By Matthew Hastings, Principal Resource Geologist, SRK and Alex Wilson, Lead Data Scientist, Seequent
Unlocking latent value: an artificial intelligence reinterpretation of a legacy gold deposit
Even well-drilled deposit datasets can hold surprises, and as the data density increases, important geological patterns often emerge, challenging early interpretations and opening new possibilities for building higher quality models.
That was the situation facing SRK Principal Resource Geologist Matthew Hastings as he revisited a client project that would benefit from a new look at a historic interpretation. The site has been in care and maintenance for several years and is now undergoing a reassessment for its latent potential. This review was prompted by the realisation that, despite extensive drilling, sampling, and mining, the supporting geological work remains outdated and based on legacy assumptions.
These limitations stem largely from early 2000s software capabilities and inherent constraints within the original dataset. As the project potentially begins a new life, SRK noted that updating the resource using legacy methods, which are dependent on unsupported software, would be impractical and time consuming. Thus, a challenge to take a legacy resource and bring it into the modern era evolved.
In collaboration with the client, SRK sought to use the opportunity to move beyond traditional manual methods of interpretation and explore whether AI could assist in the rapid creation of a high-quality implicit domain model in Leapfrog Geo. Although a geological model existed for the project, it historically held little influence over the grade distributions, which were modelled using rigid global indicator shells based on gold (Au) cut-off values.
The geological setting
SRK was presented with two adjacent deposits which show gold mineralisation occurring within quartz shears that crosscut metamorphic intrusive and sedimentary host rocks.
In the first deposit, mineralisation is strongly controlled by a regional structure which occurs within and along the margins of the deposit area. Although the interaction with high-angle faulting related to this regional structure is not well-understood, the mineralisation appears to trend along a general low angle regional strike and dip.
The second deposit contains mineralisation within several discrete shears, each exhibiting slightly different dipping-plane orientations. Historical reverse circulation drilling and legacy data collection practices have made detailed geological interpretation of the grade relationships difficult, amounting to a relatively simplistic legacy structural interpretation of the mineralisation trend. This interpretation approximated the mineralisation within each area as a global strike and dip, gently dipping and broadly following the major lithological boundaries.
AI analysis with Evo and Driver
A first look at the dataset in Leapfrog Geo highlighted the limitations of the legacy structural model. While this interpretation was acceptable for an initial approach, SRK deemed it would be insufficient for addressing the deposit’s more intricate geological features which suggested notable local complexities, including potential changes in the primary dip angle, abrupt grade discontinuities, and the presence of several, broad, open fold-like structures.
SRK first heard about Evo and Driver in a demonstration at a mining conference and immediately thought of the applications for this challenge. SRK quickly partnered with Seequent Lead Data Scientist Alexander Wilson to review the project and explore these new cloud capabilities, alongside a new connected workflow developed in Leapfrog Geo.
‘I was very curious to try out one of Seequents’ new modelling tools-Driver, which uses AI to generate quantitative local grade trends directly from the raw drilling data. I was keen to see if I could quickly validate the legacy global dip hypothesis and potentially improve on it by introducing some local structural controls that better reflected the geological complexity’ Hastings said.
Driver is built on Seequent Evo, which is a geoscience data and compute platform that enables integrated workflows and collaboration across Seequent and third-party products. Using Evo’s connected capabilities within Leapfrog Geo, Hastings was able to quickly upload the drilling data to the cloud, where it was automatically formatted and ready for a Driver analysis. He then set up a new project and started to explore the data under the guidance of Driver’s advanced algorithms.
Driver brings new AI capabilities to structural deposit modelling (Figure 1). The core algorithm, Spatial Continuity Mapping (SCM), runs an unsupervised learning procedure that quantifies a spatial representation of grade continuity, directly from the input dataset, which can include grade assays, or categorical information like lithology logs converted to binary arrays. The continuity detected by Driver is represented as local ellipsoids, which are positioned, oriented and shaped to represent the local geological trends formed by the feature of interest (eg enhanced-grade values). These trends can then be integrated into the geological interpretation and downstream models, providing additional support or challenge to prior assumptions of grade continuity.
Figure 1: Driver, a new web application built on Seequent Evo, brings new AI capabilities to structural deposit interpretation and modelling. Figure shows planar continuities of high-grade samples automatically segmented into spatially consistent clusters representing individual veins.
Uncovering a new story with Driver
Driver’s data-driven observations broadly validated the legacy, shallow dipping structural interpretation, which supports the notion that mineralisation continuity is generally gently dipping in a consistent way (Figure 2). However, Driver also provided a rapid assessment of the Locally Varying Anisotropy (LVA) which yield key insights into local structural complexities that were not captured in the legacy global interpolants. The automatically generated Driver ellipsoids indicate the presence of a broad open syncline in the northern part of the deposit and numerous areas where the local dips of shear zones deviate by more than 10-15 degrees away from the primary overall trend.
Most critically, in an area which is expected to be the target of near-term future mining, the spatial continuity ellipsoids revealed a dramatic deviation from the global norm, with a steepening of grade continuity outlining the apparent trace of a tight, overturned fold. While the relationship of this apparent fold is not clearly understood, the outputs from Driver provide compelling evidence of potential structural complexity related to the north-south, regional-scale faulting and deformation (Figure 2).
‘The more I looked, the more I started seeing evidence of a highly deformed system. Driver was clearly showing the mineralisation curving upwards into an overturned structure, before returning to the typical shallow dip in the west’ Wilson said.
Figure 2: Driver analysis results colored by automatic cluster groupings. A) Stereonet showing poles to planes of the legacy global mineraliszation trend (yellow star) and the Driver-generated local ellipsoids (coloured dots). B) Cross section slice showing drilling data within the proposed future mining area, colored by Au assay. C) Driver-generated local ellipsoids delineating the potential presence of an overturned fold. D) Cross section slice of dipping shears in the second deposit, coloured by automatic cluster groupings.
None of this complexity was represented in the legacy grade distributions, opening the opportunity for Driver to provide a data-driven foundation for building a more geologically coherent model.
‘The apparent fold structure which emerged from the analysis fit well with the regional structural setting. It certainly suggests that, given the proposed near term mine expansion/production, that this area merits further investigation using modern drilling to test the possible new interpretation. In addition, the variable anisotropy models from Driver do a much better job at rapidly capturing the local variability in the regional shallowly dipping trends in both deposit areas, which could be used to significantly improve the local continuity of the final grade model’ Hastings said.
An integrated, data-driven modelling pipeline
Within a few minutes, the Driver analysis generates a form of quantitative structural information that can be used for deposit-scale observations, and as inputs into a new integrated modelling pipeline shared with Leapfrog Geo. Each local ellipsoid designates the expected continuity of a geological feature (in this case, elevated Au grade), extending away from the local analysis locations. The collection of ellipsoids constitutes a locally varying anisotropy field, which can now be directly integrated into Leapfrog’s Structural Trend, where it can guide and regulate the RBF™-based implicit-modelling engine.
After filtering local ellipsoids with low confidence and poor data support metrics, Hastings created a range of grade-based indicator volumes at various cutoffs and probability thresholds. By using the new Triaxial-Blending Structural Trend as an input (available in Leapfrog 25.1), the resultant implicit surfaces vary the strength and magnitude of local anisotropy as a function of the grade trends detected by Driver.
Figure 3: Comparison of Leapfrog implicit domain and numeric models constructed using the Triaxial-blending Structural Trend informed by Driver (purple and hot colours), compared with the legacy global dip (red). Results delineate the overturned fold structure in the proposed future mining area (A). and a broad open syncline structure in the north (B).
The new approach, which combines Driver and Leapfrog Geo, offers significant advantages over the legacy interpretation by efficiently providing models with enhanced structural integrity. The generated indicator volumes more closely reflect the deposit’s structural complexities, clearly delineating features like the tight, overturned fold in the proposed mining area and the broad, open syncline structure in the north (Figure 3). This model, with enhanced local flexibility is achieved within a few minutes, and with significantly less manual intervention in the form of digitised polylines, creating structural controlling discs and building form interpolants — a process that previously could take hours or even days to complete by hand.
‘What stood out to me was how efficiently the new structural inputs could be generated,’ Hastings said. ‘As a consultant working across many projects and on tight time frames, I am excited to be able to build a domain model that is more geologically coherent and connected with the data, far faster and efficiently than I ever could have with a conventional implicit or explicit approach.’
The hybrid ai and implicit modelling workflow delivered by Evo, Driver and Leapfrog Geo offers a practical way to enhance the structural relevance of models with little manual effort while remaining fast, objective, and fully dynamic.