Operational Oil Spill Modelling: From Science to Engineering Applications in the Presence of Uncertainty - Mathematical Modelling and Numerical Simulation of Oil Pollution Problems - page 106

grids [ 9 ], so the wind forcing and diffusivity parameters will need to be different
(typically higher). In particular, for large-scale oceanographic models the effects of
submesocale instabilities are poorly modelled and must be parameterized [ 17 ].
The challenge facing any oil spill forecast system is that the parameterization
of these different aspects are interdependent. Obtaining a “best” value of any given
parameter is impossiblewithout considering the systemmodelled, the types ofmodels
applied, and choices in the model setup (e.g. grid spacing). From the standpoint of
uncertainty evaluation, determining the best value is less important than estimating
a range of reasonable parameters for a given system. In Sect. 5.8 we discuss some
of the ways that hindcast modelling and drifter data can be used to improve our
understanding of these parameters.
5.6 Systems for Real-Time Forecast Uncertainty
Emergency responders need estimates of spill forecast accuracy and likely outcomes,
such as when and where a spill might make landfall. Ideally, forecasts should contain
a range of results, such as the earliest time landfall is expected or the widest range of
beaches that could be affected. A single model forecast cannot provide the necessary
range of data for effectively deploying emergency response equipment. Obtaining
systematic estimates of real-time forecast uncertainty requires an operational system
that evaluates the key uncertainty sources outlined in Sect. 5.4 , above. Some uncer-
tainties can be minimized in model construction, but the remaining uncertainties
need to be evaluated and reported to emergency managers with visualization tools
that are easy to use and understand. Although some progress has been made (e.g.
[ 18 , 39 , 41 ], presently there are no operational systems that can evaluate the accu-
mulation of uncertainty from the wind forecast through hydrodynamic, wave, and
oil spill modelling. Fortunately, the tools and technology to build such a system are
now available.
Perhaps the simplest way to evaluate forecast uncertainty is a brute-force multi-
model approach that takes the real-time forecast system of Fig. 5.1 and creates mul-
tiple model instances in a hierarchical series of solutions (Figs. 5.2 and 5.3 ). Amulti-
model operational system provides a range of forecast oil spills that can be used to
develop probability maps instead of a prediction cloud. Some number, N wind ,ofwind
forecasts are created based on a primary wind forecast and likely perturbations. For
each wind forecast a set of independent hydrodynamic models is run. These models
use some N wave different wave model coefficients and some N hydro different hydro-
dynamic model conditions (e.g. different wind drag coefficients, tidal forecasts, or
turbulence parameters). For each hydrodynamic model, a set of independent oil spill
models is run with N IC different oil spill initial conditions and N oil different oil
model parameters. This system requires a set of N wind N wave N hydro hydrodynamic
models and a total of N wind N wave N hydro N IC N oil oil spill models.
Three perturbations at each system level could be used to represent the expected
parameters along with high and low sets that bound the expected values. Note that