Environmental Correlation and Digital Soil Mapping

Landscape soils and surface environments - Week 3 Workshop 2b

Lewis Walden

2026-03-04

Overview of Workshop 2b

Recap: In Workshop 2a we learned about kriging, which uses spatial autocorrelation to make maps

Now: A complementary approach that uses environmental relationships

Key questions:

  • How do soil-forming factors help predict soil properties?
  • What is digital soil mapping (DSM)?
  • When to use DSM vs kriging?

What we’ll cover:

  • Why maps matter in ecology
  • DSM approach and SCORPAN framework
  • DSM workflow and methods
  • Case study and video on mapping ecosystems

Maps and Ecology


Video explaining the importance of maps in ecology, and how they can be used to understand spatial variation and processes.

Quantifying spatial variability, predictive or digital soil mapping

Different approach from kriging: Predict soil properties using environmental relationships

Important

Remember Jenny’s equation from Week 2?

\(S = \mathscr{f}(cl, o, r, p, t)\)

Soil = function of climate, organisms, relief, parent material, time

Digital Soil Mapping (DSM) operationalises this concept

We use spatial datasets (covariates) representing these factors to predict soil properties at unsampled locations.

The SCORPAN framework for DSM

SCORPAN is the modern formulation of Jenny’s equation for spatial prediction:

\(S = f(s, c, o, r, p, a, n) + \varepsilon\)

  • s = soil (existing soil maps, point data)
  • c = climate (rainfall, temperature grids)
  • o = organisms (NDVI, vegetation maps)
  • r = relief (elevation, slope, aspect from DEMs)
  • p = parent material (geology maps, gamma-ray data)
  • a = age (time since disturbance, land use history)
  • n = spatial position (coordinates, distance to features)

These factors are represented as digital spatial layers (covariates) in a GIS.

Digital or Predictive Mapping workflow

  1. Collect training data

  2. Assemble environmental covariates:

    • Climate: rainfall, temperature grids
    • Vegetation: NDVI, land cover maps
    • Topography: DEM, slope, wetness index, etc
    • Parent material: geology maps, gamma-ray data

  3. Extract covariate values at sampling locations

  4. Build model: (regression, random forest…)

  5. Validate model, then apply model and create maps

Comparison: Kriging vs Digital Mapping

Aspect Kriging Digital or Predictive Mapping
Uses Spatial autocorrelation Environmental relationships
Data needed Samples + coordinates Samples + environmental covariates
Sampling Dense/grid preferred Can work with sparser samples
Extrapolation No extrapolation beyond sampled area Can extrapolate if covariates available
Process knowledge Implicit (spatial structure) Explicit (soil-environment relationships)
Interpretation Less process insight Reveals environmental controls

Note

Choice depends on objectives, data availability, and whether you want process understanding.

Which method to use? Depends on objectives and data!

Key decision points:

  • Need a map or just summary statistics?
  • Do you have good environmental covariate data?
  • Is the objective process understanding or just spatial interpolation?

Kriging and DSM are valid—often used together (e.g. kriging residuals from DSM model)

Case study – Mapping Australia’s terrestrial & blue carbon

Study: Walden et al. 2023, Multi-scale mapping of Australia’s terrestrial and blue carbon stocks and their drivers. Read the full paper here

  • Measurement: lab + spectroscopy SOC + remote sensing covariates

  • Sampling: broad, stratified network across ecosystems/regions

  • Prediction: DSM + environmental correlation to map SOC and its drivers with uncertainty

  • Key results: Climate and vegetation dominate SOC patterns at continental scale; within bioregions, ecosystem type, terrain, clay/mineralogy and nutrients become more important drivers.

Open the case study presentation (PDF, ~4 MB)

Video: How mapping ecosystems can help restore them

  • Video showing how scientists use environmental covariates and machine learning to map ecosystems globally and guide restoration decisions.

  • You’ve just seen how we map soils and carbon in Australia using environmental covariates—–this shows the same approach applied at a global scale.

Watch: How mapping ecosystems can help restore them (CNN, 5 min)

Bringing it all together: Week 3 synthesis

Complete spatial analysis workflow:

Workshop 1

  • 3a: HOW to measure (direct, remote, proximal)
  • 3b: WHERE to sample (random, stratified, grid)

Workshop 2 (today)

  • 3a: Kriging (spatial autocorrelation → maps + uncertainty)
  • 3b: DSM (environmental relationships → process insights)

You now have a complete toolkit for landscape-scale spatial studies

Key takeaways

1. DSM uses environmental relationships

SCORPAN factors rather than spatial autocorrelation alone

2. Workflow:

training data → covariates → model → validate → map

3. Choice depends on:

  • Data: grid + coordinates → kriging | samples + covariates → DSM
  • Goal: interpolation → kriging | process understanding → DSM

4. Often combined: DSM model + kriging residuals = best of both worlds

Leave a comment or question