Optimising grassland Above-Ground biomass Estimation for managed grasslands: A Gaussian process regression approach for Sentinel-2 and Planet Scope in Northern Italy

Abstract

Abstract Context Accurate and regular estimation of above-ground biomass (AGB) in grassland ecosystems is essential for sustainable grazing management, feed planning, and carbon accounting. However, AGB mapping in heterogeneous grasslands remains challenging due to the spatial and temporal variability of vegetation and management practices. Aims This study explores the potential of Gaussian Process Regression (GPR) models combined with multispectral imagery from Sentinel-2 and PlanetScope to predict AGB across different grassland systems in Northern Italy. Methods Extensive field measurements (n = 954) were collected over 18 months across meadows, lowland pastures, and alpine grasslands, covering a range of altitudes, management regimes, and canopy structures. Spectral predictors from Sentinel-2 and PlanetScope were used to train independent GPR models and evaluate their predictive performance at both pixel and field scales. Key Results At the pixel level, GPR models achieved R 2 = 0.520 (Sentinel-2) and R 2 = 0.514 (PlanetScope) with mean absolute errors (MAE) of ~400 kg DM ha −1 , consistent with the high heterogeneity of grassland canopies. Aggregating predictions at the field scale markedly improved accuracy (R 2 = 0.972 and 0.968; MAE = 60–120 kg DM ha −1 , ≤10% relative error). These results are comparable to those of commercial pasture monitoring platforms. Conclusion The integration of high-resolution multispectral imagery and non-parametric GPR modeling allows robust AGB estimation in heterogeneous grasslands, reducing uncertainty through field-scale aggregation. Implications and Impacts This research provides a scalable and transferable framework for operational biomass monitoring, offering a practical tool for digital decision support systems (DSS) and a scientific basis for integration into carbon Measurement, Reporting, and Verification (MRV) protocols. The novelty of the study lies in demonstrating the combined use of Sentinel-2 and PlanetScope data within a unified GPR framework for multi-site grassland systems, validated through extensive field observations.

Plant phenotyping relevance

衛星マルチスペクトル画像とGPRを統合して草地の地上部バイオマスという植物群落形質を推定し、広範な実測データで画素・圃場スケールの性能を検証しているため、方法開発・検証が中心である。

Code and data availability

The paper describes 954 field AGB measurements and GPR models trained on Sentinel-2 and PlanetScope imagery, but no blocks contain a public deposit of the field dataset, imagery subsets, analysis code, or trained models. The Copernicus and Planet portals are generic data sources, not paper-specific assets, and no data或

No evidence-backed public reproduction asset is currently recorded.