Re-visiting projections of PCBs in Lower Hudson River fish using model emulation

Remedial decision making at large contaminated sediment sites with bioaccumulative contaminants often relies on complex mechanistic models to forecast future concentrations and compare remedial alternatives. Remedial decision-making for the Hudson River PCBs Superfund site involved predictions of future levels of PCBs in Upper Hudson River (UHR) and Lower Hudson River (LHR) fish. This study applied model emulation to evaluate the impact of updated sediment concentrations on the original mechanistic model projections of time to reach risk-based target thresholds in fish in the LHR under Monitored Natural Attenuation (MNA) and the selected dredging remedy. The model emulation approach used a combination of nonlinear and linear regression models to estimate UHR water PCBs as a function of UHR sediment PCBs and to estimate fish concentrations in the LHR as a function of UHR water PCBs, respectively. Model emulation captured temporal changes in sediment, water, and fish PCBs predicted by the mechanistic model over the emulation period. The emulated model, using updated sediment concentrations and a revised estimate of recovery rate, matched the trend in annual monitoring data for white perch and largemouth bass in the LHR between 1997 and 2014. Our best predictions based on the emulated model indicate that the projected time to reach fish tissue risk-based thresholds in the LHR will take decades longer than the original mechanistic model projections.