Quantifying time-series of leaf morphology using 2D and 3D photogrammetry methods for high-throughput plant phenotyping

Developed novel computer vision algorithms for time-series plant morphology in 2D and 3D.We used 2D leaf area to quantify leaf nastic movements.3D leaf area showed a reliable time-series leaf growth trend.The throughput of the pipeline is very high comparing to previous studies. Conventional phenotyping methods impose a significant bottleneck to the characterization of genotypic and environmental effects on trait expression in plants. In particular, invasive and destructive sampling methods along with manual measurements widely used in conventional studies are labor-intensive, time-consuming, costly, and can lack consistency. These experimental features impede large-scale genetic studies of both crops and wild plant species. Here, we present a high-throughput phenotyping pipeline using photogrammetry and 3D modeling techniques in the model species, Arabidopsis thaliana. We develop novel photogrammetry and computer vision algorithms to quantify 2D and 3D leaf areas for a mapping population of 1050 Arabidopsis thaliana lines, and use 2D areas to analyze plant nastic movements and diurnal cycles. Compared to the 2D leaf areas, 3D leaf areas show an uncorrupted growth trend regardless of plant nastic movement. With optimized algorithms, our pipeline throughput is very computationally efficient for screening a large number of plants. The pipeline not only supports measurement of organ-level growth and development over time, but also enables analysis of whole-plant phenotypes and, thus, identification of genotype-specific performance. Further, the accuracy results evaluating the relationship between physical dimensions and 3D measurements indicate an R2=0.99, and the average 3D area processing time per plant is 0.02s. Our algorithms provide both high accuracy and throughput in plant phenotyping, thereby, enabling progress in plant genotypic modeling.