Abstract
The VERE framework was designed and developed in Python to generate hazard confidence maps for seismic-induced landslides, leveraging advanced data analysis and machine learning capabilities. A Virtual Environment (VE) and a Real Environment (RE) containing, respectively, datasets and map sets, are the core of the framework. The Virtual Environment (VE) comprises datasets including morphometric, geotechnical, and hydrological metadata, which are generated assuming a normal distribution, based on representative recurrent values of these parameters in the study area. The Real Environment (RE) includes grid datasets with a common resolution, obtained through analytical preprocessing of various spatial data distributions, including InSAR (Interferometric Synthetic Aperture Radar) data. This data is processed to detect ongoing slope instability and the activity state of surveyed landslides. The framework employs numerical machine learning, trained on meta-solutions derived from an advanced simplified physical model. The model accounts for viscoplastic behavior as well as the reduction of shear strengths toward the residual state during seismic-induced sliding. Hazard confidence maps are produced through an ML-based prediction, considering co-seismic displacements and post-seismic mobility under different initial porewater pressures and seismicity scenarios. The test-site region is the Sele River valley located in an inter-Apennine sector of southern Italy, a seismic-prone area known for its recent seismic activity.
•VE and RE environments modular computation enhances data management and analysis.•InSAR data detects precise spatial slope instability and landslide activity.•Meta-Solutions for training using a viscoplastic model for seismic and post-seismic sliding.•Numerical learning matches complex probability patterns for hazard confidence.•Considers aleatory and epistemic uncertainties, enhancing prediction robustness.