SAGE-XGBoost Maps Natural Hazards with Scarce Data.

Mapping natural hazard susceptibility is often hampered by a lack of sufficient labeled data, which limits the effectiveness and generalizability of traditional machine learning models and complex deep learning approaches. To overcome this, researchers have developed SAGE-XGBoost, a novel framework designed for robust prediction in data-scarce environments. SAGE (Spatially Augmented Graph Embeddings) is a feature-engineering component that enhances prediction by generating structurally informed features. It achieves this through a combination of controlled noise-based data augmentation and neighborhood-based graph embeddings. A K-nearest neighbor graph is constructed to derive local spatial statistics, which are then reduced using principal component analysis and integrated with environmental covariates and spatial coordinates. These enriched features are then fed into an XGBoost model. The SAGE-XGBoost framework was rigorously tested for both landslide and wildfire susceptibility mapping, consistently outperforming conventional and spatially explicit machine learning models. It achieved impressive AUC values (around 0.97 for landslides and 0.95 for wildfires) and demonstrated an absolute improvement of over 33 percentage points compared to Spatial XGBoost, confirming the significant contribution of graph embeddings to prediction accuracy and spatial coherence.