Parameter Sensitivity Analysis of a Root Architecture Model: Field Scale Simulation of Root Systems and Virtual Field Sampling

Traits of the plant root architecture play a key role in crop performances, especially under non-optimal environmental and climatic conditions. Therefore, plant phenotyping methods based on architectural root traits are becoming increasingly important to develop new genotypes. However, the main challenge in root phenotyping programs is the limited accessibility to the root system because it is hidden in the soil. Classical field root sampling methods such as coring, trenching, and minirzhizotrons provide only aggregated data about the root system architectures. Architectural root traits can be represented by parameters of root architecture models. When describing the root architecture using a model, methods like sensitivity analysis and inverse modelling can be deployed to investigate which parameters are sensitive to and can be derived from aggregated data. We investigate which parameters are highly sensitive to data obtained from different types of field root sampling schemes. The root growth model “CRootBox” was used to simulate virtual 3-D root systems of wheat and maize plants in field plots. The ground truth of the virtual experiment was established for coring, trench profile and minirhizotron methods. From these data, root length density, root intersection density profiles and arrival curves for rhizotubes were computed and considered to be the observed data. We then performed a sensitivity analysis to find out which RSA parameters have the significant influence to the virtual field sampling results. Finally, we analyzed the spatial/temporal correlation of the field sampling data to find out the influence on the likelihood function used for optimization. We found that the numbers of branches, maximum length, internodal distance, elongation rate, insertion angles of main roots are the most sensitive and the parameters of branching roots are less sensitive to the model. Moreover, different sampling schemes and respective error function show different degrees of sensitivities of root architecture parameters. Most sensitive parameters could be retrievable using suitable optimization algorithm. This is an important step in the characterization of root traits from field observations. These root architectures can be used in models that simulate water and nutrient uptake so as to evaluating the performance of root systems with certain traits