New methods for the integration of multisource landslide monitoring systems

A key concept in landslide early warning systems (LEWS) is the timely detection of ground deformations deemed hazardous and the reliable prediction of imminent collapses. These aspects constitute two fundamental pillars of LEWS: lead time—the time available before the final paroxysmal acceleration, which is essential for evacuation—and landslide predictability—the required degree of reliability for spatial and temporal predictions. In reality, however, there are few regional-scale monitoring systems, and they often struggle to distinguish genuine early-warning signals from seasonal variations and noise. Furthermore, once hazardous phenomena are identified, there is a lack of refinement procedures based on prediction methods grounded in the mechanical and physical principles of failure-prone systems, as well as a lack of forecasting tools that inherently quantify prediction reliability. In this context, this doctoral dissertation presents a multiscale approach with a modular structure that enables transferability and increased robustness for landslide hazard detection and spatio-temporal failure forecasting. The proposed framework is composed of three complementary modules that are calibrated, developed, and validated using satellite-based Interferometric Synthetic Aperture Radar (Sat-InSAR) data covering the June 24, 2017 Xinmo landslide in China. In this case, the entire methodology is applied sequentially, with the output of each module serving as the input for the next, enabling progressive refinement of hazard identification and prediction confidence through an increasing scale of analysis—from regional to slope scale, and from multivariate to univariate analysis. The individual modules have also been applied to independent case studies to further test the overall robustness. Overall, the results demonstrate that integrating regional-scale deformation screening, physically informed spatial analysis, and confidence-based temporal forecasting significantly improves the reliability, robustness, and operational value of automated LEWS frameworks.