Multitemporal field-based plant height estimation using 3D point clouds generated from small unmanned aerial systems high-resolution imagery

• Published in: International journal of applied earth observation and geoinformation Vol. 64; pp. 31 - 42
• Main Authors: Malambo, L., Popescu, S.C., Murray, S.C., Putman, E., Pugh, N.A., Horne, D.W., Richardson, G., Sheridan, R., Rooney, W.L., Avant, R., Vidrine, M., McCutchen, B., Baltensperger, D., Bishop, M.
• Format: Journal Article
• Language: English
• Published: Elsevier B.V 01.02.2018
• Subjects: 3D point cloud; Above-ground level; Field-based; High-throughput phenotyping; Maize; Multitemporal; Pix4D; Plant height; Sorghum; Structure from motion; Terrestrial laser scanning; Unmanned aerial systems; Multitemporal; Structure from motion; Above-ground level; Field-based; Pix4D; 3D point cloud; Sorghum; Terrestrial laser scanning; Maize; Plant height; Unmanned aerial systems; High-throughput phenotyping
• ISSN: 1569-8432; 1872-826X
• DOI: 10.1016/j.jag.2017.08.014

•Demonstrates production of repeated crop height data using unmanned aerial systems.•Evaluates structure from motion (SfM) data against manual and LiDAR measurements.•Presents a plot-based plant height estimation approach using SfM point clouds.•Demonstrates SfM-based height estimation’ effectiveness and inconsistency over time.•Discusses changing field environment and its impact on SfM 3D reconstruction. Plant breeders and agronomists are increasingly interested in repeated plant height measurements over large experimental fields to study critical aspects of plant physiology, genetics and environmental conditions during plant growth. However, collecting such measurements using commonly used manual field measurements is inefficient. 3D point clouds generated from unmanned aerial systems (UAS) images using Structure from Motion (SfM) techniques offer a new option for efficiently deriving in-field crop height data. This study evaluated UAS/SfM for multitemporal 3D crop modelling and developed and assessed a methodology for estimating plant height data from point clouds generated using SfM. High-resolution images in visible spectrum were collected weekly across 12 dates from April (planting) to July (harvest) 2016 over 288 maize (Zea mays L.) and 460 sorghum (Sorghum bicolor L.) plots using a DJI Phantom 3 Professional UAS. The study compared SfM point clouds with terrestrial lidar (TLS) at two dates to evaluate the ability of SfM point clouds to accurately capture ground surfaces and crop canopies, both of which are critical for plant height estimation. Extended plant height comparisons were carried out between SfM plant height (the 90th, 95th, 99th percentiles and maximum height) per plot and field plant height measurements at six dates throughout the growing season to test the repeatability and consistency of SfM estimates. High correlations were observed between SfM and TLS data (R2=0.88–0.97, RMSE=0.01–0.02m and R2=0.60–0.77 RMSE=0.12–0.16m for the ground surface and canopy comparison, respectively). Extended height comparisons also showed strong correlations (R2=0.42–0.91, RMSE=0.11–0.19m for maize and R2=0.61–0.85, RMSE=0.12–0.24m for sorghum). In general, the 90th, 95th and 99th percentile height metrics had higher correlations to field measurements than the maximum metric though differences among them were not statistically significant. The accuracy of SfM plant height estimates fluctuated over the growing period, likely impacted by the changing reflectance regime due to plant development. Overall, these results show a potential path to reducing laborious manual height measurement and enhancing plant research programs through UAS and SfM.