Local Universal Kriging
Also known as: local UK, local kriging with trend, local KED, local kriging with external drift
Local Universal Kriging is a geostatistical interpolation method that combines a spatially varying deterministic trend with a stochastic residual, estimated using only nearby observations within a defined search neighborhood. It generalizes local ordinary kriging by explicitly modeling and removing a polynomial or covariate-driven drift before interpolating the residual surface.
Key highlights
- Explicitly accounts for non-stationary mean structure (trend/drift), producing less biased predictions than ordinary kriging in trended data.
- Local search neighborhood makes the method scalable to large spatial datasets and computationally feasible.
- Provides prediction uncertainty (kriging variance) at every location, enabling probabilistic spatial mapping.
- Flexible trend specification: polynomial in coordinates or regression on external covariates (kriging with external drift).
- Unbiased by construction — kriging weights automatically satisfy the unbiasedness constraint even in the presence of trend.
Intuition
This section is available to Pro members. Upgrade to Pro
How it works
This section is available to Pro members. Upgrade to Pro
When to use it
Use Local Universal Kriging when your spatial variable exhibits a clear deterministic trend or gradient (e.g., elevation, temperature, precipitation increasing with latitude or altitude) in addition to local spatial autocorrelation in the residuals. It is appropriate when the dataset is large enough to require a local search window for efficiency, yet the underlying process has non-stationary mean structure. It outperforms local ordinary kriging when a significant trend is present, and is preferred over global universal kriging when the trend itself varies across the study region. Do not use it when there is no detectable spatial trend (ordinary kriging suffices), when the trend model is highly uncertain or misspecified, or when data are too sparse in local windows to reliably estimate both trend and variogram parameters.
Strengths & limitations
- Explicitly accounts for non-stationary mean structure (trend/drift), producing less biased predictions than ordinary kriging in trended data.
- Local search neighborhood makes the method scalable to large spatial datasets and computationally feasible.
- Provides prediction uncertainty (kriging variance) at every location, enabling probabilistic spatial mapping.
- Flexible trend specification: polynomial in coordinates or regression on external covariates (kriging with external drift).
- Unbiased by construction — kriging weights automatically satisfy the unbiasedness constraint even in the presence of trend.
- Requires explicit specification of the trend model; a misspecified trend can introduce systematic bias throughout the interpolated surface.
- Estimating both trend parameters and variogram parameters from the same data can be unstable, especially with small local windows.
- Computationally more demanding than local ordinary kriging due to the expanded kriging system that includes drift constraints.
- Assumes second-order stationarity of the residuals after trend removal, which may not hold in all applications.
- Performance is sensitive to the choice of neighborhood size and shape; too small a window may give insufficient data for reliable estimation.
Common pitfalls
This section is available to Pro members. Upgrade to Pro
Applications
This section is available to Pro members. Upgrade to Pro
Frequently asked
What is the difference between Local Universal Kriging and Local Ordinary Kriging?
Ordinary kriging assumes a constant but unknown mean and estimates it implicitly. Universal kriging explicitly models a spatially varying mean (trend) using polynomial coordinates or external covariates, making it appropriate when data display a systematic gradient or drift that ordinary kriging cannot adequately capture.
How do I choose the search neighborhood size?
The neighborhood must contain enough points to estimate the kriging system — at minimum more than the number of trend parameters plus one. In practice a minimum of 10-20 nearby observations per prediction is common. Balance local adaptivity (smaller window) against estimation stability (larger window) using cross-validation metrics such as mean squared prediction error.
Can external covariates be used as the trend instead of coordinates?
Yes. When the trend is a function of a secondary variable (e.g., elevation used as a covariate to interpolate temperature), the method is called Kriging with External Drift (KED). This is a subcase of universal kriging and inherits the same local implementation approach.
Does local universal kriging guarantee exact interpolation at data points?
Yes, like all standard kriging variants, local universal kriging is an exact interpolator — the predicted value at a sample location equals the observed value, provided that location falls within its own search neighborhood.
When should I prefer Geographically Weighted Regression over Local Universal Kriging?
Geographically weighted regression (GWR) is better suited when the primary goal is understanding spatially varying relationships between a response and predictors, and spatial autocorrelation in residuals is secondary. Local Universal Kriging is preferable when the primary goal is optimal spatial interpolation of a variable with a trend, leveraging spatial autocorrelation in residuals to minimize prediction variance.
Sources
- 1.Goovaerts, P. (1997). Geostatistics for Natural Resources Evaluation. Oxford University Press.ISBN 9780195115383
- 2.Chiles, J.-P., & Delfiner, P. (1999). Geostatistics: Modeling Spatial Uncertainty. Wiley.ISBN 9780471083153
You have read it. What now?
Cite this page
ScholarGate. (2026, June 3). Local Universal Kriging. ScholarGate. https://scholargate.app/spatial-analysis/local-universal-kriging