Exploring the limitations of automated classification and feature extraction: A case study examining clay kilns in Kabul, Afghanistan

Authors: Kelsey O'Pry*, Natural Systems Analyst, Inc., Jessica D DeWitt, United States Geological Survey
Topics: Remote Sensing, Middle East, Natural Resources
Keywords: remote sensing, image interpretation, image classification, Afghanistan, artisanal small-scale mining (ASM)
Session Type: Poster
Day: 4/6/2019
Start / End Time: 8:00 AM / 9:40 AM
Room: Lincoln 2, Marriott, Exhibition Level
Presentation File: Download


Clay and clay bricks are important industrial materials used in the Kabul, Afghanistan region. Reconstruction and growth of infrastructure in Kabul has prompted a corresponding growth in clay mining and brick kiln development. However, lack of regulation has significant worker health and safety concerns as well as environmental implications. Additionally, illegal taxes and control by criminal or terrorist elements has been reported. Recent government reorganization has the potential to improve regulation of the mining sector, but due to its informality, no record of clay mine or kiln locations currently exists to facilitate oversight and monitoring. This study explores various remote sensing techniques, including spectral- and object-oriented classification methods, to map the locations of clay kilns from multispectral imagery, as compared to the manual image interpretation or more dangerous field mapping methods. Specifically, band ratio analysis, random forest and support vector machine classification techniques, and object-based image analysis methods were tested using Landsat 8, Sentinel-2, and WorldView-3 imagery. Short-wave infra-red bands available in Landsat 8 and Sentinel-2 are often used to differentiate surficial geology and clay minerology, however the spatial resolution of these sensors is too coarse for accurate kiln classification. WorldView-3 has comparatively fine spatial resolution but only visible and near-infrared data are available in the study area. The automated classification methods are also inhibited by the pervasiveness of clay in the semi-arid steppe landscape of the Kabul Basin. Ultimately, this study found that manual interpretation produced the most accurate data for mapping brick kilns in Kabul.

Abstract Information

This abstract is already part of a session. View the session here.

To access contact information login