Find your next learning step
A pedotransfer function translates soil information that is relatively easy to obtain—such as texture, bulk density, organic carbon and depth—into an estimate of a property that is slower or costlier to measure. Its output can support modelling, but it remains a calibrated prediction. The decisive question is whether the new soil, measurement method and scale resemble the data that taught the function.
Open the chapter contents · Soil Science root · Gold Standard of Education · How X Works
Chapter contents
Choose the route closest to the decision, or read from Chapter 1. The chapters remain open; these grouped routes are a navigable index.
Route 1 · Define the decision
Route 2 · Work the ledgers
Route 3 · Transfer the method
Route 4 · Check limits and evidence
Route 5 · Ask and continue
CHAPTER 1 OF 12 · DEFINE THE DECISION
1. Define the prediction task before choosing a function
Name the target: water content at a stated pressure head, saturated hydraulic conductivity, a retention-curve parameter or another property. A function trained for θ at −33 kPa is not automatically a function for field capacity under drainage in a particular profile.
Also name the output scale, units, horizon and sample support. Volumetric and gravimetric water content are different; log conductivity and conductivity have different error behaviour; point estimates and full parameter curves require different validation.
CHAPTER 2 OF 12 · DEFINE THE DECISION
2. Separate measured predictors from inferred targets
Common predictors include sand, silt and clay fractions, bulk density, organic carbon, depth, pH or cation-exchange capacity. Their availability makes them useful, not error free. Texture methods, sample preparation and horizon aggregation influence the values sent into the function.
The target may itself contain laboratory and model-fitting uncertainty. A pedotransfer function learns the relationship represented in its calibration database, including the database’s measurement protocols and sampling choices.
CHAPTER 3 OF 12 · WORK THE LEDGERS
3. Work a transparent toy function
For teaching only, suppose a toy rule predicts volumetric water content as θ*=0.08+0.004C+0.002O, where C is clay percentage and O is organic-matter percentage. With C=30 and O=2, θ*=0.204.
| Clay C | Organic matter O | Toy prediction θ* | Change from baseline |
|---|---|---|---|
| 30% | 2% | 0.204 | 0 |
| 35% | 2% | 0.224 | +0.020 |
| 30% | 4% | 0.208 | +0.004 |
| 25% | 1% | 0.182 | −0.022 |
CHAPTER 4 OF 12 · WORK THE LEDGERS
4. Propagate input uncertainty rather than hiding it
If clay content is reported as 30±3 percentage points in the toy rule, clay uncertainty alone contributes about ±0.012 to θ*. Uncertainty in organic matter, coefficients and residual model error must be added using an appropriate joint method.
Predictors can be correlated and constrained: sand, silt and clay should sum to approximately 100 percent. Varying each independently across impossible combinations exaggerates or misdirects uncertainty. Use covariance-aware simulation or a model designed for compositional inputs.
CHAPTER 5 OF 12 · WORK THE LEDGERS
5. Change the soil beyond the training domain
Assume a function was calibrated for clay from 5–45%, bulk density from 1.10–1.60 g cm−3 and organic carbon from 0.2–6%. A volcanic soil with 70% clay-equivalent fine material, bulk density 0.85 and unusual mineralogy sits outside several observed ranges.
| Predictor | Calibration range | New soil | Transfer flag |
|---|---|---|---|
| Clay | 5–45% | 70% | Outside |
| Bulk density | 1.10–1.60 g cm−3 | 0.85 g cm−3 | Outside |
| Organic carbon | 0.2–6% | 4% | Inside |
| Mineralogy | Mostly represented classes | Andic properties | Poorly represented |
CHAPTER 6 OF 12 · TRANSFER THE METHOD
6. Compare class and continuous pedotransfer functions
A class pedotransfer function assigns typical hydraulic values to a texture or soil class. It is simple and transparent but can conceal large within-class variation. A continuous function uses measured predictors and may preserve more detail.
Machine-learning functions can fit nonlinear interactions and several input combinations, yet flexibility does not guarantee transferability. Independent validation, applicability-domain checks and calibrated prediction intervals matter more than algorithm prestige.
CHAPTER 7 OF 12 · TRANSFER THE METHOD
7. Connect the established soil-water routes
The soil-water potential guide defines the hydraulic states that retention targets describe. The hydraulic-conductivity guide separates conductivity from storage.
Use the Richards-equation guide to see where predicted functions enter a flow model, and the retention-curve reading guide to keep measurement branch, units and field-capacity conventions visible.
CHAPTER 8 OF 12 · TRANSFER THE METHOD
8. Validate at the decision-relevant scale
Randomly splitting one database can overstate performance when neighbouring horizons or related sites appear in both training and test sets. Spatial, site-wise or region-wise validation better tests transfer to genuinely new conditions.
Evaluate the quantity used by the later model. Small pointwise water-content error may still yield poor available-water difference, conductivity, drainage timing or runoff. Propagate pedotransfer uncertainty through the downstream simulation before judging practical adequacy.
CHAPTER 9 OF 12 · CHECK LIMITS AND EVIDENCE
9. Know the limits and common misconceptions
A pedotransfer estimate is not a measurement and a narrow algorithmic interval is not automatically a full uncertainty interval. Database coverage, measurement protocol, residual variation and structural model error must be represented.
More predictors can improve accuracy in one database while reducing portability or increasing missing-data substitutions. Soil maps also aggregate within-unit variation. A function validated at core scale should not be promoted to a field or region without a scale argument.
CHAPTER 10 OF 12 · CHECK LIMITS AND EVIDENCE
10. Use an evidence and source routine
Record target definition, units, measurement method, predictor methods, calibration database, sample support, range and joint distribution of inputs, missing-data handling, model version, independent validation, applicability domain, prediction interval and downstream sensitivity.
A USDA Agricultural Research Service study shows how pedotransfer accuracy and uncertainty depend strongly on calibration and evaluation databases. The European Commission Joint Research Centre EUPTF3 record documents input combinations, random-forest functions and prediction-uncertainty information for European hydraulic properties.
CHAPTER 11 OF 12 · ASK AND CONTINUE
11. Frequently asked questions
Can a pedotransfer function replace all laboratory measurements?
No. It can screen, initialise or fill gaps, but local measurement remains important when the decision is sensitive or the soil lies outside the calibration domain.
Does a higher R² guarantee useful predictions?
No. Check bias, absolute error, interval calibration, domain coverage and the downstream quantity. A strong global fit can still fail for one soil group.
Should texture percentages be treated as exact?
Usually not. Measurement and spatial uncertainty matter, and the three fractions form a constrained composition rather than independent inputs.
CHAPTER 12 OF 12 · ASK AND CONTINUE
12. Continue with a regional transfer check
Take a pedotransfer function trained on temperate mineral topsoils and apply it to a tropical subsoil with low bulk density and iron-rich mineralogy. List which predictors remain in range, which soil-forming features are absent and which local measurements would most efficiently test transfer.
Continue through the Soil Science root, the Gold Standard of Education and the How X Works hub.
Top · Contents · Previous · How X Works