When is a soil volume large enough to speak for the property?
Representative elementary volume analysis connects pore-scale images to continuum claims. The crucial discipline is to treat scale convergence as property-specific evidence, then keep image resolution and site sampling bias separate.
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. Why a measurement needs a representative volume
- 2. Read the classic three-scale picture carefully
- 3. Different soil properties converge at different rates
- 4. Design nested and translated windows
- 5. Worked example: convergence of an illustrative porosity estimate
- 6. A plateau requires an explicit tolerance
- 7. Small windows suffer boundary and finite-size bias
- 8. Voxel resolution and REV answer different questions
- 9. Representative volume is not representative site sampling
- 10. Changed-context check: move from porosity to saturation
- 11. Evidence that a reported REV can travel
- 12. The Gold Standard scale decision
1. Why a measurement needs a representative volume
CHAPTER 1 OF 12 · Orient
A soil image can resolve pores at micrometre scale yet still describe too little material to support a field-scale claim. A representative elementary volume, or REV, is a sampling scale at which the measured property becomes sufficiently stable for the intended continuum description.
The word representative is property- and tolerance-specific. Porosity may stabilise in a smaller window than water saturation, connectivity or air–water interfacial area. There is therefore no single permanent REV stamped onto a soil.
2. Read the classic three-scale picture carefully
CHAPTER 2 OF 12 · Build
At very small windows, the estimate fluctuates strongly because moving the window crosses individual grains and pores. At an intermediate range, variation may settle enough to support a homogeneous continuum value. At still larger scales, genuine layers or structural trends can reappear.
That middle plateau is evidence for a useful REV only within the sampled domain and property. If the material is hierarchical or nonstationary, the apparent plateau can be short, directional or absent.
3. Different soil properties converge at different rates
CHAPTER 3 OF 12 · Build
Porosity is an additive volume fraction and often settles comparatively early. Permeability depends on connected pathways and bottlenecks, so rare throats can keep it variable at larger windows. Saturation and interfacial area also respond to phase configuration, wetting history and image resolution.
A volume that represents the solid fraction does not automatically represent flow. Reusing a porosity REV for hydraulic conductivity without a new convergence check is a category error.
4. Design nested and translated windows
CHAPTER 4 OF 12 · Measure
Nested windows reveal how one location changes as the observation volume grows. Translated windows of the same size reveal location-to-location variation. A defensible study uses both: otherwise a conveniently centred nested sequence can hide spatial bias.
Where possible, repeat the procedure along multiple orientations and across independent specimens. Cubic windows are convenient, but anisotropic soil structure may require direction-aware dimensions rather than one scalar side length.
5. Worked example: convergence of an illustrative porosity estimate
CHAPTER 5 OF 12 · Work
Imagine a segmented soil volume analysed with many translated cubic windows. At each side length L, compute the mean porosity, standard deviation across positions and coefficient of variation. The numbers below are invented to demonstrate the decision process.
Suppose the rule requires the mean to change by less than 0.005 between successive sizes and the positional coefficient of variation to fall below 2%. The 4 mm mean is close to the 8 mm mean, but its coefficient of variation is still about 2.3%. Under this rule, 4 mm has not yet passed.
| Cubic side L | Mean porosity | SD across positions | Coefficient of variation |
|---|---|---|---|
| 1 mm | 0.370 | 0.060 | 16.2% |
| 2 mm | 0.355 | 0.025 | 7.0% |
| 4 mm | 0.348 | 0.008 | 2.3% |
| 8 mm | 0.347 | 0.003 | 0.9% |
6. A plateau requires an explicit tolerance
CHAPTER 6 OF 12 · Work
Visual flattening alone invites wishful judgement. State a tolerance tied to the downstream decision: absolute change, relative change, confidence interval width, coefficient of variation or prediction error. Report how many translated windows support the estimate.
The criterion should be chosen before inspecting the final curve when possible. A threshold tightened after seeing the data can turn one scan into a self-confirming result.
7. Small windows suffer boundary and finite-size bias
CHAPTER 7 OF 12 · Audit
Cropping breaks pores, throats and phase interfaces at window faces. Counts based on connected components, Euler characteristic or spanning pathways can be especially sensitive. The issue is not only greater scatter; the mean estimate itself can be biased.
Periodic boundaries in simulations may reduce one artefact while introducing another idealisation. In physical images, use boundary corrections where defined, compare interior and edge windows, and report the convention.
8. Voxel resolution and REV answer different questions
CHAPTER 8 OF 12 · Audit
Growing a window does not recover pores smaller than the voxel size. Resolution controls what structures enter the measurement; volume controls how much of the resolved structure is sampled. A large low-resolution scan can be representative of the wrong geometry.
Use a two-axis audit: first test whether the property is stable as resolution changes, then test whether it is stable as volume changes. Multi-resolution imaging or synthetic down-sampling can reveal which axis drives the result.
9. Representative volume is not representative site sampling
CHAPTER 9 OF 12 · Transfer
An REV within one core says nothing about whether the core represents the field. Preferentially choosing intact, convenient or visually uniform cores can exclude cracks, roots, aggregates and horizon transitions that control field behaviour.
The sampling design therefore has two levels: select locations and specimens to represent the population, then select an observation volume large enough to represent the chosen specimen for the target property.
10. Changed-context check: move from porosity to saturation
CHAPTER 10 OF 12 · Transfer
A dry packing can show stable porosity while a partially saturated state contains patchy water clusters. Applying the dry-porosity REV to moisture saturation may under-sample those clusters and bias interfacial-area estimates.
Repeat the translated-window analysis for each phase property and relevant wetting state. Hysteresis means drainage and imbibition can require different volumes even at the same bulk saturation.
11. Evidence that a reported REV can travel
CHAPTER 11 OF 12 · Audit
Report specimen origin, preparation, dimensions, voxel size, segmentation method, window shape, positions, orientation, property formula, tolerance and uncertainty. Show the full convergence curve rather than only the chosen scale.
A useful transfer statement names the range: this volume was adequate for this property, material, state, resolution and tolerance. It does not silently become a universal soil constant.
| Property | Why convergence may differ | Required evidence |
|---|---|---|
| Porosity | Local volume fraction | Mean and positional variance |
| Saturation | Patchy phase occupancy | State-specific windows |
| Permeability | Connected bottlenecks | Directional flow and boundary tests |
| Interfacial area | Resolution and phase geometry | Voxel and window sensitivity |
12. The Gold Standard scale decision
CHAPTER 12 OF 12 · Use
Start with the decision, not the biggest available scan. Define the property and error tolerance, sample translated and nested windows, test mean and variability, inspect boundary and resolution effects, and repeat across independent specimens.
When no stable plateau appears, report that result. A larger domain, a directional REV, a hierarchical model or explicit random field may be more honest than forcing a continuum number.
A changed-context transfer check
A 4 mm cube met a porosity-mean tolerance in one homogeneous sand scan. The same team wants to use it for air–water interfacial area during imbibition. The claim does not transfer automatically: the phase interface is resolution-sensitive, spatially patchy and history-dependent.
The correct check repeats translated windows for interfacial area at that wetting state, varies voxel resolution where possible, and evaluates whether mean and variability both settle. If they do not, report the available scale range instead of inventing an REV.
Frequently asked questions
Is the REV the smallest scanned volume?
No. It is a property- and tolerance-specific scale supported by convergence evidence. The smallest scan may sit in the fluctuating pore-scale region.
Can one REV serve every soil property?
Usually not. Additive fractions, connected transport and interfacial geometry can converge at very different scales.
Does a stable mean prove representativeness?
Not by itself. Positional variability, independent specimens, boundary effects, resolution and population sampling still matter.
What if no plateau appears?
Use a larger or directional domain, a hierarchical or stochastic model, or report that the available volume did not establish an REV.
Authoritative sources and evidence route
- Costanza-Robinson, Estabrook and Fouhey: REV estimation for porosity, saturation and air–water interfacial area
- Yuan, Chareyre and Darve: finite-size effects and representative volume in pore-scale drainage simulations
- Middlebury College repository copy of the REV study
- Minkowski functionals, Euler characteristic and geometric state closure
- Persistent homology, filtration scales and pore connectivity