Optimizing floating treatment wetland and retention pond design through random forest: A meta-analysis of influential variables

Abstract

Floating treatment wetlands (FTWs), artificial systems constructed from buoyant mats and planted with emergent macrophytes, represent a potential retrofit to enhance the dissolved nutrient removal performance of existing retention ponds. Treatment occurs as water flows through the dense network of roots suspended in the water column, providing opportunities for pollutants to be removed via filtration, sedimentation, plant uptake, and adsorption to biofilms in the root zone. Despite several recent review articles summarizing the growing body of research on FTWs, FTW design guidance and strategies to optimize their contributions to pollutant removal from stormwater are lacking, due in part to a lack of statistical analysis on FTW performance at the field scale. A meta-analysis of eight international FTW studies was performed to investigate the influence of retention pond, catchment, and FTW design characteristics on effluent concentrations of nutrients and total suspended solids (TSS). Random forest regression, a tree-based machine learning approach, was used to model complex interactions between a suite of predictor variables to identify design strategies for both retention ponds and FTWs to enhance treatment of nutrient and sediment. Results indicate that pond design features, especially loading ratio and pond depth (which should be limited to 200:1 and 1.75 m, respectively), are most influential to effluent water quality, while the benefits of FTWs were limited to improving mitigation of phosphorus species and TSS which was primarily influenced by FTW coverage and planting density. Findings from this work inform wet retention pond and FTW design, as well as guidance on scenarios where FTW implementation is most appropriate, to improve dissolved nutrient and sediment removal in urban runoff.