University of Twente Student Theses


Analysis of opencast mining characteristics using multi-temporalremote sensing images and Google Earth Engine

Shankaraiah, Jitendra (2021) Analysis of opencast mining characteristics using multi-temporalremote sensing images and Google Earth Engine.

[img] PDF
Link to full-text:
(only accessible for UT students and staff)
Abstract:Opencast coal mining is the most economical and also the most intense mining practice that drastically alters the land use land cover (LULC) in and around the mining areas. Coal fires are found in coal mines where combustion of coal is extrinsically affected by mining practices. Recent launches of Earth Observation satellites have provided open access to their data thereby giving the opportunity to continuously map opencast coal mines, coal fires and their rapid changes using relatively high spatial and temporal resolution images. This study focuses on using the computational resources of Google Earth Engine (GEE) and its cloud platform for developing web-based application to demonstrate the characteristics of opencast coal mining at Jharia coal field (JCF) in India. Openly accessible multi-temporal data from Sentinel-2 and ECOsystem Spaceborne Thermal Radiometer Experiment on Space Station (ECOSTRESS) were used to respectively characterize active coal mining areas, areas affected by coal fires, and their changes between 2019 and 2020. A Random Forest (RF) classification algorithm was used to classify coal mining areas and other LULCclasses with samples generated fromrandom sampling approach, and applying non-heuristic data balancing techniques to address class imbalance problem. The classification achieved overall accuracy of 73.09% and 76.5% for 2019 and 2020 respectively. The change detection results found that the coal mining activity decreased from 16.30 km2 in 2019 to 12.51 km2 in 2020. Two dynamic threshold based methods were applied to detect areas affected by coal fires using Land Surface Temperature (LST) images. First method used images with minimum cloud cover and achieved 90% accuracy at detecting coal fires in 2019. A detection ratio was calculated to measure how consistently the pixels were detected as coal fires. The first method achieved 70% accuracy for a detection ratio above 0.5. The LST variability in day and night time images showed that the diurnal images could be independently analysed. The second method used all images to create a 95th percentile composite of day and night images, and achieved 75% accuracy at detecting coal fires. Both methods showed an increase in the extent of the coal fires during the observation period with new fires propagating around the active fires. The results of this study show effective use of medium-high resolution Sentinel-2 imagery and fine-scale ECOSTRESS LST images to provide consistent information regarding changes in coal mining areas and coal fire areas. The ability to process high volume of data and perform analyses in real-time shows a considerable advance in remote sensing technology towards adopting cloud-based applications in mining operations. Integration of these mining characteristics in an interactive, free and openly accessible application can facilitate stakeholders and policy makers in informed decision making and planning mining operations. The application also provides a framework to facilitate future research through sharing of scripts and ensuring reproducibility.
Item Type:Essay (Master)
Faculty:ITC: Faculty of Geo-information Science and Earth Observation
Programme:Geoinformation Science and Earth Observation MSc (75014)
Link to this item:
Export this item as:BibTeX
HTML Citation
Reference Manager


Repository Staff Only: item control page