Article ID | Journal | Published Year | Pages | File Type |
---|---|---|---|---|
6922324 | Computers & Geosciences | 2016 | 9 Pages |
Abstract
With the Amazon EC2 Cloud becoming available as a viable platform for parallel computing, Earth System Models are increasingly interested in leveraging its capabilities towards improving climate projections. In particular, faced with long wait periods on high-end clusters, the elasticity of the Cloud presents a unique opportunity of potentially “infinite” availability of small-sized clusters running on high-performance instances. Among specific applications of this new paradigm, we show here how uncertainty quantification in climate projections of polar ice sheets (Antarctica and Greenland) can be significantly accelerated using the Cloud. Indeed, small-sized clusters are very efficient at delivering sensitivity and sampling analysis, core tools of uncertainty quantification. We demonstrate how this approach was used to carry out an extensive analysis of ice-flow projections on one of the largest basins in Greenland, the North-East Greenland Glacier, using the Ice Sheet System Model, the public-domain NASA-funded ice-flow modeling software. We show how errors in the projections were accurately quantified using Monte-Carlo sampling analysis on the EC2 Cloud, and how a judicious mix of high-end parallel computing and Cloud use can best leverage existing infrastructures, and significantly accelerate delivery of potentially ground-breaking climate projections, and in particular, enable uncertainty quantification that were previously impossible to achieve.
Related Topics
Physical Sciences and Engineering
Computer Science
Computer Science Applications
Authors
E. Larour, N. Schlegel,