kriging weights, Oil & Gas Glossary

Kriging weights are the set of numerical coefficients, written as lambda values, that a geostatistical estimator assigns to each measured control point when it predicts the value of a reservoir property at an unsampled location. The defining feature of kriging, which separates it from simpler interpolation schemes such as inverse-distance weighting, is that these weights are not chosen by a fixed geometric rule but are calculated to minimize the estimation variance while keeping the estimate unbiased. To do this the method first models spatial continuity with a variogram or its covariance equivalent, which describes how the squared difference between paired measurements grows with separation distance and direction. The fitted variogram supplies the spatial covariance terms that populate the kriging system, a set of simultaneous linear equations whose solution is the weight vector. In ordinary kriging the weights are forced to sum to one through a Lagrange multiplier, the mathematical expression of the unbiasedness constraint, which guarantees that the average estimation error across many estimates is zero. Minimum variance means the squared error is made as small as the data and the variogram permit. The resulting weights behave intuitively in most respects: a control point close to the target generally receives a larger weight than a distant one, and a tight cluster of points is automatically declustered so the group does not over-count, a behaviour called the screening effect. Less intuitively, kriging weights can be negative, which lets the estimator extrapolate beyond the local data range but can also produce physically impossible negative porosity or permeability if not constrained. In the Western Canadian Sedimentary Basin these weights are the engine behind property models built for the Montney, Duvernay, Cardium, Viking, and Mannville, where engineers and geoscientists interpolate core and log measurements of porosity, permeability, net pay, water saturation, and structural top across hundreds of wells. Because each weight reflects both distance and the modelled anisotropy of the formation, a Cardium model with strong northwest-southeast channel continuity will pull more heavily from wells aligned along that trend than from equally distant wells across the channel axis. The weights, and the kriging variance they generate, feed directly into volumetric reserves estimates and into the uncertainty bounds reported under reserves disclosure standards.

Key Takeaways

Solved, not assigned: Kriging weights come from solving the kriging system, a matrix equation built from the variogram covariance values between every data pair and between each datum and the target. Unlike inverse-distance weighting where the analyst picks the power exponent, the weights are the unique solution that minimizes estimation variance, so two analysts using the same variogram and data get identical weights.
Unbiasedness costs one constraint: Ordinary kriging forces the weights to sum to exactly one, enforced by a Lagrange multiplier added as an extra row and column in the kriging matrix. This is the unbiasedness condition: it removes systematic over or under estimation so the long-run average error is zero, at the small cost of slightly higher estimation variance than simple kriging, which assumes a known mean.
Screening and declustering: A point that sits behind a closer point relative to the target is partly shielded, receiving reduced weight because the closer point already carries that directional information. This screening effect means three wells clustered in one township do not collectively dominate a Montney porosity estimate the way a naive distance average would let them.
Negative weights are real: The covariance structure can produce negative kriging weights for screened or distant points, which is legitimate and improves the estimate, but unchecked it can return negative porosity or permeability. Practitioners either accept and post-process these or apply non-negativity constraints, accepting a small loss in optimality to preserve physical validity.
Anisotropy steers the weights: When the variogram range is longer along depositional strike than across it, kriging weights favour control points aligned with the longer correlation direction. In WCSB channel and shoreface reservoirs this captures real geology, so a Viking shoreface model weights along-shore wells more than across-shore wells at equal distance.

How the Kriging System Produces the Weights

The kriging equations set the weighted covariances between each control point and the others equal to the covariance between each control point and the estimation target, then add the unbiasedness row. Written in matrix form it is a square system the size of the number of conditioning data plus one for the Lagrange term. Inverting that matrix and multiplying by the target covariance vector yields the weight vector and the Lagrange multiplier in a single step. The same inversion also returns the kriging variance, a map-able measure of confidence that is highest near wells and rises in undrilled gaps. For a 150-well Cardium model the system is solved at every grid node, often within a moving search neighbourhood so only the nearest 20 to 40 wells condition any one estimate, which keeps the matrix small and the computation fast.

Why the Weights Matter for Reserves and Development

Because kriging weights translate scattered well measurements into a continuous property field, they directly shape original-oil-in-place and original-gas-in-place volumes. A model that over-weights a single high-porosity well inflates net pay across a wide area and overstates reserves, a risk regulators and reserves auditors scrutinize under AER and securities disclosure rules. The kriging variance that accompanies the weights also defines where new wells reduce uncertainty most, guiding appraisal drilling. In a Montney development a high-variance corner of the lease, far from existing control, is exactly where an operator places a delineation well to sharpen the weights and tighten the reserves range before committing to a full pad program.

Fast Facts

Kriging is named after Danie Krige, a South African mining engineer who in the early 1950s analyzed Witwatersrand gold grades and noticed that estimates based on raw sample averages were systematically biased. The French mathematician Georges Matheron formalized Krige's empirical work into the theory of regionalized variables in 1960, coining the term kriging in his honour. The mathematics that began with gold reef estimation now underpins nearly every commercial reservoir modelling package used across the WCSB, from porosity mapping to seismic-guided property simulation.

Kriging weights cannot be computed without a variogram, which quantifies the spatial correlation the weights depend on, so the two terms are inseparable in practice. The output property fields feed reservoir characterization, the broader discipline of describing rock and fluid distribution between wells. Where kriging gives one smooth best estimate, geostatistics also offers stochastic simulation to generate many equally probable realizations that honour the same weights and variogram while preserving realistic heterogeneity for uncertainty analysis.

Real-World WCSB Scenario: Conditioning a Cardium Porosity Model near Pembina

An operator developing a Cardium tight-oil property in the Pembina field of west-central Alberta builds a porosity model from 180 vertical and horizontal penetrations. Core-calibrated log porosity ranges from 6 to 14 percent, and a variogram fitted to the data shows a 2.8 km range along the northwest-southeast depositional strike and only 1.1 km across it. Feeding this anisotropy into ordinary kriging, the system assigns larger weights to wells aligned along strike, producing a porosity map that mirrors the known shoreface trend rather than smearing values into a circular bullseye around each well. The kriging variance map flags a 1.5-section gap in the northeast quadrant where confidence is poor.

Acting on the variance map, the operator drills a single CAD 4.2 million delineation well into the flagged gap. The new control point cuts the local kriging variance roughly in half, narrows the booked net-pay range, and confirms that the gap holds commercial porosity, supporting sanction of a six-well pad and avoiding a multi-well program in an area the model had previously left in doubt.