Module 3 · Digital soil mapping

Digital Soil Mapping
Modelling and mapping soil properties with Quantile Regression Forest: covariate selection (Boruta), training, accuracy assessment and tiled prediction with uncertainty.

Format: 2 sessions · Level: Intermediate · Runs in: SoilFER Learn lab (JupyterLab · RStudio · QGIS) · Material: SoilFER Training Manual + Training Resources

What you will learn

  • build the regression matrix from soil points and covariates
  • select covariates and tune a Quantile Regression Forest
  • assess accuracy with cross-validation
  • predict mean and uncertainty maps by tiles and mosaic them

Notebooks

Use the workspace your trainer picks. In a notebook, run the Setup cell first, then the cells in order.

  • JupyterLab opens the notebook in your SoilFER Learn lab, with the project, packages and data already there.
  • RStudio: the same exercises as the original R scripts. RRStudio opens the SoilFER-Training-Resources project; the scripts are in 02_scripts/module3.
Session Open in
Helper · Soil property classes (pH and clay × pH)

JJupyterLab.ipynb

Modelling and mapping SOC with Quantile Regression Forest

JJupyterLab.ipynb

TipRun time

The Kansas example runs in the SoilFER Learn lab. For a whole country, reduce the area or the resolution, or ask your trainer for a larger lab.

TipSee your maps

The last cell of the mapping notebooks, Explore your results, plots your maps and packs them into soilfer_outputs.zip. Open them in QGIS in your Lab (QGIS desktop), or download the zip and drop it on the Map viewer.

First time here? Read Getting started.

Back to top