Toward sustainable crop residue management: A deep ensemble learning approach

Abstract

For the existence of biorefineries, a consistent supply of sustainable crop residues is critical. Crop residue removal may adversely affect the overall sustainability, i.e., soil productivity and economics of crop residue removal. Therefore, it is necessary to develop a tool or method to estimate the sustainable removal of crop residues which is critical to various stakeholders including farmers, biorefinery, and policymakers. We proposed a robust deep ensemble machine learning(ML) model to estimate soil sustainability indicators [i.e., soil-erosion-factor(SEF), soil-conditioning-index(SCI), and organic-matter-factor(OMF) as well the sustainable crop residues removal rates (RRR) using topography, soil properties, climate, and crop management practices. The results showed that proposed ML models are having very high predictability [SEF,R2 = 0.999; SCI,R2 = 0.996; OMF,R2 = 0.996, and RRR,R2 > 0.98) with low error for sustainability indicators and RRR. The proposed ML model can be used as a tool for estimating sustainable RRR and serve as a guide to assess sustainability indicators in real-time.