The annual LEO Coastal Predictive Skill Experiments provided a wellsampled ocean environment in whic

1
Validation of an Atmosphere-Ocean Forecast Model
at the Longterm Ecosystem Observatory
Validation of an Atmosphere-Ocean Forecast Model
at the Longterm Ecosystem Observatory
C. Sage Lichtenwalner, Scott M. Glenn, Hernan
Arango, Dale Haidvogel, John Wilkin Institute of
Marine and Coastal Sciences, Rutgers University,
New Brunswick, N.J.
Assimilation Datasets
Introduction
Downwelling Case Study
Model Transport Metrics
Metric Development
Quantitative model evaluation metrics for a
two-layer system were developed to compare the
mooring observations with the forecast results.
Based on a typical temperature profile (Figure
20), the boundary between the two layers is
assumed to be at the depth of the maximum
gradient (DMG). With this definition, the
surface, average upper layer and average bottom
layer temperatures can be objectively calculated
for both the mooring and the model profiles.
Plots of these parameters, along with the depth
and slope (SMG) of the thermocline are plotted in
Figure 21 for each of the 4 forecasts in the
downwelling case ensemble.
Why does an increased wind stress improve the
forecast offshore movement of the downwelling
front? Some clues are obtained by comparing the
transports in each layer with those observed by
the ADCPs at the mooring sites. Figures 25-27
illustrate the along-shore barotropic transport
for the full water column and the cross-shore
baroclinic transport in the upper layer as
defined by the maximum temperature gradient.
(The baroclinic transport in the bottom layer is
the mirror image of the surface layer.) The
standard wind and bottom stress forecast (Figure
25) indicates that the along-shore barotropic
transport and the cross-shore baroclinic
transport is underpredicted by the model during
the frontal passage on July 19. Later on July
20, the transports are in better agreement. The
decreased bottom friction hindcast (Figure 26)
indicates that the along-shore and cross-shore
transports are in better agreement throughout.
However, this forecast has trouble maintaining
the downwelling fronts distance offshore on July
20 when the baroclinic cross-shore transport is
properly predicted to go to zero. The
increased wind stress hindcast (Figure 27)
transports are in the best agreement during the
time period of the frontal passage when the
temperature metrics performed best. Later in the
forecast when the downwelling front is properly
maintained offshore of COOL-2 by the increased
wind stress, both transports are overpredicted.
On July 20, the model predicts strong onshore
surface layer and offshore bottom layer
transports when the observations say this
baroclinic transport is shut down. This
indicates that even though the increased wind
stress provides better predicted thermal
structure nearshore, it may be for the wrong
reason.
Extensive real-time datasets are available at the
Longterm Ecosystem Observatory (Figure 6) during
the annual Coastal Predictive Skill Experiments.
Assimilation data for the ROMS forecasts included
satellite-derived sea surface temperatures
(Figure 7), a nested grid of CODAR-derived
surface currents from a long-range regional array
(Figure 8) and a high-resolution local array
(Figure 9), and CTD data from an undulating
shipboard tow-body (Figures 10 and 11) and an
undulating autonomous underwater Glider (Figure
12). Ocean forecasts from each ROMS ensemble
were evaluated in real-time using ADCP and
remotely-profiled CTD data from the LEO
underwater nodes.
The most significant oceanographic event of the
2001 Coastal Predictive Skill Experiment was the
rapid downwelling event that occurred on July 19
during forecast Cycle 3 (July 18-21). The series
of cross-shelf temperature sections (Figure 16)
illustrate the progression of the downwelling
front past COOL-2 to about the location of COOL-3
by July 20. Stations COOL-4 through COOL-5
retain the sharp thermocline characteristic of
the two-layer system, but with a deeper
thermocline than at the start. Atmospheric
forecasts during this forecast cycle were
especially accurate, capturing both the timing
and magnitude of the switch to downwelling
favorable winds (see poster by Bowers et al.).
The fourth and final Coastal Predictive Skill
Experiment (CPSE) at the Rutgers University
Longterm Ecosystem Observatory (LEO) was
conducted from July 11 through August 7, 2001.
Ensembles of atmosphere and ocean forecasts were
generated twice per week for four consecutive
weeks. This poster describes the validation
procedures developed and applied to the Regional
Ocean Modeling System (ROMS).
Figure 1 LEO Research Area offshore Tuckerton,
New Jersey.
Figure 2 ROMS bathymetry.
Figure 20 Temperature profile validation metric
definitions.
Figure 6 Long-term Ecosystem Observatory during
the 2001 Coastal Predictive Skill Experiment.
ROMS Configuration
The most telling metric for this ensemble of
forecasts is the average bottom layer
temperature. For the ROMS forecasts forced with
the Operational COAMPS, the bottom layer warming
begins immediately, over a full day too early.
The average bottom layer temperature never warms
in the ROMS forecast forced with the Experimental
COAMPS using the MY closure. The same forcing
with the KPP closure produces a bottom layer
temperature that begins warming at the right
time, but is slower than the observations in its
final progression to isothermal.
Figure 16 Cross-sections of observed
temperature during the Cycle 3 downwelling event.
Cycle 3 Ocean Forecast Ensemble
The ROMS forecast domain for the Coastal
Predictive Skill Experiments covered the full New
Jersey continental shelf (Figure 2). Nominal
configuration is as follows
Time series of temperature and velocity profiles
were extracted from each ROMS forecast at the
locations of the validation moorings. Figure 17
displays the temperature profile time series at
COOL-2 from the four forecasts in the real-time
ensemble. The two ROMS forecasts forced by the
Operational COAMPS warm rapidly at the bottom on
July 18 with significant mixing as the
thermocline broadens. Both ROMS forecasts forced
by the Experimental COAMPS maintain a sharp
thermocline. The forecast with MY closure
maintains two-layers throughout the cycle,
indicating the downwelling front remains inshore
of COOL-2. With KPP closure, the downwelling
front moves past COOL-2 late in the day on July
19, becoming isothermal by mid-day on July 20.

• Tidal Forcing Specified along the outer
  boundary using the ADCIRC harmonic forcing.
• Atmospheric Forcing Operational (27 km
  resolution) and Experimental (6 km resolution)
  COAMPS.
• Air-Sea Fluxes COARE, Fairall et al., 1996.
• Ocean Surface Boundary Layer KPP, Large et al.,
  1994, and MY2.5.
• Ocean Bottom Boundary Layer Inverted KPP,
  Durski et al., 2001, and MY2.5.
• Bottom Stress Wave-Current Boundary Layer with
  Rippled Bed, Styles and Glenn, 2001.

Figure 7 Satellite Sea Surface Temperature Image.
Figure 8 Long-Range CODAR Surface Currents.
Figure 9 High-Resolution CODAR Surface Currents.
Figure 25 Transport metrics at COOL-2 for the
standard wind and bottom stress forecast.
Figure 26 Transport metrics at COOL-2 for the
decreased bottom stress hindcast.
Figure 27 Transport metrics at COOL-2 for the
increased wind stress hindcast.
Figure 3 ROMS boundary layers.
Wind-Current Correlations
Real-time Ensemble Forecasts
Figure 12 Undulating autonomous underwater
Glider equipped with CTD.
Figure 10 Undulating towbody equipped with CTD
and Flouromter.
Figure 11 Typical temperature section collected
with the towbody.
Figure 21 Time series of temperature profile
validation metrics for the Cycle 3 ensemble.
Blue lines are the ocean observations, red lines
are the ROMS forecasts.
Because the coastal ocean response during the
stratified season is highly correlated with the
wind (see Kohut et al. talk), additional clues to
model behavior may be gained by comparing the
model correlations with the observed. Figure 28
shows the average wind during forecast Cycle 3,
and the average current at each depth measured by
the ADCPs deployed at moorings COOL-2 through
COOL-5. The downwelling favorable wind produces
currents that are generally along-shore to the
south but turn to the left with increasing depth.
The least amount of turning is experienced
inshore at COOL-2, where the water column is
unstratified for about half of the
forecast. Average currents in the standard wind
and bottom stress forecast (Figure 29) are
underpredicted, especially offshore. Turning with
depth is to the left at the two inshore stations.
While surface and bottom current directions
agree with the observations offshore, mid-depth
current predictions are opposite of the
observations. However, the mid depth currents
moving in the wrong direction are also
uncorrelated with the wind. Figure 30 indicates
that the bottom stress reduction only slightly
increases the magnitude of the average currents
and has little effect on their vertical structure
compared to the standard forecast. Figure 31,
however, demonstrates that the increased wind
stress causes a fundamental change in the current
response. All currents are now of similar
magnitude as the observations, and all turn to
the left with depth.
Validation Array
The Navy Operational Global Atmosphere Prediction
System (NOGAPS) provided initial and boundary
conditions for two regional atmospheric forecasts
with the Coupled Ocean Atmosphere Mesoscale
Prediction System (COAMPS). Both the
low-resolution Operational COAMPS and a
high-resolution Experimental COAMPS were used to
force the Regional Ocean Modeling System (ROMS).
The ROMS forecasts were run with two closure
schemes, K-Profile Parameterization (KPP) and
Mellor-Yamada 2.5 (MY). The resulting ensembles
of 3 atmospheric forecasts and 4 ocean forecasts
were compared to available data to determine
which forecasts were most on-track. The best
ocean forecast was used for adaptive sampling
mission planning for aircraft, ships and
autonomous underwater vehicles, and as forcing
for the EcoSim bio-optical model.
Sensitivity Studies and Metrics
A cross-shelf array of 6 moorings were deployed
at 4 km spacing through the middle of the 30 km x
30 km LEO research area (Figure 13). Each
mooring location was occupied by a thermister
string and a bottom-mounted ADCP. Thermisters
where placed every 1/3 m at COOL-1, and every
meter at COOL-2 through COOL-5. COOL-1 was
deployed alongside the LEO Node ADCP. COOL-6 was
an extensive offshore bio-optical mooring (Chang
et al. 2002).
Model temperature metrics indicate that the
real-time ROMS ocean forecast using the KPP
turbulence closure and forced with
high-resolution Experimental COAMPS winds
reproduced the Cycle 3 downwelling event with
sufficient accuracy to improve adaptive sampling.
But why in Cycle 3 was the offshore propagation
of the downwelling front too slow? Was the
surface stress to weak, or the bottom stress to
strong? A matrix of sensitivity hindcasts that
increased the wind stress and decreased the
bottom stress was performed.
Figure 28
Figure 22 shows the model and metric results for
the hindcast with reduced bottom stress. The
initial timing of the downwelling front reaching
the location of COOL-2 is well matched, but the
front does not maintain its distance offshore,
creeping back inshore of COOL-2 on July
20. Figure 23 illustrates the model and metric
results for the hindcast with increased wind
stress. The downwelling front passes through
COOL-2 at the right time and remains offshore
throughout the forecast. The average bottom
layer temperature metric shows that by this
measure, this model performance is best during
the frontal passage.
Time Series Data
Figure 17 Ensemble of ROMS forecast temperature
profile time series at COOL-2.
Figure 13 Model validation mooring array
locations.
Validation data from the COOL-2 thermister string
is plotted in Figure 18. The thermocline
remains sharp until the downwelling front moves
past COOL-2 on July 19, remaining isothermal from
that point on. Based on this comparison, the
ROMS forecast with the KPP turbulence
parameterization forced with the high-resolution
Experimental COAMPS looks the best.
Figure 29
Time series of temperature and 30-hour low-pass
filtered cross-shelf and along-shelf velocity are
shown for COOL-2 (Figure 14) and COOL-5 (Figure
15). Offshore at COOL-5, a persistent sharp
thermocline is found between depths of 8 to 16 m.
Nearshore at COOL-2, the two layer system
responds to the winds by alternating between
downwelling and upwelling regimes. Downwelling
is sufficiently strong and sustained during the
observation period to move the downwelling front
offshore of COOL-2 three times, resulting in a
single warm layer from top to bottom. Only for
brief periods during this deployment are
upwelling winds persistent enough to move the
upwelling front offshore of COOL-2, resulting in
a single cold layer from top to bottom. Filtered
cross-shelf velocities have a pronounced
baroclinic response at COOL-5 throughout the
deployment and at COOL-2 when two layers are
present. Along-shelf velocities are strongly
barotropic at both sites, with the least vertical
variation inshore at COOL-2.
Figure 4 Ensemble forecast flowchart.
Figure 22 Temperature profile and model metric
time series at COOL-2 for the reduced bottom
stress hindcast.
Figure 30
Figure 18 Observed temperature profile time
series at COOL-2.
Figure 19 displays the vertical eddy viscosity
calculated from KPP and MY for the Experimental
COAMPS forcing. Initially the surface layer
response is similar, with a broad mid-layer peak
between the surface and the top of the
thermocline. Later in the forecast, the MY
surface layer looses track of the deepening main
thermocline, and instead begins to track a minor
mid-depth temperature difference. The KPP
surface layer fills the full water column late in
the forecast as the forecast temperature field
turns isothermal. The greatest difference,
however, occurs in the bottom boundary layer,
which is barely visible in the MY closure.
Conclusions

• The annual LEO Coastal Predictive Skill
  Experiments provided a well-sampled ocean
  environment in which to evaluate forecast model
  formulations rather than sensitivities to
  initial conditions.
• The validation domain was predominantly
  characterized as a 2 layer system with
  along-shore barotropic and cross-shore baroclinic
  transport.
• The high-resolution experimental COAMPS with KPP
  closure provided the best forecast of the
  downwelling front.
• Decreased bottom friction improved the
  along-shore and cross-shore transports a lot but
  did not improve the prediction of the downwelling
  front.
• Increased wind stress provided the best forecast
  of the front location and along-shore transport,
  but caused too much mixing offshore.
• Increased wind stress produced the best mean
  current profiles and correlation across the
  validation array.

Figure 23 Temperature profile and model metric
time series at COOL-2 for the increased wind
stress hindcast.
Figure 31
While the increased wind stress dramatically
improves the temperature forecast nearshore, does
the same hold offshore? Figure 24 compares the
observed temperature profile time series at
COOL-5 with the standard wind stress forecast and
the increased wind stress hindcast. The standard
wind stress forecast maintains a sharp
thermocline, while the increased wind stress
forecast results in too much mixing across the
thermocline. Thus arbitrarily increasing the
wind stress everywhere improves the forecast
nearshore but causes problems offshore.
Figure 14 Temperature, cross-shelf and
along-shelf velocity at COOL-2. Positive
cross-shelf velocity is offshore, and positive
along-shelf velocity is poleward.
Figure 5 Ensemble of 3 atmospheric forecasts and
4 ocean forecasts.
Figure 24 Temperature profile time series
offshore at COOL-5 for the observations, the
standard wind stress forecast, and the increased
wind stress hindcast.
Acknowledgements
Support provided by ONR, NOPP, NSF, NOAA/NURP and
the Great State of New Jersey. Special thanks to
Mike Crowley, Josh Kohut, Liz Creed, Louis
Bowers, and the rest of the COOL gang.
Figure 19 Vertical Viscosity predicted by the
KPP and MY closure schemes for the
high-resolution Experimental COAMPS Forecast.
Find this and more on the web http//marine.rutge
rs.edu/cool/coolresults/agu2002/
Figure 15 Temperature, cross-shelf and
along-shelf velocity at COOL-5.