Spatial distribution and temporal variation of tropical mountaintop vegetation through images obtained by drones

• Published in: Frontiers in environmental science Vol. 11
• Main Authors: Medeiros, Thaís Pereira de, Morellato, Leonor Patrícia Cerdeira, Silva, Thiago Sanna Freire
• Format: Journal Article
• Language: English
• Published: Lausanne Frontiers Research Foundation 10.02.2023; Frontiers Media S.A
• Subjects: Algorithms; Artificial intelligence; Biodiversity; Cameras; Classification; Climate change; Drone aircraft; Drone vehicles; Ecology; Ecosystems; Environmental science; Fieldwork; Grasslands; heterogeneous vegetation; Image analysis; Image processing; Machine learning; random forest; Remote sensing systems; rupestrian grassland; Sensors; Spatial discrimination; Spatial distribution; Spatial resolution; Sustainable development; Temporal variations; UAS; unmanned aerial system; Unmanned aerial vehicles; Vegetation; Vegetation mapping; Wildlife conservation
• ISSN: 2296-665X; 2296-665X
• DOI: 10.3389/fenvs.2023.1083328

Modern UAS (Unmanned Aerial Vehicles) or just drones have emerged with the primary goal of producing maps and imagery with extremely high spatial resolution. The refined information provides a good opportunity to quantify the distribution of vegetation across heterogeneous landscapes, revealing an important strategy for biodiversity conservation. We investigate whether computer vision and machine learning techniques (Object-Based Image Analysis—OBIA method, associated with Random Forest classifier) are effective to classify heterogeneous vegetation arising from ultrahigh-resolution data generated by UAS images. We focus our fieldwork in a highly diverse, seasonally dry, complex mountaintop vegetation system, the campo rupestre or rupestrian grassland, located at Serra do Cipó, Espinhaço Range, Southeastern Brazil. According to our results, all classifications received general accuracy above 0.95, indicating that the methodological approach enabled the identification of subtle variations in species composition, the capture of detailed vegetation and landscape features, and the recognition of vegetation types’ phenophases. Therefore, our study demonstrated that the machine learning approach and combination between OBIA method and Random Forest classifier, generated extremely high accuracy classification, reducing the misclassified pixels, and providing valuable data for the classification of complex vegetation systems such as the campo rupestre mountaintop grassland.