Assessment of UAV and Ground-Based Structure from Motion with Multi-View Stereo Photogrammetry in a Gullied Savanna Catchment

: Structure from Motion with Multi-View Stereo photogrammetry (SfM-MVS) is increasingly used in geoscience investigations, but has not been thoroughly tested in gullied savanna systems. The aim of this study was to test the accuracy of topographic models derived from aerial (via Unmanned Aerial Vehicle, ‘UAV’) and ground-based (via handheld digital camera, ‘ground’) SfM-MVS in modelling hillslope gully systems in a dry-tropical savanna, and to assess the strengths and limitations of the approach at a hillslope scale and an individual gully scale. UAV surveys covered three separate hillslope gully systems (with areas of 0.412–0.715 km 2 ), while ground surveys assessed individual gullies within the broader systems (with areas of 350–750 m 2 ). SfM-MVS topographic models, including Digital Surface Models (DSM) and dense point clouds, were compared against RTK-GPS point data and a pre-existing airborne LiDAR Digital Elevation Model (DEM). Results indicate that UAV SfM-MVS can deliver topographic models with a resolution and accuracy suitable to deﬁne gully systems at a hillslope scale (e.g., approximately 0.1 m resolution with 0.4–1.2 m elevation error), while ground-based SfM-MVS is more capable of quantifying gully morphology (e.g., approximately 0.01 m resolution with 0.04–0.1 m elevation error). Despite difﬁculties in reconstructing vegetated surfaces, uncertainty as to optimal survey and processing designs, and high computational demands, this study has demonstrated great potential for SfM-MVS to be used as a cost-effective tool to aid in the mapping, modelling and management of hillslope gully systems at different scales, in savanna landscapes and elsewhere. This study tests the accuracy of topographic models derived from aerial (via UAV) and ground-based (via handheld digital camera) SfM-MVS in modelling hillslope gully systems located in a dry-tropical savanna catchment tributary to the GBR, and assesses the strengths and limitations of the approach at a hillslope scale and an individual gully scale.


Introduction
Gully erosion is a globally significant land degradation process that has detrimental environmental, ecological and economic impacts across a range of landscape systems [1][2][3][4][5][6][7]. Effective management of gully erosion requires high-resolution topographic information that can be used to accurately quantify the spatial distribution and density of gully systems, gully morphology and morphologic change, and the catchment characteristics that influence gully evolution [2,[8][9][10]. Recent advances in remote sensing techniques have greatly improved our ability to collect high-resolution topographic data at a range of This study tests the accuracy of topographic models derived from aerial (via UAV) and ground-based (via handheld digital camera) SfM-MVS in modelling hillslope gully systems located in a dry-tropical savanna catchment tributary to the GBR, and assesses the strengths and limitations of the approach at a hillslope scale and an individual gully scale.

Study Site
The study sites are located in the Weany Creek sub-catchment (13.6 km 2 ) within the Upper Burdekin Catchment (~36,000 km 2 ) and part of the Burdekin River Basin (130,000 km 2 ), which drains to the GBR in northeast Australia ( Figure 1). The area has granodiorite lithology [82,83] and is dominated by chromosol soil, a sandy clay loam, known locally as red goldfields soil. Chromosol soils, which cover approximately 12% of the Burdekin River basin, contain numerous gullies and are a priority for management intervention [84]. The terrain is dissected by many drainage depressions, although relief and slope gradients are low (median gradient is 2.3% [85]). The climate is dry-tropical with two distinct seasons; a summer wet season between October and March, and a dry season between April and September [86]. The mean-annual rainfall (1900-2012) is 686 mm year −1 , but exhibits high interannual variability linked to the phase of the Southern Oscillation, with drought conditions coinciding with El Niño/Southern Oscillation (ENSO), and wetter periods accompanying anti-ENSO (La Niña) events [87,88]. Vegetation is generally characterised by a discontinuous upper stratum of Eucalyptus spp. and a more continuous understory of annual and perennial grasses, forbs, and small shrubs [89] (Figure 2). Vegetation grows rapidly during the wet season, followed by a period of desiccation in the dry season [90][91][92]. Cattle grazing commenced in the area after 1850 and remains the dominant land use today.
Within the Weany Creek sub-catchment, measurements were focused in three separate hillslope gully systems (known as 'G1', 'G3', and 'G6'). These systems contain a diversity of gully features and ground cover conditions, considered representative of the broader sub-catchment. UAV surveys covered each hillslope system, while ground-based ('ground') surveys covered a single gully within each system (known as 'G1 gully', 'G3 gully' and 'G6 gully'). The individual gully sites were selected because they had been previously studied [57,93] and contain contrasting morphological features. The survey areas of hillslopes G1, G3, and G6 are 0.715, 0.412 and 0.451 km 2 , respectively. The survey areas of gullies G1, G3, and G6 are 650, 350, and 750 m 2 , respectively. G1 gully is a V-shaped gully with steep sidewalls (45-90 • ), a narrow gully floor (0.2-1 m wide) and two distinct headcuts (Figure 3a). The headcuts are slightly overhanging in the upper 0.5-1 m of the incision profile, with total headcut depths ranging from 1-1.5 m. G3 gully is a U-shaped gully, with very steep sidewalls (~70-90 • ) ranging in height from 1 to 2.5 m, a slightly wider gully floor (1-1.5 m) and a single distinct headcut (0.5 m) that sits below a small (1 × 0.5 m deep) incised drainage depression (Figure 3b). G6 gully is more variable in shape, with sidewalls ranging from steeply dipping (45-90 • ) to more gently sloping (<45 • ), and gully bed widths ranging from 0.2-1 m. The main headcut of G6 is~1 m deep and sits approximately 5 m below two smaller headcuts (~0.5 m deep) (Figure 3c). Evidence of mass wasting at the headcuts is clearly visible in all three gullies, with scouring around the base of the headwall, and sediment deposits on the gully floor. Rill fluting occurs along portions of gully sides. Weathered bedrock is exposed in sections of gully wall, typically in the lower 0.5 m of the profile. At the time of the ground survey (October 2016), grass (predominantly Indian Couch (Bothriochloa pertusa) and woody legumes (Stylosanthes spp.); height of~30-50 cm) occupied much of the area along the gully edges and in some sections of the more gently sloping gully wall, obscuring the bare soil. Leaves, bark and woody debris also partially covered the ground. Grass was not present in the gully floor. Thicker pockets of low-lying shrubs (~1 m tall × 1 m wide) occupied some sections of gully edge and walls.
conditions coinciding with El Niño/Southern Oscillation (ENSO), and wetter periods accompanying anti-ENSO (La Niña) events [87,88]. Vegetation is generally characterised by a discontinuous upper stratum of Eucalyptus spp. and a more continuous understory of annual and perennial grasses, forbs, and small shrubs [89] (Figure 2). Vegetation grows rapidly during the wet season, followed by a period of desiccation in the dry season [90][91][92]. Cattle grazing commenced in the area after 1850 and remains the dominant land use today.  Within the Weany Creek sub-catchment, measurements were focused in three separate hillslope gully systems (known as 'G1', 'G3', and 'G6'). These systems contain a diversity of gully features and ground cover conditions, considered representative of the broader sub-catchment. UAV surveys covered each hillslope system, while ground-based ('ground') surveys covered a single gully within each system (known as 'G1 gully', 'G3 gully' and 'G6 gully'). The individual gully sites were selected

LiDAR
The airborne LiDAR data, covering the area around and including the hillslope site, was acquired in 2013 as part of a larger airborne LiDAR survey campaign by the Queensland Department of Science, Information Technology, Innovation and the Arts [76]. The sensors were configured to sample the gully environment with an average pulse density of 4.2 pulses per m 2 (each pulse recorded up to five returns) and an overlap of 50% between flight runs to minimise the impact of occlusion from variable terrain and vegetation. This resulted in an average of 8 pulses per m 2 . DEM surfaces were interpolated using the natural neighbour algorithm [94] at a spatial resolution of 0.5 m using the LiDAR returns classified as ground. More detail on LiDAR capture specifications and DEM generation are given in Tindall et al. [76]. The amount of erosion across the hillslopes sites in the three years between the airborne LiDAR survey and UAV survey was minimal, due to a limited number of erosive rainfall events over this period.

UAV and Ground Survey
In August 2016, a UAV survey of the hillslope sites was conducted using a DJI Phantom 3 Professional Drone with photographs taken with a nadir (vertical) viewing direction [95] ( Table 1). The flying mission was planned using the Map Pilot App by Maps Made Easy [96]. GPS measurements recorded by the UAV were not used in the processing of the data. Ground Control Points (GCPs) were positioned approximately in the corners and centre of each hillslope gully system catchment area, in a layout considered sufficient to adequately constrain the SfM-MVS model (Figures 4a and 6). The GCPs consisted of a 1.2 × 1.2 m black and white "iron cross" vinyl marker. The GCPs and a number of random validation points (used to evaluate the accuracy of the SfM-MVS topographic models), were surveyed using a CHC X91 Real Time Kinematic GPS (RTK GPS) receiver. The RTK GPS had a mean horizontal error of 0.014 m (standard deviation, SD of 0.004 m), and a mean vertical error of 0.030 m (SD of 0.010 m). Further details about the UAV survey are outlined in Table 2.
In October 2016, the ground survey of the gully sites was conducted using a Panasonic GH3 digital single-lens reflex (DSLR) camera (Table 1). No geocoding information was collected during ground photo acquisition. GCPs were distributed on the ground surface throughout the gully system along the gully edge, wall and floor ( Figure 4b). The GCPs consisted of a laminated A4 page with four coded circular targets, produced using the 'Print Markers' tool within AgiSoft PhotoScan Professional V1.3 (PhotoScan) [97]. The GCPs and a number of random validation points were surveyed using an Ashtech Magellan Promark 500 RTK GPS. The RTK GPS had a mean horizontal error of 0.018 m (SD of 0.003 m) and a mean vertical error of 0.026 m (SD of 0.004 m). Further details about the ground survey are outlined in Table 2.
Initial attempts at the ground survey, involving significantly fewer photos and a non-systematic approach to image capture (e.g., random path of photographer, images captured non-consecutively) resulted in poor surface reconstruction. Here we outline the method of image capture found to produce the best surface reconstruction (Figure 4c,d). The photographer started at the gully head, approximately 3 m from the gully edge, and worked their way along the gully perimeter, capturing one image approximately every meter, focusing obliquely on the opposite bank toward the downstream end. Upon reaching the gully outlet, the photographer followed the same path back toward the gully head, again focusing obliquely on the opposite bank, but this time toward the upstream end. Upon reaching the gully head (i.e., the starting point), the photographer moved to the gully edge and repeated the process. Once images were captured on both sides of the gully boundary, the photographer then captured images through the centre of the gully, walking along the gully bed. Here the photographer started at the gully head and captured a series of images focused directly downstream. Upon reaching the gully outlet, the photographer followed the same path back toward the gully head, this time capturing images focused directly upstream. The photographer then repeated this process, but changed the point of focus to be obliquely toward the left and right bank, respectively, collecting images in both the upstream and downstream direction. Following guidance provided by Smith et al. [13], the photographer tried to achieve full 360 • coverage of gully features with a high degree of overlap between images; avoided large angular changes of >25-30 • between adjacent camera locations and large jumps in scale; and minimised the interval between images to reduce the effect of changes in lighting and shadow conditions. provided by Smith et al. [13], the photographer tried to achieve full 360° coverage of gully features with a high degree of overlap between images; avoided large angular changes of >25-30° between adjacent camera locations and large jumps in scale; and minimised the interval between images to reduce the effect of changes in lighting and shadow conditions.

SfM-MVS Workflow, 3D Model, and Ortho-Photo Mosaic Generation
Agisoft PhotoScan software was used to implement the SfM-MVS workflow. Details of the processing parameters and processing times are provided in Supplementary Table S1. The SfM-MVS workflow, described in detail elsewhere [13,19], can be summarised in three main steps. The first step involves: (i) the identification and matching of features in a set of images, utilising an algorithm based on the scale-invariant feature transform (SIFT) object recognition system [98]; and (ii) the implementation of bundle adjustment algorithms to estimate the 3D geometry (or structure) of a scene, as well as the internal and external camera orientation parameters [97], producing a sparse,   Supplementary Table S1. The SfM-MVS workflow, described in detail elsewhere [13,19], can be summarised in three main steps. The first step involves: (i) the identification and matching of features in a set of images, utilising an algorithm based on the scale-invariant feature transform (SIFT) object recognition system [98]; and (ii) the implementation of bundle adjustment algorithms to estimate the 3D geometry (or structure) of a scene, as well as the internal and external camera orientation parameters [97], producing a sparse, unscaled 3D point cloud in arbitrary units. The second step involves: (i) a linear similarity transformation to scale and georeference the point cloud, utilising reference coordinates of GCPs distributed throughout the study site; and (ii) point cloud optimisation, a process in which camera parameters and 3D points are adjusted to minimise the sum of the reprojection error and the georeferencing error [97]. In this study the precise location of the GCPs were automatically identified by PhotoScan and manually refined where appropriate. The final step involves the implementation of multi-view stereo (MVS) image matching algorithms to build a dense 3D point cloud. An additional step can be added to the SfM-MVS workflow to generate textured 3D models and ortho-photo mosaics derived from the dense point clouds. In this study, PhotoScan processing ran on a computer that carried an 8 GB NVIDIA Quadro K5200 graphics card, two Intel Xeon CPU E5 K5200 v2 @2.5 GHZ processors, and 128 GB RAM.

Digital Surface Model (DSM) Generation
DSMs, derived from the dense point clouds, were generated using the 'rasterize' tool in CloudCompare (v2.8) [99]. The minimum elevation point at each pixel was used to model the terrain surface, as it has the greatest chance to represent the surface within vegetated areas [26]. The UAV DSMs were generated at 0.1 m pixel size, whereas the ground DSMs were generated at 0.01 m pixel size.

Conversion to the Australian Height Datum
The airborne LiDAR DEM elevation values were referenced to the Australian Height Datum (AHD), whereas the RTK validation points and UAV and ground DSMs had elevation values referenced to the ellipsoid GDA94. To allow comparison among the datasets, the elevation values of the DSMs and RTK validation points were converted to AHD, using the AusGeoid09 transformation [100], prior to accuracy assessment.

Elevation Accuracy of SfM-MVS Topographic Models
The elevation accuracy of the SfM-MVS topographic models was evaluated using point-to-raster, point-to-point, and raster-to-raster comparisons. In the point-to-raster comparison, elevation values were extracted from the DSM and compared to concordant RTK validation points. In the point-to-point comparison, elevation values were extracted from the dense point cloud (prior to rasterization) and compared to concordant RTK validation points, using the 'compute cloud/cloud distance' tool in CloudCompare. Point-to-point comparison is useful in topographically complex environments [101] such as gullies, where steep sides and overhangs are common [46], as a given set of x and y coordinates can have multiple z values [42,102]. In the raster-to-raster comparison, the UAV DSM was subtracted from the airborne LiDAR DEM, allowing examination of the spatial distribution of error. Here we assume the airborne LiDAR DEM is more accurate than the UAV DSM.
Elevation accuracy is evaluated for: (i) all validation points (excluding GCPs) within the hillslope sites; (ii) all validation points (excluding GCPs) within the gully sites; and (iii) individual cross-sections across the gully site. Accuracy metrics include: root mean square error (RMSE), mean absolute error (MAE), and mean error (ME) [102].

Georeferencing Error of Ground Control Points
Georeferencing error of the GCPs is low for all gully systems and similar across survey platforms (Table 3). Total georeferencing error ranges from 0.010 to 0.027 m (0.140 to 3.351 pixels), with most of the error associated with the Z (elevation) coordinate. The high total pixel error reported for G3 ground is due to one GCP with a much higher pixel error (7.134 pixels) compared to the other GCPs (0.095 to 0.521 pixel). This GCP was not removed prior to processing in order to maintain a good distribution of GCPs over the survey area, and its inclusion did not reduce model accuracy. The low georeferencing error across all survey platforms and study sites suggests georeferencing error is not a major source of error in the SfM-MVS topographic models. Table 3. Georeferencing error of GCPs for the UAV and ground survey. Error (m) is calculated as the root mean square error between the actual position of the GCPs, derived from RTK measurement, and the estimated position of the GCPs, derived from the SfM-MVS model. Error (pixel) is calculated as the root mean square reprojection error for the GCPs calculated over all photos where the GCPs are visible.

DSM Elevation Error
At the hillslope scale, the UAV DSM elevation error is moderate for G3 (MAE = 0.406 m), and relatively large for G6 and G1 (MAE = 0.851 and 1.220 m, respectively, Table 4). The distribution of elevation error is negatively skewed for G1 ( Figure 5b) and G3 (Figure 5d), with >84% of the error between 0.5 and 3 m (i.e., the DSM is lower than the RTK validation points). In contrast, 80% of the validation points in G6 (Figure 5f) have an error of between −0.5 and −3 m (i.e., the DSM is higher than the RTK points). Comparison of the UAV dense point cloud to the RTK validation points reduced the MAE reported at the hillslope scale by 0.35, 0.16, and 0.24 m for G1, G3, G6, respectively (Table 5), suggesting some error may have been introduced during rasterization of the point cloud.   Vegetation is estimated to cover~80-90% of the survey area and includes irregularly distributed grass, shrubs and trees. Visual inspection of the spatial variability of error across the ortho-photo mosaics (Figure 6b,d,f) does not reveal any distinct patterns related to vegetation effects on error; high errors are observed in both bare and vegetated patches. However, examination of the error model between the airborne LiDAR DEM and UAV DSM (particularly Figure 6c,e) clearly identify higher errors in areas corresponding to the position of dense vegetation patches (such as tree canopies and bushes). The high degree of vegetation cover at the time of survey is therefore likely to have introduced error into the UAV DSMs during image processing. mosaics (Figure 6b,d,f) does not reveal any distinct patterns related to vegetation effects on error; high errors are observed in both bare and vegetated patches. However, examination of the error model between the airborne LiDAR DEM and UAV DSM (particularly Figure 6c,e) clearly identify higher errors in areas corresponding to the position of dense vegetation patches (such as tree canopies and bushes). The high degree of vegetation cover at the time of survey is therefore likely to have introduced error into the UAV DSMs during image processing.  Table 4) than the UAV DSM. The distribution of elevation error is more normally distributed for the ground DSMs with 95%, 94%, and 97% of validation points having an error of between −0.1 m and 0.1 m for G1, G3, and G6, respectively (Figure 5a,c,e). Similar to the UAV dense point cloud, comparison of the ground dense point cloud  At the individual gully scale, the ground DSMs have much lower elevation error (MAE = 0.052 m, 0.046 m, and 0.038 m, for G1, G3 and G6, respectively, Table 4) than the UAV DSM. The distribution of elevation error is more normally distributed for the ground DSMs with 95%, 94%, and 97% of validation points having an error of between −0.1 m and 0.1 m for G1, G3, and G6, respectively (Figure 5a,c,e). Similar to the UAV dense point cloud, comparison of the ground dense point cloud with the RTK validation dataset reduced the elevation error by 0.021, 0.024 and 0.012 m for G1, G3, and G6, respectively (Table 5), suggesting the dense point cloud better represents the complex 3D gully geometry than the DSM. No discernible pattern in elevation error related to landscape position or vegetation is apparent; errors are similar across the gully walls and gully floor, and in bare versus vegetated patches (Figure 7a,c,e). The ground DSM cross-sectional profiles closely follow the shape of the RTK profiles, providing a better representation of gully shape than the UAV DSM (Figure 7b,d,f).

Comparison of Time and Resource Requirements
The UAV surveys of the hillslope took between 3 and 4 h to complete (generally <1 h for image capture), while the ground surveys of individual gullies took between 1 to 2 h (Table 6). Processing time for ground-based images was significantly longer than for UAV images owing to the greater number of images captured and the complexity of the structures being modelled. Total time required to produce the DSMs presented in this study varied from approximately 18 to 32 h for the UAV DSMs and approximately 31 to 120 h for the ground DSMs.

Discussion
Here we evaluate the strengths and limitations of UAV and ground SfM-MVS in modelling hillslope gully systems in tropical savannas. The strengths, limitations and opportunities are summarised in Table 7.

Application
Cost-effective high-resolution gully and ground cover mapping at a hillslope scale (e.g., 1-100 ha).
Cost-effective very high resolution 3D modelling of gully morphology at an individual gully scale (e.g., 0.01-0.2 ha).

Strengths
• Provides high-resolution DSMs at a hillslope scale, with elevation error comparable to, although larger than, airborne LiDAR.

•
Provides high-resolution ortho-photo mosaics that can be used to assess ground cover (e.g., ground cover spatial arrangement, cattle trails, roads) and is valuable for land management communication.

•
Instrument costs are very low compared to airborne LiDAR.
• Provides very high resolution DSMs at an individual gully scale, with elevation error comparable to TLS. • Provides very high resolution 3D models that can be used to assess gully morphology and describe key erosion features. • Instrument costs are very low compared to TLS.

•
The complex structure, pattern and movement of vegetation in tropical savanna negatively affects model accuracy.

•
The accuracy of the final topographic models strongly depends on the methodological approach implemented. There remains much uncertainty as to optimal survey and processing designs for different landscapes and geomorphological applications.

•
The high number of images captured, mean that data storage, processing and analysis is computationally demanding.

Opportunities
Further research is needed to: • Determine SfM-MVS survey repeatability and its ability to detect geomorphic change.

Resolution and Accuracy
Results indicate UAV SfM-MVS can deliver topographic models with a resolution and accuracy suitable to define gully systems at a hillslope scale (e.g.,~0.1 m resolution with~0.4-1.2 m elevation error), while ground-based SfM-MVS is more capable of quantifying gully morphology (e.g.,~0.01 m resolution with~0.04-0.1 m elevation error). These results agree with other studies that have used aerial platforms to capture images across large areas in complex terrain (RMSE values ranging from 0.05 m to~1 m [18,36,47,[101][102][103]), and ground-based approaches to model gully systems (RMSE values ranging from 0.025 m to 0.155 m [40][41][42]46,47,49,104,105]). While the overall elevation errors of the UAV topographic models were large relative to a pre-existing airborne LiDAR dataset and RTK validation points, cross-sections extracted from the UAV DSM broadly match the shape of gully profiles, although are vertically offset. The elevation errors of the ground topographic models were substantially lower than those derived from UAV and are similar to those reported by recent TLS surveys of gully systems in the region [errors of~0.05 m, 15,106]. While further work is needed to ascertain survey repeatability and the ability to detect geomorphic change [15,50,78,106], our results suggest there is great potential for both UAV and ground SfM-MVS to deliver high-resolution topographic information in gullied savanna catchments, at both the hillslope and individual gully scale.

Low Survey Instrument Costs and Survey Time
A major strength of SfM-MVS is the low survey instrument costs. In our study, we utilised a lightweight UAV (<2 kg) and camera (total cost including accessories was~$AUD 2500) to capture aerial imagery at the hillslope scale, and a DSLR camera (total cost including accessories was~$AUD 2500) to capture ground-based imagery at the individual gully scale. The SfM-MVS instrument costs are considerably lower than airborne LiDAR and TLS, both of which cost >$AUD 100,000 to purchase outright or, if data are provided by an external consultant, can cost between $AUD 1000 and 2000 per km 2 [106]. The low SfM-MVS survey instrument costs mean they can be purchased outright and deployed rapidly. In Australia, lightweight drones (<2 kg) can be used without a remote pilot license [107], representing a substantial cost saving of several thousand dollars. In our study, the UAV and ground surveys took approximately four and two hours of survey time, respectively, comparable to the survey times of equivalent LiDAR approaches [106]. Information generated from SfM-MVS survey therefore has the potential to enable cost-effective prioritisation of gully erosion hotspots at a hillslope scale (via aerial surveys), and to be used as a tool in the design, monitoring and evaluation of gully remediation activities at individual gully scales (via ground-based survey).

High-Resolution Ortho-Photo Mosaics and 3D Models
High-resolution ortho-photo mosaics and 3D models produced as part of the SfM-MVS workflow provide a wealth of additional information that can be used to improve understanding of hillslope processes and aid in gully erosion management. Visual interpretation of the 3D model of the gully site, for example, allows detailed description of gully morphology and erosion features in a more systematic way than can be achieved in the field (Figure 3). Similarly, the ortho-photo mosaics provide valuable insights into landscape factors that may influence gully development (e.g., ground cover type and spatial arrangement, cattle trails, roads). Ortho-photo mosaics and 3D models also provide excellent resources to visually communicate the benefits of improved land management to a range of stakeholders.

Vegetation
Vegetation presents a significant challenge for the generation of accurate topographic models for use in environmental modelling [108]. Vegetation is particularly problematic for SfM-MVS in tropical savannas, which are characterised by a grassy understory of variable density and structure and a discontinuous layer of trees and shrubs. Vegetation obstructs the view of the ground, is prone to movement (key-point matching algorithms rely on a static scene), and has a complex structure that is difficult to reconstruct [13,101,102]. Examination of the error model between the airborne LiDAR DEM and UAV DSM clearly identify higher errors in areas corresponding to the position of dense vegetation patches (such as tree canopies and bushes). It is likely that the high ground cover at the time of survey reduced the accuracy of the topographic models as highlighted in other studies [18,26,101,103]. For example, Cook [101] found that in sparsely vegetated areas, the SfM-MVS point cloud closely matched the ground returns of LiDAR, but struggled to accurately represent ground surface in areas of denser grass, bush, and trees. Similarly, Javernick [26] reported much higher RMSE values for vegetated surfaces (0.78 m) compared to bare areas (0.17 m). We tested several methods of vegetation classification and removal (e.g., using PhotoScan's 'classification' procedure [97,109]), resampling the point cloud at larger grid sizes where vegetation clusters are observed, extracting the minimum value within the wider area [26], and point cloud classification based on multi-scale dimensionality criteria [110]). None of these methods adequately classified vegetation across our study site. Instead, the minimum elevation value point within each pixel was used to provide the best approximation of the terrain surface within vegetated areas, acknowledging that this can sometimes result in the extraction of false elevation values as a result of point scatter below the ground surface [29]. Vegetation filtering from the SfM-MVS datasets was not the focus of this study, but is an active area of research, e.g., [111][112][113]. If suitable vegetation filtering methods can be developed and easily applied to SfM-MVS datasets, it will greatly enhance the ability of SfM-MVS to accurately reconstruct ground topography in densely vegetated areas.

Methodological Uncertainty
While implementation of the SfM-MVS workflow is relatively straightforward, the accuracy of the final topographic models depends on the methodological approach used. General guidelines for SfM-MVS surveys have been described, e.g., [18], but there remains much uncertainty as to optimal survey and processing designs for different landscapes and geomorphological conditions and applications. SfM-MVS practitioners must therefore test and adapt their methodological approach to suit their particular landscape and survey goals prior to implementing broad-scale surveys. Here we highlight two key methodological considerations, namely: the number and distribution of GCPs and the method of image capture.
The number and distribution of GCPs has a strong influence on survey time and the quality of surface reconstruction [13,19,44,102,114]. In our UAV datasets, despite very low georeferencing error of GCPs, much greater elevation errors were observed in the final topographic models. It is likely that the number and distribution of GCPs was insufficient to properly parameterize the 3D model during bundle adjustment (explained in [18]). While a minimum of three GCPs are required to account for model rotation, translation, and scale during the implementation of bundle adjustment algorithms, research suggests that collection of more GCPs is better and these should adequately cover the area of interest [17,44,115,116]. Specific GCP requirements will, however, vary depending on site characteristics, quality of the image network used in surface reconstruction, and required accuracy of the final topographic model [19,44]. Given the effect of GCPs on surface reconstruction quality and the considerable field effort required to collect them, determining the optimal number and distribution of GCPs remains an important research priority.
The method of image capture also has a strong bearing on survey and processing time and the quality of surface reconstruction. During our UAV survey for example, images were captured at a height of~85-100 m above the ground, with a single nadir (vertical) viewing direction, and a single orthographic grid flight path. This flight strategy was selected as it provided a good compromise between the extent of the survey area covered and survey time, but may have contributed to the error observed in the UAV topographic models. Recent studies have reported a decrease in model accuracy with increasing survey distance [18,19] and also demonstrated that using vertical imagery alone can introduce systematic broad-scale error into topographic models, expressed as vertical 'doming' [44,115,117]. The addition of convergent (non-nadir) imagery into aerial surveys has recently been shown to improve model accuracy [18,115]. Similarly, the addition of a second set of perpendicular flight lines (e.g., in a 'double-grid' pattern), is increasingly utilised in UAV topographic surveys to provide better image overlap and coverage of the area of interest and thus improve model performance [118]. Further testing to determine the optimum number, overlap, height and angle of image of acquisition, appropriate for both hillslope and individual gully surveys, would enable faster and more efficient survey and processing time and help to improve model accuracy. It is also important to consider the camera sensor used to capture images. Most light-weight consumer UAVs, including the DJI Phantom 3 Professional, come equipped with a rolling shutter camera, which is prone to image distortion when images are recorded from a moving platform [119]. This may have introduced further error into our UAV topographic models. Cameras equipped with global shutters can mitigate this problem and should be considered in future UAV surveys.

Computational Demands
The high number and resolution of images captured during SfM-MVS surveys means that data storage, processing, and analysis is computationally demanding. For example, data from the UAV and ground survey took approximately 11-24 h and 26-114 h, respectively, to process on a computer with 8 GB graphics card, two CPU cores, and 128 GB RAM. Such computational demands may limit the scale at which SfM-MVS is currently applied and to practitioners who have access to high-performance computers. However, rapid advances in computing capability, for example, through improvements to Graphics Processing Units and the implementation of parallel computing, are revolutionizing SfM-MVS workflows [19]. As high-performance computers become more readily available, broader-scale implementation of both aerial and ground-based SfM-MVS approaches can be expected.

Conclusions
Structure from Motion with Multi-View Stereo photogrammetry (SfM-MVS) is increasingly used as a cost-effective method of rapidly acquiring high-resolution topographic data across a range of scales and in diverse landscapes and geomorphic settings, but has not been thoroughly tested in savanna systems. Results from our study, conducted in a gullied dry-tropical savanna catchment in Upper Burdekin catchment, indicate UAV SfM-MVS can deliver topographic models with a resolution and accuracy suitable to define gully systems at a hillslope scale (e.g.,~0.1 m resolution with 0.4-1.2 m elevation error), while ground-based SfM-MVS is more capable of quantifying individual gully morphology (e.g.,~0.01 m resolution with~0.04-0.1 m elevation error). High-resolution 3D topographic models and ortho-photo mosaics produced as part of the SfM-MVS workflow allow detailed description of gully morphology and contributing area characteristics (e.g., ground cover type and spatial arrangement), providing valuable information for land management and restoration practices. Modest survey instrument costs (<$AUD 3000) and rapid survey times (~4 and 2 h for UAV and ground survey respectively) mean that SfM-MVS is highly cost-effective compared to equivalent Light Detection and Ranging (LiDAR) techniques. While SfM-MVS offers great potential, this study has also identified some important limitations of the methodology. For example, accurately reconstructing ground topography in densely vegetated areas, common across tropical savanna landscapes, remains a major challenge for SfM-MVS workflows and requires the development of improved vegetation filtering methods. There remains uncertainty as to optimal survey and processing designs to achieve the best model accuracy and reduce processing times. Finally, high computational demands for image processing may limit the spatial scale at which SfM-MVS is currently implemented and the number of practitioners who can apply it. However, as high-performance computing becomes more readily available, broader-scale application of SfM-MVS is expected. Overall, this study has demonstrated great potential for SfM-MVS to be used as a cost-effective tool to aid in the mapping, modelling, and management of complex hillslope gully systems at different scales in tropical savanna landscapes and elsewhere.
Supplementary Materials: The following are available online at www.mdpi.com/2220-9964/6/11/328/s1. Table S1: PhotoScan processing parameters and approximate times for UAV and Ground survey datasets.  Figure 1. We gratefully acknowledge Rob and Sue Benetto of Virginia Park Station for allowing access onto their property to undertake this work, and the two anonymous reviewers who provided comments on early versions of this manuscript.