1 Introduction

Dune management is considered a prime coastal management tool [60] and whilst prior monitoring efforts to capture fine-scale features required laborious field surveys over relatively limited spatial scales (due to the intensive labour requirements) [35, 59], the past two decades have shown marked advances in describing, and quantifying, the dynamic processes affecting coastal beach and estuarine systems. From the early works of Brock et al. [9], Gutierrez et al. [29], and Stockdon et al. [59], airborne light detection and ranging (LiDAR) (from manned aircraft) has been increasingly used and become an important tool in coastal geomorphology through facilitating the acquisition of detailed, and accurate, topographic data over broad coastal regions and enabling geomorphic analysis over a continuum of scales [29]. Mitasova et al. [45], for example, utilized airborne multi-temporal LiDAR to detect, analyse, and quantify topographic changes in rapidly evolving coastal landscapes. Similarly, the robust studies of Revell et al. [50], Allen et al. [7], and Loftis et al. [43] utilized airborne LiDAR for coastal terrain modelling and change detection. More recently, Dong [20] employed airborne LiDAR and terrain modelling to automate measures of dune migration whilst Brownett and Mills [12] and Lalimi et al. [39] have combined airborne LiDAR with hyperspectral imagery to map and classify vegetation patterns in coastal dunes.

As the costs associated with airborne LiDAR have declined, the use has increased as evidenced by Brock and Purkis [10] and such studies as Claudino-Sales et al. [16], Splinter et al. [58], and Turner et al. [62] who utilized airborne LiDAR data, flown pre- and post-storm events, to understand, and predict, along-shore variable sand dune erosion. As expressed by Pikelj et al. [49], however, airborne LiDAR from manned aircraft can remain prohibitively expensive and thereby potentially limit temporal-coverage. As technology has advanced, however, the utilization of unmanned aerial vehicles (UAVs) in coastal monitoring has increased and are proving to be an efficient, cost-effective, tool for topological mapping and measurement with deployment flexibility and spatial-temporal resolution (‘microscales’ [54]) not previously feasible within the coastal zone [21, 62]. To date, studies utilizing UAVs have focused almost exclusively on the use of structure from motion photogrammetry (SfM) and whilst this has shown to be very effective in creating, detecting, and measuring change using Digital Surface Models (DSMs) with sub-metre resolution [26, 33, 48, 53, 61], such studies may be somewhat limited by the characteristics of the surface being measured, presence of vegetation, water, and small-scale texture contributing to inaccuracies [17, 34].

Unlike that of the SfM approach, UAV LiDAR is not restricted to the development of DSMs but is able to ‘penetrate’ dune vegetation for the development of accurate Digital Terrain Models (DTMs) and facilitate a more direct measure of dune structure relative to local processes driving geomorphic change. In addition, the high spatial resolution and capacity for regular monitoring, UAV LiDAR could better inform morphodynamic and beach dune erosion models, important to coastal scientists and engineers and provide the theory, data, models, and predictions that planners, managers, and policymakers require [54].

Whilst traditional point cloud methods are suitable for monitoring change in beach topography [21], here we describe, test, and evaluate a new approach in the deployment of UAV LiDAR across an existing blowout feature along the coastal dunes of northern Bribie Island, Queensland. Southeast Queensland is considered one of the coastal ‘hotspots’ in Australia due to increasing population pressures and threat of rising sea levels due to ocean thermal expansion [15] and the Sunshine Coast City Council (SCCC) has identified our study location as an area of concern where further degradation of the existing blowout could result in ocean flow-through to Pumicestone Passage, adversely affecting local hydrodynamic characteristics of navigable waters.

Here, we describe a methodology that, if expanded upon with further research, could provide a new tool for coastal survey applications on micro- to mesoscales (as defined by Sherman and Bauer [54]) and facilitate opportunities to ask new questions regarding the processes affecting observed changes. In addition, results from this study provide a quantitative performance assessment of the Hovermap LiDAR within a dynamic environment and assist in identifying the strengths, and limitations, of this system in the acquisition of accurate 3D point clouds associated with coastal survey and mapping research and monitoring.

2 Materials and Methods

2.1 Study Area Location and Timing

Bribie Island is the northernmost, and smallest, of the major sand islands of the Moreton Bay Marine Park, Queensland, Australia (Fig. 1(A)) and is part of an internationally recognized Ramsar wetland and migratory shorebird habitats [1]. Located approximately 70 km north of Brisbane in eastern Australia, Bribie Island (Fig. 1(B)) is separated from the mainland by Pumicestone Passage, characterized by intertidal mangrove and saltmarsh habitat, whilst the 34-km western shoreline along the Coral Sea (Southwest Pacific Ocean) consists of sandy beach and coastal dune formations [15]. Most of the island is designated as a National Park and Recreation Area (~ 5580 ha) with four-wheel drive (4WD) and camping activities allowed with corresponding access permits [2].

Fig. 1
figure 1

Location of the study area (A) on the Pacific coast of Queensland, Australia, along the (B) northern Bribie Island beach. (C) Inset of the Study Area (MGA Projection, GDA 94 Datum). Photo source: Esri World Imagery July 2017

Full size image

The Study Area (26° 50.5′ S, 153° 7.7′ E) corresponds to a segment of coastal sand dune along northern Bribie Island where washover processes have compromised the dune structure and foredune vegetation as evidence by an existing blowout and overwash fan formations (Fig. 1(C) and Fig. 2). This relatively narrow (~ 150 m wide) portion of the island was identified by the SCCC as an area of concern in that continued erosional processes might cut off northern Bribie Island and alter hydrodynamic processes of the adjacent waters of Pumicestone Passage, its shoreline, and associated infrastructure (e.g. adversely affect established navigation channels). With increased monitoring focus, this location represents ideal testbed for evaluating the utility of new monitoring techniques at microscales such as UAV LiDAR.

Fig. 2
figure 2

Northerly ground perspective of the study area centred on the blowout and overwash fan formation that bisects this relatively narrow (~ 150 m) part of northern Bribie Island

Full size image

To observe temporal variation across microscales [54], four surveys were conducted every 3 months (quarterly) from July 2017 to April 2018 (Table 1). The timing of this study was designed to commence prior-to, and run through, the official ‘cyclone season’ (1 November to 30 April). In effort to minimize variability in coastline exposure due to tidal regime, each survey was undertaken 5 days following the new moon with the exception of Survey4, which was delayed 3 days due to a local storm event. All UAV flights were executed in the mid-to-late afternoon to coincide with the falling tide in effort to minimize variability in beach exposure during the tidal cycle.

Table 1 Timing of surveys and intervals between each event within this study
Full size table

2.2 Georeferencing and Co-registration

Given that the coastal dunes on Bribie Island are a dynamic environment, it was not practical to establish fixed ground control points (GCPs) over the duration of this study. As such, each scanning event was preceded by the placement of six (7.5 cm) laser registration targets across the study area. Targets were representatively distributed with three across the foredune and dune crest, respectively. A Leica Viva GS16 GNSS ‘Smart Antenna’ with CS20 Controller (Leica [4])Footnote 1 was used in combination with a connection to the Geoconnect SmartnetAUS RTK (Real-Time Kinematic) network to georeference each target in the MGA projection, GDA 94 Datum with a < 20-mm 3D solution (Fig. 3). The GCPs were later utilized to co-register individual scans and facilitate comparative analysis.

Fig. 3
figure 3

Example imagery of the a laser registration targets and b Leica GS16 GNSS antenna with CS20 controller utilized to georeferenced and co-reference individual laser scans

Full size image

2.3 Leica P40

Similar to the work of Fabbri et al. [23] and Zhou et al. [71], a Leica P40 terrestrial laser scanner (P40) [41]Footnote 2 was utilized to capture a high-definition, high accuracy 3D image of the Study Area, and serve as a comparative ‘baseline’, just prior to the UAV flight during Survey1, like that of Elsner et al. [21]. With a reported range accuracy of ± 1.2 mm, 3D position accuracy of 3 mm (at 50 m) and scan rate up to 1,000,000 pts.·s−1, the P40 was deployed at nine locations, three along beach seaward of the foredune, three on the dune crest, and three within the central hind-dune complex. Individual scans were then co-referenced, unified, and cropped to retain only points within the Study Area boundary using the Leica Cyclone 3D Point Cloud Processing Software (v.8.1.1).

2.4 Hovermap and UAV Platform

Initially described by Kaul et al. [36], the Hovermap is a lightweight (1.5 kg) 3D LiDAR mapping payload specifically designed for small UAV platforms. Utilizing a proprietary Simultaneous Localisation and Mapping (SLAM) solution to generate 3D point clouds Hovermap does not require GNSS and therefore not subject to the same challenges as other systems that are dependent on satellite-derived positional information [66]. Although not dependent on GNSS and IMU sensor data to generate point clouds, the Hovermap was connected to, and utilized information from, the onboard GNSS system of the UAV platform (± 0.5 m vertical, ± 1.5 m horizontal) to facilitate georegistration.

Here, the Hovermap [22]Footnote 3 was deployed using the Velodyne (VLP-16) ‘Puck-LITE’ sensor. The VLP-16 is a 16-channel dual-return sensor with a scan rate up to 600,000 pts.·s−1, angular resolution (vertical) of 2.0°, field-of-view (FOV) along the z-axis of 360° × 30°, reported accuracy of ± .03 m, and range up to 100 m [64].Footnote 4 As integrated on the Hovermap, the VLP-16 is also rotated 360° about the y-axis at a rate of 0.5 Hz effecting a near 360° × 360° field-of-view of the surrounding environment, irrespective of the direction of travel, other than where obstructed (e.g. by the UAV).

All flights were conducted using a DJI ‘Matrice (M600) Pro’ UAV platform (DJI Science & Technology, China),Footnote 5 capable of flight times with the Hovermap between 20 and 25 min. As a self-contained unit, the Hovermap is easily integrated onto any suitable UAV platform and was secured to the M600 in this study using a straightforward anti-vibration mount to minimize the potential influence of high-frequency vibration (Fig. 4).

Fig. 4
figure 4

The a Hovermap LiDAR as mounted to the DJI M600 UAV platform and b shortly after takeoff on at the Study Area

Full size image

2.5 Flight Planning

Previous work by Sofonia et al. [57] concluded that the range of the Hovermap was (i.e. altitude of the UAV) the single most important variable with regard to point cloud density and accuracy, with the best overall performance observed at relatively low altitudes. Based on this, and on the previous work of Wallace et al. [65], flight altitudes for all surveys was set to 20 m above the beach ground level. This effectively minimized the UAV altitude whilst maintaining a safe separation from trees established in the hind-dune complex. To maximize point cloud density and the number of ground returns through vegetation, a ‘cross-flight’ similar to that as discussed by Gerke and Przybilla [24] was employed. Flight speed was set for each flight at 4 m·s−1 as determined as the minimum speed required to cover the estimated distance travelled in a cross-flight pattern over the Study Area (~ 4750 m2) within a conservative estimate of the maximum UAV flight time (20 min).

Each flight was planned using a geo-referenced Esri World Imagery aerial photograph of the study area and Global Mapper (v.18.2, Blue Marble Geographics, USA) software. To maximize point cloud density and canopy ‘penetration’, flight lines were plotted at 20 m intervals using the Grid tool to create the ‘cross-flight’ pattern (Fig. 5a). Exported as *.kml files, these waypoints were imported into the Autopilot software (v.3.8, Hangar Technology, USA) used to control the UAV with speed and altitude set as previously described (Fig. 5b).

Fig. 5
figure 5

Screen captures of autonomous ‘cross-flight’ planning for (a) in Global Mapper (MGA Projection, GDA 94 Datum, photo source: Esri World Imagery July 2017) with the (b) corresponding flight plan as loaded in the Hangar Autopilot software

Full size image

2.6 Marine Weather

To better understand the process forcing observed geomorphological change within the Study Area, marine wind and wave data were obtained from the nearest relevant monitoring stations covering the duration of the study [68]. Specifically, wind speed and direction measurement recorded every 3 h were sourced from the Australian Government Bureau of Meteorology ‘Spitfire Channel’ Weather Beacon (040927) [3] located (27° 2.9′ S, 153° 16.0′ E) approximately 27 km southeast (brg. 148° 35.8′) of the Study Area. Similarly, half-hourly significant wave height measures were obtained from the Queensland Government Coastal Impacts Unit, ‘Caloundra’ Wave Monitoring Buoy [5] located (26° 50.8′ S, 153° 9.3′ E) approximately 2.6 km east (brg. 100° 28.5′) of the Study Area (Fig. 1). Daily averages were calculated for all values to facilitate analysis with groups of the same size (n) for each interval, a term used here as the time between survey events.

2.7 Initial Data Processing

Sofonia et al. [57] describe method for pre-processing UAV LiDAR point clouds with the Python3 coding language [63] and discuss several benefits in doing so including a standardized approach to the removal of noise and other undesired points and improved basis for inter-flight comparisons. This script was employed here with the parameters listed in Table 2. Point clouds were subsequently filtered and cropped as illustrated in Fig. 6. Descriptive statistics for each flight were recorded including altitude (m), speed (m·s−1), and area (m2) as well as resulting point cloud density (pts.·m−2), ground sample distance (GSD, m), and ‘sampling effort variable’ (SEV, s·m−2).

Table 2 Key parameter settings utilized in data pre-processing across all flights using the Python script described by Sofonia et al. [57]
Full size table
Fig. 6
figure 6

Example outputs (local coordinate projection) from Survey1 (coloured by range) illustrating (a) the original point cloud and (b) points retained post-processing with the Python script described

Full size image

The unified cleaned and cropped P40 and Hovermap data were imported into CloudCompare (v.2.10 alpha) [25] wherein the initial georeferencing of the point clouds was achieved by utilizing the GPCs from each survey and the Align (point pairs picking) tool. As the accuracy of this process is dependent on how well each point is selected relative to the corresponding GCP, this process was repeated until the final root mean square error (RMSE) was as low as possible (\( \overline{x} \) = 0.050 ± 0.031 m). Each georeferenced point cloud was then brought into Global Mapper and cropped to the Study Area boundary (Fig. 7a). Given the objective of this study was to evaluate the ability of the Hovermap to detect microscale change in local geomorphology, the point cloud was spatially constrained to focus on that part of the Study Area that was likely to experience significant elevation changes [11] over relatively short time scales (i.e. months). Illustrated in Fig. 7b, this area is termed here as the Analysis Area and represents the portion of data exported to 3DReshaper (3DR) 3D scanning software package (v.17.0.24477.0, Technodigit, France) for further interrogation.

Fig. 7
figure 7

Example georeferenced point cloud (coloured by elevation) of (a) the Study Area post-processing with the Python script and (b) the portion of the cropped point cloud retained (Analysis Area) for detailed analysis (MGA Projection, GDA 94 Datum)

Full size image

2.8 Terrain Modelling

Primarily interested in temporal changes of the foredune stoss slope and beach surface, the Ground Extractor tool within 3DR was utilized to classify and retain only ‘ground-points’ and, similar to previous studies [7, 27, 43, 45], create DTMs of the Analysis Area. Here, DTMs were created using a slope setting of 55°, ‘local steep slopes’ strategy and refined using an average point distance of 0.020 m as derived from the observed mean GSD of the Hovermap point clouds (Fig. 8).

Fig. 8
figure 8

Example mesh creation using Hovermap point cloud data from Survey2 illustrating (a) the original point cloud, (b) points classified as ‘ground’, and (c) the resulting mesh object (DTM)

Full size image

2.9 Elevation, Slope, Deviation, and Cubature

Elevation and slope have long been common data sets for assessing the basic structure of sand dunes [42, 51, 54, 55, 67] with change detection a key monitoring aspect of understanding, modelling, and modelling geomorphic change [23, 45, 47, 59, 69]. The use of mesh DTMs (or similar) for this type of inspection has been in use since at least 2003 [46] and is now a relatively standard approach with continued improvement in software capability and PC performance [6, 23, 34, 40, 43, 71].

Here, P40 and Hovermap mesh objects of the Analysis Area were imported into 3DR where measures of foredune and beach elevation and slope were calculated using Colour Along a Direction (z-axis) and Slope Analysis tools (max slope tolerance = 90°), respectively. To aid in the visualization of meaningful differences, the default ‘continuous’ colourisation scheme was reduced to a user-specified number of ‘levels’. Here, thirteen levels were utilized for elevation (one level per 0.5 m difference) and six (one level per 15° difference) across the slope analysis. Each elevation mesh was then exported in *.asc “Vertices Only” format to retain only the x, y, and z position information for each whilst exporting slope as an ASCII *.ply file retained the position and slope data as calculated by 3DR.

In addition to elevation and slope, 3DR allows a mesh-to-mesh comparison of the difference, or deviation, between two models. Similar to ‘DEMS of Difference’ comparisons of Le Mauff et al. [40], two meshes were selected and the Compare/Inspect tool was applied with the ‘ignore points with a distance greater’ than function disabled. As we were primarily interested in vertical differences between surveys, the option of applying a 2D inspection along the z-axis was utilized. This data was then exported as a *.csv file for further interrogation. This approach is supported by the work of Zhang et al. [70] who indicate that the relative vertical accuracy between the compared DEMs is more important than the absolute estimations derived by comparing each survey to GCPs.

Similar to the analysis of Claudino-Sales et al. [16], Allen et al. [7], and Jaud et al. [34], volumetric differences between mesh models were of interest. As such, the cubature function of 3DR was also utilized along the z-axis (user defined) to quantify the approximate volume of sand removed and/or deposited (i.e. below/above reference surface) during each survey interval (Fig. 9). Approximated volumes were recorded directly as displayed within 3DR.

Fig. 9
figure 9

Example cubature analysis output with observed volumetric differences between Survey3 (green) and Survey4 (red) during which approximately (1) 239 m3 of sand was deposited on the dune crest whilst (2) 1534 m3 was removed from undercutting and erosion processes during Interval3

Full size image

2.10 Data Analysis Structure and Interrogation

In each case, the preceding survey was utilized as the reference surface, with the exception of the P40 to Hovermap comparison (Survey1), wherein the P40 data was utilized as the reference model. With each mesh comprised of tens of millions of data points, a simple Python script was written to accept the exported data in the various formats, interrogate for descriptive statistics, and generate histogram data. Microsoft Excel (v.2013, Microsoft Corp., USA) was used to process the marine weather data as well as produce relevant wind roses, histograms and charts.

3 Results

3.1 UAV Flight Performance

As listed in Table 3, UAV flight performance was in line with the prescribed parameters and relatively consistent across all surveys with the exception of the increased altitude observed in Survey4 with corresponding increase on the observed range and decreased SEV. This is considered unlikely to have been an issue with the UAV platform performance but rather a consequence of changes in foredune topography (lower beach elevation) during Interval3 as evidenced in the results from the elevation, deviation, and cubature analysis.

Table 3 Descriptive statistics of UAV flight performance over the Study Area for each survey
Full size table

3.2 Hovermap Point Clouds

Similar to that of UAV flight performance, the corresponding Hovermap point clouds were all relatively consistent across each survey (Table 4). Analogous to the observations of Sofonia et al. [57], the inverse relationship between altitude/range and point cloud density is evident with corresponding increases in GSD. The relationships between density/SEV and GSD/EDR also appears to hold, however, a greater sample size with repeated measures would be needed to determine the appropriate coefficients and strength of observed correlations.

Table 4 Descriptive statistics of Hovermap point clouds over the Study Area for each survey
Full size table

3.3 Marine Weather

Marine weather data recorded over the study period showed that daily averaged wind speeds ranged from 7.9 to 41.9 m·s−1 (x̅ = 21.4 ± 7.4 m·s−1) predominantly from the south/southeast with wave heights of 0.3–2.5 m (x̅ = 1.0 ± 0.4 m). Although no cyclonic activity occurred at the Study Area during this investigation, significantly increased wind speeds (ANOVA, F(1,180) = 6.257, p = 0.013) and wave heights (ANOVA, F(1,180) = 41.769, p ≤ 0.001) were recorded during Interval3 (Fig. 10). Descriptive statistics of the marine weather recorded during the study are provided in Table 6, Appendix A.

Fig. 10
figure 10

Daily wind (speed and direction) and significant wave heights recorded between (a) Interval1, (b) Interval2, and (c) Interval3

Full size image

3.4 Hovermap to P40 Comparison

Values across all measurement matrices, as determined from the Survey1 Hovermap data (Survey1-HM), were very similar to those derived from the P40 ‘baseline’. This not only demonstrates that Hovermap is comparable to more traditional terrestrial laser scanning techniques within the construct of this study but also provides an indication as to the suitability of this workflow in coastal monitoring applications. Specifically, minimum, maximum, and mean elevations from Survey1-HM were each within the reported ± 1.2 mm accuracy of the P40 (Table 5a). Similarly, the calculated values for slope (Table 5b) were also comparable, varying by approximately 7.4%. Mean deviation of Survey1-HM mesh to P40 mesh was 0.01 ± 0.03 m, consistent with the reported ±30 mm accuracy of the Velodyne sensor (Table 5c) with cubature reporting a net difference of 2.5 m3 across the Analysis Area (Table 5d). Results for elevation, slope, and deviation are also visualized using histograms and associated mesh models of Figs. 13a, 14, and 15a.

Table 5 Descriptive statistics for the elevation and slope analysis as well as the mesh-to-mesh comparison of vertical deviation and cubature observed across the Analysis Area
Full size table

3.5 Temporal Change

Temporal variance across the Analysis Area was relatively consistent in the measurement matrices over the course of the study, however, considerable changes were evident within the Survey4-HM model. Here, mean elevation decreased whilst both slope and deviation measured increased relative to the Survey3-HM mesh (Table 5). Most notably, a negative net loss of approximately 1295 m3 was observed in the cubature analysis during this period. Interestingly, the standard deviations observed relative to the mean values across all measures of Survey4-HM data also increased providing further evidence of a transition from a relatively uniform to more complex foredune environment.

These changes are better visualized in the histograms and mesh models of Figs. 13, 14, and 15, wherein an increased distribution of elevation and deviation values is also evident. Correspondingly, tighter topographic contours (Fig. 11d) and slope values greater than 40° were recorded in the Survey4-HM mesh (Fig. 12d), reflecting scarping of the foredune during Interval3. Whilst an overall decrease deviation values (Fig. 13d) were observed relative to the Survey3-HM mesh coinciding with the net volumetric loss of sand measured in the cubature analysis, sand deposition along the dune crest is also evident during this period.

Fig. 11
figure 11

Histogram outputs and mesh models of the Analysis Area elevation (0.5 m contours), coloured by elevation (m), observed in (a) Survey1-P40 and Survey1-HM, (b) Survey2-HM, (c) Survey3-HM, and (d) Survey4-HM. Red dashed box represents the approximate location of the existing blowout

Full size image
Fig. 12
figure 12

Histogram outputs and mesh models of the Analysis Area slope coloured by degrees (°), observed in (a) Survey1-P40 and Survey1-HM, (b) Survey2-HM, (c) Survey3-HM, and (d) Survey4-HM. Red dashed box represents the approximate location of the existing blowout

Full size image
Fig. 13
figure 13

Histogram outputs and mesh models of the Analysis Area mesh-to-mesh vertical (z-axis) deviation coloured by distance (m) between (a) Survey1-HM:Survey1-P40, (b) Survey2-HM:Survey1-HM, (c) Survey3-HM:Survey2-HM, and (d) Survey4-HM:Survey30-HM. Red dashed box represents the approximate location of the existing blowout

Full size image

Similar to that described by Short and Hesp [55], Carter and Stone [14], and Sherman and Bauer [54], the dynamic changes observed in Survey4-HM are most likely linked with the process forces associated with the relatively high-energy wind-wave climate recorded during Interval3.

3.6 Foredune Scarping and Blowout Area

As evidenced within the Survey4-HM mesh, substantial scarping of the foredune occurred during Interval3. This is best visualized when illustrating the same segments of the foredune, at reduced scale, and rotated to a perspective view of the stoss slope. Here, the change between Survey3-HM (Fig. 14a) and Survey4-HM (Fig. 14b) mesh is clearly visible and includes detail such as microscale wedge failures. These observations from the mesh data are supported by photographs of this section of the foredune during Survey4 (Fig. 14c, d).

Fig. 14
figure 14

Example undercutting and scarping of the foredune by storm waves prior during Interval3 when comparing the same segments of the (a) Survey3-HM and (b) Survey4-HM. Photos of this portion of the foredune taken during Survey4 illustrate (c) microscale post-scarp wedge failures with rhizome undermining/exposure and (d) earlier dune bedding features including (1) rapid precipitation deposition, (2) reactivation surfaces, and (3) aeolian cross bedding

Full size image

Similar to that observed across the Analysis Area, temporal change was also observed at the existing blowout. As illustrated in Fig. 15, relatively minor changes in elevation and slope (as perceived through 0.5 m contours) were detected through most of the study with the exception of Survey4-HM where scarping of the foredune and lowering of the beach were evident (Fig. 15d). Interestingly, the highest elevation (~ 4.1 m) of the blowout (southern (“-Y”) side of blowout trough) was also observed in Survey4-HM mesh associated with sand deposition on the dune crest. In addition, sand was deposited at the back of the trough slightly raising the elevation of this portion of the blowout compared with the Survey3-HM mesh.

Fig. 15
figure 15

Topographic change, coloured by elevation (m) at the existing blowout as observed in (a) Survey1-HM, (b) Survey2-HM, (c) Survey3-HM, and (d) Survey4-HM (0.5 m contour intervals)

Full size image

Although heavy scarping occurred along the foredune during Interval3 resulting in the steepening of the trough walls, the depth of the blowout remained relatively unchanged (min. ~ 2.2 m) throughout the course of the study. This suggests the presence of a deflation floor at, or near, this elevation that is potentially linked to the water table of Pumicestone Passage, an immobile layer of rock and/or, semi-permanent layer of shell, or gravel representing a deflation limit of this blowout and washover features of the foredune [13].

4 Discussion

Cracknell [18] stated that the key to success, or otherwise, of remote sensing in coastal or estuarine studies lies in the question of scale. Successful in mapping, detecting, and quantifying change on centimetre-level scales, with accuracy consistent with more traditional terrestrial laser scanning (TLS) techniques, results from this study demonstrate that the UAV LiDAR, such as the Hovermap, has strong potential as a tool for future research and monitoring of coastal dunes and other dynamic geomorphic systems.

4.1 Hovermap Performance Comparison

Similar to the UAV SfM work of Papakonstantinou et al. [48], we presented a UAV LiDAR workflow that facilitates the study of coastal changes, and accurate visualization of 3D models, of the beach and foredune over micro spatial and temporal scales. With regard to performance, the Hovermap reported a mean point cloud density of 2532 ± 170 pts.·m-2 and GSD of 0.02 ± 0.001 m across the Study Area. As a byproduct of the relatively low altitudes and flight speeds [57], the resolution achieved here was substantially higher than the 1–30 pts.·m-2 and 0.3–0.5 m GSD traditionally associated with airborne LiDAR from manned aircraft [6, 39, 45, 69]. The cost, however, is realized in the relatively limited spatial scales that can be achieved by UAVs compared with that of manned aircraft.

A closer comparison, therefore, may be made relative to previous UAV SfM studies where again the Hovermap reported consistently higher spatial resolution. Goncalves and Henriques [26], for example, reported 0.032–0.45 m resolution, whilst Jaud et al. [34] evidenced a ‘high’ spatial resolution of 0.042 m·px−1. Similarly, resolution up to 0.047 m·px−1 were reported by both Papakonstantinou et al. [48] and Topouzelis et al. [61]. Of the studies reviewed, only the work of Guillot and Pouget [28] described a spatial resolution (0.015 m·px−1) greater than what was observed in this study.

With a mean RMSE of 0.050 ± 0.31 m, the absolute accuracy of the Hovermap point clouds to GCP targets were again consistent with the 0.02–0.8 m range reported in previous manned airborne LiDAR [6, 39, 43] and UAV SfM (0.17–0.3 m) studies [17, 34, 44, 61]. According to Le Mauff et al. [40], these relatively small differences may be attributed to noise inherent to LiDAR data.

4.2 Foredune Scarping and Blowout Area

The Hovermap performance enabled the mesh models utilized here to be created with polygon size of 0.02 m and fine-scale evaluations of foredune elevation and slope as well as change detection between survey events in the deviation and cubature analysis. Utilizing the methodology in point cloud processing and mesh creation described, both nominal and substantial temporal changes in foredune were observed throughout the study with the exception of Survey4-HM data where substantial changes in the foredune were evidenced. According to Brown and McLachlan [11], physical features of beaches reflect the interaction of wave height, wavelength, and direction of the tidal regime and sediment transport available impacting sandy beaches either directly through changes in the structure of vegetation or substratum. Here, we estimated that a net loss of approximately 1295 m3 of sand was removed from the Analyis Area during Interval3 and observed corresponding increases in the maximum elevation, slope, and deviation from previous surveys. This measure of sand loss and deviation are supported by the marine weather data recorded during interval3 and correlations between storms, wind direction, wave heights, and erosional processes of numerous previous studies [14, 16, 19, 23, 37]. Similarly, in accordance with Arens [8], the increase in maximum elevation and mean slope may be contributed to scarping of the foredune and corresponding loss of vegetation where sand is transported further up the stoss faces, and this can increase as foredune height and/or steepness increases. Continued monitoring was beyond the scope of this study; however, subsequent foredune recovery will likely depend on the degree of revegetation and timeframes associated with reestablishment [52].

As described by Hesp [30], the existing blowout and overwash fan within the Study Area were likely formed by wave erosion along the seaward face of the dune with overwash hollows and fans developing into blowouts if the vegetation cover is slow in reestablishing. Interestingly, the blowout studied here demonstrated relatively minor changes in elevation and slope through most of the study with the exception of Survey4-HM where scarping of the foredune and lowering of the beach was evident with increased maximum elevations and slopes observed and consistent with trough-blowout topography that can significantly accelerate wind speeds along the deflation floor and lateral erosion walls resulting in steep stoss faces and general elongation of the blowout [31, 32]. Interestingly, the depth of the blowout remained relatively unchanged (min. ~ 2.2 m) throughout the course of the study and suggested this elevation represents the deflation limit of this blowout [13]. Chapman [15] described a ‘coffee rock’ layer underlying unconsolidated sands at Bribie Island that may be associated with the deflation limit observed. Evidence for this is presented in Fig. 16, where the dark sands along the lower right of the image of the Analysis Area indicate the presence of the coffee rock layer near the beach surface.

Fig. 16
figure 16

Photomosaic image of the Analysis Area illustrating dark sands (lower right) indicating the presence of ‘coffee rock’ near the beach surface

Full size image

4.3 Limitations and Future Research

Repeatability is a function of the stability of calibration of the instrument, accuracy of position estimation (either via GNSS or SLAM), density and completeness of point cloud coverage, and the availability and accuracy of ‘ground-truth’ information [29]. In this study, the Velodyne sensor and Hovermap system were both calibrated by the manufacturers and stable over the course of the study. At over 2000 pts.·m-2, point cloud density was consistently high and the ‘cross-flight’ pattern utilized ensured complete coverage of the Study Area during each survey. Lague et al. [38] reported that the primary sources of uncertainty in point cloud comparison are within the registration uncertainty between point clouds. Here, referencing individual point clouds to the GCPs introduced error, however, observed RMSE values were similar to values reported in airborne LiDAR from manned aircraft. Additional potential sources of error include the accuracy associated with the data collection process (i.e. robustness of the SLAM solution) and that associated with the topographic interpolation process action [37]. This is supported by Grohmann and Sawakuchi [27] who discussed the influence of DTM cell size on volumetric calculations and observed a directly proportional increase in RMSE and cell size.

The approach described here is likely to be particularly useful at microspatial and temporal scales and, similar to the work of Simpson et al. [56], may be expanded to include additional attributes of the coastal dune environment such as detecting changes in vegetation structure. Specifically, elaboration on the work of Lalimi et al. [39] though the application of UAV LiDAR could assess if derived leaf area indices improve from point cloud data that is several orders of more dense magnitude. It would also be interesting to regularly deploy UAV LiDAR over longer periods to evaluate the utility of high-spatial resolution data in the correlation of marine weather and coastal dune formation, blowout formation, and subsequent recovery processes. Such predictive modelling would also likely benefit on having high-frequency, multi-temporal data microscales to inform morphodynamic and beach dune erosion models.

5 Conclusions

Although, additional work is required to better understand point cloud response under different site conditions, the results demonstrated here confirm that UAV LiDAR is a robust tool and has great promise as a new tool among the various methods currently available to coastal scientists and engineers. The high spatial resolution and flexibility of deployment are key attributes to UAV remote sensing and will likely better inform future theories, models, and predictions, particularly on microtemporal and spatial scales. We believe the methods for evaluation described here could be applied to other UAV LiDAR systems and compared with more traditional techniques to assist future operators in better understanding system performance, and limitations, and thereby inform cost/benefit decisions when selecting instrumentation for future coastal geomorphology studies.