Spatio-temporal 4D phenotyping for automated morphological genotype differentiation of sugar beet

Abstract

Abstract 3D models are used in plant phenotyping for non-destructive quantification and analysis of morphological characteristics. Analyzing plant structure allows breeders to select for desirable traits, associated with e.g. drought tolerance or increased productivity. In sugar beet, morphological parameters depict an essential element of the variety approval for distinguishing between genotypes. However, only a limited number of measured or scored parameters are considered at a single time point. In contrast, 4D data adds a temporal component and can depict the dynamic development of 3D parameters. To explore the potential of spatio-temporal 4D phenotyping for automated crop genotype differentiation, a greenhouse experiment was conducted by us covering twelve sugar beet genotypes. High-resolution 3D models were generated twice a week over the course of two months and both common and novel 3D morphological parameters were extracted. The importance of these parameters was assessed by us, and the dataset was analyzed using unsupervised pointwise clustering and time series clustering. Varying importance of parameters depending on the time point and significantly higher importance of plant parameters compared to leaf parameters are demonstrated by our results. Moreover, increased and more stable genotype differentiation is archived using time series clustering compared to pointwise clustering. Furthermore, taproot formation of sugar beet was found to have a crucial impact on morphological development. Substantial variations in the dynamic development of 3D morphological parameters underline the importance of 4D data for plant genotype differentiation. Thus, a novel foundation for genotype differentiation in plant phenotyping is provided by our findings.

Plant phenotyping relevance

3Dモデルから植物形態形質を抽出し、時系列クラスタリングで遺伝型識別を評価する4Dフェノタイピング手法が研究の中心である。

Code and data availability

The paper publicly deposits its generated sugar beet point cloud dataset under CC BY 4.0 at a Dataverse DOI, directly reproducing the paper's phenotyping measurements. Supplementary Python codes and extracted parameter values are stated to be included with the article, but no authors' public URL for the code is present