Variational Data Assimilation

Scott Smith, Clark Amerault, Charlie Barron, Timothy Campbell, Andrey Koch, Ole Smedstad, et.al.,

U.S. Naval Research Laboratory,
QinetiQ North America,
University of Southern Mississippi

CTA: CWO | Resources: Cray XC30 [NAVY, MS]; IBM iDataPlex [NAVY, MS]

DoD Impact/Significance

3D and 4D variational systems were tested and/or transitioned to improving Navy capabilities of forecasting Ocean environments. The work on coupling the assimilation systems between the ocean, waves, atmosphere, and acoustics made progress to further improve the forecast skill of the ocean environment.

Research Objectives

To advance analysis and prediction capabilities of Navy environmental modeling and forecasting systems by improvement of assimilation software. Three assimilation systems were primarily used in this project: RELO NCOM (3D-VAR), adjointless 4D-VAR, and 4D-VAR-NCOM.

Methodology

The advancement of 4D-VAR assimilation systems was supported by the following: 6.4 4D-VAR-NCOM RTP tested robustness. 6.2 Coupled Ocean-Atmosphere Variational Assimilation and Prediction System merged four-dimensional variational (4D-VAR) capabilities of the atmospheric and oceanic components COAMPS. 6.2 Coupled Ocean-Acoustic Assimilative Model for acoustic propagation forecast, 6.2 Extending Predictability in Coastal Environments project, 6.2 Calibration of Ocean Forcing with satellite Flux Estimates project, 6.1 Propagation and Dissipation of Internal Tides on Coastal Shelves used the 4DVAR- NCOM system, 6.1 Adjointless 4D-VAR for operational Navy ocean models.

Results

Even though the 4D-VAR-NCOM system has already been transitioned to NAVO, much effort was performed this fiscal year to improve its capabilities. Advancements were also made in understanding modeling and prediction of internal tides, heat fluxes, and error covariances. Good results were obtained from adjointless modelling, showing promise for continued exploring.

Figure/Image Description

Comparison between 3D-VAR and 4D-VAR from year-long experiments in the Okinawa Trough. (A and B) Comparison of 24-hr forecast RMS profile errors between 4D-VAR (red, blue, & green) and 3D-VAR (black) temperature (A) and salinity profiles (B). (C & D) Comparison of 24-hr (solid) and 96-hr (dashed) forecast RMS profile errors between 4D-VAR (red & blue) and 3D-VAR (black) experiments for temperature (C) and salinity profiles (D). (E & F) 2D difference histograms in Sonic Layer Depth counts between 3D-VAR and 4D-VAR analyses (left) and 96-hr forecasts (right). Blue (red) squares signify 4D-VAR has more (less) SLD combination counts than 3D-VAR.

Comparison between 3D-VAR and 4D-VAR from year-long experiments in the Okinawa Trough. (A and B) Comparison of 24-hr forecast RMS profile errors between 4D-VAR (red, blue, & green) and 3D-VAR (black) temperature (A) and salinity profiles (B). (C & D) Comparison of 24-hr (solid) and 96-hr (dashed) forecast RMS profile errors between 4D-VAR (red & blue) and 3D-VAR (black) experiments for temperature (C) and salinity profiles (D). (E & F) 2D difference histograms in Sonic Layer Depth counts between 3D-VAR and 4D-VAR analyses (left) and 96-hr forecasts (right). Blue (red) squares signify 4D-VAR has more (less) SLD combination counts than 3D-VAR.