Convergence of Ensemble Simulations for Environmental Risk Assessment

Ensemble simulations of oil fate and transport for a hypothetical oil spill, sometimes referred to as "stochastic simulations", are frequently used for environmental risk assessment related to offshore operations. In this study we investigate the importance of the number of simulations and the effects of different sampling strategies on the results. We focus particularly on stranded oil, as this result is often assessed as a worst case criterion. We have run three ensembles for different discharge durations (3, 24 and 48 hours), all having the same simulation duration of 10 days. The ensembles were created by starting one simulation every hour throughout the four year period 2009 - 2019, resulting in a total number of 35064 simulations for each of the three durations. The complete set, as well as smaller subsets, were then used for further investigations. We find large variations in the probability distributions for amount of oil in the environmental compartments atmosphere, ashore, sea surface, water column and sediments and for the biodegraded fraction. When calculating the autocorrelations of the series of simulation results, we found the shorter releases to have shorter correlation times. Based on these findings, we discuss the relationships between release duration sampling rate and expected deviation caused by missing samples. Furthermore, we looked into sampling strategies, using both uniform and random sampling to create subsets from the complete set of simulations. We also briefly discuss the length of environmental data series to use for an environmental risk assessment, and briefly outline some future work