Abstract
The Mediterranean seagrass Posidonia oceanica, which provides highly valuable ecosystem services, is subject to increasing anthropogenic pressures, causing habitat loss or fragmentation. Whilst airborne images and acoustic data can be used for monitoring seagrass coverage at a macro-scale and over long time periods, monitoring its health in the short term requires precision mapping in order to assess current regression/progression of individual meadows. However, current fine-scale underwater techniques in the field are imprecise and time-demanding. We propose an automatic classification approach based on underwater photogrammetry for an operational, cost- and time-effective fine-scale monitoring method. The method uses a property of the sparse cloud generated during bundle adjustment-the reconstruction uncertainty-to map seagrass patches. The mean precision, recall and Fl score of the method over 21 study sites with different morphologies were 0.79, 0.91 and 0.84, respectively. However, the fragmentation level of the meadows had a significant negative effect on classification performances. The temporal monitoring of 3 sites using this method proved its operability and showed a positive evolution index of the corresponding meadows over a period of 3 yr. This method is generalizable for most encountered configurations and can be integrated in a large monitoring system, as it enables the production of numerous seagrass maps over a short period of time. Moreover, our methodology could be generalized and applied in the study of other submerged aquatic vegetation by adjusting the method's parameters.