Experience and lessons learned regarding configuration and control of an advanced 4 dimensional variational satellite data assimilation system
Scott Longmore, CIRA/Colorado State Univ., Fort Collins, CO; and A. Jones, A. Carheden, T. Vukicevic, and T. H. Vonder Haar
The Regional Atmospheric Modeling and Data Assimilation System (RAMDAS) is a 4 dimensional variational analysis (4DVAR) data assimilation algorithm developed at the Cooperative Institute for Research in the Atmosphere (CIRA) at Colorado State University (CSU). The current version of RAMDAS assimilates satellite observations from the Geostationary Operational Environmental Satellite (GOES) and assimilation of other satellite observations are in progress using the same generalized 4DVAR framework. This work focuses on the computational and configuration aspects of the RAMDAS 4DVAR implementation within the CIRA university research environment. Aspects of our experience could be relevant to other 4DVAR research and operational environments (govt., military, private, etc).
RAMDAS is implemented on an Intel processor-based High Performance Computing (HPC) 18-node cluster using the SuSE Linux operating system and is connected through a gigabit Ethernet switch. The RAMDAS software is a set of executables utilizing the Message Passing Interface (MPI) library for parallelization and is invoked through a master and set of sub-command scripts. The system is written in FORTRAN 77/90, C, and BASH, and is built using the Portland Group and GNU compilers. System output is generated in the HDF5 data format and MATLAB software tools are used for end-user visualization and analysis purposes.
The utilization of RAMDAS by academic users is facilitated through several software packages and research staff-developed processes and conventions. The RAMDAS software is stored and accessed using the Subversion version control system and is built using a set of GNU Makefiles within the RAMDAS repository. RAMDAS usage including master and subscript modification, satellite and model data set preparation, observation operator development. Output diagnostic and analysis examples as well as system configuration information are documented and updated using a MediaWiki on-line content management system. Training and education of new users is an important part of the process.
The RAMDAS development cycle includes the following steps: 1) user checkout and modification of a local copy of RAMDAS source code, 2) user local copy modification testing, 3) transfer and merge of user local copy source code modifications to RAMDAS administrator's local copy, 4) component and system testing of merged code, 5) submission to the RAMDAS repository. This work focuses on a description of these steps and the necessary components to make the 4DVAR system work within the CIRA research environment.
Extended Abstract (940K)
Poster Session 2, Poster Session II
Wednesday, 17 January 2007, 2:30 PM-4:00 PM, Exhibit Hall C
Previous paper Next paper
Browse or search entire meeting
AMS Home Page