Download the paper from the conference proceedings
Abstract
Numerical simulations are often used to approximate a real-world geoscientific problem. Such simulations generally require both highly specific domain knowledge, and an in-depth knowledge of a specific numerical solver. Additionally, these simulations often run in very specific computational environments (eg Windows pools, Linux supercomputers, Linux clusters) that themselves require a steep learning curve in terms of gaining accounts, logging in, data formatting, library dependencies, moving data etc.
For a geoscientist to become fluent in all of the above areas requires a significant investment of time and effort outside of the scientific field of interest. Numerical solvers are normally specific to particular problem domains (eg 'mechanical deformation', 'fluid flow' 'chemical deposition' etc), and are usually highly complex pieces of code with their own specific language and syntax. Hence changing the numerical solver (for instance to a newer code that is faster or more scale-appropriate) becomes a daunting and rarely-done task. Even within the one solver, there are often so many ways to construct simulations for the same problem, that consistency and repeatability can suffer. A similar issue arises for using alternative computational resources to actually run the simulation on.
The Desktop Modelling Toolkit (DMT) buffers the user from these issues. It is a tool that allows the user to specify a geo-scientific problem on a conceptual level, without recourse to specific knowledge of the actual solver that will run the simulation; solver-specifc code is auto-generated by the DMT. It also manages interactions with Grid-Computing resources on the user's behalf, insulating the user from the specific requirements of the computational resource that the simulation will run on.
The DMT is aimed at modellers who wish to run simulations of mechanical deformation, fluid flow, thermal regimes and/or chemical interactions in the Earth’s crust; however, it is flexible enough that it could be modified to suit a range of other modelling domains, such as seismic processing, gravity inversion etc. It provides a fast-track to setting up and running multiple simulations (often required for parameter sweeps or inversion techniques) and is particularly useful for training new researchers or students. Through the enforcing of consistent algorithms and coding standards, it is also an important method of ensuring experimental repeatability.
About the speaker
Gordon German is a Research Scientist with CSIRO Exploration and Mining based at the Australian Resources Research Centre (ARRC) at Kensington, Perth, Western Australia, and is responsible for software engineering in the Computational Geoscience group. Gordon joined CSIRO in 2000. From 2000-03, he started and was responsible for the 3DMACS software project. Since then, he has been primarily responsible for the development of the Desktop Modelling Toolkit , which aims to provide a desktop interface for the running of large geoscientific numerical simulations. From 1999-2000, Gordon held a postdoctoral position at the School of Computing, Curtin University of Technology in Perth, Western Australia, and developed the Discrete Output Neural Network (DONNET) classifier. He also worked in the joint CSIRO/ Curtin University research program into software mineral modelling, object translation and Geographic Information System (GIS) interoperability. Prior to this, he was a lecturer for Curtin University’s Schools of Computing and Spatial Sciences.