Ocean State Estimation Projects

Fits and Forecasts in the CalCOFI Region:
Mesoscale Dynamics

The California Current System (CCS) is among the most biologically productive, and economically important regions in the ocean. A specific subregion of the CCS has been studied extensively since 1949 by the California Cooperative Oceanic Fisheries Investigations (CalCOFI). However, because of the low space and time resolution of the sampling, the data from this program have been unable to resolve the mesoscale variability which dominates the biological productivity in this region. The focus of this project is to fit an ocean model to the available remotely sensed and in situ data to obtain a dynamically consistent representation of the flow for a given 3-week CalCOFI survey. Then we will test the model in forecast mode to quantify predictability limits and validate the hindcast fit physics.

The CalCOFI data provide a very long time series (1949-present) of hydrographic measurements (including nutrients and chlorophyll) in the Southern California Bight, with the present quarterly sampling plan (Fig 1) extending back to 1984. A key barrier synthesizing or modeling the observed physical and biological distributions distributions in this region has been the lack of a circulation model which could well resolve the three-dimensional flow fields over time and to which an ecosystem model could be coupled. Another principal problem in using that data for dynamical studies of the CCS has been the coarse spatial sampling (roughly 70 km) of the surveys which can often lead to aliasing of the highly energetic mesoscale variations now known to be ubiquitous in the CCS. Other datasets which exist now or are planned complement the CalCOFI data, such as underway ADCP data with GPS correction to absolute velocity; surface drifter observations, which carry additional sensors for measuring sea surface temperature and atmospheric pressure, chlorophyll fluorecence; and, remotely sensed sea surface height (TOPEX/Poseidon), sea surface temperature (AVHRR), ocean color (SeaWIFS), and wind (NSCAT and SSM/I, ERS-1/2).

One way to overcome the CCS hydrographic sampling problem is to combine the CalCOFI data, which yields large-scale structure and vertical profiling, with the ADCP and sea surface height (T/P and ERS-1/2) data, which provides near-surface mesoscale resolution over limited areas, with a numerical model which can simulate the time evolution of the complete data set while remaining constrained to ocean dynamics. This mix of data types should provide some redundancy necessary for validation of the circulation models.

Many different circulation modeling studies have attempted to reproduce the mean and seasonal structure of the California Current (e.g., Philander and Yoon, 1982; McCreary et al., 1992), its instabilities (e.g., Batteen et al., 1989; Auad et al., 1991; Haidvogel et al., 1991; Allen et al., 1991; Hurlburt et al., 1992) and its locally and remotely forced variations (e.g., Pares-Sierra and O'Brien, 1989; Ramp et al., 1996; Miller et al., 1996). However, little attempt has been made to reproduce the time history of the temperature and current structure in detail (see Robinson et al., 1984 and Rienecker et al., 1987 for noteworthy exceptions).

The availability of the T/P and ERS-1/2 data provides an opportunity to attempt to follow the evolution of the flow between CalCOFI surveys and to consequently enhance the net spatial and temporal resolution of observed CCS features. If an eddy-resolving model, initialized from observed conditions, can reproduce observations over some time interval, the dynamics of observed CCS meanders could be definitively diagnosed for the first time. The fit of the model to the data provides a quantitative test of the model quality, either by explicitly including error terms in the fit or by observing larger than expected residuals after the fit. These errors are part of the bounds on the forecast skill of the model, and identifying them can point out ways to improve the model. Ideally, the model could be given sufficient skill to preserve information between quarterly CalCOFI surveys to allow redundant checks of the baroclinic structure. The eddy-resolving model can then be coupled to an ecosystem model to determine what physical forces and biological relationships are responsible for the mesoscale and large scale variability in the ecosystem as observed with both the CalCOFI field surveys and the ocean color sensors.

Model Development and Data Preparation

We use ROMS, the UCLA Regional Ocean Modeling System (ROMS), which is a descendant of the Rutgers SCRUM model, for our modeling studies. The ROMS is a sigma coordinate primitive equation model that has been parallelized and recoded for many improvements (e.g., improved pressure gradients and KPP mixing scheme, open boundary conditions, etc.). Our choice of grid, for the Southern California Bight (CalCOFI) Region from San Francisco to Baja, from the coast 700 km offshore, at 10km resolution, fits into the larger UCLA West Coast Domain.

We have run the model in seasonal cycle mode for many years to get some statistics of the model eddy evolution. We also have the model running from arbritrary initial conditions with arbitrary perturbations for model fitting tests. Our first test involved the July 1997 CalCOFI survey and is discussed below.

Model Fitting Tests

We initialized the model July 1997 CalCOFI temperature and salinity data by merging objective analyses of the T-S data with historical Levitus data outside the survey region. Geostrophic initial currents were estimated by assuming a level of no motion at 500m. Sea level height was estimated from the dynamic height of the geostrophic currents.

We started tests of our Green's function fitting strategy by using sine and cosines in the domain as the horizontal structure of the test perturbations, and vertical EOFs of temperature structure as the vertical structure functions.

We first tested for linearity in the model response by adding a domain-scale perturbation to temperature with first baroclinic mode structure and tiny amplitude 10exp(-3) C). One run with positive amplitude and one run with negative amplitude compared to the base run with no perturbation gives a measure of nonlinearity (second derivative) of the model sensitivity. We found substantial transfer of energy to grid scale occurring both on the shelf region, with strong nolinearity after 1 day, and in the open ocean, with strong nonlinearity after 5 days. We traced the source of the nonlinearity to the KPP scheme which computes the vertical diffusivity. We retreated to a constant vertical diffusivity and nearly completely eliminated this nonlinear effect. So we presently are following this route for model fits.

We tested our fitting procedure in `identical twin' experiments to be sure it was capable of retrieving the (here) known coefficients of the expansion functions for the initial (here specified) error.

We then tested the model fitting strategy on the real T-S data from CalCOFI July 1997. Using only 350 sine and cosine functions (up to circular truncation k_max=3) for a single vertical mode and for both T and S simultaneously, we achieved a 55% reduction in model-data misfit variance. This matched almost exactly the linear inverse expected misfit variance reduction. Thus, we feel the technique is successfully working in the CalCOFI domain and more detailed fitting can commence.

Miller, A. J., J. C. McWilliams, N. Schneider, J. S. Allen, J. A. Barth, R. C. Beardsley, F. P. Chavez, T. K. Chereskin, C. A. Edwards, R. L. Haney, K. A. Kelly, J. C. Kindle, L. N. Ly, J. R. Moisan, M. A. Noble, P. P. Niiler, L. Y. Oey, F. B. Schwing, R. K. Shearman, and M. S. Swenson, 1999:
Observing and modeling the California Current System.
Eos Trans. AGU, in press.

Mr. Emanuele Di Lorenzo (CRD/SIO)

Dr. Arthur J. Miller (CRD/SIO)

Dr. Bruce D. Cornuelle (PORD-CRD/SIO)

Dr. Douglas J. Neilson (CRD/SIO)

Prof John R. Moisan (Long Island University)

Dr. Teri Chereskin (PORD/SIO)

Prof Peter Niiler (PORD/SIO)

[ ROMS/SCRUM Adjoint | CALCOFI Mesoscale | CALCOFI Ecosystem | ECPC ]
[ CORC | Iceland/Faroe Fit | Estimation Projects Home Page ]
[Team Members | Publications ]
Email us at dneilson@ucsd.edu