Morpho–Sedimentary Monitoring in a Coastal Area, from 1D to 2.5D, Using Airborne Drone Imagery

Coastal areas are among the most endangered places in the world, due to their exposure to both marine and terrestrial hazards. Coastal areas host more than two-thirds of the world’s population, and will become increasingly affected by global changes, in particular, rising sea levels. Monitoring and protecting the coastlines have impelled scientists to develop adequate tools and methods to spatially monitor morpho-sedimentary coastal areas. This paper presents the capabilities of the aerial drone, as an “all-in-one” technology, to drive accurate morpho-sedimentary investigations in 1D, 2D and 2.5D at very high resolution. Our results show that drone-related fine-resolution, high accuracies and point density outperform the state-of-the-science manned airborne passive and active methods for shoreline position tracking, digital elevation model as well as point cloud creation. We further discuss the reduced costs per acquisition campaign, the increased spatial and temporal resolution, and demonstrate the potentialities to carry out diachronic and volumetric analyses, bringing new perspectives for coastal scientists and managers.


Interest of Morpho-Sedimentary Monitoring in the Global Change Context
Due to their location at the land-sea interface and the increase of the human activity they have undergone for decades, coastal areas are among the most exposed areas worldwide to the effects of global change, especially to the ongoing and predicted sea level rise [1] as well as to the potential increase of storm intensity [2]. In this context, coastal sediment deposits, like beaches, sand dunes, gravel barriers or tidal ridges, have an important role to play to protect human stakes, by acting as a natural buffer zone, thanks to their wave attenuation ability [3][4][5]. The morpho-sedimentary monitoring of these sediment structures across time series is crucial to understand their dynamics in order to account for them into an integrated coastal management.

Common Data Acquisition and Monitoring Methods
To date, several monitoring methods, in situ (handborne) and ex situ (airborne and spaceborne), exist to examine the geomorphology of coastal sediments.

Handborne Monitoring Methods
The most widely used (and oldest) method for morpho-sedimentary monitoring in coastal areas is the manned acquisition of data on the field. This method can be divided into two categories: position tracking of the sediment deposit, and topographic monitoring.
The position tracking of the sediment deposit consists of the recording of the position of a previously determined indicator (shoreline, slope failure, vegetation line, etc.), across several measurement campaigns, using a global navigation satellite system (GNSS) or a more accurate differential GNSS (DGNSS). The accuracy of this monitoring method depends on the precision induced by the material used. This method cannot get information about the topography, the subsequent deposit's volume and its evolution.
The handborne topographic monitoring can be carried out through several methods and the use of different tools. Firstly, some topographic profiles can be surveyed with a tachometer or a DGNSS (Table 1). These topographic profiles allow observation of the surface elevation evolution of the entities along some punctual cross-or long-shore transects [6]. Handborne topographic monitoring can also be done in 2D through a digital elevation model (DEM) using a DGNSS figure field, which enables the volume evolution of the sediment deposit to be estimated, in addition to the topographic profiles [6]. Airborne data source can be used to monitor the coastal areas and its sediment deposit. These data can be used for 1D, 2D and 2.5D coastal monitoring depending on the passive or active nature of the sourced data.
In the case of airborne imagery data, the monitoring methodology consists in a 1D position tracking using a photointerpretation method. This kind of data can be used in the same way as handborne data to study the spatiotemporal evolution of sediment deposit and their shape evolution [7]. Passive imagery presents the advantage to cover a large scale study site, in contrast to handborne methods, and is relatively cost-effective in terms of time and effort. However, the airborne imagery resolution does not allow neither a very high spatial resolution monitoring due to the pixel resolution (Table 1), nor a very high temporal resolution due to the weather constraints.
In the case of airborne light detection and ranging (LiDAR) data, 2D and 2.5D morpho-sedimentary monitoring can be achieved using the DEM [8] product or the raw data in the form of the point cloud. LiDAR data leverage a very high spatial accuracy (Table 1) and a large area covered during the acquisition, but this acquisition method is also very expensive, resulting in a low temporal resolution. The monitoring methods using spaceborne data (satellite imagery and IceSat-2 LiDAR), echoing the airborne data, consist of 1D, 2D and 2.5D monitoring of the sediment deposit through imagery data using a photointerpretation method for 1D position tracking [7,9] and LiDAR data to generate a DEM (2D) or a point cloud (2.5D). The spatial accuracy and resolution of the spaceborne method is negatively correlated with the acquisition altitude of the imagery data (Table 1). The latter one is significantly higher for satellite imagery than for airborne imagery. The digitalization precision used for the monitoring is of a lower quality, but the area covered by a single satellite image remains larger. Disadvantages of the spaceborne method are the same as for airborne imagery data acquired using a MAV (Table 1).

Interest of the Use of Drone (or Unmanned Aerial Vehicle) Imagery for Coastal Monitoring
The drone development for civilian use [10,11] opens new perspectives for coastal expertise by overcoming the limits observed with the previous monitoring methods. This technology offers the possibility of a very high spatial (VHR, <0.05 m) and temporal resolution (<1 month) monitoring of sedimentary structures [12][13][14].
The purposes of this study are to evaluate the airborne drone technology to achieve morpho-sedimentary products in coastal areas, by resolution and accuracy comparisons with those of other state-of-the-science airborne monitoring methods. This work will investigate the drone capabilities for (1) the 1D morpho-sedimentary monitoring (shoreline position tracking) in comparison with VHR airborne passive imagery, (2) the 2D DEM creation (DEM) and (3) the 2.5D point cloud analysis, both latter in comparison with VHR airborne LiDAR data. Furthermore, the potential of drone technology to perform diachronic analyses of the shoreline evolution, topographic profiles and sediment volume change induced by winter storms, will be discussed.

Study Site Presentation
In order to illustrate the use of drone imagery for morpho-sedimentary monitoring, the study will focus on some sedimentary deposits mainly composed of siliceous sand and calcareous shells that play an active role into the coastal protection in the bay of Mont-Saint-Michel, France (BMSM).

General Context of the Bay of Mont-Saint-Michel
The BMSM is located deep in the Normandy-Brittany Gulf between the Cotentin peninsula and the northern coast of Brittany called "Emerald Coast" (Figure 1). This bay, subjected to a strong tidal regime, belongs to the top six areas hosting the world's highest tide [15]. Provided with a tidal range of approximately 14 m, the intertidal zone extends over an important area of 250 km 2 . The bay is bounded in the South from the lowlands of the Dol Marsh polder by the Duchess Anne's dyke, in front of which numerous sediment deposits can be observed, contributing to the coastal protection as natural barriers against waves [3,4]. Due to its lowland feature, the Dol Marsh area is particularly exposed to marine flood risk. The monitoring of this protective geomorphic feature is thereby a key element for the local coastal risk management [3].

Ground Control Points' Positioning
Prior to the imagery data acquisition using a drone, it is necessary to take some ground control points (GCPs), needed for the georeferencing (in x, y, z) of the raw data and by-products. GCPs were distributed in different locations across the study field (Figures 2 and 3). Their spatial reference was recorded using a DGNSS (Topcon hiper V) and then postprocessed using the RTK Lib software, to obtain data with a horizontal accuracy (xy) of 0.05 m and a vertical accuracy around 0.1 m. For this study, 8 GCPs were taken into account on the field (GCPs' root mean square error (RMSE) for this study case: x = 0.03 m, y = 0.05 m, z = 0.08 m).

Ground Control Points' Positioning
Prior to the imagery data acquisition using a drone, it is necessary to take some ground control points (GCPs), needed for the georeferencing (in x, y, z) of the raw data and by-products. GCPs were distributed in different locations across the study field (Figures 2 and 3). Their spatial reference was recorded using a DGNSS (Topcon hiper V) and then postprocessed using the RTK Lib software, to obtain data with a horizontal accuracy (xy) of 0.05 m and a vertical accuracy around 0.1 m. For this study, 8 GCPs were taken into account on the field (GCPs' root mean square error (RMSE) for this study case: x = 0.03 m, y = 0.05 m, z = 0.08 m).

Ground Control Points' Positioning
Prior to the imagery data acquisition using a drone, it is necessary to take some ground control points (GCPs), needed for the georeferencing (in x, y, z) of the raw data and by-products. GCPs were distributed in different locations across the study field (Figures 2 and 3). Their spatial reference was recorded using a DGNSS (Topcon hiper V) and then postprocessed using the RTK Lib software, to obtain data with a horizontal accuracy (xy) of 0.05 m and a vertical accuracy around 0.1 m. For this study, 8 GCPs were taken into account on the field (GCPs' root mean square error (RMSE) for this study case: x = 0.03 m, y = 0.05 m, z = 0.08 m).

Imagery Acquisition Material
The drone platform was a quadcopter DJI Mavic Pro Platinum with a 3-axis stabilized gimbal system and a 3-channel Red Green Blue (RGB) camera (Table 2). In order to compare the monitoring methods achieved using the drone to the common methods, this paper will refer to two other state-of-the-science data sources (Table 3).

Imagery Acquisition Material
The drone platform was a quadcopter DJI Mavic Pro Platinum with a 3-axis stabilized gimbal system and a 3-channel Red Green Blue (RGB) camera (Table 2). In order to compare the monitoring methods achieved using the drone to the common methods, this paper will refer to two other state-of-the-science data sources (Table 3).

Data Acquisition Method
Drone data were collected using a flight plan following previously determined parameters (Table 4).

Data Processing
All drone data processing steps presented here were carried out using the photogrammetry software Pix4D, the total duration of data processing is approximately one hour.

•
The first step of the raw data processing consists into image calibration and image pair matching. A data set of 63 images calibrated with a median of 21,483 key points per images. Pair matching is calibrated using an exact geolocation and orientation method. It was achieved using the original image scale, and correspondence between images was conducted using the acquisition time and triangulation of the image's geolocation.

•
Previously recorded GCPs (see Section 2.2.1) are added to the result of image pair matching to correctly georeference the data set.

•
Dense point cloud is generated with a high point density (number of densified points on the study area: 72 578 307). • 2.5D textured mesh is produced at VHR from the dense point cloud. This mesh is made using a triangulated irregular network.
LiDAR data processing steps were done using the software ENVI LiDAR. The 2.5D mesh is created using a triangulated irregular network (TIN) from the raw point cloud and then DEM is performed by converting this TIN into a raster.

1D Monitoring
The shoreline delineation was based on visual inspection of the seaward foot of vegetation at the same site for both UAV and MAV datasets.
Overlapped drone imageries allow the creation of an orthomosaic, geometrically corrected representation of the Earth's surface adjusted for topographic relief, lens distortion and camera tilt.
The position tracking method by photointerpretation using the drone orthomosaic ( Figure 4) is strictly the same as the common position tracking methods using airborne passive imagery ( Figure 5). The drone orthomosaic permitted to delineate the shoreline with a VHR [16,17], leveraging a pixel size of 0.016 m, whereas the 0.5 m passive airborne imagery (Ortho-littorale v2) impeded an easy visual extraction.

2D Monitoring and Derived Product
The assessment of the DEM creation was made possible by comparing the horizontal and vertical accuracies derived from the 2019 drone and 2018 LiDAR surveys.

2D Monitoring and Derived Product
The assessment of the DEM creation was made possible by comparing the horizontal and vertical accuracies derived from the 2019 drone and 2018 LiDAR surveys.

2D Monitoring and Derived Product
The assessment of the DEM creation was made possible by comparing the horizontal and vertical accuracies derived from the 2019 drone and 2018 LiDAR surveys.
The drone-based DEM ( Figure 6) showed higher accuracy and spatial resolution (horizontal accuracy: 0.032 m; vertical accuracy: 0.048 m; spatial resolution: 0.016 m) than the LiDAR-based DEM (Figure 7; horizontal accuracy: < 2 m; vertical accuracy: < 0.40 m; spatial resolution: < 0.20 m, [18]). Derived topographic profiles (Figures 8 and 9, respectively), important for understanding the cross-shore morpho-sedimentary mechanisms, displayed continuous lines for the drone dataset, contrasting with terrace-like segments for LiDAR. cross-shore morpho-sedimentary mechanisms, displayed continuous lines for the drone dataset, contrasting with terrace-like segments for LiDAR.   cross-shore morpho-sedimentary mechanisms, displayed continuous lines for the drone dataset, contrasting with terrace-like segments for LiDAR.

2.5D Monitoring
In addition to topographic measurements using the 2D products, the drone data allowed the achievement of a point cloud ( Figure 10) with a very high point density (~2.572 points/m 2 , Table 5) in comparison with the point cloud made from the LiDAR (Figure 11;~30 points/m 2 , Table 5).
Drones 2019, 3, x FOR PEER REVIEW 10 of 16 Figure 9. Location of the topographic profiles sliced in the DEM derived from LiDAR data (elevation datum: RAF09). (a), (b), (c) and (d) correspond respectively to the location of topographic profiles presented in Figure 6.

2.5D monitoring.
In addition to topographic measurements using the 2D products, the drone data allowed the achievement of a point cloud ( Figure 10) with a very high point density (~2.572 points/m 2 , Table 5) in comparison with the point cloud made from the LiDAR (Figure 11; ~30 points/m 2 , Table 5).   4. Discussion

Shoreline Analysis
Because of its centimetre-scale spatial resolution, the orthomosaic created using drone data is a better data source (31 times finer) to achieve more accurate shoreline monitoring, based on the foot

2.5D monitoring.
In addition to topographic measurements using the 2D products, the drone data allowed the achievement of a point cloud ( Figure 10) with a very high point density (~2.572 points/m 2 , Table 5) in comparison with the point cloud made from the LiDAR (Figure 11; ~30 points/m 2 , Table 5).

Shoreline Analysis
Because of its centimetre-scale spatial resolution, the orthomosaic created using drone data is a better data source (31 times finer) to achieve more accurate shoreline monitoring, based on the foot Figure 11. Point cloud using the MAV LiDAR data.

Shoreline Analysis
Because of its centimetre-scale spatial resolution, the orthomosaic created using drone data is a better data source (31 times finer) to achieve more accurate shoreline monitoring, based on the foot of vegetation, compared to the decimetre-scale passive airborne imagery. The use of the orthomosaic created by drone data is also very interesting for pixel-based shoreline identification (e.g., the "wet" and "dry" parts of the beach) because of its centimetre-scale spatial resolution [19].
Furthermore, the drone imagery has also the potential to easily achieve a diachronic monitoring by spatially estimating the shoreline retreat value with a potential high to very high temporal resolution (Figure 12; here the two shorelines were timely separated by seven months). of vegetation, compared to the decimetre-scale passive airborne imagery. The use of the orthomosaic created by drone data is also very interesting for pixel-based shoreline identification (e.g., the "wet" and "dry" parts of the beach) because of its centimetre-scale spatial resolution [19].
Furthermore, the drone imagery has also the potential to easily achieve a diachronic monitoring by spatially estimating the shoreline retreat value with a potential high to very high temporal resolution (Figure 12; here the two shorelines were timely separated by seven months).

Topographic Profiles' Analysis
Owing to the better accuracies (62 and 8 times for XY and Z, respectively) and spatial resolution (25 times better) of the DEM achieved using drone data compared to the DEM created from LiDAR data, the drone achieved better topographic analyses through sliced topographic profiles than LiDAR-based DEM by showing the microtopography along the topographic profiles [20].
Moreover, the drone data has the potential to implement a very high temporal resolution monitoring, making it possible to observe subtle changes after each tide or storm event ( Figure 13).

Topographic Profiles' Analysis
Owing to the better accuracies (62 and 8 times for XY and Z, respectively) and spatial resolution (25 times better) of the DEM achieved using drone data compared to the DEM created from LiDAR data, the drone achieved better topographic analyses through sliced topographic profiles than LiDAR-based DEM by showing the microtopography along the topographic profiles [20].
Moreover, the drone data has the potential to implement a very high temporal resolution monitoring, making it possible to observe subtle changes after each tide or storm event ( Figure 13). The analysis of distinct topographic transects highlights the cross-shore sediment dynamics in action in the BMSM during the winter season. On the four measurement transects previously presented, a North-South sediment movement can be observed, from the foreshore to the saltmarsh area. This spatiotemporal pattern corroborates the findings of [21].
A significant elevation levelling of what constituted the crest of the sediment deposit in October The analysis of distinct topographic transects highlights the cross-shore sediment dynamics in action in the BMSM during the winter season. On the four measurement transects previously presented, a North-South sediment movement can be observed, from the foreshore to the saltmarsh area. This spatiotemporal pattern corroborates the findings of [21].
A significant elevation levelling of what constituted the crest of the sediment deposit in October 2018 was observed, in favour of the low zone located on the South part of the study area, whose surface rose considerably between the two measurement periods. This might be due to the sediment remobilization under the influence of the 18 tides, combined with wind-driven waves and currents.

Point Cloud and DEM Analyses
Stemming from a point cloud 86 times denser, the drone-based DEM leverages better spatial resolution and accuracies than the DEM extracted from LiDAR data. It is, once again, more interesting to use it instead of the LiDAR-DEM to evaluate the spatially-explicit morphological changes of a coastal study area between several periods or events ( Figure 14). For example, the DEM comparison, easily achieved by subtracting the October 2018 DEM to the May 2019 DEM, confirms the topographic transects' analyses. A sediment movement from the North to the South part of the study area was observed, from the foreshore to the area located behind the initial sediment deposit. For example, the DEM comparison, easily achieved by subtracting the October 2018 DEM to the May 2019 DEM, confirms the topographic transects' analyses. A sediment movement from the North to the South part of the study area was observed, from the foreshore to the area located behind the initial sediment deposit.

Possible Improvements in Drone Data Acquisition
From a technical point of view concerning the experimental monitoring using the drone imagery, it appears that the overlap ratio, between images, opted for the flight plan has to be improved for the next data acquisition. Indeed, the 60% overlap ratio necessitated a long processing time and an important computation capacity, especially for the image pair matching. It will be facilitated with an 80% to 90% overlap ratio, knowing that an increase of the overlap ratio in the range 80% to 90% will also lead to an increase of the accuracy of the 2.5D modelling software used for the data processing [22]. However, this change would generate the disadvantage of increasing the data processing time, thus the need for high computation capacities.

Conclusions
The data acquisition using an airborne drone, purposed to monitor morpho-sedimentary patterns in a coastal area, presents many interests and advantages in comparison to MAV passive imagery, active LiDAR data, but also other common source data ( Figure 15). time and an important computation capacity, especially for the image pair matching. It will be facilitated with an 80% to 90% overlap ratio, knowing that an increase of the overlap ratio in the range 80% to 90% will also lead to an increase of the accuracy of the 2.5D modelling software used for the data processing [22]. However, this change would generate the disadvantage of increasing the data processing time, thus the need for high computation capacities.

Conclusions
The data acquisition using an airborne drone, purposed to monitor morpho-sedimentary patterns in a coastal area, presents many interests and advantages in comparison to MAV passive imagery, active LiDAR data, but also other common source data ( Figure 15).
Considering the 1D position tracking method using the imagery-based vegetation line (see 1.2.2, 1.2.3 and 3.1), the drone orthomosaic strongly outperformed (31 times finer) the MAV imagery, while being the least expensive source of data. Furthermore, because of its easy deployment, the drone potentially permits a very high temporal resolution (which was only presented here with a seven-month resolution). This capability to finely record changes between two close events is not possible using spaceborne or airborne imagery due to the low acquisition frequency and related cost, respectively. However, spaceborne and airborne imagery can cover larger areas (for a single acquisition campaign) than those of the drone and thus can be useful to monitor some regional scenes.
Regarding the 2D and the 2.5D morpho-sedimentary monitoring, the drone approach greatly surpassed the LiDAR one, with horizontal and vertical accuracies 62 and 8 times better, spatial resolution 25 times better, and point cloud density 86 denser (Table 5). Being much more cost-efficient than a LiDAR campaign (100 times cheaper), the drone holds great promise to survey a very frequent coverage of sites.  Considering the 1D position tracking method using the imagery-based vegetation line (see Sections 1.2.2, 1.2.3 and 3.1), the drone orthomosaic strongly outperformed (31 times finer) the MAV imagery, while being the least expensive source of data. Furthermore, because of its easy deployment, the drone potentially permits a very high temporal resolution (which was only presented here with a seven-month resolution). This capability to finely record changes between two close events is not possible using spaceborne or airborne imagery due to the low acquisition frequency and related cost, respectively. However, spaceborne and airborne imagery can cover larger areas (for a single acquisition campaign) than those of the drone and thus can be useful to monitor some regional scenes.
Regarding the 2D and the 2.5D morpho-sedimentary monitoring, the drone approach greatly surpassed the LiDAR one, with horizontal and vertical accuracies 62 and 8 times better, spatial resolution 25 times better, and point cloud density 86 denser (Table 5). Being much more cost-efficient than a LiDAR campaign (100 times cheaper), the drone holds great promise to survey a very frequent coverage of sites.