Inverse Estimation of Root Architectural Model Parameters Based on Field Sampling Methods

Root architecture of different phenotypes at different life time stages is important information to assess e.g. their resource uptake efficiency. Obtaining individual root architectural traits, e.g. number of primary roots, distance between lateral roots, branching angles, etc. from field data such as root length density from soil cores or root count densities in trenches remains a challenge as they represent more aggregated information about the root system. Virtual experimental studies on root sampling methods showed that the sensitivity of the model output to the different root architectural parameters varies with the sampling method and observed model output (such as root length density at different depths in the soil profile, maximal rooting depth etc.). Based on characteristics of the observed root distributions that are sensitive to certain parameters, we defined objective functions and developed a method for estimating root architecture parameters from field data.

We use our python version of the Markov chain Monte Carlo (MCMC) DREAMzs algorithm parallelized to estimate within a Bayesian framework 17 root architecture parameters from three different types of synthetic field data: coring, trench root counting and minirhizotron methods. The used forward model is RootBox and the inverted data include maximum rooting depths, root length density with depth, root intersection counts, and arrival times to the rhizotubes. To minimize the stochasticity of RootBox during the inversion, each time a new parameter set is evaluated the associated model outputs are averaged over 128 forward runs. Our results based on virtually simulated sampling data show that different sampling schemes along with respective root distribution characteristic functions allow retrieving the most sensitive parameters of the root system architecture by inversion. Our approach is an important step towards retrieving root architecture of plants by inversion of field root sampling data.