Phenotypic Trait Monitoring of Victoria amazonica Plants Using Unmanned Aerial Vehicles

Abstract

The study investigates the potential of unmanned aerial vehicles (UAVs) for acquiring phenotypic trait parameters of Victoria amazonica in an open pond environment. Sequential 2D images using UAVs were acquired from multiple views on a weekly basis. The structure from motion (SfM) technique was then used to build high-resolution orthomosaic and 3D models of the mapping area. Measurements corresponding to typical leaf, petiole, and flower growth were made using digital models. It was observed that the digital models could represent the actual ground truth values for all the traits with a ground sample distance ranging between 0.25 and 0.4 cm/pixel. A comparison of digital and manually measured phenotypic trait data revealed that the UAV-based measurements could predict on par with the conventional manual measurements. Additionally, linear regression fits generated for digital and manual trait data resulted in adjusted coefficients of determination (Adj. R2) of atleast 0.98 for all parameters. The trait data were also statistically analyzed to assess the growth rates of various parameters during the monitoring period. It is observed that the leaf rim height and petiole parameters are the highly sensitive and varying traits (COV range: 53-78%) for Victoria amazonica species. Besides addressing the problems with manual phenotyping, the proposed methodology provides an easy, flexible, frequent, accurate, contactless, non-destructive and cost-effective solution for aquatic plant research.