Predicting Dry-Basis Carbon Content in Raw Plant Biomass from Ash and Volatile Matter

Abstract

Carbon content is needed for fuel characterization and carbon accounting, but measuring every incoming biomass lot by ultimate analysis is not always practical. This study examined whether routine proximate data could provide a preliminary value that remains easy to audit. The source table contained 1,689 solid-fuel records. Filters for feedstock type, completeness, physical limits, proximate closure, and duplication reduced this to 643 samples of untreated wood, grass or plant, organic residue, husk or shell, and straw. A category-stratified test set was reserved before comparing ridge regression, four nonlinear regressors, and three published correlations by nested cross-validation. Ridge based on ash and volatile matter returned the lowest mean root mean square error (RMSE; 2.883 wt.% C). For the 129 held-out samples, R² was 0.707, RMSE was 2.412 wt.% C, and mean absolute error (MAE) was 1.767 wt.% C. Its lead over the alternatives was small. The nominal 90% conformal interval covered 95.3% of test samples, although its mean width was 11.841 wt.% C; leave-one-category-out RMSE reached 4.621 wt.% C. The equation is limited to preliminary screening and quality assurance. Certified elemental analysis is still required for high-stakes decisions.