Peatland restoration age (Scotland, UK) can be better reproduced by a classification model based on Sentinel-2 than with high resolution aerial imagery

Damage to peatland globally causes significant contributions to the current net greenhouse gas emissions and pose a further future risk as such damaged peatlands are vulnerable to future climatic stress. Globally, peatland restoration efforts are rapidly increasing in scale as natural climate solutions, yet relatively little effort has been it into effective monitoring of landscape scale restoration projects. We developed a classification model that uses remote observations (Sentinel-2 or national scale aerial imagery from Getmapping) to detect restoration efficacy by training it against a dataset from a chronosequence of spatially collocated peatland restoration sites that had previously been converted to plantation forestry. The Sentinel-2 based model greatly outperformed the aerial imagery-based model (RGB and IR, 25 and 50 cm, respectively). Adding slope to the classification improved kappa by less than 0.02. Prediction of the starting (forestry) and target (restored) state was very robust, and both recent and the oldest restoration sites were spatially well predicted. The main model uncertainties lie with sites of intermediate age, where on-the-ground restoration trajectories based on vegetation composition also differ the most, and with sites where additional layers of management after the initial restoration management have been applied.