Evaluation of Sentinel-2 time-series for mapping floodplain grassland plant communities

Abstract Monitoring grassland plant communities is crucial for understanding and managing biodiversity. Previous studies indicate that mapping these natural habitats from single-date remotely sensed imagery remains challenging because some communities have similar physiognomy. The recently launched Sentinel-2 satellites are a promising opportunity for monitoring vegetation. This article assesses the advantages of Sentinel-2 time-series for discriminating plant communities in wet grasslands. An annual Sentinel-2 time-series was compared respectively to single-date and single-band datasets derived from this time-series for mapping grassland plant communities in a temperate floodplain located near Mont-Saint-Michel Bay, which is included in the long-term ecological research network “ZA Armorique” (France). At this 475 ha site, 123 vegetation releves were collected and assigned to seven plant communities to calibrate and validate the Sentinel-2 data. Satellite images were classified using support vector machine (SVM) and random forest (RF) classifiers. Results show that the SVM classifier performs slightly better than the RF classifier (overall accuracy 0.78 and 0.71, respectively). They highlight that accuracy is lower when using single-date (0.67) or single-band images (0.70). The results also reveal that discrimination of plant communities is more sensitive to temporal resolution (Δ = 0.34 in overall accuracy) than spectral resolution (Δ = 0.12 in overall accuracy).