ALBERT

Description: Optical remote sensing has in the last two decades been extensively tested for the detection of hydrocarbons at the Earth's surface. The spectral absorption features of seepage-related hydrocarbons can easily be confused with those of man-made bituminous surfaces such as tarred roads. The characteristic low albedo of bituminous surfaces can, at the same time, easily be confused with other dark surfaces such as shade. This paper presents the results of two pixel-based classifications that have been carried out on hyperspectral imagery acquired over seepage areas. The first classification algorithm is a minimum distance to class means' (MDC), which is sensitive to spectral absorption features as well as albedo differences. The second algorithm is a spectral angle mapper' (SAM), which is not sensitive to albedo differences. Both algorithms are applied for the detection of crude oil resulting from macroseepage and an anomalous halo of bare soil resulting from microseepage. The results show that, at best, only 48% and 29% of the pixels that respectively contain crude oil and seepage-related bare soil could be detected, with the inclusion of many false anomalies. Confusion mainly results from the physical characteristics of the anomalies, as these are not unique to seepages. It is concluded that remote sensing of natural hydrocarbon seepages can be improved by image processing algorithms that make use of spatial information.