Grasslands provide a wide array of ecosystem services, such as forage for cattle and habitat for different types of organisms. Interactions between these services, land management, and biodiversity are complex and scale dependent. New generations of remote sensing sensors and machine learning approaches can predict grassland characteristics with varying accuracy. However, studies often fail to cover a sufficiently broad range of environmental conditions, and evidence suggests that prediction models are often case-specific. To address this gap, we have modelled above-ground biomass, plant height and species richness across the three Biodiversity Exploratories. These are a set of 150 grassland plots in three regions in Germany that follow a North-South gradient. They are characterized by different soil types, topography, elevation, climatic conditions, historical contexts, and management intensities. As predictors we use Sentinel-2 time series of surface reflectance, which are passed through a feed-forward deep neural network (DNN). Predictions of biomass, plant height and species richness show good levels of accuracy (r2 = 0.40-0.50) and could generalize well across regions. This study represents an important step forward in generating robust spatially explicit predictions of plant attributes and biodiversity variables across large areas, environmental gradients, and phenological stages. Better results are expected with the incorporation of UAV imagery and data augmentation schemes. The outputs produced can, in turn, be used in mechanistic ecological models to increase our understanding of the interaction between land management practices and ecosystem functions at regional scales.
