Real-time data processing and inversion procedure

C. Naimie and C. Flagg

The numerical model at the center of the ship-board computational system implemented during the Real Time Data Assimilation study on Georges Bank during the 1999 field season was QUODDY – a 3-D, nonlinear, prognostic, tide-resolving model with advanced turbulence closure (Lynch et al 1996). This model was operated on a bank-scale domain. An iterative nonlinear inversion procedure constituted the forecast system, using QUODDY as the forward model and TRUXTON (Lynch et al 1998) as the inverse model. In this talk, I will address four important aspects of the forecast: the Best Prior Estimate of the circulation; the inversion of velocity data; the posterior estimate (or hindcast) of the circulation; and the forecast. Other aspects of the modeling system and results from its application are addressed by Lynch (herein), Werner et al (herein), and McGillicuddy (herein).

The Best Prior Estimate (BPE) of the circulation was computed using QUODDY simulation, forced as follows. Initial Conditions were established by a blend of hydrography via Objective Analysis from a variety of sources: the climatology, recent observations, and results from recent hindcasts. Meteorological Forcing was specified using hindcast/forecast wind stress and heat flux from the NCEP AVN model (see Werner et al (herein)). Boundary Conditions were determined by the sum of the climatological pressure signature from the Dartmouth archive (tidal plus seasonal mean) and the far-field Northwest Atlantic response to large-scale atmospheric forcing as computed using ADCIRC (see Werner et al (herein)).

Discrepancies between the observed horizontal velocity data (from the ship-board ADCP measurements and short-term drifter observations) and the BPE (“sampled” in a manner consistent with the velocity data) were inverted using TRUXTON (Lynch et al 1998) to improve the pressure boundary conditions. As indicated above, TRUXTON was executed within an iterative loop that included the nonlinear forward model QUODDY, yielding a nonlinear inversion (see Figure 1).

The posterior estimate (or hindcast) of the circulation was determined from the final forward model solution in the nonlinear iteration. The typical for these hindcasts was 1-5 days. The forecast is the projection into the future of this simulation for an additional 3 days, using forecast atmospheric and oceanic products and persistence of boundary conditions. The daily modeling system comprises of the following steps:

  1. Download NCEP AVN hindcast/forecast simulation results,

  2. Run ADCIRC on NW Atlantic domain to obtain far-field pressure response,

  3. Sample ADCIRC results at locations corresponding to boundary of local-scale domain,

  4. Transmit atmospheric and far-field boundary forcing to ship,

  5. Collect velocity data (ship-mounted ADCP and drifters) to be used during inversion,

  6. Run local-scale hindcast/forecast system,

  7. Archive results of simulation and provide ship-board scientists with standard displays of the model predictions.

Steps 1-3 occurred at the University of North Carolina, as described in Werner et al. (herein). Steps 5-7 occurred on ship. The modeling procedure was followed once per day and published via html for shipboard use. The total time-delay for the modeling system was typically 12-15 hours for simulations on Georges Bank. Thus, daily 3-day forecasts combined with 3-5 day hindcasts are archived on a daily basis.

Figure 1

Figure 1: Bank-scale finite element mesh and iterative nonlinear inversion procedure