4DVar Assimilation (physics) in ROMS ESPreSSO* John Wilkin, Julia Levin, Javier Zavala-Garay

1
4DVar Assimilation (physics) in ROMS
ESPreSSOJohn Wilkin, Julia Levin, Javier
Zavala-Garay

• 2006 reanalysis (SW06)
• Operational system for OOI CI OSSE (ongoing)
• Assimilating altimeter SLA satellite IR SST
  CODAR surface currents climatology glider
  T,S T,S from XBT/CTD, Argo, NDBC (via GTS)

• Use methodology developed for spring 2006 LaTTE
  reanalysis Zhang et al., Ocean
  Modelling, submitted 2009
• Skill assessed in forecast window
• several days for T,S
• 1-2 days for velocity

Experimental System for Predicting Shelf and
Slope Optics
2
(No Transcript)
3
ROMS ESPreSSO configuration

• 5 km horizontal resolution Cape Cod to Cape
  Hatteras
• 36 levels in traditional ROMS s-coordinate
  (stretching1)
• 4th order Akima T,S advection 3rd order upwind
  u,v advection
• Bathymetry, land-sea mask from NGDC Coastal
  Relief Model
• Open boundary data HyCOMNCODA (no bias
  correction)
• stiff boundary nudging in forward simulations
• Meteorology forcing 3-hourly 12-km NCEP NAM-WRF
• 72-hour forecast window
• sea level atmospheric conditions bulk formulae
  fluxes
• use NCEP NARR in 2006 reanalysis
• River daily average discharge USGS gauges
• adjusted for ungauged fraction of watershed
• Tides TPXO0.7 tides (5 harmonics)

We are working in a data rich location for 4DVar
assimilation
4
Mid-Atlantic Regional Coastal Ocean Observing
System (MARCOOS) CODAR, gliders, moorings,
tide gauges, drifters, satellites
5
IS4DVAR

• Given a first guess (the forward trajectory)
• and given the available data

Incremental Strong Constraint 4-Dimensional
Variational data assimilation
6
IS4DVAR

• Given a first guess (the forward trajectory)
• and given the available data
• what change (or increment) to the initial
  conditions (IC) produces a new forward trajectory
  that better fits the observations?

7
The best fit becomes the analysis
assimilation window
ti analysis initial time
tf analysis final time
The strong constraint requires the trajectory
satisfies the physics in ROMS. The Adjoint
enforces the consistency among state variables.
8
The final analysis state becomes the IC for the
forecast window
assimilation window
forecast
tf analysis final time
tf t forecast horizon
9
Forecast verification is with respect to data not
yet assimilated
assimilation window
forecast
verification
tf t forecast horizon
10
(No Transcript)
11
(No Transcript)
12
(No Transcript)
13
LaTTE 2006 reanalysis (60 days)
LaTTE domain and observation locations.
Bathymetry of the New York Bight is in
grayscale and dashed contours. yellow star is
location of Ambrose Tower green squares are
CODAR HF Radar sites
14
LaTTE 2006 reanalysis (60 days)
Comparison of observed and modeled sea surface
temperature and current at 0700 UTC 20 April 2006.
15
LaTTE 2006 reanalysis (60 days)
2-D histograms comparing observed and modeled
temperature, salinity, and u-component of
velocity model before (control simulation) and
after (analysis) data assimilation. Color
indicates the log10 of the number of
observations.
16
Ensemble average of the DA skill for analysis and
forecast periods for different data withheld from
systemVertical bars are 95 confidence.Vertic
al dashed line is boundary between analysis and
forecast.
LaTTE 2006 reanalysis (60 days)
Removing any data from the analysis system has
more or less predictable negative impact on the
forecast. More data is always better.
17
LaTTE 2006 reanalysis (60 days)

• Ensemble average of the DA skill for analysis and
  forecast periods for different data withheld from
  system, evaluated with respect to
• glider-measured temperature and
• satellite-measured SST Vertical bars are 95
  confidence.Vertical dashed line is boundary
  between analysis and forecast.

Satellite SST is crucial to forecast skill for
this skill metric namely, comparison to new
observations in the forecast window.
18
4DVar Assimilation (physics) in ROMS ESPreSSO

• 2006 reanalysis (SW06)
• Basis for experiments with ecosystem and
  bio-optical modeling
• 2009- operational System for OOI CI OSSE
  (ongoing)
• 72-hour forecast (NAM-WRF meteorology)
• tides, rivers, OBC HyCOM NCODA etc.
• assimilates
• altimeter along-track SLA
• satellite IR SST
• CODAR surface currents
• climatology
• glider T,S
• GTS XBT/CTD, Argo, NDBC

19

• Work flow for operational ESPreSSO/MARCOOS 4DVar
• Analysis interval is 0000
  2400 UTC
• Input data preparation commences 0100 EST (0500
  UT)
• RU CODAR is hourly - but with 4-hour delay
• RU glider T,S where available (approx 1 hour
  delay)
• USGS daily average flow available 1100 EST
• persist in forecast
• AVHRR IR passes (approx 2 hour delay)
• HyCOM NCODA forecast updated daily
• Jason-2 along-track SLA via RADS (4 to 16 hour
  delay)
• GTS XBT/CTD, Argo, NDBC from AOML (intermittent)
• T,S climatology (MOCHA)

20

• Work flow for operational ESPreSSO/MARCOOS 4DVar
• Input
  preprocessing
• RU CODAR de-tided (harmonic analysis) and binned
  to 5km
• variance within bins and OI combiner expected
  u_err used for QC
• ROMS tide solution added to de-tided CODAR this
  approach reduces tide phase error contribution to
  cost function
• RU glider T,S averaged to 5 km horiz. and 5 m
  vertical bins
• developed thermal lag salinity correction using
  constrained parametric fit to minimize statically
  unstable profiles
• AVHRR IR individual passes 6-8 per day
• use Matt Olivers cloud mask bin to 5 km
  resolution
• 2006 reanalysis uses REMSS daily SST OI
  combination of AVHRR, GOES, AMSR
• Jason-2 alongtrack 5 km bins (no coastal
  corrections)
• MDT from 4DVAR on mean model (climatology 3D
  T,S, usurface, twind)

21
Comparison of HF Radar observed and modeled M2
tide in LaTTE.
22
(No Transcript)
23
Work flow for operational ESPreSSO/MARCOOS

• Input preprocessing completes approximately 0500
  EST
• 4DVAR analysis completes approx 0800 EST
• 24-hour analysis is followed by 72-hour forecast
  using NCEP NAM 00Z cycle available from NOMADS
  OPeNDAP at 0230 UT (1030 pm EST)
• Forecast complete and transferred to OPeNDAP by
  0900 EST
• Effective forecast is 60 hours

OPeNDAP http//tashtego.marine.rutgers.edu8080/t
hredds/catalog.html ncWMS http//tashtego.ma
rine.rutgers.edu8081/ncWMS/godiva2.html
24
Output OPeNDAP http//tashtego.marine.rutgers.edu
8080/thredds/catalog.html ncWMS
http//tashtego.marine.rutgers.edu8081/ncWMS/god
iva2.html
25
ESPreSSO operational system
26
(No Transcript)
27
(No Transcript)
28
(No Transcript)
29
(No Transcript)
30
Lessons from operational IS4DVAR for ESPreSSO

• More and diverse data is better
• use all available observations and platforms
• Quality control
• outliers in CODAR
• cloud clearing from IR
• coastal altimetry
• High resolution regional climatology
• removes bias from Open Boundary Condition
• improves representation of dynamic modes and
  adjoint-based increments
• IR SST individual passes work best with 4DVAR
• time variability is explicitly resolved
• implications for optics
• Useful skill for operational applications
• glider reachability forecast
• Physics analysis affects ecosystem/optics model
  skill

31
Small scales may be important to large scale
dynamics (and ecosystem)
RU Endurance Line glider transect May 18-24, 2006
32
assimilation window
tf analysis final time
33
The analysis final hours becomes the data for the
high-res 4DVAR, then forecast
high-res assimilationwindow
high-res forecast
tf t forecast horizon
assimilation window
tf analysis final time
34
Issues/Tasks ahead for 4DVAR ESPreSSO

• High frequencies
• Filter inertial oscillations/tides in increment
  when updating outer loop
• High frequencies in coastal altimetry (keep or
  remove IB correction?)
• Background error covariance
• Can we use multi-variate balance constraint in
  coastal ocean?
• Wide shelf, steep slope, anisotropic variability
• Ecosystem / bio-optics?
• Ecosystem/bio-optics assimilation in 4DVar
• have Adjoint for NZPD model (ecosystem emphasis)
  
• have Adjoint for simple bio-optical model (IOP
  emphasis)
• Climatology? Initialization?
• need dense data set for assimilation development
• twin experiments?
• optics in Community Sediment Transport Model?
• Couple ecosystem/optics with thermodynamics
• interaction is significant on NJ inner shelf
  (just do it)
• Downscaling
• 1 km resolution grid (better bathymetry and
  land/sea mask)