Cyberinfrastructure for next-generation atmospheric models

Morning lectures in the Main Seminar Room of NCAR's Mesa Lab. CISL's people, technology, and facilities played a key role in the 2008 NCAR Advanced Study Program (ASP) colloquium entitled Numerical Techniques for Global Atmospheric Models. The colloquium, which was jointly supported by NSF, NASA, and the Department of Energy, was held at NCAR on 1-13 June in Boulder, Colorado. This working group of graduate students, model developers, and scientists analyzed 13 atmospheric dynamical cores that are being considered for next-generation models.

Dynamical cores solve the equations that describe the properties of the atmosphere over time, including its motion. Their development and selection is a critical part of the process of creating new atmospheric models. They are especially important now, because increased computer power is making new types of atmospheric models possible. The limitations of past supercomputers required separate development paths for regional-scale weather models and global-scale climate models. Current supercomputers that operate at tens of teraflops -- and the coming petascale systems running at thousands of teraflops -- will allow a single model to simulate atmospheric phenomena on multiple scales, from local to global. However, spanning these scales requires high computational efficiency and numerical accuracy of the dynamical cores. ASP colloquium participants performed hundreds of test runs to analyze these 13 cores, yielding insights into the advantages and disadvantages of each.

The workshop demonstrated the relevance of the emerging Earth System Knowledge Environment, a strategic priority within CISL that seeks to integrate and expand modeling and data services. The results from the workshop are available to the research community via a new science gateway developed collaboratively by the Earth System Curator and Earth System Grid (ESG) projects in CISL. The effort was also supported by NCAR's Cyberinfrastructure Strategic Initiative, which is developing a Science Gateway Framework that can be used to build a variety of custom portals. The teams jointly extended the Earth System Grid portal by adding structured information, or metadata, about the scientific and technical properties of the dynamical cores and by building new tools, such as comparison tables, that utilize that metadata. Modelers who examine the portal will find not just datasets for download, but a rich store of semantic information that enables them to better understand how and why results were obtained.

FY2008 accomplishments

Key accomplishments of FY2008 include:

Development of a prototype web portal that links output datasets with model descriptions, experiment specifications, and other semantic information
Deployment of the portal for the 2008 ASP colloquium entitled Numerical Techniques for Global Atmospheric Models
Consulting and facilities support for this colloquium, including computing time, data storage, and new user training

FY2009 plans

During FY2009, the collaborators who developed the prototype portal will turn to a larger project: the next Intergovernmental Panel on Climate Change (IPCC) assessment. There is a need for collaborative metadata development and for further evolution of portal capabilities that were introduced for the ASP colloquium. One of the major challenges is coordination with a broad international community. Collaborators anticipate working closely with the European project Metafor (Common Metadata for Climate Modelling Digital Repositories), using international bodies such as the World Climate Research Programme and the Global Organization for Earth System Science Portals (GO-ESSP) to help facilitate the interactions.

Project description

During all afternoon tutorial sessions, the students, mentors, and researchers ran models using the dynamical cores, then archived data and metadata on the NCAR Mass Storage System. This ASP colloquium attracted graduate students with backgrounds in atmospheric science, applied mathematics, and/or computer science, and it introduced them to the latest developments in weather and climate modeling. An elite group of lecturers, model developers, and mentors provided input and guidance for the two weeks of intensive analysis. A total of 38 students, 13 modeling mentors, 18 lecturers, and 4 organizers participated in this event that produced significant benefits for both the attendees and the atmospheric research community. The graduate students received training and experience in atmospheric science, modeling, and computer architectures, and they performed and archived more than 350 simulations from models that incorporated the 13 dynamical cores. The organizers now have a large archive of data for their research. The modeling community has a rich and timely intercomparison dataset -- described in detail using metadata and available through a science gateway -- that may be used to compare and improve the dynamical cores.

The NSF-funded Curator project provided front-line support for the colloquium. The goal of Curator is to contribute to comprehensive science portals in which model source code, simulations, datasets, and individual model components are linked through metadata. This benefits model intercomparison projects, which require that modelers have information about the algorithms and codes they are analyzing. A specific benefit to modelers is the effort to standardize how datasets are linked to models: how they were generated, what flags were set, which input datasets were used, what initial conditions were used, and so on. Collaborators working on other aspects of the Curator project include the Georgia Institute of Technology, the Massachusetts Institute of Technology, Princeton University, and the European Metafor project.

Rather than trying to build a computing environment from scratch, the Curator project enhanced the Earth System Grid portal, which had an initial ontology -- a structured collection of metadata -- in place. Sylvia Murphy of Earth System Curator coordinated the joint effort, organized the development of new metadata with Luca Cinquini of ESG, and was the primary point of contact for the colloquium organizers. CISL software engineer Julien Chastang created several specialized tools for use with the enhanced metadata. One is a "trackback" page that details the model configuration associated with a particular dataset. It shows the scientific and technical properties for a model as a whole, and for each constituent component in the model. A second tool enables modelers to dynamically create comparison tables for a select set of models or components and a select set of properties. For the ASP colloquium, these tools helped participants track and understand the characteristics of each core, including information about the resolution, time step, spatial filtering, and input datasets that were used for each simulation.

Since ESG will be a primary portal for hosting IPCC Fifth Assessment data, and the enhancements from Curator will be folded into the main ESG portal, the improvements in metadata and tools arising from this ASP colloquium will benefit the next IPCC assessment.

Timeframe for the project

The ASP Colloquium was a one-time event. The portal developed to support it was the culmination of the Curator project, a three year, NSF-funded effort whose first phase ended during FY2008. The project will continue for another two years with NASA funding for the NCAR participants.

Accomplishments overview

CISL deployed a new science portal for archiving data and metadata for the colloquium. Building on CISL's strategic effort to develop a flexible Science Gateway Framework, the new portal was developed jointly by CISL staff working on the Earth System Curator and Earth System Grid projects. CISL also provided facilities support: for this colloquium alone, the NCAR Mass Storage System (MSS) received a 1.1 TB archive of model output. CISL staff wrote software to automatically load the data from the MSS onto the new portal, publish it on the web, and upload the metadata for each test case.

Supporting 38 new users and providing the appropriate cyberinfrastructure for the project required a significant commitment from CISL:

Allocated dedicated time on NCAR's 624-processor 4.7-TFLOPS supercomputer bluevista for colloquium participants
Oriented the new users to the NCAR computing facility
Provided training tailored to the project
Presented a lecture on the challenges of employing massively parallel processing technology
Implemented a new job queue for the project
Issued user names and one-time password devices to new users
Produced daily usage reports for the project
Provided laptops for some of the participants

Evaluation measures and outcomes

Output of nine models running test case 5-0-0 (Mountain-induced Rossby wave): 700 hPa zonal wind at day 15. The test starts with balanced and isothermal initial conditions. A 2-km-high Gaussian-hill-shaped mountain is placed at [90°E, 30°N] (not shown) to trigger Rossby waves. The test evaluates the treatment of the orography and reveals numerical noise (especially at later days). The four primary colloquium organizers will prepare a research paper for publication describing how this approach to testing dynamical cores can benefit the field (the full test case formulation used in the colloquium will be available as an NCAR technical report shortly). Christiane Jablonowski (University of Michigan) and Peter H. Lauritzen (NCAR, ESSL, CGD) created a specific set of idealized test cases designed to highlight strengths and find weaknesses in the cores. Christiane and Peter are now analyzing the data generated during the colloquium to understand how both the test suite and the cores performed. The testing approach used for the colloquium has the potential to become a community standard for testing dynamical cores. The ongoing analysis and their paper will further develop this idea. The organizers are editing a book based on contributions from the colloquium lecturers that is going to be published in the Springer series Lecture Notes in Computational Science and Engineering. Curator collaborators have also prepared a paper on their methodology and findings, submitted to the Earth Science Informatics Journal.

About NCAR's Advanced Study Program

NCAR's Advanced Study Program encourages the development of young scientists in the fields of atmospheric and related science, and it directs attention to timely scientific areas needing special emphasis. The ASP also helps to organize new science initiatives, supports interactions with universities, and promotes continuing education at NCAR. Sponsored by the National Science Foundation and NASA, the 2008 ASP summer colloquium's primary organizers were Peter H. Lauritzen (NCAR, ESSL, CGD), Christiane Jablonowski (University of Michigan), Mark Taylor (Sandia National Laboratories), and Ramachandran D. Nair (NCAR, CISL, IMAGe). Four distinguished lecturers presented the four keynotes: Professor John Thuburn, Professor of Applied Mathematics in the School of Engineering, Computer Science, and Mathematics at the University of Exeter, U.K.; Professor Dale Durran, Professor and Chair of Atmospheric Sciences, Adjunct Professor of Applied Mathematics, Atmospheric Sciences at the University of Washington; Dr. Todd Ringler, Climate, Ocean, and Sea Ice Modeling Group, Theoretical Division at the Los Alamos National Laboratory; and Professor David A. Randall, Department of Atmospheric Science at Colorado State University.

Some of the participants in NCAR's 1-13 June 2008 ASP Colloquium on Numerical Techniques for Global Atmospheric Models.

Director's Message