FORECASTING DRIFT
513
2.3
Stochastic approach to drift prediction
In predicting the drift of objects on the sea surface, we are faced with
the challenge of accounting for uncertainties in almost all aspects of the
task. We have already seen that the models for drift of objects and ships
utilize empirical parameterizations (or empirically calibrated formulae)
and imperfect approximations to the hydrodynamical laws. In addition,
we often lack information about the object itself and where it is (or was
at some time). And even if we did have this information, there is still
the unavoidable uncertainty in the wind, wave and current data we use
to drive the drift models. Accounting for uncertainty is most readily
approached in a probabilistic framework. By assigning probabilities to
the relevant parameters, an ensemble of numerical integrations can be
performed where the various parameters are perturbed in a stochastic
fashion. The perturbations are dictated by the pertinent probability distributions. Thus, we get a cloud of
”
candidate” positions for the drifting
been extensively used across many scientific disciplines (see e.g., Press
et al., 1993).
The last known position. The first task in any real SAR or tracking
operation is to determine the last known position (LKP). For SAR, this
is a critical step, since the accuracy of this information is decisive for the
outcome of the search. In the case of ship drift, the position at any time is
often known very accurately. In the stochastic approach, an uncertainty
is then assigned to the LKP, both in space and time. If the LKP is
assumed to be very precise (e.g., a distress call is received from a ship
with a GPS unit), a small radius may be assigned to the datum and all
candidate objects (ensemble members) can be released at the same point
in time. In the other extreme - a situation where little is known about
the time and location of the accident - then a wide radius and a long
period of time must be used. This will result in a cloud of possible initial
positions scattered over a large portion of the sea surface released over
an extended period of time. Thus, the various members of the ensemble
will meet very different fates under the influence of differing current,
wind and wave conditions. Obviously, the choice of initial distribution
of ensemble members will affect the future search area seriously. It is the
task of the Rescue Coordinator to estimate the LKP and its distribution.
Uncertainties in forcing fields and drift properties. In addition to the
uncertainty assigned to the LKP, we also need to address the uncertainties present in the forcing data that are used and the drift properties
of the object. The spread of the ensemble is thus a function of perturobject. This cloud is itself a measure of search object’s most probable
location. Such a technique is known as a Monte Carlo integration and has
513
2.3
Stochastic approach to drift prediction
In predicting the drift of objects on the sea surface, we are faced with
the challenge of accounting for uncertainties in almost all aspects of the
task. We have already seen that the models for drift of objects and ships
utilize empirical parameterizations (or empirically calibrated formulae)
and imperfect approximations to the hydrodynamical laws. In addition,
we often lack information about the object itself and where it is (or was
at some time). And even if we did have this information, there is still
the unavoidable uncertainty in the wind, wave and current data we use
to drive the drift models. Accounting for uncertainty is most readily
approached in a probabilistic framework. By assigning probabilities to
the relevant parameters, an ensemble of numerical integrations can be
performed where the various parameters are perturbed in a stochastic
fashion. The perturbations are dictated by the pertinent probability distributions. Thus, we get a cloud of
”
candidate” positions for the drifting
been extensively used across many scientific disciplines (see e.g., Press
et al., 1993).
The last known position. The first task in any real SAR or tracking
operation is to determine the last known position (LKP). For SAR, this
is a critical step, since the accuracy of this information is decisive for the
outcome of the search. In the case of ship drift, the position at any time is
often known very accurately. In the stochastic approach, an uncertainty
is then assigned to the LKP, both in space and time. If the LKP is
assumed to be very precise (e.g., a distress call is received from a ship
with a GPS unit), a small radius may be assigned to the datum and all
candidate objects (ensemble members) can be released at the same point
in time. In the other extreme - a situation where little is known about
the time and location of the accident - then a wide radius and a long
period of time must be used. This will result in a cloud of possible initial
positions scattered over a large portion of the sea surface released over
an extended period of time. Thus, the various members of the ensemble
will meet very different fates under the influence of differing current,
wind and wave conditions. Obviously, the choice of initial distribution
of ensemble members will affect the future search area seriously. It is the
task of the Rescue Coordinator to estimate the LKP and its distribution.
Uncertainties in forcing fields and drift properties. In addition to the
uncertainty assigned to the LKP, we also need to address the uncertainties present in the forcing data that are used and the drift properties
of the object. The spread of the ensemble is thus a function of perturobject. This cloud is itself a measure of search object’s most probable
location. Such a technique is known as a Monte Carlo integration and has