Kriging¶

A best-linear-unbiased spatial prediction method whose weights derive from a modeled covariance or variogram under stated mean assumptions.

Version: v3 · 2026-09-19 · History
Domain-specific #: 2145
Domain group: Formal Sciences
Origin domain: Experimental Design & Statistics
Subdomains: Geostatistics, Spatial Prediction → Experimental Design & Statistics
Aliases: Gaussian process regression in geostatistics

Core Idea¶

Kriging is a best-linear-unbiased spatial prediction method whose weights derive from a modeled covariance or variogram under stated mean assumptions.

Kriging predicts a spatial random field at an unsampled location with a linear combination of observed values whose weights are derived from a variogram or covariance model and stated mean assumptions. Ordinary, simple, universal, and other forms differ in their trend constraints; under the model, the predictor is best linear unbiased and carries a kriging variance.

When the mean or trend, covariance kernel, Gaussian prior, observation and noise model, and conditioning target are aligned, the kriging predictor coincides with the corresponding Gaussian-process posterior mean, with uncertainty correspondence under the same assumptions. Without Gaussianity, kriging retains its best-linear-unbiased interpretation but not a full posterior-distribution equivalence.^[1]

Its operative boundary is not supplied by the name alone. Preserve this identity: A best-linear-unbiased spatial prediction method whose weights derive from a modeled covariance or variogram under stated mean assumptions.

How would you explain it like I'm…

Smart Guessing Between Spots

Imagine you measured how deep the snow is at a few spots in a park and want to guess the depth at a spot you didn't measure. Kriging is a careful way of guessing. It uses a rule about how much nearby spots tend to be alike to decide how much each measurement counts. It also tells you how unsure the guess is.

Weighted Map Guessing

Kriging is a way to guess a value at a place you didn't measure, like soil depth or rainfall, using places you did measure. Its guess is a weighted mix of the measurements. The weights aren't picked by distance alone. They come from a model of how similar values are at different separations. The weights are chosen so the guess isn't tilted too high or too low on average, and so the error is as small as possible. Kriging also gives a number saying how uncertain each guess is.

Best Linear Spatial Prediction

Kriging predicts a spatial quantity at an unsampled location as a weighted sum of observed values. The weights come from a model of spatial correlation, a covariance function or variogram describing how similarity changes with distance, together with assumptions about the mean. They are chosen so the predictor is unbiased and has the smallest possible error variance among linear predictors, which is called 'best linear unbiased'. Versions like simple, ordinary, and universal kriging differ in what they assume about the mean or trend. Each prediction comes with a kriging variance measuring its uncertainty. That separates it from generic smoothing or spline interpolation, which don't come from these unbiasedness and covariance-based conditions.

Kriging treats the data as a realization of a spatial random field and predicts its value at an unsampled location by a linear combination of observations. The weights solve a constrained optimization: minimize the prediction error variance computed from a modeled covariance or variogram, subject to unbiasedness constraints set by the mean assumptions. Simple kriging assumes a known mean, ordinary kriging an unknown constant mean, and universal kriging a trend in specified covariates. Under the model the result is the best linear unbiased predictor (BLUP), and the minimized variance is the kriging variance. When the trend, covariance kernel, Gaussian prior, observation and noise model, and prediction target are all aligned, the kriging predictor coincides with the Gaussian-process posterior mean, and the uncertainties correspond as well. Without Gaussianity it keeps its BLUP interpretation but is not a full posterior distribution. Generic smoothing or spline interpolation does not qualify unless its weights satisfy the kriging conditions.

Scope of Application¶

The abstraction recurs literally within environmental science, geology, hydrology, agriculture, and other fields predicting spatially correlated quantities. The following habitats preserve the same recognition machinery; they are not invitations to extend the name metaphorically.

Ore estimation. sample grades predict blocks within a deposit.
Environmental mapping. monitoring stations predict pollutant surfaces.
Soil science. sampled properties are interpolated across fields.
Hydrology. rainfall or groundwater variables are mapped with spatial uncertainty.
Remote sensing. sparse ground observations calibrate spatial products.

Clarity¶

Declare the kriging variant, coordinate system, target support, trend, variogram fit, anisotropy, neighborhood, and transformation. The kriging variance is conditional on the spatial design and covariance model and usually does not include all parameter or measurement uncertainty.

A practical identification audit begins with the typed roles rather than the title: establish the spatial field, verify the sampled locations, then test the remaining conditions and exclusions.

Manages Complexity¶

Kriging reduces an entire spatial dependence model and sample geometry to a local weighted predictor and uncertainty. The same equations expose extrapolation, clustering, anisotropy, and support mismatch rather than hiding them behind distance alone.

The compression remains accountable because each simplification has a named failure condition. Disagreement can be localized to a missing role, an invalid assumption, an ambiguous measurement, or a neighboring abstraction instead of being hidden inside an unanalyzed label.

Abstract Reasoning¶

R1. Define the prediction support and stochastic trend assumptions. R2. Explore spatial dependence and fit a defensible variogram or covariance model. R3. Construct the covariance system for samples and target. R4. Solve the unbiased minimum-variance equations for weights. R5. Validate residuals and predictions, then report conditional variance and model sensitivity.

Knowledge Transfer¶

The name transfers literally to spatial or spatiotemporal BLUP systems with the kriging equations and dependence model. Measurement and optimization are parents; weighted averaging by itself is not kriging.

The transfer boundary is explicit: DOMAIN-SPECIFIC PASS / PRIME FAIL: The method recurs across spatial datasets, locations, variables, and covariance or variogram models. Literal recognition retains the specialist vocabulary and validity conditions of geostatistics and spatial interpolation; outside that setting only broader parent operations transfer.

Relationships to Other Abstractions¶

Current abstraction Kriging Domain-specific

Parents (2) — outgoing typed relations

Kriging is a kind of Measurement Prime

Measurement (prime:measurement).
Kriging is a kind of Optimization Prime

Optimization (prime:optimization).

Hierarchy paths (2) — routes to 2 parentless roots

Kriging → is a kind of Measurement

Show alternative path (1)

Neighborhood in Abstraction Space¶

Kriging sits in a moderately populated region (51^st percentile for distinctiveness): it has near-neighbors but no dense thicket of look-alikes.

Family — Statistical Learning & Model Failure Modes (41 abstractions)

Nearest neighbors

Computed from structural-signature embeddings · 2026-10-25

References¶

[1] Carl Edward Rasmussen and Christopher K. I. Williams, Gaussian Processes for Machine Learning, MIT Press, 2006. Authoritative reference for Gaussian-process priors, covariance functions, observation-noise models, conditioning, posterior means, and posterior variances. registry ↩