Estimation of wheat architectural development dynamics from high-throughput field phenotyping data using a hybrid AI-3D plant model approach

Summary

This estimation will be achieved thanks to a hybrid approach consisting in building an AI-based learner aware of a dynamic 3-dimensional (3D) structural model of wheat (AdelWheat).

The LiDAR (Light Detection and Ranging) sensors and RGB cameras onboard the Phenomobile proved effective in recovering the dynamics of canopy level structural traits such as, canopy height, canopy leaf area index, vertical and horizontal distributions of leaf area, head density or average head size. This thesis aims at projecting those observations in a plant-based parametric space using an AI trained with AdelWheat to estimate the best set of parameters that fits the observed high-throughput data. The resulting set of parameters consists of organ-based traits (individual leaf dimensions, leaf insertion height, tiller number…) and development rates (phyllochron, stem elongation rate…) that will ease the future use of field HTP observations for GxE analysis using crop growth models to characterize genotype functioning, since these traits are also crop model parameters or variables.

To achieve this goal, the thesis will address three specific objectives:

1. identify the subset of AdelWheat parameters that most controls the Phenomobile’s observations,
2. develop an AI learning scheme that allows the estimation of AdelWheat parameters from observations,
3. validate the accuracy of advanced traits with existing field measurements.

The thesis will rely on multi-site Phenomobile measurement campaigns conducted between 2022 and 2024 (ANR FFAST project) for 10 wheat cultivars, and two more programmed in 2025 and 2026 for 40 wheat cultivars, including the FFAST panel (Horizon Europe PHENET project).

Contact

• Aurélien Besnier