Level: Master’s
Institution: University of Tübingen
Department: Geosciences (Soil Science & Geomorphology)
My contribution: Digital Soil Mapping (DSM) module (lectures + practical sessions)

Overview

This course introduces students to spatial approaches in pedology and geomorphology, with a dedicated module on Digital Soil Mapping (DSM). My teaching focuses on how modern statistical learning and geospatial workflows can be used to produce reproducible, interpretable, and uncertainty-aware soil maps. The module combines theoretical foundations with hands-on training using R and point-based soil observations.

Learning objectives

By the end of the DSM module, students will be able to:

• understand the DSM workflow from problem definition to map production and reporting
• explain core machine-learning concepts relevant to environmental modelling (bias–variance trade-off, overfitting, validation)
• implement and compare common modelling approaches for DSM, including:
  • Decision Trees and Random Forest
  • basic geostatistical concepts and spatial structure in residuals
• prepare and engineer covariates for spatial prediction (terrain, remote sensing, climate/land use where available)
• evaluate model performance using appropriate strategies (e.g., cross-validation and spatial considerations)
• produce soil property maps and communicate prediction uncertainty in a transparent way

Teaching format

The module is structured in two complementary parts:

1) Theoretical sessions

Topics include:

• DSM concepts and the role of environmental covariates (e.g., SCORPAN-inspired thinking)
• supervised learning for spatial prediction: regression and model generalization
• model interpretation: variable importance and response patterns (intro level)
• spatial dependence and why it matters in validation and uncertainty assessment
• uncertainty quantification and good practice in reporting DSM results

2) Practical sessions (R-based)

Students work with point-based soil datasets and covariates to:

• clean and explore soil observations (EDA, outliers, transformations where needed)
• build predictive models (baseline models → Random Forest / tree-based approaches)
• validate models and compare alternatives using robust performance metrics
• generate prediction grids and create maps
• summarize uncertainty and limitations, and document a reproducible workflow

Key topics (DSM module)

• Digital Soil Mapping workflow (from data to map)
• Machine learning fundamentals for environmental prediction
• Decision Trees, Random Forest, and model tuning (intro)
• Geostatistical concepts (spatial autocorrelation, residual structure)
• Model evaluation and spatial validation considerations
• Uncertainty quantification and communication

Tools

• R (data handling, modelling, validation, mapping)
• Common spatial packages (depending on course setup): sf, terra/raster, caret/mlr3, visualization tools

If you are a student in this course and need access to scripts, data links, or assignment instructions, please contact me.