Practical Part

Predicting regional-scale groundwater levels at high spatial resolution using spatial Random Forest models

Description: This practical component follows two theoretical lectures on challenges in groundwater data and modeling. It operationalizes those ideas into a reproducible, hands-on workflow that students can use to predict groundwater levels (GWLs), fill data gaps, and generate monthly 1 km maps for Brandenburg (or a comparable region). Based on theoretical understanding, this module emphasises spatially advised machine learning, especially Random Forest Spatial Interpolation (RFSI).This integrates nearby observations and their distances with environmental covariates to account for spatial dependence explicitly for groundwater resources assessments.

The practical shows how to:

• Large spatiotemporal data handling structures
• Implement and evaluate conventional ML (RF, SVM) versus spatially explicit ML (RFSI)
• Apply spatiotemporal cross-validation to avoid overly optimistic accuracy
• Produce high-resolution, monthly maps
• Gap-fill well time series

Practical part flow (content):

1. Overview of project setup and data sourcing
2. Curate and harmonize groundwater time series; identify and address quality issues
3. Geospatial data handling; groundwater time series and covariates
4. Conventional ML – RF
5. spatially advised ML – RFsp
6. spatially advised ML – RFSI
7. Performance evaluation of ML models
8. Time Series Mapping of Groundwater Levels