An Approach for Coupling Diverse Geophysical and Dynamical Models



J.D. Dykes, R.A. Allard, C.A. Blain, B. Estrade, T. Keen,
L. Smedstad, and A. Wallcraft
Oceanography Division
M. Bettencourt
Center of Higher Learning/University of Southern Mississippi
G. Peggion
University of New Orleans

Introduction: Successful military operations in littoral waters rely on timely analyses and forecasts of a diverse range of environmental variables associated with geophysical and dynamical processes. The complex, interdependent processes of the littoral environment are captured using separate numerical models that simulate circulation and sediment dynamics driven by wind, waves, tides, and buoyancy over spatial scales that extend over kilometers to meters. Typically, meteorological and oceanographic models run in a loosely coupled fashion, each model sequentially provides information to the other through a cascade of individual forecasts. Data between models are exchanged via written files, with each model responsible for handling differences in grid structure, desired quantities, and/or file formats. For this case, model interdependence is usually limited to one-way and the time to forecast depends on all models.

Ideally, we would like to process various computations simultaneously and exchange information as needed without regard to variations in computational grids or file formats. One could envision providing the ability to connect any model to a common server capable of handling any grid configuration and providing the computed information to any other model in the needed form. This methodology has now become possible with the advent of the Model Coupling Environmental Library (MCEL). The implementation of MCEL for model coupling provides the following benefits: (1) allows one-way or two-way model interaction; (2) assumes the burden of handling grids of different structure; and (3) requires minimal intrusion in existing code. Other benefits include a reduction in development time for a coupled model system and more efficient computations due to the parallel, distributed construct of the MCEL.

Under the recent High Performance Simulation of Littoral Environments (HFSoLE) Challenge Project, we configured two applications to provide a rigorous test for MCEL-based model coupling. Success is defined as the appropriate and accurate transfer of information from one model to another and improved efficiency as compared to file-based, sequential coupling. All models in the coupled model suite have been individually validated via model-data comparisons. The results demonstrate enormous potential for the MCEL to revolutionize the simulation of complex environments having diverse, interdependent processes by facilitating the coupling of models that represent these processes.

Description of Tests: Two applications designed to investigate the MCEL performance involved the coupling of five geophysical and dynamical models. The first model, the Hybrid Coordinate Ocean Model (HYCOM), captures the large-scale wind and thermodynamic-driven ocean circulation using a structured horizontal grid and a unique hybrid vertical gridding strategy. HYCOM simulations provide open-ocean forcing for the shelf-scale ocean circulation simulated by the Navy Coastal Ocean Model (NCOM), also a wind and thermodynamic-driven ocean circulation model. Results from the NCOM model in turn provide water-level forcing for the Advanced Circulation model (ADCIRC), a hydrodynamic model that is particularly well-suited for simulating tide and surge dynamics in the coastal areas at very high resolution. ADCIRC computations occur over an unstructured triangular finite-element mesh. Currents and water elevations from ADCIRC then drive a shallow-water wave model, the Simulating Waves Nearshore (SWAN) model in the nearshore. The Littoral Sediment Optical Model (LSOM), a sediment transport and water clarity model developed at NRL, is also focused on very shallow coastal waters, receiving wave information from SWAN and currents from the ADCIRC model.

The two domains for which the coupled model system is applied are the Mississippi Bight in the northeast Gulf of Mexico and the Spanish coastal waters, both shown in Fig. 3. One-way coupling is depicted in the conceptual diagram of Fig. 4, which highlights the connections between each model and the role of the MCEL server in facilitating coupling. To ensure portability of the coupled model system and the MCEL server, the applications are run on three different high-performance computer architectures at two DOD Major Shared Resource Centers.

Figure 3 Image
FIGURE 3
Five models running over domains of varying size and structure are connected to the MCEL server, which handles the interpolation of gridded fields regardless of domain diversity.
Figure 4 Image
FIGURE 4
These participating models were connected to the MCEL server. The black arrows depict the flow of data as model output to the server. The green arrows show the flow of certain interpolated parameters, as required by the receiving model.

Results: We are able to validate the coupled model computations using the MCEL server by comparing to identical file-based coupling applications. One would expect that connection to the server would result in minimal degradation of accuracy; indeed this is observed with errors limited only by model precision. We found across computational platforms differences on the order of machine precision, an excellent indication of the robustness of MCEL. Significant improvement in computational time is realized using the MCEL server for the five-model coupled system. The typical coupled model application on the IBM SP4 experienced a reduction of more than 33% when using the MCEL server. Validation of the coupled model system is accomplished by comparison to available observations. Figure 5 shows sample comparisons for ADCIRC and SWAN vs buoy data. Reasonable representations of observed dynamics further reinforce the validity of the MCEL approach for model coupling.

Figure 5 Image
FIGURE 5
Sample comparisons between model output and observational data gave the assurance that, under these test conditions, the models are functioning as they were originally designed. The error bars in the SWAN vs buoy comparison are ±0.15 m.

Conclusions: The efficiency, portability and accuracy of the MCEL approach for model coupling has been demonstrated. MCEL connects geophysical and dynamical models that represent a range of processes and spatial scales and handles a diversity of grid structures. MCEL requires minimal intervention within existing model codes, which facilitates its ease of use. The potential of the MCEL server is unlimited. Increasingly complex coupled processes are capable of being represented by numerical models today. The MCEL server releases researchers from the mechanics of model coupling, allowing their focus to turn to the unknowns of coupled processes.

Acknowledgments: The HFSoLE Portfolio was funded under the Common High Performance Software Support Initiative (CHSSI) of the Department of Defense (DOD) High Performance Computing Modernization Program (HPCMP). Special thanks to Mr. Carl Szczechowski of NAVOCEANO for serving as an independent reviewer of the MCEL coupled model system.

[Sponsored by ONR]

Reference
1R. Allard, A. Wallcraft, C.A. Blain, M. Cobb, L. Smedstad, P. Hogan, T. Keen, S. Howington, C. Berger, J.M. Smith, and
M. Bettencourt, "Providing the Warfighter Information Superiority in Littoral Waters," High Performance Computing Contributions to DOD Mission Success 2002, March 2003, pp. 78-80.