Can one representative soil volume support a directional flow tensor?
A scalar property may appear stable while directional flow is still changing with window size. For cracked, layered or structured soil, representative elementary volume work must test the permeability tensor, its principal directions and its sensitivity to boundaries—not only the mean porosity.
This guide belongs to The Gold Standard of Soil Science, within The Gold Standard of Education and the verified How X Works hub.
Choose a reading route
Chapter contents
Open all twelve chapters
- 1. A representative volume belongs to a property and a protocol
- 2. Anisotropic Darcy flow uses a permeability tensor
- 3. Test magnitude, anisotropy and orientation separately
- 4. Worked example: a layered cracked-soil sequence
- 5. Principal directions prevent coordinate-dependent overclaiming
- 6. Off-diagonal terms can survive averaging
- 7. Rotate the tensor, not its entries one by one
- 8. A directional REV can change with hydraulic state
- 9. Directional REV does not replace field sampling design
- 10. Common shortcuts fail under strong anisotropy
- 11. Build a directional upscaling evidence record
- 12. The Gold Standard treats representativeness as tensor convergence
1. A representative volume belongs to a property and a protocol
CHAPTER 1 OF 12 · Orient
A representative elementary volume, or REV, is not a universal cube size attached to a soil. It is the scale at which a specified effective property becomes sufficiently stable under a specified measurement and boundary protocol. Porosity, water content and hydraulic conductivity can therefore reach apparent plateaux at different sizes.
Directional permeability raises the standard. The object is a tensor, so convergence must include its components or eigenstructure. A volume whose scalar average has stabilised may still be too small to represent preferential cracks, platy fabric or layered horizons that carry flow differently by direction.
2. Anisotropic Darcy flow uses a permeability tensor
CHAPTER 2 OF 12 · Build
For an isotropic medium, Darcy flux is parallel to the hydraulic gradient and a single permeability is sufficient. In an anisotropic medium, the relation is vectorial: the flux in one coordinate direction can respond to gradients in several directions. A symmetric positive-definite permeability tensor is the common effective description under standard assumptions.
Diagonal entries in a chosen coordinate system describe responses along those axes; off-diagonal entries describe coupled directions. Rotating the axes changes the component table but not the physical tensor. The principal directions are the eigenvectors, and the corresponding eigenvalues are the principal permeabilities.
3. Test magnitude, anisotropy and orientation separately
CHAPTER 3 OF 12 · Build
A directional REV test should track at least three quantities across nested and translated windows: the overall permeability scale, an anisotropy ratio such as largest to smallest eigenvalue, and the orientation of the principal eigenvectors. Stable diagonal entries alone can be misleading if the principal frame rotates.
Choose tolerances before reading the plots. A practical rule might require every principal permeability to change by less than ten per cent, the anisotropy ratio by less than five per cent and the leading direction by less than five degrees across the largest two window sizes and several translations. The numbers must come from the modelling purpose, not from a generic standard.
4. Worked example: a layered cracked-soil sequence
CHAPTER 4 OF 12 · Work
Consider cubic windows with sides 0.10, 0.20, 0.40 and 0.80 metres. An illustrative upscaling exercise gives horizontal principal permeabilities of 9.2, 7.1, 6.5 and 6.4 units, while the vertical values are 1.4, 1.8, 2.0 and 2.0. The leading horizontal direction rotates from 31 degrees to 18, 13 and 12 degrees.
Between 0.40 and 0.80 metres, the horizontal change is about 1.5 per cent, the vertical value is unchanged and orientation changes by one degree. Those results support a 0.40-metre directional REV under the stated tolerances. They do not prove that 0.40 metres represents another horizon, saturation state or boundary protocol.
| Window side | Largest principal k | Smallest principal k | Leading direction |
|---|---|---|---|
| 0.10 m | 9.2 | 1.4 | 31° |
| 0.20 m | 7.1 | 1.8 | 18° |
| 0.40 m | 6.5 | 2.0 | 13° |
| 0.80 m | 6.4 | 2.0 | 12° |
5. Principal directions prevent coordinate-dependent overclaiming
CHAPTER 5 OF 12 · Interpret
Reporting only Kxx, Kyy and Kzz ties the result to the laboratory axes. If the soil fabric is oblique, off-diagonal terms may be substantial and the diagonal entries do not identify the fastest physical pathway. Eigenvalues and eigenvectors provide a coordinate-aware summary.
The eigenvectors also need uncertainty. When the two largest eigenvalues are nearly equal, their directions can rotate sharply under small perturbations even though the tensor is stable. In that case a principal plane may be identifiable while a single azimuth is not.
6. Off-diagonal terms can survive averaging
CHAPTER 6 OF 12 · Explain
A slanted crack family can drive horizontal flux under a vertical gradient, producing non-zero cross terms in the chosen axes. Averaging does not guarantee that those terms vanish. They disappear only in a coordinate system aligned with the principal directions or because symmetry and sampling make the coupling negligible.
Forcing the tensor to be diagonal before testing can erase real structure. A better sequence is to estimate the full admissible tensor, examine reciprocity and positive definiteness, then rotate to its principal frame. Any imposed symmetry should be stated as a modelling assumption.
7. Rotate the tensor, not its entries one by one
CHAPTER 7 OF 12 · Transfer
When an upscaled tensor moves from sample coordinates into a landscape model, it must be transformed by the coordinate rotation: K′ = R K Rᵀ. Copying the old diagonal numbers into new axes is correct only when the frames are already aligned. The same rule applies when a tilted core is mapped to vertical and horizontal field coordinates.
Tensor rotation preserves eigenvalues but changes components. This gives a useful check: if the reported principal permeabilities change merely because the axes were relabelled, the transformation has been implemented incorrectly.
8. A directional REV can change with hydraulic state
CHAPTER 8 OF 12 · Limit
Saturated intrinsic permeability, unsaturated hydraulic conductivity and relative permeability are not interchangeable properties. In unsaturated soil, connected water pathways and air–water interfaces change with saturation and history. A tensor that is representative at one state may not be representative at another.
Crack opening, shrink–swell and stress can also change geometry. A defensible report therefore records moisture state, stress, temperature and any conditioning sequence. It should not describe a state-dependent tensor as a permanent material constant.
9. Directional REV does not replace field sampling design
CHAPTER 9 OF 12 · Clarify
A cube can be internally representative of one soil fabric yet come from an unrepresentative place in the field. REV analysis handles within-window homogenisation; site sampling handles variation among horizons, landforms and management zones. The two scales answer different questions.
Use multiple translated windows inside each specimen and multiple specimens across the site. Nested windows diagnose convergence, while replicated locations diagnose population variability. Combining them prevents a precise tensor from one convenient core being mistaken for a field-wide law.
10. Common shortcuts fail under strong anisotropy
CHAPTER 10 OF 12 · Audit
A porosity plateau does not prove permeability convergence. One pressure-gradient direction does not identify a tensor. A single centred window does not reveal translation sensitivity. A diagonal fit does not demonstrate that cross terms are zero. These shortcuts all replace evidence with symmetry assumptions.
Another limit is scale separation. If a few domain-spanning cracks control flow, no comfortable plateau may exist before the model domain is reached. The honest conclusion may be that an equivalent continuum tensor is not supported and explicit fracture representation remains necessary.
11. Build a directional upscaling evidence record
CHAPTER 11 OF 12 · Check
Archive the segmented or structural input, voxel or measurement resolution, window sizes, translations, boundary conditions, load cases and solver checks. Report the full tensor before rotation, its eigenvalues, eigenvectors, anisotropy ratio and uncertainty. State the tolerance that defined convergence.
Then test transfer: change window origin, rotate the coordinate frame and compare a second boundary protocol. A reported REV should survive these checks within the stated tolerance, or its limitations should be carried into the larger model.
| Evidence item | Minimum record | Failure it prevents |
|---|---|---|
| Domain series | sizes and translations | a plateau caused by one convenient origin |
| Flow tests | independent gradients and boundaries | an underidentified tensor |
| Tensor checks | symmetry, positivity, eigenstructure | nonphysical or coordinate-bound results |
| Transfer check | rotation, state and second protocol | exporting an REV beyond its experiment |
12. The Gold Standard treats representativeness as tensor convergence
CHAPTER 12 OF 12 · Use
For anisotropic soil, the decision is not “the mean stopped moving.” It is that the effective tensor, its principal magnitudes and its relevant directions are sufficiently stable across size, translation and justified boundary choices for the intended hydraulic state.
That conclusion is deliberately conditional. It names the property, state, geometry, tolerance and model use. When those conditions change, representativeness must be tested again rather than inherited.
A changed-context transfer check
A horizontal core yields a stable diagonal tensor at 0.40 metres. The values are inserted unchanged into a hillslope model whose axes follow the slope and vertical. This silently rotates the physical problem without rotating the tensor.
The correct transfer constructs the rotation matrix from core to model coordinates and applies K′=RKRᵀ. It then checks whether the new domain and moisture state still satisfy the convergence evidence. If off-diagonal terms appear after rotation, that is expected geometry—not a defect to delete.
Frequently asked questions
Is the REV always a cube?
No. Shape and orientation are part of the averaging protocol. An elongated or directional domain may be needed when correlation lengths differ.
Can porosity and permeability share one REV size?
Sometimes, but it must be shown. Connectivity-sensitive permeability often converges differently from scalar porosity.
Must a permeability tensor be diagonal?
Only in its principal coordinate frame or under a justified symmetry. In arbitrary axes, off-diagonal terms may be physically meaningful.
What if no plateau appears?
Report that a continuum tensor is not supported at the available scale, enlarge the domain if possible or retain explicit structural detail.
Authoritative sources and evidence route
- Li and colleagues: permeability tensor and representative elementary volume of saturated cracked soil
- Rong and colleagues: permeability tensor and REV of fractured rock masses
- Comparison of REV size and tensor characteristics for fractured-rock conductivities
- Representative elementary volumes, scale convergence and sampling bias