Daniel C. Szuba

Concern over polybrominated diphenyl ethers (PBDEs) - which are used as flame retardants in a wide range of products - have risen due to its ubiquitous presence in the environment, and due to the fact that since the 1970s levels of this contaminant in North Americans has risen 100 fold (from nearly 0 parts per billion (ppb) per lipid in 1973 to between about 61.7 to 79.7 ppb per lipid currently). It has been estimated that over 100 tonnes of PBDEs are present in Great Lakes sediments. This study used geographic information systems (GIS), and spatial statistical techniques to produce various maps expressing the concentrations of three selected PBDE congeners (BDE-47, BDE-153, and BDE-209) in Lake Ontario. Two different interpolation methods, Ordinary Kriging and Inverse Distance Weighting (IDW), were also compared to see which is more effective at producing such maps when a small number of samples is available.
For the kriging analyses log-transforming the dataset produced cross-validation statistics closer to optimal, with the best models being Gaussian, Spherical, and Gaussian for BDE-47, BDE-153, and BDE-209 respectively. Comparing the Mean Prediction Error (MPE) and Root Mean Square Prediction Error (RMSPE), it was determined there may be less interpolation bias and more interpolation accuracy in the IDW analyses when the log-transformed data were used. The validity of the prediction surfaces for both interpolation methods, however, needs to be questioned. Not only was the sample size potentially too small and unevenly distributed to accurately portray the surface; default settings were used for both methods; and it is possible log-transforming the data may have improved the cross-validation statistics based on a scaling effect of the resulting smaller values, and not necessarily from improvement in interpolation accuracy or bias. The MPE, RMSPE, and Correlation Coefficient (r) - calculated from the actual sample location values and the prediction values for each model, were also used to directly compare the best kriging results to the IDW results. It was found that for our study, kriging was the more statistically valid interpolation method to use, although for both techniques the overall spatial trends were similar. The overall spatial patterns of PBDEs in Lake Ontario observed in this study can be explained by the location of urban centres, as well as by the circulation of the lake. Although there were limitations to this study due to the small sample size used, interpolation techniques can be helpful tools in understanding the spatial distribution of PBDEs in Lake Ontario and in the Great Lakes in general.