Answers are grounded in this paper and cite their sources.
Crop residues support soil health by reducing erosion, improving water retention, and contributing to carbon sequestration. Accurate estimation of crop residue biomass is essential for understanding residue distribution patterns and improving sustainable land management practices. Remote sensing, especially high-resolution UAV-based imaging, is a powerful tool for monitoring residue over agricultural fields, and many studies use remote sensing datasets for mapping residue cover (a 2D metric). However, few studies have evaluated residue biomass using remote sensing, despite biomass being more ecologically informative. This study uses high-resolution UAV multispectral imagery to predict crop residue biomass using feature selection and machine learning. Candidate predictors included raw bands, spectral indices, texture metrics, and digital-surface-model-derived topographic variables. Three feature selection methods—recursive feature elimination with cross-validation, Pearson correlation screening, and least absolute shrinkage and selection operator regression, were applied on the training set to identify informative predictors. Four machine learning models (Random Forest Regression, Support Vector Regression, CatBoost, and k-Nearest Neighbors [kNN]) were evaluated individually and in combination using simple averaging, weighted averaging, and stacked ensemble strategies. Results show that Pearson-selected features paired with kNN achieved the best performance (R 2 = 0.61, RMSE = 188.71 g m -2 ). Ensemble approaches did not outperform the best individual model, suggesting limited benefit from meta-learning under small-sample conditions. Across selection methods, red- and blue-band-related predictors were consistently retained, while textural and topographic variables were selected more selectively, indicating context-dependent contributions. Overall, simpler models with targeted feature selection can outperform more complex ensembles for UAV-based crop residue biomass estimation.
No takes yet. Share an insight, caveat, or question.
Yang et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: