Mapping Shallow Water Coastal Habitats Using Optical Remote Sensing and Machine Learning: Evaluating Drone, Airplane, and Satellite-Based Imagery

Herkül, K.; Vahtmäe, E.; Kutser, T.; Svirgsden, R., and Rohtla, M., 2024. Mapping shallow water coastal habitats using optical remote sensing and machine learning: Evaluating drone, airplane, and satellite-based imagery. In: Phillips, M.R.; Al-Naemi, S., and Duarte, C.M. (eds.), Coastlines under Global Change: Proceedings from the International Coastal Symposium (ICS) 2024 (Doha, Qatar). Journal of Coastal Research, Special Issue No. 113, pp. 519-523. Charlotte (North Carolina), ISSN 0749-0208. Mapping habitats in shallow coastal waters with conventional site inspections is costly and time-consuming due to the need for numerous sampling points, hindered by shallow water, rocks, and vegetation. Remote sensing (RS) imagery and mathematical modeling can reduce in situ efforts while improving mapping accuracy and resolution. This study tested various optical RS methods for mapping shallow water habitats in a coastal lagoon in the eastern Baltic Sea. On-site sampling, including bed substrate and macrophytes, was done through direct visual observations and underwater video recordings. Different RS imagery types were tested: drone and airplane-based orthophotos in RGB colors, airplane-based hyperspectral imagery, and Sentinel-2 multispectral satellite imagery. The Random Forest (RF) machine learning algorithm was used for supervised habitat distribution modeling. Field observation data and RS reflectance data were used to train RF models. Using the trained models, spatial predictions were made to cover the full extent of study areas with seamless maps of habitat variables. Cross-validation showed that all RS data types achieved moderate to nearly perfect prediction accuracy, effectively mapping the lagoon's habitat patchiness. It was concluded that the choice of RS method depends on the study area's size, characteristics, and desired spatial resolution.