Thu, Aug 05, 2021:On Demand
Background/Question/Methods
Root traits are characteristics of an individual plant that reflect or affect ecosystem properties such as the soil carbon cycle, microbial activity, and mycorrhizal colonization. There are various methods for characterizing root structure and traits; however, these methods remain labor-intensive and time-consuming. Moreover, these methods typically characterize roots in a two-dimensional (2D) perspective, and like all living organisms, roots are three-dimensional (3D) objects. Thus, new approaches are needed to fully characterize root structure. Here, we propose using the computer vision technique of Structure from Motion to fully characterize roots in 3D by constructing a computational root structure model. In order to create such photogrammetric 3D models of roots, we used a Go-Pro Hero 7 and a cellphone camera to take photographs and videos in a systematic way (every ~10°) of roots in two different settings: 1) underwater and 2) suspended in the air. We processed the photographs using AgiSoft, aligned the matching features of all the photographs to create a cloud of points, and then built a 3D model. From the resulting 3D model, we measured the root length (n=12) of a potted plant in AgiSoft and manually.
Results/Conclusions From linear regression analysis, we found a strong relationship between the root length measured manually compared with the 3D model (r2 = 0.96, p<0.001). Our results suggest that root traits, in our case root length, can be effectively measured from the constructed photogrammetric 3D model. Furthermore, we could potentially use the 3D models to measure other traits like root diameter, length density, and root order. Moreover, with 3D models, we will be able to characterize other important root traits more precisely, such as root volume, that might be difficult to reconstruct from 2D measurements. We believe that our proposed methodology for scientists interested in assessing root traits leveraging 3D photogrammetry, providing a scalable approach for better characterizing roots, and take us a step closer to a better understanding of root ecology.
Results/Conclusions From linear regression analysis, we found a strong relationship between the root length measured manually compared with the 3D model (r2 = 0.96, p<0.001). Our results suggest that root traits, in our case root length, can be effectively measured from the constructed photogrammetric 3D model. Furthermore, we could potentially use the 3D models to measure other traits like root diameter, length density, and root order. Moreover, with 3D models, we will be able to characterize other important root traits more precisely, such as root volume, that might be difficult to reconstruct from 2D measurements. We believe that our proposed methodology for scientists interested in assessing root traits leveraging 3D photogrammetry, providing a scalable approach for better characterizing roots, and take us a step closer to a better understanding of root ecology.