Dataset and Code for Rapid In-line Monitoring of Full-Scale Anaerobic Co-Digestion Using Diffuse Reflectance Spectroscopy and Machine Learning

In-line diffuse reflectance spectroscopy (DRS) with partial least squares regression (PLS) was piloted at a full-scale municipal anaerobic co-digestion (AcoD) facility to evaluate rapid monitoring of heterogeneous feedstocks and digestate. Models developed from 42 high-strength waste and 146 digestate samples successfully predicted volatile solids, fats, carbohydrates, chemical oxygen demand, volatile fatty acids, and alkalinity with operationally useful accuracy (RMSE% 15–30; R2 up to 0.96). Protein predictions were less reliable due to limitations in reference data. Importantly, predicted volatile acid:alkalinity ratios provided rapid indicators of digester stability. Downsampling analysis demonstrated that effective models could be developed with fewer than 50 training samples for several parameters, highlighting opportunities to reduce analytical costs. Field deployment during periods of digester instability, including foaming and failure, further validated the robustness of DRS models under dynamic operating conditions. These results establish DRS as a potentially cost-effective tool for improving process stability, biogas yield, and decision-making at full-scale AcoD facilities.