Deciphering the dynamic interactions among ammonia emissions and composting parameters in sewage sludge composting using multi-stage machine learning

During sewage sludge composting, the dynamic and nonlinear interactions among composting parameters and ammonia (NH3) emissions present challenges to conventional statistical methods, hindering stage-specific mitigation of NH3 emissions. This study employed multi-stage machine learning to investigate the associations among composting parameters and cumulative NH3-N emissions (cNH3) across different composting stages. The stage-specific models demonstrated high predictive accuracy (R2 = 0.85-0.91) on independent test sets. Shapley Additive Explanations analysis identified composting time, aeration rate, and pH as the features most strongly associated with cNH3 during the mesophilic and thermophilic stage, while aeration rate, pH, and organic matter were the predominant factors during the cooling and mature stage. Bivariate partial dependence plots revealed optimal parameter ranges and interactions, including a synergistic relationship between organic matter and nitrate levels, which was linked to lower cNH3 in the model outputs. These findings illuminate the evolving relationship between key composting parameters and NH3 emissions throughout the composting process, providing a scientific basis for developing stage-specific strategies to minimize NH3 emissions based on model-derived patterns.