Evaluating the potential of freely available multispectral remotely sensed imagery in mapping American bramble (Rubus cuneifolius)

Abstract

Globally, alien invasive plant species are considered a serious threat to native flora and fauna. In the eastern parts South Africa, the American bramble (Rubus cuneifolius) has been identified as one of the major threats to social and ecological systems. Optimal management and mitigation of American bramble spread requires reliable and cost effective approaches to determine invaded spatial extents. In this study, we test the value of the recently launched, freely available Sentinel-2 (S2), as opposed to conventional Landsat 8 imagery in mapping the American bramble. Using the Support Vector Machine classification algorithm, we seek to identify the optimal season for mapping the American bramble as well as the most influential bands in the classification process. Results show that Sentinel-2 out-performed Landsat 8 in all seasons, with summer providing the highest classification accuracy (77% accuracy). The study also shows that strategically placed Sentinel-2 bands of Near Infrared, Red edge and Short Wave Infrared significantly contribute to an increase in overall bramble mapping accuracy. This study demonstrates the value of freely available multispectral imagery in mapping American bramble at large spatial extents, hence valuable for cost-effective operational use.

Keywords:

• Support Vector Machine (SVM)
• American bramble
• multi-spectral remote sensing
• Sentinel-2
• Landsat 8
• Rubus cuneifolius