276
J. Kjellsson et al.
in a manner not represented in the RCO model. However, due to the drogue anchored
at 12 to 18 m depth, the SVP drifters in the World Ocean have been shown to follow
the currents at the drogue depth with only small deviations (Niiler et al. 1995).
A difference between the open ocean and the Baltic Sea is that the latter has a
much shallower thermocline, which in some cases is situated at the drogue depth.
This could result in a shear on the drogue, in which case the effect on the drifter
is uncertain. For this reason, it would be interesting to equip future drifters with a
temperature sensor at both the top and bottom of the drogue to identify these events.
Multiplying the velocities of the RCO model output by a factor 1.25 or including parameterized subgrid turbulence resulted in a clearly better agreement between
several statistical characteristics of observed and simulated drifters such as distribution of velocities, absolute dispersion and mean displacement. Although both methods yielded similar results for mean displacement, the results were very different in
terms of some other metrics. The Lagrangian integral time scale was severely shortened after adding subgrid turbulence. The resulting time scales were shorter than
those of the recorded drifter segments.
The random motions introduced through the turbulence parameterization (Döös
et al. 2011) do not take the original velocity into account, thus changing both the
velocity and properties of the trajectory somewhat. The transport speed can thus be
improved with this parameterization, but at the cost of changes in, e.g., the transport
direction. If one needs to tune the motion of simulated drifters for single-particle
statistics, multiplying the simulated velocities by a constant factor would then be
a reasonable choice: it increases the speed but does not alter the properties of the
trajectory. This choice is however poor if the relative dispersion needs to be adjusted
since small, subgrid scale changes in the direction are needed to separate initially
closely located simulated drifters.
Furthermore, even when the simulated drifters were separated by at least one
grid box, the further separation rates were lower than the average rate for the
SVP drifters. Therefore, subgrid parameterization is needed also on larger scales
(D 2
R > 4 km). In this study, simulated drifters were tuned to resemble the mean displacement and absolute dispersion of observed SVP drifters. It would be possible
to instead tune them to yield good fit to observed relative dispersion, depending on
whether transport or spreading is the most crucial metric for the study. It is possible that using a higher-order turbulence scheme such as ‘Markov 1’ or ‘Markov 2’
(Rupolo 2007), thereby taking, e.g., velocity autocorrelation into account, would
give better results by combining the benefits of both methods used in this study.
The higher-order Markov models have been found to better describe the motions
of water particles than the ‘Markov 0’ model (Griffa 1996). Such schemes may be
implemented in TRACMASS in the future.
The comparison between simulated drifters and observed SVP-B drifters indicates that the RCO model simulates correctly neither the mean flow nor the turbulent
flow. Using values roughly estimated from Figs. 8.6 and 8.10, some implications
for Lagrangian modelling without tuning can be identified. If the simulated drifters
have 4/5 of the mean displacement of the SVP drifter segments, this would mean
that if simulated particles, on average, travel 100 km in 10 days, a drifter, or a real
J. Kjellsson et al.
in a manner not represented in the RCO model. However, due to the drogue anchored
at 12 to 18 m depth, the SVP drifters in the World Ocean have been shown to follow
the currents at the drogue depth with only small deviations (Niiler et al. 1995).
A difference between the open ocean and the Baltic Sea is that the latter has a
much shallower thermocline, which in some cases is situated at the drogue depth.
This could result in a shear on the drogue, in which case the effect on the drifter
is uncertain. For this reason, it would be interesting to equip future drifters with a
temperature sensor at both the top and bottom of the drogue to identify these events.
Multiplying the velocities of the RCO model output by a factor 1.25 or including parameterized subgrid turbulence resulted in a clearly better agreement between
several statistical characteristics of observed and simulated drifters such as distribution of velocities, absolute dispersion and mean displacement. Although both methods yielded similar results for mean displacement, the results were very different in
terms of some other metrics. The Lagrangian integral time scale was severely shortened after adding subgrid turbulence. The resulting time scales were shorter than
those of the recorded drifter segments.
The random motions introduced through the turbulence parameterization (Döös
et al. 2011) do not take the original velocity into account, thus changing both the
velocity and properties of the trajectory somewhat. The transport speed can thus be
improved with this parameterization, but at the cost of changes in, e.g., the transport
direction. If one needs to tune the motion of simulated drifters for single-particle
statistics, multiplying the simulated velocities by a constant factor would then be
a reasonable choice: it increases the speed but does not alter the properties of the
trajectory. This choice is however poor if the relative dispersion needs to be adjusted
since small, subgrid scale changes in the direction are needed to separate initially
closely located simulated drifters.
Furthermore, even when the simulated drifters were separated by at least one
grid box, the further separation rates were lower than the average rate for the
SVP drifters. Therefore, subgrid parameterization is needed also on larger scales
(D 2
R > 4 km). In this study, simulated drifters were tuned to resemble the mean displacement and absolute dispersion of observed SVP drifters. It would be possible
to instead tune them to yield good fit to observed relative dispersion, depending on
whether transport or spreading is the most crucial metric for the study. It is possible that using a higher-order turbulence scheme such as ‘Markov 1’ or ‘Markov 2’
(Rupolo 2007), thereby taking, e.g., velocity autocorrelation into account, would
give better results by combining the benefits of both methods used in this study.
The higher-order Markov models have been found to better describe the motions
of water particles than the ‘Markov 0’ model (Griffa 1996). Such schemes may be
implemented in TRACMASS in the future.
The comparison between simulated drifters and observed SVP-B drifters indicates that the RCO model simulates correctly neither the mean flow nor the turbulent
flow. Using values roughly estimated from Figs. 8.6 and 8.10, some implications
for Lagrangian modelling without tuning can be identified. If the simulated drifters
have 4/5 of the mean displacement of the SVP drifter segments, this would mean
that if simulated particles, on average, travel 100 km in 10 days, a drifter, or a real