When you hear of coal production in America, what comes to mind? Perhaps you imagine a rugged man with a miner’s lamp on his helmet descending into a tunnel several hundred feet below the ground. Or maybe you picture giant machines removing topsoil and bedrock from a forested West Virginia mountain. But what if I told you most of the coal produced in America is mined from the arid grasslands of Wyoming and Montana?
According to the 2017 Federal Coal Program, “85% of production occurs in the arid region of Wyoming and Montana known as the Powder River Basin”. The Gillette coalfield in Wyoming contains the largest deposits of low-sulfur sub-bituminous coal. The area is flat, and the coal seams are very thick and close to the surface, making it much easier (and cheaper) to extract from open-pit mines, compared to the cost and effort of removing Appalachian mountaintops.
Mountaintop removal mining (MTR) is reshaping the Appalachian landscape. In the spring of 2016, Duke University and SkyTruth created a Google Earth Engine script to process satellite imagery and derive an accurate, annually updated map and GIS dataset of MTR operations across Appalachia. Google Earth Engine is a cloud-based geospatial processing platform with access to satellite imagery archives. For this work, we used Landsat imagery from 1985 to 2015. A band ratio was used on the imagery to identify active mining operations and to discriminate bare surfaces from vegetated land. A normalized difference vegetation index (NDVI) is a ratio of the red band to the infrared band. We chose this band ratio because vegetation will use red light but reflect infrared, while bare rock and soil strongly reflect both. The script determines an NDVI threshold based on testing the results against thousands of manually classified control points randomly scattered throughout the project area. If the NDVI value of a given pixel falls below the automatically determined threshold, it is classified as active mining.
Part of my summer internship was devoted to adapting this process for mining operations in the Powder River Basin. The first step in applying this script was to create a mask. Its purpose is to mask out everything that could be misclassified as mining because it’s a bare surface, like lakes, streams, roads, railroads, urban areas, etc. This data was collected from US Census TIGER shapefiles and merged to generate a raster mask. However, unlike Appalachian MTR operations, Powder River Basin coal mining is also surrounded by natural gas and oil drilling sites. To mask out these fracking pads, well permits were downloaded from the Wyoming Oil & Gas Conservation Commission, then added to the mask. Variables such as coal mining permits, and county boundaries (Converse and Campbell) were added for Wyoming.
The vastly different climate proved difficult in this adaptation. While Appalachia is mostly mountainous deciduous forest, the high plains of eastern Wyoming are flat and semi-arid. There are naturally many small barren areas or badlands in this region that mimic mining operations, at least from the satellite’s perspective. My solution was to filter the results by eliminating any area classified as active mining that was less than 300,000 square meters (m2) in size. This threshold was determined during some post-process editing, where I examined all of the areas classified as mining that fell outside the boundaries of mining permits, and the largest was 300,000 m2. The resulting output only retained the larger vectors located within the permits and can be seen below.
As you can see, this approach yields reliable results. I’m confident the methodology we demonstrated in Appalachia can work for coal mining out West. It is worth experimenting with changing the NDVI threshold to see if we can come up with a better tradeoff between identifying the active mining areas, and misclassifying badlands and other non-mining barren areas.