From Signal to Significance in Mine Geotechnical Engineering

What deserves to change our understanding and our decisions?

​

Modern mines can observe ground behaviour with increasing detail. Open pits may combine slope radar, prisms, piezometers, InSAR, photogrammetry and LiDAR, while underground mines may use seismic monitoring, extensometers, convergence measurements and damage mapping. These observations sit alongside geological, hydrogeological and numerical models, mine plans and operational experience.

​

For engineering purposes, the important distinction is not simply whether an observation is credible, but what it changes. A measurement may be technically sound yet add little to the current interpretation. Another may appear modest but materially change confidence in a plausible mechanism, our assessment of its consequences, or the decision that follows.

Once an observation is considered credible, the next question is therefore one of engineering significance. Does it strengthen or weaken a plausible interpretation? Does it change our understanding of the potential consequence? Does it affect the decision that follows? Or does it mainly confirm what is already understood?

​

This is the sense in which signal and noise are used in this essay. A signal is a credible observation that materially changes our confidence in one or more plausible interpretations, our assessment of their consequences, or the decisions that depend on them. Noise, in this context, may also be credible information, but information that adds little to the engineering question being considered at that point in time.

​

That distinction is not fixed. An observation with little significance today may become important as the mine develops, as new evidence emerges, or as the decision changes.

​

From observation to mechanism

​

Consider an open-pit slope where radar identifies increasing movement. Once the movement is confirmed, the important question is no longer whether something has changed, but why. The response may reflect blasting, changing pore pressure, structurally controlled deformation, progressive rock-mass damage, or factors not adequately represented in the current ground model.

​

Underground, increased seismic activity following a change in extraction sequence may reflect expected stress redistribution, controlled yielding, localisation of damage, structural activation or behaviour not anticipated by the current geotechnical understanding.

In both settings, establishing that the observation is credible is only the first step. Its engineering significance depends on what it reveals about the mechanism and the consequences that mechanism may have for the mine.

​

Monitoring is most useful when it helps distinguish between plausible explanations. Suppose two explanations for slope movement differ mainly in their expected groundwater response. Another displacement sensor may confirm the movement more precisely but contribute little to distinguishing between them. A strategically placed piezometer may be more informative because groundwater response is what separates the two hypotheses.

​

For seismicity, if alternative interpretations differ primarily in source or stress-transfer mechanism rather than event frequency, better event characterisation may be more informative than detecting additional events.

 

Monitoring design should therefore begin, where practical, with the engineering question rather than the instrument: What observation would strengthen one interpretation, weaken another, or reveal that both are incomplete?

​

Context and discrepancy

​

Geotechnical interpretation rarely rests on one observation. Confidence comes from how different aspects of evidence fit together.

​

Converging evidence can strengthen an interpretation, particularly where the observations are sufficiently independent. Contradictory evidence can be more important still.

​

If displacement accelerates while pore pressures remain stable, mapped structures do not support the assumed mechanism, and movement develops somewhere the model did not predict, the discrepancy may indicate that the current interpretation is incomplete.

​

In a partially observed system, evidence that does not fit can be more informative than repeated confirmation. Several observations may be compatible with the same incorrect explanation; one credible contradiction may force the conceptual model to be reconsidered.

​

This is the practical space between prediction and observation discussed in Geotechnical Judgement. The relevant question is not simply whether the observations agree with what we expected, but what their agreement or disagreement tells us about our understanding of the ground.

​

Competent engineers may still interpret the same evidence differently because they begin with different assumptions or give different weight to particular observations. That is a disagreement about the physical interpretation and the confidence attached to it.

​

A different issue arises when the interpretation is shared but views differ on acceptable exposure. That is not a disagreement about what the ground is doing; it is a question of risk acceptance and decision governance. Keeping the two separate helps prevent technical uncertainty from being confused with differences in operational or corporate risk tolerance.

​

The underlying logic is consistent with the observational method: establish an expectation of behaviour, observe what develops, compare it with what was anticipated and respond when the evidence requires it. Formal application goes further by defining expected behaviour and responses to foreseeable departures in advance.

​

Consequence, scale and time horizon

​

Understanding a mechanism does not by itself establish its importance.

​

One source of uncertainty may be large yet have little influence on a particular decision. Another may appear relatively small but be capable of changing slope geometry, mining sequence or a control strategy.

​

The largest uncertainty is therefore not necessarily the most consequential one.

​

Scale matters as well. Mines experience many small failures, localised deformations and damage events. Their occurrence alone says little about their wider significance.

​

A small failure may be inconsequential in itself but important if it reveals structural continuity, progressive damage or a mechanism capable of propagating beyond its immediate footprint. Conversely, a larger event may remain locally contained and have little relevance to the wider design.

​

Significance also changes with time horizon. A deformation mechanism may have little effect on today's production but materially influence the next pushback. A seismic response may require immediate access restrictions while having little bearing on life-of-mine strategy. A structure exposed today may not alter current operations but could materially change confidence in a future design.

​

The useful question is therefore: Important to which decision, at what scale, and over what horizon?

​

Short-term sequencing, annual planning and life-of-mine design can legitimately place different weight on the same evidence. Geotechnical interpretation therefore needs to remain connected to mine planning rather than operate as a separate technical stream.

​

The attention problem in the AI era

​

As monitoring systems become more integrated, access to information becomes less of a constraint. Attention and interpretation become more important.

​

Observations do not all demand the same level of attention. Some materially affect the current interpretation or operational response; others become important only as conditions, geometry or the engineering question changes.

​

This creates a distinction between immediate engineering significance and longer-term information value.

​

Retaining information is not enough. Its future value depends on preserving the context needed to interpret it: mine geometry, blast or extraction sequence, instrumentation changes, dewatering conditions, mapping revisions, model assumptions and the engineering questions being considered at the time. Without that context, an archive can preserve measurements while losing much of their meaning.

 

Artificial intelligence will make this issue more important. AI can already assist with tasks such as pattern recognition, event classification, anomaly detection and the analysis of large monitoring datasets. It can also help retrieve historical information and compare evidence across sources. More advanced applications are beginning to support model interrogation, identification of inconsistencies and comparison of alternative interpretations, but these depend heavily on data quality, engineering context and appropriate validation. 

​

The important point is not that AI will replace interpretation. It is that it will expand the amount of evidence and the number of possible relationships that can be examined. Some findings will strengthen existing understanding. Others may expose discrepancies worth investigating. Many may have little consequence for the decision in front of us.

​

AI increases our capacity to find relationships faster than it increases our ability to establish their engineering significance.

​

Operational response and engineering review

​

Trigger Action Response Plans are designed to remove ambiguity when action is required. If a defined trigger is reached, the prescribed response should follow. That is not the time to reopen the interpretation behind the trigger.

​

The basis of the TARP still needs to be reviewed as conditions evolve. Mine geometry changes, trends develop, trigger levels may be approached repeatedly, and behaviour may emerge outside the conditions for which the thresholds were originally set. None of that requires waiting for an alarm to be exceeded before asking whether the trigger remains appropriate.

​

Numerical models need the same discipline. When observed behaviour departs from prediction, recalibration may be justified, but agreement should not be restored simply for its own sake. The discrepancy may point to an unrecognised structure, hydraulic connection, stress-transfer mechanism, unsuitable boundary condition or another weakness in the conceptual model.

​

The task is to understand why the prediction failed. That usually means returning to the evidence, the conceptual model and the assumptions behind it.

​

The same issue applies to a geotechnical digital twin. A model that automatically adjusts its parameters whenever observations depart from prediction may achieve a better fit while embedding the wrong explanation.

​

Updating is useful when it improves understanding, not merely when it makes prediction and observation agree again.

​

Conclusion

​

Better monitoring, models and AI are giving mines an increasingly detailed view of ground behaviour. The harder problem is deciding what that information means and whether it matters.

 

Engineering significance is not inherent in an observation. It depends on what the observation tells us about mechanism, scale and consequence, and on the decision being made.

 

The question is therefore: What have we learned that changes our understanding, and what, if anything, should change because of it?