Statistical Modeling of Phosphorus Removal in Horizontal Subsurface Constructed Wetland

A horizontal subsurface flow constructed wetland (HSSF-CW) was constructed to improve the water quality of an artificial lake in Beijing wildlife rescue and rehabilitation center, Beijing, China. Multiple Regression Analysis (MRA) and Artificial Neural Networks (ANNs) including Multilayer Perceptron (MLP) and Radial Basis Function (RBF) were used to model the treatment performance of total phosphorus (TP). In order to increase the model efficiency, input parameters were selected as influent TP concentration, hydraulic retention time, wastewater temperature, month of the year, porosity, area, precipitation and evapotranspiration based on the methods of principal component analysis (PCA) and redundancy analysis (RDA). Genetic algorithm and cross-validation were utilized to find the optimal network architecture and parameters for ANNs. The overall performance of the models was validated using different datasets from the case study spanning 3 years. The results implied that modeling using adequate but crucial parameters can provide an efficient and robust tool for predicting performance. By comparing the three models in terms of model fitness when applied to the prediction, ANNs seemed to be more efficient than MRA in modeling of the areal TP removal and RBF (R2: 0.829, p = 0.000) produced the most accuracy and efficiency indicating strong potential for modeling the TP treatment processes in HSSF-CW systems.