Estimation of annual grassland yields with Sentinel-2 time series

Grassland ecosystems shape the landscape in large parts of Germany and provide numerous services that are relevant for the carbon cycle, water quality and biodiversity, apart from being the main source of fodder for the dairy and meat industry. Annual yields between grasslands vary strongly because their productivity depends on the management and environmental conditions. Information on grassland yields are not freely and extensively available in Germany but would be relevant for comprehensive assessments of grassland ecosystem services including the impact of extreme events on yields. With satellite remote sensing, grassland productivity and yields can be extensively and multi-temporally estimated. Within our project (SUSALPS, https://www.susalps.de/en/), grassland yields are estimated in a grassland-dominated area in southern Germany using ground-truth measurements of above-ground biomass and Sentinel-2 time series data. Field data was collected on 12 differently used grassland parcels in the region in 2019-2021. We aim to overcome limitations of previous research – caused by the heterogenous nature of grasslands due to varying use intensities in Germany – by including management information and a large gradient of field samples trough multiple measurements throughout the vegetation growth period into the modelling. We tested empirical model based on the field and accompanying Sentinel-2 data (n=74) to estimate grassland biomass. The best model was applied to all available Sentinel-2 scenes in the region in 2019. Random Forest and Artificial Neural Network models showed the highest accuracy (R²_cv = 0.7). A novel input feature was the mowing date which is available as 6-year dataset (Reinermann et al. 2022 & 2023). Next, the multi-temporal biomass estimations are aggregated to annual yield estimates to enable spatially discrete and multi-annual yields are estimated and compared (2018-2023). First results show that the inclusion of mowing date information supports the reliable estimation of grassland yields and its assessment on fine spatial scale substantially. In the future, the results are coupled with modelled plant biodiversity information to gain a complementary picture on grassland ecosystem services.