Static and Dynamic Anisotropy in Resource Estimation: Methods and Implications

Anisotropy – directional dependence of spatial continuity – is fundamental in geostatistical resource estimation. Whether anisotropy is treated as static (fixed over a domain) or dynamic (locally varying) controls how grades, geological properties, and associated uncertainty are interpolated and simulated. This has direct implications for mine design, drilling, and risk management (Chilès and Desassis, 2018; Boisvert and Deutsch, 2011).

Static Anisotropy: Concept, Strengths, and Limits

In standard kriging workflows, anisotropy is represented by a single variogram model and ellipsoid per domain. Directional variograms are computed, and principal continuity axes (ranges and orientations) are fitted. This assumes a stationary structure across the domain (Chilès and Desassis, 2018). It is computationally simple, robust, and works adequately when geology is relatively homogeneous or layer‑like. This applies in many sedimentary settings where a single preferred direction of continuity dominates (Chilès and Desassis, 2018).

Static anisotropy is built into many extensions of kriging and random field models. For example, Gaussian Markov random field and SPDE approaches can incorporate anisotropy through a constant coefficient matrix governing directional smoothness (Chilès and Desassis, 2018). In such cases, static anisotropy still enables efficient computation (e.g., very large 2D grids in seconds). It honors a fixed set of preferred directions.

However, in complex folded, faulted, or vein‑hosted systems, a global anisotropy ellipsoid may misrepresent continuity. Structures can curve, bifurcate, or rotate with depth. A single orientation can then (i) connect points that are not geologically related and (ii) ignore continuity along real structural trends. Misplacing high‑grade zones or hydraulic pathways is a risk (Pizzella et al., 2021; Boisvert and Deutsch, 2011).

Dynamic / Locally Varying Anisotropy

Dynamic anisotropy (or locally varying anisotropy, LVA) relaxes stationarity by allowing anisotropy parameters to change with location. Each block (or cell) is assigned its own orientation and, in some methods, its own anisotropy ratio. These are inferred from structural data, folded surfaces, or external covariates (Pizzella et al., 2021; Chilès and Desassis, 2018; Boisvert and Deutsch, 2011).

A widely used geostatistical strategy is to define an exhaustive field of anisotropy orientations and magnitudes. Covariance is then computed using shortest-path distances that follow these local directions. Multidimensional scaling is applied to ensure valid covariance matrices. Standard kriging or sequential Gaussian simulation proceeds in the transformed space (Boisvert and Deutsch, 2011). On US CO₂ emissions data, this locally varying anisotropy approach improved cross-validation. It reproduced nonlinear spatial structures more realistically than static models (Boisvert and Deutsch, 2011).

In airborne geophysics, nested anisotropic kriging with non‑stationary anisotropy grids has been applied to gridding of electromagnetic and magnetic surveys. Locally informed anisotropy “anchor points” are propagated across the domain with smoothing kernels. This reduces artifacts such as boudinage (“string‑of‑beads”) along flight lines and sharpens cross‑line features (Davis, 2021). Model selection tools based on Laplace approximation help decide where increased anisotropy complexity is justified by data. This helps avoid over-parameterization (Davis, 2021).

Beyond kriging, dynamic anisotropy is also embedded in warped-process and SPDE frameworks. Spatial warping is used to accommodate nonstationary anisotropy and varying correlation lengths with depth in 3D seabed property models. This outperforms stationary competitors in predictive accuracy (Bertolacci et al., 2025). SPDE approaches allow the anisotropy matrix to vary in space. Thus, correlation directions follow complex layer geometries inferred from structural models (Chilès and Desassis, 2018).

Role of Structural Data

Structural indicators (e.g., fold axes, hinge lines) are powerful sources of anisotropy information. Implicit geological modelling methods demonstrate that adding fold-axis constraints as higher-order derivatives helps define local anisotropy directions parsimoniously. This avoids arbitrary control points and better captures complex folded structures (Pizzella et al., 2021). This same philosophy underpins dynamic anisotropy in resource estimation. Local structural orientation fields guide the direction and strength of continuity.

Advantages of Static Anisotropy

• Simpler implementation and parameterization.
• Less data‑hungry and robust in regular, gently varying geology.
• Computationally efficient and easily integrated into legacy workflows (Chilès and Desassis, 2018).

Advantages and Risks of Dynamic Anisotropy

• Improved geological realism: better reproduction of curved, channelized, or folded patterns, and of nonlinear features (e.g., emission corridors, palaeochannels, or narrow orebodies) (Davis, 2021; Bertolacci et al., 2025; Boisvert and Deutsch, 2011).
• Better predictive performance: cross-validation and conditional simulations typically show reduced error. They also show more realistic uncertainty structures relative to static models in complex settings (Davis, 2021; Bertolacci et al., 2025; Boisvert and Deutsch, 2011).
• Targeted complexity: model-selection schemes and parsimonious use of structural constraints help add local anisotropy only where data support it. This limits overfitting (Davis, 2021; Pizzella et al., 2021).

Risks include mis‑specified anisotropy fields (leading to biased connectivity and artefacts) and higher computational and conceptual complexity. Careful construction and validation of orientation fields, cross‑validation of predictions, and sensitivity testing of anisotropy parameters are therefore essential.

Outlook

Static anisotropy remains appropriate for many relatively simple deposits and for early-stage assessments. As structural complexity and decision stakes grow, dynamic anisotropy methods, ranging from locally varying kriging to SPDE and warped spatial processes, offer significant gains in realism, prediction quality, and confidence in resource models. This is provided that anisotropy information is well constrained and judiciously applied.