Assessment of removal rate coefficient in vertical flow constructed wetland employing machine learning for low organic loaded systems

Secondary datasets of 42 low organic loading Vertical flow constructed wetlands (LOLVFCWs) were assessed to optimize their area requirements for N and P (nutrients) removal. Significant variations in removal rate coefficients (k(20)) (0.002-0.464 md(- 1)) indicated scope for optimization. Data classification based on nitrogen loading rate, temperature and depth could reduce the relative standard deviations of the k(20) values only in some cases. As an alternative method of deriving k(20) values, the effluent concentrations of the targeted pollutants were predicted using two machine learning approaches, MLR and SVR. The latter was found to perform better (R-2 = 0.87-0.9; RMSE = 0.08-3.64) as validated using primary data of a lab-scale VFCW. The generated model equations for predicting effluent parameters and computing corresponding k(20) values can assist in a customized design for nutrient removal employing minimal surface area for such systems for attaining the desired standards.

Automated summary

Secondary datasets of 42 low organic loading Vertical flow constructed wetlands (LOLVFCWs) were assessed to optimize their area requirements for N and P (nutrients) removal. Significant variations in removal rate coefficients (k(20)) (0.002-0.464 md(- 1)) indicated scope for optimization. The effluent concentrations of the targeted pollutants were predicted using two machine learning approaches, MLR and SVR, with SVR performing better. The generated model equations can assist in a customized design for nutrient removal and attaining the desired standards.