Multi-resolution Multi-sensor Fusion of Remote Sensing Data to investigate Rock Slope Instabilities

The detailed characterization of rock slopes along highway corridors affected by slopeinstabilities is vital for mitigation design and hazard reduction. In this regard, it is essential torapidly identify the predisposing factors (i.e., rock slope lithologies and mineralogicalcomposition) to evaluate the slope instabilities. However, the traditional approach of fi eld datacollection is laborious, costly, and unfeasible for those territories where landslides orembankment failures are frequent and widespread. The highway networks in the mountainousstate of Colorado encounter several landslides and rockfalls. The use of Unmanned AircraftSystems (UAS) has proven to be an alternative solution to extensive fi eld-based mapping,improving the effi ciency of identifying and classifying geological features of interest. For thecurrent study, a pilot test site at Red Hill, located near Carbondale (CO), was selected as arepresentative case of a road segment exposed to frequent rockfalls. A multi-sensor datasetconsisting of optical, thermal, multispectral, and hyperspectral data was collected from three UAS platforms and processed into high-resolution geospatial products (i.e.,orthoimages and a DEM). A deep learning algorithm was used to integrate and combine such heterogeneous data andcomplementary information provided by the different sensors. The proposed model allowed amulti-sensor and multi-resolution (both spectral and spatial) fusion of the data and subsequentlyperformed an unsupervised classifi cation. We successfully identifi ed the lithological featuresinside of the rock mass that were most likely to cause the slope deformation. The presentworkfl ow will be applied to other site-specifi c locations to assess the approach's effectiveness forrapidly classifying soils and rock types in different geological context and to assist in prioritizingunstable slopes in an asset management framework.