A comparative study of neural network and mechanistic models for surface complexation

Abstract This paper demonstrates the use of a feed-forward neural network model to quantify the partitioning of phosphate onto water treatment residual (WTR) as a function of pH. Reasonably good results were obtained with a limited amount of experimental data. The neural network models were essentially as good as the specific mechanistic model used. Comparison of the neural network models with simple models obtained from statistical regression shows the neural network models to be superior. Quantification of the distribution of phosphate in this system may allow accurate prediction of available phosphate in a land application scenario. In surface complexation studies where mechanistic models are not available, it is recommended that neural network models be used.