1 of 12

�Encouraging the Accumulation of Soil Organic Carbon in Soil: General Strategies and Soil-Specific Considerations

Mark Chappell, Ben Kocar, and Yoko Slowey

Environmental Laboratory

U.S. Army Engineer Research & Development Center (ERDC)

Vicksburg, MS

1

DISCOVER | DEVELOP | DELIVER

US Army Corps of Engineers ∙ Engineer Research and Development Center

UNCLASSIFIED

2 of 12

Soil OC storage

Influenced by with:
Soil properties

Soil texture:

Finer soil particle size distribution – attributed to greater surface area for binding humics.
Statistical Pearson correlations with SOC on fine (< 20 µm) particles. Otherwise, no correlations exist

Soil depth (influenced by topography, weathering, and parent material)
Soil mineralogy
Soil “type”

Climate: Temperature and precipitation
Management practices

2

US Army Corps of Engineers ∙ Engineer Research and Development Center

UNCLASSIFIED

3 of 12

SOC thresholds (in theory)

Assuming 1^st order degradation kinetics of SOC, the classical model suggest increases in total C stock is associated with greater C inputs.

The SOC stock will eventually maximize based on the level that an unmanaged soil can sustain.

SOC stocks can further increase through management by adding greater C inputs

However, the soil will eventually maximize the SOC that can be retained
SOC stocks > C saturation point are “unprotected” and expected to degrade rapidly.

SOC threshold represents the difference between the managed and unmanaged maximums.

3

US Army Corps of Engineers ∙ Engineer Research and Development Center

UNCLASSIFIED

4 of 12

Current state of art: Soil behavioral predictions

The conventional goal:

Develop mechanistic knowledge for predicting behavior in soils on any geospatial scale of interest.

Problem with this approach:

Accuracy falls off for areas outside of calibration space. Therefore, mechanistic relationships are globally unreliable.

Theoretical reasons for failures in soil predictions:

Current models are overwhelmingly agnostic to material “type” (statistically referred to as unsupervised).

Based on the assumption that the system is homogeneous – a condition impossible to meet in natural solid phase materials

Soil components are inherently covariate (i.e., inextricably interdependent) in nature, and cannot be arbitrarily separated.
Models treat soils as mono-functional, ignoring the obvious connections between different behaviors

For example, altering the SOC content and type may alter the soil exchange complex and soil acidity.

US Army Corps of Engineers ∙ Engineer Research and Development Center

UNCLASSIFIED

5 of 12

Soil formulation theory (SFT)

Commercial formulations are classified by both their components and their application. We propose the same for soils:
If we consider the application as a “label” that designates function or utility, then:

A different label infers a different array of ingredients. Similarly, a change in the ingredients refers to a different formulation (label).
Thus, combining the label with the ingredients in a model allow us to consider the “formulation” (i.e., material type) in a supervised way.

Soil components are covariately associated with each other.

Formulations defy classical thermodynamic descriptions because (in part) the overall behavior is not the sum of the individual ingredients but the result of their multivariate interactions.
Arbitrary exclusion of ingredients infers a different formulation.

Considering soil “formulations” in terms of “active” vs. “inactive” ingredients may be helpful in context of the full composition of ingredients.

Soils exhibit multimodal behavior that must be accounted for:

As solid-phase materials, soils exhibit a range of diverse geochemical reactions including, Surface redox, ion exchange, surface acidity/alkalinity, dispersion, etc.

File Name

5

US Army Corps of Engineers ∙ Engineer Research and Development Center

UNCLASSIFIED

6 of 12

Soil Taxonomy: Based on morphological features

Inceptisols

Alfisols

Entisols

Ultisols

Soils are most accurately labeled based on pedomorphological descriptions

Allows for inclusion of characteristics related to a soil horizonation
Forms of the basis of soil taxonomic systems (e.g. NRCS, WRB, etc.)
E.g., Fine-silty, mixed, active, thermic Typic Hapludalfs

Advantages of using morphological labels

Labels are “universal”, independent of project data (such as in an unsupervised approach), and highly discriminating
For those trained in soil morphology, these labels can be reliably reproduced from expert to expert.

Disadvantages of morphological labels

No global taxonomy based in morphology
Some labels are more descriptive than others

US Army Corps of Engineers ∙ Engineer Research and Development Center

UNCLASSIFIED

7 of 12

SOC thresholds for different soil “types”

7

US Army Corps of Engineers ∙ Engineer Research and Development Center

UNCLASSIFIED

8 of 12

Formal SFT expression: Compositional Data Analysis (CoDA)

Soils data must be “closed”

Solution: Log-ratio transformations (Aitchinson, 1986, 1999)

Chappell et al., 2019 Geoderma

Solid-phase Formulation Theory: CoDA prohibits univariate consideration of soil components (commonly done in model development). Instead, CoDA enforces the view that soil components are covariate, similar to a formulation, which is appropriate for solid-phase materials.

Proportional data exhibits n-simplex geometry. Common statistical techniques cannot be used until data transformed into normal Cartesian space.

US Army Corps of Engineers ∙ Engineer Research and Development Center

UNCLASSIFIED

9 of 12

Formal SFT expression: Capturing multimodal behavior

Aqueous

measurement

Solid-phase

measurement

Physical

measurement

Weak-acid

extraction

Katseanas et al., 2016; Chappell et al., 2019, PLOS One; Chappell et al., 2019 Geoderma

US Army Corps of Engineers ∙ Engineer Research and Development Center

UNCLASSIFIED

10 of 12

Formal SFT expression: Labels for modeling soil applications

Comparison of “unstacked” vs. “stacked” ensemble ML models

Here, taxonomic designation represents a separate model

Both characterization data (composition) and the taxonomic label are necessary for predictions in view of global soil diversity.

The taxonomic label is especially important for “weak” or poorly relevant characterization data

10

Chappell et al., Geoderma. 2022. 422:115924

US Army Corps of Engineers ∙ Engineer Research and Development Center

UNCLASSIFIED

11 of 12

Future work: SOC sequestration studies

ERDC possesses a soil morphological library of about 500, taxonomically identified samples collected around the world.

This library will be used to create an stacked ensemble model showing the affect of soil taxonomic designation and compositions on C_{max U}

This assumes the soils were collected in their natural C stock asymptote.

C _{max M}can be approximated by organic carbon adsorption experiments.
SOC seq threshold predicted by difference.

11

US Army Corps of Engineers ∙ Engineer Research and Development Center

UNCLASSIFIED

12 of 12

SFT implications for SOC sequestration

SFT overcomes the inherent limitations of natural, solid-phase heterogeneity by explicit inclusion of soil “class” in the prediction model.

Creating a stacked ensemble model that leverages classical soil taxonomic information with site-specific geochemical information provides important increase in model accuracy

Some value out of using even an ambiguous classification scenario (e.g., USGS) over a completely unsupervised approach (but not recommended).

Extending the Critical Zone for ecosystem classifications is key to Climate Change predictions

12

US Army Corps of Engineers ∙ Engineer Research and Development Center

UNCLASSIFIED