Predicting Soil Sorption Coefficients of Phenanthrene Using a Neural Network Model

Abstract

Background: Sorption coefficient modeling is an effective technique for investigating fate and behavior of environmental pollutants. As a polycyclic aromatic hydrocarbon (PAH), Phenanthrene is an important organic pollutant, mainly due to its health risks for humankind. Objectives: To offer an alternative for laborious and high- priced experimental measurements, this study aimed to introduce an accurate artificial intelligence-based model, using minimum input data, to predict soil sorption coefficients (Koc and Kd) of phenanthrene. Materials and Methods: The required data were derived from previous studies carried out on soil samples taken from an under pasture paddock at Flaxley agriculture centre, mount lofty ranges, South Australia (Ahangar et al., 2008). An eight-fold cross-validation technique was also used to choose the best performance model and to obtain more authentic and precise results. Results: Multilayer perceptron (MLP) artificial neural network (ANN) model with a 1-6-1 structure was chosen which explained 97% and 95% of Kd and Koc variances, respectively. The only input data was soil organic carbon content. Conclusions: Based on this study, the ANN method is a promising alternative for conventional methods in modeling and estimating sorption coefficients in relation to soil organic carbon.