glossary¶

short definitions of the terms these pages use. elsewhere, hover a dotted term to see its definition.

angular tolerance: the half-angle of the cone around a direction within which pairs count toward a directional variogram.
anisotropy: continuity that depends on direction, with a longer range along some directions than others.
anisotropy ellipse: the ellipse, or ellipsoid in 3D, whose axes are the variogram ranges in the principal directions.
arrow: a standard in-memory column format that polars, pandas and pyarrow read directly.
azimuth: a horizontal direction in degrees, measured clockwise from north.
back-transform: the inverse of a transform, returning simulated scores to the original units.
block kriging: kriging of the average value over a block instead of the value at a point.
boundary edge: a mesh edge used by only one triangle, the rim of a hole.
coefficient of variation: the standard deviation divided by the mean, a unitless measure of spread.
collar: the point where a drill hole starts at the surface, with its coordinates and total length.
composite: a drill-hole interval of fixed length whose grade is the sampled-length-weighted mean of the assays it covers.
composites: drill-hole intervals of fixed length whose grades are the sampled-length-weighted means of the assays they cover.
conditional bias: a systematic error that depends on the estimate itself, with high estimates too high and low ones too low.
contained metal: tonnes × mean grade above a cutoff: the amount of metal the resource holds.
cross-validation: re-estimating each sample from the other samples and comparing the estimates with the actual values.
CRS: coordinate reference system: the named map projection or local grid the coordinates refer to.
cutoff: the grade below which a block does not pay to mine; you report resources above it.
declustering: weighting samples by the area or volume each represents, to correct statistics for clustered sampling.
declustering effect: the sharing of weight among clustered samples, which kriging treats almost as one.
degenerate face: a triangle with a repeated corner or zero area.
desurveying: turning a hole's collar and downhole survey angles into 3D coordinates along the hole.
dip: the angle in degrees below the horizontal; positive points down.
discretization: splitting a block into a regular grid of points to measure what fraction lies inside a solid.
domain: a volume of rock whose values you treat as one population, such as one lithology or one vein.
E-type: the mean of the realizations at each node, a smooth estimate comparable to kriging.
experimental variogram: half the mean squared difference between samples, computed for pairs grouped by lag.
grade-tonnage curve: tonnes and mean grade above each cutoff, read as a curve over the cutoffs.
h-scatterplot: a plot of the value at one end of each pair of samples a lag h apart against the value at the other end.
inverse distance: an estimator that weights each sample by a power of the inverse of its distance to the target.
kriging efficiency: share of the block variance an estimate resolves: 1 for a perfectly known block, 0 or less for one known no better than the mean.
kriging variance: the error variance kriging predicts at a target, from the sample layout and the variogram alone.
lag: the separation vector between two samples, a distance and a direction.
lag tolerance: how far a pair's distance may differ from a lag center and still count toward that lag; half the lag spacing.
misplaced volume: labeled volume outside its solid plus solid volume outside its labeled blocks.
non-manifold edge: a mesh edge shared by three or more triangles.
normal scores: values replaced by the standard gaussian values with the same cumulative probability.
nugget: variogram value just above zero lag: variability at scales shorter than the closest samples, plus measurement error.
ordinary kriging: kriging with an unknown mean that is constant near the target; its weights sum to one.
outlier: a value far from the rest of the distribution, often real but with outsized influence.
parent block: a block of the regular grid a sub-blocked model starts from.
parquet: an open, compressed file format for tables that most data tools read.
partial block: a block that a solid's surface crosses, so it lies partly inside and partly outside.
positive definite: property of a variogram model that guarantees a positive variance for any weighted combination of samples.
preferential sampling: sampling that places more samples in some zones, usually the rich ones, than in others.
random path: the random order in which sequential simulation visits the grid nodes.
range: the lag at which the variogram reaches its sill; beyond it, samples are uncorrelated.
realization: one simulated map that honors the samples and reproduces the histogram and variogram of the data.
realizations: simulated maps, each honoring the samples and reproducing the histogram and variogram of the data.
screen effect: the loss of weight by a sample hidden behind a closer sample in the same direction.
seed: the number that starts the random-number generator, so the same seed gives the same realizations.
sentinel: a placeholder number such as -999 that a file uses to mark a missing value.
SGS: sequential gaussian simulation: it visits the nodes in random order, krigs each one, draws from the gaussian that kriging gives and adds the draw to the data.
sill: the plateau of the variogram, reached when samples lie too far apart to be related.
skewness: a measure of the asymmetry of a distribution; positive when a long tail runs to high values.
slope of regression: the slope of the true values regressed on the estimates; 1 means no conditional bias.
smoothing: the loss of variance in an estimate, which averages neighboring samples and so varies less than the true grades.
solid: a closed triangle mesh that bounds a volume, such as an ore body or a vein.
solids: closed triangle meshes that each bound a volume, such as ore bodies or veins.
sub-block: a smaller cell a parent block is split into where a contact crosses it.
sub-blocks: smaller cells a parent block is split into where a contact crosses it.
sub-cell: one cell of the regular sub-grid inside a parent block; neighbors with one label merge into a sub-block.
support: the volume a value represents: a point, a length of core or a block.
swath plot: mean grade of blocks and samples in parallel slices along one direction, to find local bias.
top cut: a cap: values above it are set to it, to limit the influence of a few extreme values.
value object: an object that holds model parameters and no data, such as a variogram or a search; it never changes after you build it.
variogram map: the experimental variogram computed for every direction and lag, drawn as an image.
volume-variance effect: averaging values over larger supports keeps their mean and reduces their variance.
winding: the order of a triangle's corners; counter-clockwise seen from outside makes its normal point out.
wireframe: a triangle mesh drawn as its edges; a closed one is a solid.