October 2019

Delving deeper into deep earth imaging

  • By Tim Munday, Lead, Deep Earth Imaging Future Science Platform, CSIRO

CSIRO’s Deep Earth Imaging Future Science Platform continues to lead the way in conducting vital research into the imaging, conceptualisation and prediction of resources

Since its inception in 2017, CSIRO’s Deep Earth Imaging Future Science Platform (DEI FSP) has conducted fundamental research in the imaging, conceptualisation and prediction of water, energy and mineral resources. In one of the largest recent recruitment campaigns in the earth sciences in Australia, a transdisciplinary  team of 20 early career researchers, four mid-career researchers and a Platform Director was created with the aim of contributing to the national mission of securing the future supply of resources from the subsurface. DEI FSP is now well established on a path that will provide Australia and its mining industry with the next generation of algorithms and tools to support the sector in exploring at greater depths across the continent and offshore. 

Our techniques and their applications

DEI FSP’s imaging techniques, employing passive and active methods, can be used to identify/characterise a wide range of subsurface features. This includes the major crustal scale structures underpinning mineral systems, the architecture of sedimentary basins hosting hydrocarbons and mineral deposits, and near surface characteristics such as the water table. 

On the continental scale, our team has focused on value-adding to precompetitive datasets. Geoscience Australia’s extensive active seismic acquisition campaign over the past decades has resulted in multiple crustal seismic transects across the continent. However, the broad application of seismic velocity model building, or more generally construction of geological models, has not kept pace with the release of these high-quality datasets due to the labour-intensive process of analysing them. By combining machine learning techniques and geophysical inversion, DEI is developing automated workflows to construct geological models from seismic data. 

Recordings of the seismic wavefield contain a wealth of information about the earth. Traditionally, only a small subset of this information is extracted and used to constrain the subsurface. DEI has developed new methods that allow prediction of the full wavefield response of an earth model, which is key to improving the quality/resolution of subsurface images.  We have combined forward modelling techniques with a Bayesian approach, and by using high-performance computing infrastructure we have been able to produce images of the complex geology beneath places such as the Lord Howe Rise deep sea plateau off eastern Australia. 

‘DEI is developing automated workflows to construct geological models from seismic data.’

The increased interest in passive seismic data sets (both acquired and applied) across the minerals and oil and gas sectors has also been a focus for DEI FSP researchers. To extract the maximum amount of subsurface information from these unique data, particularly those acquired at continental scale by universities and government agencies, early career researchers within DEI have been developing workflows to extract additional pieces of information from the wavefield that will help the creation of the next generation of seismic velocity models. Ultimately, these techniques are applicable across scales, ranging from mapping the boundaries of cratons all the way to defining the geometry and extent of the regolith at the mine-scale. In the future, we will seek to progress methods that will combine the information from both passive and active seismic data sets, with the goal of  improving the spatial resolution of our models. 

Making the most of magnetotelluric data

Another important source of information about the subsurface is magnetotelluric data. Measuring the ratio of the electric and magnetic field variations provides an indirect measure of the electrical resistivity distribution in the subsurface. Understanding the robustness of images inferred from these measurements depends on understanding how the noise in our recordings of the electrical and magnetic field affects processing and analysis of the data. DEI has therefore developed novel processing techniques for magnetotelluric data that permit the quantification of the noise in the data, thereby helping users to more comprehensively assess the robustness of models of the subsurface electrical resistivity. Understanding the robustness of models is particularly important for our fusion of different sources of information – for example, point estimates of cover thickness obtained using multiple geophysical techniques. Our cover thickness estimate assimilator provides a cover thickness map and associated uncertainties while accounting for the uncertainties of the individual point estimates of cover thickness and breaks in spatial correlation resulting from faults.

Direct and indirect observations are only one source of information when it comes to deriving subsurface models. Current exploration approaches commonly seek to image the subsurface. DEI is developing workflows centred around the numerical simulation of processes related to the formation of sedimentary basins, aquifers and alteration zones associated with a mineral system. Most geophysical inverse problems are non-unique; ie, if one model can be found that fits the available data, then it is likely that there is a wide range of alternative models with some different characteristics that can fit these data equally well. Some of these models may fit the data well but might not be plausible from a geological perspective. Therefore, they may have little relevance in a resource exploration context. 

Our ability to simulate the formation of structures that host resources or orebodies allows us to determine which models are geologically plausible. This then becomes a valuable source of prior information when it comes to inferring models of the subsurface from geoscience data sets, particularly geophysical data. An ability to generate plausible models and geophysical observations for these models also allows us to generate training datasets for machine learning algorithms. Once the training of these algorithms is complete, they provide a rapid screening tool to identify prospective domains of the subsurface prior to employing or developing more accurate and comprehensive inference methods.  

The challenge of hidden variables

A common theme around DEI’s development of methods for the inference of subsurface models involves addressing the challenge posed by hidden or latent variables.  Two examples of this are: 

  • where for well log data, we need to identify lithological units prior to inferring rock physics models for each lithology
  • where in the calibration of a reactive transport model for a mineral system, we must first identify alteration zones based on our geochemical field observations. 

Traditional approaches to these types of inference problems are sequential, with the first step being a qualitative interpretation of the data seeking to label or classify the data, and the subsequent step being an inference of a subsurface model from the classified data. This labelling, or classification of the data, is often a time-consuming step that is subject to qualitative interpretation. By using methods that combine unsupervised learning with model inference, we can treat this as a joint problem and derive objective models of the subsurface that do not rely on a potentially subjective, and sometimes erroneous, interpretation of the data.

Multiple applications of the workflows and algorithms

While Deep Earth Imaging aims to develop methods that answer specific questions for a specific resource problem, rather than focusing on more generic approaches, the workflows and algorithms under development do have multiple applications. For example, extracting information about available groundwater resources through the interpretation of geodata (eg airborne electromagnetics, magnetics, digital elevation models, Interferometric Synthetic Aperture Radar (InSAR)) can involve a determination of regolith thickness, palaeovalley location and fault networks. 

On a fundamental level these methods are applicable to other questions an explorer may have – for example, information about palaeovalley structures is also central to understanding the formation of channel iron ore deposits. Due to its transdisciplinary nature, the DEI FSP can readily exploit such synergies. 

Supporting the next generation of earth scientists

As much as the Future Science Platform is about developing and applying its science, it is also about supporting opportunities for the next generation of earth scientists to transition into positions in industry, academia and government. The early career researchers who joined DEI have taken advantage of training in a wide variety of subjects including project management and mathematical inverse theory. Combined with this has been the opportunity for DEI to engage with world-leading scientists from Australian and international universities, research institutes and government bodies. Nine PhD projects have been supported by the Platform at selected Australian universities on topics of relevance to deep earth imaging. This further contributes to the development of a nationally relevant geoscience capability. 

This emergence of a collaborative geoscience innovation hub at DEI FSP was the result of CSIRO recognising the challenges facing Australia in securing its resource base. Over the next three years, the DEI FSP will continue its effort to address these challenges under the three pillars we refer to as Imaging, Conceptualisation and Prediction.


An ever-increasing volume of geodata drives the development of novel techniques that can produce snapshots of resource systems. Future breakthroughs will be based on advances in sensor networks and computational techniques to extract the maximum amount of information from our observations. 

‘The Future Science Platform is supporting opportunities for the next generation of earth scientists.’


Our understanding of mineral, energy and groundwater systems only increases when we combine images of the subsurface with geological knowledge. Confidence in predictions will increase if they are underpinned by formal interpretations of images and transparent conceptualisations of geological processes.


Accurate predictions are the key to de-risking exploration in geological complex settings and managing water and hydrocarbon resources. Improving prediction will require advancing inference capabilities so we can turn images and conceptualisations into insight and understanding. 

Underpinning this focus will be an increased effort in developing the next generation of inference techniques that combine advances in data processing, machine learning, model parametrisation, data collection, forward modelling, inversion and predictive applications. We will become more outward focused and the technologies we are developing will transition from a proof of concept stage to delivery for industry and government. 

Linked to this increased external engagement, the DEI will convene its inaugural  interdisciplinary subsurface conference (Sub20) around the themes of imaging, conceptualisation and prediction of water, energy and mineral resources, to be held in Perth in February 2020. Over two days, we will focus on the science required, developed and deployed by academia, industry and government to prospect today for the resources that will underpin our low energy future. 

The conference will include a combination of keynotes, panel discussions, presentations and networking opportunities. Emphasis will be placed on the next generation of techniques to acquire knowledge about the subsurface; by for example combining machine learning, forward modelling, inverse theory and predictive applications. We look forward to your participation.

Image 1: Johan Swanepoel/Shutterstock.com. 

Image 2: Edward Haylan/Shutterstock.com.

Share This Article