Transition from Sulfur Autotrophic to Mixotrophic Denitrification : Performance, Microbial Community and Artificial Neural Network Modeling

Abstract

To address the limitations inherent in both sulfur autotrophic denitrification (SAD) and heterotrophic denitrification (HD) processes, this study introduces a novel approach. Three carbon sources (glucose, methanol, and sodium acetate) were fed into the SAD system to facilitate the transition towards mixotrophic denitrification.  Batch experiments were conducted to explore the effects of influencing factors (pH, HRT) on the denitrification performance of the mixotrophic system. Carbon source dosages were varied at 1/8, 1/4, and 1/2 of the theoretical amounts required for HD (18, 36, and 72 mg/L, respectively). The results showed distinct optimal dosages for each of the three organic carbon sources. The mixotrophic system, initiated with sodium acetate at 1/4 of the theoretical value, demonstrated the highest denitrification performance, achieving NO3−-N removal efficiency of 99.8%. In contrast, the corresponding systems utilizing glucose (at 1/4 of the theoretical value) and methanol (at 1/2 of the theoretical value) achieved lower removal efficiency of 77.0% and 88.4%, respectively. Following the transition from SAD to a mixotrophic system, the abundance of Thiobacillus decreased from 78.5% to 34.4% at the genus level, and the system cultivated a variety of other denitrifying bacteria (Thauera, Aquimonas, Azoarcus, and Pseudomonas), indicating an enhanced microbial community structure diversity. The established artificial neural network (ANN) model accurately predicted the effluent quality of mixotrophic denitrification, which predicted values closely aligning with experimental results (R2 > 0.99). Furthermore, initial pH exerted greater relative importance for COD removal and sulfur conversion, while the relative importance of HRT was more pronounced for NO3−-N removal.

Keywords: Carbon sourceSulfur autotrophic denitrification Mixotrophic denitrificationDenitrification performanceMicrobial community Artificial neural network modeling