Mowing event detection in permanent grasslands: Systematic evaluation of input features from Sentinel-1, Sentinel-2, and Landsat 8 time series

• DOI: 10.1016/j.rse.2021.112751
• Publication Date: 2021-10-22T11:49:17Z
• Authors: Felix Lobert; Ann-Kathrin Holtg...; Marcel Schwieder; Marion Pause; Juliane Vogt; Alexander Gocht; Stefan Erasmi
• Keywords: Backscatter (email)

The intensity of land use and management in permanent grasslands affects both biodiversity important ecosystem services. Comprehensive knowledge about these intensities is a crucial factor for sustainable decision-making landscape policy. For meadows, the can be described by proxies such as mowing frequency, usually, higher number cuts indicate intensities. Dense time series medium resolution (10–30 m) remote sensing data are suitable detection events. However, existing studies revealed general lack consensus most appropriate input set consistent reliable detection. We systematically evaluated synergistic acquisitions from Sentinel-1, Sentinel-2, Landsat 8 to detect occurrence, date events an indicator grassland intensity. NDVI (Sentinel-2 8), γ0 backscatter, backscatter cross-ratio, second-order texture metrics well 6-day interferometric coherence (Sentinel-1) were used features. All possible combinations features tested train one-dimensional convolutional neural network, which enables enhanced exploitation temporal domain data. evaluation was conducted on 64 meadows overall 257 2017 2019 Germany. Our results that combination improves performance. highest accuracy reached NDVI, with F1-Score 0.84. frequency predicted mean absolute error 0.38 per year, while missed 3.79 days average. alone mostly underperformed comparison optical/SAR but clearly outperformed input-sets solely based SAR proposed model performed low further testing recommended highly intensive managed parcels. demonstrate additional value fusing three present Earth observation systems deliver freely available global coverage surface at resolution.