A generalized phenomenological approach for real-time automatic time of failure forecasting of landslides

Human activities, tectonic shifts, and global climatic changes have heightened both the frequency and severity of landslides, creating a growing need for reliable failure time prediction. However, the complexity and variability of monitored displacement data present formidable obstacles to the automated assessment of landslides' time of failure (TOF). As such, a generalized real-time TOF forecasting model is developed in this study to minimize dependency on subjective judgments. Initially, the mutation and trend characteristics of displacement data before and after the onset of acceleration (OOA) from landslides with the exponential deformation pattern were analyzed. Subsequently, a multi-loop time forecast (MLTF) model incorporating exponential moving average (EMA), dynamic typical displacement trend (DTDT) test, automatic OOA identification based on major mutation detection and deformation speed ratio (DSR), and the inverse velocity method (INV) is proposed and validated for reliability with 25 historic landslides' datasets. The results indicate that the MLTF model can automatically identify the OOA of landslides with a low error rate. It significantly narrows the alarm time window and predicted TOF window, which helps reduce false alarms. Additionally, compared to manual INV, the predicted TOF is more accurate and shows better consistency with the actual TOF. Generalizability assessments demonstrate that it can be widely applied in automatic OOA identification and TOF prediction across various deformation patterns of landslides with high reliability.