Event Detail

Event Type: 
Department Colloquium
Date/Time: 
Monday, April 9, 2012 - 09:00 to 10:00
Location: 
Kidder 364

Speaker Info

Institution: 
Naval Postgraduate School
Abstract: 

Our group at the Department of Applied Mathematics at the Naval Postgraduate School is constructing numerical models primarily targeted for numerical weather prediction on petascale to exascale computers. Our goal is to develop both atmospheric and ocean models that are: high-order accurate, efficient, and highly scalable on CPU and GPU-based massively parallel computers. To this end, we have developed a suite of numerical methods including spectral element and discontinuous Galerkin methods, as well as implicit-explicit time-integrators in order to advance our model equations in time. So far we have developed two models.


The first model, the Nonhydrostatic Atmospheric Model of the Atmosphere (NUMA), is a three-dimensional unified model that has been shown to scale to tens of thousands of processors on CPU-based machines – up to the limit of one element per processor (strong scalability). This model solves the compressible Navier-Stokes equations and is used for representing both global (the entire planet) and mesoscale (local regions of the planet) in a unified way. NUMA is capable of using either spectral elements (SE) (continuous Galerkin methods) or discontinuous Galerkin (DG) methods; both methods are very similar but the DG method scales better on massively parallel computers even for small problems. NUMA has been verified on a suite of both two- and three-dimensional idealized test cases including tests with microphysics (simple moisture).


The second model, the Discontinuous Galerkin Coastal Ocean Model (DGCOM), is a fully unstructured (and eventually adaptive) model used to study both tsunami and storm-surge problems. This model is currently only two-dimensional and serial but a GPU-parallel (CUDA) version has been developed. DGCOM has been verified on a variety of idealized test cases and has been validated using real observational data from the 2004 Indian Ocean tsunami.


The aim of this talk is to give an overview of the models and methods that we are developing and to explore possible interests in either the models or methods. We are open to collaborations and are willing to work with interested researchers to explore new applications/developments of this modeling paradigm currently under development.