Statistical analysis and modeling of root measures for the description of spatiotemporal root patterns, using experimental and simulated image data gained by X-ray CT and root architecture models

点击翻译按钮获取中文摘要

The 3D microstructure of roots plays a key role for biological, chemical and physical processes that drive rhizosphere and root structure formation and function. X-ray computed tomography (CT) is a powerful technology to study spatiotemporal root growth patterns in 3D. However, simulated root architectures provide additional insights, e.g. through faster data acquisition and higher temporal resolution. In both cases, i.e. in experimental and virtual investigations of root growth patterns, large amounts of complex image data are generated, which need to be statistically analyzed and modeled using as few as possible model parameters. In a recent publication together with the group of D. Vetterlein (UFZ), we proposed a root distance model, which is able to describe root growth patterns throughout all stages in the first weeks of growth of Vicia faba. In a further paper with the group of A. Schnepf (FZJ), we investigated the connection between the input parameters of the 3D root architecture model CRootBox and various measures of the simulated root systems, like root length density and volume of the convex hull.The aim of the present project is to continue and extend the fruitful collaborations with the Vetterlein and Schnepf groups. First, we continue to statistically analyze (experimentally observed and simulated) root growth patterns from the soil perspective. In addition to analyzing entire root systems via root distance models, we develop local root distance models with respect to specific classes of root segments, e.g. segments which are older (proximal to the seed) or segments being at, or near, the tips of roots. This will give us more detailed insight into the dynamics of root growth and function. Furthermore, quantitative relationships will be established between the input parameters of the 3D root architecture model CRootBox and various root measures. Multivariate approaches such as copulas provide the mathematical tools to build parametric meta-models for vectors of (correlated) root measures. The results will be used to develop a universally applicable approach for the target-oriented calibration of root architecture models. In particular, we will show how methods of machine learning can be combined with the results obtained, in order to calibrate CRootBox by means of tomographic root image data or derived measures. An additional topic is the statistical description of geometrical root patterns to distinguish between purely random, even and clustering morphologies. Methods of stochastic geometry provide yet another perspective in the analysis of growing root systems and will be used to study, e.g. the correlation of root piercing point patterns in planar (e.g. vertical or horizontal) sections of soil with chemical 2D maps. Last but not least, we will perform a comparative statistical analysis of root measures in constrained and unconstrained root architectures.