Predictive Modeling of Debris Flow Initiation Using Multi-Sensor Data Fusion and Graph Convolutional Networks for Enhanced Early Warning Systems

**Abstract:** Existing debris flow early warning systems (DEWS) often struggle with high false alarm rates and limited predictive accuracy, particularly in complex terrains. This research proposes a novel framework utilizing multi-sensor data fusion, specifically integrating rainfall intensity, soil moisture, groundwater levels, displacement data from inclinometers, and LiDAR-derived topographic features. These data streams are processed using a Graph Convolutional Network (GCN) that models the spatial relationships and interdependencies between sensors and terrain characteristics, allowing for a highly localized and nuanced prediction of debris flow initiation. The model demonstrably improves predictive accuracy and reduces false alarms compared to traditional threshold-based or machine learning approaches, offering a pathway toward more reliable and actionable DEWS.