Improving ADM1 predictions via Bayesian analysis for continuous anaerobic digestion

The Anaerobic Digestion Model No.1 (ADM1) application for continuous anaerobic digestion is often constrained by challenges in reliably calibrating model parameters, especially when long-term data are unavailable. This study presents a Bayesian inference-based framework that enables ADM1 calibration using only initial-stage digester performance data. A custom Python implementation was developed, integrating modules for global sensitivity analysis, Bayesian calibration and parameter identifiability evaluation. Key microbial and ionic parameters were refined through Random Balance Designs-Fourier Amplitude Sensitivity Test (RBD-FAST), identifying sugar/acetate degraders and cation/anion levels in the inoculum as critical drivers of steady-state performance. With informative priors derived from similar ADM1 studies, the model was calibrated with less than two hydraulic retention times of data and validated against steady-state performance data. It predicted pH and total chemical oxygen demand (tCOD) with mean percentage errors of 1.10 % and 5.38 % respectively. Biogas production trends were captured within the 95 % credible interval for 63.14 % of observations. Compared to uniform priors, the Bayesian approach with informative priors improved predictive accuracy. Jensen-Shannon divergence revealed that hydrolysis rates is the most identifiable for thermally hydrolysed sludge. Unlike conventional ADM1 calibration approaches that require long-term steady-state data, this Bayesian framework achieves reliable predictions using only early-stage observations. By enabling accurate simulation of organic contaminant degradation and system stability from limited data, the framework supports risk-informed design and operation of anaerobic digesters and offers a solution for data-scarce industrial settings to improve safety, sustainability, and optimisation.