Conference Agenda

The removal of trace-level organic chemicals (TrOCs) from stormwater is critical due to their persistent, toxic, and mobile nature. Stormwater biofilters have shown promise in mitigating TrOCs through physical, chemical, and biological mechanisms. However, detecting TrOCs is time- and resource-intensive, underscoring the need to identify effective surrogates for real-time monitoring. This study evaluates the predictive potential of 11 surrogates, including nine water quality parameters (e.g., Total Organic Carbon (TOC), Dissolved Oxygen (DO), UVA₂₅₄) and two operational parameters (e.g., oxidation-reduction potential (ORP) and soil moisture), in stormwater biofilters under varying rainfall events. Using machine learning (ML) models such as Random Forest and XGBoost, acceptable prediction accuracy was achieved (R² > 0.5) for TrOCs, e.g., Caffeine, DEET, and Diuron. Additionally, a LSTM deep learning model was introduced, leveraging soil moisture and operational time (e.g., water release schedules) to predict ORP under different Real Time Control (RTC) biofilters (with averaged testing NSE above 0.7). Integrating parameters such as soil moisture and ORP, alongside operational time, provided new insights into biofilter behaviour, enabling accurate and efficient TrOCs removal predictions. This approach significantly advances automating biofilter performance monitoring, reducing the reliance on labour-intensive sampling, and enhancing the practical applicability of biofilters in stormwater management.