A Hybrid UAV Ground-Truthing Methodology for Mapping Invasive Plant Species in Australian Rangelands.

Abstract

Accurate ground truth data underpins the development of robust machine-learning models for unmanned aerial vehicle (UAV) based weed detection. In remote Australian rangelands, however, acquiring such data is constrained by limited mobile connectivity, heterogeneous terrain and the visual similarity of co-occurring invasive species. This paper presents a hybrid field methodology for collecting ground truth data to support UAV-based mapping of Chinee apple (Ziziphus mauritiana) and Prickly acacia (Vachellia nilotica) weeds across four sites near Bowen, Queensland. Two complementary ground-truthing technologies were evaluated and combined: a low-cost manual method using colourcoded
cardboard markers, and an automated, surveygrade real-time kinematic (RTK) Global Navigation Satellite System (GNSS) using a DJI D-RTK 3 Mobile Station. A total of 192 markers were placed across three connectivity-limited sites, while centimetre-level coordinates were logged for ten individuals of each species at the RTK site. The two technologies were not flown and compared at the same site, and this paper is therefore presented as a preliminary, descriptive assessment of their respective feasibility and data yield rather than a controlled comparison of mapping accuracy.