1. Adjoint Sensitivity Studies on the US East Coast
Julia Levin, Rutgers University
Gordon Zhang
John Wilkin
Hernan Arango
September,
ONR Regional Review, IMCS

2. Long-Term Objective
Development of coastal ocean observing system that integrates coastal model with data from NEOS observational network to improve forecast
Use variational data assimilation tools that has been recently developed for ROMS model
Assimilation data (CODAR installations, cabled observatories, autonomous gliders and satellite imagery)
Observing system design based on sensitivity and stability analysis of the forecast system.

3. Approach
Model Setup is based on LaTTE domain (New Jersey Bight, New York Bight, Hudson Shelf Valley) to simulate the dispersal of the Hudson River Plume;
Idealized sensitivity study to examine relationship between various aspects of circulation to learn the adjoint techniques and debug the code
Adjoint sensitivity studies to determine the impact of information contained in CODAR surface velocity data on different aspects of ocean circulation; this forms the basis for observing system design
Implement 4dVar assimilation of CODAR data into ROMS

4. Forecast System Configuration
Domain:
central Long Island to mouth of Delaware Bay, out to approximately the 100-m isobath
1 km horizontal resolution; 30 vertical s-levels
Forcing:
NCEP ETA hourly forecast u10 winds, Tair, qair, cloud from OPeNDAP server
Daily mean Hudson River flow from USGS Mohawk and Fort Edward gauges + persistence (for forecast)

5. Idealized Sensitivity Experiments
We choose SST anomaly as a forcing in the adjoint simulation. Then the results of the adjoint simulation show the sensitivity of SST anomaly at the region of interest to various other parameters throughout the model domain and back through time
The result of the idealized forward (nonlinear) experiment is used as a background state in the linearized adjoint simulation.
Different adjoint fields show the sensitivity of SST anomaly to variations in different model parameters.
Spatial and temporal detail in the adjoint fields give quantitative information on the relationship between SST anomaly and other model fields.

6. Sensitivity experiment 1 steady low Hudson River transport and no wind

7. Sensitivity experiment 1 steady low Hudson River transport and no wind
Adjoint SST
Surface Temperature, Velocity
This is sensitivity of SST anomaly in the marked box to the changes in ocean temperature over preceding 5 days throughout the model domain
Results show that SST anomaly comes from the Hudson River

8. Sensitivity Experiment 2 Hudson River Transport and Northward wind
Adjoint SST
Surface Temperature, Velocity
Hudson River Plume propagates to the East
SST at a marked box is influenced mostly by the wind and wind induced upwelling

9. Sensitivity Experiment 2 Hudson River Transport and Northward wind
Adjoint Temperature
Temperature

10. Sensitivity Experiment 3 Hudson River Transport and Southward wind
Adjoint SST
Surface Temperature, Velocity
Wind induced secondary circulation pushes the plume closer to the coast. SST anomaly is influenced by the Hudson River Plume (on a shorter time scale) and on the local SST (on a longer time scale) due to the wind induced downwelling

11. Next Steps
More work on interpretation of idealized sensitivity results
Realistic adjoint and sensitivity studies to determine the impact of information contained in CODAR surface velocity data in preparation of 4dVar data assimilation
Assimilation of synthetic data from a twin experiment
Assimilation of CODAR data