Ever wondered if satellite crop biomass can be as precise as a drone’s view?

Winter cover crops benefit soil, yet assessing their impact at large scales is challenging. Satellites provide broad coverage, but their coarse pixels and spatial mismatch limit field‑level insights. Scientists merged UAS (drone) data with satellite imagery to refine winter rye biomass estimates. Two machine‑learning models—XGBoost and Random Forest—were trained in three scenarios: using only UAS data with field‑measured biomass, using only Sentinel‑2 data with satellite‑derived vegetation indices and texture measures, and a hybrid approach where drone‑derived biomass acted as surrogate ground truth to calibrate satellite features. Results show hybrid models cut errors by 20‑30% compared to direct satellite training. Drones capture fine‑scale details (5 cm) that satellites (10 m) miss, boosting accuracy and spatial fidelity. This method could simplify the adoption of cover crops by delivering reliable, scalable biomass maps for entire landscapes.

The research tackles a common challenge in precision agriculture: estimating crop biomass at landscape scales with enough accuracy to support policy and management decisions. The proposed hybrid framework merges the strengths of high‑resolution UAS data (5 cm) and extensive satellite coverage (Sentinel‑2, 10 m). Two machine‑learning algorithms—Extreme Gradient Boosting (XGBoost) and Random Forest (RF)—were trained on three complementary datasets. In the first scenario, models were trained solely on UAS‑derived vegetation indices (VIs) and crop height, paired with field‑measured biomass, delivering fine‑scale learning and RMSE values around 56 g m⁻² (XGBoost) and 62 g m⁻² (RF). The second scenario relied exclusively on satellite‑derived VIs and grey‑level co‑occurrence texture measures, with field biomass as target, resulting in markedly higher errors—up to 83 g m⁻² (XGBoost) and 80 g m⁻² (RF)—highlighting the loss of spatial detail when pixel size exceeds plant scale. The third, synergistic approach used drone‑estimated biomass as a surrogate ground truth to calibrate satellite features, effectively bridging the resolution gap. This yielded RMSE of 60 g m⁻² (XGBoost) and 61 g m⁻² (RF), demonstrating that UAS data can compensate for satellite limitations. Quantitatively, the hybrid models improve accuracy by 20‑30% and preserve spatial fidelity, making them suitable for operational precision‑agriculture pipelines. Periodic UAS flights could update satellite‑based maps, providing dynamic, high‑resolution biomass assessments for policymakers, agronomists, and farmers alike.