Improving data quality for environmental fate models:: A least-squares adjustment procedure for harmonizing physicochemical properties of organic compounds

Physicochemical properties (vapor pressure, aqueous solubility, octanol solubility, Henry's law constant, and octanol-air and octanol-water partition coefficients) and their temperature dependencies are required for fate modeling of environmental pollutants. To be internally consistent, measured values for these properties often must be adjusted. The goal of adjusting the property values for consistency is to more accurately estimate the true values. However, consistency and accuracy are not synonymous. If there are systematic errors in one property, then adjustment for consistency may reduce the accuracy of other property data. Here, we provide methods for achieving consistency and improving accuracy in the selection of partitioning properties from literature sources. First, we show that a widely used procedure does not always minimize the adjustments of property values derived from the literature when harmonizing them according to thermodynamic constraints. In such cases, the final adjusted values (FAVs) are unnecessarily different from the literature-derived values (LDVs) selected from measurements. We present an improved procedure based on the theory of least squares that minimizes the adjustment of LDVs and allows quantitative propagation of uncertainty from LDVs to FAVs. When this procedure is applied to partitioning properties for 30 organic chemicals, FAVs obtained differ by up to 30% from those calculated with the current adjustment procedure. Second, we point out that the adjustment procedure is only appropriate for correcting random errors in measurement data. Biased LDVs must be identified and corrected prior to harmonization. Using a set of 16 PCB congeners as a case study, we provide methods to identify biased data and discuss possible sources of bias. We present a new interpretation of property data for the PCBs and a new set of internally consistent properties and quantitative structure-property relationships that we recommend as the best currently available.