Electrical · Interpretation
The section is smoother than the ground
A resistivity section looks like a picture of the subsurface. It is not. It is the simplest arrangement of colour that would have produced the readings, and knowing the difference is most of what separates a useful survey from a misleading one.
Electrical resistivity tomography produces the most immediately persuasive image in near-surface geophysics. Warm colours, cool colours, a boundary running across the section at a sensible depth. It reads like a photograph of a trench wall. Almost everything that goes wrong with these surveys goes wrong at that moment of recognition.
What the instrument actually recorded
The field data is not a section. It is a few hundred readings of apparent resistivity, each one a single number produced by injecting current through one pair of electrodes and measuring the potential difference across another. Every reading samples a large, fuzzy, overlapping volume of ground. None of them belongs to a particular depth.
Turning that into a section is the job of inversion: a computer builds a model of the subsurface, calculates what readings it would have produced, compares them with the real ones, adjusts, and repeats until the two agree closely enough. The section you are handed is the model it stopped at.
Why it comes out smooth
Many different arrangements of ground would fit the same readings to the same tolerance. The inversion has to choose one, and the usual rule it is given is to prefer the smoothest model that fits — the one with the least abrupt change in it.
That rule is sensible and it has consequences. A genuinely sharp boundary, a clay layer sitting on sand with a knife edge between them, is rendered as a gradient spread over a depth interval. A small strong target is smeared into a larger weak one. The section is not being cautious or imprecise; it is answering a different question from the one most readers think it is answering. It shows the smoothest ground consistent with the data, not the ground.
The number that tells you whether to believe it
Every inversion produces a misfit: a figure for how well the model's predicted readings match the real ones. It belongs on the section, and if it is absent from a report that is worth asking about.
A high misfit means the model does not explain the data, and the structure drawn on it is not to be trusted. A very low misfit is not automatically better either — an inversion pushed to fit the data too closely starts fitting the noise as well, and produces detailed-looking features that are artefacts of the measurement rather than the ground.
Depth is the least certain axis
Resistivity resolves lateral position considerably better than it resolves depth. A boundary's horizontal position across the line is usually reliable; the depth attached to it carries a real uncertainty band, and that band widens with depth.
This is why a single borehole changes the value of a whole survey. Tie the section to one logged hole and the depth scale is calibrated rather than modelled, and the interpretation across the rest of the line tightens accordingly. One hole will often do it. Where boreholes exist already, logging one is cheap next to what it resolves.
Two methods, independent properties
The strongest check on an ERT interpretation is a second method measuring a different physical property over the same ground. An air-filled void is highly resistive and also a deficit of mass, so microgravity sees it through density while resistivity sees it through conduction. Agreement between two independent properties is evidence. One anomaly on one instrument is a hypothesis.
Where a project cannot afford a false negative — a void under a foundation, a dissolution feature under a runway — that is not belt and braces. It is the only defensible way to run it.
What to ask for
Ask for the model misfit on the section. Ask what inversion constraint was used, and whether anything was used to constrain it besides smoothness. Ask what the depth uncertainty is at the horizon you care about, rather than reading a depth off the colour scale. And ask what the interpretation would look like if the second-most-likely model were drawn instead — because there always is one.