FORECASTING PCB CONCENTRATIONS IN LAKE-MICHIGAN SALMONIDS - A DYNAMICLINEAR-MODEL APPROACH

Ecological time-series data often consist of highly variable observati ons collected from ecosystems whose structure and function are changin g, making trend assessment and forecasting difficult. Dynamic linear m odels (DLMs) were used to study time trends in annual average PCB conc entrations in five species of Lake Michigan salmonids, using data coll ected from 1972 to 1994 by both the Michigan and Wisconsin Departments of Natural Resources. DLMs use an adaptive fitting procedure to track trends in the rate of decline over time, in contrast to other approac hes that fit fixed parameters. We used this method to make forecasts o f PCB concentrations in these fishes, along with Bayesian credible-int erval estimates associated with these forecasts. Point estimates of PC B concentrations in all five species show a decline through a 10-yr fo recast horizon. However, upper bounds on the 90% credible-interval unc ertainty estimates for chinook salmon and lake bout indicate that stea dy or slightly increasing annual average PCB concentrations are within the 90% credible interval for the IO-yr forecast horizon. DLM forecas ts corroborate other modeling results for Lake Michigan salmonids that suggest that PCBs will decline very slowly over the next decade.