Analysis of Topographic Surveys with RPAS in Steep Coastal Dunes

Abstract

The current use of photogrammetric systems with RPAS (remotely piloted aircraft systems) for the acquisition of topographic data in every type of coastal area has the benefit of a low risk for the personnel involved, good precision, increased productivity, and lower costs. However, their accuracy has not yet been researched in areas with steep terrain. In this paper, we study dune areas with slopes between 20 and 30%. The objective of this work is to examine the influence of the inclination of both the flight and the camera and to compare the results with those obtained using constant-height flights. With the data collected during three field campaigns, a total of 18 cases were studied. Among the results obtained, in the case of the horizontal flights, a vertical error of 0.048 m was detected for a 0° inclination of the camera versus an error of 0.086 m for a 10° inclination, thus an increase in the camera inclination decreased the accuracy by 44%. Moreover, the flight inclination did not lead to a significant reduction in the error. Therefore, as a main conclusion, the planning of horizontal flights as well as the non-tilt of the camera are recommended.

1. Introduction

Coastal zones are areas with high tourist value where the proper management of coastal conservation is important. The Cadiz coast has high ecological and economic value; hence, numerous studies have been carried out on it. In the field of beach regeneration, publications [1,2,3] stand out.

A three-dimensional reconstruction of the study area is necessary for accurate coastal modeling [4]. Digital terrain models (DTMs) with high spatial resolution are used to illustrate coastal modeling. A multitemporal surface can be understood as a geomorphological condition [5]. In addition, various procedures and methodologies might be employed depending on the beach area to be mapped (dry zone, intertidal zone, or submerged zone) [6]. Direct topography approaches have been used to map the intertidal zone and the dry beach [7]. The total station, including an electronic transit theodolite integrated with electronic distance measurement (EDM), and an on-board computer were later employed in place of a tachymeter to gather data and perform triangulation calculations. However, GPS methods are currently used to complete this work.

The current use of photogrammetric systems with RPASs (remotely piloted aircraft systems) for the acquisition of topographic data in every type of coastal area has the benefit of low risk for the personnel involved in capture; in addition, such systems provide good accuracy, increased productivity, and lower costs. Numerous studies have explored the use of topographic techniques with RPASs and have supported their use [8,9,10,11,12,13]. This method enables the monitoring of more coastlines for a greater length of time.

Remotely piloted aircraft systems (RPAS) are a practical and affordable choice for monitoring spatial–temporal variations in an environment that is changing rapidly. Nevertheless, studies characterizing beach geomorphology in relation to potential reflectance in the visible spectrum are scarce [14].

For the creation of focused land management operations that preserve biodiversity and ecological services, the topographic mapping of a dune field is crucial. The findings show the benefits and drawbacks of each technology and emphasize the effectiveness of RPAS, in particular, for the topographic modeling of coastal dune complexes [15].

However, this technique has difficulties due to the high reflectance of sand and its visual homogeneity. In addition to this issue, which has been addressed in previous studies [16,17,18], there are also difficulties when a beach has steep areas such as high dunes or cliffs. Some authors have researched cliffs, e.g., Barlow et al. (2017) [19] or Eboigbe et al. (2022) [20]; nevertheless, these authors have not found any research about the errors in the use of drones on steep dune slopes.

Other research [21] outlined a method for collecting and analyzing high-resolution data for coastal regions using a VTOL unmanned aerial vehicle (UAV) fitted with a small commercial camera. The suggested methodology combined image processing methods for analysis with artificial vision algorithms for 3D representation. Structure-from-motion (SfM) algorithms were utilized in computer vision, while geographic object-based image analysis (GEOBIA) with fuzzy categorization was employed for image processing. In Spain, regulations prevent drones from flying above 120 m. Added to this limitation is a technical condition for photogrammetry: the maintenance of a constant flight height with respect to the ground surface during data collection. This condition is enforced so that neighboring frames have the same coating. In two contiguous frames, the coverage will vary depending on the difference between the levels of their photo centers. It is possible, therefore, that this will directly affect the precision obtained in areas with steep slopes.

The uses and benefits of this methodology in beach areas with minimal unevenness are known. The authors of [22] presented some recommendations for near-horizontal beaches with the following parameters: the number of ground control points (GCPs), the flight height, and the percentage of side and forward overlap. The work developed by the authors of [23] analyzed the precision obtained in topographic surveys that were conducted using RPAS techniques in an area with artificial dunes with steep slopes. This study made use of a scraped artificial dune that was constructed on the sandy beach of Porto Garibaldi (Comacchio, Italy) to act as a barrier against wintertime sea storms and to better understand how elevation data uncertainty and uniform thresholds for change detection (TCD) affect the interpretation of volume change estimates. In order to evaluate the case study morphodynamics under non-extreme sea and wind conditions, this application depended on products acquired from unmanned aerial vehicle (UAV) surveys and on the evaluation of the uncertainty associated with volume change estimations. Digital elevation models (DEMs) created with UAVs were compared with orthophotos and GNSS data, with the root mean squared error (RMSE) being less than 0.05 m. This study complements the results obtained in [21,22], seeking to define a process that improves the accuracy of the current methodology. Our study focuses on dune areas with slopes between 20% and 30%.

Although opportunities for improvement have been identified [18,22,23,24], thus far, no solutions have been found. With this study, we intended to address this need and find an optimal methodology that would allow us to apply a UAV–SfM cartographic solution in dune areas with steep slopes. These areas are usually those that need the most surveillance and study due to their compromised state and great dynamism [25,26,27].

This study examined the analysis of flight plan trajectories to optimize battery use. Other researchers explored techniques for reducing a power plant’s battery consumption based on brushless motors, which are frequently employed in small unmanned aerial vehicles [28].

In another work [22], a strategy for photogrammetric flight planning for RPAS missions is described. The authors discovered many issues that arise in complex situations or challenges resulting from the project that need to be overcome in order to build practical functions or require instruments to address them. This strategy entails the enhancement of some standard photogrammetric flight operations and the recommendation of new flight plans for specific applications and scenarios. Combination flights, which combine elements of traditional block and corridor flights into a single mission, and a polygon extrusion mode for structures and vertical objects are two examples of these particular designs [29].

The objective of this work is to introduce inclined flights and to compare the results of this analysis with constant-height flights. The topographic data collection methodology involves the use of photogrammetric methods with unmanned aircraft (drones) in coastal areas with pronounced orography.

2. Study Area

The Valdevaqueros Dune, chosen for its precipitous nature (Figure 1), is an area with special protection that has a high value in terms of migratory processes [30,31].

Located on the southernmost tip of Spain (the Gibraltar Strait) at Valdevaqueros (Cadiz), the research area is a coastal dune. About 10 km separate the city of Tarifa from the Valdevaqueros Dune (Cadiz).

Strong easterly winds in the Strait of Gibraltar are correlated with high migration rates in the area of the Valdevaqueros Dune in southern Spain. The system is situated in an area with high pressure from human activities, and because of the way it interacts with land use, it is highly influential not only from a scientific and technical standpoint but also from a media and social perspective.

Sand from the submerged zone is brought into the system by westerly wind waves. This sediment builds up and causes the dry beach to become wider. The finer fraction on the dry beach is eroded by the easterly winds and carried to the dune.

The degree of urbanization is low, and the dune is composed of gold-colored sand with a D50 value of 0.34 mm. Thus, sediments are medium-grained sand, showing well-sorted materials, composed mostly of quartz (90–95%) and a carbonate (bioclastic) fraction (5–10%) [32]. The berm or dry beach of Valdevaqueros is the source of sediments that feeds and maintains the advance of the dune [33].

The study area (Figure 1b) is in a good state of environmental conservation [34] and covers an area of 48 Ha (800 m × 600 m). The dune stands among one of Spain’s windiest regions. The system is composed of sand accumulations that migrated from the shore as a result of strong easterly winds. The end result is a sand deposit with a NE–SW orientation, as seen in Figure 1b, with steep sand profiles. Data on the slopes can be found in [35]. From a scientific and sociological perspective, it is highly impactful [33]. Here, easterly winds occur frequently, creating a dune that is quite dynamic. Because of the fact that 70% of the winds have speeds exceeding the sediment threshold velocity, high sand migration rates are observed in the dune area. Due to the presence of a cove and a lengthy shoreline (more than 40 km) preceding the dune, the climate in this area is unusual [36].

3. Methodology

The visual homogeneity of coastal regions is the main challenge for photogrammetric survey methods. Because of this effect, there are fewer homologous points between neighboring frames which precludes the best possible correlation. The brief RPAS flight time is one challenge. The surface to be flown over, the flight height, and the overlap in the desired photographs must all be weighed against the standard flight time (20–25 min).

The different variables necessary for the correct planning of flights (the flight height, transversal and longitudinal overlaps, flight time, and number of GCPs/Ha) were defined according to Contreras-de-Villar et al. [22], and the flight parameters required to minimize the time and the topographic errors are shown in

Table 1. Flight parameters.

Flight Height (m)	100
Longitudinal overlap	70%
Transversal overlap	85%
GCPs/Ha	7
Flight time	08:00 a.m.

.

In order to complement previous studies [34,35,37], two new parameters were considered in this work:

• The inclined flight.
• Variation of the camera angle.

This study proposes conducting flights parallel to the surface of the dune and applying different angles to the camera used to take the shots. Constant-height flights were also planned, and camera inclination was applied. Figure 2 shows the flight plans and camera tilt angles. Subsequently, the differences in errors found in the different cases were analyzed [38]. A total of 18 cases, which are summarized in

Table 2. Case studies.

Field campaigns (20 June/20 October/21 April)	3
Flight height (Constant vs. inclined or sloped)	2
Camera angle (0°, 5°, or 10°)	3
Number of cases considered	18

, were considered.

3.1. Data Collection

The study area measured 300 by 86 m; therefore, its surface area was approximately 25,800 m². The flight area was extended by 30 m in the longitudinal direction of the flight and 10 m in the transversal direction, resulting in a total flight area of 30,000 m².

The data collection was carried out with a Phantom 4 Pro drone in three stages (see

Table 3. Aircraft specifications.

DJI Phantom 4 Pro
Take-off weight		1375 g
Max flight speed		20 m/s
Max flight time		30 min
Hovering accuracy	Horizontal	±0.5 m
	Vertical	±1.5 m

for aircraft specifications):

• A total of 16 GCPs were distributed on the beach to improve the accuracy and calibration of the camera. Despite the claims of some authors [39] that the use of GCPs may be reduced via direct georeferencing using highly accurate camera locations and GNSS receivers, landmarks were nevertheless employed as GCPs and georeferenced. The plastic GCPs had dimensions of 24 cm × 30 cm, a height of around 5 mm, and a geometry that helped us to establish their locations in the sand. Their backs were painted with a grid of alternating black and white squares (see Figure 3).

The positions of the GCPs were chosen in an attempt to obtain the optimal locations according to the criteria of [22]. There was enough shelter along and across the dune, and the site’s four corners—at the highest and lowest elevations—were utilized. Additionally, the sites were kept constant during the two flight campaigns so that positional changes would not impact the outcomes (see Figure 4).

2.

A topographic survey of the area was carried out using GPS topography in the RTK (real-time kinematic) mode, which provided a vertical precision of around 3 cm. To increase this accuracy, each topographic reading was taken three times in quick succession. When differences between the three readings were less than 1 cm in planimetry and 2 cm in altimetry, they were examined and validated. Based on his or her training and experience, the person holding the GPS surveying stick was also able to identify every location where there were obvious variations in the beach’s topography. Therefore, the density of GPS points in these areas was increased. A total of 120 survey points were obtained. This density of points is common on this type of terrain.

The objective was twofold: first, to complete a topographical survey of a dune surface for comparison with photogrammetric data, and second, to find the GCPs’ coordinates so that the RPAS could be post-processed. The leveling was calculated with reference to the Spanish Datum using the EGM2008 geoid provided by the National Geographic Institute, and the points were specified using the European UTM ETRS89 coordinates [40].

3.

Planning for the flight mission was undertaken previously. The Mapilot program (Automotive Data Research, 2021) was employed for this purpose. In this method, the area to be flown is defined, the flight height and the lateral and forward coverage are established, and the flight course is determined to save time; Google Earth^® aerial pictures are used as a foundation. The flight speed and the gap between photo captures are computed with the program. A schematic of the flight plan is shown in Figure 5.

Regardless of flight height, two flight plans were defined. The first plan involved flights with the camera in the zenithal position (Figure 5a), and in the second plan, flights with an inclination of the camera were conducted (Figure 5b). See Table 2 for a more specific definition of the different studied cases.

The purpose of doubling the number of trajectories (as shown in Figure 5b) was to guarantee that the processed frames would maintain the same angle of inclination with respect to the dune surface. The angle of the camera was defined with respect to the vertical direction and the direction of advance of the RPAS. When the RPAS changed its trajectory, the angle was reversed with respect to the original. Figure 6 shows the positioning of the camera according to the direction of advance. Likewise, the RPAS diagram shows the angle of the camera and how the camera changed its inclination with respect to the profile of the dune when the flight path was altered by 180°.

Frames taken with the wrong camera position were removed later in post-processing. Figure 7 shows the post-processing scheme with the frames.

There were six flights in each campaign. The meteorological conditions and the geography were the same in each case, thus avoiding external influences on the study findings. When establishing the parameters of the area to be photographed, one must define the camera’s specifications and undertake flight planning, including the flight height, the direction of flight, and the overlap of the frame in the lateral and forward directions. Thus, three horizontal flights and three flights with defined slopes (10°, 5°, and 0°, respectively) were carried out. The duration of the flights with the tilted camera was almost doubled due to the fact that more trajectories had to be flown.

The general characteristics of the camera can be found in Contreras et al. [22]. Nevertheless, some aspects should be mentioned here. The mechanical shutter speed is 3:2, the electronic shutter speed is 4:3, and the aspect ratio of the image is 16:9. Moreover, the lens has a format equivalent to 35 mm (autofocus from 1 m to infinity).

Table 4. Data on the type of flight, camera angle, GSD, frame characteristics, frame count, and flight time.

Type of Flight	Angle Camera (°)	Flight Height (m)	GSD (cm/Pixel)	Frame Size 5472 × 3648 Pixels	Frame Count	Flight Time (min)
Sloped	0	Min 100	2.73	149.4 m × 99.6 m	45	7
		Max 126	3.27	178.9 m × 119.3 m
	5	Min 100	2.73	149.4 m × 99.6 m	47	13
		Max 126	3.27	178.9 m × 119.3 m
	10	Min 100	2.73	149.4 m × 99.6 m	47	13
		Max 126	3.27	178.9 m × 119.3 m
Horizontal	0	100	2.73	149.4 m × 99.6 m	45	7
	5		2.73	149.4 m × 99.6 m	47	13
	10		2.73	149.4 m × 99.6 m	47	13

displays the number of frames and the lengths of the flights. The ground sampling distance (GSD), which was directly linked to the flight height and camera parameters, is shown in the fourth column. The distance between two successive pixel centers, as measured on the ground, was used to define the GSD. The spatial resolution of the images and the visibility of their details decrease with increasing GSD values (www.support.pix4d.com, accessed on 17 May 2023).

3.2. Determination of the Errors Obtained

The methodology for obtaining the DTMs involved the use of images taken with the RPAS. These images were processed with the SfM (structure-from-motion algorithm). The software used was Agisoft-Metashape Professional Educational^®.

Three-dimensional information was obtained from the unstructured aerial photos that were gathered through quick, low-cost, and highly automated image processing.

The RPAS–SfM partnership produced a high-quality cartographic output.

First, using the EXIF information from each shot, a group of frames was roughly oriented after they were input into the software. An exchangeable image file, or EXIF, is a standard format for storing interchange information in JPEG-compressed digital photographic image files. This information mostly depends on the camera’s focal length, the time the photo was taken, and the GPS coordinates.

Following the ordering of the entire block, the software was used to search for tie points between adjacent frames. We could then specify the level of accuracy required, the key points, and the maximum tie points to be used in each frame in order to carry out the operation.

This procedure produced a world point cloud that contained each tie point from the flight frameset. The program had already produced a three-dimensional point cloud by this stage. The GCPs were used to modify, georeference, and correct these point clouds for lens distortion. The steps in this process involved entering the GCP coordinates (X, Y, and Z) and graphically demonstrating each frame in which they appeared. The adjusted point cloud would have the same coordinate system as the GCPs points since they were specified with coordinates using the UTM-ETRS89 system. The methodology was similar to that used in [13,36,37].

Comparing the results with the DTM that was defined using the topographic data obtained with the GPS in RTK as a reference, each of the 18 DTMs (generated from the cloud of points obtained with the RPAS) had its error determined in order to assess the overall quality of each flight (Equation (1)). A grid size of 1 m × 1 m was chosen to enable comparison as all of the DTMs had the same dimensions. Additionally, a reliability parameter was calculated, specifically the percentage of grids that had at least one data point.

The comparison produced another DTM, whose distinguishing feature was the difference in altitude between the flight and topographic DTMs (GPS on the ground) or the vertical error (ε) in each grid. The error obtained from comparing the DTMs with different camera inclinations vs. GPS terrain survey are shown in Figure 8 and Figure 9.

$$\epsilon =Z_{flight}-Z_{ground}$$

(1)

However, because some positives and negatives might cancel each other out and produce a false impression of accuracy, this average of the vertical inaccuracies deteriorated. For this reason, the RMSE (Equation (2)), another statistical measure, was computed.

The Federal Geographic Data Committee recently proposed the National Standard for Spatial Data Accuracy (NSSDA), which is applicable to both analog and digital cartographic data [41]. This standard assumes a normal distribution of ε and uses the root mean square error (RMSE) as the most common and valid statistical measure for the evaluation of products obtained via photogrammetry and remote sensing.

$${RMSE}_Z=\sqrt{\frac{1}{n}\sum _{i=1}^n{\left(Z_{flight}-Z_{ground}\right)}^2}$$

(2)

4. Results and Discussion

As already mentioned in the Section 3, since positive and negative values might cancel each other out, the average of the vertical errors (Equation (1)) yields a number that is not particularly useful, but the graphical representation (Figure 8 and Figure 9) helps the discussion. In the figures, we can observe how the errors are smaller in the flights with the nadiral camera regardless of the configuration of the flight height.

Thus, the RMSE was determined (Equation (2)) from the data obtained in each of the cases studied. The results of the three campaigns that were carried out (in June, October, and April) corresponding to the RMSEs are shown in

Table 5. Comparison of the RMSE results for the three campaigns with either an inclined or horizontal flight and the three camera inclinations.

Campaign	Horizontal Flight			Inclined or Sloped Flight
	Camera Inclination
	(10°)	(5°)	(0°)	(10°)	(5°)	(0°)
June 2020	0.090	0.078	0.046	0.073	0.048	0.047
October 2020	0.085	0.073	0.055	0.078	0.055	0.053
April 2021	0.087	0.075	0.043	0.070	0.043	0.041
Average	0.086	0.074	0.048	0.074	0.050	0.047

.

A clearer way of demonstrating the few differences between the three campaigns is with graphics. The RMSEs, in relation to the type of flight and the angle of inclination, are shown in Figure 10.

Although the methodology has been based on the calculation of vertical errors, it should be noted that the RMSE obtained (in X and Y coordinates) was 0.016 and 0.025 m, respectively. These values are similar to the error found in non-steep surfaces. Moreover, there is no difference in the horizontal errors for either a constant height flight or an inclined flight.

It is not the object of this research to compare the three seasons of the year. The reason for carrying out three field campaigns was to have a representative sample of data. By analyzing the angle of inclination of the camera, it can be seen that the lowest RMSE values were obtained with the zenithal or vertical shot (0°). In the case of the horizontal flights, there was a change from an error of 0.048 for a 0° inclination to 0.086 for a 10° inclination; thus, the accuracy of the results decreased by 44%. In the case of the inclined or variable-height flights, there was a change from an error of 0.047 for a 0° inclination to 0.074 for a 10° camera inclination, resulting in a 36% deterioration in the results.

Moreover, to ensure that the angle of inclination of the camera was constant with respect to the dune’s surface, we needed to double the number of passes (Figure 5). This is why the duration of the flight was almost doubled (from 7 to 13 min; see Table 3) and another reason why we did not tilt the camera.

Regarding the effect produced as a result of the type of flight (horizontal or sloped), we only compared the tests with the zenith angle (0°) since we already showed that the inclination of the camera deteriorated the results. The hypothetical improvement in the mean error when performing a sloped flight instead of a horizontal flight was not significant since it only decreased from 0.048 to 0.047 (see Table 5). Therefore, it was not worth carrying out this type of “sloped” flight since a vertical error of 0.04 m is considered acceptable. This accuracy is consistent with other RPAS surveying papers. For instance, errors ranging from 0.053 to 0.064 m appear in the investigation of Taddia [39]. In the same way, Laporte [42] reports similar accuracies (0.050–0.067) m.

5. Conclusions

The topographic mapping of a dune field is essential for the development of focused land management operations. Regrettably, classic topographical surveys of steep dunes are costly due to human effort and the consequent long time required. That is the reason why, in recent years, RPAS surveys have been used more and more frequently.

In this study, some methodological changes for the topographic surveys of dune areas with steep slopes are presented. The flight inclination variable together with the camera inclination variable are introduced to study their influence on the vertical error.

Three field campaigns were performed to obtain a representative amount of data.

Though the campaigns were carried out at different times of the year, no significant differences were found; therefore, errors were independent of the season of the year.

The lowest errors were achieved with no inclination of the camera. In addition, the flights with a tilted camera angle lasted almost twice as long as those without.

Finally, the flight inclination did not lead to a significant reduction in the error; thus, the planning of horizontal flights is recommended.

For all of the cases outlined above, the proposed methodology for dune surveys of steep dunes is to perform horizontal flights with the camera in the zenithal position.