Integrating a scalable and effcient semi-Lagrangian multi-tracer transport scheme in HOMME

详细信息	查看全文 | 推荐本文 |

• 作者：Christoph Erath^a ; ^b ; ^{erath@ucar.edu} ; Peter H. Lauritzen^c ; Jose H. Garcia^b ; Henry M. Tufo^a ; ^b
• 关键词：transport scheme ; spherical geometry ; cubed-sphere grid ; semi-Lagrangian ; parallel scalability ; performance
• 刊名：Procedia Computer Science
• 出版年：2012
• 期刊代码：pc74_18770509
• 类别：cp
• 出版时间：2012
• 卷：9
• 期：Complete
• 页码：994-1003
• 文件大小：286 K

摘要

With the inclusion of a scalable spectral-element-based version of the High-Order Method Modeling Environment (HOMME) in NCAR's Community Atmosphere Model (CAM) and the need for on the order of 100 prognostic tracers in next generation climate modeling efforts, there is an urgent need for a scalable, monotonic, and conservative multitracer advection scheme in HOMME. To address this, we integrate a Conservative Semi-LAgrangian Multi-tracer transport scheme (CSLAM) into HOMME. The algorithm is specifically re-designed for multiple processors and the cubed-sphere grid with the element-wise domain decomposition as employed by HOMME. Based on these spectral elements, we create a finite volume mesh and add a halo zone of cells since a higher-order CSLAM needs a patch of tracer values for a reconstruction that extend the element. Therefore, we extend HOMME to also be capable of communicating cell center data. Our data structures and design allow us to provide a numerical scheme that still conserves mass locally to machine precision and is also monotone on multiple processor systems, but reduces the communication of tracer values by one for each time step. This is based on the idea that we calculate the reconstruction coeffcients redundantly, in a cell of an element and in the corresponding cell in the halo zone. Our algorithm of CSLAM in HOMME is the first one for multiple processor systems and multi-tracers that is highly scalable and effcient.