Scientific abstract:

Towards automated mapping of global mining land use Advances in the quality and accessibility of satellite imagery have prompted rapid growth in research mapping the land footprint of mining. Multiple research teams have recently compiled open datasets with more than 150,000 polygons covering mining activities worldwide. These data help to explain the size, spread and nature of land use challenges linked to global material supply chains. Yet so far, it has only been viable to gather such data through a time-consuming manual process that requires trained analysts to visually recognise and delineate mine areas. Consequently, published updates on global mining land use are limited to approximately every two years. Meanwhile, mines are highly dynamic, constantly changing and expanding into new land. To keep pace with the real-time changes in mine areas globally, efforts to automate the task are needed. This presentation outlines recent advances in the use of machine learning algorithms to automatically detect mine areas in satellite imagery. Building from this, we will discuss barriers and progress towards automating the global mapping of mine areas. Through a series of mapping case studies, we will also illustrate what levels of geometric and categorical accuracy can be achieved for different types of mine features, and for different parts of the world. Finally, we will discuss the implications of access to timely global mine land use data on broader field of industrial ecology, on governments, and the mining industry itself.