Data-driven prediction of methane potential of food waste collected from diverse waste sources

Understanding the drivers of methane (CH₄) yield variability in food waste (FW) is essential for predicting anaerobic digestion performance. This study evaluated the influence of feedstock origin and quality attributes on CH₄ yield and developed multiple linear regression models to estimate bioenergy potential. FW samples were collected quarterly over seven sampling periods from grocery, restaurants, schools, and landfill-bound residential streams. Samples were characterized for solids, pH, alkalinity, organics, and structural composition, and were tested using standardized biochemical CH₄ potential assays with inter-laboratory validation. Methane yield differed significantly among sources (p < 0.001): grocery FW produced the highest yield (802 mL CH₄/g VS), restaurant, and school FW ranged from 500-650 mL CH₄/g VS, and residential waste produced the lowest yields. Inter-laboratory reproducibility was acceptable, with relative standard deviations of 11-16.2 % and relative ranges of 15.6-22.9 %. Permutational multivariate analysis of variance showed source, time, and their interaction explained 75 % of variation, with source accounting for the largest contribution (30 %), indicating feedstock origin influenced CH₄ yield more than collection month. Source, solids, chemical oxygen demand, ORP, alkalinity, and protein content showed the strongest influence on CH₄ production. Model performance improved from the simple model (R² = 0.57; RMSE = 134 mL CH₄/g VS) to the practical model (R² = 0.63; RMSE = 121 mL CH₄/g VS), while the full model achieved the highest accuracy (R² = 0.73; RMSE = 107 mL CH₄/g VS). These findings support renewable energy prediction using feedstock attributes to enhance FW pretreatment and management across diverse waste sources.