Post‐wildfire coastal dunefield response using photogrammetry and satellite indices

Fire has been suggested to be an initiation mechanism of landscape instability and coastal dune transgression, but modern evidence showing a shift to a transgressive dune phase is lacking. Following the largest wildfire in historical records on Kangaroo Island, South Australia, bimonthly uncrewed aerial vehicle (UAV) surveys were conducted on three coastal dune sites to study their post‐fire responses. The three sites studied here represent the landscape diversity of the temperate dunes of Kangaroo Island with both active coastal and inland relict stabilised dune fields studied. UAV surveys were used to reconstruct landscapes with structure from motion (SfM) photogrammetry and compared over time to illustrate significant changes in the landscape. The geomorphic and vegetation changes are compared in net and intra‐survey comparisons to illustrate the post‐fire dunefield response and trends towards stabilisation. Because of a lack of reliable baseline pre‐fire data, satellite geomedians are used to compute spectral indices to show the trajectory of ground cover in the study sites in the years preceding and following the fire. Satellite indices are used to separate 3D changes according to ground cover types and show their differing post‐fire responses. Local and regional wind, temperature and rainfall records are presented to provide weather patterns of the years preceding and following the fire, illustrating the wet and mild post‐fire weather. The overall results indicate no significant landscape instability across the studied sites and that the ground cover of vegetation is nearing pre‐fire baselines, showing that a severe fire has not caused a transgressive dunefield to develop.

Although many vegetation communities benefit from fires (Keeley, 1995), shifts in fire regimes (frequency and severity) may lead to disruptions in vegetation diversity (Wright & Clarke, 2007) or landscape stability (Barchyn & Hugenholtz, 2013). Fires are a common and natural occurrence across the world, including most of the Australian continent (Levin et al., 2021), but climate change is driving shifts in weather patterns and aridity, which may exacerbate fires and their effects on the landscape (Van Oldenborgh et al., 2021). With the forecast of a warmer climate that is conducive to more frequent and severe fires, it is unclear as to what extent previously stabilised and semi-stabilised dunefields may be reactivated or altered, especially in temperate climates. Further study is necessary to show how these complex natural environments respond in the short-to medium-term following disturbance by fire.
Uncrewed aerial vehicles (UAVs) are used in many disciplines across the physical sciences  and in coastal dune research (Hilgendorf et al., 2021;Walker et al., 2022). Geomorphic studies use UAV platforms and spatial data to produce 2D and 3D datasets to track landform, topographic or ground cover changes over time (Gonçalves & Henriques, 2015;Shumack et al., 2022). UAVs have been used to survey post-fires landscapes, deploying after a fire to observe the extent of burnt areas and monitor post-fire processes (Ellett et al., 2019;Fernández-Guisuraga et al., 2018). UAVs equipped with imaging cameras can be used to systematically capture overlapping images and construct 3D models of landscapes using the tenets of structure from motion (SfM) photogrammetry (Fonstad et al., 2013). Ortho-images and point clouds can be generated with varying resolution, precision and accuracy, depending on the methods and hardware used (Poley & McDermid, 2020). Certain steps in respect to camera calibration and flight design (Luo et al., 2020;Nesbit & Hugenholtz, 2019;Sanz-Ablanedo et al., 2020) tied with spatial location data ensure accurate and repeatable surveys James, Robson, D'oleire-Oltmanns, & Niethammer, 2017). Spatial data are used for georeferencing, often collected with high accuracy Real Time Kinematic/Extended Global Navigation Satellite Systems (RTK/X-GNSS) and used as ground control or check control points (GCPs and CCPs). GCPs ensure robust geometry for the model, and CCPs provide an independent assessment of accuracy (James, Robson, D'oleire-Oltmanns, & Niethammer, 2017).
The expected resolution limit for confidently detecting change from surface differencing is the level of detection (LoD), which is determined from estimated error, both systematic and random (James, Robson, & Smith, 2017). Research methods and sensors must account for their expected LoD to definitively show significant changes. 2D surface differencing from digital elevation models (DEMs) is limited in topographically complex landscapes with high relief, overhanging areas and/or surfaces with significant roughness such as vegetation (Lague et al., 2013). 3D point clouds can be used to measure change along axes unique to the surface normal, therefore preserving the local shape, texture and complexity (Williams et al., 2021). Lague et al. (2013) developed the Multiscale Model to Model Cloud Comparison (M3C2) algorithm that computes the difference between point clouds in the horizontal (X and Y) and vertical (Z) directions. M3C2 uses a vector orthogonal to the surface and calculates the difference between the mean position of points of each cloud and has been used extensively in geomorphic studies (Williams et al., 2021). James, Robson, and Smith (2017) further improved M3C2 by estimating the spatially variant error from photogrammetric and georeferencing uncertainties to generate precision maps (M3C2-PM) to vary the levels of detection across the point clouds.
In this study, post-fire coastal dunefield response was explored using 3D datasets from in situ surveys and 2D spectral changes from satellite images. The in situ UAV surveys were conducted across three representative dune types to capture landscape-level dynamics and describe them in relation to their local topography in the year following a severe wildfire. Significant changes are described according to the drone surveys hardware and method's respective error (LoDs) by quantifying changes in derived point clouds. Similar to recent studies (Hilgendorf et al., 2021Konlechner & Hilton, 2022;Walker et al., 2022), both vegetation and surfaces changes are of research interest as they show the post-fire ecogeomorphic response of landscapes and indicate stability, dynamism or transgression. Satellite indices are used to separate landcover types and to assess the 3D changes between pre-fire sandy active areas, and regions that were burnt or unburnt. For a pre-fire temporal context, satellite imagery is used to calculate pixel-based geomedian images (Roberts et al., 2017) for a time series analysis to illustrate changes in ground cover. Three separate spectral indices are computed for the years preceding and following the fire to show the post-fire spectral response.
The 2019-2020 Australian wildfire season burned thousands of hectares across Australia (Levin et al., 2021), with a significant portion of South Australia's Kangaroo Island's (KI) coastal dune systems severely affected. The Black Summer wildfires on KI were the largest fires according to contemporary records that also show an increase in recent fire frequency (Bonney et al., 2020). The broader aim of this paper is to test whether a severe fire has triggered any significant landscape instability or a shift to a transgressive dune phase in temperate South Australian dunefields. Although this study focuses on local coastal dune changes, it provides insights into the resiliency of landscapes to wildfire events and explores the short-term changes following a disturbance that has been suggested to initiate landscape instability and dunefield transgression.

| STUDY SITES
This study focuses on three coastal dune sites on the southern coast of KI, which burned in the Black Summer wildfires (Figure 1)

| Geography and climate
In all three sites, and along the southern coast of KI, indurated Pleistocene aeolian calcarenite is overlain by Holocene-aged unconsolidated sands (Bourman, 2016). Study sites are located amongst complex transgressive and parabolic dunefields that, pre-fire, were mostly stabilised by vegetation but contain active areas closer to the coast (Short & Fotheringham, 1986). Sediments are predominately sands, fine-to medium-grained and carbonate rich.
The vegetation in the dune sites is highly heterogenous and patterned based on its relative exposure to aeolian and coastal forces.
Vegetation surveys were conducted throughout this study during site visits, with dominant vegetation species noted in Table 2 for each site.
Exposed regions of the coastal sites (A and B) were dominated by rhizomatous grasses (Spinifex longifolius) and invasives (Euphorbia paralias and Thinopyrum junceiforme) with predominantly woody varieties from the Acacia, Melaleuca and Eucalyptus genus in the stabilised or sheltered regions ( Table 2). The in-land site C was characterised by the epicormic re-growth of dense mallee Eucalyptus species (plants with multiple stems from an underground lignotuber) within relatively low-lying regions as described in later sections. See Northcote (2002) for additional descriptions of vegetation communities pre-fire.
Weather information was sourced from several locations due to inconsistent long-term records and destruction following the fire. For the study period, local wind data were sourced from a permanent weather station, installed in December of 2020 at Maupertuis Bay, with additional wind observations sourced from the station at Cape Borda, which is maintained by the Bureau of Meteorology (BoM; see location in Figure 1). Monthly mean temperature and rainfall  were sourced from the BoM's datasets from Cape Borda and Flinders Chase at Rocky River, respectively. These monthly means were compared with the observation data at Cape Borda (2019-2021) and modelled rainfall monthly totals from interpolated rainfall gauge data (Beutel et al., 2019) across the UAV survey regions. Wind and weather patterns are influenced by the island's proximity to the Southern Ocean and its cold waters (Peace, 2012), with a complex multi-directional wind system ( Figure 2). According to the Koppen-Geiger classification (Beck et al., 2018), the climate is classified as Csb, indicating a temperate climate with warm and dry summers with most rain occurring outside of the summer months. Fires are F I G U R E 1 Sites of analysis for this study on the dunefields of Kangaroo Island (KI). Sentinel 2 (S2) satellite image from 30 January 2020 shows KI in bands 12, 8 and 4 highlighting regions burnt in the 2019/20 Black Summer wildfires. Black triangle in the northwest corner of the satellite image indicates the location of Cape Borda weather station. Inset maps of A, B and C show the extent of drone surveys in the dotted lines, with near-transparent overlays of burnt, unburnt and sand classifications (S2) as explained further in the methods [Color figure can be viewed at wileyonlinelibrary.com] T A B L E 1 Descriptions of dune sites, showing their site name and labels in this study, their state of stability before the fire and a description of the dominant landscape features and area. common and occur annually, principally from intentionally humaninitiated burn-offs and dry lightning strikes (Peace, 2012  3 and 4 (grid columns Z and X), was mostly unburnt and reflects pre-fire vegetative cover (Figure 1). Throughout the rest of the site, vegetative cover was highly heterogenous with species presence/absence strongly driven by the degree of exposure. The parabolic dunes, characterised by their trailing ridges and U or V shapes (Figures 3 and 4), have developed according to the multidirectional wind regime ( Figure 2). Significant sediment is supplied from a high-energy surfzone-near-shore wave environment.
To illustrate the changes that occurred in these sites, landform units are described in representative panels with changes shown between multiple time series groups. Within Maupertuis Bay, these are shown in Table 3. These five regions are composed of landforms units on both previously active dunes and stable or accretionary areas (Table 3 and Figures 3 and 4). Active dunes are shown by panels A and E, illustrating a foredune-blowout complex and an active blowout on the crest of a parabolic dune. Pre-fire, panels C and D were fully  Table 4. These five regions are composed of landform units on both previously active dunes and stable sheltered areas (Table 4 and Figures 5 and 6). Active dunes are shown by panels A, B and D, illustrating a foredune near the intermittent mouth of an estuary, a blowout cutting across the estuarine barrier and a foredune-blowout complex. Pre-fire, panels C and E were semi-stabilised illustrating a dune ridge within a relatively sheltered sand plain and a higher dune ridge behind the primary foredune-blowout complex.

| Snake Lagoon-Site C
Site C is located $7 km inland from Maupertuis Bay in Flinders Chase National Park (Figure 1). It is located within the Gantheaume Dunes formation (Northcote, 2002), with the study site covering a Holocene stabilised relict parabolic dune (Figures 7 and 8). Pre-fire vegetation was characterised as dense mallee Eucalyptus, with differential vegetation cover according to the relative location within the landscape.
The representative landform units of Snake Lagoon are shown in

| SfM methods
The SfM methods workflow is presented in Figure 9, with detailed parameters of individual surveys (

| Flight design and spatial data acquisition
UAV flight surveys were designed to cover the three sites (Table 1) with specific flight information presented in Table 6 as noted in Table 6.
UAV image collection consisted of multiple flight lines at 50 m above flight location in a crosshatch pattern with nadir and off-nadir (70 degree) imagery using a hover and capture flight mode. Multiple mission start locations within each site were utilised, to maintain line of site of the UAV and to ensure that the starting flying height above ground was maintained. Site C's take-off location was centrally located within the study, next to the GCP in Figure 8 (grid 2/Y), which resulted in reduced photo overlap on the surrounding higher points.
The effects of the reduction in photo overlap will be discussed in further sections. Forward and lateral overlap of missions was set to a minimum of 75/60 respectively, with two orthogonal blocks forming a double grid. Variable image totals between surveys (Table 6) are attributed to dynamic survey conditions and shifting target areas, which were ultimately extracted to coverage according to spread of ground control. Local lighting conditions are summarised, and average wind speed is sourced from the nearest operating weather station as described in Table 6.

| SfM processing
Surveys were processed in MS with uniform processing parameters (Table 7). Georeferencing was prioritised on control points in images where targets were centrally located to reduce edge distortions. There  (Table 7) were chosen after extensive testing and according to works using a similar sensor (Cooper et al., 2021; Daakir, et al., 2020).

| SfM error
Clear and comprehensive error reporting shows the significance of observations with respect to the limits of spatial resolution and precision of the sensors and methods used. Following the guidelines of James et al. (2019), the error metrics in this work were designed to illustrate both the inherent systematic (bias) and random errors present in this work. Care was taken to conservatively interpret significant changes between surveys due to uncertainties arising from the methods used, as discussed in further sections.
Measurements of accuracy are from MS's total error, which includes individual RTX point precisions, and generated from the distance between control/check and its resulting modelled position in space giving the root mean square error (RMSE).
Precision estimates were generated from bundle block uncertainties across the surveys, exported with tie-points from MS and doubled as suggested by James et al. (2020). These spatially variant precision estimates were interpolated to point clouds in CC and used as X, Y and Z precision maps (PMs) within the CC workflow (Table 8). given by where subscripts N1 and N2 denote surveys with error of one sigma (σ) around points that are aligned to corresponding axes (X, Y or Z). In Equation (1) A-E extent indicators are described in Table 5, shown in Figure 7 and subsequent result figures [Color figure can be viewed at wileyonlinelibrary. com] where σ N1 2 and σ N2 2 was the total error of X, Y and Z directions of check points (RMSE) from MS.
Residual errors from GCPs were tested for systematic error according to James et al. (2020). The dense cloud, reference data and sparse cloud as exported from MS were tested in Matlab 2020b.
Where significant error was detected through check points, selections of CCPs were adjusted to areas exhibiting errors to adequately reflect realistic error propagation bounds into Equation 2.

| SfM change detection
To measure significant surface and vegetation changes between surveys, the M3C2-PM plugin of CC was used. M3C2 was developed by Lague et al. (2013) to perform a direct change comparison between point clouds, with PMs added by James, Robson, and Smith (2017)

| Ground cover changes
NBR uses the combination of near-infrared (NIR) and shortwave infrared (SWIR) bands as they are sensitive to water content, woody biomass and greenness. The normalised difference vegetation index (NDVI) using bands 8 842 and 4 655 is given by NDVI uses the NIR and red bands to estimate greenness and chlorophyl concentrations. The brightness adjusted disturbance index (BADI) is given by

| RESULTS
Results are divided into five sections: the first gives an overview of errors associated with the SfM processing and field data collection.
The next three sections illustrate the net and gross changes that F I G U R E 1 0 Changes in M3C2-PM distance as extracted for areas with significant changes above error thresholds. These results show the total changes measured in this study site as per the first and last surveys conducted. The grid columns Z:T and grid rows 1:4 correspond to the spatial extent of Figure 4 with the extent indicators showing panel regions A:E as described in Table 3 [Color figure can be viewed at wileyonlinelibrary.com]

| SfM error
The error between SfM surveys is a product of Equation (1) (LoD 95% ) and used to interpret significant changes between dates, including the estimated precision (PMs) and measured accuracy (CCP RMSE) errors.
The measured errors from SfM surveys are presented in Table 9.
Measures of the GCP and CCP error were sourced from MS total errors, which include individual RTX-GNSS precisions. The range of RMSE shows close alignment with RTX average horizontal and vertical precisions (1.5/3 cm), suggesting that the majority of surveys were relatively good fits without significant systematic error. All surveys were tested for systematic doming or dishing errors (James et al., 2020), with no surveys presenting significant error that could be modelled (P < 0.05). Spatially distributed check points (CCPs) were used, instead of GCPs, to quantify accuracy between surveys to include within error propagation (Equation 2).
The final survey of site C presented considerable check point errors in certain regions, located along high points within the survey. The areas with increased error correspond to regions with reduced photo overlap as flight plans had a constant height above take-off location, as described in methods. Additionally, substantial vegetation re-growth reduced reliable tie-points and network geometry across the survey.

| Maupertuis Bay
Maupertuis Bay (  F I G U R E 1 1 Changes in M3C2-PM distance as extracted to areas with significant changes above error thresholds between surveys. The panels A:E correspond to the same spatial extent of the panels in Figure 10 with descriptions in Table 5 [Color figure can be viewed at wileyonlinelibrary.com] representative landform units in panels A:E (Table 3) with net change in Figure 10, intra-survey gross changes in Figure 11 and volumetric totals in Figure 12.
The results show little evidence of geomorphic surface changes outside of highly exposed regions as shown in panels A and E of  Figure 11 shows the detailed intra-survey surface and vegetation changes that occurred during this study. Panels A and E of Figure 11 show the differential rates of erosion within this study's duration. The majority of negative surface changes occurred in the first 4 months post-first survey, corresponding to the spring and early summer months. Conversely, panel C of Figure 11 shows an increased rate of vegetation regrowth in sheltered areas in the latter two frames of F I G U R E 1 2 Shows the M3C2-PM surface gain (>0) and loss (<0) from significant changes according to satellitederived classifications of pre-fire sand, burnt or unburnt pixels. Exact volumetric change (m 3 ) totals and area summary statistics are provided in Tables S10 and S11. [Color figure can be viewed at wileyonlinelibrary.com] F I G U R E 1 3 Changes in M3C2-PM distance as extracted to areas with significant changes above error thresholds between the first and last survey. These results show the total changes measured in this study site. The grid columns Z:T and grid rows 1:2 correspond to the spatial extent of Figures 5 and 6 with the extent indicators showing panel regions A:E as described in Table 3 (Tables S10 and S11).

| Hanson Bay
Hanson Bay (site B) is a semi-stabilised coastal dunefield on the southern coast of KI. This site was mostly burnt, with fire removing above-ground biomass and vegetation across the site (Figure 1), and with just a few relatively small patches unburnt on the stoss side of the foredune. Changes that occurred in the site are shown and described in representative landform units in panels A:E (Table 4) with net change shown in Figure 13, intra-survey changes in Figure Figure 14 shows the intra-survey surface and vegetation changes.
Panels A and D of Figure 14 show the shifting trends of positive and negative surface changes that occurred during this study. Considerable surface changes occurred in panel B of Figure 14 during the F I G U R E 1 4 Changes in M3C2-PM distance as extracted to areas with significant changes above error thresholds between surveys. Panels A:E correspond to the same spatial extent of Figures 5, 6 and 13 with descriptions in Table 4 (Tables S12 and S13).

| Snake Lagoon
The Snake Lagoon dunes are the landward portion of the Maupertuis dunefield. Before the fire, Snake Lagoon (site C) was described as a stabilised transgressive-parabolic dunefield complex. All vegetation was completely burnt as shown in Figure 1. Changes that occurred in the site are shown and described in representative landform panels A:E (Table 5) with net change shown in Figure 16, intra-survey changes in Figure 17 and volumetric totals in Figure 18.
The results show no evidence of significant surface changes that suggests instability, with differential regrowth of vegetation across the study site. Figure 16 shows mostly positive M3C2-PM distances indicating that the net change is predominantly vegetation regrowth or accretion of sediments. Overall, net changes suggest differential regrowth rates, with distinct regions within the study area exhibiting varying degrees of positive changes. As shown by panels A and E of    (Tables S14 and S15).

| DISCUSSION
The fires of 2019/2020 burned coastal dunefields across KI, altering the vegetation cover on both semi-stabilised and fully vegetated dunefields. Fire has been suggested to be a potential initiation mechanism for dune transgression in the literature (Barchyn & Hugenholtz, 2013;Filion, 1984;Filion et al., 1991;Hesp et al., 2022;Matthews & Seppälä, 2013) with increases in vegetation with differential regrowth in the F I G U R E 1 6 Changes in M3C2-PM distance as extracted to areas with significant changes above error thresholds between the first and last survey at Snake lagoon. The grid columns Z:T and grid rows 1:2 correspond to the spatial extent of Figure 5 with the extent indicators showing panel regions A:E as described in Table 5 [Color figure can be viewed at wileyonlinelibrary.com] sheltered regions of coastal sites (panels C, Figures 10 and 13) and in the relatively low-lying areas of Snake Lagoon (panels A, C and, E, Figure 16) compared with higher sites in the landscape in similarity to the results from post-fire dune research (Cowling & Pierce, 1988;Vestergaard & Hastings, 2001), north and south facing slopes show distinct differences in vegetation regrowth and ground cover. The intra-survey results show the variable regrowth rates related to seasonality, with higher 3D changes observed in the late summer to early autumn months (Figures 11 and 17), which is supported by the seasonal increases seen in the spectral signatures ( Figure 19). The trend of increasing M3C2-PM totals from burnt pixel areas in the later months of the survey (Figures 12 and 18) corresponds to the seasonal increases in vegetation growth of the sheltered and low-lying areas shown in Figures 11 and 17.

| Vegetation cover changes from optical satellites
A time series of three space-based optical indices is presented to show the relative spectral index trajectory before, during and after the time series of UAV surveys, to place the index values in context to pre-fire patterns (Figure 19). Although spectral change interpreted as ground cover change is not able to illustrate the diversity or health of F I G U R E 1 8 Shows the M3C2-PM surface gain (>0) and loss (<0) from significant changes according to burnt pixels of site C. exact volumetric (m 3 ) totals and area summary statistics are provided in Tables S14 and S15 [Color figure can be viewed at wileyonlinelibrary. com] F I G U R E 1 7 Changes in M3C2-PM distance as extracted to areas with significant changes above error thresholds between surveys. The panels A:E correspond to the same spatial extent of Figure 16 with descriptions in Table 5 [Color figure can be viewed at wileyonlinelibrary.com] an ecosystem, it provides a spectral-temporal baseline of vegetation that is used here to demonstrate the progression of these dunefields as they return to a post-fire state.
Results show the increasing trend of ground cover that occurred during this study, with all sites approaching pre-fire index minimums according to 4-month geomedians. Geomedians retain the spectral relationships between bands and facilitate the computation of indices in time series (Roberts et al., 2017). Two spectral indices are chosen to align with established research methods (Hislop et al., 2018) (Gascon et al., 2017) in contrast to nonmetric UAV sensors which are limited by inconsistent illumination in orthomosaics and spectral resolutions (Eltner & James, 2022).
The increases in spectral indices, interpreted as increasing vegetation cover (Figure 19), correspond with the results from the UAV surveys. The trend of dunefield revegetation and landscape stability after severe fire follows similar findings of other post-fire dune studies who have found little evidence of dunefield reactivation post-fire (Barchyn & Hugenholtz, 2013;Shumack & Hesse, 2018;Shumack et al., 2017). Evidence of dunefield destabilisation following fires is limited to interpretations of the stratigraphic record (Boyd, 2002;Filion, 1984Filion, , 1987Filion et al., 1991;Kotilainen, 2004;Seppälä, 1995;Tolksdorf et al., 2013), and central to the suggestion of fire and landscape instability is a coincident increase in aridity or drought conditions post-fire acting as a catalyst for reactivation (East & Sankey, 2020;Fisher & Hesse, 2019). Figure 20 shows the monthly mean maximum temperature for Cape Borda and monthly mean rainfall totals for Rocky River between 1960 and 2019, plotted against modelled rainfall and observation data for the corresponding 2016-2022 period in Figure 19.
Modelled rainfall monthly totals were interpolated from rainfall gauge data (Beutel et al., 2019) across the extents of the UAV surveys. Following the fire, rainfall totals show that mean precipitation was high, whereas temperatures were below average, and thus, a wet and mild summer occurred throughout the time of the UAV surveys. The rainfall data show a wet month following the fire, which likely resulted in removal of surface sediments. Six months later, significant sheet flows and gullying of sediments within the burnt landscape provide direct evidence of the impacts of rainfall-induced erosion ( Figure 21). Studies have shown that fire drastically alters the susceptibility of landscapes to surface water erosion, showing that even short-duration rainfalls with moderate intensities can trigger significant post-fire surface water erosion (Lopes et al., 2020), although the high rate of recurrence of fire in Australian landscapes may reduce the geomorphological role of wildfires in producing significant sediment mobilisation (Shakesby et al., 2007).

| SfM change detection and error
The change detection workflow used here is based on two central  et al., 2013) and its PMs  were developed for surface changes, the workflow used here and the above assumptions benefit from its ability to robustly detect changes across point clouds with spatially variant errors as in the case of SfM generated datasets. Additionally, SfM datasets are widely used for vegetation-related applications, such as calculating above-ground biomass (Cunliffe et al., 2022) and in multitemporal studies of canopy changes (Zhang et al., 2021). For this study, the increases in vegetation height reduced the efficacy of SfM point cloud generation but also illustrated the lack of dunefield instability or transgression following severe fire.
The principal limitations of the methods used here are attributed and reviewed according to the following: reduction of functional photo-overlap due to vegetation regrowth, sensor-specific hardware issues and operational from inconsistent above-ground flying heights.
The increasing surface roughness of regrowing leafy vegetation reduces effective overlap, tie-point precisions and subsequent point cloud reconstructions (James et al., 2020). Through the MS workflow outlined in Table 7, tie-points were iteratively reduced according to their error or uncertainty, with point precisions over high roughness regions representing the highest values of uncertainty in PMs. As uncertainty increases, tie-points are reduced and reconstruction in regions of dense vegetation is less robust. Additional limitations of this study arose from the use of the Mavic 2 Pro. Its rolling shutter sensor has been shown to reduce image clarity and is not recommended for monitoring environmental changes (Eltner & James, 2022;Stark et al., 2021). Although the precise magnitude of error directly from the rolling shutter is difficult to isolate, the use of rolling shutter corrections attempts to model sensor movement within the bundle block adjustment and has been shown to improve results (Zhou, Rupnik, et al., 2020).
After iterations of reducing uncertain tie-points (Table 7), camera calibration coefficient correlations displayed high correlations between focal (f) and radial (k1-k3) lens distortion coefficients and between principal point shifts (Cx, Cy) and tangential (p1, p2) lens distortion coefficients (see error reports linked in the Supporting Information) suggesting potential over-parameterisation. Although some parameters are acknowledged to have inherent correlations (Senn et al., 2022), it is suggested to remove individual parameters if their error exceeds the coefficient value . Camera calibration coefficients in this study show errors that are significantly lower than their coefficient values, with errors representing a miniscule fraction of the corrections applied. Although there is not consensus regarding the optimal parameters available in SfM software (Śledź & Ewertowski, 2022), flight design and acceptable error bounds need to be fit for purpose to answer the research objectives considering the scale, resolution, precision and accuracy of the methods (Anderson, 2019).
In this study, flight planning incorporated a double grid pattern, including nadir and oblique imagery, and distributed RTX control and checks points to ensure reliable geometric camera calibration (Eltner & James, 2022). The inclusion of off-nadir imagery in aerial grid surveys has been shown to improve scale variation within and between images and improves self-calibration within the bundle-block adjustment (Luo et al., 2020;Nesbit & Hugenholtz, 2019;Sanz-Ablanedo et al., 2020). The high-relief landscapes of the dune sites also increased the scale variation within and between images, which benefits network geometry (Fonstad et al., 2013;Fraser, 2013). In sur-

| CONCLUSIONS
This study focuses on the post-fire response of burnt coastal dunes on KI, South Australia, using a variety of remote sensing techniques and methods. The three sites studied here represent the landscape diversity of both active, partially vegetated and relict stabilised dune fields in a temperate climate. Fire has been suggested to be an initiator of coastal dunefield transgression, but the results of this work show no evidence of aeolian destabilisation or widespread landscape instability in the months following a wildfire. In situ bi-monthly UAV surveys are used to reconstruct landscapes and compared to describe the significant geomorphic and vegetation changes and assess potential dunefield transgression. The UAV surveys show that negative surface changes associated with aeolian erosion and destabilisation occurred in the coastal sites, predominantly occurring in the spring and early summer months in highly exposed regions. Positive 3D changes show vegetation regrowth occurring across all sites, with distinctly increased rates in sheltered and relatively low-lying regions.
The high temporal resolution illustrates the variable vegetation regrowth rates and landscape dynamics related to seasonality, which are observed in the spectral signatures from satellite indices. Errors in this method are reviewed with the limitations of the UAV methods discussed specific to the objective of the research question. Without pre-fire baseline data or rapid post-fire assessments, changes that occurred in the immediate 6 months following the fire are unknown.
The logistics of rapid response surveys in protected wild areas precluded earlier UAV surveys so alternative data were used.
Satellite composite images were computed to calculate spectral indices to assess the ground cover changes that occurred in the years before and following the fire. During the time period of the UAV surveys, spectral indices increased and approached pre-fire baseline values. The results from the satellite time series agree with the UAV surveys showing that vegetation ground cover in dune sites is steadily increasing and suggests a continued trend of landscape stability and re-vegetation.
Overall, the results of this work correspond to similar recent research in post-fire coastal dune fields showing landscape stability following fire. The literature reviewed here suggesting fire as an initiation mechanism for landscape instability and dunefield transgression is largely sourced from stratigraphic data, suggesting that fire followed by changing or atypical climatic conditions produces instability. The short term of this study illustrates the structural and ground-cover trends changes in these dunefields but demonstrates a trend towards vegetation re-growth and increasing stability over time. We note that the short time period of the drone surveys in this study is a limitation in regard to explaining longer-term landscape dynamism.
As reviewed in the introduction, fires are natural and cyclical in Australian landscapes but shifts in fire regimes or climate may alter dunefield structure in the longer term. Following the most extensive fire on record, however, the results presented here do not suggest landscape instability or a re-activation of previously stabilised transgressive or parabolic dunefields. This may be due to the occurrence of an immediate post-fire climate, which was mild and wet; alternatively, it may indicate that stratigraphic interpretations of widespread landscape instability following fire may need to be re-assessed.