Environmental Engineering Reference
In-Depth Information
“forget” that it started with the wrong velocities [
21
]. Hydrodynamic spin-up time
can vary from days to several weeks, depending on the scales of the system. Testing
for spin-up time is relatively straightforward by starting the model from successively
older time points. When starting at an older time does not change the prediction for
today, the model results are effectively independent of the starting conditions and
the geophysical IC uncertainty is essentially zero. The key point is that operational
oil spill models
must
start from a hydrodynamic model that is already running and
past its spin-up time; for any significant domain, a hydrodynamic model cannot be
expected to reach spin-up in time for an accurate forecast if the model is only started
after a spill is reported.
5.4.4 Boundary Condition Uncertainty
Boundary condition (BC) uncertainty also includes both oil spill and geophysical
forcing components. At the oil spill itself, any ongoing leakage and its subgrid (near-
field) oil distributions are typically uncertain. Where the oil hits a land boundary,
the processes by which the oil adheres or remobilizes are poorly understood, again
resulting in highly uncertain BC. In geophysical modelling, the 3D currents andwater
surface elevations at the computational domain's edges are never known exactly,
but modelers have developed sophisticated methods to reduce the effects of these
uncertainties (e.g. [
59
]). Nevertheless, it is necessary that the hydrodynamic model's
artificial boundaries should be as far as possible from the location of an oil spill to
minimize BC effects.
Wind plays a major role in BC uncertainty. The spatial and temporal fluctuations
of the wind field are never precisely known (even in hindcast), and the modelling of
wind-driven waves, turbulence, and currents is strongly affected by empirical para-
meter choices and model structures. Added to these effects is the inherent uncertainty
in the overall wind forecast speed and direction. It can be argued that the BC uncer-
tainty associated with how energy from the wind affects waves, currents, and the oil
spill is the dominant form of uncertainty for any spill [
15
].
In a more general sense, the uncertainties above can be divided into “epistemic”
and “aleatoric” classes [
32
]. For our purposes, the former can be thought of as uncer-
tainty developed in modelling system through lack of either adequate models or data
(i.e. things we either know or should know); whereas the latter can be thought of as
uncertainty associated with the chaotic behavior of highly nonlinear systems, which
is deterministically unknowable (i.e. things we can only “know” stochastically) [
38
].
This classification concept can be used to focus model development efforts on reduc-
ing epistemic uncertainties, whereas system operational efforts can be focused on
evaluating effects of irreducible aleatoric uncertainties. For example, part of the
uncertainty in near-surface ocean currents in a coarse-grid hydrodynamic model is
epistemic uncertainty, which can be reduced by using finer grid—a choice made dur-
ing model development. In contrast, the effects of aleatoric uncertainty associated
with forecast wind conditions can only be evaluated for a particular event (e.g. by
Search WWH ::
Custom Search