Evaluation of potential of multiple endmember spectral mixture analysis (MESMA) for surface coal mining affected area mapping in different world forest ecosystems

Surface coal mining (SCM) has undergone dramatic changes in the last 30years. Large-scale SCM practices are at the center of an environmental and legal controversy that has spawned lawsuits and major environmental investigations. SCM techniques extract multiple coal seams by removing an area of many square kilometers and creating serious environmental problems. Information about mining activities location is essential for environmental applications, specifically the temporal and spatial patterns of land cover/land use change (LCLUC). Advancements in satellite imagery analysis provide possibilities to investigate new approaches for LCLUC detection caused by SCM globally. However there is no study that analyzes the changes produced for SCM at a global scale. Our work examines three areas of coal extraction in the world: Spain, United States of America (USA), and Australia. We used Multiple Endmember Spectral Mixture Analysis (MESMA) applied to Landsat Thematic Mapper (TM) data to map SCM affected area. Endmember spectra of vegetation, soil, and impervious surfaces were collected from the Landsat TM image with the help of a fine resolution orthophotographs and the pixel purity index (PPI). Reference endmembers from an Airborne Visible-Infrared Imaging Spectrometer (AVIRIS) spectral library were utilized as well. An unsupervised classifier was applied to the fraction images to obtain an estimation of active SCM affected area. Classification accuracy was reported using error matrixes and κ statistic using active SCM affected area perimeters digitized from fine resolution orthophotographs as reference data. In addition, we compared the accuracy of the MESMA based estimation to estimates using Spectral Mixture Analysis (SMA), and a spectral index traditionally used as Normalized Difference Vegetation Index (NDVI) testing statistical significance using a Ζ-test of their κ statistics. Results showed a significant improvement in the accuracy of the SCM affected area using MESMA with an average increase of the κ statistic of 31%. We conclude that MESMA-based approach is effective in mapping SCM active affected area.