Evaluating LUCAS in-situ and Sentinel data for grass height and vigour of vegetation in Europe

Quantifying grassland biomass is essential for monitoring forage availability in livestock-based agricultural systems and for supporting sustainable land management and policy decisions across Europe. Remote sensing offers extensive capabilities for monitoring grasslands, but the lack of adequate in-situ data for training and validation constrains the evaluation of regression models at a continental scale. The dedicated grassland module in the 2022/23 rollout of the Land Use/Cover Area frame Survey (LUCAS) constitutes a unique, large-scale and geographically comprehensive data source covering the European Union (EU). This study investigates the suitability of the LUCAS variables average grass height and vigour of vegetation for remote sensing-based analysis by conducting machine learning regression and classification on more than 4,000 transects with a cloud-free Sentinel-2 scene available. The EU-wide grass height Support Vector Regression model, based on Sentinel-2 bands and vegetation indices, delivers a root mean square error (RMSE) of approximately 19 cm and an R 2 of 0.25. Predicting the height of tall grassland presents greater challenges due to increased saturation of vegetation indices and the onset of plant senescence. The vigour of vegetation – a visual assessment on an ordinal scale – only allows for classification analysis that leads to frequent misclassification despite class aggregation. In conclusion, the average grass height is the preferred variable within the LUCAS database for biomass-related and remote sensing-based modelling. Rigorous filtering using ground images and metadata is strongly recommended to account for the environmental and management heterogeneity across the EU. • More than 4,000 transects with a fitting cloud-free Sentinel-2 image across the EU • Weak linear correlation between Sentinel-2 vegetation indices and grass height • Tall grass heights and strong variation complicate grass height regression • Index saturation, varying phenology and grassland type heterogeneity are main challenges • Limited agreement between Sentinel-2 data and LUCAS vigour of vegetation