Improving the unsupervised mapping of riparian bugweed in commercial forest plantations using hyperspectral data and LiDAR
- Resource Type
- article
- Authors
- Kabir Peerbhay; Onisimo Mutanga; Romano Lottering; Na’eem Agjee; Riyad Ismail
- Source
- Geocarto International, Vol 36, Iss 4, Pp 465-480 (2021)
- Subject
- unsupervised random forest
anomaly detection
hyperspectral
lidar
Physical geography
GB3-5030
- Language
- English
- ISSN
- 1010-6049
1752-0762
10106049
Accurate spatial information on the location of invasive alien plants (IAPs) in riparian environments is critical to fulfilling a comprehensive weed management regime. This study aimed to automatically map the occurrence of riparian bugweed (Solanum mauritianum) using airborne AISA Eagle hyperspectral data (393 nm–994 nm) in conjunction with LiDAR derived height. Utilising an unsupervised random forest (RF) classification approach and Anselin local Moran’s I clustering, results indicate that the integration of LiDAR with minimum noise fraction (MNF) produce the best detection rate (DR) of 88%, the lowest false positive rate (FPR) of 7.14% and an overall mapping accuracy of 83% for riparian bugweed. In comparison, utilising the original hyperspectral wavebands with and without LiDAR produced lower DRs and higher FPRs with overall accuracies of 79% and 68% respectively. This research demonstrates the potential of combining spectral information with LiDAR to accurately map IAPs using an automated unsupervised RF anomaly detection framework.