Orientations and fracture sets
By now you have orientation measurements — typed in the field, digitised as planes on the model, or estimated from traces — or you can extract them from the faces the outcrop exposes, as the first section below describes. This page turns that cloud of measurements into fracture sets: the two, three or four preferred orientations that actually describe the outcrop.
Everything downstream keys off sets. One set becomes one fracture set in a DFN, with its own mean orientation, dispersion, size distribution and intensity.
Orientations from exposed faces
Where the rock has broken along its fractures, the fracture surfaces themselves are exposed as planar faces, and Grow Planar Patches measures them all in one run. Select the mesh, then Extract → Grow Planar Patches. The dialog measures the mesh and suggests its own parameters, and Apply divides the mesh into planar patches, each with a fitted plane.
- Tick Create Objects to turn every patch into an orientation measurement. From then on they behave like any others: they plot on the stereonet, and contour and cluster as described below. Each patch is one measurement however large it is, so raise Minimum Area to keep slivers out of the statistics.
- Set Cluster into Sets to the number of sets you expect, and the tool also sorts the patches into sets, weighted by area, with one group per set. Treat that as a first answer to check against the contoured net.
- To collect a single set, use the dialog's Orientation Filter: pick a set by number with From Set, or give a dip, an azimuth and an angle, and only the patches of that orientation are produced. See Extracting One Fracture Set.
A fracture shows as a face only where the rock has broken along it, which favours fractures at a low angle to the exposure; fractures at a high angle to it show as traces instead. The sampling bias described in the overview therefore runs the other way for faces, so combine faces and traces rather than choosing one.
Get them on the stereonet
Open a Stereonet Window from Home → Windows. Your orientations plot as poles.
The first thing to do is contour them, in the Properties panel under Contour Properties. A scatter of poles will always look like it has clusters in it; contouring is what tells you whether they are real.
| Method | Use it to ask |
|---|---|
| Modified Kamb | "Is this cluster real?" Contours are in σ — standard deviations above a uniform distribution — so the contour value answers the question directly. |
| Schmidt (1%) | "How does this compare?" A fixed 1%-area counting circle contoured in % of data — the classic method, comparable against published nets and against datasets of a different size. |
Start with Modified Kamb. Raise Significance (σ) for broader, smoother contours and lower it for tighter, more detailed ones; if a cluster only appears at low σ, be suspicious of it.
Density contouring assumes an equal-area (Schmidt) projection, which is the default. If you switch Equal Area off for an equal-angle (Wulff) net, the contours are no longer a valid density estimate.
Splitting into sets
Two routes, and they are complementary rather than competing.
By hand, on the net
The Stereonet ribbon tab has Rectangle, Ellipse, Polygon and Sector selection tools. Draw round a cluster and the poles inside are selected — and because selection is shared across VRGS, they are simultaneously selected in the 3D view and in the trees.
That last part is what makes hand-picking worth doing at least once. Select a cluster on the net, look at the model, and see where those fractures are. If the cluster turns out to be forty measurements off one small face, you have learnt something the algorithm would not have told you. Hide Unselected on the same panel makes this easier.
Having selected a cluster, right-click in the tree → Group to put it in a folder of its own, or give each measurement a Set in the Properties panel.
Automatically, by k-means or a Fisher mixture
Select the orientation groups (or the loose measurements) in the tree, then right + click → Auto-cluster Selected Orientations….
- Choose the Method: k-means, quick and needing only k, or Fisher mixture (EM), which fits each set as a Fisher distribution with its own concentration, can put stray measurements in a noise class that belongs to no set, and can choose k by BIC for you.
- Give the Number of sets (k). The default is 3; k must be a whole number from 2 to half the number of measurements.
- It needs at least 4 distinct measurements.
- Clustering is on axial pole direction, so a plane and its opposite pole are treated as the same orientation, which is what you want for fractures. Lineations are clustered separately, never mixed with planes.
- k-means seeds the clusters with k-means++ and keeps the best of 20 runs, so it is not thrown by an unlucky start; the same selection always gives the same answer with either method.
- Each measurement is tagged with a new Set, J1, J2 and so on, listed in Collections → Sets, each in its own colour. Nothing is copied or moved in the tree.
To see the result, set the stereonet's Colour by to Set. With Choose k by BIC, the messages panel lists the BIC for each k it tried; a clear minimum is evidence for that number of sets, a flat run of values is a warning that the data do not settle it. Running it again with a different k simply re-tags the same measurements with new sets, so try two or three values and compare against the contoured net. k-means will happily return four sets from a population that has two; the contours are what tell you it has overreached.
Contour first to decide how many sets there are, then auto-cluster with that k to do the sorting. Picking k off the stereonet and letting k-means do the tedious part is faster than either alone, and it keeps the judgement where it belongs.
Reading the set statistics
In a Stereonet window, set Sets From to Project Sets and tick Set Statistics (both under Analysis Display Properties). Each set's mean pole is marked in its colour with its α95 cone and a variability cone, and the key lists every set's N, mean Plane, Fisher K and α95. If the outcrop face sampled some sets badly, turn on Terzaghi Correction too; see Set statistics.
When you run Create DFN from Selected Groups, which works on groups, VRGS also shows a per-group preview before it commits to anything:
NE joints: 48 measurements -> dip 82.4, az 137.1, K 24.6
NW joints: 31 measurements -> dip 79.0, az 044.9, K 11.2
That is the Fisher summary — mean dip, mean azimuth, and K, the concentration parameter. It is worth reading even if you never build a DFN, because it is the most compact honest description of a set you will get.
As a rough reading of K: 1–5 is strongly dispersed, 10–20 is moderate clustering, and 50+ is a tight set. A group needs at least 3 measurements to have statistics computed at all; fewer and it is skipped.
If a cluster comes back down in the dispersed range, go back to the net. Usually it is either two sets that k-means merged into one, or a genuinely diffuse population — and the contours distinguish those.
Getting the numbers out
- Right-click the stereonet → Export writes SVG Document → Stereonet Diagram (or Rose Diagram) at publication quality, or Spreadsheet for the plotted orientations as a table.
- Right-click any interpretation group → Spreadsheet gives a tabular view of every measurement in it, which is usually faster than exporting when the question is just "what were those numbers?".
- For a figure that travels with its statistics, collect the net into an analysis report.
See also
- Stereonet User Guide — the complete property sheet, the rose and dip plots, and the 3D stereosphere.
- Grow Planar Patches — extracting orientations from the planar faces of a mesh, and its method.
- DFN: Matching Interpretations — the mathematics behind the Fisher and k-means steps.
Next: measuring fracture intensity.