Multi-resolution remote sensing for flark area detection in boreal aapa mires

Abstract

ABSTRACT Peatlands have suffered significant degradation globally due to human impacts, which has increased the need to monitor the condition and changes in peatland ecosystems. With remote sensing, point-based in-situ observations can be upscaled to larger areas but there is a need to develop scalable monitoring methods. We predicted wet flark area extent, a key ecological indicator for patterned flark fens, in five sites in central Finland. Our primary aim was to test how the spatial and spectral resolution of different optical satellite image products (Landsat 8–9, Sentinel-2, PlanetScope) affect flark area coverage prediction. We also tested if there were seasonal or site-specific differences in prediction accuracy. Lastly, we upscaled the flark area coverage to entire mire areas. We used unmanned aerial vehicle (UAV)-derived flark area classification as a ground reference data to compare satellite sensors’ prediction accuracies. We predicted flark area coverage using spectral bands and indices as predictor variables using random forest regression. All sensors provided accurate results with some differences between Landsat (pseudo-R2 32−84%, root-mean squared error (RMSE) 10 − 18%), Sentinel-2 (R2 61−92%, RMSE 6−14%), and PlanetScope (R2 46 − 92%, RMSE 6 − 17%). The shortwave infrared bands of Landsat and Sentinel-2 did not increase the prediction accuracy. There were notable site-specific variations in prediction accuracy despite all the sites having typical open aapa mire wet flark – dry string patterns. With single-site models, the prediction accuracies were similar for early and late summer data, but when transferring the models to the other sites, performance significantly decreased, especially with the models using the late-summer imagery. Finally, we successfully upscaled the flark area coverage across entire mire areas. Our results demonstrate that the combination of UAV and satellite imagery can be used successfully to monitor peatland conditions and changes in them.