Grassland management and phenology affect trait retrieval accuracy from remote sensing observations

Grasslands, the most widespread terrestrial biome, are subject to various management practices that influence their biodiversity and ecological functions. Remote sensing offers a promising tool for monitoring these impacts, but challenges persist in heterogeneous grassland systems. This study combines radiative transfer model (RTM) and machine learning algorithms to assess the efficacy of the model inversion in predicting plant functional traits under different grassland management regimes. The model was applied to intensively and extensively managed grasslands using field-collected hyperspectral data. Results show that while RTM inversion effectively predicts traits such as leaf area index (LAI) and pigment concentrations in homogeneous, intensively managed systems, its accuracy diminishes in diverse, extensively managed grasslands, particularly for traits like leaf mass per area (LMA) and pigment content. These limitations stem from the model’s assumption of homogeneous canopy scattering, which fails to account for the heterogeneity in mixed green and brown vegetation, especially at senescence. Despite these challenges, the study highlights the potential of hyperspectral remote sensing to capture grassland management based on a solely empirical approach. Future research should focus on refining models to better account for canopy heterogeneity and integrating in-situ data to improve trait prediction in complex ecosystems. • Used machine learning to predict plant traits of differently managed grasslands. • An accurate prediction of leaf area index and pigments in intensive grasslands. • Model performance deteriorates in extensive, species-rich grasslands. • Assumption of a homogeneous canopy structure limits the prediction accuracy. • Potential for monitoring grassland management using hyperspectral data.