Answers are grounded in this paper and cite their sources.
Summary When two gravity inversion models of the same ground are built from surveys of different spatial extent, they disagree, and it is seldom clear what the disagreement means. We show where it arises and what controls it, using ten models of the Eastern Goldfields Superterrane, Western Australia, with survey widths from 28.5 to 288.5 km, inverted with the open-source platform Tomofast-x. Each model comprises a core, covered by the survey, and a surrounding padding zone. For 22 model pairs we compute the RMS difference of density and of density gradient over a series of shrinking, centred comparison windows. Three contributions can be distinguished. Over the full extent of the smaller model, the density difference is dominated by its padding zone and decreases approximately as the inverse square root of the core width, the form that the ratio of padding volume to total volume gives for wide cores. Toward the edge of the core both metrics rise, the gradient metric strongly, and for the gradient metric the extent of this edge zone scales with the core width. In the interior of the core, the density difference follows a difference between the two data sets: because the regional trend is removed from each survey over its own footprint, the data sets differ, to a close approximation, by a plane, of up to 14 mGal here. This interior density difference, which ranges from about 2 to 40 kg m−3, can be estimated from the data before any inversion is run. With no fitted parameter, dividing the RMS data difference by the gravitational attraction, per unit density, of a uniform block with the core’s width and the model’s depth reproduces it to within about 30 per cent in the central half of the core. The gradient metric in the interior responds to the same data difference, largely because the density shift that it produces increases with depth, and is described by an empirical two-term relation. We recommend comparing models within a window of 0.75 times the smaller core width, which excludes the part of the edge zone where the gradient difference exceeds twice its interior value, at every core width studied.
No takes yet. Share an insight, caveat, or question.
Ogarko et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: