Date of Graduation


Document Type


Degree Type



Eberly College of Arts and Sciences


Geology and Geography

Committee Chair

Timothy A Warner

Committee Co-Chair

Jamison Conley

Committee Member

J Steven Kite

Committee Member

Michael P Strager

Committee Member

Nicholas P Zegre


The aim of this dissertation was to investigate the potential for mapping land cover associated with mountaintop mining in Southern West Virginia using high spatial resolution aerial- and satellite-based multispectral imagery, as well as light detection and ranging (LiDAR) elevation data and terrain derivatives. The following research themes were explored: comparing aerial- and satellite-based imagery, combining data sets of multiple dates and types, incorporating measures of texture, using nonparametric, machine learning classification algorithms, and employing a geographical object-based image analysis (GEOBIA) framework. This research is presented as four interrelated manuscripts.;In a comparison of aerial National Agriculture Imagery Program (NAIP) orthophotography and satellite-based RapidEye data, the aerial imagery was found to provide statistically less accurate classifications of land cover. These lower accuracies are most likely due to inconsistent viewing geometry and radiometric normalization associated with the aerial imagery. Nevertheless, NAIP orthophotography has many characteristics that make it useful for surface mine mapping and monitoring, including its availability for multiple years, a general lack of cloud cover, contiguous coverage of large areas, ease of availability, and low cost. The lower accuracies of the NAIP classifications were somewhat remediated by decreasing the spatial resolution and reducing the number of classes mapped.;Combining LiDAR with multispectral imagery statistically improved the classification of mining and mine reclamation land cover in comparison to only using multispectral data for both pixel-based and GEOBIA classification. This suggests that the reduced spectral resolution of high spatial resolution data can be combated by incorporating data from another sensor.;Generally, the support vector machines (SVM) algorithm provided higher classification accuracies in comparison to random forests (RF) and boosted classification and regression trees (CART) for both pixel-based and GEOBIA classification. It also outperformed k-nearest neighbor, the algorithm commonly used for GEOBIA classification. However, optimizing user-defined parameters for the SVM algorithm tends to be more complex in comparison to the other algorithms. In particular, RF has fewer parameters, and the program seems robust regarding the parameter settings. RF also offers measures to assess model performance, such as estimates of variable importance and overall accuracy.;Textural measures were found to be of marginal value for pixel-based classification. For GEOBIA, neither measures of texture nor object-specific geometry improved the classification accuracy. Notably, the incorporation of additional information from LiDAR provided a greater improvement in classification accuracy then deriving complex textural and geometric measures.;Pre- and post-mining terrain data classified using GEOBIA and machine learning algorithms resulted in significantly more accurate differentiation of mine-reclaimed and non-mining grasslands than was possible with spectral data. The combination of pre- and post-mining terrain data or just pre-mining data generally outperformed post-mining data. Elevation change data were shown to be of particular value, as were terrain shape parameters. GEOBIA was a valuable tool for combining data collected using different sensors and gridded at variable cell sizes, and machine learning algorithms were particularly useful for incorporating the ancillary data derived from the digital elevation models (DEMs), since these most likely would not have met the basic assumptions of multivariate normality required for parametric classifiers.;Collectively, this research suggests that high spatial resolution remotely sensed data are valuable for mapping and monitoring surface mining and mine reclamation, especially when elevation and spectral data are combined. Machine learning algorithms and GEOBIA are useful for integrating such diverse data.