Robust Universal Kriging
Also known as: RUK, robust kriging with external drift, outlier-resistant universal kriging, robust geostatistical regression kriging
Robust Universal Kriging (RUK) is a geostatistical interpolation method that combines a spatially varying deterministic trend with a stochastic residual surface, while using robust estimators to protect the variogram and trend coefficients from the distorting influence of outlying observations.
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 Robust Universal Kriging when you have georeferenced continuous measurements (soil contamination, temperature, rainfall, pollutant concentrations) that contain outliers or heavy-tailed distributions and when the mean of the variable changes systematically across the study area due to spatial covariates. It is the preferred choice over standard Universal Kriging whenever data quality is uncertain, sampling devices occasionally malfunction, or extreme values are likely to be measurement errors rather than genuine extremes. Do NOT use it when your dataset is small (fewer than ~50 locations), when no reliable spatial covariates are available for the trend (use Ordinary Kriging instead), or when computation time is critical and the data are known to be clean.
Strengths & limitations
- Resistant to outliers and measurement errors that would otherwise distort the variogram and bias predictions across the entire map.
- Accommodates a spatially varying mean through external covariates, making it more flexible than Ordinary Kriging in heterogeneous landscapes.
- Retains the full kriging framework: provides both a best linear unbiased predictor and a quantified prediction variance.
- Robust variogram estimators (e.g. Cressie-Hawkins) are well-established in the geostatistical literature and implemented in standard software.
- Can substantially improve map accuracy compared to standard Universal Kriging in real-world environmental datasets where contamination is common.
- Fitting is iteratively more complex than classical kriging: robust variogram estimation and robust GLS trend estimation require additional computation and convergence monitoring.
- Requires sufficient spatial replication — at least ~50 observation locations — to reliably estimate the variogram at multiple lag distances.
- Choosing the right robust estimator and its tuning constants (breakdown point, influence function shape) demands geostatistical expertise.
- If outliers are genuine extreme events rather than errors, down-weighting them introduces bias and can under-predict extremes on the map.
- Software support is less uniform than for classical kriging; some platforms require scripting in R (gstat, robgstat) or Python.
Frequently asked
How is Robust Universal Kriging different from ordinary Universal Kriging?
Standard Universal Kriging uses classical method-of-moments variogram estimation and ordinary GLS for trend fitting, both of which are sensitive to outliers. Robust Universal Kriging replaces these with influence-limiting estimators (e.g. the Cressie-Hawkins variogram, M-estimators for β), so that a small fraction of atypical observations cannot dominate the fitted model.
What robust variogram estimator should I use?
The Cressie-Hawkins estimator (based on the fourth root of absolute differences, 1980) is the most widely cited and is available in R's gstat package. For very heavy-tailed data, the median-based estimator of Genton (1998) offers even higher breakdown-point robustness.
Can I use Robust Universal Kriging when outliers are genuine extreme values, not errors?
Not without caution. If the extremes are real (e.g. a genuine contamination hotspot), down-weighting them will bias your map toward lower values at those locations. In that case, consider modelling the data on a transformed scale or using an indicator kriging approach instead.
How many data points do I need?
A minimum of roughly 50 spatially distributed observations is recommended so that the robust variogram can be estimated reliably across several lag distance classes. With fewer points the robust estimators become unreliable, and classical kriging with expert-specified variogram parameters may be preferable.
Which software packages implement Robust Universal Kriging?
The R packages gstat and robgstat support robust variogram estimation and kriging. Python users can combine scikit-gstat (which includes the Cressie-Hawkins estimator) with custom GLS routines. Standalone geostatistical platforms such as GSLIB provide scripting interfaces for robust estimation.
Sources
- Cressie, N. A. C. (1993). Statistics for Spatial Data (revised ed.). Wiley-Interscience, New York. ISBN: 978-0471002550
- Genton, M. G., & Rousseeuw, P. J. (1995). The change-of-variance curve and optimal redescending M-estimators. Journal of Computational and Graphical Statistics, 4(4), 411-432. link ↗
How to cite this page
ScholarGate. (2026, June 3). Robust Universal Kriging. ScholarGate. https://scholargate.app/en/spatial-analysis/robust-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.
- Geographically Weighted RegressionSpatial analysis↔ compare
- Ordinary KrigingSpatial analysis↔ compare
- Spatial Lag ModelSpatial analysis↔ compare
- Universal KrigingSpatial analysis↔ compare