Graduation Semester and Year
Summer 2026
Language
English
Document Type
Dissertation
Degree Name
Doctor of Philosophy in Civil Engineering
Department
Civil Engineering
First Advisor
Dr. Melanie L. Sattler, P.E.
Second Advisor
Dr. Victoria Chen
Third Advisor
Dr. Arpita Bhatt
Fourth Advisor
Dr. Hyeok Choi
Fifth Advisor
Dr. Niloofar Parsaeifard
Abstract
Food waste constitutes the single largest component of municipal solid waste landfilled in the United States — approximately 22% of 292.4 million tons generated in 2018 — and its anaerobic decomposition releases methane, a greenhouse gas with global warming potential approximately 27–30 times that of carbon dioxide over a 100-year horizon (IPCC, 2021). Anaerobic digestion (AD) offers an alternative management pathway that recovers energy and produces nutrient-rich digestate, but AD performance varies by as much as five-fold across food waste streams (130–630 m3 CH4 per Mg VS added), and existing predictive tools either treat food waste as a homogeneous substrate or require dozens of mechanistic parameters that are rarely available in practice.
This dissertation develops a component-resolved predictive model for biogas production from food waste co-digested with sewage sludge, grasses, leaves, and waste paper. Waste paper is a new addition to this class of co-digestion study: clean paper is a recyclable commodity and is normally kept out of the waste stream, but paper contaminated with food — pizza boxes, napkins, paper towels, coated cups and molded-fiber containers — is rejected by fiber recyclers because grease and food residue disrupt pulping, and is landfilled. Because that fraction is generated, collected, and disposed of alongside the food waste itself, it is available to a food-waste digester at no cost in recovered fiber, and it supplies the only strongly carbon-rich lever available for correcting the low C/N ratio of protein-rich food waste. Sixty Biochemical Methane Potential (BMP) bottles were operated at 30 °C across three mixing ratios — Set A (10 % food / 90 % co-digestant), Set B (30 % / 70 %), and Set C (50 % / 50 %) — using six food waste categories (meat/fish, dairy/eggs, bread/cereal, fruits, vegetables, and FOG) and four co-digestants. Cumulative methane was measured by GC-FID over an extended mesophilic incubation (199–299 days). Linearized first-order decay (FOD) fits, computed against the experimentally measured per-bottle headspace volume (1–140 mL) with the rigorous current-headspace + prior-removed cumulative formulation and a per-bottle adaptive truncation of the linearization window (active-phase 5–95 % window for 41 of 60 bottles; saturation-only 0–95 % window for 19 slow-k bottles still in active growth at end of monitoring), yielded ultimate methane potentials L0 ranging from 341 to 77,505 µg CH4/g substrate and rate constants k from 0.0016 to 0.0342 d-1.
The fitted L0 and k values were modeled as functions of substrate composition using Multiple Linear Regression (MLR, SAS PROC REG) and Multivariate Adaptive Regression Splines (MARS, R earth v5.3.5, additive, GCV-pruned). The production Adelegan-MLR equation, selected by SAS PROC REG Best Subsets (maximum Adj R2) on the absolute-fraction parameterisation, produced (after a July 2026 AIC-parsimony re-fit per the statistics advisor) a 4-predictor L0 model with Adj R2 = 0.611 and a 5-predictor k model with Adj R2 = 0.640; the 9-fraction Sludge-reference fits (L0 Adj R2 = 0.623; k Adj R2 = 0.641) are retained as parameterisation-consistent baselines. The Adelegan-MARS equation achieved superior fit on the same data: R2 = 0.767 (Adj R2 = 0.730) for L0 through 8 piecewise-linear basis functions, and R2 = 0.837 (Adj R2 = 0.803) for k through 10 piecewise-linear basis functions. The k regression uses a per-bottle adaptive truncation rule (active-phase 5–95 % window for 41 of 60 bottles; saturation-only 0–95 % window for the 19 slow-k Set C bottles that never reached plateau), which removes lag-phase contamination and unlocks substantial non-linear structure including paired-hinge tents on Paper at 5 %, FOG at 6 %, and Grasses at 6.6/17.9 %. Direct interpretation of the MARS hinge functions identified FOG and grasses as the dominant drivers of k and L0, respectively, with supplementary post-hoc Sobol variance and SHAP cross-checks reported in Appendix F. Three pre-registered hypotheses were tested and all three were supported: H1 — Adelegan-MARS outperforms the Best-Subsets MLR baseline on both responses (ΔAdj R2 = +0.119 for L0 and +0.163 for k, each clearing the pre-specified 0.05 strong-improvement floor); H4 — the per-bottle decay rate k decreases monotonically with food-waste loading across Sets A, B, and C; and H7 — the sludge fraction exerts a significantly positive marginal effect on k. The validated **Adelegan-MLR equation** (linear baseline) and **Adelegan-MARS equation** (non-linear primary) were embedded in a Flask-based decision-support tool that enables AD facility operators to predict cumulative methane production, optimize co-digestion ratios, and screen project economics from substrate composition alone. The combined experimental database, regression equations, and web tool together provide a practical and reproducible pathway from waste characterization to methane forecasting for food-waste AD systems.
Keywords
Anaerobic digestion, Food waste, Biochemical methane potential, Methane, Co-digestion, First-order decay, Multivariate adaptive regression splines, Waste paper, Decision support tool, Latin hypercube design
Disciplines
Civil Engineering | Environmental Engineering | Industrial Engineering
License
This work is licensed under a Creative Commons Attribution-NonCommercial-No Derivative Works 4.0 International License.
Recommended Citation
Adelegan, Opeyemi E., "ANAEROBIC DIGESTION OF FOOD WASTE COMPONENTS: MODELING BIOGAS PRODUCTION" (2026). Civil Engineering Dissertations. 15.
https://mavmatrix.uta.edu/civilengineering_dissertations2/15
Included in
Civil Engineering Commons, Environmental Engineering Commons, Industrial Engineering Commons