MAPPING THE UNKNOWN NEIGHBORHOOD: A SIMPLE APPROACH COMBINING TRADITIONAL GEOLOGICAL MAPS AND REMOTE SENSING DATA

Lithological units have been delimited in the traditional geologic maps using qualitative assessments of aerial photography and ground-based observations. Here, I evaluate the applicability of Landsat-8 data for reproducing a traditional geological map and establishing a framework for mapping nearby unexplored areas. I present a case study on mapping Cretaceous sedimentary rocks in Zapatoca, Santander Province of Colombia. The mapping approach described here is a straightforward process: a base—geologic map is used to establish the number and spatial extent of lithostratigraphic units (i.e., lithological classes). The working area is a subset of the map where all lithological classes occur. The remaining area is called the unknown neighborhood and it is used for cross validation. Training samples (n=180), including lithological classes (i.e., Giron, Los Santos, Rosablanca, and Paja) and Rural, Vegetation and Water land cover classes, are collected from Landsad-8 data for the working area. The location of those training samples was stablished using Google Earth imagery. Finally, a supervised classification based on a machine learning algorithm is performed over the full Landsat-8 data (i.e., working area + unknown neighborhood). This approach provides geologically plausible predictions in nearby areas beyond the geological map extent. In average, 44% of the pixel of the Landsat-8 image were classify into the correct lithostratigraphic unit. However, Paja Formation could not be identified and only a few areas of Rosablanca Formation (36%) matched the base—geologic map reflecting the limitations of Landsat-8 data for mapping sedimentary basins with complex tectonic history, high vegetal coverture and extensive weathering.