The use of mathematics and computer technology to model, simulate and visualize the process of plant growth modeling could help us to explore the law of plant growth and predict the final yield of plants, which is of great significance for guiding agricultural practices. As one of the representatives of functional-structural models of plant growth, GreenLab has been gradually applied to various crops and tree growth simulation. In general, calibration is needed before models can be applied. Hidden parameters in GreenLab model, which cannot be measured directly, should be fitted using measurement by model inversion. Traditional fitting process utilizes the least squared method to approximate the model output to field data. However, fitting methods based on gradient descent frequently converge to local minimum and rely on the initial values of parameters chosen for iteration, which often lead to poor fitting result. In summary, sensitivity analysis of the GreenLab model was studied in this paper and the accomplishments of this work include the following aspects: (1) Sensitivity analysis of model with multiple outputs was carried out by analyzing the source-sink ratio at each growth cycle without missing the global information, which reflects the advantages of the GreenLab model. (2) Model parameters were fitted automatically and sequentially by choosing appropriate subset based on the sensitivity analysis result. Experiment showed that the fitting process is less dependent on prior knowledge of possible parameter values.
修改评论