Satellite multispectral images analysis to develop a rock classification method

The non-access areas are a very problematic issue in Geological survey, that cause lack of data providing low quality results. The first step of the study is the analysis of the 13 bands (1,2,3,4,5,6,7,8,8A,9,11,12) of the Sentinel project multispectral images, in order to create a recognizable reflectance footprint of every outcropping lithology, in eastern Sicily. The second step start with a machine learning training by the ESRI ArcGIS software with the aim of separate every lithological outcrop pictured in the orthophotos from the vegetation, soil and anthropic objects. The main problem is the different resolutions of the orthophotos and the multispectral images (0.25m and 30m, respectively). Consequently we create a fishnet (Grid) 30m x 30m, corresponding to a pixel of the multispectral images, and we extrapolate, for each cell, the number of pixels representing each category (e.g. vegetation) utilizing the categorized orthophoto. Finally, using the Microsoft Excel software, we extrapolate only the reflectance footprint referred exclusively to the lithological outcrops in every cell for the 13 bands. Mediating the data, we will obtain a reflectance range for every recognizable lithologies in outcrop. In order to validate the elaborated data will be necessary compare the reflectance footprint elaborated with our methodology with the values measured in the field by an optical spectrometer. The main goal of the study, after the acquisition of thousands of field data for every lithology is the mapping of the outcropping terranes of an area using only satellite images. Another item is to recognize a variation of reflectance in the fault rocks outcropping along the main regional tectonic lineaments.