Civil Engineering Departmental Seminar - Landslide Susceptibility Mapping using Physics-Guided Machine Learning (PGML): An Application to A Debris Flow Event in the Colorado Front Range

Physics-guided machine learning (PGML) is a new paradigm for applying machine learning (ML) techniques to solve engineering problems. This talk will demonstrate PGML's capability to improve traditional ML models for mapping and predicting debris-flow hazards through a case study of debris flows triggered by a storm event in the Colorado Front Range. The spatial distribution of the debris flows was assessed using three distinct classes of methods: a physics-based model, a traditional ML model, and a PGML model. For the physics-based model, the infinite-slope model under submerged conditions was used to calculate the factor of safety and assess debris-flow risk. For the PGML model, a physics-guided loss function was implemented to incorporate debris-flow physics into traditional ML models. It was found that PGML can improve ML model performance for mapping debris flow hazards by enhancing physical consistency and making ML models more generalizable across ecoregions.