing or a methodology containing (systematic/measure) errors.

Arguably, the interaction between dunes, topography and on aeolian landforms is best examined by considering the parameters in concert. For example, there are localities where we observe an increase in dune junction frequency and a decrease in interdune spacing, and vice versa. Qualitatively, the SW Kati Thanda (Lake Eyre) dunefield appears to have more instances where dune junction density and spacing values increase, which we propose may relate to the role both of underlying and surrounding topography as physical barriers to dune migration. Applying our workflow to dunefields with less geomorphic variability over a small area may yield contrasting insights, indicating different mechanisms such as sediment supply or reduction thereof, deflection of wind regimes by the surrounding relief, or underlying topography influencing the kinetic energy required to transport the sand up or down slopes. Our results provide some first insights and hypotheses into the nature of these interactions, and despite their limitations can nevertheless be considered to advance the state of knowledge of large-scale dunefield dynamics.