Monitoring the Topography of a Dynamic Tidal Inlet Using UAV Imagery Nathalie Long, Bastien Millescamps, Benoît Guillot, Frédéric Pouget, Xavier Bertin

Home , Digital camera

To cite this version:

Nathalie Long, Bastien Millescamps, Benoît Guillot, Frédéric Pouget, Xavier Bertin. Monitoring the Topography of a Dynamic Tidal Inlet Using UAV Imagery. Remote Sensing, MDPI, 2016, Special Issue Remote Sensing in Coastal Environments, 8 (387), 10.3390/rs8050387. halshs-01340509

HAL Id: halshs-01340509 https://halshs.archives-ouvertes.fr/halshs-01340509 Submitted on 1 Jul 2016

HAL is a multi-disciplinary open access L’archive ouverte pluridisciplinaire HAL, est archive for the deposit and dissemination of sci- destinée au dépôt et à la diffusion de documents entific research documents, whether they are pub- scientifiques de niveau recherche, publiés ou non, lished or not. The documents may come from émanant des établissements d’enseignement et de teaching and research institutions in France or recherche français ou étrangers, des laboratoires abroad, or from public or private research centers. publics ou privés. remote sensing

Article Monitoring the Topography of a Dynamic Tidal Inlet Using UAV Imagery

Nathalie Long 1,*, Bastien Millescamps 1, Benoît Guillot 2, Frédéric Pouget 1 and Xavier Bertin 1

1 Littoral, Environnement et Sociétés, Université de la Rochelle—CNRS, 2 rue Olympe de Gouges, La Rochelle 17000, France; [email protected] (B.M.); [email protected] (F.P.); [email protected] (X.B.) 2 Environnements et Paléoenvironnements Océaniques et Continentaux, Université de Bordeaux—CNRS, Allée Geoffroy Saint-Hilaire, Pessac 33615, France; [email protected] * Correspondence: [email protected]; Tel.: +33-05-4650-7633

Academic Editors: Deepak R. Mishra, Richard W. Gould Jr. and Prasad S. Thenkabail Received: 31 December 2015; Accepted: 27 April 2016; Published: 6 May 2016

Abstract: Unmanned Aerial Vehicles (UAVs) are being increasingly used to monitor topographic changes in coastal areas. Compared to Light Detection And Ranging (LiDAR) data or Terrestrial Laser Scanning data, this solution is low-cost and easy to use, while allowing the production of a Digital Surface Model (DSM) with a similar accuracy. Three campaigns were carried out within a three-month period at a lagoon-inlet system (Bonne-Anse Bay, La Palmyre, France), with a flying wing (eBee) combined with a digital camera. Ground Control Points (GCPs), surveyed by the Global Navigation Satellite System (GNSS) and post-processed by differential correction, allowed georeferencing DSMs. Using a photogrammetry process (Structure From Motion algorithm), DSMs and orthomosaics were produced. The DSM accuracy was assessed against the ellipsoidal height of a GNSS profile and Independent Control Points (ICPs) and the root mean square discrepancies were about 10 and 17 cm, respectively. Compared to traditional topographic surveys, this solution allows the accurate representation of bedforms with a wavelength of the order of 1 m and a height of 0.1 m. Finally, changes identified between both main campaigns revealed erosion/accretion areas and the progradation of a sandspit. These results open new perspectives to validate detailed morphological predictions or to parameterize bottom friction in coastal numerical models.

Keywords: UAV photogrammetry; coastal monitoring; tidal inlet; sandspit

1. Introduction Due to their exposure to both marine and terrestrial natural processes as well as anthropogenic activities, coastal environments can exhibit fast morphological changes. To improve the knowledge about these processes, a ﬁrst step is often to perform repetitive topographic surveys. In particular, the morphology of small estuaries and inlets evolves very quickly and can change drastically within several weeks only. Therefore, accurate data with a high temporal frequency are often required [1–4] and several solutions have emerged over the last decade. Satellite images were used to delineate coastal landforms and to demonstrate their temporal changes [5–8]. Digital Surface Models (DSM) can also be generated from tri-stereo images, as was recently achieved with Pleiades images [9,10]. Light Detection and Ranging (LiDAR) and Terrestrial Laser Scanning (TLS) allow the generation of accurate DSMs in coastal areas with a comparable spatial resolution [11–13]. However, the availability of satellite images is not guaranteed under bad weather conditions while the high cost of LiDAR and TLS acquisition can limit the number of campaigns. Atlantic European coasts are exposed to winter storms that can induce coastal damage [14,15]. In France, the retreat of the coastline can reach several tens of meters per year [16] and impact

Remote Sens. 2016, 8, 387; doi:10.3390/rs8050387 www.mdpi.com/journal/remotesensing Remote Sens. 2016, 8, 387 2 of 18 Remote Sens. 2016, 8, 387 2 of 18 infrastructure located on the beachfront, with significant socio-economic consequences. To monitor theseinfrastructure rapid and located substantial on the morphological beachfront, with changes, signiﬁcant an Unmanned socio-economic Aerial consequences. Vehicle (UAV) To combined monitor withthese a rapid digital and camera substantial appears morphological as an attractive changes, solution. an UnmannedIndeed, several Aerial studies Vehicle have (UAV) demonstrated combined withthe performance a digital camera of this appears technique as anin coastal attractive areas, solution. with a Indeed,vertical severalaccuracy studies of the DSM have of demonstrated the order of ±10the performancecm [17–19]. ofWhile this techniqueUAVs are in limited coastal areas,by mete withorological a vertical conditions accuracy of (wind the DSM speed of the has order to be of typically˘10 cm [ 17lower–19]. than While 70–80 UAVs km/h are limitedalong with by meteorological no rain) and the conditions size of the (wind study speed area has (typically to be typically of the orderlower of than 1 km²), 70–80 they km/h are alongable to with collect no images rain) and that the allow size the of the generation study area of a (typically three-dimensional of the order (3D) of 1point km² ),cloud they areand able DSM to collectby a process images thatof photogramme allow the generationtry. The of accuracy a three-dimensional of the resulting (3D) point DSM cloud was alreadyand DSM shown by a processto be similar of photogrammetry. to that of LiDAR The accuracydata [20]. of Several the resulting types DSMof UAVs was alreadywith different shown on-boardto be similar sensors to that are of available, LiDAR data adapted [20]. Severalfor each types type ofof UAVsenvironment with different [21,22]. on-board Compared sensors to other are methodsavailable, discussed adapted for above, each typeUAVs of combine environment several [21 ,advantages22]. Compared that to were other summarized methods discussed by Gonçalves above, andUAVs Henriques combine several[17]: (1) advantages a high level that of wereautomation summarized of photographic by Gonçalves survey; and Henriques(2) a very [low17]: operating (1) a high cost;level of(3) automation a high repeatability of photographic of the survey; survey; (2) (4) a very the lowpossibility operating to cost;obtain (3) aerial a high photography repeatability ofwith the centimetricsurvey; (4) theresolution. possibility In toadditi obtainon, aerialUAV photographymapping methods with centimetricare evolving resolution. quickly and In addition, are subjected UAV intensemapping developments. methods are evolving Thus, Jutzi quickly et al. and [23] are combined subjected a intensecamera developments. and an active Thus,sensorJutzi (lightweightet al. [23] laser-scannercombined a camera line) to and produce an active DSMs sensor in complex (lightweight environments laser-scanner without line) ground to produce control DSMs points. in complex Other innovationsenvironments concern without visualizing ground control data from points. UAVs Other in innovations3D on a virtual concern globe, visualizing both in real data time from and UAVs after landingin 3D on [24]. a virtual globe, both in real time and after landing [24]. This study aims at assessing th thee applicability of photogrammetr photogrammetryy from from UAV to a lagoon-inlet system, to generate high resolution DSMs and provideprovide a detailed description of the morphological changes that the inlet experienced over a three-month summer period.period.

2. Study Study Area Area The study study area area corresponds corresponds to to the the Bonne-Anse Bonne-Anse Lago Lagoon-Inleton-Inlet system, system, located located at the at themouth mouth of the of Girondethe Gironde Estuary, Estuary, in the in central the central part partof the of Bay the of Bay Biscay of Biscay (Atlantic (Atlantic coast, coast, La Palmyre, La Palmyre, France). France). This Thislagoon-inlet lagoon-inlet was selected was selected because because of the offast the evolution fast evolution of this ofsystem this system,, mostly mostly intertidal intertidal under spring under tides.spring The tides. mouth The mouthof the ofbay the is bay2 km is 2long km longand and1.5 km 1.5 kmwide. wide. This Thistidal tidal inlet inlet is composed is composed of well-developedof well-developed flood ﬂood and and ebb ebb deltas deltas and and a amain main channel, channel, which which allows allows navigation navigation to to the small recreational harbor of La Palmyre City (Figure 11).). TheThe shapeshape ofof thethe inletinlet itselfitself evolvesevolves very quickly, moving from a convex to a concave shape. Over the past 15 years, the main channel migrated to the southeast at a mean rate of 93 mm·year¨ year−´1 1withwith a amaximum maximum value value of of 193 193 m·year m¨ year−1 ´[7,8].1 [7, 8].

Figure 1. Location of the study area (A) on the Atlantic coast of France; (B) the Bonne-Anse Figure 1. Location of the study area (A) on the Atlantic coast of France; (B) the Bonne-Anse Lagoon-Inlet; Lagoon-Inlet;and (C) the study and area(C) the (Lambert study area 93 Projection). (Lambert 93 Projection).

Remote Sens. 2016, 8, 387 3 of 18 Remote Sens. 2016, 8, 387 3 of 18

3. Materials Materials and and Methods Methods

3.1. UAV Characteristics The UAV data used in this study were obtained using the eBee flyingflying wing, developed by the SenseFly Company (Cheseaux-sur-La (Cheseaux-sur-Lausanne,usanne, Switzerland). Switzerland). This This flyi flyingng wing wing is is a a very very light light UAV UAV (700 g with the camera) and its wingspan is 96 cm (Figure 22).). eBeeeBee isis anan autonomousautonomous UAVUAV withwith anan on-board artificialartificial intelligenceintelligence system, system, which which analyzes analyzes data data from from an Inertial an Inertial Measurement Measurement Unit (IMU) Unit (IMU)and an and on-board an on-board GPS to GPS optimize to optimize every aspectevery aspect of its flight.of its flight. The main The advantagemain advantage of this of flying this flying wing wingcompared compared to a multicopter to a multicopter is the is size the ofsize the of overflowthe overflow area area per per flight flight but but the the wind wind speed speed has has to beto belower lower than than 40 40 km/h. km/h. A A lithium lithium polymer polymer battery battery providesprovides atat leastleast 5050 min of continuous operation. operation. A radio linked (2.4(2.4 GHz)GHz) byby aa modemmodem allowsallows communicationscommunications betweenbetween thethe software/pilotsoftware/pilot and the ® UAV up to a a distance distance of of 3 3 km. km. The The flight flight is isoperated operated with with the the eMotion eMotion® software,software, provided provided with with the eBee.the eBee. This This software software allows allows planning planning the theflight flight before before the the mission mission and and to to interact interact with with the the UAV UAV during the flight. flight. All the parameters, such as the he heightight of the flight, flight, the overlap between the images or the images’ spatial resolution, are user-specifieduser-specified before each flight.flight. A flight flight simulation, where a fixedfixed wind speed and direction are considered, allows verifying the overflow overflow area and the autonomy autonomy of the UAV (or ifif severalseveral flightsflights areare needed).needed). The number of photos, the flyingflying time and the surface of the covered covered area area are are also also computed computed through through this this simulation. simulation. After After the the flight, flight, the the same same software software is usedis used to togeoreference georeference all all the the images images according according to to the the flight flight effectively effectively performed performed recorded recorded by by the onboard GNSS. The UAV UAV is equipped equipped with with a a CANON CANON Powershot Powershot EL ELPH110PH110 HS HS RGB RGB camera camera with with a resolution a resolution of 16.1of 16.1 Mpixel. Mpixel. Its Itsfocal focal length length ranges ranges from from 4.3 4.3mm mm and and 21.5 21.5 mm, mm, which which for a for flight a flight height height of 150 of 150m, for m, example,for example, yields yields a ground a ground sampling sampling distance distance of 4.69 of 4.69 cm. cm.

Figure 2. eBee UAV during a field campaign, in flight and the hardware. Figure 2. eBee UAV during a ﬁeld campaign, in ﬂight and the hardware.

3.2. Photogrammetry Process To monitor coastal coastal topography, topography, a a3D 3D analysis analysis is isrequired. required. For For a a3D 3D reconstruction, reconstruction, a photogrammetrya photogrammetry process process is used, is used, based based on the on Structure the Structure From From Motion Motion (SFM) (SFM) algorithm. algorithm. The photogrammetryThe photogrammetry is a istechnique a technique which which allows allows reconstructing reconstructing a arelief relief from from several several stereoscopic images of the same object. Basically, Basically, the SFM algorithmalgorithm allows reconstructing a 3D scene geometry from a set set of of images images of of a a static static scene scene by by matching matching features features on on multiple multiple images. images. A A3D 3D point point cloud cloud is generatedis generated and and georeferenced georeferenced using using ground ground control control points points (hereafter (hereafter GCP). GCP). The The SFM SFM algorithm is based on a multi-view of the scene and the redundan redundancycy of the information allows the success of this process [25–27]. Mancini et al. [19] have compared the accuracy of the DSMs generated from TLS data and UAV data using the SFM algorithm and showed that the elevation accuracies were similar.

Remote Sens. 2016, 8, 387 4 of 18 process [25–27]. Mancini et al. [19] have compared the accuracy of the DSMs generated from TLS data andRemote UAV Sens. data 2016, using8, 387 the SFM algorithm and showed that the elevation accuracies were similar. 4 of 18 The SFM algorithm is available in several softwares to generate DSMs and orthomosaics. Free or The SFM algorithm is available in several softwares to generate DSMs and orthomosaics. Free open-source software such as Cloud Compare, Mic-Mac [28] or Opensource Photogrammetry can or open-source software such as Cloud Compare, Mic-Mac [28] or Opensource Photogrammetry can also be used. We have chosen the Agisoft Photoscan® Professional Edition software (version 1.1.6), also be used. We have chosen the Agisoft Photoscan® Professional Edition software (version 1.1.6), which is well suited to UAV image processing [29]. The workflow proposed by Photoscan to generate which is well suited to UAV image processing [29]. The workflow proposed by Photoscan to a DSM is divided into different steps and uses a well-known photogrammetric approach (Figure3). generate a DSM is divided into different steps and uses a well-known photogrammetric approach The first stage is the image alignment. At this stage, common points on images are identified and (Figure 3). The first stage is the image alignment. At this stage, common points on images are matched, as well as the position of the camera for each image. The camera calibration parameters are identified and matched, as well as the position of the camera for each image. The camera calibration refined. The values for the initial camera parameters are taken from the EXIF (EXchangable Image file parameters are refined. The values for the initial camera parameters are taken from the EXIF Format) header: (EXchangable Image file Format) header: ‚● fx,fx, fy:fy: focalfocal lengthlength inin x-x- andand y-dimensionsy-dimensions measuredmeasured inin pixels,pixels, ‚● cx,cx, cy: principal principal point point coordinates, coordinates, i.e.,i.e. coordinates, coordinates of lens of lensoptical optical axis interception axis interception with sensor with sensorplane, plane, ‚● skew:skew: skewskew transformationtransformation coefficient,coefficient, ‚● k1,k1, k2,k2, k3:k3: radialradial distortiondistortion coefficients,coefficients, ‚● p1,p1, p2:p2: tangentialtangential distortiondistortion coefficients.coefficients.

GCPsGCPs areare introducedintroduced inin thethe modelmodel toto performperform thethe imageimage alignment.alignment. Each GCPGCP isis manuallymanually assignedassigned toto oneone imageimage where where the the target target is is visible visible and and recognizable, recognizable, and and then then automatically automatically assigned assigned to everyto every other other image image that that contains contains the the same same GCP. GCP. This This step step is is concluded concluded by by a a last last optimization optimization imageimage alignment.alignment. The possible non-linear deformations of the model can bebe removedremoved byby optimizingoptimizing thethe estimatedestimated pointpoint cloud cloud and and camera camera parameters parameters based base ond on the the known known reference reference coordinates. coordinates. During During this optimization,this optimization, estimated estimated point point coordinates coordinates and camera and camera parameters parameters are adjusted are adjusted to minimize to minimize the sum the of reprojectionsum of reprojection errors and errors reference and refere coordinatence coordinate misalignment misalignment errors. errors.

FigureFigure 3.3. GeneralGeneral overviewoverview ofof methods,methods, startingstarting withwith thethe preparationpreparation ofof overﬂightoverflight dronedrone campaignscampaigns untiluntil thethe determinationdetermination ofof DSMDSM andand orthomosaicorthomosaic accuracyaccuracy byby GNSSGNSS datadata (proﬁle(profileand and ICPs). ICPs).

Then,Then, aa densedense pointpoint cloudcloud isis computed,computed, georeferencedgeoreferenced inin aa real-worldreal-world coordinatecoordinate system,system, andand aa triangulartriangular meshmesh is is built. built. Lastly, Lastly, the mosaicthe mosaic of images of images is draped is ondraped the mesh on tothe produce mesh anto orthomosaic.produce an Theorthomosaic. results (DSM The andresults orthomosaic) (DSM and areorthomosaic) exported accordingare exported to a according selected projection to a selected system projection and a spatialsystem resolution. and a spatial resolution.

Remote Sens. 2016, 8, 387 5 of 18

3.3. Field Campaigns and Data Acquisition

3.3.1. Image Acquisition To study the spatial variability of the topography of the Bonne Anse Lagoon-Inlet, three campaigns were executed in June, September and October 2015. The same flight plan was used: three flights were necessary to cover the whole area. Campaigns 1 (16 June 2015) and 2 (28 September 2015 and 2 October 2015) cover the whole study area (three flights per campaign). On the 2 October 2015, an additional flight was realized at a lower altitude (50 m) over the flood delta (campaign 3), in order to provide a higher spatial resolution (Figure3). The height of the flights was about 149 m above the ground level (AGL) on average for the two first campaigns, which yields images with a 4.6 cm spatial resolution. For the third campaign, the 50 m AGL yielded a spatial resolution of 2 cm. The latitudinal overlap was 60% and the longitudinal overlap was 75%. The flight plans were prepared on the eMotion® software and the mission area was saved in order to repeat the same mission from one campaign to another. Only the direction of the flight was adapted to the wind direction, considering that the UAV has to fly against the wind to provide better-quality images. Around 700 photos were recorded to cover this area of about 400 ha (Figure3). The meteorological conditions and the tidal level varied between the three campaigns: for campaign 1, the wind was very irregular with strong gusts; the UAV lost its trajectory several times and had to come back on the right trajectory. For campaigns 2 and 3, the wind speed was moderate but regular, and in October, the wind speed was very low so the flights occurred without any problem. The tidal range was slightly different: 3.95 m for campaign 1 against 4.75/4.10 m for campaigns 2 and 3 (Table1).

Table 1. Meteorological conditions, tidal ranges and tidal levels with respect to marine charts datum for the three campaigns.

Date Wind Speed (m/s) Wind Direction (˝) Tidal Range (m) Tidal Level (m) 16 June 2015 5–7 20 3.95 1.15 28 September 2015 9–11 70 4.75 0.80 2 October 2015 2 70 4.10 1.15

3.3.2. GNSS Surveys GCPs are required for the georeferencing. In natural environments in general and in coastal areas in particular, a few time-invariant objects can usually be identified. Therefore, artificial targets were also used: white sheets of paper were deployed on the ground and were partially buried because of the wind. These artificial targets were placed and their coordinates were surveyed, just before low tide, to realize the flight before the rising tide. Forty-six and 56 targets were placed and surveyed using GNSS receivers for the two first campaigns, respectively. For the third campaign, 24 GCPs were acquired (Figure3). GCPs were roughly equally dispatched with two GNSS receivers (Figure4). For campaign 2, a 677 m GNSS profile with 529 points was also surveyed, crossing the flood delta in front of the harbor entrance (Figure4b). Then, to evaluate vertical and horizontal errors, complementary GNSS measurements were carried out in February 2016 on stable areas of the study site (sea promenade, bike path and parking lot). A total of 25 independent check points (ICPs) were surveyed (Figure4a,b). Remote Sens. 2016, 8, 387 6 of 18 Remote Sens. 2016, 8, 387 6 of 18

Figure 4. Orthomosaics of the three campaigns: ( (aa)) campaign campaign 1; 1; ( (bb)) campaign campaign 2 2 and and ( (cc)) campaign campaign 3. 3. Location of the GCPs (black circles), the ICPs (black(black triangles) and the profile proﬁle used to estimate the accuracy ofof thethe DSMs. DSMs. The The grey grey areas areas correspond correspon tod areasto areas where where the photogrammetric the photogrammetric processes processes failed faileddue to due the lackto the of tielack points. of tie Whitepoints. areas White correspond areas correspond to areas whereto areas no where data were no data available. were Blueavailable. areas Bluecorrespond areas correspond to subtidal/water to subtidal/water areas. areas.

The following methodology was applied to perform and post-process the GNSS surveys: a base station waswas settledsettled immediately immediately near near the the zone zone of study,of study, less less than than 3 km 3 ofkm the of farthest the farthest region region of the of ﬂying the flyingzone. Beforehand,zone. Beforehand, this station this wasstation the was object the of object a long-term of a long-term measurement measurement with differential with differential correction (withcorrection a permanent (with a permanent GPS network), GPS whichnetwork), allowed which determining allowed determining its coordinates its coordinates (X, Y and ellipsoidal(X, Y and ellipsoidalheight) with height) a precision with a betterprecision than better 10 cm. than A 10 ﬁxed cm. GNSS A fixed receiver GNSS receiver was placed was onplaced this baseon this station base stationduring allduring GNSS all measurements. GNSS measurements. At the end At of the the end session, of the a differential session, a correctiondifferential with correction regard towith the regard to the fixed station was performed for all the observations obtained on the ground. The first GNSS receiver used was a GeoXH (Trimble) decametric GNSS and it was used only for campaign 1.

Remote Sens. 2016, 8, 387 7 of 18

fixed station was performed for all the observations obtained on the ground. The first GNSS receiver used was a GeoXH (Trimble) decametric GNSS and it was used only for campaign 1. The data was post-processed following a Post-Processing Kinematic (PPK) method, with the base station of “Royan” (at a distance of 13 km to the southeast of the study area), using the GPS Pathfinder (Trimble) software. This base station is a permanent GNSS station, part of the French Geographical Institute GNSS Network (IGN-RGP). Considering the extension of the surveyed area, a second GNSS receiver, a Topcon Hiper Pro, was used to complete the topographic acquisitions. The data acquired with this receiver was post-processed with the same PPK method. The software used for this post-processing was RTKLib, an open-source software [30]. The measuring time was 2 min for each point, with a time step of 1 s. The respective heights of the GNSS antenna and the base were measured for each campaign and used for the post-processing. All field data were acquired in WGS84. Applying this methodology, an interruption of the radio connection between the base and the receiver does not disturb the survey and allows a guaranteed vertical absolute precision better than 8 cm and an horizontal absolute precision around 1 cm, according to comparisons previously performed by our group at several geodetic points.

3.4. Vertical and Horizontal Discrepancy To assess the vertical accuracy of UAV-derived DSMs, the arithmetic average discrepancy and the root mean square discrepancy (RMSD) were calculated based on the difference between the ellipsoidal height measured with GNSS (ICPs and proﬁle) and extracted from DSMs at these coordinates (Figure3). These data were also compared against LiDAR data, originating from a topographic survey performed from 16 June 2010 to 13 October 2010 by the French Geographical Institute (IGN) using an airborne laser. The digital terrain model is in the form of a regular grid with a cell size of 1 m ˆ 1 m, derived from LiDAR points. To assess the horizontal accuracy, the distance between the location of ICPs determined with GNSS and the position given by orthomosaics (white markings on the ground) was measured. This assessment was made possible thanks to the high spatial resolution of orthomosaics (4.6 cm/pix).

3.5. Spatial Analysis The DSMs and the orthomosaics were analyzed within a geographical information system (GIS) based on the ArcGIS® software 10.2 (Esri Company, Redlands, CA, USA). The results were exported in a Lambert 93 projection. First, the difference between the DSMs of the first two campaigns was computed and the areas of erosion and accretion were quantified. In order to perform a consistent comparison between both DSMs, a mask was produced to keep only surfaces that were emerged for both campaigns. Changes in the volume of the sandspit located to the south of the study area were also quantified. Finally, a profile was extracted at the end of the sandspit to illustrate the potential of the UAV solutions to quantify detailed morphological changes. This spatial analysis was first performed on a large scale (400 ha), in order to identify the main morphological changes. A finer-scale analysis was also performed over a smaller area (33 ha) to highlight small 3D morphological features that are hardly identifiable from satellite images or other data.

4. Results The first two campaigns were composed of three flights per campaign, which were processed all at once (Figure5). The last campaign was composed of one flight. The following section presents the DSM and orthomosaic results and their accuracy, evaluated against the GNSS profile and ICPs. Remote Sens. 2016, 8, 387 8 of 18 Remote Sens. 2016, 8, 387 8 of 18

Figure 5.5. IllustrationIllustration of of the the different different stages stages of theof the photogrammetric photogrammetric process process at the at flood the flood delta delta sand bank:sand alignmentbank: alignment of images, of images, creation creation of point of cloud, point creationcloud, creation of dense of point dense cloud, point creation cloud, creation of model of texture. model texture. 4.1. Image Processing 4.1. Image Processing For each flight, images were georeferenced using the Post-flight Manager tool of the eMotionFor each software. flight, All images bright were surfaces georeferenced and water usin surfacesg the werePost-flight masked Manager to avoid tool error of the during eMotion the photogrammetrysoftware. All bright process. surfaces Image and water alignment surfaces was we performedre masked with to avoid “high” error accuracy during andthe photogrammetry “reference” pair preselectionprocess. Image options alignment of Photoscan was performed software (Figurewith “hig5). Ah” higher-accuracy accuracy and “reference” setting helps pair to obtainpreselection more accurateoptions of image Photoscan position soft estimates.ware (Figure In the 5). “reference” A higher-accuracy preselection setting mode, helps the photosto obtain overlapping more accurate pairs wereimage selected position based estimates. on the In measured the “reference” image locations.preselection The mode, GCPs the were photos imported, overlapping correctly pairs assigned were onselected each imagebased manuallyon the measured and the orientationimage locations. process The was GCPs optimized. were imported, Optimizing correctly images assigned during theon photogrammetriceach image manually process and resulted the orientation in small changesprocess ofwas camera optimized. parameters Optimizing (Table2 ).images during the photogrammetric process resulted in small changes of camera parameters (Table 2). Table 2. Optimization of the camera parameters (pixel). Table 2. Optimization of the camera parameters (pixel). Camera Adjusted Adjusted Adjusted Camera Initial Adjusted Adjusted Adjusted ParametersInitial (Campaign 1) (Campaign 2) (Campaign 3) Parameters (Campaign 1) (Campaign 2) (Campaign 3) fx 3212.47 3272.45 3274.76 3267.49 fx fy3212.47 3212.47 3273.173272.45 3275.59 3274.76 3268.28 3267.49 fy cx3212.47 2304 3273.17 2334 2334.12 3275.59 2333.6 3268.28 cx cy2304 1728 1805.71 2334 1808.93 2334.12 1811.37 2333.6 cy skew1728 0 1.441951805.71 1.362241808.93 1.19702 1811.37 k1 0 ´0.0418731 ´0.0422142 ´0.0409088 skew 0 1.44195 1.36224 1.19702 k2 0 0.0426406 0.0424654 0.0421629 k1 k30 0 ´−0.02209190.0418731 ´0.0217889−0.0422142´0.0223425 −0.0409088 k2 k40 00.0426406 0 00.0424654 0 0.0421629 k3 p10 0 0.00422248−0.0220919 0.00434259−0.0217889 0.00435362 −0.0223425 k4 p20 0 0.00256945 0 0.00258894 0 0.00246941 0 p1 0 0.00422248 0.00434259 0.00435362 The resultingp2 residual error0 computed0.00256945 on GCPs is very close0.00258894 between the three0.00246941 campaigns and around 0.3 pixel on average, with one outlier at 0.97 pixel for the first campaign (Table3). However, the errorThe resulting distribution residual ranges error from computed 0.1 and 1on for GCPs campaign is very 1 close with between a modal the class three at 0.3 campaigns pixel; for and the secondaround campaign,0.3 pixel on the average, error distributionwith one outlier is narrower at 0.97 pixel and centeredfor the first on campaign the value of(Table 0.4 pixel. 3). However, For the thirdthe error campaign, distribution the modal ranges class from is 0.2 0.1 pixel and (Figure1 for campaign6). The adverse 1 with weathera modal conditions class at 0.3 in pixel; June forfor thethe firstsecond campaign campaign, may the explain error thedistribution outlier of is campaign narrower 1. and centered on the value of 0.4 pixel. For the third campaign, the modal class is 0.2 pixel (Figure 6). The adverse weather conditions in June for the first campaign may explain the outlier of campaign 1.

Remote Sens. 2016, 8, 387 9 of 18 Remote Sens. 2016, 8, 387 9 of 18 Remote Sens. 2016, 8, 387 9 of 18

Figure 6. Histograms of errors ofof imagesimages georeferencinggeoreferencing from from GCPs GCPs for for (a ()a campaign) campaign 1; 1; (b ()b campaign) campaign 2; Figure 6. Histograms of errors of images georeferencing from GCPs for (a) campaign 1; (b) campaign 2;and and (c )(c campaign) campaign 3. 3. 2; and (c) campaign 3.

TableTable 3. Assessment of the image ge georeferencingoreferencing error from GCPs for the three campaigns.campaigns. Table 3. Assessment of the image georeferencing error from GCPs for the three campaigns. No. Photo No. Tie X Error Y Error Z Error Error Error Campaigns No.No. PhotoPhoto No.No. Tie Tie XX Error Error Y ErrorY ErrorZ ErrorZ Error Error Error CampaignsCampaigns Error (m) Error (pixel) UsedUsed PointsPoints (m)(m)(m) (m)(m)(m) (m) (m)(m) (m)(m) (pixel)(pixel) CampaignCampaignCampaign 11 672672672 12017381201738 0.0410.041 0.0770.0770.077 0.0180.0180.018 0.0890.0890.089 0.3040.3040.304 CampaignCampaignCampaign 22 643643643 19483131948313 0.0120.012 0.0090.0090.009 0.0120.0120.012 0.0190.0190.019 0.3160.3160.316 CampaignCampaignCampaign 33 301301301 351977351977 0.0040.004 0.0840.0840.084 0.0160.0160.016 0.0920.0920.092 0.2920.2920.292

BeforeBefore investigatinginvestigating the the accuracy of the DSM producedproduced by by Photoscan, Photoscan, it it is is essential essential to to visualize visualize thethe nadirnadir andand overlapoverlap ofof thethe imagesimages for thethe threethree campaignscacampaignsmpaigns (Figure (Figure 7 7).7).). As As previously previously noted, noted,noted, flight ﬂightflight plansplans hadhad aa laterallateral overlapoverlap ofof 60%60% and a longitudinal overlap overlap of of 75%. 75%. According According to to the the overlapping overlapping maps,maps, thethe image image number number decreasesdecreases from from nine nine in in thth theee studystudy study areaarea area centercenter center toto to twotwo two along along along the the the periphery. periphery. periphery. DuringDuring thethe thirdthird campaign,campaign, thethe recoveryrecovery did not exceedexceed eight eight images. images.

FigureFigure 7. 7. ImageImageImage recovery recoveryrecovery maps maps for the three campaigncampaign campaigns.s.s. BlackBlack pointspoints correspond correspond to to nadir nadir images. images.

4.2.4.2. Digital Digital SurfaceSurface Model Model

4.2.1.4.2.1. Construction Construction of of the the Digital Digital Surface Model TheThe densedense pointpoint cloudcloud waswas performedperformed using the the “medium “medium quality” quality” and and an an “aggressive” “aggressive”“aggressive” depth depth filteringfilteringfiltering optionsoptionsoptions of ofofPhotoscan. Photoscan.Photoscan. These These settings settings help helphelp to to reduceto reducereduce processing processingprocessing time timetime when when when dealing dealing dealing with with with a large a a largenumberlarge numbernumber of tie points, ofof tietie points,points, as is the asas case isis the here. case Then, here.the Then,Then, mesh thethe of meshmesh the ground ofof thethe ground heightground washeight height computed was was computed computed from the fromdensefrom the the cloud dense dense without cloudcloud interpolation withoutwithout interpolationinterpolation of the data of (Figurethethe datadata5 ),(Figure(Figure which 5),5), avoids which which filling avoids avoids in filling areasfilling in where in areas areas data where where are datalacking.data areare Onlacking.lacking. the sand, OnOn thethe and sand,sand, particularly and particularly on white on and whitewhite smooth andand sand smoothsmooth areas, sandsand the areas, areas, generation the the generation generation of the DSM of of wasthethe DSM DSM difficult was was because difficultdifficult the becausebecause model the did model not find did a nono sufficienttt findfind aa sufficientsufficient number of numbernumber tie points. of of tie tie points. points.

Remote Sens. 2016, 8, 387 10 of 18

Remote Sens. 2016, 8, 387 10 of 18 Figure8 shows the DSMs obtained for each campaign. Except in areas where not enough tie points wereFigure found 8 shows or waterthe DSMs coverage obtained differed for each between campaign. campaigns Except 1 in and areas 2, the where overlap not enough between tie the points were found or water coverage differed between campaigns 1 and 2, the overlap between the two full DSMs is good. Outside the anthropized area, the ground is relatively ﬂat and smooth and two full DSMs is good. Outside the anthropized area, the ground is relatively flat and smooth and its its height ranges from 45 and 47 m, except at the end of the sandspit where some dunes reach 49 m. height ranges from 45 and 47 m, except at the end of the sandspit where some dunes reach 49 m. In In order to demonstrate the relevance of this high spatial density of data, a DSM was generated on the order to demonstrate the relevance of this high spatial density of data, a DSM was generated on the central sand bank with a grid size of 2 cm and several levels of zoom are shown in Figure8c. The ﬁrst central sand bank with a grid size of 2 cm and several levels of zoom are shown in Figure 8c. The twofirst levels two oflevels zoom of inzoom the in left the panels left panels show show the development the development of tidal of tidal dunes dunes with with a wavelengtha wavelength of of the orderthe oforder 10 m of and 10 am height and ofa height 0.3 m. Theof 0.3 third m. zoomThe third level inzoom the rightlevel panelin the shows right thepanel development shows the of superimposeddevelopment bedformsof superimposed with a bedforms wavelength with of a the wavele orderngth of 1of m the and order a height of 1 m ofand 0.10 a height m. of 0.10 m.

Figure 8. DSMs obtained from the three campaigns: (a) campaign 1; (b) campaign 2; and (c) campaign Figure 8. DSMs obtained from the three campaigns: (a) campaign 1; (b) campaign 2; and (c) campaign 3. 3. The spatial resolutions are 20 cm for (a) and (b) and 2 cm for (c). Isolines computed over The spatial resolutions are 20 cm for (a) and (b) and 2 cm for (c). Isolines computed over 5-m-resolution 5-m-resolution DSMs were superimposed every 0.5 m to improve the representation of the DSMs were superimposed every 0.5 m to improve the representation of the morphology on ﬁgures a–c. morphology on figures a–c. Grey areas correspond to lack of tie points. White areas correspond to no Grey areas correspond to lack of tie points. White areas correspond to no data (inside the sand banks). data (inside the sand banks).

Remote Sens. 2016, 8, 387 11 of 18

Remote Sens. 2016, 8, 387 11 of 18 4.2.2. Vertical Accuracy of Digital Surface Models 4.2.2. Vertical Accuracy of Digital Surface Models ‚● WithWith the the GNSS GNSS profile profile ToTo assess assess the the vertical vertical accuracy accuracy of the of DSMs,the DSMs, the ellipsoidal the ellipsoidal heights heights of the GNSS of the profile GNSS surveyed profile duringsurveyed campaign during 2campaign (Figure4) were2 (Figure compared 4) were against comp theared heights against of the pointsheights extracted of the points from theextracted DSM atfrom the samethe DSM coordinates at the same (Figure coordinates9a). The DSM(Figure heights 9a). The are closeDSM toheights that of are the close GNSS to profile,that of althoughthe GNSS withprofile, a slight although positive with bias a slight of the positive order of bias 10 cm. of Somethe order larger of differences10 cm. Some appear larger at differences the beginning appear of the at profilethe beginning (first 100 of meters) the profile and (first correspond 100 meters) to the and trough correspond of bedforms to the where trough water of bedforms was still where present water and causedwas still disturbance present and in thecaused photogrammetry disturbance process.in the photogrammetry The histogram of process. the height The difference histogram between of the theheight DSM difference and the GNSS between profile the showsDSM and that the the GNSS main partprof ofile the shows vertical that errors the main ranges part from of the 5 to vertical 15 cm (Figureerrors ranges9b). The from arithmetic 5 to 15 cm average (Figure discrepancy 9b). The arit ishmetic 6.78 cm average and the discrepancy root mean squareis 6.78 cm discrepancy and the root is 9.44mean cm. square The precisiondiscrepancy of the is 9.44 GNSS cm. surveys The precis andion the of precision the GNSS of thesurveys DSM and were the added precision to produce of the aDSM conservative were added cumulative to produce error a conservative of ˘20 cm. Sincecumulati theve GNSS error profile of ±20 has cm. only Sinc beene the surveyedGNSS profile during has campaignonly been 2,surveyed this cumulative during campaign error is only 2, this valid cumulative for this campaign. error is only valid for this campaign.

(a) (b)

FigureFigure 9.9.( (aa)) EllipsoidalEllipsoidal heightsheights ofof thethe GNSSGNSS proﬁleprofile (in(in black)black) andand extractedextracted fromfrom thethe DSMDSM (in(in red)red) ofof campaigncampaign 2 2 and and ( b(b)) histogram histogram of of the the ellipsoidal ellipsoidal height height differences. differences.

‚● WithWith independent independent control control points points

ToTo determinedetermine thethe verticalvertical accuracyaccuracy ofof DSMsDSMs forfor thethe entireentire study,study,error errorcalculations calculations werewere mademade onon thethe ICPs.ICPs. ICPsICPs werewere surveyedsurveyed withinwithin thethe wholewhole anthropizedanthropized areaarea locatedlocated toto thethe northnorth ofof thethe studystudy area.area. ThoseThose ICPsICPs consistconsist ofof groundground markingsmarkings thatthat cancan bebe consideredconsidered asas invariantinvariant betweenbetween campaignscampaigns (Figure(Figure4 ).4). TheThe scatterscatter plotplot ofof thethe ICPs’ICPs’ ellipsoidalellipsoidal heightheight betweenbetween DSMDSM datadata andand GNSSGNSS datadata (Figure(Figure 1010)) showsshows ﬁrstlyfirstly thatthat thethe scatter is small and and comparable comparable for for the the three three datasets datasets considered considered (campaigns (campaigns 1, 1,2 2and and LiDAR). LiDAR). The The root root mean mean square square discrepancy discrepancy (RMSD) (RMSD) is is 17 17 cm cm for campaign 1 and 1616 cmcm forfor campaigncampaign 2,2, withwith aa slightslight negativenegative biasbias ofof´ −0.080.08 mm andand ´−0.07 m,m, respectively.respectively. TheThe LiDARLiDAR datadata areare slightlyslightly less less accurate,accurate, with with a a RMSD RMSD of of 0.21 0.21 mm andand aa slightslight positivepositive biasbias ofof 0.080.08 m.m. Here,Here, wewe referredreferred toto discrepancydiscrepancy ratherrather thanthan errorerror becausebecause GNSSGNSS datadata havehave theirtheir ownown errors.errors. SinceSince thethe samesame GNSSGNSS werewere usedused to to position position GCPs GCPs used used in thein the photogrammetry photogrammetr process,y process, the errors the errors associated associated with DSM with and DSM GNSS and dataGNSS are data not are strictly not strictly independent, independent, and therefore and therefore the computation the computation of total of errortotal iserror not is a trivialnot a trivial task. Alternatively,task. Alternatively, we adopted we adopted a conservative a conservative approach whereapproach we linearlywhere we summed linearly the summed error estimation the error of theestimation GNSS data of the and GNSS the RMSD data and of the the DMS RMSD data of described the DMS above.data described This process above. resulted This process in a cumulative resulted errorin a cumulative of ˘27 cm. Becauseerror of ICPs±27 cm. were Because surveyed ICPs on were stable surveyed areas in on time, stable and areas therefore in time, used and in alltherefore DSMs, thisused cumulative in all DSMs, error this is cumulative used to deﬁne error a verticalis used to accuracy define a of vertical UAV-derived accuracy DSMs. of UAV-derived DSMs.

Remote Sens. 2016, 8, 387 12 of 18

Remote Sens. 2016, 8, 387 12 of 18 Remote Sens. 2016, 8, 387 12 of 18

Figure 10. Scatter plot of ICPs’ ellipsoidal heights between DSMs data and GNSS data. The blue dots correspond to campaign 1 while the red dots correspond to campaign 2.

4.3. Horizontal Accuracy of Orthomosaics

For campaign 1, planimetric errors are distributed between 3.55 and 34.14 cm, with a root mean squareFigure discrepancy 10. Scatter plot of 12.84 of ICPs’ cm ellipsoidal (Figure 11). heights The betweenbe valuestween obtainedDSMs data for andand campaign GNSSGNSS data.data. 2 are The similar blue dots and rangecorrespond from 2.78 toto campaigncampaignto 25.29 cm, 11 whilewhile with thethea root redred mean dotsdots correspondcorrespond square discrepancy toto campaigncampaign of 2.12.362. cm. The horizontal error given for the GNSS receiver (1 cm) and the operator error when measuring 4.3.the Horizontal difference Accuracy between of ICPs Orthomosaics and orthomosaics (2 pixels which is 9.2 cm) must be added to the first error,For which campaign results 1, planimetricin a xy cumulative errors areerror distributed of ±23 cm. betweenBecause ICPs 3.55 andwere 34.14 surveyed cm, withon stable a root areas mean squareused discrepancydiscrepancyin all orthomosaics, of of 12.84 12.84 cmthis cm (Figure cu(Figuremulative 11 ).11). Theerror Th values ewas values obtainedused obtained to defin for campaignefor a campaignhorizontal 2 are 2 similaraccuracy are similar and of rangethe and orthomosaics. fromrange 2.78 from to 2.78 25.29 to cm, 25.29 with cm, a with root meana root squaremean square discrepancy discrepancy of 12.36 of cm. 12.36 cm. The horizontal error given for the GNSS receiver (1 cm) and the operator error when measuring the difference between ICPs and orthomosaics (2 pixels which is 9.2 cm) must be added to the first error, which results in a xy cumulative error of ±23 cm. Because ICPs were surveyed on stable areas used in all orthomosaics, this cumulative error was used to define a horizontal accuracy of the orthomosaics.

FigureFigure 11. 11.Horizontal Horizontal differences differences between between GNSSGNSS ICPs and orthomosaics orthomosaics from from campaigns campaigns 1 and 1 and 2. 2.

4.4.The Morphological horizontal Changes error given for the GNSS receiver (1 cm) and the operator error when measuring the differenceAt the scale between of the ICPs whole and tidal orthomosaics inlet, the compar (2 pixelsative which study is 9.2of these cm) must DSMs be shows added that to thesignificant first error, whichchanges results arein restricted a xy cumulative to a few areas error only, of ˘23 while cm. the Because main ICPspart of were the surveyedsystem remained on stable stable areas within used in allthe orthomosaics, accuracy of thisour cumulativemethod (Fig errorure 12). was Among used to the define areas a where horizontal significant accuracy changes of the occurred, orthomosaics. the terminalFigure portion 11. Horizontal of the sandspit differences of Bonne-Anse between GNSS Ba yICPs progressed and orthomosaics by about from100 m campaigns while it significantly 1 and 2. 4.4.eroded Morphological updrift. ChangesAround the inlet main channel, a new channel appeared, cutting the ebb delta sand 4.4.bank. MorphologicalAt theAlso, scale the of mainChanges the channel whole tidal globally inlet, enlarged the comparative over the studied study ofperiod. these DSMs shows that significant changes At the are scale restricted of the to whole a few tidal areas inlet, only, the while compar the mainative part study of theof these system DSMs remained shows stable that significant within the accuracychanges are of our restricted method to (Figure a few areas12). Among only, while the areas the main where part significant of the system changes remained occurred, stable the terminal within portionthe accuracy of the of sandspit our method of Bonne-Anse (Figure 12). Bay Among progressed the areas by about where 100 significant m while itchanges significantly occurred, eroded the updrift.terminal Aroundportion of the the inlet sandspit main channel,of Bonne-Anse a new Ba channely progressed appeared, by about cutting 100 the m ebbwhile delta it significantly sand bank. Also,eroded the updrift. main channel Around globally the inlet enlarged main channel, over the a studiednew channel period. appeared, cutting the ebb delta sand bank. Also, the main channel globally enlarged over the studied period.

Remote Sens. 2016, 8, 387 13 of 18

Over the main part of the study area, changes are included within the error interval (±0.27 m) (Figure 12), which shows that the mouth of the bay remained globally stable between June and RemoteOctober Sens. 20162015., 8 ,The 387 largest morphological changes took place at the sandspit, with vertical differences13 of 18 Remoteranging Sens. from 2016 ,− 81.77, 387 to + 2.27 m. 13 of 18

Over the main part of the study area, changes are included within the error interval (±0.27 m) (Figure 12), which shows that the mouth of the bay remained globally stable between June and October 2015. The largest morphological changes took place at the sandspit, with vertical differences ranging from −1.77 to + 2.27 m.

FigureFigure 12. 12.Ellipsoidal Ellipsoidal height height difference difference duringduring the studied period period (from (from campaign campaign 1 1to to campaign campaign 2). 2). The results are presented on the orthomosaic of the campaign 2. The shade of red corresponds to The results are presented on the orthomosaic of the campaign 2. The shade of red corresponds to areas areas where erosion occurred and the shade of green corresponds to areas where accretion occurred. where erosion occurred and the shade of green corresponds to areas where accretion occurred. Light Light grey means that the changes are not significant according to the margin of error computed grey means that the changes are not significant according to the margin of error computed previously. previously. Dark grey areas correspond to lack of tie points and white areas to no data. The arrow Dark grey areas correspond to lack of tie points and white areas to no data. The arrow corresponds to corresponds to the location of the topographic profile plotted on Figure 13. theFigure location 12. ofEllipsoidal the topographic height difference profile plotted during on the Figure studied 13 .period (from campaign 1 to campaign 2). The results are presented on the orthomosaic of the campaign 2. The shade of red corresponds to areasA first where result erosion of thisoccurred analysis and th ise shadethe loss of green of sediment corresponds upstream to areas where and accretionthe gain occurred.of sediment Over the main part of the study area, changes are included within the error interval (˘0.27 m) downstream.Light grey The means sedimentary that the changes budget are of notthe sandspitsignificant area according is positive, to the with marg ain gain of error of about computed 13,000 m3. (Figure 12), which shows that the mouth of the bay remained globally stable between June and October Erosivepreviously. areas (yellow, Dark grey orange areas correspondand red in to Figure lack of 12) tie pointslost about and white37,300 areas m3, towhile no data. areas The in arrow accretion 2015. The largest morphological changes took place at the sandspit, with vertical differences ranging (shadescorresponds of green) to earnedthe location about of the50,300 topo graphicm3. The profile cross-shore plotted onprofile Figure of 13. Fi gure 13 illustrates that this fromsediment´1.77 accretion to + 2.27 m.reached about 1.6 m vertically over a distance of 115 m. AA first first result result of thisof analysisthis analysis is the is loss the of loss sediment of sediment upstream upstream and the gainand ofthe sediment gain of downstream. sediment 3 Thedownstream. sedimentary The budget sedimentary of the sandspit budget of area the issandspit positive, area with is positive, a gain of with about a gain 13,000 of about m . Erosive 13,000 m areas3. (yellow,Erosive orange areas (yellow, and red orange in Figure and 12 red) lost in about Figure 37,300 12) lost m3 about, while 37,300 areas m in3, accretion while areas (shades in accretion of green) earned(shades about of green) 50,300 earned m3. The about cross-shore 50,300 m profile3. The cross-shore of Figure 13 profile illustrates of Figure that this13 illustrates sediment that accretion this reachedsediment about accretion 1.6 m verticallyreached about over 1.6 a distance m vertically of 115 over m. a distance of 115 m.

Figure 13. Ellipsoidal height profiles, extracted from the DSM of campaigns 1 (blue) and 2 (red) and difference between both (black).

FigureFigure 13. 13.Ellipsoidal Ellipsoidal height heightproﬁles, profiles, extractedextracted from the DSM DSM of of campaigns campaigns 1 1(blue) (blue) and and 2 2(red) (red) and and differencedifference between between both both (black). (black).

Remote Sens. 2016, 8, 387 14 of 18

5. Discussion The morphological changes of a lagoon-inlet system were monitored over a three-month period for the ﬁrst time by means of UAV-based photogrammetry. Compared to similar studies realized in a coastal environment [17,19], our UAV solution was deployed over a larger surface area (400 ha vs. a few tens of ha) and repetitively, which allowed characterizing topographic changes with a high resolution and quantifying erosion/accretion patterns. This section discusses the relevance of our method compared to classical LiDAR and GNSS methods and also provides an example of interpretation of the observed morphological changes.

5.1. Relevance of the UAV Method Compared to GNSS and LiDAR The vertical accuracy of the method presented in this study was shown to be of the order of 17 cm, which is comparable to LiDAR or TLS surveys but slightly less accurate than GNSS-based surveys [19,31]. Compared to LiDAR surveys, the method proposed here is several orders of magnitude cheaper, while providing two orders of magnitude finer spatial resolution [11–13]. Compared to GNSS surveys, the new method is much faster while providing much denser spatial information, although with a slightly lesser accuracy [31]. This vertical accuracy was calculated from ICPs located in the limit of the area covered by the UAV flight where the number of images decline to values between two and four (Figure7): it is thus expected that our error estimate is conservative. Furthermore, the strong crosswind during the campaign in June generated oblique images unusable during the photogrammetric process. In the next field missions, it will be important to extend the flight plan in order to have a better recovery of peripheral areas. The comparison between DSMs of the two main campaigns and LiDAR from 2010 confirms the quality of the DSMs derived from the UAV images (Figure 10). Lagoon-inlet systems correspond to particular environments where access is difficult due to the presence of many channels. The solution used in this study allows computing a DSM at very high resolution while being non-intrusive. In addition, classical LiDAR and GNSS topographic surveys, which usually allow computing DSMs with a spatial resolution in the range of 1–10 m, would offer a very rough representation of the largest tidal dunes imaged with our UAV solution (Figure8) while the smallest bedforms would not be represented at all. This decisive improvement would, for instance, allow investigating residual sediment transport, which is indicated by the slip face of the dunes. Another perspective would be to improve the parameterization of bottom friction in hydrodynamic numerical models based on the dimension of these bedforms.

5.2. Interpretation of the Morphological Changes at the Sandspit As shown in the previous section, the main morphological change during the surveyed period concerns the development of the sandspit by more than 100 m. Even if GCPs are absent close to the erosion area (to the south of the sandspit, shade of red color on Figure 12), it can be assumed that the accuracy of the DSM in this area is similar to that computed previously. Indeed, the image recovery on this area is up to six images for both campaigns, suggesting that the photogrammetry process was accurate. This sediment accretion roughly balances the erosion that occurred updrift, although the sediment balance remains positive over this three-month period. While the development of sandspits is well documented and explained by the presence of longshore currents driven by the breaking of oblique waves, the amount of sand moved for a summer period remains exceptional for the study area. Indeed, Chaumillon et al. [7] showed that, during the summer period, offshore swells tend to be smaller, of shorter period and originate from the northerly direction so that the study area was exposed to residual waves with signiﬁcant height in the range of 0.1–0.5 m. Such small waves would not be energetic enough to move volumes of sand of the order of 50,000 m3. In order to better understand this surprising behavior, we analyzed a time series of wave height measured offshore of the study area (Figure1, 1.83 ˝W; 45.91˝N, water depth = 50 m) over 2014–2015. Figure 14a reveals the occurrence Remote Sens. 2016, 8, 387 15 of 18

Remote Sens. 2016, 8, 387 15 of 18 of several events inducing wave heights larger than 5 m in August and September, which is rather localuncommon effect of for these the study uncommon area and storm corresponds waves in to summer, winter conditions we extended [16]. the To evaluatewave hindcast the local presented effect of inthese Chaumillon uncommon et al. storm [7] up waves to December in summer, 2015. we This extended computat the waveion revealed hindcast firstly presented the occurrence in Chaumillon of twoet eventsal. [7] up in toAugust December and 2015.September, This computation characterized revealed by waves firstly of the significant occurrence height of two exceeding events in 1.5 August m in frontand September, of the inlet, characterized which usually by waves occurs of in significant winter. We height also exceeding forced the 1.5 msimple in front Coastal of the Engineering inlet, which Researchusually occurs Center in winter.(CERC) We[32] also longshore forced the transpor simple Coastalt formula Engineering with wave Research parameters Center at (CERC) breaking [32] extractedlongshore from transport this hindcast formula and with followed wave parametersthe methodology at breaking described extracted in Bertin from [33] this (Figure hindcast 14b). The and cumulatedfollowed the longshore methodology transport described was estimated in Bertin [ 33to] be (Figure 31,000 14 b).m3 Theover cumulated the studied longshore period, transportwhich is aboutwas estimated three times to belarger 31,000 compared m3 over to the classical studied summers. period, which Indeed, is the about cumula threeted times longshore largercompared transport overto classical the studied summers. period Indeed, represents the cumulated about 20% longshore of the yearly transport value, overwhile the the studied summer period contribution represents of theabout yearly 20% longshore of the yearly transport value, whileis usually the summerless than contribution 10% considering of the the yearly 36-year longshore hindcast transport of Bertin is [33].usually However, less than morphological 10% considering changes the 36-year are controlled hindcast of by Bertin the divergence [33]. However, of sand morphological fluxes and changesnot the potentialare controlled longshore by the transport divergence as of computed sand fluxes here. and notIn the potentialpresent case, longshore our computation transport as computedcan only explainhere. In that the presentthe sediment case, ouraccretion computation that occurred can only at explainthe tip of that the thesandspit sediment results accretion from a that combination occurred ofat the tiplongshore of the sandspit transport results that takes from place a combination along the ofwhole the longshore spit and the transport erosion that area takes that placedeveloped along immediatelythe whole spit updrift. and the The erosion physical area processes that developed responsible immediately for these morphological updrift. The physicalchanges processeswill have toresponsible be analyzed for thesein detail morphological using the process-based changes will havemodeling to be system analyzed under in detail deve usinglopment the in process-based our lab [34]. Themodeling detailed system morphological under development changes evidenced in our lab from [34 ].our The UAV detailed image morphological mapping will changesbe of great evidenced help to validatefrom our our UAV modeling image mapping system. will be of great help to validate our modeling system.

Figure 14. ((aa)) SigniﬁcantSignificant wave wave height height measured measured offshore offshore of Oléronof Oléron Island Island (red) (red) and simulatedand simulated in front in frontof the of inlet the (blue) inlet between(blue) between July 2014 July and 2014 December and December 2015; (b) 2015; Longshore (b) Longshore transport estimatedtransport withestimated wave withparameters wave parameters at breaking at extracted breaking from extracted the wave from hindcast. the wave hindcast.

6. Conclusions Conclusions This studystudy demonstrateddemonstrated the the potential potential of of UAV UAV methods methods using using a photogrammetry a photogrammetry approach approach (SFM (SFMalgorithm) algorithm) to monitor to monitor tidal inlet tidal environments. inlet environments. This solution This allowssolution surveying allows surveying inter- and supra-tidalinter- and supra-tidaltopography topography with a non-intrusive with a non-intrusive method, overmethod, a large over geographic a large geographic area (400 area ha) (400 with ha) very with dense very densespatial spatial information information (4.6 cm horizontal(4.6 cm horizontal resolution). resolution). From three From campaigns three spanningcampaigns a three-monthspanning a three-monthperiod, DSMs period, and orthomosaics DSMs and orth wereomosaics produced. were GCPs produced. were used GCPs to georeferencewere used to the georeference DSMs and the DSMscomparison and the with comparison GNSS data with (proﬁle GNSS and data ICPs) (profile demonstrated and ICPs) the demonstrated good accuracy the of our good method, accuracy with of a ourroot method, mean square with discrepancya root mean lower square than discrepancy 17 cm, which lower is slightlythan 17 better cm, which than thatis slightly of the LiDARbetter than data thatacquired of the in LiDAR this area data in 2010.acquired Based in onthis this area accuracy, in 2010. a Based differential on this map accuracy, was computed, a differential which map revealed was computed,that the main which part revealed of the study that area the remainedmain part stable, of the although study area the remained inlet channel stable, tended although to enlarge the inlet and channelthe sandspit tended prograded to enlarge by more and thanthe 100sandspit m. The prograded very detailed by morphologicalmore than 100 changes m. The revealed very detailed by our morphological changes revealed by our method will be very useful to further analyze the causes for this sandspit accretion. In addition, our method allows for the mapping of tidal dunes up to a wavelength of the order of 1 m and a height of 0.1 m, which opens new perspectives to

Remote Sens. 2016, 8, 387 16 of 18 method will be very useful to further analyze the causes for this sandspit accretion. In addition, our method allows for the mapping of tidal dunes up to a wavelength of the order of 1 m and a height of 0.1 m, which opens new perspectives to understanding residual sand fluxes on sand banks and also to better parameterize bottom friction in process-based models. The methodology developed here was shown to be suitable for such a coastal area subjected to a strong tidal influence, which promotes the development of bedforms almost everywhere. These bedforms provides a large density of visual landmarks on most images, which optimizes the generation of tie points in the SFM algorithm. The efficiency of this methodology will therefore have to be verified for wave-dominated beaches or eolian dunes where lower contrast and fewer landmarks may be found. A possible improvement would be to pre-process images in order to increase their contrast and therefore enhance the photogrammetric process. Also, an algorithm will be developed to identify and automatically mask water areas, a process that was done manually in this study.

Acknowledgments: This study was conducted in the scope of the project DYNAMO, funded by the French Agency for Research (Grant agreement n˝ ANR-12-JS02-00008-01) and the project EVEX, funded by Region Poitou-Charentes in the framework of the Regional Chair program. Antoine Dumon and Nicolas Lachaussée are acknowledged for their help on the field and Médéric Gravelle provided valuable advices to process GNSS data. Author Contributions: All authors contributed in a substantial way to the manuscript. N.L. and X.B. proposed and conceived the experiments; N.L., B.G., F.P. and X.B. performed the field work; B.M. analyzed the data; All authors contributed to the drafting of the paper; N.L. supervised the study at all stages. Conflicts of Interest: The authors declare no conflict of interest.

References

1. Thieler, E.R.; Danforth, W.W. Historical Shoreline Mapping (I): Improving Techniques and Reducing Positioning Errors. J. Coast. Res. 1994, 10, 549–563. Available online: http://www.jstor.org/stable/4298252 (accessed on 15 December 2015). 2. Quartel, S.; Addink, E.A.; Ruessink, B.G. Object-oriented extraction of beach morphology from video images. Int. J. Appl. Earth Obs. Geoinf. 2006, 8, 256–269. [CrossRef] 3. Pardo-Pascual, J.E.; Almonacid-Caballer, J.; Ruiz, L.A.; Palomar-Vazquez, J. Automatic extraction of shorelines from Landsat TM and ETM+ multi-temporal images with subpixel precision. Remote Sens. Environ. 2012, 123, 1–11. [CrossRef] 4. Ford, M. Shoreline changes interpreted from multi-temporal aerial photographs and high resolution satellite images: Wotje Atoll, Marshall Islands. Remote Sens. Environ. 2013, 135, 130–140. [CrossRef] 5. Ryu, J.H.; Won, J.S.; Min, K.D. Waterline extraction from Landsat TM data in a tidal ﬂat: A case study in Gomso Bay, Korea. Remote Sens. Environ. 2002, 83, 442–456. [CrossRef] 6. Maiti, S.; Bhattacharya, A.K. Shoreline change analysis and its application to prediction: A remote statistics based approach. Mar. Geol. 2009, 247, 11–23. [CrossRef] 7. Chaumillon, E.; Ozenne, F.; Bertin, X.; Long, N.; Ganthy, F. Wave climate and inlet channel meander bend control spit breaching and migration of a new inlet: La Coubre Sandspit, France. J. Coast. Res. 2014, 70, 109–114. [CrossRef] 8. Long, N.; Pouget, F.; Bellec, A.; Chaumillon, E.; Bertin, X. Automatic extraction of shorelines from multi-temporal and very high resolution images to monitor coastal ecosystem. In Proceedings of the 5th Geographic Object-based Image Analysis Conference, Thessaloniki, Greece, 21–24 May 2014; Volume 3, pp. 245–250. 9. Perko, R.; Raggam, H.; Gutjahr, K.; Schardt, M. Assessment of the mapping potential of Pleiades stereo and triplet data. In Proceedings of the ISPRS conference on ISPRS Annals of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Zurich, Switzerland, 5–7 September 2014; Volume II-3, pp. 103–109. 10. Poli, D.; Remondino, F.; Angiuli, E.; Agugiaro, G. Evaluation of Pleiades-1A triplet on Trento testﬁeld. In International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Proceedings of the ISPRS Hannover Workshop, Hannover, Germany, 21–24 May 2012; Volume XL-1, pp. 287–292. Remote Sens. 2016, 8, 387 17 of 18

11. White, S.A.; Wang, Y. Utilizing DEMs derived from LIDAR data to analyze morphologic change in the North Carolina coastline. Remote Sens. Environ. 2003, 85, 39–47. [CrossRef] 12. Nagihara, S.; Kevin, R.; Mulligan, R.; Xiong, W. Use of the tree-dimensional laser scanner to digitally capture the topography of sand dunes in high spatial resolution. Earth Surf. Process. Landf. 2004, 29, 391–398. [CrossRef] 13. Letortu, P.; Costa, S.; Maquaire, O.; Delacourt, C.; Augereau, E.; Davidson, R.; Suanez, S.; Nabucet, J. Retreat rates, modalities and agents responsible for erosion along the coastal cliffs of Upper Normandy: The contribution of terrestrial laser scanning. Geomorphology 2015, 245, 3–14. [CrossRef] 14. Castelle, B.; Marieu, V.; Bujan, S.; Splinter, K.D.; Robinet, A.; Sénéchal, N.; Ferreira, S. Impact of the winter 2013–2014 series of severe western Europe storms on a double barred sandy coast: Beach and dune erosion and megacusp embayments. Geomorphology 2015, 238, 135–148. [CrossRef] 15. Taveira-Pinto, F.; Silva, R.; Barbosa, J. Coastal Erosion Along the Portuguese Northwest Coast Due to Changing Sediment Discharges from Rivers and Climate Change. In Global Change and Baltic Coastal Zones; Schernewski, G., Hofstede, J., Neumann, T., Eds.; Global Change and Baltic Coastal Zones, Coastal Research Library-Series; Springer: Dordrecht, The Netherlands, 2011; Volume 1, pp. 135–151. 16. Bertin, X.; Castelle, B.; Chaumillon, E.; Butel, R.; Quique, R. Estimation and inter-annual variability of the longshore transport at a high-energy dissipative beach: The St Trojan beach, SW Oléron Island, France. Cont. Shelf Res. 2008, 28, 1316–1332. [CrossRef] 17. Gonçalves, J.A.; Henriques, R. UAV photogrammetry for topographic monitoring of coastal areas. ISPRS J. Photogramm. Remote Sens. 2015, 104, 101–111. [CrossRef] 18. Casella, E.; Rovere, A.; Pedroncini, A.; Mucerino, L.; Casella, M.; Cusati, L.A.; Vacchi, M.; Ferrari, M.; Firpo, M. Study of wave runup using numerical models and low-altitude aerial photogrammetry: A tool for coastal management. Estuar. Coast. Shelf Sci. 2014, 149, 160–167. [CrossRef] 19. Mancini, F.; Dubbini, M.; Gatelli, M.; stecchi, F.; Fabbri, S.; Gabbianelli, G. Using Unmanned Aerial Vehicles (UAV) for the high-resolution reconstruction of topography: the Structure for Motion approach on coastal environments. Remote Sens. 2013, 5, 6880–6898. [CrossRef] 20. Haala, N. Comeback of Digital Image Matching. In Photogrammetric Week; Wichmann Verlag: Heidelberg, Germany, 2009; pp. 289–301. 21. Colomina, I.; Molina, P. Unmanned aerial systems for photogrammetry and remote sensing: A review. ISPRS J. Photogramm. Remote Sens. 2014, 92, 79–97. [CrossRef] 22. Klemas, V.V. Coastal and environmental remote sensing from unmanned aerial vehicles: On overwiev. J. Coast. Res. 2015, 31, 1260–1267. [CrossRef] 23. Jutzi, B.; Weinmann, M.; Meidow, J. Weighted data fusion for UAV-borne 3D mapping with camera and line laser scanner. Int. J. Image Data Fusion 2014, 5, 226–243. [CrossRef] 24. Stødle, D.; Borch, N.; Solbø, S.; Storvold, R. High performance visualisation of UAV sensor and image data with raster maps and topography in 3D. Int. J. Image Data Fusion 2014, 5, 244–262. [CrossRef] 25. Clapuyt, F.; Vanacker, V.; Van Oost, K. Reproducibility of UAV-based earth topography reconstruction based on Structure-from-Motion algorithms. Geomorphology 2015. in press. [CrossRef] 26. Robertson, D.P.; Cipolla, R. Structure from Motion, in Practical Image Processing and Computer Vision; Varga, M., Ed.; John Wiley & Sons: Milton, Australia, 2009; Chapter 13; p. 24. 27. Ullman, S. The Interpretation of Structure from Motion; The Royal Society: London, UK, 1979; p. 35. Available online: http://publications.ai.mit.edu/ai-publications/pdf/AIM-476.pdf (accessed on 15 December 2015). 28. Deseilligny, M.P.; Clery, I. Apero, an open source bundle adjustment software for automatic calibration and orientation of set of images. In Proceedings of the ISPRS International Workshop on 3D Virtual Reconstruction and Visualisation of Complex Architectures, Trento, Italy, 2–4 March 2011; pp. 269–276. 29. Agisoft LLC. Agisoft Photoscan User Manual, Professional Edition, Version 0.9.0; Agisoft LLC: St. Petersburg, Russia, 2012. 30. Takasu, T. RTKLIB Version 2.4.2 Manual 2013; p. 183. Available online: http://www.rtklib.com/prog/ manual_2.4.2.pdf (accessed on 10 February 2016). 31. Lee, J.M.; Park, J.-Y.; Choi, J.-Y. Evaluation of su-aerial topographic surveying techniques using total station and RTK-GPS for application in macrotidal sand beach environment. J. Coast. Res. SI 2013, 65, 535–540. [CrossRef] Remote Sens. 2016, 8, 387 18 of 18

32. CERC. Shore Protection Manual, 4th ed.; US Army Engineer Waterways Experiment Station, Coastal Engineering Research Center, US Government Printing Ofﬁce: Washington, DC, USA, 1984. 33. Bertin, X. Dynamique Hydro-Sédimentaire Des Côtes Dominées par la Houle et par la Marée. Habilitation Thesis to Coordinate Research Activities; La Rochelle University: La Rochelle, France, 2015; p. 135. 34. Guérin, T.; Bertin, X.; Dodet, G. A numerical scheme for coastal morphodynamic modelling on unstructured grids. Ocean Modell. 2016. accepted. [CrossRef]

© 2016 by the authors; licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC-BY) license (http://creativecommons.org/licenses/by/4.0/).