Land use and land cover (LULC) is a key variable of the Earth’s system and has become an important indicator to evaluate the impact of human activities on the Earth’s ecosystems. With the increasing demand of mine resources, widespread opencast mining has led to significant changes in LULC and caused substantial damage to the environment. An efficient approach of detecting mining activities at large scales is of critical importance in mitigating their potential impacts on downstream settlements and in assessing LULC characteristics. In this study, we present a novel approach for enabling large-scale automatic detection of opencast mining areas by integrating multitemporal digital elevation models (DEMs, including the SRTM DEM and the recently released TanDEM-X DEM) and multispectral imagery in object-based image analysis and random forest (RF) algorithms. A sequence of data preparation, image segmentation, threshold analysis, calculation of metrics, and influence factor regulation was developed and tested on the Landsat 8 sample dataset in Inner Mongolia in China, which is a mineral-rich area. Aside from spectral metrics, such as brightness and reflectance value, introduced topographical features enhanced the modeling and classification significantly, and the overall performance is greatly influenced by feature selection (the out-of-bag error rate in the RF algorithm is 7.54% for the integrated DEM method in comparison with 12.70% for the only-optical images method). The integrated use of spectral imagery and multitemporal DEMs reveals that the identified mining area is about 1100 km2 in the study area and period, and the topographic changes of opencast mining in terms of elevation difference is between −258 and 162 m. The results show that the method can map the locations and extents of mining areas automatically from spectral and DEM data and can potentially be applied to larger areas.
Read full abstract