SEAGRASS MAPPING USING SENTINEL-2 IMAGERY AND REMOTE SENSING TECHNIQUES: A CASE STUDY FROM CROATIA

Abstract

Nowadays, remote sensing techniques play an important role in the rapid acquisition of a large amount of spatial environmental data. The development of sensors in the last decade has led to the development of Earth observation satellite missions, e.g., Sentinel, PlanetScope. Numerous machine learning and deep learning methods are used in nowadays research to classify satellite imagery to enable rapid environmental mapping (e.g., land cover and land use, water bodies). Posidonia oceanica is considered the most important and best-studied seagrass species in the Mediterranean Sea. The objective of this preliminary research is to test the applicability of machine learning image classification methods for rapid seagrass mapping based on Sentinel-2 imagery. The research was conducted in the study area located in the north part of Dugi Otok in the central Adriatic in Croatia. Accuracy assessment of the mapped seagrass emphasises that Cart, Random Forest (RF), and Support vector machine (SVM) overperformed Naive Bayes (NB) method. Further, detailed visual analysis of seagrass map and accuracy assessment shows that RF and Cart give the best results. This research was done as part of the project Climate HIDROLAB (KK.01.1.1.04.0053) � Integrated hydrographic system for sustainable development of the marine ecosystem.