Unsupervised multiple endmember spectral mixture analysis-based detection of opium poppy fields from an EO-1 Hyperion image in Helmand, Afghanistan

• We used EO-1 Hyperion hyperspectral data to detect opium fields in Afghanistan.
• We proposed an unsupervised methodology to detect opium fields.
• The retrieved opium field distribution map showed agreement with groundtruth data.
• The 2- and 3-endmember models performed better than 2-, 3- and 4-endmember models.
• Multiple endmembers SMA performed better than simple SMA.

Since 1992, Afghanistan has gradually become the primary illicit opium producer in the entire world. To efficiently eradicate the opium poppy, it is crucial for the United Nations Office on Drugs and Crime (UNODC) and Ministry of Counter Narcotics of Afghanistan to monitor opium poppy cultivation timely. In situ detection of opium fields, however, is often expensive, time-consuming and dangerous in Afghanistan. To overcome the constraints of inaccessibility of opium fields, high-resolution (≤ 1 m) images, like pan-sharpened IKONOS, have been applied in previous studies. Unfortunately, these high-resolution images are expensive when monitoring a large area. In contrast, EO-1 Hyperion imagery, the only source of spaceborne hyperspectral data, has a coarse resolution (30 m), but it is free of charge. Moreover, Hyperion's large number of channels may increase the detection capability of subpixel size targets. Until now, however, little research has been found that identified opium fields from spaceborne or aerial hyperspectral images. Therefore, this study attempts to detect opium fields from a Hyperion image covering a study area in Southwest Afghanistan in a situation where training samples were not available. A proposed methodology based on unsupervised endmember-selection and multiple-endmember spectral mixture analysis can detect opium fields directly from the Hyperion image. The number of poppy pixels was overestimated by 12%.