Background

Improvements in oceanographic and meteorological scientific understanding as well as predictions and projections of the earth's physical system are increasingly reliant on higher resolution models, model ensembles, and explicit coupling of the physical components into simulations where the ocean, atmosphere, wave and ice fields are allowed to co-evolve during the forward time integration. However, the increased computational demand that these efforts require is beyond current research and operational computational capacity and is projected to grow faster than traditional computational infrastructure increases can support. Additionally, capacity growth in the current generation of multi-core systems is becoming limited not only by acquisition cost but also by power, cooling, memory, and communications rather than processor speed. Emerging "commodity-based" processors such as Many-Integrated Cores (MIC) and Graphical Processor Units (GPUs) show promise for addressing these trends with projected 10-times theoretical improvement and likely at least 2-4 times improvement in realized cost and power requirements; and there are several national-level efforts to provide Peta-flop computing for computationally intensive research areas such as the environmental modeling and simulation community.

While developmental models can be structured "in-stride" in the underlying numerics and dynamics to fully exploit these new computational architectures, there is a need to revise the established environmental prediction models whose supporting physics, dynamics, data assimilation, accuracy and performance for different processes and case studies is well understood, stable and well characterized at a range of spatial resolutions and forecast lead times. Additionally, next generation coupled ensemble forecast systems will likely require common source code that can optionally take advantage of both legacy and emerging computational capacity to address reliability and consistency of results from these highly complex models on heterogeneous architectures.

Description

A significantly improved capability to simulate and predict the coupled global air-ocean-wave- land-ice system at eddy-resolving spatial scales in a computationally and operationally efficient and massively parallel architecture towards real-time, predictions is desired. Responders to this announcement proposed work in collaboration with the Naval Research Laboratory (NRL), Department of Energy (DOE), National Oceanic and Atmospheric Administration (NOAA), and/or National Center for Atmospheric Research (NCAR) laboratories; and should consider an interdisciplinary team of computer scientists, oceanographic and meteorological scientists, numerical methods experts, and software engineers. Projects use the Navy, DOE and NOAA's component models and conduct research towards the following scientific and engineering goals:

• Predictive simulations on heterogeneous architectures Central Processing Unit (CPU), MIC, GPU: identification of representative code patterns that either look particularly amenable or intractable to refactoring; establishment of pathways to maintain single source code compatible with multiple platforms; and determination of mechanisms to achieve optimal performance and portability.
• Identification of key bottlenecks in component models such as Navy Global Environmental Model (NAVGEM), Global Forecast System (GFS), Non-hydrostatic Unified Model of the Atmosphere (NUMA), Model for Prediction Across Scales (MPAS), High Resolution Atmospheric Model (HIRAM), Non-Hydrostatic Icosahedral Model (NIM), Community Earth System Model (CESM), Hybrid Coordinate Ocean Model (HYCOM), Modular Ocean Model (MOM), Community Ice Code (CICE), and WAVEWATCH 3 and minimization of the impact in resulting systems.
• Identification or development of software tools to facilitate code porting to maximize performance, assess memory and communication infrastructures efficiently, and minimize cross- platform communication and data copies.
• Addressing of language based or directive based compiler solutions in coordination with a model developers working group across similar national efforts at NOAA, NCAR, DOE, and National Aeronautics and Space Administration (NASA).
• Assessment of the performance in terms of speed and accuracy of the resulting multi-platform versions of the component models, coupling interfaces, data-assimilation, and post-processing against the current modeling suite.
• Use of these new multi-platform versions in the representation of air-sea coupled processes such as tropical cyclones, the Madden-Julian Oscillation, El Nino-Southern Oscillation (ENSO), Atlantic Meridional Mode (AMM), or the sub-seasonal to annual Arctic sea-ice evolution in coordination with the Earth System Prediction Capability (ESPC) initiative (www.espc.oar.noaa.gov).
• Creation of a working group of dynamical and observational environmental scientists, process modelers, model developers, and computer scientists to assess model performance and provide recommendations.

One technical goal of this topic is, by the end of the collective efforts, to produce a data- assimilative, eddy-resolving, high-resolution air-ocean-land-ice coupled prediction system suitable for medium-range to seasonal forecasts that might begin the transition process to U.S. operational environmental prediction centers. The NOPP-funded groups collectively perform the background scientific research necessary to design and test such a model, including the data assimilation component.