Detection and spatial analysis of selective logging with geometrically corrected Landsat images

Abstract

The Brazilian Amazonian rain forests are under imminent threat of serious degradation and ultimately deforestation. Human activities such as selective logging are an important cause. Selectively logged locations are difficult to detect from medium-resolution Landsat images, due to their relatively small sizes and subtle spatial patterns. Spectral linear unmixing provides an effective tool for the purpose. The orientation of geometrically corrected images, however, artificially introduces zero-reflectance background pixels. These change the variance–covariance structure of the image bands and hinder the identification of pure endmembers. In this study, we compare image cropping and image rotation as two alternative approaches. Selectively logged forests were detected in northern Rondônia state, north-western Mato Grosso state and south-eastern Amazonas state in Brazil by applying spectral unmixing. The study shows that image rotation is a better approach as it preserves the image extent and thus provides information on forest degradation over a wider region. Spatial statistical analysis of the detected locations shows strong clustering within the study area. We conclude that the endmembers used in this study represent basic components of a degraded forest environment. As spectral unmixing of remote-sensing images avoids collection of field data, it may broadly be applied towards other Amazonian regions as well.