Machine Learning for Natural Hazard Prediction

Overview

This tutorial provides a hands-on introduction to machine learning-based landslide susceptibility prediction using scikit-learn in Python.
Participants will learn how to prepare balanced datasets, select and evaluate models, and analyze the effects of negative sample selection, cross-validation strategies, and ensemble modeling on prediction robustness and generalization.
All exercises are implemented in Google Colab for reproducibility and easy access.

Key Topics

• Fundamentals of Landslide Susceptibility Mapping (LSM)
• Preparation of geomorphic and environmental features
• Negative sample selection: comparing random sampling vs. Positive–Unlabeled (PU) Bagging
• Validation strategies: random vs. spatial cross-validation
• Training and evaluating ML models with scikit-learn: LR, SVM, DT, RF, GBM
• Ensemble modeling and whole-region inference for large-scale prediction
• Uncertainty estimation for model interpretability and reliability
• Performance metrics: Accuracy, Precision, Recall, F1, and AUC

Who Should Attend

Researchers, students, and practitioners interested in:

• Geohazard modeling and landslide risk assessment
• Machine learning applications in geoscience and remote sensing
• Data-driven slope stability and regional hazard prediction

No prior experience in engineering geology or machine learning is required;
basic familiarity with Python and scikit-learn is helpful but not mandatory.

Learning Outcomes

By the end of the tutorial, participants will be able to:

• Prepare balanced datasets for landslide susceptibility analysis
• Understand importance of reliable negative samples for modeling with presence-only data
• Compare model robustness under random and spatial cross-validation
• Train and evaluate multiple scikit-learn models and ensembles for LSM
• Perform whole-region susceptibility inference using ensemble outputs
• Estimate and visualize prediction uncertainty to assess confidence levels
• Interpret the physical and statistical implications of model predictions

Resources

Landslide Susceptibility Prediction Tutorial (Google Colab):
Open in Colab

• [Relevant Papers]
  • Pei, T., & Qiu, T. (2024). Landslide susceptibility mapping using physics-guided machine learning: a case study of a debris flow event in Colorado Front Range. Acta Geotechnica, 19, 6617–6641. https://doi.org/10.1007/s11440-024-02384-y
  • Goldberg, R., Pei, T., Moshary, F., & Tian, Y. (2025). Advancing landslide susceptibility mapping across heterogeneous regions with deep learning-based domain adaptation. In Geotechnical Frontiers 2025 (pp. 30–39). https://doi.org/10.1061/9780784485989.004
  • Liu, J., Shen, C., Pei, T., Kifer, D., & Lawson, K. (2025). The value of terrain pattern, high-resolution data and ensemble modeling for landslide susceptibility prediction. Journal of Geophysical Research: Machine Learning and Computation, 2, e2024JH000460. https://doi.org/10.1029/2024JH000460