STATISTICAL MODELING OF THE PARTITIONING OF NONYLPHENOL IN SOIL

摘要

Partition coefficients K_P of nonylphenol (NP) in soil were determined for 193 soil samples which differed widely in content of soil organic carbon (SOC), hydrogen activity, clay content, and in the content of dissolved organic carbon (DOC). By means of multiple linear regression analysis (MLR), pedotransfer functions were derived to predict partition coefficients from soil data. SOC and pH affected the sorption, though the latter was in a range significantly below the pK_a of NP. Quality of soil organic matter presumably plays an important but yet not quantified role in sorption of NP. For soil samples with SOC values less than 3 g kg~(-1), model prediction became uncertain with this linear approach. We suggest that using only SOC and pH data results in good prediction of NP sorption in soils with SOC higher than 3 g kg~(-1). Considering the varying validity of the linear model for different ranges of the most sensitive parameter SOC, a more flexible, nonlinear approach was tested. The application of an artificial neuronal network (ANN) to predict sorption of NP in soils showed a sigmoidal relation between K_P and SOC. The nonlinear ANN approach provided good results compared to the MLR approach and represents an alternative tool for prediction of NP partitioning coefficients.