Estimating Dry Biomass Partitioning Coefficients in a Dynamic Vegetation Model.

Evaluating ecosystem services and sustainability of bioenrgy crops requires a detailed understanding of their potential for carbon sequestration, nutrient cycling and water balance. Dynamic vegetation models can be used to address these issues, but often there are limitations due to the data needed to parameterize these models as well as lack of algorithms for parameter estimation.  Here I propose a dynamic vegetation model (BioCro) and algorithms for parameter estimation in perennial crops. Although many parameters can be easily estimated using general optimization algorithms others pose additional challenges due to linear restrictions and correlations among parameters. In BioCro biomass partitioning is controlled by 24 parameters which determine how the carbohydrates assimilated through photosynthesis are distributed into plant structures such as rhizome, roots, stems and leaves.  The 24 parameters correspond to 4 coefficients for each of 6 phenological stages. How do we estimate dry biomass partitioning coefficients in a dynamic vegetation model?  I developed three methods for parameter estimation. The first one is based on a reparameterization of the model in which the new parameters are not constrained and a general non-linear optimization method is used. The second method uses a constrained optimization algorithm which imposes a logarithmic barrier on a general unconstrained optimization routine. The third method is a hybrid approach which starts with a simulated annealing algorithm and transitions to a Markov chain Monte Carlo with a Metropolis-Hasting acceptance rule. To evaluate the relative merit of the three approaches data will be generated from the model with known parameters and the ability and efficiency of the methods to recover these parameters will be assessed. I will also show a sensitivity analysis to illustrate other parameters which exert a strong control on the estimation process.