Global Universal Kriging
Also known as: universal kriging (global), global UK, kriging with external drift (global), global trend kriging
Global Universal Kriging is a geostatistical interpolation method that models a spatially varying trend (drift) as a deterministic function of coordinates and uses the entire dataset to fit both the trend coefficients and the residual variogram simultaneously. It produces optimal linear unbiased predictions together with pointwise estimation uncertainty, accounting for a large-scale spatial gradient across the full study region.
Read the full method
Sign in with a free account to read this section.
Method map
The neighbourhood of related methods — select a node to explore.
When to use it
Use Global Universal Kriging when the variable of interest exhibits a clear large-scale spatial trend (e.g., elevation increasing from coast to mountains, temperature decreasing with latitude) that must be modelled explicitly before interpolating local variation. It is most appropriate when the trend is stationary (consistent direction and magnitude) across the entire study area, the dataset is moderately sized (roughly 50–5 000 points) so that a single global variogram is representative, and prediction uncertainty maps are needed alongside the interpolated surface. Avoid it when the trend changes direction or strength across sub-regions (use local or multiscale variants instead), when the dataset is very large (tens of thousands of points) and the global kriging system becomes computationally prohibitive, or when there is no detectable drift and Ordinary Kriging suffices.
Strengths & limitations
- Explicitly models and removes a large-scale spatial trend, yielding unbiased predictions even when a drift is present.
- Uses all available data to estimate both the trend and the variogram, maximising information use and producing globally consistent estimates.
- Provides rigorous, spatially varying uncertainty estimates (kriging variance) at every prediction location.
- Predictions are the Best Linear Unbiased Estimators (BLUE) under the model assumptions.
- Well-suited for study areas with a dominant directional or polynomial gradient (e.g., elevation, rainfall, temperature).
- Computationally expensive for large datasets: solving the full n×n kriging system has O(n^3) complexity.
- Assumes the trend is correctly specified; a misspecified trend function leads to biased predictions and distorted variograms.
- The global variogram assumes stationarity of residuals across the entire domain, which may be violated in heterogeneous landscapes.
- Requires sufficient data spread to estimate the trend reliably — a sparse or clustered dataset can lead to unstable trend coefficients.
- More complex to implement and interpret than Ordinary Kriging; requires decisions on trend order and variogram model form.
Frequently asked
What is the difference between Ordinary Kriging and Global Universal Kriging?
Ordinary Kriging assumes a constant but unknown mean across the study area, whereas Global Universal Kriging allows the mean to vary as a deterministic function of spatial coordinates (the drift or trend). When a spatial trend is present, using Ordinary Kriging produces biased predictions; Universal Kriging accounts for and removes the trend before interpolating residuals.
How do I choose the trend order in Universal Kriging?
Start with a linear trend (first-order polynomial) and check whether the variogram of generalised residuals looks stationary. If systematic large-scale variation remains, try a second-order polynomial. Avoid high-order polynomials unless clearly justified, as they risk over-fitting. Cross-validation statistics (RMSE, standardised error) help select the best order.
When should I prefer Local Universal Kriging over Global Universal Kriging?
When the trend changes direction or intensity across the study area, fitting a single global polynomial is inappropriate. Local Universal Kriging uses a moving neighbourhood so the trend is re-estimated at each prediction point, capturing spatial non-stationarity at the cost of greater computation and potential edge effects.
Can Global Universal Kriging handle very large datasets efficiently?
Standard implementations require inverting an n×n matrix, which becomes prohibitive beyond a few thousand points. For large datasets, options include sparse approximations, covariance tapering, low-rank methods, or switching to a local neighbourhood variant that only solves smaller sub-systems at each prediction location.
What software implements Global Universal Kriging?
R packages gstat and geoR support Universal Kriging with selectable neighbourhood size (set to all data for global mode). Python's PyKrige library also provides Universal Kriging. ArcGIS Geostatistical Analyst and GSLIB implement the method as well.
Sources
- Journel, A. G., & Huijbregts, C. J. (1978). Mining Geostatistics. Academic Press, London. ISBN: 978-0123910608
- Chiles, J.-P., & Delfiner, P. (1999). Geostatistics: Modeling Spatial Uncertainty. Wiley, New York. ISBN: 978-0471083153
How to cite this page
ScholarGate. (2026, June 3). Global Universal Kriging. ScholarGate. https://scholargate.app/en/spatial-analysis/global-universal-kriging
Which method?
Set this method beside its closest kin and read them side by side — the library lays the books on the table; the choice is yours.
- Co-krigingSpatial analysis↔ compare
- Local Universal KrigingSpatial analysis↔ compare
- Ordinary KrigingSpatial analysis↔ compare
- Universal KrigingSpatial analysis↔ compare