Current generation supercomputers provide geophysicists such as myself with unprecedented capabilities to obtain accurate maps of the Earth’s subsurface. The key ingredient for this is our ability to model the seismic response of the subsurface with sufficient realism, so that data recorded by geophones can be directly compared with synthetic data obtained from simulations.
In the past, simulations were limited to 2D slices of the subsurface, or just some parts of the physical response. A good example of this is the travel time of seismic energy from source to receiver. Current generation algorithms and computer clusters, however, can obtain the full 3D response in realistic scenarios at scales which are meaningful for seismic exploration of the first few kilometres of the subsurface.
Even more interesting is the fact that such modelling capabilities can give rise to some very advanced imaging algorithms, among which the most powerful is called Full Waveform Inversion (FWI). FWI directly maps slight differences in data with respect to our simulations into corrections to the physical properties of our 3D subsurface model. The problem is extremely complicated because there is not a single “best” solution to the problem and, furthermore, we can only improve our data to synthetic fit step by step, thus recursively improving the resolution of our image of the subsurface. Not unlike medical imaging, we are finding the opportunity to evaluate the properties of kilometres of subsurface without having to actually drill through them, which has a huge potential impact for geological resource management (e.g. hydrocarbons, mining, aquifers) as well as for our understanding of geological processes such active faults that may result in earthquakes. It is important to remark that this revolutionary improvement in our knowledge of the subsurface has been achieved from supercomputing and algorithmics alone, without a need for changing the amount or quality of the dataset used.
The capabilities of supercomputers today allow us to use methodologies such as FWI to reveal the properties of the Earth with unprecedented detail. We are currently solving problems with 200 million unknowns by fitting 500 billion data points, which is a huge number by any standards. Current work is leading towards breaking the billion unknowns barrier (that is similar detail to the resolution of a few hundred full HD displays stacked one on top of the other to build up an “image cube”, if you can picture that!).
A project such as HPC4E helps drive next generation technologies to the goal of obtaining faster algorithms, more realistic simulation capabilities and, overall, an increase in resolution of our maps of the subsurface. As a secondary benefit, and due to the relentless development of off-the-shelf hardware, our advanced imaging tools might be commonplace in few years, which means that a whole bunch of future applications might benefit from our enhanced capacity to “see” below our feet. This will be thanks, in part at least, to the current effort of the geophysical and computational science communities today in projects such as HPC4E.
Josep de la Puente
Geophysical Applications Group Leader, BSC
This article also appeared in the LinkedIn Group. Join us!
Other LinkedIn articles: