Species-Level Classification of Peatland Vegetation Using Ultra-High-Resolution UAV Imagery

Abstract

Peatland restoration projects are being employed worldwide as a form of climate change mitigation due to their potential for long-term carbon sequestration. Monitoring these environments (e.g., cover of keystone species) is therefore essential to evaluate success. However, existing studies have rarely examined peatland vegetation at fine scales due to its strong spatial heterogeneity and seasonal canopy development. The present study collected centimetre-scale multispectral Uncrewed Aerial Vehicle (UAV) imagery with a Parrot Sequoia camera (2.8 cm resolution; Parrot Drones SAS, Paris, France) in a temperate peatland over a complete growing season. Supervised classification algorithms were used to map the vegetation at the single-species level, and the Maximum Likelihood classifier was found to perform best at the site level (69% overall accuracy). The classification accuracy increased with the spatial resolution of the input data, and a large reduction in accuracy was observed when employing imagery of >11 cm resolution. Finally, the most accurate classifications were produced using imagery collected during the peak (July–August) or early growing season (start of May). These findings suggest that despite the strong heterogeneity of peatlands, these environments can be mapped at the species level using UAVs. Such an approach would benefit studies estimating peatland carbon emissions or using the cover of keystone species to evaluate restoration projects.