NOPP Project: Boundary conditions, data assimilation, and predictability in coastal ocean models

NOPP Project:

Boundary conditions, data assimilation, and predictability in coastal ocean models

OSU: R. M. Samelson (lead PI), J. S. Allen, G. D. Egbert, A. Kurapov, R. N. MillerNRL: J. C. KindleNCAR: C. Snyder

Objectives:

•Determine impact of open ocean boundary conditions from GODAE Pacific Ocean models on Oregon coastal ocean models by comparison with in situ measurements

•Determine impact on Oregon coastal ocean simulations of assimilating satellite remote sensing observations

•Quantify uncertainty and predictability in coastal ocean simulations

+ Interactions with ongoing ONR, CIOSS, and GLOBEC funded projects on coastal ocean/atmosphere modeling and data assimilation

Coastal Model daily ave. surface T, velocities(7 Sept. 2005). Using

ROMS at 2 km resolution, forced with output from atmospheric ETA model

AVHRR Satellite SST(24 Aug. 2003)Isobaths: 100 and 200 m.

Spatial variability and flow-topography interactions in coastal flows:

Stronger interactions between coastal and interior oceanic flows in the observed fields than in the coastal model

Improvements in coastal transition zone (CTZ) modeling require nesting in larger scale models

The SSH snapshot from the 1/12o resolution Pacific HYCOM (15 July 2001) [J. Metzger, NRL].

The boundary of the regional West Coast HYCOM (also NCOM-CCS)

ROMS-based Oregon coastal model in white

Interaction of the shelf flows and interior ocean seen in model SSH

Model improvement in the CTZ should also result from assimilation of SSH, SST, and long-range HF radar data

Model SST and surf. currents (August 2000):

Coastal ocean interacts with interior ocean

Separation near Heceta Bank, Cape Blanco

Upwelling is weaker than observed (should be improved using higher resolution winds)

GLOBEC study: analysis of CCS model simulationModel solution for 2000 provided by E. Curchitser, currently being analyzed by CIOSS-supported post-doc B. J. Choi

Location of time series observations to be used for assimilation and validation

Black contours: H= 100, 200, and 1000 m

Daily ave. HF radar surface currents, (courtesy of P. M. Kosro at OSU).

Linearized models + rigorous (variational) DA

1) Theoretical models: help formulate model error statistics for practical applications [Scott et al., JPO, 2000, Kurapov et al., Mon. Wea Rev., 2002]

2) Internal tide, HF radar surface velocity data [Kurapov et al., JPO, 2003]

Fully nonlinear model + suboptimal, sequential DA (Optimal Interpolation)

1) HF radar surface velocity data [Oke et al., JGR, 2002]

2) moored ADP velocities [Kurapov et al., JGR, 2005a, 2005b, JPO, 2005]

“Dual Approach”

Application of tangent linear and adjoint ROMS for variational DA

Barotropically unstable jet in a channel [Kurapov and Di Lorenzo, 2005]

Forced-dissipative flows in the nearshore: ongoing research

Three-dimensional, stratified flows on shelf: ongoing research

Present focus: merger of these approaches:

Development of DA methods for the Coastal Ocean (research supported by ONR)

Advantages of variational DA: the study of M2 internal tide off Oregon

• HF radar data for summer 1998 (provided by P. M. Kosro) are assimilated in a linear frequency-domain model [Kurapov et al. JPO, 2003]

• DA corrects open boundary (OB) baroclinic tidal currents

• Inverse solution minimizes penalty function: J(u)= || OB error ||2 + || data error ||2

HF

HF

ADPADP

Representer-based minimization optimally projects surface observational information to 3D

(Improvement is verified to be obtained at ADP site)

Day, 1998

Dep

th

Tidal ellipses of the horizontal current (for a series of overlapped 2-week time windows)

depth-ave

Deviations from

depth-aveValidation ADP

DA

no DA

Time-ave baroclinic KE:

Surface Bottom

lat 45.65N lat 45.55N

Inverse solutions provide a uniquely detailed picture of the spatial and temporal variability of the M2 internal tide

Surface tidal ellipses Day 139

Deviations from depth-ave

(gray ellipses rotate CW)

Depth-ave (white ellipses rotate CCW)

Experience assimilating data into the fully nonlinear, primitive eqn. model of wind-driven shelf circulation (studies of summer upwelling)

Dynamics: Princeton Ocean Model (free surface, nonlinear, primitive eqn., w/ turbulence parameterization [Mellor & Yamada 1982])

-Realistic bathymetry-Boundary conditions: periodic (south to north)-Forcing: alongshore wind stress and heat flux

HF radars (HF radars (Kosro))

Moorings (ADP, T, S: Moorings (ADP, T, S: Levine, Kosro, Boyd))

Optimal Interpolation (OI) Data assimilation:

- sequential, optimal interpolation (OI) - correction is added in small increments every time step

,a ft t t POMν ν

( )a f ft t t t ν ν G obs Hν

matrix matching observations to state vector

- correction only to u:

-correction term is present in momentum equations

-however, equations for T, S, q2, q2l are dynamically balanced (which facilitates their term balance analysis)

- Approximate gain matrix obtained from an ensemble of model runs

Time-invariant gain matrix

||Error|| model w/out DA

DAforecast

analysis

Time

Effects of DA: improvement in near-bottom currents

alongshore velocity at NH10 mooring [Kosro]

close to surface

close to bottom

model-data corr.: 0.52 (no DA) 0.83 (DA)

rms error: 8.1 (no DA) 4.4 cm s-1 (DA)Assimilation sites:

Effects of ADP velocity DA: improvement in near-shore SSH time series

Assimilation of velocity observations in shelf circulation models can improve accuracy of SSH maps in the coastal zone, where altimetry is not available

comparison with coastal tide gauge data near Newport

obs, no DA, DA

SSH, surf v, no DA SSH, surf v, DA surf v, HF radar [Kosro]Flow control over Stonewall Bank(Day 166, 2001)

Optimal Interpolation: limitations

OI corrects the ocean state, not forcing limited control over source of error

OI assumes time-invariant forecast error covariance, used to compute the gain matrix. State-dependent covariance is needed to predict events.

Observations (such as satellite SSH, SST, HF radar) will generally have to be processed into maps (without spatial or temporal gaps) before using OI-DA

GIM has potential of resolving these and some other deficiencies of OI.

Methodology has been developed for using GIM efficiently with nonlinear oceanic models [Chua and Bennett, 2001]. This technology is yet to be tried in the context of coastal ocean circulation modeling.

To use GIM, tangent linear and adjoint models have to be developed.

Our ongoing research is focused on GIM assimilation into nonlinear coastal models

Variational, representer based, generalized inverse method (GIM)