Title: Ensemble data assimilation experiments for the coastal ocean: Impact of different observed variables
1Ensemble data assimilation experiments for the
coastal ocean Impact of different observed
variables
An ensemble Kalman filter approach to data
assimilation for the NY Harbor.
- Ross N Hoffman1, Rui M Ponte1,
- Eric Kostelich2, Alan Blumberg3, Istvan
Szunyogh4, - and Sergey V Vinogradov1
- 1Atmospheric and Environmental Research, Inc.
- 2Arizona State University
- 3Stevens Institute of Technology
- 4University of Maryland
- IGARSS 2008 (Boston)
- FR3.111.4, Friday, 11 July 2008, 1420
2Estuarine and Coastal Ocean Model ECOM
- Based on Princeton Ocean Model POM
- 3d, sigma coordinate, curvilinear, C grid
- Currents, temperature, salinity, water level
- Turbulence energy, length scale
- Mellor-Yamada, level 2.5
- High-resolution model grid, allows 50m resolution
in rivers - Real-time application
- Realistic inter-tidal zone
- Comprehensive catalogue of fresh water and
thermal sources 241 treatment plants, 39 power
plants, 91 river systems
3LETKF Local Ensemble Transform Kalman Filter
- Kalman filter minimizes data misfit and
propagate uncertainty consistent with model
dynamics and prior information - Ensemble error covariance from N forecasts
- Local each grid point analyzed locally
- Transform minimize cost function in space
spanned by the forecast ensemble - LETKF is efficient and effective
- No change required to ocean model in these
experiments no adjoint needed - Used quasi-operationally with NOAA and NASA
atmospheric models
4Ensemble data assimilation approach
- The ensemble mean is our best estimate the
ensemble spread captures uncertainty - 16 sets of ECOM initial conditions are
established by sampling a validated model
simulation (nature) - 16 3hr ECOM forecasts made
- Nature errors gives observations
- 10 of grid points for each variable are observed
- Errors standard deviations 10 cm, 0.5ºC, 5
cm/s, 1 psu - LETKF optimally combines forecasts and
observations - For comparison, a free running forecast from mean
IC uses no observations.
5Nature run (True SST evolution)
SST 06 UTC 27 April 2004
SST 16 UTC 28 April 2004
NYC
LI
NJ
- Large change in plume of fresh/warm water over 34
h - Dynamically challenging test case
6Time-height cross sections
ECOM/LETKF Analysis
Free Running Forecast
Location
Truth (Nature Run)
T (degC)
S (psu)
Bathymetry Map
7Evolution of T and h Error
FRF
Analysis
8Surface Salinity Analysis Error
Analysis FRF
- Map view of SSS error
- Analysis errors much smaller than FRF errors
- S.D. of error for hours 48-96
- Grid point view of SSS error
- Shows rivers and inner harbor
9Findings
- Most useful for variables with slower times
scales - T, S are slow u, v, h are fast and adjust
quickly to tide and wind forcing so there is
little room for improvement - Errors and biases greatly reduced by the
assimilation - Sensitivity experiments
- Works well at all data densities examined
- As data density increases, the ensemble spread,
bias, and error standard deviation decrease - As ensemble size increases, the ensemble spread
increases and error standard deviation decreases - Increases in the size of the observation error
lead to a larger ensemble spread but have a small
impact on the analysis accuracy
10Data type impact experiments
11Temp. 10 km S of LI
12Salinity _at_ GW Bridge
13Observations in random columns
Baseline
Mobile
Fixed
Slow decrease In errors
T Error
Time
Filter appears to be diverging
T Spread
14Simulated observing network
Ferry
SST
CODAR
Buoy
15Layer 1 temperature spread trend
oC/hr
Filter divergence is only in unobserved river
head waters. These areas eliminated in following
statistics.
16Naive vs tuned localization
T Bias
Naive
Tuning eliminates filter divergence
Tuning improves errors
Time
T Error
T Spread
Tuned
Tuning very quickly removes bias
17Future work
- Real data
- Quality control
- Forecast uncertainty provides ruler for O-B
(obs-bkgrd) - Verification of forecasts and probability
forecasts - Model and data bias estimation
18Extensions
- Retrieval, ambiguity removal, data analysis at
once - ECOM modules include waves, biology, intertidal
zone, sediment transport, chemistry transport - LETKF allows general nonlinear obs operators,
bias correction for model and observations - Improved ocean forecasting (h,T,u,v,S) will
improve forecasting of all other properties and
vice versa - Ocean color, turbidity, wave statistics
- Not wave observations maybe wave statistics
- Brightness temperatures (SST info)
- CODAR line of sight currents
- Acoustic data (travel time)
- Drifters/gliders (trajectories positions)
- SAR, scatterometer
- Targeted observations
19Conclusions
- ECOM/NYHOPS is near real-time, and has
observation data base verification tools - LETKF is fully 4-d, efficient (mpi), req. no
adjoints - Experiments show LETKF is most useful for T, S
- u, v, h adjust quickly to tide and wind forcing
so there is little room for improvement - We see only weak coupling between T and S
analyses - More realistic simulation experiments indicate
tuning of localization is important - Many interesting extensions need exploring
- Complex obs operators accommodate unusual data,
targeted observations, bias correction
20End
- Contact rhoffman_at_, www.aer.com
- References
- A. F. Blumberg, L. A. Khan, and J. P. St. John,
Threedimensional hydrodynamic simulations of the
New York Harbor, Long Island Sound and the New
York Bight, J. Hydrologic Eng., vol. 125, pp.
799816, 1999. - I. Szunyogh, E. J. Kostelich, G. Gyarmati, E.
Kalnay, B R. Hunt, E. Ott, E. Satterfield, and J.
A. Yorke, A local ensemble transform Kalman
filter data assimilation system for the NCEP
global model, Tellus A, vol. 60, pp. 113130,
2008. - R. N. Hoffman, R. M. Ponte, E. J. Kostelich, A.
Blumberg, I. Szunyogh, S. V. Vinogradov, and J.
M. Henderson, A simulation study using a local
ensemble transform Kalman filter for data
assimilation in New York Harbor, J. Atmos.
Oceanic Technol., 2008, In press.