Coastal zone monitoring is an important task in environmental protection, while coastline detection is fundamental for coastal management [1
]. Coastal management requires up-to-date accurate information since coastal movements are of primary importance in evaluating coastal erosion. Remote sensing plays a significant role in coastal observation as it is one of the most valuable tools used for detecting and monitoring coastlines. The scope of this paper is to detect the coastline and to identify the coastline zones by applying geographic object-based image analysis (GEOBIA) to high-resolution orthophotos produced by an unmanned aerial vehicle (UAV) combined with a structure from motion (SfM) algorithm, and to reconstruct the study areas in a three-dimensional (3D) representation to examine if other beach characteristics, such as swash zones, wrack lines, and berm zones, can be identified.
Although the spatial resolution of satellite imagery has significantly improved in the last decade, the data collected is still not sufficient for medium to small coastal changes (centimetric accuracy). Conversely, UAVs in combination with SfM, which is also called a UAV-SfM pipeline, can provide high-resolution information at a low cost for a small area [2
]. The miniaturization of sensors and the increase in the flight capabilities and agility of UAVs, as well as the high-quality imagery tools that are available have combined to produce a spatial data acquisition tool for environmental monitoring [4
]. Several recent publications have described methods and techniques that measure changes in coastal morphology zones using UAVs [4
]. Currently, UAVs are a viable option for collecting remote sensing data for a wide range of practical applications, including scientific, agricultural and environmental applications [4
] Unmanned aerial systems (UAS), which consist of a UAV and a sensor, provide digital images with spatial and temporal resolutions that are capable of overcoming some of the limitations of spatial data acquisition using satellites and airplanes. The resolution of the data commonly available using conventional platforms, such as satellites and manned aircraft, is typically in the range of 20 to 50
]. For example, the acquired spatial data resolution via a manned aircraft ranges from 10-100 cm when the satellite system provides a resolution greater than 50 cm [12
Datasets produced by UAV-based remote sensing have such a high spatial resolution (2–5 cm) that characteristics and changes of the landscape, such as coastal morphology, coastal zones, and beach morphological characteristics, can be mapped in detail in two (2D) and three dimensions (3D) [3
]. However, such small changes are not distinguishable at the spatial resolutions generally obtained using manned aircraft and satellite systems. Furthermore, the high level of automation, the ease of deployment, the ease of survey repeatability, and the low running costs of UAVs in comparison with other traditional remote sensing methods, allows for frequent missions that provide spatial datasets with a resolution of less than 5 cm and a high temporal repetition due to the ease of survey deployment [12
]. Georeferenced orthophotos and digital surface models (DSMs) are used to measure and depict the morphology of a beach in 2D and 3D, which allows the assessment of changes to the beach due to extreme wave phenomena [6
Recent advances in computer vision include SfM algorithms, which have been successfully used for the reconstruction of large uncontrolled photo collections [24
]. Studies have described the adaptation of SfM techniques to process UAV data photography to create an accurate georeferenced 3D point cloud and to generate a high-resolution DSM; SfM techniques have also been used in the orthorectification of georeferenced images that have been joined together to form a mosaic [6
Hay and Castilla stated in 2008 that geographic object-based image analysis (GEOBIA) is a new discipline that involves the object-based analysis of Earth’s surface using remote sensing imagery. They defined GEOBIA as “as a sub-discipline of Geographic Information Science (GIScience) devoted to developing automated methods to partition remote sensing imagery into meaningful image-objects, and assessing their characteristics through spatial, spectral and temporal scales, so as to generate new geographic information in GIS-ready format
]. Using GEOBIA for coastline zone extraction involves exporting the geographic information in vector format with the classes of water and beach zones assigned through an automatic procedure [28
The present work uses the UAV-SfM pipeline for spatial data acquisition, high-resolution geoinformation production, and 3D visualization of coastal morphology assisted by GEOBIA for the identification of coastal zones. More specifically, the proposed methodology uses GEOBIA of high-resolution orthophotos for the definition of coastal zones. The combination of these innovative techniques is applied in two diverse coastal areas to examine the efficiency of the developed method.
The potential use of the UAV-SfM pipeline has been evaluated in several studies [5
]. However, references about the use of GEOBIA with the geographic information produced by the UAV-SfM methodology for coastal monitoring are still scarce in the recent scientific bibliography.
A coastline is defined as the line that forms the boundary between the land and the sea; the detection of a coastline in high resolution includes recognizing several discontinuities [30
]. However, the boundary between water and land is not easily distinguishable in clear water. For the proper study of coastal areas, the four typical beach zones should be examined. These beach zones include the swash zone, beach face, wrack line, and berm. The swash zone is the area of land that is alternately covered by the sea or exposed to wave runup. The beach face includes the sloping section of land that is exposed to the swash of the waves. The wrack line is the highest reach of the daily tide. The berm is the area of land that is almost horizontal; it is the dry portion of the coast that usually contains sand dunes.
3. Results and Discussion
The UAV-SfM pipeline was performed for Eressos beach and Neapoli beach. All the geoinformation that was produced was referred to the GGRS87 / Greek Grid (EPSG: 2100). The dense point clouds produced for Eressos beach and Neapoli beach consisted of 456.7 and 581.171
, respectively. The accuracy of the geoinformation produced was evaluated by calculating the residuals, the root sum of the squares error (RSSE), and the root mean square errors (RMSE) for all GCPs used in both survey missions. The produced orthophoto pixel resolution was 2.34 and 2.41
for Neapoli beach and Eressos beach, respectively. Furthermore, the produced DSM pixel resolution from the Neapoli survey mission was 4.35
, while the DSM pixel resolution from the Eressos survey was 4.68
. The GCPs error statistics for the Eressos survey mission are summarized in Table 1
. The GCPs error statistics for the Neapoli survey mission are presented in Table 2
. The RSSE for all GCPs in the Eressos survey mission was from 1 to 7 cm; the RSSE for the Neapoli survey mission was found from 3 to 16 cm. The UAV-SfM pipeline accuracy and estimated error for both surveys are summarized in Table 3
. The orthophoto and DSM RMSE for both cases was calculated as 0.187 and 0.585 pixels, respectively. The difference in accuracy between the two survey missions is due to the different spatial distribution and the number of GCPs used for georeferencing the orthophoto maps and DSM. These differences could be attributed to the use of 14 GCPs in the Erresos survey compared with the 8 GCPs used in the Neapoli survey. Furthermore, the low quality of the camera lens increased the accuracy error due to high image residuals. However, the achieved accuracy met the requirements for creating highly detailed 2D/3D visualizations and for defining coastal zones, which was the main focus of this study.
The generated orthophoto from the Neapoli study site had a resolution of 2.41 cm (Figure 6
). The inset in Figure 6
illustrates the swash zone, wrack lines, and berm zone in 3D. The level of detail of the generated 3D visualizations clearly shows the structure of the coastal zones. Additionally, the DSM produced by the workflow presented in this paper had a spatial resolution of
cm for both beaches. Figure 7
shows the orthophoto and dsm for Eressos beach.
, Figure 7
and Figure 8
show the high degree of detail that was achieved using the UAV-SfM methodology. Aside from the coastline, small variations, such as small rocks and small benches on the beach, are visible over the whole extent of the mapped area (refer to the inset in Figure 6
and Figure 8
Several important beach characteristics, including the swash zone, beach face, wrack line, and berm, are visible in high-resolution orthophoto maps, but are not visible in high-resolution satellite images. Orthophotomaps cannot be processed using the remote sensing pixel-based classification methodologies (e.g., nearest neighbour) because spectral representatives are similar for the different classes. The segmentation of the orthophoto maps is a critical step since the characteristics of various objects can be used in the classification process. Since the segmentation size, shape, and compactness determine the geometry of an object in the multiresolution segmentation, the characteristics should be carefully assessed. Water clarity and wave state are causes of coastline misclassification. Coastline detection in orthophoto maps with a spatial resolution of centimeters depends on the complexity of the beach. In cases with a simple structure, such as a sand beach with a large face, the detection of the swash zone and beach face can be achieved with high accuracy (Figure 9
). However, in complicated cases where the beach is a mixture of composition, such as shingles with various sizes, dead algae, tree shadows, and erosion gaps, the detection of the coastline is more difficult (Figure 10
). DSMs can help in the determination of vectors used to describe beach characteristics.
The first case study, Eressos beach, is a large sandy beach affected by large north waves. Figure 9
shows an example of the developed methodology used for a small portion of the beach. Using the image, objects could be classified into four zones according to their spectral characteristics. The most seaward zone is the swash zone where the wave runup makes the sand wet and, therefore, easily distinguishable from the rest of the zones. The next three zones belong to the beach face, although each one has different characteristics. The green zone represents an old swash zone (perhaps from the previous day) or an extreme wave runup due to large waves. The brown zone is a result of large waves from the previous days/weeks and consists of small pebbles and dead shells. The last purple zone of the beach face consists only of sand, and it is the result of large waves during winter. Several other zones are visible up to the northern part of the beach where the berm area and the sand dunes begin.
The second case study, Neapoli beach (Figure 10
), is a small beach with a complicated environment where organic and inorganic debris is deposited by wave action. In this case, the beach zones are not easily detected due to tree shadows and dead algae formations. However, four zones can be classified and manually corrected where necessary. Shown in black are the benching shadows, which are planes at different heights created by erosion during large wave effects. There are three different planes detected between the sea and the road network. Next, the old swash zone is purple. The swash zone is interrupted by tree shadows and is a mixture of areas covered by dead algae (red) and bare soil with pebbles (pink). The mixture of sand and gravel is shown in gray. In this situation, an expert photo interpretation can help with the classification to obtain reliable results in terms of coastline detection and zone characterization.
As this is the first attempt to identify coastal (beach) zones in the literature, no comparisons can be made with previous studies. Our classification should be compared with the classification determined using measurements, such as DGPS points across the zone and/or a terrestrial laser scanner, that have the same or better resolution than the orthophoto that we used. A laser scanner was not been included in this study, although the use of one is planned for future works. DGPS points were selected, but only for calculating the accuracy of the orthophoto map and were mainly placed inside the zones and not on the borders. However, since the accuracy of the orthophoto is 2–3 cm zones, it is expected to have the same spatial displacement. This work aims to create a reference map of the costal zones that will be used as a base to compare other lower resolution data, such as high-resolution satellite images of 0.30 cm to 1 m resolution, or cameras that have spatial variability across a line of view.
Although only two small-scale areas were studied, it is clear that UAV-based remote sensing bridges the gap in scale and resolution between ground observations and imagery acquired from conventional manned aircrafts and satellite sensors. Additionally, the computer vision SfM and MVS techniques supplement the need for accurate geodata for 2D and 3D analysis. UAV-SfM derived products can become an important and accurate tool for analyzing changes in landscape and landscape erosion in various spatial and temporal scales.
As the resolution and precision of the orthophoto maps and 3D visualizations cannot be achieved from satellite datasets and Laser Imaging, Detection and Ranging (LIDAR) high-resolution data is not yet available for Lesvos Island, the UAV-based data acquisition provides an efficient and rapid framework for remote sensing of coastal morphology.
In this paper, we used a UAV-SfM technique to generate accurate, high-resolution geoinformation and 3D geovisualizations of two study areas. Furthermore, GEOBIA was applied to the geoinformation to identify coastal zones in two different beaches. UAVs provide a platform for close range aerial photography using small sensors and can become a useful tool for spatial and temporal coastal observation and environmental monitoring, as well as monitoring erosion in coastal zones. This study applied SfM techniques combined with computer vision algorithms to aerial photos acquired from a quad-rotor UAV to determine the 3D representations of two coastal sites in Lesvos, Greece. The two sites were chosen because the coastlines are gradually eroding due to extreme winds and wave conditions. The speed and height at which the UAV flew and the use of the SfM technique allowed for the capture of detailed representation of coastal features, such as hollows, small sand dunes, swash zones, and beach faces, on a centimeter scale.
The results showed that a georeferenced point cloud, orthophoto, and DSM with subpixel RMSE cm can be obtained from imagery acquired using a compact camera attached to a quad-rotor UAV flying 100 m AGL. The image acquired had the required spatial resolution, a high level of survey automation, and a high survey repeatability for mapping and monitoring natural coastal phenomena and changes. Additionally, UAVs are capable of obtaining spatial data for inaccessible areas and allowing the easy repetition of a survey.
The UAV-SfM pipeline uses a combination of spatial data acquisition from a UAV and computer vision and image processing techniques to automate the detection procedures and to create in-situ measurements with large spatial cover for monitoring a variety of environmental phenomena.
This study demonstrated how the combination of two new remote sensing technologies in the form of UAV-SfM and GEOBIA methods can be successfully combined to map coastline zones and to create accurate 3D geovisualizations of coastal morphology. The use of UAVs in combination with SfM methodology to create accurate 2D and 3D geoinformation can also be applied to coastal surveying.
The main conclusion derived from this study is that the high spatial resolution geoinformation produced from a UAV-SfM pipeline in combination with GEOBIA creates new structural information for coastal monitoring. Specific vectors for the coastline, swash zone, wrack lines, berm crests, swash limits, and berm areas can be detected as a result of the wave runup in different time frames. Thus, coastal geomorphologists that require accurate topographic information of the coastal systems to perform a reliable simulation and mapping of coastal erosion can use the spatial data produced by the proposed methodology. Furthermore, the management of coastal areas, port installations, and coastal energy projects are areas that could also use such methodology. Further studies of this methodology should include infrared cameras to correct for sun glint and hyperspectral sensors for deriving spectral signatures of debris.