Seagrass meadows around the world are threatened by anthropogenic activity, causing habitat destruction and global warming. There is a need for better high-resolution mapping to increase monitoring abilities and fill knowledge gaps regarding these vulnerable shallow coastal ecosystems. In this study, an underwater hyperspectral imager (UHI) was deployed by an unmanned surface vehicle (USV) to map spatial and seasonal distribution of marine vegetation (seagrass and macroalgae) in a Zostera (eelgrass) habitat in Hopavågen, a semi-enclosed bay in Agdenes, Norway. The UHI enables species-specific identification based on characteristic pigment composition of the organism of interest (OOI), resulting in a spectral reflectance curve (i.e. optical signature) detected by the sensor, and obtained in each image pixel with spatial resolution of 1 cm2 and spectral resolution of 2.2 nm. Three transect lines from September, December and February were recorded with an altitude of ~1.6 m above the area of interest on the seafloor.

After the georeferencing, radiometric processing and radiance conversion of the UHI data, the marine vegetation was identified using the supervised classification algorithm Support Vector Machine (SVM) in order to create distribution maps and estimate percent areal coverage of the OOIs. Additionally, the SVM-classifier was compared to Band Ratio and Decision Tree classifications. The efficiency and reliability of this mapping technique were assessed by looking at the classification accuracy, time use and the ability to revisit the same location. Since turf algae (filamentous epiphytic algae) is a major threat to seagrass health and grows rapidly due to eutrophication, the study also investigated the potential for detection and quantification of turf algae growth, but further work is needed to separate specific brown macroalgae species with confidence.

The SVM classification successfully separated seagrass from macroalgae, but performed differently according to the pre-processing and quality of the UHI data. Downwelling spectral irradiance and inherent optical properties of the seawater play an essential role in the signal-to-noise ratio, influencing correct classification. Thus, the Band Ratio classification is considered to be the most reliable and time efficient classifier for seagrass mapping in different seasons, under the conditions outlined by this thesis, using the characteristic reflectance maximum at 550 nm and reflectance minimum at 665 nm to extract information of the distribution of photosynthesizing biomass absorbing wavelengths corresponding to Chlorophyll a and b. However, withered seagrass was found to have a similar optical signature as brown algae, and this must be kept in mind when interpreting the classification results. Turf algae of small sizes were also difficult to classify with confidence.

By demonstrating USV-based UHI mapping, the study is a contribution to establish and validate different methods for acquiring and translating UHI data into ecologically important information, which can be applied to seagrass research and aid ecosystem management and conservation in the upcoming years.