Development of the data-driven models for accessing the impact of design variables on heavy metal removal in constructed wetlands

Abstract

AbstractConstructed wetlands are a type of green infrastructure commonly used for urban stormwater treatment. Previous studies have shown that the various design characteristics have an influence on the outflow heavy metal concentrations. In this study, we develop a Bayesian linear mixed model (BLMM) and a Bayesian linear regression model (BLRM) to predict the outflow concentrations of heavy metals (Cd, Cu, Pb and Zn) using an inflow concentration (Cin) and five design variables, namely media type, constructed wetland type (CWT), hydraulic retention time, presence of a sedimentation pond (SedP) and wetland-to-catchment area ratio (Ratio). The results show that the BLMM had much better performance, with the mean Nash–Sutcliffe efficiency between 0.51 (Pb) and 0.75 (Cu) in calibration and between 0.28 (Pb) and 0.71 (Zn) in validation. The inflow concentration was found to have significant impacts on the outflow concentration of all heavy metals, while the impacts of other variables on the wetland performance varied across metals, e.g., CWT and SedP showed a positive correlation to Cd and Cu, whereas media and Ratio were negatively correlated with Pb and Zn. Results also show that the 100-fold calibration and validation was superior in identifying the key influential factors.