Predicting agricultural waste generation in the U.S.: A Data-Driven guide for smarter resource use

U.S. crop residue is a major underutilized bioenergy resource, yet current national models often rely on outdated data or broad assumptions, limiting localized planning for renewable energy and soil conservation. This study uses machine learning to quantify and classify county-level agricultural residue generation across the U.S. An XGBoost regression model achieved R² ≈ 0.69, showing that cropland acreage, USDA-estimated residue supply, local biorefinery potential, and regional clustering together explain much of the variation in predicted biopower potential. A Random Forest binary classifier reached ~83% balanced accuracy in identifying high- versus low-residue counties, while multiclass classification of residue behavior clusters achieved ~57% accuracy, with strongest performance in moderate-output, small-acreage areas. These findings highlight clear spatial patterns: high-residue counties are concentrated in the Midwest corn belt, while other regions display mixed or lower-output profiles. Corn was the primary contributor to residue volume, far surpassing crops like wheat and soybeans. The resulting reuse readiness maps provide actionable insights for farmers, policymakers, and renewable energy planners, offering a scalable tool to guide sustainable bioenergy investment while supporting long-term soil health and circular economy goals.