514
BRUCE HACKETT ET AL.
bations to the time-invariant leeway coefficients (accounting for experimental variance) and time-varying perturbations to the wind field. The
A discussion of more
ad
McWilliams, 2002. Particularly the random flight model is useful when
studying surface drift. However, for SAR purposes, it suffices to observe
that the random walk will represent an upper bound on the inflation of
search areas as it represents the maximum dispersion of particles in a
given flow. Using random walk means erring on the side of caution in
the sense that the size of the search areas is not underestimated. As
search areas should be conservative (rather too large than too small),
this makes sense with an operational SAR model.
The orientation of a drifting object. A final random factor is the orientation of the object with regard to the local wind direction. As the leeway drift for most objects contains a substantial cross wind component,
there will be a significant discrepancy between the downwind direction
and the direction of propagation of the object. Whether the object drifts
to the left or to the right of the wind cannot be known in advance and
unless more is known about the object we must assign equal probability
to the two outcomes. Search areas are thus naturally bimodal, meaning
that there will be two disjoint areas of high probability.
Furthermore, one could even
”
perturb” the object class if nothing is
known about the object. In practice, however, this is done by running
several integrations from the same initial conditions, once for, say, a
person in water (PIW), then for a life raft, then for a swamped boat,
etc. Overlaying the different trajectories will give a total search area.
Implementation of stochastic initial positions. The ensemble is O(500)
and all members are positioned using a 2D normal distribution with a
standard deviation equal to half the radius input by the user. The mean
position of the particle cloud is a great circle arc and the release time
varies linearly throughout the ensemble. This approach is flexible: it
allows on one hand for point release in space and time and, in the other
extreme, a contiuous release in space and time in a
”
trumpet” shaped
area with one radius in one end and another radius in the other end of
the seeding area (see figure).
3.
Oil spill fate modeling
Oil spill fate models tend to be considerably more complicated than
the models for surface objects, due to the range of oil types and the
complex chemical processes that oil undergoes in the ocean (weathering).
We will in the following outline the types of models in use and the
latter represent a random walk perturbation.
vanced stochastic methods can be found in Griffa, 1996 and Berloff and
BRUCE HACKETT ET AL.
bations to the time-invariant leeway coefficients (accounting for experimental variance) and time-varying perturbations to the wind field. The
A discussion of more
ad
McWilliams, 2002. Particularly the random flight model is useful when
studying surface drift. However, for SAR purposes, it suffices to observe
that the random walk will represent an upper bound on the inflation of
search areas as it represents the maximum dispersion of particles in a
given flow. Using random walk means erring on the side of caution in
the sense that the size of the search areas is not underestimated. As
search areas should be conservative (rather too large than too small),
this makes sense with an operational SAR model.
The orientation of a drifting object. A final random factor is the orientation of the object with regard to the local wind direction. As the leeway drift for most objects contains a substantial cross wind component,
there will be a significant discrepancy between the downwind direction
and the direction of propagation of the object. Whether the object drifts
to the left or to the right of the wind cannot be known in advance and
unless more is known about the object we must assign equal probability
to the two outcomes. Search areas are thus naturally bimodal, meaning
that there will be two disjoint areas of high probability.
Furthermore, one could even
”
perturb” the object class if nothing is
known about the object. In practice, however, this is done by running
several integrations from the same initial conditions, once for, say, a
person in water (PIW), then for a life raft, then for a swamped boat,
etc. Overlaying the different trajectories will give a total search area.
Implementation of stochastic initial positions. The ensemble is O(500)
and all members are positioned using a 2D normal distribution with a
standard deviation equal to half the radius input by the user. The mean
position of the particle cloud is a great circle arc and the release time
varies linearly throughout the ensemble. This approach is flexible: it
allows on one hand for point release in space and time and, in the other
extreme, a contiuous release in space and time in a
”
trumpet” shaped
area with one radius in one end and another radius in the other end of
the seeding area (see figure).
3.
Oil spill fate modeling
Oil spill fate models tend to be considerably more complicated than
the models for surface objects, due to the range of oil types and the
complex chemical processes that oil undergoes in the ocean (weathering).
We will in the following outline the types of models in use and the
latter represent a random walk perturbation.
vanced stochastic methods can be found in Griffa, 1996 and Berloff and