how AI is transforming our geophysics data processing
From IP inversions to seismic refraction — a look at how machine intelligence is helping Groundsearch deliver faster, more accurate subsurface images.
Geophysics has always been a data-intensive discipline. A single induced polarisation (IP) or resistivity survey across a mineral exploration licence can generate hundreds of thousands of electrode readings. A multi-channel seismic reflection or MASW line produces equally dense waveform datasets. The science of turning those raw numbers into a reliable subsurface picture has traditionally relied on the experience of skilled interpreters — and weeks of iterative processing.
At Groundsearch, we have spent more than three decades developing that expertise across Southeast Asia, the Pacific, and beyond. Now, we are integrating artificial intelligence and machine learning directly into our processing workflows — not to replace the geophysicist, but to make their judgement faster, sharper, and better-informed.
"The goal is not automation for its own sake. It is getting a higher-quality answer into the client's hands in less time, with greater confidence."
We have identified three areas in our current workflows where AI-assisted tools are delivering measurable improvements.
Induced polarisation surveys are among our most powerful tools for mineral exploration and groundwater investigations. The technique measures both the resistivity of the ground and its chargeability — the tendency of earth materials to store electrical charge. Sulphide mineralisation, graphite, and clay-rich aquifer zones all produce distinctive chargeability signatures that resistivity alone cannot resolve.
The inversion of IP data is notoriously sensitive to noise and to the choice of starting model. We are now using a convolutional neural network, trained on a library of forward-modelled IP responses across geologically plausible models, to generate starting model candidates ranked by their likelihood of producing a stable inversion. This reduces the number of inversion iterations required by a meaningful margin and produces final models that our senior geophysicists validate in less review time.
Equally important, the AI flags zones in the inverted section where the data fit is poor — indicating either genuine geological complexity or residual noise — so the interpreter can direct attention where it is needed rather than reviewing every metre of a long section uniformly.
Traditional IP and resistivity inversion is an iterative, computationally expensive process. Choosing the right regularisation parameter, identifying and suppressing noise spikes, deciding where to truncate the model depth — these have always required human judgement backed by field experience. The same is true for seismic processing: velocity picking, static corrections, noise filtering, and first-break picking on refraction lines are time-consuming steps where small errors compound.
When a junior processor makes a conservative or incorrect parameter choice, the inversion can converge on a geologically implausible model. When first-break picks drift on noisy seismic data, velocity models become unreliable. In both cases, the client receives a less accurate result — and often does not know it.
Our seismic work spans MASW (multichannel analysis of surface waves) for geotechnical site investigations and refraction seismic for depth-to-bedrock mapping. Both methods depend critically on accurate velocity picking — and both generate data volumes that make manual picking genuinely burdensome on large projects.
We have integrated a supervised learning picker trained on labelled field records from Cambodia, Laos, and New Zealand. The model handles the broad spectrum of surface conditions we encounter across the region — from soft alluvial plains to residual tropical soils to hard rock outcrop — and produces first-break picks that our processors then quality-check and adjust rather than generate from scratch. The result is a faster turnaround without any reduction in the rigorous QC that defines our deliverables.
For MASW specifically, AI-assisted dispersion curve picking and automated mode identification are helping us extract more of the useful frequency bandwidth from each shot gather, which translates directly into greater depth penetration and better-resolved shear-wave velocity profiles.
"Southeast Asia's varied geology is precisely why our AI tools need to be trained on regional data. A model calibrated on temperate-zone datasets will underperform on the deeply weathered profiles we routinely encounter in Cambodia and Laos."
We want to be clear about what AI does and does not do in our workflow. It does not replace geological reasoning, field experience, or the professional responsibility of a qualified geophysicist. What it does is handle the repetitive, parameter-sensitive, high-volume aspects of data processing so that our team can spend more of their time on the parts of the work that genuinely require expert judgement — designing the right survey geometry, assessing the geological plausibility of an inverted model, integrating geophysical results with borehole, geochemical, and geological mapping data.
The client receives the same rigorous, independently quality-checked deliverable. What changes is how efficiently we get there — and in many cases, the quality of the underlying processing is higher because the AI-assisted workflow catches noise and artefacts that manual processing at speed can miss.
Whether you are exploring for gold or base metals in Southeast Asia, investigating a groundwater source in Cambodia, or commissioning a geotechnical site investigation ahead of construction, improved data processing means a clearer picture of what lies beneath your site — and a faster answer. It means fewer reprocessing cycles, more transparent uncertainty reporting, and results that hold up to independent review.
We are continuing to develop and refine these capabilities, with a particular focus on integrating AI-assisted QC directly into our field acquisition software so that data quality issues can be identified and addressed in real time, before the survey crew leaves site.
If you would like to discuss how our AI-enhanced processing workflows can benefit your next geophysics programme, we would be pleased to talk through your project requirements.
WhatsApp Groundsearch
Need help? Our team is just a message away