Debris flows are natural hazards that can damage ecosystems and infrastructure, especially in mountainous areas after wildfires. Debris flow mapping and inventory development provide the groundwork for understanding the frequency, spatial distribution, and key influencing parameters that help identify high-risk areas. This study evaluates parameters associated with post-fire debris flow mapping by analyzing lidar differencing, satellite imagery, and machine learning (ML) applications. The study area includes seven watersheds in the Columbia River Gorge, Oregon, USA. The combination of wildfires in 2017, followed by an intense rainstorm in 2021, led to multiple debris flow events in the area. Lidar and satellite imagery datasets from 2018 and 2021 are analyzed using machine learning algorithms: Random Forest, XGBoost, SVM, and logistic regression. In implementing these models, we focus on feature selection optimization and handling class imbalance. Random Forest and XGBoost performed best, achieving an F1 score of 85% to 90% in mapping debris flow locations. Spatial visualization of the results validated the models against historical data. Adding soil burn severity to the analysis highlights its influence on sediment erosion and vegetation patterns. The study showed that the watersheds with higher burn severity areas experienced more significant sediment erosion.
