Multi-scale seagrass mapping in satellite data and the use of UAS in accuracy assessment

Abstract

Seagrass meadows play a vital role in coastal ecosystems health as constitute an important pillar of the coastal environment. So far, regional scale habitat mapping was implemented with the use of freely available medium scale satellite images (Sentinel-2 or Landsat-8). The Unmanned Aerial Systems (UAS) have increase the spatial resolution of the observation from meter to sub-decimeter. Using sub-decimeter imagery, seagrass can be mapped in great detail revealing significant habitat species and detect new habitat patterns. In the present study, we suggest a multi-scale image analysis methodology consisting of georeferencing, atmospheric and water column correction and Object-Based Image Analysis (OBIA). OBIA process is performed using nearest neighborhood and fuzzy rules as classifiers in three major classes, a) seagrass, b) shallow areas with soft bottom and c) shallow areas with hard bottom (reefs). UAS very high-resolution data treated as in situ observations and used for training the classifiers and for accuracy assessment. The methodology applied in two satellite images Sentinel-2 and Landsat-8 with 10m and 30m spatial resolution respectively, at Livadi beach, Folegandros Island, Greece. The results show better classification accuracies in Sentinel-2 data than in Landsat-8. There was a great difficulty in the detection of the reef habitat in satellite images because it covered a small area. Reef habitat was clearly detected only in the UAS data. In conclusion, the present study highlights the necessity of new high precision geospatial data for examining the habitat detection accuracies on satellite images of different resolutions.