A neural network method for analysis of hyperspectral imagery with application to the Cassas landslide (Susa Valley, NW-Italy)

Geomorphological analysis is an essential element of many landslide monitoring and hazard assessment studies. Remote sensing techniques have long been an effective tool for such analyses. This paper presents results regarding a geomorphological investigation performed exploiting information derived from high-resolution hyper-spectral airborne sensor MIVIS (Multi-spectral Infrared and Visible Imaging Spectrometer) images. The authors propose a workflow based on neural network algorithms used both for image geometric correction and classification. The case study area is the Middle Susa Valley (Italian Western Alps). It includes the Cassas landslide, a well-known complex landslide affecting slopes on the southern side of the valley, near the town of Salbertrand. At this location several “lifelines” (railway, main roads, hydroelectric tunnels, etc.) are concentrated, and they suffer from a constant threat of landslide activity. Traditional geomorphological characterization of the landslide slope was improved by comparing field campaigns with further information extracted from remote-sensed datasets. Presented results include the identification and characterization of major elements of the present-day active unstable slope (debris-covered areas, fractured/disjointed rock walls, landslide accumulation borders), as well as of individual structural features and landforms (major faults and fractures, trenches, elongated depressions, counterslope scarps), related to long-term deep-seated gravitational slope deformation.