Spatio-temporal modelling of high-throughput phenotyping data

Abstract

High throughput phenotyping (HTP) platforms and devices are increasingly used to characterise growthand developmental processes for large sets of plant genotypes. This dissertation is motivated by the needto accurately estimate genetic effects over time when analysing data from such HTP experiments. TheHTP data we deal with here are characterised by phenotypic traits measured multiple times in thepresence of spatial and temporal noise and a hierarchical organisation at three levels (populations,genotypes within populations, and plants within genotypes). To that aim, we propose two approaches.The first proposal divides the problem into two stages. The first stage (spatial model) focuses oncorrecting the phenotypic data for experimental design factors and spatial variation, while the secondstage (hierarchical longitudinal model) aims to estimate the evolution over time of the genetic signal. Thesecond proposal is to face the problem simultaneously (one-stage approach). That is, modelling thelongitudinal evolution of the genetic effect on a given phenotypic trait while accounting for the temporaland spatial effects of environmental and design factors (spatio-temporal hierarchical model). We followthe same modelling philosophy throughout the thesis and propose multidimensional P-spline-basedhierarchical approaches. The challenge is to balance efficient statistical models and computationalsolutions to deal with the complexity and dimensionality of the experimental data. We provide the userwith appealing tools that take advantage of the sparse model matrices structure to reduce computationalcomplexity. All our codes are publicly available on the R-package statgenHTP andhttps://gitlab.bcamath.org/dperez/htp_one_stage_approach. We illustrate the performance of our methodsusing spatio-temporal simulated data and data from the PhenoArch greenhouse platform at INRAEMontpellier and the outdoor Field Phenotyping platform at ETH Zürich. In the plant breeding context, weshow how to extract new time-independent phenotypes for genomic selection purposes.