Kriging

Kriging is a spatial prediction method that estimates values at unsampled locations from nearby observations and a model of spatial dependence, choosing weights to minimize prediction variance.

Semivariogram: A function describing how the expected squared difference between observations varies with their separation. Kriging uses this spatial dependence function to determine prediction weights.
Spatial autocorrelation: Dependence between values associated with locations, often stronger for nearby locations. Kriging relies on spatial autocorrelation to infer unsampled values from measured locations.
Geostatistics: The statistical study of spatially or temporally distributed data, especially in Earth sciences. Kriging is a foundational prediction method in this field.
Inverse distance weighting: An interpolation method that assigns weights according to inverse powers of distance from sampled locations. Unlike kriging, it does not estimate weights from a spatial dependence model.
Covariance function: A function specifying how covariance between random variables changes with their separation or other inputs. Kriging can derive its weights from covariance between observed and predicted values.
Stationarity: A property of a stochastic process whose specified statistical features remain unchanged under shifts in location. Common kriging models assume stable means or dependence across the modeled region.
Ore reserve estimation: The estimation of the quantity and spatial distribution of mineral resources in a deposit. Mining uses kriging to estimate grades between drilling locations.
Spline interpolation: An interpolation method that fits a smooth function through or near observed data. Splines prioritize smoothness rather than minimizing model-based prediction variance.
Best linear unbiased prediction: A linear prediction that is unbiased and has minimum variance among linear unbiased predictors. Ordinary kriging is the spatial form of this variance-minimizing prediction principle.
Intrinsic hypothesis: An assumption that spatial increments have zero mean and variance depending only on separation. It supports variogram-based kriging without requiring a stationary covariance function.

Linked from 5 pages