Machine learning approach for optimizing pesticide degradation in wastewater using a hybrid approach with ozone treatment and biological degradation

This study investigates the treatment of atrazine-contaminated wastewater through a hybrid approach combining ozone pretreatment and anaerobic degradation using an Upflow Anaerobic Sludge Blanket Reactor (UASBR). Atrazine, a widely used herbicide for controlling weeds in crops such as maize, sorghum, and sugarcane, is an emerging organic contaminant and a potential endocrine disruptor due to its complex structure and recalcitrant nature. In this study, synthetic wastewater was prepared with an atrazine concentration of 100 ppb, based on the maximum levels reported in environmental studies. The initial chemical oxygen demand (COD) of the solution was measured at 6930 mg/L, and the pH was adjusted to 10.3 ± 0.1 using NaOH and H2SO4. Ozonation was performed at varying ozone concentrations and durations to determine optimal conditions, ultimately identifying 9.4 mg/L ozone dosage for 40 min as most effective. This pretreatment reduced atrazine concentration to 66.5 ppb and significantly improved the biodegradability of the sample, facilitating subsequent biological treatment. The ozonated sample was then subjected to anaerobic treatment in a UASBR to evaluate and optimize atrazine removal performance. Ion chromatography was used to assess changes in ionic by-products pre- and post-ozonation. Increases in ions such as Cl−, NO3−, SO42−, and F−indicated the degradation of atrazine into intermediate compounds. To further analyze the interrelations among operational parameters, wastewater characteristics, and atrazine concentrations, advanced machine learning models – Multivariate Adaptive Regression Splines (MARS) and the Group Method of Data Handling (GMDH) – were employed. The integrated treatment system demonstrated atrazine and COD removal efficiencies of 83.6% and 85.6%, respectively. Among the models, MARS exhibited superior predictive performance, achieving R2 values of 0.930, 0.926, and 0.924 during training, validation, and testing phases, respectively, surpassing GMDH. Model evaluation using Taylor diagrams, the Comprehensive Measure (COM), and rank analysis further confirmed MARS as the more accurate and robust predictive tool for atrazine removal from the pesticide wastewater.