A Future Vision for Electromagnetic Geophysical Data Processing and Modelling on High Performance Computing
ALCS 2019 aims to bring together scientists from five key computational science disciplines in Australasia. ALCS aims to share knowledge about HPC and HPD in practice within and across these powerful research communities: astronomy, genomics, geosciences, climate and weather, and materials science.
Electromagnetic (EM) surveys collect data imaging the conductivity of Earth’s subsurface. EM surveys have been conducted in Australia since the 1970’s at local, regional and continental scales and at depths varying from 10s of meters to 100s of kilometers. Very little of the primary time series data collected so far with public research funding is FAIR (findable, accessible, interoperable and reusable) online. Currently EM data is typically processed on a survey by survey basis and only secondary data products are released, and consequently the full potential of Australian EM data has yet to be realised for either academic research projects or for potential resource industry applications (minerals, petroleum).
Nevertheless, the increased availability of computational power at NCI provides new opportunities, such as the creation of multiscale conductivity models using different EM data types (airborne electromagnetics, magnetotellurics, geomagnetic soundings) that characterise the Australian subsurface with different resolutions, from the upper few hundred meters to lower crust and upper mantle depths. Massive parallelized inversions could be run at once, using both newly acquired data and reprocessed legacy datasets. Achieving this ambitious vision first requires that relevant geophysical datasets are available in a high performance data format (HPD), usable on HPC infrastructure with sufficient metadata to enable (re)use and aggregation into multiple, composite datasets. Processing and inversion codes will need to be modernised for scalable computing, and probabilistic modeling algorithms will need to be developed to incorporate uncertainty into the derived models, which can be updated as new data are acquired.