Assessment of Non-Linearity in Functional-Structural Plant Models

Qiongli Wu, Jessica Bertheloot, Amélie Mathieu, Bruno Andrieu, Paul-Henry CournèdeSPIRE 2010

Abstract: International audience

Introduction

Functional-Structural Plant Models (FSPMs) represent a sophisticated class of simulations that integrate plant physiological processes (functioning) with three-dimensional architectural development (structure). While powerful, these models are inherently complex due to the non-linear feedback loops between growth and form. This study investigates a novel method to quantify this non-linearity, providing a "red flag" system for researchers to identify critical biological transitions and optimize experimental design.

Methodology: The Linearity Index

The researchers utilized Global Sensitivity Analysis (SA), specifically the Standardized Regression Coefficients (SRC) method. The core innovation lies in treating the coefficient of determination ($R^2$) as a dynamic Linearity Index:

  • High Linearity ($R^2 \approx 1$): The model is predictable; inputs explain outputs through simple additive or multiplicative relationships.
  • Low Linearity ($R^2 < 0.5$): The model is highly non-linear, indicating that parameters are interacting in complex, non-additive ways.

Case Study: GreenLab Maize

In the GreenLab Maize model, which focuses on carbon source-sink allocation, the linearity index remains high for most of the growth cycle. However, a significant dip occurs around Growth Cycle (GC) 17.

Linearity Index ($R^2$) across growth cycles for the GreenLab Maize model, showing a characteristic dip during the transition to grain filling.

This mathematical dip serves as a marker for a major biophysical transition: the shift from leaf area expansion to grain filling. As illustrated in the biomass generation data, this is the point where the "Cob" becomes the primary sink for resources, fundamentally changing the model's internal dynamics.

Biomass generation by organ type in the Maize model, highlighting the shift in carbon allocation that drives model non-linearity.

Feedback and Complexity in Poplar and Wheat

The study extended this analysis to models with increasing structural and physiological complexity:

  • GreenLab Poplar: This model incorporates structural plasticity. Once "trophic competition" begins (around GC 5) and branching starts, the linearity index stabilizes at a low 0.5. The feedback loop between available biomass and the creation of new organs introduces significant non-linearity that simple linear approximations cannot capture.
  • NEMA Wheat: This mechanistic model tracks both Carbon and Nitrogen (N). While predicting grain dry mass (Carbon) remains highly linear ($R^2 \approx 1$), Nitrogen uptake is highly non-linear and erratic, reflecting the complex feedback loops between soil nitrate levels and plant activity.

Conclusion and Strategic Experimental Design

Quantifying model linearity allows for more strategic experimental design. During highly linear phases, researchers can afford "lighter" experimental observation. Conversely, during non-linear phases—such as the onset of branching or grain filling—frequent and detailed measurements are essential to capture complex parameter interactions.

While the SRC-based linearity index identifies when complexity arises, it does not fully decompose specific parameter interactions. Future work will employ Sobol’s method to explore higher-order interactions, providing a deeper map of the genetic determinants of plant growth and aiding in the design of virtual "ideotypes" optimized for specific environments.

Summary updated

Abstract

International audience
BibTeX
@inproceedings{wu2010assessment,
  author = {Qiongli Wu and Jessica Bertheloot and Amélie Mathieu and Bruno Andrieu and Paul-Henry Cournède},
  title = {{Assessment of Non-Linearity in Functional-Structural Plant Models}},
  booktitle = {SPIRE},
  year = {2010},
}

Loading related content...

Figure