When one soil layer is really two: splitting geotechnical units

Updated: 2026-08-13 · ISKRIT

A geotechnical unit is a claim: that every sample inside it comes from one population and can be described by one set of numbers. The claim is testable, and a surprising share of units drawn by eye do not survive the test.

The test

A unit is provisionally drawn from the field descriptions and the borehole logs, on geology and origin. Only then are the numbers checked. Two things disqualify a provisional unit: scatter above the limit set for that property, and a systematic trend of the property with depth or with position on the site.

The distinction matters. Scatter without a trend means the samples are mixed, and the fix is to look for a criterion that separates them. A trend with low scatter around it means the unit is real but not uniform, and the fix is either to split it into sub-units or to describe the property as a function of depth.

Why averaging over a trend is worse than either option

Consider a sand whose deformation modulus rises from 18 MPa at the top of the unit to 34 MPa at the base, with little scatter at any one depth. The mean is 26 MPa. Used at the base of a foundation resting near the top of the unit, 26 MPa overstates the stiffness by nearly half. Used for a pile toe near the bottom, it understates it by a quarter. The averaged number is wrong in both directions and safe in neither.

What a high coefficient of variation is telling you

The order of operations

Geology first, statistics second. Splitting a unit purely because the numbers scatter, with no geological reason for the boundary, produces units that cannot be mapped between boreholes and that no one can identify on the next site visit. The statistics are there to refute a unit, not to invent one.

The unit delineation tool tests each provisional unit for scatter and for a depth trend, reports which property failed and by how much, and proposes the split rather than applying it.