Run Tab

The Run tab holds the complete set of model parameters, starts the interpolation, and gives you the result. If you have worked through the Variogram tab, most of these fields are already filled in for you and you are only here to check them and press the button.

Model Parameters

Kernel

The kernel is the covariance model used for the estimate - the shape of the variogram curve. The first three correspond exactly to the variogram model shapes described in Kriging Concepts, and choosing a model on the Variogram tab sets the kernel here to match.

The spheroidal and cauchy families exist because the choice between spherical and gaussian is a coarse one. If spherical looks slightly too rough and gaussian is unusably smooth, step through the numbered variants until the result looks like the geology. They cannot be fitted on the Variogram tab, so pick the fitted spherical or exponential parameters and then substitute the kernel here.

Range, Sill and Nugget

• Range - the distance along the anisotropy major axis at which samples stop influencing each other.
• Sill (partial) - the variance reached at the range, excluding the nugget.
• Nugget - the variance at zero distance: measurement error plus structure finer than your sample length.

These are the fitted values from the Variogram tab, and editing them here edits the same numbers. A zero nugget forces the estimate through every sample exactly; a larger nugget lets it pass near them instead, which is usually what you want with assay data.

A Gaussian kernel with a zero nugget will usually fail to solve, and the dialog warns you before you try. Set the nugget to about 1% of the sill.

Anisotropy

Aniso az / plunge / dip and Aniso ratios define the continuity ellipsoid. The scan on the Variogram tab fills them in; Anisotropy explains what each number means and how to set them by hand.

Drift

• Ordinary kriging (constant mean) assumes the mean is unknown but locally constant. This is the normal choice and the default.
• Linear drift adds a regional trend to the mean, and is what other software calls universal kriging. Use it when the values genuinely trend across the deposit - a depth-related density, a weathering profile - rather than fluctuating about a level.

Linear drift extrapolates that trend beyond your data with enthusiasm, so if you use it, watch the extrapolation section of the Quality tab and consider blanking the unsupported nodes.

Fit Tolerance and Max Iterations

The model is fitted to the samples by successive refinement. Fit tolerance is how close the fit must get, in the units of your values, before it stops; Max iterations caps how long it tries. The default tolerance is scaled from your data, so it is already appropriate for parts per million and for percentages alike.

Tighten the tolerance for a final run if you like - it costs time and changes little. Loosen it if a fit will not converge. If a run reports a convergence failure, raising both is the first thing to try, after checking for duplicate samples at the same position.

Isotropic Distance Mask

Every grid is written with a second array recording each node's distance to the nearest sample. By default that distance is measured in the anisotropy's own frame - in major-axis metres, directly comparable with the range - so a node far down plunge counts as near, matching how far the model can actually see. Tick this box to record plain straight-line distance instead.

Running

Run interpolation starts the job. The output path is shown above the button; if the run cannot start, an amber line says why instead. Common reasons are that no samples have been extracted yet, that the selection has changed since they were, that a quality check is still running - fits run strictly one at a time - or that a parameter is out of range.

The run has four phases, reported above the progress bar.

1. fitting - fitting the model to the samples. No progress can be reported for this phase, so the bar is indeterminate and only the elapsed time moves. It is usually the longest phase.
2. evaluating - estimating every grid node. Real progress, with an estimated time remaining.
3. distance mask - measuring each node's distance to the nearest sample.
4. writing - writing the .vti file.

Cancel stops the job. During evaluation and masking it takes effect almost immediately. During the fit it can take a little longer, and the dialog tells you it is waiting to stop.

The log pane beneath the progress bar records each phase and its timing, along with progress messages from the fit, which are worth a glance when something takes longer than expected.

The Result

When the run finishes, the output path and file size are shown with three actions.

• Load result into scene adds the grid as a volume layer with the interpolated attribute and the samples' coordinate system already set. This is undoable: Ctrl+Z removes the layer, and the file stays on disk either way.
• Load with options... opens the standard volume loader instead, so you can choose which array to load - the distance mask, for instance - or set the coordinate system by hand.
• Open containing folder shows the file in your file manager.

Once loaded, the grid behaves like any other volume: isosurfaces, block cut-offs, clipping and the rest.

Staleness

Change the data selection, the grid or the parameters after a run and an amber warning appears saying the result no longer matches. The result is kept - nothing is thrown away - until a new run replaces it. The same applies to the variogram, the scan and the quality results, each of which knows what it was computed from.

This is deliberate: recomputing automatically would throw away expensive work every time you nudged a field. When you see the warning, decide whether it matters and rerun if it does.

Also See

• Quality tab - checking the result you just made
• Anisotropy
• Troubleshooting