The Rutgers Ocean Modeling Group are running a "moderate" resolution
(5km horizontal, 30 vertical levels) regional ROMS forecast/analysis
model with 4-dimensional variational data assimilation of:
* all glider observations temp/salt (RU COOL)
* shipboard CTDs and XBTs of opportunity on the transit legs
  (thank you to all Chief Sci and WHOI staff who assisted this effort)
* scanfish profiles from Endeavor (Gawarkiewicz)
* underway thermosalinograph temp/salt from the first 2 Knorr legs
  (other Chief Sci we would love to have access to your underway data)
* daily composite SST (RU COOL)
* gridded altimeter SSH (AVISO) 

Meteorological forcing is NCEP/NAM 12-km 3-hourly forecast data. A
companion run with WRF forecast data will proceed soon. Daily average
USGS-observed Hudson River discharge is used. Tide open boundary
conditions are imposed from the Oregon State OTPS harmonic analysis.

Plots from the full hindcast sequence can be viewed at:
http://marine.rutgers.edu/po/sw06/roms/  We hope this page is compact
enough to be viewed by those of you at sea. Feedback on this is
welcome.

Plotted results are grouped by 2-day cycle. Observation location plots
show where in-situ obs were made, but do not distinguish the actual
data source - sorry, that comes later.

We assimilate data over a 2-day interval iterating the initial
conditions to minimize the model-data misfit over the cycle. The
subsequent 2-day cycle begins with first guess initial conditions
being the conclusion of the preceding interval. Consequently, a jump
in the analysis state can occur crossing from one cycle to the next.

For those of you with broad bandwidth there are links to Quicktime
animations at http://marine.rutgers.edu/po/sw06 and also links to NRL
NCOM ocean forecast plots. We are experimenting with applying boundary
constraints from NCOM.

This analysis started from the Linder and Gawarkiewicz NJ shelf
climatology with a crude estimate of total velocity, followed by
relaxation to the kinematic constraints of the model domain.

Unfortunately, climatology is a poor estimator of the initial state in
2006 given the extreme precipitation and Hudson River discharge we
experienced just prior to the beginning of our modeling cycles.
Therefore, what we observe in the output is that the model quickly
adjusts to in situ observations in the central SW06 region, strongly
influenced by the low salinities observed by the gliders. In the
absence of other data in the far field, the adjustment is
unsatisfyingly local at first which is something we will address
during reanalysis after the real-time experiment concludes.

As the model time proceeds, the lowered salinity region advances
slowly northward and eastward indicating that the adjoint is
propagating the model-data misfit information upstream, as it
should. The salinity update is not being advected downstream by the
forward model as quickly as we expected, and we suspect that
barotropic transport is not well constrained by the assimilation of
predominantly temp/salt data (constrains thermal wind OK) but no
absolute velocity data. CODAR and ADCP (moorings and vessels), and
potentially depth-average velocity from the gliders, we be used in the
reanalysis assimilation.

The model maintains sharp frontal features captured in the data, and
we overall we are pleased with this demonstration of operational 4DVAR
assimilation in ROMS for a contineental shelf observing system.

John Wilkin 
   for the Rutgers ROMS team:
   Hernan Arango, John Evans, Naomi Fleming, Julia Levin, 
   Javier Zavala and Gordon Zhang