Brief

Researchers have developed a new unsupervised machine learning framework called spatiotemporal Local Intrinsic Dimensionality (st-LID) to improve the early detection of catastrophic slope failures. This method enhances traditional LID by incorporating velocity to capture deformation rates and using Bayesian spatial fusion to aggregate data across neighborhoods, accounting for noise and spatial correlations. Additionally, a temporal modeling component (t-LID) characterizes long-term displacement dynamics, enabling the identification of complex, multi-stage failure zones that current methods often miss. Experiments demonstrate st-LID's superior performance in detection precision and lead-time compared to existing unsupervised techniques, offering a more robust foundation for landslide early warning systems. AI