Estimation of organic carbon normalized sorption coefficient (K-oc) for soils using the fragment constant method

A fragment constant model for prediction of K-OC was developed and evaluate d with a diverse database of 592 chemicals belonging to 17 classes. The ran ge of experimental K-OC covered 7.65 log-units. The 592 chemicals were rand omly divided into a training set and a testing set for model development an d validation. A general model was then established using the entire databas e having 74 fragment constants and 24 structural factors. Statistically, th e regression model accounted for as much as 96.96% of the variation in the measured log K-OC. The mean residual between the experimental and predicted K-OC values was 0.366 log-units. In more than 74% of the chemicals studied the residual values were less than 0.5 log-units. The robustness of the re gression model, with respect to either specific individual chemicals or par ticular compound classes, was evaluated through use of jackknife tests. The experimental results confirmed the ability of the fragment model to predic t K-OC for a wide variety of untested chemicals.