Pioneer vegetation as an indicator of the geochemical parameters in abandoned mine sites using hyperspectral airborne data

A general problem that is associated with open-pit mining is the total displacement of the former landscape’s geocomponents. Overburden conveyor bridges have an irreversible impact on the environment. The consequence of this is a completely new system—the so-called “post-mining landscapes”. These are characterized by extreme geochemical milieus, especially acidity, which is caused by weathering processes (acid mine drainage). These extreme environments are often left to natural succession because of the enormous costs and poor prospects of a successful reclamation. This study investigated the relationship between pioneer vegetation (e.g., Calamagrostis epigejos) and the geochemical properties of heap material using hyperspectral airborne HyMap data in the west Bohemian lignite district near Sokolov (Czech Republic). Thus, a fully constrained linear spectral unmixing algorithm was applied to link the geochemical parameters with fractional abundances. Due to the limitation of pure surfaces in the data set of such fragmented and mixed areas, a three-stage semiautomated endmember extraction was developed. The first stage included the selection of pure surface types using a decision tree with a self-developed narrow band and normalized difference indices. The second stage extracted the potential endmembers using the sequential maximum angle convex cone algorithm. The third stage identified the final endmembers and eliminated the spectral outliers by comparing the extractions to field/laboratory spectral libraries. The abundance fraction image of the pioneer vegetation, which was generated by the linear spectral unmixing, showed the spatial distribution and density of the vegetation. A linear regression model was used to calibrate the abundance image against the pH value. The regression model successfully predicted the pH values of the soil (R ² = 0.67, RPD = 1.42). This method demonstrated that hyperspectral remote sensing data could be used for ecological medium- and long-term monitoring, as well as for spatial assessments in post-mining areas. In addition, this approach works with a small amount of geochemical measurements and the sparse availability of endmembers.