Bayesian sunken oil tracking with SOSim v2: Inference from field and bathymetric data

Application of the SOSim v2 Model to Spills of Sunken Oil in Rivers

“…Uncertainty estimates on the parameters in the model are conducted by a −2 log-likelihood ratio test. Essentially, SOSim v2 uses the same basic approach to forecast the relative concentration of sunken oil in space and time in river settings as described in Jacketti et al [23] for marine settings, in which the model was previously validated versus data from the ITB (Integrated Tank Barge) DBL-152 oil spill. SOSim v2 does not incorporate hydrodynamic data as input because field observations on the location and concentration of sunken oil reflect current riverine conditions, and are used in inferring model parameters.…”

“…Sunken oil transport along f /H contours [24] was not considered in the riverine model, due to differing hydrodynamics relative to continental shelf locations. Thus, the algorithms described in Jacketti et al [23] for marine settings were adapted to more accurately predict the concentration of sunken oil in space and time in a river, considering their generally smaller scale and unique transport mechanisms. Accordingly, the probabilities derived based on gravitational forcing were scaled consistent with the relative probabilities of finding oil calculated based only on the available field concentration data.…”

“…The resulting field data are shown in Figure 2. Only a portion of these data, representing roughly one half of the study area, were used as input to SOSim to allow demonstration of the model's ability to locate sunken oil across the entire study area, similar to previous work [21,23]. Several assumptions were made when extracting the data from Figure 2 as input.…”

“…In this demonstration, ten times more bathymetric data points were sampled than the number of available field data points, to obtain a representative sample of the local bathymetry. The tenth root of the product of the prior likelihood functions was then taken to obtain one prior likelihood function, as described in Jacketti et al [23], to avoid weighting the evidence towards bathymetry and away from field data. The optional inputs and outputs for the M/T Athos I case study are described in Table 2.…”

“…In addition, although forecasts may be less reliable when data are available from only one sampling date, due to uncertainties in the inferred velocity and diffusion coefficients, a second prediction was made 24 days after the spill (12 days after the sampling campaign). This forecast was based on default parameters for the initial advection and dispersion of the oil during settling [23], to test the model's ability to forecast the movement of sunken oil in a river setting based on limited data. Figure 3a,b represent assessments of the location of sunken oil 12 and 24 days after the initial spill, respectively.…”