Authors

Campos JC, Brito JC

Abstract

 

Arid and semi-arid regions comprise high levels of habitat heterogeneity that usually remain undetected in available global land cover (GLC) maps. The Sahara-Sahel is a critical example of how GLC maps continue to oversimplify the complex landscape structure of deserts and arid environments. In this work, we aimed to overcome a generalized knowledge gap concerning the LC heterogeneity of arid regions, using the largest warm desert in the world as a study case. We intended to generate a 30x30m land cover map for the Wet Sahara-Sahel and compare the results with currently available GLC maps (ESA GlobCover and GlobeLand30). To do this, we collected an extensive series of GPS field control points (n = 48,857) and associated descriptive traits. We included the control points in a Hierarchical Cluster Analyses (HCA), and the resulting groups were used as LC classes in Landsat 8 image classification. Independent control points (n = 10,082) suggested a robust regional classification (83.3% correctly classified) of land cover. The Sahara was the most representative ecoregion (around 70% of the study area) and exhibited the highest classification accuracy (91%). The final map is composed by a total of 18 classes, providing a higher number of classes for arid regions than currently available GLC maps. Differences were evident amongst the arid regions of the Sahara, in which the derived map presented a more complex land cover in comparison to the analysed GLC maps. The map derived in this study constitutes framework data for mapping local land cover information and for improving the effectiveness of the assessment and management of natural resources for both local human populations and biodiversity. These results highlight the prevalent need for improving local and regional land cover categorization of arid and semi-arid regions, areas whose land cover heterogeneity is still underrepresented by most of the available GLC maps.

.

 

Journal: ISPRS Journal of Photogrammetry and Remote Sensing

DOI: 10.1016/j.isprsjprs.2018.09.012