Reading a fault map: what the scaling laws say about your picks

A fault map is a set of assertions. Each one says: there is a surface here, it is this long, and the rocks either side of it have moved this far past each other. Most interpretation software will draw those assertions beautifully and check none of them.

The check exists, and it comes from outside your survey entirely.

Displacement scales with length

Across published fault populations spanning seven orders of magnitude, from centimetres to hundreds of kilometres, maximum displacement is roughly proportional to length: @@D_{max} \approx \gamma L@@, with @@\gamma@@ near 0.03. The scatter is real and wide, about an order of magnitude, because faults grow by linking segments and no two histories are identical.

That width is exactly why the relationship is useful as a check rather than a prediction. A fault sitting anywhere in the band is unremarkable. A fault sitting far outside it is usually not a discovery about the geology; it is a statement about the picking.

What each departure means

Far below the trend, carrying much less displacement than its length implies, is most often two separate faults picked as one continuous segment. Merging doubles the length while the throw stays where it was.

Far above the trend is the reverse: either the fault continues past where it was picked, so the length is understated, or the throw was measured across something that is not a single fault. Measuring across a relay zone, where two faults overlap, gives you the sum of both.

And the profile catches what the scaling plot cannot. A single fault is an ellipse of displacement: maximum near the middle, dying to zero at both tips. A flat-topped profile means the surface was carried at a constant offset rather than followed. Two peaks with a saddle between them means two faults, and this is the only view that can see it, because merging halves the ratio and half of trend is still inside the natural scatter.

The population is evidence too

Fault lengths follow a power law: many small faults, few large ones. If your length histogram is missing its smallest bin, you have not mapped a fault population. You have mapped the faults that were easy to see, and any statement you make about fault density, seal or compartmentalisation inherits that bias.

None of this needs an answer key. The reference is published, not local, which is why the same three views work on real data where nobody can tell you whether you were right.