We develop accurate, efficient, robust and usable high performance computational models of coastal ocean hydrodynamics and transport that can be applied to real engineering problems.
These models compute water surface elevations, currents and transport of heat, salinity, pollutants and sediment within ocean basins, the continental shelf, estuaries, inlets, channels and adjacent floodplains.
These computational models evaluate and design for
- Coastal flooding due to hurricanes
- Currents for shipping operations, dredging, and harbor design
- Shoreline erosion/accretion and coastal morphology
- Sewage and waste heat disposal
- Water resources management – fisheries management
Our focus is on the ADCIRC community coastal ocean model. ADCIRC represents the multi-scale multi-process physics of the coastal environment by:
- Incorporating the relevant processes including riverine flows, tides, wind,
atmospheric pressure, wind waves, rainfall and hydrologic runoff, and
- Defining the physical system as it is observed
- Numerically resolving the physical system and the energetic scales of
- Applying correct boundary conditions
- Applying accurate non-diffusive discretization algorithms
With the recent rapid growth in computing power, high resolution observational descriptions of the system and processes, as well as developments in accurate, fast and robust adaptive algorithms, truly predictive multi-process and multi-scale simulations can now be performed that allow the various processes to interact and evolve as they are observed.
Algorithm Development and Analysis
As computational power rapidly increases, we continue to refine grid resolution and add physical processes to our computations. This results in faster and higher gradient flows, necessitating further refinements in the underlying algorithms in terms of consistency, accuracy and stability.
We develop, analyze, test and verify accurate, robust, stable and efficient unstructured mesh algorithms for our highly energetic scale content rich flows.
We apply unstructured mesh finite element based methods that allow us to resolve the mesh locally where gradients in the solutions are large. These methods are ideally suited for the wide range of scales that occur in coastal flow and transport problems. We also investigate how to define optimal finite element meshes that efficiently resolve the necessary energetic scales of motion.
We work with both Continuous Galerkin (CG) methods as well as newer Discontinuous Galerkin (DG) methods. DG algorithms appear to have significant advantages in terms of advection dominant and propagation problems, can easily implement h-p adaptivity and are mass conservative on the elemental level.
Accurate algorithms and meshes lead to better model physics.
High Performance Code Development
We are developing scalable codes that apply 10,000's of computational cores to allow us to compute more processes and to better resolve these processes within 15 minutes of wall clock time per day of simulation.
Coding paradigms are rapidly evolving as PetaFlop computers are becoming a reality. We focus on developing massively parallel code using domain decomposition and MPI and loop level optimization for vectorization and cache utilization.
In addition, we are using separate processors to manage disk output for our very large unstructured meshes as well as separate processors to produce GMT based visualizations during the simulation.
Optimal performance allows the model to apply more detailed meshes/more computational points and therefore to better resolve the physics.
Verification and Validation
We focus on proving that we are correctly solving the governing partial differential equations without introducing artificial damping or artificial modes that alter the stated physics. We prove consistency and convergence rates by examining problems with a range of dynamic balances so that our analytical tests mimic our real world applications. We base error estimates on comparisons to simplified problems with analytical solutions or on Richardson extrapolation.
We use our models to solve real world problems applying observed geometry and boundary conditions, universal air-sea momentum transfer coefficients and frictional dissipation parameters, ideally based on small scale process specific measurements, so that a thorough evaluation of the model's performance can be made. We compute model to measurement error estimates and when possible measurement error estimates to define model errors.
Our codes and models have been extensively applied in worldwide applications. Our current work focuses on the Gulf of Mexico and we have developed tide, riverine flow and hurricane storm surge models for Louisiana, Mississippi and Texas. Our SL15 model of Southern Louisiana and Mississippi is being applied to redesign the Hurricane Protection System in Southern Louisiana as well as being used to establish flood elevation levels for the FEMA Digital Flood Insurance Rate Maps (DFIRMS). Our Texas storm surge model is also being applied to establish FEMA DFIRMS for coastal Texas.
ADCIRC elevation contours (m) and wind vectors (m s-1) for Hurricane Katrina
at 0700 UTC 29 August 2005, for southeastern Louisiana