Sensitivity Analysis and Surrogate Modeling of a Sulfur-Driven Autotrophic Denitrification Model

This study performed a global sensitivity analysis of a sulfur-driven autotrophic denitrification (SdAD) model, focusing on both biological and operational parameters that influence reactor performance, including nitrogen removal efficiency, sulfate concentration, soluble microbial product (SMP) production, and microbial dynamics. A preliminary sensitivity screening using the Morris method identified the most influential input parameters. Subsequently, surrogate models were developed employing machine learning techniques - specifically Random Forest and XGBoost algorithms - and were rigorously trained and validated using Sobol and Halton sampling sequences. The performance of the surrogate models was assessed using a number of metrics and tools, including error indices, adequacy plots, feature importance analysis, SHAP values, and Partial Dependence Plots. These assessments consistently confirmed the high accuracy and reliability of the surrogate models. The results highlighted the key role of parameters related to sulfur hydrolysis and autotrophic growth, as well as critical operational factors like target COD concentration and the timing of COD injection within the reactor. By improving the interpretability of the sulfur-driven autotrophic denitrification model, this approach enables more effective optimization of system performance and offers a robust framework for managing the complexity of this biotechnology.