Tree Species Mapping in Mangrove Ecosystems Using UAV-RGB Imagery and Object-Based Image Classification

• Published in: Journal of the Indian Society of Remote Sensing Vol. 51; no. 10; pp. 2095 - 2103
• Main Authors: Miraki, Mojdeh, Sohrabi, Hormoz, Immitzer, Markus
• Format: Journal Article
• Language: English
• Published: New Delhi Springer India 01.10.2023; Springer Nature B.V
• Subjects: Accuracy; Algorithms; Classification; Color imagery; Earth and Environmental Science; Earth Sciences; Image classification; Mangrove trees; Mapping; Plant species; Remote Sensing/Photogrammetry; Research Article; Segments; Unmanned aerial vehicles; CHM; Random forest; Mangrove forest; UAV; Sirik Azini estuary
• ISSN: 0255-660X; 0974-3006
• DOI: 10.1007/s12524-023-01752-7

Unmanned aerial vehicles (UAV) imagery has proved to be useful in the efficient protection and management of mangrove forests. However, there have been few attempts to show that UAV-RGB images may be used for mapping trees at the species level. Our objective, in this study, is to identify two mangrove species using object-based classification. Height information was used to segment trees to obtain maximum spectral purity in each segment for classification using the Canopy Height Model (CHM) in Sirik mangrove forest (Azini Creek) located in southern Iran. The object-based classification (using a random forest algorithm) of UAV imagery with dominant mangrove features (i.e., Rhizophora mucronata , Avicennia marina , ground/sand, and water) achieved an overall accuracy (OA) of 98% and Kappa coefficient of 0.97. The results showed that the overall accuracy and Kappa were upgraded from 94 to 98% and 0.91–0.97, respectively. The water and ground classes were identified with a producer’s accuracy of 100%. The random forest algorithm accuracy for both of the trees was more than 90% (produce accuracy 95 and 98% and user accuracy 98 and 97% for R. mucronata and A. marina , respectively). The results demonstrated proof for the potential and usefulness of spectral data, i.e., UAV–RGB derived orthomosaic, and structural, i.e., CHM data for mangrove trees identification.