Landslide Susceptibility Mapping

Landslide susceptibility mapping estimates the relative likelihood of landslide initiation based on terrain, geology, hydrology, and land use. Unlike runout models, it focuses on where slides may start, not how far they will go. Methods range from expert heuristic overlays to statistical and machine-learning models (logistic regression, random forests, gradient boosting) trained on inventories of past slides and non-slide controls. Predictor variables commonly include slope, aspect, curvature, lithology, soil depth, fault proximity, precipitation, NDVI, and road or river proximity. Because training data can be biased toward accessible areas, balanced sampling and spatial cross-validation are essential. The output is often a susceptibility index categorized into classes for planning. Uncertainty and temporal dynamics (e.g., post-fire periods) should be explicit.