Evaluation of Posidonia oceanica Map Generated by Sentinel-2 Image: Gulbahce Bay Test Site

Abstract

Remote Sensing offers an effective and low-cost solution for mapping the extent of seagrasses and temporally monitoring them routinely. In an in-situ well-explored Gülbahçe Bay test site, we used a high-resolution Sentinel-2 image to map the distribution of P. oceanica. Atmospheric and water column corrections were applied to the raw image. Thematic maps of the area were obtained by supervised classification and their accuracy was evaluated by cross-validation. The produced maps were also compared to a previously generated high accurate thematic map generated by the combination of sub-bottom profiler, side-scan sonar, and very high-resolution satellite image data. The classification achieved nearly 88% user accuracy at best. The assessment of the accuracy was also carried out in terms of water transparency. For this purpose, Secchi disk depths (SDDs) over the study area were inverted from the satellite-derived vertical attenuation coefficients by using a model. This model was retrieved by tuning the coefficients to 22 local SDD measurements. Sentinel-2 twinsatellite with its high spatial and temporal resolution data can contribute to mapping the P. oceanica cover in shallow waters. Their multispectral data can also be utilized for deriving the water transparency in order to determine the application limitations of the generated map.

Antalya Körfezi’ndeki Posidonia oceanica Deniz Çayırının Yaprak Boyu ve Biyokütle Miktarının Akustiksel Tahminleri

“…The detection of vegetation cover and distribution is particularly important, but previous studies have only employed standard methods (observations, sampling or marking with SCUBA, photography with the quadrat technique, satellite) (Pergent-Martini et al, 2005b;Yücel-Gier et al, 2020). Recently, acoustic devices (i.e., side-scan sonar, multi-beam echo sounder) and remote-controlled vehicles have been used for highly effective seabed mapping and characterization (Bonacorsi et al, 2013;Montefalcone et al, 2013;Buchet, 2015;Duman et al, 2019).…”

“…In addition, on-site observation, visual census, and visual monitoring are laborious and time consuming. In contrast, demands on human effort and time are reduced when using remote sensing systems (Komatsu et al, 2003;Buchet, 2015;Yücel-Gier et al, 2020). There is a need for new, fast, cost-effective, and approved strategies and methods to monitor and protect existing changes in susceptible species, especially those with a tendency to decline rapidly, such as P. oceanica.…”

Branching Algorithm to Identify Bottom Habitat in the Optically Complex Coastal Waters of Atlantic Canada Using Sentinel-2 Satellite Imagery

“…Therefore, three separate data sources were used in model building/evaluation to maximize the amount of information available to the classification algorithm. The number of data points used in the current study lies within the number of points used for other coastal Sentinel-2 studies ranging from <150 (e.g., Fauzan et al, 2017;Poursanidis et al, 2019) to >1,000 (e.g., Traganos et al, 2018;Yucel-Gier et al, 2020).…”

“…If a poor maximum depth for image classification is chosen, then all habitat below this depth will be classified as vegetated due to the relatively dark spectra compared to optically shallow regions. Generally, the maximum depth is either specified to coincide with the maximum depth that field survey data were collected (Yucel-Gier et al, 2020), or analytical modeling is used to determine the maximum depth bottom reflectance that can be detected (O'Neill and Costa, 2013;Poursanidis et al, 2019). In Nova Scotia, eelgrass beds have been documented up to 12 m deep (DFO, 2009) and kelp beds to 20 m deep (Johnson and Mann, 1988), yet we masked water deeper than 10 m to coincide with the maximum depth at which field survey points were collected during our study.…”