Towards a Next generation Ocean Model in the Gung-Ho framework: 2D test cases (G-Ocean:2D)

点击翻译按钮获取中文摘要

This project aims to develop and test a new a software approach to coding ocean models that can exploit the next generation of computer architectures. Ocean models form a vital component of the climate models that produce future climate projections, for example, for the Inter-Governmental Panel on Climate Change. They are also important tools for exploring all aspects of the marine environment from coastal to shelf sea and global scales. The use of ocean models relies on national computer facilities that are among the fastest computers in the world. Computer power tends to approximately double every year, and as these facilities improve then so does the potential for ocean models to provide more accurate simulations, with benefits for climate and weather forecasting, as well as our understanding of the marine environment. However, increases in computer power are now occurring primarily through increased parallelism, with more computer cores per chip and more chips per computer. Hence to exploit increases in computer power we must develop models that can exploit as many different forms of parallelism as possible. While there are many ways of achieving this, they are generally at the expense of the ease of the development of the model, so that in time only a computer science expert would be able to develop the ocean model - an unreasonable expectation. One of the particular ways ocean scientists would want to use this increase in computer power is to target horizontal resolution where the scientific understanding dictates or where it is particularly important for the application.A solution to these computational issues has been identified in an on-going project to develop a new atmospheric model for the UK Met Office (GungHo) to meet many of the same challenges and opportunities identified here. The proposed solution is to separate the model computer code into layers, each requiring different expertise to develop, and so isolate the natural scientist from the complexities of the computer science aspects. While oceans and the atmosphere show many similarities in their physics, there are some important differences, notably the large changes in depth of the ocean and the presence of land leading to complex boundaries since the oceans do not cover the whole globe as the atmosphere does. This naturally leads to ocean modellers not necessarily making the same choices in model design as atmospheric modellers. Hence, the aim of this project is to apply the 'layered approach' to a simple ocean case to prove the concepts:1. That the computational framework under development in GungHo is sufficiently flexible to accommodate the natural choices of grids and solution approaches for ocean models.2. That, when coded within this framework, conventional solution approaches can perform at least as well as the existing models in terms of their efficiency and scalability, and also have benefits of ease of use and development when highly optimisedThis work will provide a first view of how an ocean model built and designed in the GungHo framework is likely to perform. The tools built here will allow us to explore ocean model design and help answer the question: Are the approaches being developed for GungHo appropriate for the ocean? Or will alternatives be needed? The long term impact of this work is potentially very far reaching. The vision is that this is the first step on the route to an ocean model that runs efficiently on hundreds of thousands to millions of computational cores and has flexibility to change resolution as the science or user interest dictates, but is also readily usable by oceanographers of many disciplines. Realising this vision would represent a step change in Earth System Modelling and Regional System Modelling capability that would be truly world leading.