The cumulative human-induced threats on marine biodiversity and ecosystem goods and services are increasing worldwide while habitat loss and degradation, pollution, overfishing and climate change effects are classified among the most important threats that affect the marine realm [1
]. Yet, the successful and effective protection and conservation of the marine ecosystem largely depends on our knowledge regarding the spatial extent, the geographical range and the ecological characteristics of the resource or the habitat of interest [3
]. Thus, accurate and high-resolution habitat mapping can be considered as a prerequisite in spatial marine management and conservation planning. In recent years the urgent need for accurate habitat maps of the seafloor has been largely acknowledged within the framework of the relative legislation such as the European Habitats Directive (92/43/EEC) and Marine Strategy Framework Directive (MSFD; 2008/56/EC). Likewise, marine spatial planning, systematic conservation planning, biodiversity and marine resources management also demand good and comprehensive spatio-temporal knowledge of the seascape and the habitat types [4
Swath mapping echosounder is the preferred and most efficient way to rapidly map benthic habitats at all ranges and depths [6
]. A significant part of the survey efforts rests on acquiring acoustic backscatter intensity of the seafloor with sidescan sonars (SSS) and, more recently, multibeam echosounders (MBES). Variations in surveying approaches limit repeatability and comparability of the mapping results, but MBES offers the opportunity of building standardized procedures for automated seafloor mapping [9
]. There is a growing literature that makes use of MBES for benthic habitat mapping [11
], taking advantage of the wide spectrum of information that it provides, including co-registered backscatter and bathymetry and angular backscatter intensity. Bathymetric and backscatter texture features are long being used for quantitative analysis of the seabed components [13
], but also Backscatter Angular Response Analysis is gaining popularity over the years and tends to be a popular procedure for seafloor characterization [15
]. Among the disadvantages of traditional MBES is the relatively low resolution that they provide, although newer systems (e.g., interferometric sonars) tend to compete with SSS systems in spatial sample density in shallow waters. The above limits severely the backscatter texture content of the recorded seafloor, while angular responses are traditionally analyzed in regard to a full swath strip [9
], limiting the spatial density of the extracted statistical parameters. In addition, MBES systems lack in seafloor coverage in shallow waters, as their swath width is proportional (3–4 times) to the depth. Significant technology advances of MBES rest on standardization of the collection and processing of calibrated backscatter data and the introduction of multispectral backscatter [9
The acoustic signature of seagrass, and in particular Posidonia oceanica
, is well known, e.g., [22
], modulated by the efficiency of photosynthesis [24
], i.e., related to the time of day and season at which the seagrass was mapped and how individual responses from algae combine over meadows of different densities [25
]. It is also affected by gas content [26
] and general health [27
], with clear differences between frequencies and imaging angles [28
]. Information necessary to reliably identify seagrass, reefs and thin or thick sedimentary layers is limited by the angle of imaging, the spatial resolution and other intrinsic parameters such as beam aperture and pulse length but most importantly by the frequency e.g., [29
]. However, the frequency dependent visibility of benthic habitats, including Posidonia
, is not yet known and experiments on multi-frequency dependencies on benthic habitats is a recent field of evolving research, with recent examples being [30
]. As a general rule, the choice of imaging frequency not only directly influences the resolution of the system and the time it takes to map an area, but it is also related to the acoustic response (backscatter intensity and texture) of a seafloor type.
Development of multispectral MBES, although highly promising, is still in its infancy and therefore multi-frequency datasets are still mostly acquired by successive seafloor scans of the same area with different transducers, which is generally operationally inefficient or even impossible [21
], and/or on using multiple MBES systems on a single survey [20
]. On the contrary, the majority of modern SSS have a dual-frequency capacity and so they can offer a multispectral perspective to seafloor mapping [34
]. Recently, SSS systems with the capability of simultaneously producing three frequencies have been developed and demonstrated [36
]. Additionally, SSS systems offer much higher resolution backscatter images of the seafloor, suitable for detailed habitat mapping through texture analysis [11
]. SSS can acquire data from very high grazing angles, not limited to specific beam angles as MBES are. This makes it more suitable for shallow waters. What is considered an important drawback of SSS is its lack of secondary information regarding bathymetry and angular dependency of returning echoes. This way, SSS measurements, as opposed to MBES, cannot be accurately corrected for radiometric and geometric artifacts [40
], and it is also hard to investigate angular backscatter dependencies for habitat discrimination.
As MBES advances lead to vastly improved maps of benthic habitats, MBES tends to replace SSS for seafloor mapping. Although there is work comparing these two systems [11
], there seems to be a research gap concerning if and how SSS and MBES could be used jointly to complement each other in terms of spatial and spectral resolution and mapping coverage, especially in shallow habitats. The present work uses a single frequency MBES and a dual frequency SSS spanning three frequencies (100, 180 and 400 kHz) during a single survey. Texture analysis is applied to each frequency’s respective backscatter mosaic, thus extracting a range of textural features from each. In addition, the MBES bathymetry is processed to extract bathymetric features, while well accepted approaches are applied for the extraction of statistics on angular backscatter response curves (ARCs). The above features constitute the basis for building supervised seabed classifications. All the pieces of the above dataset needed to be analyzed under a common scale, achieving stationary information in all extents of the area. This common scale was achieved through a segmentation process on the 100 kHz mosaic (being the only full seabed coverage dataset), which provided the means for an Object Based Image Analysis (OBIA). This way, regardless of the different data densities and non-stationarity of some data sources, all their features can be joined into common minute regions covering the full seafloor scene.
This work is an effort to exploit the synergistic use of MBES and SSS as complementary sources of information for improved benthic habitat mapping with emphasis on Mediterranean seagrass habitats. By weighing their relative importance for habitat discrimination, we identify optimal mapping strategies, informing future surveys and extending interpretation of the huge amount of existing datasets. At the same time, a high-detail habitat map of a very important Mediterranean MPA, the National Marine Park of Zakynthos (NMPZ), is presented for the first time.
The remainder of this paper is organized as follows: Section 2
describes the ecological and geological settings of the study area (NMPZ); Section 3
describes the data acquisition materials and methods and the detailed benthic habitat cartographic work derived by the expert interpretation of the available data. Section 4
, Section 5
, Section 6
and Section 7
present the data processing methods, regarding the feature extraction and supervised seafloor classification approaches. Finally, Section 8
provides an in-depth analysis of the results along with a discussion for general guidelines and future perspectives.
2. Study Area
The National Marine Park of Zakynthos (NMPZ) (Figure 1
) represents the first marine protected area (MPA) in the Mediterranean for the protection of sea turtles, including no-take zones. It was established in 1999 with the primary goal of protecting and managing one of the most important nesting rookeries of the loggerhead sea turtles Caretta caretta
in the region [41
]. The MPA mainly encompasses four habitat types that are all included in the EU Habitats Directive 92/43/EEC; namely, Posidonia oceanica
(EU habitat code—EUhc: 1120—“Posidonia beds”), rocky reefs (EUhc: 1170—“Reefs”), Sandbanks (EUhc 1110), soft substrates (EUhc 119A—“Unvegetated soft bottoms” and 119B—“Vegetated soft bottoms”) and marine caves (EUhc: 8330) westernmost of the NMPZ [42
]. Previous research efforts on marine habitats in the MPA of the NMPZ were rather limited [42
]; they predominantly mapped seagrass, especially Posidonia oceanica
]. The offshore geology of the NMPZ area is characterized by diapiric salt intrusions (salt domes) forming extensive reefs, up to 20 m higher than the soft bottom, as shown in the chirp SBP stratigraphy of Figure 2
. As in many cases worldwide, Zakynthos’ salt diapirs are associated with a wider hydrocarbon province, apparent in various free hydrocarbon seeps in the area off and on shore [46
]. The same SBP profiles, on the northern shallow parts of NMPZ, clearly show sandbanks, as well as soft sediments on top of two sedimentary basins of significant thickness (>25 m), formed between the salt dome outcrops, on its deeper parts. Posidonia oceanica
beds are typically found between 10 m and 35 m deep, and mostly develop on shallow sandbanks (Figure 2
3. Data Acquisition and Manual Habitat Mapping
The NMPZ has been mapped in a multi-platform manner (Figure 1
), encompassing acoustic systems ranging from SSS and MBES to a chirp Sub-Bottom Profiler (SBP). An extensive supplementary visual seafloor inspection survey, using a towed underwater camera, offered the means for ground-truthing and validation of the mapped seafloor components. During the coupled SSS-MBES survey, 36 survey lines, with a total length of 262 km, were carried out using an Edgetech 4200 SP dual-frequency SSS, transmitting simultaneously at 100 and 400 kHz and a dual-head MBES Elac Nautic Seabeam 1185, transmitting at 180 kHz, covering a total area of 84 km2
(see Figure 3
). To meet MBES survey requirements, vessel motion was acquired using an SMC IMU-108 MRU system while sound velocity (SV) in the water column was recorded using Valeport MiniSVS as a keel SV probe and Valeport MIDAS SVP as an SV profiler, realizing down-casts on a frequent basis.
The MBES data was acquired through Hypack 2016, while Elac’s HydroStar software was used as the inertial beam-forming, navigation and attitude system. A Real Time Kinematics (RTK) GPS was used to obtain a 10 cm lateral positioning accuracy. To avoid changes in source levels, pulse lengths, receiver gains and directivity patterns, all above parameters were set constant throughout the survey. Automatic Gain Control (AGC) and the Time Varied Gain (TVG) were turned off. Pulse length and frequency were constant through the swath, with no sector dependence, and automatic adjustment of the pulse length as a function of depth was set off to 1 ms. The swath coverage sector was changing between 130° and 150° depending on depth (130° at areas deeper than 50 m) with an along- and across-ship beam of 1.5°, eventually producing 106 or 126 equiangular soundings per swath respectively, corresponding to data point densities ranging from 0.4 to 3.2 points/m.
Adapting to the needs of a time-limited field-work schedule and to the inherent low coverage of MBES in shallow waters, as it is proportional to depth, full coverage data was only feasible through the 100 kHz SSS dataset (see Figure 3
), whose slant range was set to 200 m. The lower the frequency the lower the sound attenuation in the water column and thus low frequency SSS can achieve full coverage seabed surveys in a fraction of time than with higher frequency SSS. Given the above, the effective slant range of the 400 kHz SSS was found to be 100 m, thus leaving systematic data gaps between the successive survey-lines, corresponding to about 50% of the surveyed area (see Figure 3
). The equivalent mean data coverage of the MBES was about 70%, ranging from <30% shallower than −20 m and >90% deeper than −30 m. A 20-km long ground-truth survey (see Figure 1
) was concurrently run using a SeaViewer 950 analog towed camera with an attached GoPro to validate acoustic bottom types. The ground-truth survey was planned so that all distinct seafloor types, having been recognized by expert interpreters, were visually inspected (see Figure 4
). Due to the lack of an underwater positioning system, adequate positioning of the underwater imagery was performed at post-processing using landmarks, collecting salient bottom features, evident both in the video and in the backscatter mosaic, calculating the layback of the tow-camera to the vessel and accounting for the mid-distances using standard spline interpolation.
Careful interpretation of all the available marine geoacoustical and ground-truth datasets led to a detailed manual habitat map of the NMPZ, on the basis of the Natura habitat classification scheme (Figure 4
). The Natura classes considered are the Posidonia oceanica
beds (Natura code category 1120), “Sandbanks” (1110), “Reefs” (1170), “Unvegetated soft bottoms” (119A) and “Vegetated soft bottoms” (119B). Vegetated soft bottoms, in the case of NMPZ, corresponds to Cymodocea nodosa
beds have further been separated into low relief (“meadows”) and high relief (“matte”) (Figure 4
). The division of Natura code category 1120 was done because, although Posidonia
meadows and matte are morphological deviations of the same species, they constitute totally different habitat types, both in backscatter texture and in ecological importance.
6. MBES Backscatter Angular Response Features
Backscatter angular response approaches use the angular response curves (ARCs) to extract statistical parameters that describe their overall structure and sometimes even directly link them to modelled responses of certain sediment granulometries. This work followed the approach of [65
]. ARC statistics are extracted from seafloor patches, defined as average per 2.5° angular bin (range of grazing angles) of several consecutive sonar pings, set here to 30. The stacked angular responses are then divided in three angular ranges: Near, Far and Outer. Finally, slopes and intercepts are extracted for each range per channel (port and starboard) (Figure 7
a). In our case ARCs are calculated directly from the recorded raw data files and they are corrected for transmission losses (spreading and absorption) and for the volume scattering due to the ensonified footprint following [66
Factor Analysis application to the ARCs, using the consecutive stacks of pings as cases and the respective angular bins as variables, clearly revealed 3 angular ranges with highly correlated responses. Those ranges, defined as Near: 0–25°, Far: 25–45° and Outer: 45–65° (Figure 7
b), nearly validates the work in [65
], which defined them as Near: 0–25°, Far: 25–55° and Outer: 55–85°.This way, a calibration is realized, especially in the Far and Outer ranges, so that those ranges are better adapted to our dataset. Following the above findings, slopes and intercepts where extracted from the ranges identified by Factor Analysis application, finally constructing an ARC derived feature vector of 6 features (Figure 5
A thorough evaluation process clearly showed that automated acoustic habitat mapping can be highly accurate when using multi-platform swath-sonars for shallow habitat mapping. MBES alone proved capable of separating most seafloor types, but it suffered from low coverage in shallow waters. This made lower-frequency SSS a very important complementary system, as its transmit geometries allow for full coverage even in very shallow habitats. We take advantage of the very low grazing angles that low-frequency SSS can achieve to maximize the scanned area. At the same time, we gain dense bathymetric information along with additional backscatter data from the MBES. This combines into four layers of information; three backscatter mosaics at different frequencies and bathymetry. Although SSS and MBES data cannot be precisely co-registered, how important this is for habitat mapping studies needs to be compared with the desired resolution. For each dataset, the accuracy in the positions of sonar measurements on the ground must ideally be smaller than their spatial resolution. In our case, where quite large seafloor segments define the baseline spatial scale for the analysis, positional accuracies are always comparatively higher.
Regarding the system selection per mapping subject and area, SSS offers the fastest possible mappings in terms of survey time, retaining wide swaths in shallow depths at lower frequencies and it preserves high textural content in the acquired images. The above makes it suitable for shallow habitats with highly textured patchy seagrasses on a plain seafloor. At the same time, common dual-frequency SSS offer the opportunity for a multi-spectral perception of the seafloor. While low frequencies show the highest overall contrast between different seafloor types, a consideration of all frequencies permits for an improved interpretation of subtle sea-floor features. Multi-spectra backscatter proved to be advantageous for the separation of seabed types where the volume scattering versus the isotopic scattering is of importance, i.e., when there is some shallow layering just below the seafloor as in the case of sparse seagrasses with extended rhizomes (e.g., Cymodocea nodosa
). However, due to the lack of auxiliary depth and incidence angle information, traditional SSS offers less capabilities for the application of a full range of geometric and radiometric corrections and the classification of habitats in rough terrains (e.g., reefs and steep slopes) is less accurate. In the contrary, despite the lower resolution and coverage that common MBES provide, they have been proven to offer the best trade-off between habitat discrimination and survey cost, as it is a many in one system. Given the recent advances in MBES Mills-Cross technology, offering snippet imagery with centimeter-accuracy and wideband operation e.g., [32
], one can get much more predictive results than the ones derived by the older Elac system used in this study. One should consider though that it would have been impossible for any MBES to provide a full coverage mapping without the object-oriented analysis that was made feasible through the full coverage low frequency SSS mosaic. The above arguments imply that while MBES takes the lead in habitat mapping works at present, SSS still remains an invaluable tool for higher resolutions and coverages in shallow waters that is always worth employing.
Regarding the habitat composition in NMPZ, measurements show that “Posidonia beds” cover 27% (22.5 km2) of the NMPZ seafloor, C. nodosa beds (“Vegetated soft bottoms”) 6% (4.8 km2), “Reefs” 9% (7.8 km2) and soft sediments 58% (48 km2), highlighting the ecological importance of the area. For the first time, a very detailed benthic habitat map was produced of this area, with classifications suitable for any future spatial planning or monitoring works in the NMPZ.
Effective protection and conservation of the marine realm and the goods and services that it provides to humanity, should follow an ecosystem-based management (EBM) approach [73
]. However, in order to successfully reach the EBM objectives detailed, accurate and easy to use habitat maps are required by both the scientists for the needs of integrated ecological assessments and the decision makers for marine spatial planning and management purposes, respectively [75
]. Therefore, the approach that we propose here for high accuracy habitat mapping through automated procedures may facilitate the enhancement of management effectiveness of MPAs in protecting species and habitats in the EBM context. In this respect, fine scale habitat mapping, in conjunction with the spatial information on human activities, can further enhance our knowledge in managing the conflicts between human uses and conservation objectives while mitigating the cumulative impacts of stressors [77
] both inside and outside the MPAs. Yet such information is valuable in identifying, designing and prioritizing networks of MPAs at various spatial scales [79
]. Still, the thorough knowledge of the seafloor topography through accurate habitat maps can also be of use in a fisheries management context [81
It is it well known that seagrasses and especially Posidonia oceanica
meadows are declining in the Mediterranean at alarming rates due to human activities and climate change and thus they have become one of the main targets of protection and management [82
]. Yet in the Mediterranean Sea the Natura 2000 marine sites have been designated under this protection status mainly due to the presence of the protected seagrass Posidonia oceanica
(priority habitat—Habitat Type 1120: “P. oceanica
]. Thus, the accurate mapping of this habitat-forming species will allow precise detection of any change in the distribution range size, demographic status and mortality events of P. oceanica
meadows through space and time and consequently to assist conservation efforts.
A fully automated shallow habitat classification approach was proposed, making use of multi-source, high-resolution swath sonars, for improved seafloor classifications in a total area of 84 km2. We exploit the efficacy of the combined use of MBES and SSS for gaining more acoustic backscatter spectral bands in a single survey, as well as dense bathymetric and angular backscatter responses information. Integration of MBES and SSS is an operationally efficient practice, quite common in marine geophysical surveys, to gain maximum information from a single survey. Such a multi-platform approach can offer at least three acoustic backscatter spectral bands on a single survey with varying data resolutions and mapping scales, supplemented by bathymetry and angular backscatter response information. Combination of the above systems proved to offer a good trade-off between survey effort and information gain, mainly in shallow waters. The poor textural content and coverage in shallow waters of the MBES was supplemented by the higher equivalents of SSS, offering an improved classification of the bottom classes. As a rule, MBES although providing poorer coverage in shallow waters, it provided the best single-system seafloor classification accuracies, but it needed to be combined with an object-oriented analysis made available through the full coverage SSS mosaic. On the other end, individual frequency backscatter mosaics, no matter which system they were derived from, were unable to provide adequate habitat discrimination. This was much improved when all backscatter frequencies were used in synergy, but it was still below the accuracy levels achieved with the fully integrated data model, combining both SSS and MBES data derivatives. NMPZ, as an MPA of high local importance, being the first no-take zone in Greece, stood as a perfect case study to showcase the above findings. It is a shallow, high diversity MPA, hosting extensive fields of the protected phanerogams P. oceanica and C. nodosa that have been fully and in detail mapped, offering the background for local management stakeholders to set the standards for an optimized marine spatial planning of the area.