In mid-August, NIVA researchers set out to undertake SeaBee’s first officially funded assignment from the Norwegian Environment Agency: developing techniques for drone-based mapping of kelp forests. The location was the Vega archipelago. One of the aims was to explore how deep into the water the drone sensors could detect kelp forests. Earlier, SeaBee has successfully mapped benthic habitats, like seagrass and rockweeds, at 5 meters depth, but in Vega the aim was to map marine vegetation down to 10 meters depth.
“Traditional techniques for mapping kelp forests are time consuming and expensive,” says marine biologist Hege Gundersen, SeaBee’s project co-lead and leader for the KELPMAP project.
NIVA researchers were collecting ground truth data while the SeaBee drones were surveyingthe area. The KELPMAP team in Vega was composed of Maia R. Kile (NIVA), Hartvig Christie (NIVA), Medyan Ghareeb (NIVA), Robert N. Poulsen (SpectroFly), and Toms Buls (SpectroFly). Photos by Maia R. Kile and Hartvig Christie.
Traditionally, field personnel use drop-cameras lowered down to the sea floor from a boat and observations are made manually. What we see then is only a very small part of the seafloor, whereas with a drone we can cover much larger areas in much less time. In the Vega archipelago, SeaBee’s drones were able to map up to 3 km2 of the coastal zone during one single day. But Hege also points out that it is still necessary to do manually field work and collect ground truth data to verify the observations made by the drones. This is needed both for training algorithms for image classification and habitat recognition and to validate their predictions. So, until our machine learning algorithms are sufficiently trained, we will still need skilled biologists in the field to identify species and habitats.
An example of a single image captured by one of the SeaBee drones during the KELPMAP fieldwork in the Vega archipelago. Photo by SeaBee.
After collecting images and data in the field, the next step is to create orthomosaics of the images, by stitching them together for a full georeferenced overview. Thereafter, the orthomosaics are analysed by Norsk Regnesentral, who leads SeaBees image analysis development using Machine Learning technology to create detailed maps of the marine habitats.
“These new remote sensing techniques are an exciting development for monitoring of underwater habitats and environments like kelp forests. Further, the project aims to use satellite data to upscale the drone-based habitat maps even further,” Hege added. Hopefully this project will bring us a step closer to using remote sensing for monitoring and mapping of the extensive kelp forests along Norway’s long coastline.