ESTIMATING THE SUCCESS OF TRAJECTORY FORECASTS¹

A common tool in response to oil spills is trajectory forecasting. While the essential equations of oil spill trajectory modeling have not changed in four decades, the assessment of their effectiveness has only recently undergone quantitative analysis. The authors assert that estimating the success of any forecast can be approached in two different ways: (1) operationally, where the measured location of the spill is used to evaluate previous forecasts and improve the trajectory model in future forecasts, and (2) assessing the utility of the forecast, where spill location dependent cleanup decisions are affected by the model results, either favorably or unfavorably.

The challenge of determining operational success often lies in the method of estimating trajectory error in a defensible way that incorporates the uncertainty inherent in the reported slick observations and highlights the flaws in either the intrinsic model parameters or environmental input. The authors discuss the limitations of simply using the distance between the center of masses of the observed and forecasted spill locations. Some alternative supplementary approaches are discussed.

Adoption of a metric for spill modeling error allows the ability to assign anticipated accuracy to the forecast and highlights the input or model parameters most sensitive to forecast improvement. However, it does not measure the increased value of a better forecast to the actual response. To have an impact, spill forecasts must affect spill response decisions by providing additional information, often in the form of conditional probabilities, regarding oil amount and location. The authors review the likelihood that these probabilities would be, or could be, known, given different spill scenarios. If known, these probabilities can be combined with probabilities of success and anticipated rewards for different cleanup scenarios. By handling the problem in a probabilistic manner, trajectory analysis can assist analysts using risk-reward calculations to determine maximum possible gain while minimizing risk. Assessment in improved trajectory prediction would then be translatable into assessment in improved spill response. Whether existing technology is compatible with such an approach to trajectory forecasting and spill response remains an unanswered question.