Modelling temporal and spatial changes of PCBs in fish tissue from Lake Huron

A model has been developed to assess temporal and spatial changes in the concentration of polychlorinated biphenyl (PCB) contaminant in whole fish from Lake Huron during the years 1980 to 2004. The model uses log PCB concentration as the response variable and includes time trend, within-lake variability and dependence on age, weight and length as explanatory variables. A preliminary examination of the data revealed that some values are recorded as below detection limits (thus leading to the left censoring), and the PCB concentration appears to show declines in latter years. To this end, parametric log-location-scale regression models used in survival analysis were employed. It has been found that the Weibull model yields a better fit than the log-logistic or the log-normal models. The analysis provides strong evidence that, starting in 1996, the level of PCB concentration showed steady decline, which is most contributed by the Canadian and US governments' actions carried out earlier to reduce the load of toxic contaminants to the Great Lakes. Spatially, fish from the north of the lake is less contaminated than fish from the south. The pattern and magnitude of the estimated spatial and temporal trends can provide useful information regarding the safety of fish consumption, the setting of regularity limits, the identifiability of PCB sources and the effects of remedial actions in the future. In addition, the developed model is not restricted to the current application but could be used for the analysis of other contaminants.