Treatment wetlands for secondary domestic wastewater: An experimental and modelling approach

This thesis explores the use of treatment wetlands (TWs) for domestic wastewater (WW), addressing the need for more experimental trials in secondary treatment scenarios and cold climates. Additionally, a data-driven model, the Sparse Identification of Nonlinear Dynamics (SINDy) algorithm, was employed for the first time to identify a nonlinear system of ordinary differential equations (ODEs) to describe pollutant removal rates in TWs. A 2x2x2 full factorial experiment was conducted to examine the effects of WW strength, temperature and vegetation on lab-scale horizontal subsurface flow TWs. High-resolution monitoring of biologically relevant parameters such as oxidation reduction potential (ORP), dissolved oxygen (DO), pH, and temperature was implemented. Experimental datasets were utilized for training and validating the chemical oxygen demand (COD), ammonia and phosphates reaction rate models. Results indicated no significant difference in COD removal between high and low WW strength, indicating TWs can effectively manage raw domestic WW. Results indicated no significant difference between high and low organic loads, indicating TWs can effectively manage high-strength domestic WW. However, temperature significantly decreased the COD removal (86-96% in warm and 58-76% in cold conditions). Additionally, systems with the largest increase in ORP values exhibited the best COD removal efficiencies. However, not enough oxygenic levels were wealthy enough to enhance ammonia degradation efficiently. The pH also presented a slight increase in the systems, which could affect the phosphate removal by promoting their release in the water. Finally, ORP offered more reliable insights into oxygenic conditions than direct DO measurements. These results underscore the potential of TWs to be implemented for secondary treatment, as well as the importance of continuous monitoring of ORP and pH parameters as just-in-time indicators of TW performance. ODEs were identified for modelling pollutant removal, with simplified equations employing 3 to 6 polynomial basis functions in low-noise scenarios and up to 11 polynomial basis functions in high-noise situations. Key parameters such as COD*ammonia, ORP, and pH were found to have the most significant impact on the pollutant removal rates. Validation results demonstrated robust predictive capabilities for COD models, achieving R² values between 0.81 and 0.98 with experimental data, while ammonia and phosphate models showed varying accuracy levels. These results suggest the need for effective application of the SNDy algorithm to identify the TWs dynamics and more datasets to be tested for model validation. They also highlight the importance of incorporating ORP and pH into TWs monitoring and modelling.